asterisk/res/res_pjsip_pubsub.c

5940 lines
184 KiB
C
Raw Permalink Normal View History

/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 2013, Digium, Inc.
*
* Mark Michelson <mmichelson@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*!
* \brief Opaque structure representing an RFC 3265 SIP subscription
*/
/*** MODULEINFO
<depend>pjproject</depend>
<depend>res_pjsip</depend>
<support_level>core</support_level>
***/
#include "asterisk.h"
#include <pjsip.h>
#include <pjsip_simple.h>
#include <pjlib.h>
#include "asterisk/mwi.h"
#include "asterisk/res_pjsip_pubsub.h"
#include "asterisk/module.h"
#include "asterisk/linkedlists.h"
#include "asterisk/astobj2.h"
#include "asterisk/datastore.h"
#include "asterisk/uuid.h"
#include "asterisk/taskprocessor.h"
#include "asterisk/sched.h"
#include "asterisk/res_pjsip.h"
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
#include "asterisk/callerid.h"
#include "asterisk/manager.h"
#include "asterisk/cli.h"
#include "asterisk/test.h"
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
#include "res_pjsip/include/res_pjsip_private.h"
#include "asterisk/res_pjsip_presence_xml.h"
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
/*** DOCUMENTATION
<manager name="PJSIPShowSubscriptionsInbound" language="en_US">
<synopsis>
Lists subscriptions.
</synopsis>
<syntax />
<description>
<para>
Provides a listing of all inbound subscriptions. An event <literal>InboundSubscriptionDetail</literal>
is issued for each subscription object. Once all detail events are completed an
<literal>InboundSubscriptionDetailComplete</literal> event is issued.
</para>
</description>
</manager>
<manager name="PJSIPShowSubscriptionsOutbound" language="en_US">
<synopsis>
Lists subscriptions.
</synopsis>
<syntax />
<description>
<para>
Provides a listing of all outbound subscriptions. An event <literal>OutboundSubscriptionDetail</literal>
is issued for each subscription object. Once all detail events are completed an
<literal>OutboundSubscriptionDetailComplete</literal> event is issued.
</para>
</description>
</manager>
<manager name="PJSIPShowResourceLists" language="en_US">
<synopsis>
Displays settings for configured resource lists.
</synopsis>
<syntax />
<description>
<para>
Provides a listing of all resource lists. An event <literal>ResourceListDetail</literal>
is issued for each resource list object. Once all detail events are completed a
<literal>ResourceListDetailComplete</literal> event is issued.
</para>
</description>
</manager>
<configInfo name="res_pjsip_pubsub" language="en_US">
<synopsis>Module that implements publish and subscribe support.</synopsis>
<configFile name="pjsip.conf">
<configObject name="subscription_persistence">
<synopsis>Persists SIP subscriptions so they survive restarts.</synopsis>
<configOption name="packet">
<synopsis>Entire SIP SUBSCRIBE packet that created the subscription</synopsis>
</configOption>
<configOption name="src_name">
<synopsis>The source address of the subscription</synopsis>
</configOption>
<configOption name="src_port">
<synopsis>The source port of the subscription</synopsis>
</configOption>
<configOption name="transport_key">
<synopsis>The type of transport the subscription was received on</synopsis>
</configOption>
<configOption name="local_name">
<synopsis>The local address the subscription was received on</synopsis>
</configOption>
<configOption name="local_port">
<synopsis>The local port the subscription was received on</synopsis>
</configOption>
<configOption name="cseq">
<synopsis>The sequence number of the next NOTIFY to be sent</synopsis>
</configOption>
<configOption name="tag">
<synopsis>The local tag of the dialog for the subscription</synopsis>
</configOption>
<configOption name="endpoint">
<synopsis>The name of the endpoint that subscribed</synopsis>
</configOption>
<configOption name="expires">
<synopsis>The time at which the subscription expires</synopsis>
</configOption>
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
<configOption name="contact_uri">
<synopsis>The Contact URI of the dialog for the subscription</synopsis>
</configOption>
<configOption name="prune_on_boot">
<synopsis>If set, indicates that the contact used a reliable transport
and therefore the subscription must be deleted after an asterisk restart.
</synopsis>
</configOption>
<configOption name="generator_data">
<synopsis>If set, contains persistence data for all generators of content
for the subscription.
</synopsis>
</configOption>
</configObject>
<configObject name="resource_list">
<synopsis>Resource list configuration parameters.</synopsis>
<description>
<para>This configuration object allows for RFC 4662 resource list subscriptions
to be specified. This can be useful to decrease the amount of subscription traffic
that a server has to process.</para>
<note>
<para>Current limitations limit the size of SIP NOTIFY requests that Asterisk sends
to double that of the PJSIP maximum packet length. If your resource list notifications
are larger than this maximum, you will need to make adjustments.</para>
</note>
</description>
<configOption name="type">
<synopsis>Must be of type 'resource_list'</synopsis>
</configOption>
<configOption name="event">
<synopsis>The SIP event package that the list resource belong to.</synopsis>
<description><para>
The SIP event package describes the types of resources that Asterisk reports
the state of.
</para>
<enumlist>
<enum name="presence"><para>
Device state and presence reporting.
</para></enum>
<enum name="dialog"><para>
This is identical to <replaceable>presence</replaceable>.
</para></enum>
<enum name="message-summary"><para>
Message-waiting indication (MWI) reporting.
</para></enum>
</enumlist>
</description>
</configOption>
<configOption name="list_item">
<synopsis>The name of a resource to report state on</synopsis>
<description>
<para>In general Asterisk looks up list items in the following way:</para>
<para>1. Check if the list item refers to another configured resource list.</para>
<para>2. Pass the name of the resource off to event-package-specific handlers
to find the specified resource.</para>
<para>The second part means that the way the list item is specified depends
on what type of list this is. For instance, if you have the <replaceable>event</replaceable>
set to <literal>presence</literal>, then list items should be in the form of
dialplan_extension@dialplan_context. For <literal>message-summary</literal> mailbox
names should be listed.</para>
</description>
</configOption>
<configOption name="full_state" default="no">
<synopsis>Indicates if the entire list's state should be sent out.</synopsis>
<description>
<para>If this option is enabled, and a resource changes state, then Asterisk will construct
a notification that contains the state of all resources in the list. If the option is
disabled, Asterisk will construct a notification that only contains the states of
resources that have changed.</para>
<note>
<para>Even with this option disabled, there are certain situations where Asterisk is forced
to send a notification with the states of all resources in the list. When a subscriber
renews or terminates its subscription to the list, Asterisk MUST send a full state
notification.</para>
</note>
</description>
</configOption>
<configOption name="notification_batch_interval" default="0">
<synopsis>Time Asterisk should wait, in milliseconds, before sending notifications.</synopsis>
<description>
<para>When a resource's state changes, it may be desired to wait a certain amount before Asterisk
sends a notification to subscribers. This allows for other state changes to accumulate, so that
Asterisk can communicate multiple state changes in a single notification instead of rapidly sending
many notifications.</para>
</description>
</configOption>
<configOption name="resource_display_name" default="no">
<synopsis>Indicates whether display name of resource or the resource name being reported.</synopsis>
<description>
<para>If this option is enabled, the Display Name will be reported as resource name.
If the <replaceable>event</replaceable> set to <literal>presence</literal> or <literal>dialog</literal>,
the non-empty HINT name will be set as the Display Name.
The <literal>message-summary</literal> is not supported yet.</para>
</description>
</configOption>
</configObject>
<configObject name="inbound-publication">
<synopsis>The configuration for inbound publications</synopsis>
<configOption name="endpoint" default="">
<synopsis>Optional name of an endpoint that is only allowed to publish to this resource</synopsis>
</configOption>
<configOption name="type">
<synopsis>Must be of type 'inbound-publication'.</synopsis>
</configOption>
</configObject>
</configFile>
</configInfo>
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
***/
static pj_bool_t pubsub_on_rx_request(pjsip_rx_data *rdata);
static struct pjsip_module pubsub_module = {
.name = { "PubSub Module", 13 },
.priority = PJSIP_MOD_PRIORITY_APPLICATION,
.on_rx_request = pubsub_on_rx_request,
};
#define MOD_DATA_PERSISTENCE "sub_persistence"
#define MOD_DATA_MSG "sub_msg"
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
static const pj_str_t str_event_name = { "Event", 5 };
/*! \brief Scheduler used for automatically expiring publications */
static struct ast_sched_context *sched;
/*! \brief Number of buckets for publications (on a per handler) */
#define PUBLICATIONS_BUCKETS 37
/*! \brief Default expiration time for PUBLISH if one is not specified */
#define DEFAULT_PUBLISH_EXPIRES 3600
/*! \brief Number of buckets for subscription datastore */
#define DATASTORE_BUCKETS 53
/*! \brief Default expiration for subscriptions */
#define DEFAULT_EXPIRES 3600
/*! \brief Defined method for PUBLISH */
const pjsip_method pjsip_publish_method =
{
PJSIP_OTHER_METHOD,
{ "PUBLISH", 7 }
};
/*!
* \brief The types of PUBLISH messages defined in RFC 3903
*/
enum sip_publish_type {
/*!
* \brief Unknown
*
* \details
* This actually is not defined in RFC 3903. We use this as a constant
* to indicate that an incoming PUBLISH does not fit into any of the
* other categories and is thus invalid.
*/
SIP_PUBLISH_UNKNOWN,
/*!
* \brief Initial
*
* \details
* The first PUBLISH sent. This will contain a non-zero Expires header
* as well as a body that indicates the current state of the endpoint
* that has sent the message. The initial PUBLISH is the only type
* of PUBLISH to not contain a Sip-If-Match header in it.
*/
SIP_PUBLISH_INITIAL,
/*!
* \brief Refresh
*
* \details
* Used to keep a published state from expiring. This will contain a
* non-zero Expires header but no body since its purpose is not to
* update state.
*/
SIP_PUBLISH_REFRESH,
/*!
* \brief Modify
*
* \details
* Used to change state from its previous value. This will contain
* a body updating the published state. May or may not contain an
* Expires header.
*/
SIP_PUBLISH_MODIFY,
/*!
* \brief Remove
*
* \details
* Used to remove published state from an ESC. This will contain
* an Expires header set to 0 and likely no body.
*/
SIP_PUBLISH_REMOVE,
};
/*!
* \brief A vector of strings commonly used throughout this module
*/
AST_VECTOR(resources, const char *);
/*!
* \brief Resource list configuration item
*/
struct resource_list {
SORCERY_OBJECT(details);
/*! SIP event package the list uses. */
char event[32];
/*! Strings representing resources in the list. */
struct resources items;
/*! Indicates if Asterisk sends full or partial state on notifications. */
unsigned int full_state;
/*! Time, in milliseconds Asterisk waits before sending a batched notification.*/
unsigned int notification_batch_interval;
/*! Indicates whether display name of resource or the resource name being reported.*/
unsigned int resource_display_name;
};
/*!
* Used to create new entity IDs by ESCs.
*/
static int esc_etag_counter;
/*!
* \brief Structure representing a SIP publication
*/
struct ast_sip_publication {
/*! Publication datastores set up by handlers */
struct ao2_container *datastores;
/*! \brief Entity tag for the publication */
int entity_tag;
/*! \brief Handler for this publication */
struct ast_sip_publish_handler *handler;
/*! \brief The endpoint with which the subscription is communicating */
struct ast_sip_endpoint *endpoint;
/*! \brief Expiration time of the publication */
unsigned int expires;
/*! \brief Scheduled item for expiration of publication */
int sched_id;
/*! \brief The resource the publication is to */
char *resource;
/*! \brief The name of the event type configuration */
char *event_configuration_name;
/*! \brief Data containing the above */
char data[0];
};
/*!
* \brief Structure used for persisting an inbound subscription
*/
struct subscription_persistence {
/*! Sorcery object details */
SORCERY_OBJECT(details);
/*! The name of the endpoint involved in the subscription */
char *endpoint;
/*! SIP message that creates the subscription */
char packet[PJSIP_MAX_PKT_LEN];
/*! Source address of the message */
char src_name[PJ_INET6_ADDRSTRLEN];
/*! Source port of the message */
int src_port;
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
/*! Local transport type (UDP,TCP,TLS)*/
char transport_type[32];
/*! Local transport address */
char local_name[PJ_INET6_ADDRSTRLEN];
/*! Local transport port */
int local_port;
/*! Next CSeq to use for message */
unsigned int cseq;
/*! Local tag of the dialog */
char *tag;
/*! When this subscription expires */
struct timeval expires;
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
/*! Contact URI */
char contact_uri[PJSIP_MAX_URL_SIZE];
/*! Prune subscription on restart */
int prune_on_boot;
/*! Body generator specific persistence data */
struct ast_json *generator_data;
};
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
/*!
* \brief The state of the subscription tree
*/
enum sip_subscription_tree_state {
/*! Normal operation */
SIP_SUB_TREE_NORMAL = 0,
/*! A terminate has been requested by Asterisk, the client, or pjproject */
SIP_SUB_TREE_TERMINATE_PENDING,
/*! The terminate is in progress */
SIP_SUB_TREE_TERMINATE_IN_PROGRESS,
/*! The terminate process has finished and the subscription tree is no longer valid */
SIP_SUB_TREE_TERMINATED,
};
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
static char *sub_tree_state_description[] = {
"Normal",
"TerminatePending",
"TerminateInProgress",
"Terminated"
};
/*!
* \brief A tree of SIP subscriptions
*
* Because of the ability to subscribe to resource lists, a SIP
* subscription can result in a tree of subscriptions being created.
* This structure represents the information relevant to the subscription
* as a whole, to include the underlying PJSIP structure for the
* subscription.
*/
struct sip_subscription_tree {
/*! The endpoint with which the subscription is communicating */
struct ast_sip_endpoint *endpoint;
/*! Serializer on which to place operations for this subscription */
struct ast_taskprocessor *serializer;
/*! The role for this subscription */
enum ast_sip_subscription_role role;
/*! Persistence information */
struct subscription_persistence *persistence;
/*! The underlying PJSIP event subscription structure */
pjsip_evsub *evsub;
/*! The underlying PJSIP dialog */
pjsip_dialog *dlg;
/*! Interval to use for batching notifications */
unsigned int notification_batch_interval;
/*! Scheduler ID for batched notification */
int notify_sched_id;
/*! Indicator if scheduled batched notification should be sent */
unsigned int send_scheduled_notify;
/*! The root of the subscription tree */
struct ast_sip_subscription *root;
/*! Is this subscription to a list? */
int is_list;
/*! Next item in the list */
AST_LIST_ENTRY(sip_subscription_tree) next;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
/*! Subscription tree state */
enum sip_subscription_tree_state state;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
/*! On asterisk restart, this is the task data used
* to restart the expiration timer if pjproject isn't
* capable of restarting the timer.
*/
struct ast_sip_sched_task *expiration_task;
/*! The transport the subscription was received on.
* Only used for reliable transports.
*/
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
char transport_key[IP6ADDR_COLON_PORT_BUFLEN];
/*! Indicator if initial notify should be generated.
* Used to refresh modified RLS.
*/
unsigned int generate_initial_notify;
};
/*!
* \brief Structure representing a "virtual" SIP subscription.
*
* This structure serves a dual purpose. Structurally, it is
* the constructed tree of subscriptions based on the resources
* being subscribed to. API-wise, this serves as the handle that
* subscription handlers use in order to interact with the pubsub API.
*/
struct ast_sip_subscription {
/*! Subscription datastores set up by handlers */
struct ao2_container *datastores;
/*! The handler for this subscription */
const struct ast_sip_subscription_handler *handler;
/*! Pointer to the base of the tree */
struct sip_subscription_tree *tree;
/*! Body generator for NOTIFYs */
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
struct ast_sip_pubsub_body_generator *body_generator;
/*! Vector of child subscriptions */
AST_VECTOR(, struct ast_sip_subscription *) children;
/*! Saved NOTIFY body text for this subscription */
struct ast_str *body_text;
/*! Indicator that the body text has changed since the last notification */
int body_changed;
/*! The current state of the subscription */
pjsip_evsub_state subscription_state;
/*! For lists, the current version to place in the RLMI body */
unsigned int version;
/*! For lists, indicates if full state should always be communicated. */
unsigned int full_state;
/*! URI associated with the subscription */
pjsip_sip_uri *uri;
/*! Data to be persisted with the subscription */
struct ast_json *persistence_data;
/*! Display Name of resource */
char *display_name;
/*! Name of resource being subscribed to */
char resource[0];
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
};
/*!
* \brief Structure representing a publication resource
*/
struct ast_sip_publication_resource {
/*! \brief Sorcery object details */
SORCERY_OBJECT(details);
/*! \brief Optional name of an endpoint that is only allowed to publish to this resource */
char *endpoint;
/*! \brief Mapping for event types to configuration */
struct ast_variable *events;
};
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
static const char *sip_subscription_roles_map[] = {
[AST_SIP_SUBSCRIBER] = "Subscriber",
[AST_SIP_NOTIFIER] = "Notifier"
};
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
enum sip_persistence_update_type {
/*! Called from send request */
SUBSCRIPTION_PERSISTENCE_SEND_REQUEST = 0,
/*! Subscription created from initial client request */
SUBSCRIPTION_PERSISTENCE_CREATED,
/*! Subscription recreated by asterisk on startup */
SUBSCRIPTION_PERSISTENCE_RECREATED,
/*! Subscription created from client refresh */
SUBSCRIPTION_PERSISTENCE_REFRESHED,
};
AST_RWLIST_HEAD_STATIC(subscriptions, sip_subscription_tree);
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
AST_RWLIST_HEAD_STATIC(body_generators, ast_sip_pubsub_body_generator);
AST_RWLIST_HEAD_STATIC(body_supplements, ast_sip_pubsub_body_supplement);
static pjsip_media_type rlmi_media_type;
static void pubsub_on_evsub_state(pjsip_evsub *sub, pjsip_event *event);
static void pubsub_on_rx_refresh(pjsip_evsub *sub, pjsip_rx_data *rdata,
int *p_st_code, pj_str_t **p_st_text, pjsip_hdr *res_hdr, pjsip_msg_body **p_body);
static void pubsub_on_rx_notify(pjsip_evsub *sub, pjsip_rx_data *rdata, int *p_st_code,
pj_str_t **p_st_text, pjsip_hdr *res_hdr, pjsip_msg_body **p_body);
static void pubsub_on_client_refresh(pjsip_evsub *sub);
static void pubsub_on_server_timeout(pjsip_evsub *sub);
static pjsip_evsub_user pubsub_cb = {
.on_evsub_state = pubsub_on_evsub_state,
.on_rx_refresh = pubsub_on_rx_refresh,
.on_rx_notify = pubsub_on_rx_notify,
.on_client_refresh = pubsub_on_client_refresh,
.on_server_timeout = pubsub_on_server_timeout,
};
/*! \brief Destructor for publication resource */
static void publication_resource_destroy(void *obj)
{
struct ast_sip_publication_resource *resource = obj;
ast_free(resource->endpoint);
ast_variables_destroy(resource->events);
}
/*! \brief Allocator for publication resource */
static void *publication_resource_alloc(const char *name)
{
return ast_sorcery_generic_alloc(sizeof(struct ast_sip_publication_resource), publication_resource_destroy);
}
static int sub_tree_subscription_terminate_cb(void *data)
{
struct sip_subscription_tree *sub_tree = data;
if (!sub_tree->evsub) {
/* Something else already terminated the subscription. */
ao2_ref(sub_tree, -1);
return 0;
}
ast_debug(3, "Transport destroyed. Removing subscription '%s->%s' prune on boot: %d\n",
sub_tree->persistence->endpoint, sub_tree->root->resource,
sub_tree->persistence->prune_on_boot);
sub_tree->state = SIP_SUB_TREE_TERMINATE_IN_PROGRESS;
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
ao2_ref(sub_tree, -1);
return 0;
}
/*!
* \internal
* \brief The reliable transport we used as a subscription contact has shutdown.
*
* \param data What subscription needs to be terminated.
*
* \note Normally executed by the pjsip monitor thread.
*/
static void sub_tree_transport_cb(void *data)
{
struct sip_subscription_tree *sub_tree = data;
/*
* Push off the subscription termination to the serializer to
* avoid deadlock. Another thread could be trying to send a
* message on the subscription that can deadlock with this
* thread.
*/
ao2_ref(sub_tree, +1);
if (ast_sip_push_task(sub_tree->serializer, sub_tree_subscription_terminate_cb,
sub_tree)) {
ao2_ref(sub_tree, -1);
}
}
/*! \brief Destructor for subscription persistence */
static void subscription_persistence_destroy(void *obj)
{
struct subscription_persistence *persistence = obj;
ast_free(persistence->endpoint);
ast_free(persistence->tag);
ast_json_unref(persistence->generator_data);
}
/*! \brief Allocator for subscription persistence */
static void *subscription_persistence_alloc(const char *name)
{
return ast_sorcery_generic_alloc(sizeof(struct subscription_persistence), subscription_persistence_destroy);
}
/*! \brief Function which creates initial persistence information of a subscription in sorcery */
static struct subscription_persistence *subscription_persistence_create(struct sip_subscription_tree *sub_tree)
{
char tag[PJ_GUID_STRING_LENGTH + 1];
/* The id of this persistence object doesn't matter as we keep it on the subscription and don't need to
* look it up by id at all.
*/
struct subscription_persistence *persistence = ast_sorcery_alloc(ast_sip_get_sorcery(),
"subscription_persistence", NULL);
pjsip_dialog *dlg = sub_tree->dlg;
if (!persistence) {
return NULL;
}
persistence->endpoint = ast_strdup(ast_sorcery_object_get_id(sub_tree->endpoint));
ast_copy_pj_str(tag, &dlg->local.info->tag, sizeof(tag));
persistence->tag = ast_strdup(tag);
ast_sorcery_create(ast_sip_get_sorcery(), persistence);
return persistence;
}
/*! \brief Function which updates persistence information of a subscription in sorcery */
static void subscription_persistence_update(struct sip_subscription_tree *sub_tree,
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
pjsip_rx_data *rdata, enum sip_persistence_update_type type)
{
pjsip_dialog *dlg;
if (!sub_tree->persistence) {
return;
}
ast_debug(3, "Updating persistence for '%s->%s' prune on boot: %s\n",
sub_tree->persistence->endpoint, sub_tree->root->resource,
sub_tree->persistence->prune_on_boot ? "yes" : "no");
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
dlg = sub_tree->dlg;
sub_tree->persistence->cseq = dlg->local.cseq;
if (rdata) {
unsigned int expires;
pjsip_expires_hdr *expires_hdr = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_EXPIRES, NULL);
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
pjsip_contact_hdr *contact_hdr = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_CONTACT, NULL);
expires = expires_hdr ? expires_hdr->ivalue : DEFAULT_PUBLISH_EXPIRES;
sub_tree->persistence->expires = ast_tvadd(ast_tvnow(), ast_samp2tv(expires, 1));
if (contact_hdr) {
if (contact_hdr) {
if (type == SUBSCRIPTION_PERSISTENCE_CREATED) {
sub_tree->persistence->prune_on_boot =
!ast_sip_will_uri_survive_restart(
(pjsip_sip_uri *)pjsip_uri_get_uri(contact_hdr->uri),
sub_tree->endpoint, rdata);
if (sub_tree->persistence->prune_on_boot) {
ast_debug(3, "adding transport monitor on %s for '%s->%s' prune on boot: %d\n",
rdata->tp_info.transport->obj_name,
sub_tree->persistence->endpoint, sub_tree->root->resource,
sub_tree->persistence->prune_on_boot);
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR(rdata->tp_info.transport,
sub_tree->transport_key);
ast_sip_transport_monitor_register_key(sub_tree->transport_key,
sub_tree_transport_cb, sub_tree);
/*
* FYI: ast_sip_transport_monitor_register holds a reference to the sub_tree
*/
}
}
}
pjsip_uri_print(PJSIP_URI_IN_CONTACT_HDR, contact_hdr->uri,
sub_tree->persistence->contact_uri, sizeof(sub_tree->persistence->contact_uri));
} else {
ast_log(LOG_WARNING, "Contact not updated due to missing contact header\n");
}
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
/* When receiving a packet on an streaming transport, it's possible to receive more than one SIP
* message at a time into the rdata->pkt_info.packet buffer. However, the rdata->msg_info.msg_buf
* will always point to the proper SIP message that is to be processed. When updating subscription
* persistence that is pulled from persistent storage, though, the rdata->pkt_info.packet will
* only ever have a single SIP message on it, and so we base persistence on that.
*/
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
if (type == SUBSCRIPTION_PERSISTENCE_CREATED
|| type == SUBSCRIPTION_PERSISTENCE_RECREATED) {
if (rdata->msg_info.msg_buf) {
ast_copy_string(sub_tree->persistence->packet, rdata->msg_info.msg_buf,
MIN(sizeof(sub_tree->persistence->packet), rdata->msg_info.len + 1));
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
} else {
ast_copy_string(sub_tree->persistence->packet, rdata->pkt_info.packet,
sizeof(sub_tree->persistence->packet));
}
}
ast_copy_string(sub_tree->persistence->src_name, rdata->pkt_info.src_name,
sizeof(sub_tree->persistence->src_name));
sub_tree->persistence->src_port = rdata->pkt_info.src_port;
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
ast_copy_string(sub_tree->persistence->transport_type, rdata->tp_info.transport->type_name,
sizeof(sub_tree->persistence->transport_type));
ast_copy_pj_str(sub_tree->persistence->local_name, &rdata->tp_info.transport->local_name.host,
sizeof(sub_tree->persistence->local_name));
sub_tree->persistence->local_port = rdata->tp_info.transport->local_name.port;
}
ast_sorcery_update(ast_sip_get_sorcery(), sub_tree->persistence);
}
/*! \brief Function which removes persistence of a subscription from sorcery */
static void subscription_persistence_remove(struct sip_subscription_tree *sub_tree)
{
if (!sub_tree->persistence) {
return;
}
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
if (sub_tree->persistence->prune_on_boot && !ast_strlen_zero(sub_tree->transport_key)) {
ast_debug(3, "Unregistering transport monitor on %s '%s->%s'\n",
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
sub_tree->transport_key,
sub_tree->endpoint ? ast_sorcery_object_get_id(sub_tree->endpoint) : "Unknown",
sub_tree->root ? sub_tree->root->resource : "Unknown");
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
ast_sip_transport_monitor_unregister_key(sub_tree->transport_key,
sub_tree_transport_cb, sub_tree, NULL);
}
ast_sorcery_delete(ast_sip_get_sorcery(), sub_tree->persistence);
ao2_ref(sub_tree->persistence, -1);
sub_tree->persistence = NULL;
}
static struct ast_sip_subscription_handler *find_sub_handler_for_event_name(const char *event_name);
static struct ast_sip_pubsub_body_generator *find_body_generator(char accept[AST_SIP_MAX_ACCEPT][64],
size_t num_accept, const char *body_type);
/*! \brief Retrieve a handler using the Event header of an rdata message */
static struct ast_sip_subscription_handler *subscription_get_handler_from_rdata(pjsip_rx_data *rdata, const char *endpoint)
{
pjsip_event_hdr *event_header;
char event[32];
struct ast_sip_subscription_handler *handler;
event_header = pjsip_msg_find_hdr_by_name(rdata->msg_info.msg, &str_event_name, rdata->msg_info.msg->hdr.next);
if (!event_header) {
ast_log(LOG_WARNING, "Incoming SUBSCRIBE request from %s with no Event header\n",
endpoint ? endpoint : "Unknown");
return NULL;
}
ast_copy_pj_str(event, &event_header->event_type, sizeof(event));
handler = find_sub_handler_for_event_name(event);
if (!handler) {
ast_log(LOG_WARNING, "No registered subscribe handler for event %s from %s\n", event,
endpoint ? endpoint : "Unknown");
}
return handler;
}
/*!
* \brief Accept headers that are exceptions to the rule
*
* Typically, when a SUBSCRIBE arrives, we attempt to find a
* body generator that matches one of the Accept headers in
* the request. When subscribing to a single resource, this works
* great. However, when subscribing to a list, things work
* differently. Most Accept header values are fine, but there
* are a couple that are endemic to resource lists that need
* to be ignored when searching for a body generator to use
* for the individual resources of the subscription.
*/
const char *accept_exceptions[] = {
"multipart/related",
"application/rlmi+xml",
};
/*!
* \brief Is the Accept header from the SUBSCRIBE in the list of exceptions?
*
* \retval 1 This Accept header value is an exception to the rule.
* \retval 0 This Accept header is not an exception to the rule.
*/
static int exceptional_accept(const pj_str_t *accept)
{
int i;
for (i = 0; i < ARRAY_LEN(accept_exceptions); ++i) {
if (!pj_strcmp2(accept, accept_exceptions[i])) {
return 1;
}
}
return 0;
}
/*! \brief Retrieve a body generator using the Accept header of an rdata message */
static struct ast_sip_pubsub_body_generator *subscription_get_generator_from_rdata(pjsip_rx_data *rdata,
const struct ast_sip_subscription_handler *handler)
{
pjsip_accept_hdr *accept_header = (pjsip_accept_hdr *) &rdata->msg_info.msg->hdr;
char accept[AST_SIP_MAX_ACCEPT][64];
size_t num_accept_headers = 0;
while ((accept_header = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_ACCEPT, accept_header->next)) &&
(num_accept_headers < AST_SIP_MAX_ACCEPT)) {
int i;
for (i = 0; i < accept_header->count && num_accept_headers < AST_SIP_MAX_ACCEPT; ++i) {
if (!exceptional_accept(&accept_header->values[i])) {
ast_copy_pj_str(accept[num_accept_headers], &accept_header->values[i], sizeof(accept[num_accept_headers]));
++num_accept_headers;
}
}
}
if (num_accept_headers == 0) {
/* If a SUBSCRIBE contains no Accept headers, then we must assume that
* the default accept type for the event package is to be used.
*/
ast_copy_string(accept[0], handler->notifier->default_accept, sizeof(accept[0]));
num_accept_headers = 1;
}
return find_body_generator(accept, num_accept_headers, handler->body_type);
}
/*! \brief Check if the rdata has a Supported header containing 'eventlist'
*
* \retval 1 rdata has an eventlist containing supported header
* \retval 0 rdata doesn't have an eventlist containing supported header
*/
static int ast_sip_pubsub_has_eventlist_support(pjsip_rx_data *rdata)
{
pjsip_supported_hdr *supported_header = (pjsip_supported_hdr *) &rdata->msg_info.msg->hdr;
while ((supported_header = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_SUPPORTED, supported_header->next))) {
int i;
for (i = 0; i < supported_header->count; i++) {
if (!pj_stricmp2(&supported_header->values[i], "eventlist")) {
return 1;
}
}
}
return 0;
}
struct resource_tree;
/*!
* \brief A node for a resource tree.
*/
struct tree_node {
AST_VECTOR(, struct tree_node *) children;
unsigned int full_state;
char *display_name;
char resource[0];
};
/*!
* \brief Helper function for retrieving a resource list for a given event.
*
* This will retrieve a resource list that corresponds to the resource and event provided.
*
* \param resource The name of the resource list to retrieve
* \param event The expected event name on the resource list
*/
static struct resource_list *retrieve_resource_list(const char *resource, const char *event)
{
struct resource_list *list;
list = ast_sorcery_retrieve_by_id(ast_sip_get_sorcery(), "resource_list", resource);
if (!list) {
return NULL;
}
if (strcmp(list->event, event)) {
ast_log(LOG_WARNING, "Found resource list %s, but its event type (%s) does not match SUBSCRIBE's (%s)\n",
resource, list->event, event);
ao2_cleanup(list);
return NULL;
}
return list;
}
/*!
* \brief Allocate a tree node
*
* In addition to allocating and initializing the tree node, the node is also added
* to the vector of visited resources. See \ref build_resource_tree for more information
* on the visited resources.
*
* \param resource The name of the resource for this tree node.
* \param visited The vector of resources that have been visited.
* \param full_state if allocating a list, indicate whether full state is requested in notifications.
* \param display_name the display name to include with this tree node.
*
* \retval NULL Allocation failure.
* \retval non-NULL The newly-allocated tree_node
*/
static struct tree_node *tree_node_alloc(const char *resource, struct resources *visited, unsigned int full_state, const char *display_name)
{
struct tree_node *node;
node = ast_calloc(1, sizeof(*node) + strlen(resource) + 1);
if (!node) {
return NULL;
}
strcpy(node->resource, resource);
if (AST_VECTOR_INIT(&node->children, 4)) {
ast_free(node);
return NULL;
}
node->full_state = full_state;
node->display_name = ast_strdup(display_name);
if (visited) {
AST_VECTOR_APPEND(visited, resource);
}
return node;
}
/*!
* \brief Destructor for a tree node
*
* This function calls recursively in order to destroy
* all nodes lower in the tree from the given node in
* addition to the node itself.
*
* \param node The node to destroy.
*/
static void tree_node_destroy(struct tree_node *node)
{
int i;
if (!node) {
return;
}
for (i = 0; i < AST_VECTOR_SIZE(&node->children); ++i) {
tree_node_destroy(AST_VECTOR_GET(&node->children, i));
}
AST_VECTOR_FREE(&node->children);
ast_free(node->display_name);
ast_free(node);
}
/*!
* \brief Determine if this resource has been visited already
*
* See \ref build_resource_tree for more information
*
* \param resource The resource currently being visited
* \param visited The resources that have previously been visited
*/
static int have_visited(const char *resource, struct resources *visited)
{
int i;
for (i = 0; i < AST_VECTOR_SIZE(visited); ++i) {
if (!strcmp(resource, AST_VECTOR_GET(visited, i))) {
return 1;
}
}
return 0;
}
/*!
* \brief Build child nodes for a given parent.
*
* This iterates through the items on a resource list and creates tree nodes for each one. The
* tree nodes created are children of the supplied parent node. If an item in the resource
* list is itself a list, then this function is called recursively to provide children for
* the new node.
*
* If an item in a resource list is not a list, then the supplied subscription handler is
* called into as if a new SUBSCRIBE for the list item were presented. The handler's response
* is used to determine if the node can be added to the tree or not.
*
* If a parent node ends up having no child nodes added under it, then the parent node is
* pruned from the tree.
*
* \param endpoint The endpoint that sent the inbound SUBSCRIBE.
* \param handler The subscription handler for leaf nodes in the tree.
* \param list The configured resource list from which the child node is being built.
* \param parent The parent node for these children.
* \param visited The resources that have already been visited.
*/
static void build_node_children(struct ast_sip_endpoint *endpoint, const struct ast_sip_subscription_handler *handler,
struct resource_list *list, struct tree_node *parent, struct resources *visited)
{
int i;
for (i = 0; i < AST_VECTOR_SIZE(&list->items); ++i) {
struct tree_node *current;
struct resource_list *child_list;
const char *resource = AST_VECTOR_GET(&list->items, i);
if (have_visited(resource, visited)) {
ast_debug(1, "Already visited resource %s. Avoiding duplicate resource or potential loop.\n", resource);
continue;
}
child_list = retrieve_resource_list(resource, list->event);
if (!child_list) {
int resp = handler->notifier->new_subscribe(endpoint, resource);
if (PJSIP_IS_STATUS_IN_CLASS(resp, 200)) {
char display_name[AST_MAX_EXTENSION] = "";
if (list->resource_display_name && handler->notifier->get_resource_display_name) {
handler->notifier->get_resource_display_name(endpoint, resource, display_name, sizeof(display_name));
}
current = tree_node_alloc(resource, visited, 0, ast_strlen_zero(display_name) ? NULL : display_name);
if (!current) {
ast_debug(1,
"Subscription to leaf resource %s was successful, but encountered allocation error afterwards\n",
resource);
continue;
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_debug(2, "Subscription to leaf resource %s resulted in success. Adding to parent %s\n",
resource, parent->resource);
if (AST_VECTOR_APPEND(&parent->children, current)) {
tree_node_destroy(current);
}
} else {
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_debug(2, "Subscription to leaf resource %s resulted in error response %d\n",
resource, resp);
}
} else {
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_debug(2, "Resource %s (child of %s) is a list\n", resource, parent->resource);
current = tree_node_alloc(resource, visited, child_list->full_state, NULL);
if (!current) {
ast_debug(1, "Cannot build children of resource %s due to allocation failure\n", resource);
continue;
}
build_node_children(endpoint, handler, child_list, current, visited);
if (AST_VECTOR_SIZE(&current->children) > 0) {
ast_debug(1, "List %s had no successful children.\n", resource);
if (AST_VECTOR_APPEND(&parent->children, current)) {
tree_node_destroy(current);
}
} else {
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_debug(2, "List %s had successful children. Adding to parent %s\n",
resource, parent->resource);
tree_node_destroy(current);
}
ao2_cleanup(child_list);
}
}
}
/*!
* \brief A resource tree
*
* When an inbound SUBSCRIBE arrives, the resource being subscribed to may
* be a resource list. If this is the case, the resource list may contain resources
* that are themselves lists. The structure needed to hold the resources is
* a tree.
*
* Upon receipt of the SUBSCRIBE, the tree is built by determining if subscriptions
* to the individual resources in the tree would be successful or not. Any successful
* subscriptions result in a node in the tree being created. Any unsuccessful subscriptions
* result in no node being created.
*
* This tree can be seen as a bare-bones analog of the tree of ast_sip_subscriptions that
* will end up being created to actually carry out the duties of a SIP SUBSCRIBE dialog.
*/
struct resource_tree {
struct tree_node *root;
unsigned int notification_batch_interval;
};
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
/*!
* \brief Destroy a resource tree.
*
* This function makes no assumptions about how the tree itself was
* allocated and does not attempt to free the tree itself. Callers
* of this function are responsible for freeing the tree.
*
* \param tree The tree to destroy.
*/
static void resource_tree_destroy(struct resource_tree *tree)
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
{
if (tree) {
tree_node_destroy(tree->root);
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
/*!
* \brief Build a resource tree
*
* This function builds a resource tree based on the requested resource in a SUBSCRIBE request.
*
* This function also creates a container that has all resources that have been visited during
* creation of the tree, whether those resources resulted in a tree node being created or not.
* Keeping this container of visited resources allows for misconfigurations such as loops in
* the tree or duplicated resources to be detected.
*
* \param endpoint The endpoint that sent the SUBSCRIBE request.
* \param handler The subscription handler for leaf nodes in the tree.
* \param resource The resource requested in the SUBSCRIBE request.
* \param tree The tree that is to be built.
* \param has_eventlist_support
*
* \retval 200-299 Successfully subscribed to at least one resource.
* \retval 300-699 Failure to subscribe to requested resource.
*/
static int build_resource_tree(struct ast_sip_endpoint *endpoint, const struct ast_sip_subscription_handler *handler,
const char *resource, struct resource_tree *tree, int has_eventlist_support)
{
RAII_VAR(struct resource_list *, list, NULL, ao2_cleanup);
struct resources visited;
if (!has_eventlist_support || !(list = retrieve_resource_list(resource, handler->event_name))) {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(2, "Subscription '%s->%s' is not to a list\n",
ast_sorcery_object_get_id(endpoint), resource);
tree->root = tree_node_alloc(resource, NULL, 0, NULL);
if (!tree->root) {
return 500;
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
return handler->notifier->new_subscribe(endpoint, resource);
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(2, "Subscription '%s->%s' is a list\n",
ast_sorcery_object_get_id(endpoint), resource);
if (AST_VECTOR_INIT(&visited, AST_VECTOR_SIZE(&list->items))) {
return 500;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
tree->root = tree_node_alloc(resource, &visited, list->full_state, NULL);
if (!tree->root) {
AST_VECTOR_FREE(&visited);
return 500;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
tree->notification_batch_interval = list->notification_batch_interval;
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
build_node_children(endpoint, handler, list, tree->root, &visited);
AST_VECTOR_FREE(&visited);
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
if (AST_VECTOR_SIZE(&tree->root->children) > 0) {
return 200;
} else {
return 500;
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
}
static void add_subscription(struct sip_subscription_tree *obj)
{
AST_RWLIST_WRLOCK(&subscriptions);
AST_RWLIST_INSERT_TAIL(&subscriptions, obj, next);
AST_RWLIST_UNLOCK(&subscriptions);
}
static void remove_subscription(struct sip_subscription_tree *obj)
{
struct sip_subscription_tree *i;
AST_RWLIST_WRLOCK(&subscriptions);
AST_RWLIST_TRAVERSE_SAFE_BEGIN(&subscriptions, i, next) {
if (i == obj) {
AST_RWLIST_REMOVE_CURRENT(next);
if (i->root) {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(2, "Removing subscription '%s->%s' from list of subscriptions\n",
ast_sorcery_object_get_id(i->endpoint), ast_sip_subscription_get_resource_name(i->root));
}
break;
}
}
AST_RWLIST_TRAVERSE_SAFE_END;
AST_RWLIST_UNLOCK(&subscriptions);
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
static void destroy_subscription(struct ast_sip_subscription *sub)
{
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "Destroying SIP subscription from '%s->%s'\n",
sub->tree && sub->tree->endpoint ? ast_sorcery_object_get_id(sub->tree->endpoint) : "Unknown",
sub->resource);
ast_free(sub->body_text);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
AST_VECTOR_FREE(&sub->children);
ao2_cleanup(sub->datastores);
ast_json_unref(sub->persistence_data);
ast_free(sub->display_name);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
ast_free(sub);
}
static void destroy_subscriptions(struct ast_sip_subscription *root)
{
int i;
if (!root) {
return;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
for (i = 0; i < AST_VECTOR_SIZE(&root->children); ++i) {
struct ast_sip_subscription *child;
child = AST_VECTOR_GET(&root->children, i);
destroy_subscriptions(child);
}
destroy_subscription(root);
}
static struct ast_sip_subscription *allocate_subscription(const struct ast_sip_subscription_handler *handler,
const char *resource, const char *display_name, struct sip_subscription_tree *tree)
{
struct ast_sip_subscription *sub;
pjsip_msg *msg;
pjsip_sip_uri *request_uri;
msg = ast_sip_mod_data_get(tree->dlg->mod_data, pubsub_module.id, MOD_DATA_MSG);
if (!msg) {
ast_log(LOG_ERROR, "No dialog message saved for SIP subscription. Cannot allocate subscription for resource %s\n", resource);
return NULL;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
sub = ast_calloc(1, sizeof(*sub) + strlen(resource) + 1);
if (!sub) {
return NULL;
}
strcpy(sub->resource, resource); /* Safe */
sub->display_name = ast_strdup(display_name);
sub->datastores = ast_datastores_alloc();
if (!sub->datastores) {
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
destroy_subscription(sub);
return NULL;
}
sub->body_text = ast_str_create(128);
if (!sub->body_text) {
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
destroy_subscription(sub);
return NULL;
}
sub->uri = pjsip_sip_uri_create(tree->dlg->pool, PJ_FALSE);
request_uri = pjsip_uri_get_uri(msg->line.req.uri);
pjsip_sip_uri_assign(tree->dlg->pool, sub->uri, request_uri);
pj_strdup2(tree->dlg->pool, &sub->uri->user, resource);
/* If there is any persistence information available for this subscription that was persisted
* then make it available so that the NOTIFY has the correct state.
*/
if (tree->persistence && tree->persistence->generator_data) {
sub->persistence_data = ast_json_ref(ast_json_object_get(tree->persistence->generator_data, resource));
}
sub->handler = handler;
sub->subscription_state = PJSIP_EVSUB_STATE_ACTIVE;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
sub->tree = ao2_bump(tree);
return sub;
}
/*!
* \brief Create a tree of virtual subscriptions based on a resource tree node.
*
* \param handler The handler to supply to leaf subscriptions.
* \param resource The requested resource for this subscription.
* \param generator Body generator to use for leaf subscriptions.
* \param tree The root of the subscription tree.
* \param current The tree node that corresponds to the subscription being created.
*/
static struct ast_sip_subscription *create_virtual_subscriptions(const struct ast_sip_subscription_handler *handler,
const char *resource, struct ast_sip_pubsub_body_generator *generator,
struct sip_subscription_tree *tree, struct tree_node *current)
{
int i;
struct ast_sip_subscription *sub;
sub = allocate_subscription(handler, resource, current->display_name, tree);
if (!sub) {
return NULL;
}
sub->full_state = current->full_state;
sub->body_generator = generator;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
AST_VECTOR_INIT(&sub->children, AST_VECTOR_SIZE(&current->children));
for (i = 0; i < AST_VECTOR_SIZE(&current->children); ++i) {
struct ast_sip_subscription *child;
struct tree_node *child_node = AST_VECTOR_GET(&current->children, i);
child = create_virtual_subscriptions(handler, child_node->resource, generator,
tree, child_node);
if (!child) {
ast_debug(1, "Child subscription to resource %s could not be created\n",
child_node->resource);
continue;
}
if (AST_VECTOR_APPEND(&sub->children, child)) {
ast_debug(1, "Child subscription to resource %s could not be appended\n",
child_node->resource);
destroy_subscription(child);
/* Have to release tree here too because a ref was added
* to child that destroy_subscription() doesn't release. */
ao2_cleanup(tree);
}
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
return sub;
}
static void shutdown_subscriptions(struct ast_sip_subscription *sub)
{
int i;
if (!sub) {
return;
}
if (AST_VECTOR_SIZE(&sub->children) > 0) {
for (i = 0; i < AST_VECTOR_SIZE(&sub->children); ++i) {
shutdown_subscriptions(AST_VECTOR_GET(&sub->children, i));
}
return;
}
/* We notify subscription shutdown only on the tree leaves. */
if (sub->handler->subscription_shutdown) {
sub->handler->subscription_shutdown(sub);
}
}
2015-10-02 20:32:09 +00:00
static int subscription_unreference_dialog(void *obj)
{
struct sip_subscription_tree *sub_tree = obj;
/* This is why we keep the dialog on the subscription. When the subscription
* is destroyed, there is no guarantee that the underlying dialog is ready
* to be destroyed. Furthermore, there's no guarantee in the opposite direction
* either. The dialog could be destroyed before our subscription is. We fix
* this problem by keeping a reference to the dialog until it is time to
* destroy the subscription. We need to have the dialog available when the
* subscription is destroyed so that we can guarantee that our attempt to
* remove the serializer will be successful.
*/
pjsip_dlg_dec_session(sub_tree->dlg, &pubsub_module);
sub_tree->dlg = NULL;
return 0;
}
static void subscription_tree_destructor(void *obj)
{
struct sip_subscription_tree *sub_tree = obj;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "Destroying subscription tree %p '%s->%s'\n",
sub_tree,
sub_tree->endpoint ? ast_sorcery_object_get_id(sub_tree->endpoint) : "Unknown",
sub_tree->root ? sub_tree->root->resource : "Unknown");
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
destroy_subscriptions(sub_tree->root);
if (sub_tree->dlg) {
res_pjsip.c: Split ast_sip_push_task_synchronous() to fit expectations. ast_sip_push_task_synchronous() did not necessarily execute the passed in task under the specified serializer. If the current thread is any registered pjsip thread then it would execute the task immediately instead of under the specified serializer. Reentrancy issues could result if the task does not execute with the right serializer. The original reason ast_sip_push_task_synchronous() checked to see if the current thread was a registered pjsip thread was because of a deadlock with masquerades and the channel technology's fixup callback (ASTERISK_22936). A subsequent masquerade deadlock fix (ASTERISK_24356) involving call pickups avoided the original deadlock situation entirely. The PJSIP channel technology's fixup callback no longer needed to call ast_sip_push_task_synchronous(). However, there are a few places where this unexpected behavior is still required to avoid deadlocks. The pjsip monitor thread executes callbacks that do calls to ast_sip_push_task_synchronous() that would deadlock if the task were actually pushed to the specified serializer. I ran into one dealing with the pubsub subscriptions where an ao2 destructor called ast_sip_push_task_synchronous(). * Split ast_sip_push_task_synchronous() into ast_sip_push_task_wait_servant() and ast_sip_push_task_wait_serializer(). ast_sip_push_task_wait_servant() has the old behavior of ast_sip_push_task_synchronous(). ast_sip_push_task_wait_serializer() has the new behavior where the task is always executed by the specified serializer or a picked serializer if one is not passed in. Both functions behave the same if the current thread is not a SIP servant. * Redirected ast_sip_push_task_synchronous() to ast_sip_push_task_wait_servant() to preserve API for released branches. ASTERISK_26806 Change-Id: Id040fa42c0e5972f4c8deef380921461d213b9f3
2018-03-27 16:04:42 +00:00
ast_sip_push_task_wait_servant(sub_tree->serializer,
subscription_unreference_dialog, sub_tree);
}
ao2_cleanup(sub_tree->endpoint);
ast_taskprocessor_unreference(sub_tree->serializer);
ast_module_unref(ast_module_info->self);
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
void ast_sip_subscription_destroy(struct ast_sip_subscription *sub)
{
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "Removing subscription %p '%s->%s' reference to subscription tree %p\n",
sub, ast_sorcery_object_get_id(sub->tree->endpoint), sub->resource, sub->tree);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
ao2_cleanup(sub->tree);
}
static void subscription_setup_dialog(struct sip_subscription_tree *sub_tree, pjsip_dialog *dlg)
{
sub_tree->dlg = dlg;
ast_sip_dialog_set_serializer(dlg, sub_tree->serializer);
res_pjsip_pubsub: Set the endpoint on SUBSCRIBE dialogs. When SUBSCRIBE dialogs were established, we never associated the endpoint that created the subscription with the dialog we end up creating. In most cases, this ended up not causing any problems. The actual bug that was observed was that when a device that was behind NAT established a subscription with Asterisk, Asterisk would end up sending in-dialog NOTIFY requests to the device's private IP addres instead of the public address of the NAT router. When Asterisk receives the initial SUBSCRIBE from the device, res_pjsip_nat rewrites the contact to the public address on which the SUBSCRIBE was received. This allows for the dialog to have its target address set to the proper public address. Asterisk then would send a 200 OK response to the SUBSCRIBE, then a NOTIFY with the initial subscription state. The device would then send a 200 OK response to Asterisk's NOTIFY. Here's where things went wrong. When the 200 OK arrived, res_pjsip_nat did not rewrite the address in the Contact header. Then, when the PJSIP dialog layer processed the 200 OK, PJSIP would perform a comparison between the IP address in the Contact header and its saved target address for the dialog. Since they differed, PJSIP would update the target dialog address to be the address in the Contact header. From this point, if Asterisk needed to send a NOTIFY to the device, the result was that the NOTIFY would be sent to the private address that the device placed in the Contact header. The reason why res_pjsip_nat did not rewrite the address when it received the 200 OK response was that it could not associate the incoming response with a configured endpoint. This is because on a response, the only way to associate the response to an endpoint is by finding the dialog that the response is associated with and then finding the endpoint that is associated with that dialog. We do not perform endpoint lookups on responses. res_pjsip_pubsub skipped the step of associating the endpoint with the dialog we created, so res_pjsip_nat could not find the associated endpoint and therefore couldn't rewrite the contact. This commit message is like 50x longer than the actual fix. ASTERISK 24981 #close Reported by Mark Michelson Change-Id: I2b963c58c063bae293e038406f7d044a8a5377cd
2015-04-20 19:30:47 +00:00
ast_sip_dialog_set_endpoint(dlg, sub_tree->endpoint);
pjsip_evsub_set_mod_data(sub_tree->evsub, pubsub_module.id, sub_tree);
2015-10-02 20:32:09 +00:00
pjsip_dlg_inc_session(dlg, &pubsub_module);
}
static struct sip_subscription_tree *allocate_subscription_tree(struct ast_sip_endpoint *endpoint, pjsip_rx_data *rdata)
{
struct sip_subscription_tree *sub_tree;
sub_tree = ao2_alloc(sizeof *sub_tree, subscription_tree_destructor);
if (!sub_tree) {
return NULL;
}
ast_module_ref(ast_module_info->self);
if (rdata) {
/*
* We must continue using the serializer that the original
* SUBSCRIBE came in on for the dialog. There may be
* retransmissions already enqueued in the original
* serializer that can result in reentrancy and message
* sequencing problems.
*/
sub_tree->serializer = ast_sip_get_distributor_serializer(rdata);
} else {
char tps_name[AST_TASKPROCESSOR_MAX_NAME + 1];
/* Create name with seq number appended. */
ast_taskprocessor_build_name(tps_name, sizeof(tps_name), "pjsip/pubsub/%s",
ast_sorcery_object_get_id(endpoint));
sub_tree->serializer = ast_sip_create_serializer(tps_name);
}
if (!sub_tree->serializer) {
ao2_ref(sub_tree, -1);
return NULL;
}
sub_tree->endpoint = ao2_bump(endpoint);
sub_tree->notify_sched_id = -1;
return sub_tree;
}
/*!
* \brief Create a subscription tree based on a resource tree.
*
* Using the previously-determined valid resources in the provided resource tree,
* a corresponding tree of ast_sip_subscriptions are created. The root of the
* subscription tree is a real subscription, and the rest in the tree are
* virtual subscriptions.
*
* \param handler The handler to use for leaf subscriptions
* \param endpoint The endpoint that sent the SUBSCRIBE request
* \param rdata The SUBSCRIBE content
* \param resource The requested resource in the SUBSCRIBE request
* \param generator The body generator to use in leaf subscriptions
* \param tree The resource tree on which the subscription tree is based
* \param[out] dlg_status The result of attempting to create a dialog
* \param persistence
*
* \retval NULL Could not create the subscription tree
* \retval non-NULL The root of the created subscription tree
*/
static struct sip_subscription_tree *create_subscription_tree(const struct ast_sip_subscription_handler *handler,
struct ast_sip_endpoint *endpoint, pjsip_rx_data *rdata, const char *resource,
struct ast_sip_pubsub_body_generator *generator, struct resource_tree *tree,
pj_status_t *dlg_status, struct subscription_persistence *persistence)
{
struct sip_subscription_tree *sub_tree;
pjsip_dialog *dlg;
sub_tree = allocate_subscription_tree(endpoint, rdata);
if (!sub_tree) {
*dlg_status = PJ_ENOMEM;
return NULL;
}
sub_tree->role = AST_SIP_NOTIFIER;
dlg = ast_sip_create_dialog_uas_locked(endpoint, rdata, dlg_status);
if (!dlg) {
if (*dlg_status != PJ_EEXISTS) {
ast_log(LOG_WARNING, "Unable to create dialog for SIP subscription\n");
}
ao2_ref(sub_tree, -1);
return NULL;
}
persistence = ast_sip_mod_data_get(rdata->endpt_info.mod_data,
pubsub_module.id, MOD_DATA_PERSISTENCE);
if (persistence) {
/* Update the created dialog with the persisted information */
pjsip_ua_unregister_dlg(pjsip_ua_instance(), dlg);
pj_strdup2(dlg->pool, &dlg->local.info->tag, persistence->tag);
dlg->local.tag_hval = pj_hash_calc_tolower(0, NULL, &dlg->local.info->tag);
pjsip_ua_register_dlg(pjsip_ua_instance(), dlg);
dlg->local.cseq = persistence->cseq;
}
pjsip_evsub_create_uas(dlg, &pubsub_cb, rdata, 0, &sub_tree->evsub);
subscription_setup_dialog(sub_tree, dlg);
/*
* The evsub and subscription setup both add dialog refs, so the dialog ref that
* was added when the dialog was created (see ast_sip_create_dialog_uas_lock) can
* now be removed. The lock should no longer be needed so can be removed too.
*/
pjsip_dlg_dec_lock(dlg);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
#ifdef HAVE_PJSIP_EVSUB_GRP_LOCK
pjsip_evsub_add_ref(sub_tree->evsub);
#endif
ast_sip_mod_data_set(dlg->pool, dlg->mod_data, pubsub_module.id, MOD_DATA_MSG,
pjsip_msg_clone(dlg->pool, rdata->msg_info.msg));
sub_tree->notification_batch_interval = tree->notification_batch_interval;
/* Persistence information needs to be available for all the subscriptions */
sub_tree->persistence = ao2_bump(persistence);
sub_tree->root = create_virtual_subscriptions(handler, resource, generator, sub_tree, tree->root);
if (AST_VECTOR_SIZE(&sub_tree->root->children) > 0) {
sub_tree->is_list = 1;
}
add_subscription(sub_tree);
return sub_tree;
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
/*! Wrapper structure for initial_notify_task */
struct initial_notify_data {
struct sip_subscription_tree *sub_tree;
unsigned int expires;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
};
static int initial_notify_task(void *obj);
static int send_notify(struct sip_subscription_tree *sub_tree, unsigned int force_full_state);
/*! Persistent subscription recreation continuation under distributor serializer data */
struct persistence_recreate_data {
struct subscription_persistence *persistence;
pjsip_rx_data *rdata;
};
/*!
* \internal
* \brief subscription_persistence_recreate continuation under distributor serializer.
* \since 13.10.0
*
* \retval 0 on success.
* \retval -1 on error.
*/
static int sub_persistence_recreate(void *obj)
{
struct persistence_recreate_data *recreate_data = obj;
struct subscription_persistence *persistence = recreate_data->persistence;
pjsip_rx_data *rdata = recreate_data->rdata;
struct ast_sip_endpoint *endpoint;
struct sip_subscription_tree *sub_tree;
struct ast_sip_pubsub_body_generator *generator;
struct ast_sip_subscription_handler *handler;
char *resource;
size_t resource_size;
int resp;
struct resource_tree tree;
pjsip_expires_hdr *expires_header;
int64_t expires;
const pj_str_t *user;
user = ast_sip_pjsip_uri_get_username(rdata->msg_info.msg->line.req.uri);
resource_size = pj_strlen(user) + 1;
resource = ast_alloca(resource_size);
ast_copy_pj_str(resource, user, resource_size);
/*
* We may want to match without any user options getting
* in the way.
*/
AST_SIP_USER_OPTIONS_TRUNCATE_CHECK(resource);
handler = subscription_get_handler_from_rdata(rdata, persistence->endpoint);
if (!handler || !handler->notifier) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Could not get subscription handler.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
generator = subscription_get_generator_from_rdata(rdata, handler);
if (!generator) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Body generator not available.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
ast_sip_mod_data_set(rdata->tp_info.pool, rdata->endpt_info.mod_data,
pubsub_module.id, MOD_DATA_PERSISTENCE, persistence);
/* Getting the endpoint may take some time that can affect the expiration. */
endpoint = ast_sorcery_retrieve_by_id(ast_sip_get_sorcery(), "endpoint",
persistence->endpoint);
if (!endpoint) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: The endpoint was not found\n",
persistence->endpoint);
2015-08-06 17:48:07 +00:00
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
/* Update the expiration header with the new expiration */
expires_header = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_EXPIRES,
rdata->msg_info.msg->hdr.next);
if (!expires_header) {
expires_header = pjsip_expires_hdr_create(rdata->tp_info.pool, 0);
if (!expires_header) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Could not update expires header.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
ao2_ref(endpoint, -1);
return 0;
}
pjsip_msg_add_hdr(rdata->msg_info.msg, (pjsip_hdr *) expires_header);
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
expires = (ast_tvdiff_ms(persistence->expires, ast_tvnow()) / 1000);
if (expires <= 0) {
/* The subscription expired since we started recreating the subscription. */
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "Expired subscription retrived from persistent store '%s' %s\n",
persistence->endpoint, persistence->tag);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
ao2_ref(endpoint, -1);
return 0;
}
expires_header->ivalue = expires;
memset(&tree, 0, sizeof(tree));
resp = build_resource_tree(endpoint, handler, resource, &tree,
ast_sip_pubsub_has_eventlist_support(rdata));
if (PJSIP_IS_STATUS_IN_CLASS(resp, 200)) {
pj_status_t dlg_status;
sub_tree = create_subscription_tree(handler, endpoint, rdata, resource, generator,
&tree, &dlg_status, persistence);
if (!sub_tree) {
if (dlg_status != PJ_EEXISTS) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Could not create subscription tree.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
}
} else {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
struct initial_notify_data *ind = ast_malloc(sizeof(*ind));
if (!ind) {
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
goto error;
}
ind->sub_tree = ao2_bump(sub_tree);
ind->expires = expires_header->ivalue;
subscription_persistence_update(sub_tree, rdata, SUBSCRIPTION_PERSISTENCE_RECREATED);
if (ast_sip_push_task(sub_tree->serializer, initial_notify_task, ind)) {
/* Could not send initial subscribe NOTIFY */
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
ao2_ref(sub_tree, -1);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_free(ind);
}
}
} else {
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
error:
resource_tree_destroy(&tree);
ao2_ref(endpoint, -1);
return 0;
}
/*! \brief Callback function to perform the actual recreation of a subscription */
static int subscription_persistence_recreate(void *obj, void *arg, int flags)
{
struct subscription_persistence *persistence = obj;
pj_pool_t *pool = arg;
struct ast_taskprocessor *serializer;
pjsip_rx_data rdata;
struct persistence_recreate_data recreate_data;
/* If this subscription used a reliable transport it can't be reestablished so remove it */
if (persistence->prune_on_boot) {
ast_debug(3, "Deleting subscription marked as 'prune' from persistent store '%s' %s\n",
persistence->endpoint, persistence->tag);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
/* If this subscription has already expired remove it */
if (ast_tvdiff_ms(persistence->expires, ast_tvnow()) <= 0) {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "Expired subscription retrived from persistent store '%s' %s\n",
persistence->endpoint, persistence->tag);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
memset(&rdata, 0, sizeof(rdata));
pj_pool_reset(pool);
rdata.tp_info.pool = pool;
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
if (ast_sip_create_rdata_with_contact(&rdata, persistence->packet, persistence->src_name,
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
persistence->src_port, persistence->transport_type, persistence->local_name,
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
persistence->local_port, persistence->contact_uri)) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: The message could not be parsed\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
if (rdata.msg_info.msg->type != PJSIP_REQUEST_MSG) {
ast_log(LOG_NOTICE, "Failed recreating '%s' subscription: Stored a SIP response instead of a request.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
/* Continue the remainder in the distributor serializer */
serializer = ast_sip_get_distributor_serializer(&rdata);
if (!serializer) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Could not get distributor serializer.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
return 0;
}
recreate_data.persistence = persistence;
recreate_data.rdata = &rdata;
res_pjsip.c: Split ast_sip_push_task_synchronous() to fit expectations. ast_sip_push_task_synchronous() did not necessarily execute the passed in task under the specified serializer. If the current thread is any registered pjsip thread then it would execute the task immediately instead of under the specified serializer. Reentrancy issues could result if the task does not execute with the right serializer. The original reason ast_sip_push_task_synchronous() checked to see if the current thread was a registered pjsip thread was because of a deadlock with masquerades and the channel technology's fixup callback (ASTERISK_22936). A subsequent masquerade deadlock fix (ASTERISK_24356) involving call pickups avoided the original deadlock situation entirely. The PJSIP channel technology's fixup callback no longer needed to call ast_sip_push_task_synchronous(). However, there are a few places where this unexpected behavior is still required to avoid deadlocks. The pjsip monitor thread executes callbacks that do calls to ast_sip_push_task_synchronous() that would deadlock if the task were actually pushed to the specified serializer. I ran into one dealing with the pubsub subscriptions where an ao2 destructor called ast_sip_push_task_synchronous(). * Split ast_sip_push_task_synchronous() into ast_sip_push_task_wait_servant() and ast_sip_push_task_wait_serializer(). ast_sip_push_task_wait_servant() has the old behavior of ast_sip_push_task_synchronous(). ast_sip_push_task_wait_serializer() has the new behavior where the task is always executed by the specified serializer or a picked serializer if one is not passed in. Both functions behave the same if the current thread is not a SIP servant. * Redirected ast_sip_push_task_synchronous() to ast_sip_push_task_wait_servant() to preserve API for released branches. ASTERISK_26806 Change-Id: Id040fa42c0e5972f4c8deef380921461d213b9f3
2018-03-27 16:04:42 +00:00
if (ast_sip_push_task_wait_serializer(serializer, sub_persistence_recreate,
&recreate_data)) {
ast_log(LOG_WARNING, "Failed recreating '%s' subscription: Could not continue under distributor serializer.\n",
persistence->endpoint);
ast_sorcery_delete(ast_sip_get_sorcery(), persistence);
}
ast_taskprocessor_unreference(serializer);
return 0;
}
/*! \brief Function which loads and recreates persisted subscriptions upon startup when the system is fully booted */
static int subscription_persistence_load(void *data)
{
struct ao2_container *persisted_subscriptions = ast_sorcery_retrieve_by_fields(ast_sip_get_sorcery(),
"subscription_persistence", AST_RETRIEVE_FLAG_MULTIPLE | AST_RETRIEVE_FLAG_ALL, NULL);
pj_pool_t *pool;
pool = pjsip_endpt_create_pool(ast_sip_get_pjsip_endpoint(), "rtd%p", PJSIP_POOL_RDATA_LEN,
PJSIP_POOL_RDATA_INC);
if (!pool) {
ast_log(LOG_WARNING, "Could not create a memory pool for recreating SIP subscriptions\n");
return 0;
}
ao2_callback(persisted_subscriptions, OBJ_NODATA, subscription_persistence_recreate, pool);
pjsip_endpt_release_pool(ast_sip_get_pjsip_endpoint(), pool);
ao2_ref(persisted_subscriptions, -1);
return 0;
}
/*! \brief Event callback which fires subscription persistence recreation when the system is fully booted */
static void subscription_persistence_event_cb(void *data, struct stasis_subscription *sub, struct stasis_message *message)
{
struct ast_json_payload *payload;
const char *type;
if (stasis_message_type(message) != ast_manager_get_generic_type()) {
return;
}
payload = stasis_message_data(message);
type = ast_json_string_get(ast_json_object_get(payload->json, "type"));
/* This subscription only responds to the FullyBooted event so that all modules have been loaded when we
* recreate SIP subscriptions.
*/
if (strcmp(type, "FullyBooted")) {
return;
}
/* This has to be here so the subscription is recreated when the body generator is available */
ast_sip_push_task(NULL, subscription_persistence_load, NULL);
/* Once the system is fully booted we don't care anymore */
stasis_unsubscribe(sub);
}
typedef int (*on_subscription_t)(struct sip_subscription_tree *sub, void *arg);
static int for_each_subscription(on_subscription_t on_subscription, void *arg)
{
int num = 0;
struct sip_subscription_tree *i;
if (!on_subscription) {
return num;
}
AST_RWLIST_RDLOCK(&subscriptions);
AST_RWLIST_TRAVERSE(&subscriptions, i, next) {
if (on_subscription(i, arg)) {
break;
}
++num;
}
AST_RWLIST_UNLOCK(&subscriptions);
return num;
}
static void sip_subscription_to_ami(struct sip_subscription_tree *sub_tree,
struct ast_str **buf)
{
char str[256];
struct ast_sip_endpoint_id_configuration *id = &sub_tree->endpoint->id;
ast_str_append(buf, 0, "Role: %s\r\n",
sip_subscription_roles_map[sub_tree->role]);
ast_str_append(buf, 0, "Endpoint: %s\r\n",
ast_sorcery_object_get_id(sub_tree->endpoint));
if (sub_tree->dlg) {
ast_copy_pj_str(str, &sub_tree->dlg->call_id->id, sizeof(str));
} else {
ast_copy_string(str, "<unknown>", sizeof(str));
}
ast_str_append(buf, 0, "Callid: %s\r\n", str);
ast_str_append(buf, 0, "State: %s\r\n", pjsip_evsub_get_state_name(sub_tree->evsub));
ast_callerid_merge(str, sizeof(str),
S_COR(id->self.name.valid, id->self.name.str, NULL),
S_COR(id->self.number.valid, id->self.number.str, NULL),
"Unknown");
ast_str_append(buf, 0, "Callerid: %s\r\n", str);
/* XXX This needs to be done recursively for lists */
if (sub_tree->root->handler->to_ami) {
sub_tree->root->handler->to_ami(sub_tree->root, buf);
}
}
void *ast_sip_subscription_get_header(const struct ast_sip_subscription *sub, const char *header)
{
pjsip_dialog *dlg;
pjsip_msg *msg;
pj_str_t name;
dlg = sub->tree->dlg;
msg = ast_sip_mod_data_get(dlg->mod_data, pubsub_module.id, MOD_DATA_MSG);
pj_cstr(&name, header);
return pjsip_msg_find_hdr_by_name(msg, &name, NULL);
}
/* XXX This function is not used. */
struct ast_sip_subscription *ast_sip_create_subscription(const struct ast_sip_subscription_handler *handler,
struct ast_sip_endpoint *endpoint, const char *resource)
{
struct ast_sip_subscription *sub;
pjsip_dialog *dlg;
struct ast_sip_contact *contact;
pj_str_t event;
pjsip_tx_data *tdata;
pjsip_evsub *evsub;
struct sip_subscription_tree *sub_tree = NULL;
sub_tree = allocate_subscription_tree(endpoint, NULL);
if (!sub_tree) {
return NULL;
}
sub = allocate_subscription(handler, resource, NULL, sub_tree);
if (!sub) {
ao2_cleanup(sub_tree);
return NULL;
}
contact = ast_sip_location_retrieve_contact_from_aor_list(endpoint->aors);
if (!contact || ast_strlen_zero(contact->uri)) {
ast_log(LOG_WARNING, "No contacts configured for endpoint %s. Unable to create SIP subsription\n",
ast_sorcery_object_get_id(endpoint));
ao2_ref(sub_tree, -1);
ao2_cleanup(contact);
return NULL;
}
dlg = ast_sip_create_dialog_uac(endpoint, contact->uri, NULL);
ao2_cleanup(contact);
if (!dlg) {
ast_log(LOG_WARNING, "Unable to create dialog for SIP subscription\n");
ao2_ref(sub_tree, -1);
return NULL;
}
pj_cstr(&event, handler->event_name);
pjsip_evsub_create_uac(dlg, &pubsub_cb, &event, 0, &sub_tree->evsub);
subscription_setup_dialog(sub_tree, dlg);
evsub = sub_tree->evsub;
if (pjsip_evsub_initiate(evsub, NULL, -1, &tdata) == PJ_SUCCESS) {
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
pjsip_evsub_send_request(sub_tree->evsub, tdata);
} else {
/* pjsip_evsub_terminate will result in pubsub_on_evsub_state,
* being called and terminating the subscription. Therefore, we don't
* need to decrease the reference count of sub here.
*/
pjsip_evsub_terminate(evsub, PJ_TRUE);
ao2_ref(sub_tree, -1);
return NULL;
}
add_subscription(sub_tree);
return sub;
}
res_pjsip_mwi: Add voicemail extension and mwi_subscribe_replaces_unsolicited res_pjsip_mwi was missing the chan_sip "vmexten" functionality which adds the Message-Account header to the MWI NOTIFY. Also, specifying mailboxes on endpoints for unsolicited mwi and on aors for subscriptions required that the admin know in advance which the client wanted. If you specified mailboxes on the endpoint, subscriptions were rejected even if you also specified mailboxes on the aor. Voicemail extension: * Added a global default_voicemail_extension which defaults to "". * Added voicemail_extension to both endpoint and aor. * Added ast_sip_subscription_get_dialog for support. * Added ast_sip_subscription_get_sip_uri for support. When an unsolicited NOTIFY is constructed, the From header is parsed, the voicemail extension from the endpoint is substituted for the user, and the result placed in the Message-Account field in the body. When a subscribed NOTIFY is constructed, the subscription dialog local uri is parsed, the voicemail_extension from the aor (looked up from the subscription resource name) is substituted for the user, and the result placed in the Message-Account field in the body. If no voicemail extension was defined, the Message-Account field is not added to the NOTIFY body. mwi_subscribe_replaces_unsolicited: * Added mwi_subscribe_replaces_unsolicited to endpoint. The previous behavior was to reject a subscribe if a previous internal subscription for unsolicited MWI was found for the mailbox. That remains the default. However, if there are mailboxes also set on the aor and the client subscribes and mwi_subscribe_replaces_unsolicited is set, the existing internal subscription is removed and replaced with the external subscription. This allows an admin to configure mailboxes on both the endpoint and aor and allows the client to select which to use. ASTERISK-25865 #close Reported-by: Ross Beer Change-Id: Ic15a9415091760539c7134a5ba3dc4a6a1217cea
2016-03-25 03:55:03 +00:00
pjsip_dialog *ast_sip_subscription_get_dialog(struct ast_sip_subscription *sub)
{
ast_assert(sub->tree->dlg != NULL);
return sub->tree->dlg;
}
struct ast_sip_endpoint *ast_sip_subscription_get_endpoint(struct ast_sip_subscription *sub)
{
ast_assert(sub->tree->endpoint != NULL);
return ao2_bump(sub->tree->endpoint);
}
struct ast_taskprocessor *ast_sip_subscription_get_serializer(struct ast_sip_subscription *sub)
{
ast_assert(sub->tree->serializer != NULL);
return sub->tree->serializer;
}
/*!
* \brief Pre-allocate a buffer for the transmission
*
* Typically, we let PJSIP do this step for us when we send a request. PJSIP's buffer
* allocation algorithm is to allocate a buffer of PJSIP_MAX_PKT_LEN bytes and attempt
* to write the packet to the allocated buffer. If the buffer is too small to hold the
* packet, then we get told the message is too long to be sent.
*
* When dealing with SIP NOTIFY, especially with RLS, it is possible to exceed
* PJSIP_MAX_PKT_LEN. Rather than accepting the limitation imposed on us by default,
* we instead take the strategy of pre-allocating the buffer, testing for ourselves
* if the message will fit, and resizing the buffer as required.
*
* The limit we impose is double that of the maximum packet length.
*
* \param tdata The tdata onto which to allocate a buffer
* \retval 0 Success
* \retval -1 The message is too large
*/
static int allocate_tdata_buffer(pjsip_tx_data *tdata)
{
int buf_size;
int size = -1;
char *buf;
for (buf_size = PJSIP_MAX_PKT_LEN; size == -1 && buf_size < (PJSIP_MAX_PKT_LEN * 2); buf_size *= 2) {
buf = pj_pool_alloc(tdata->pool, buf_size);
size = pjsip_msg_print(tdata->msg, buf, buf_size);
}
if (size == -1) {
return -1;
}
tdata->buf.start = buf;
tdata->buf.cur = tdata->buf.start;
tdata->buf.end = tdata->buf.start + buf_size;
return 0;
}
static int sip_subscription_send_request(struct sip_subscription_tree *sub_tree, pjsip_tx_data *tdata)
{
#ifdef TEST_FRAMEWORK
struct ast_sip_endpoint *endpoint = sub_tree->endpoint;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
pjsip_evsub *evsub = sub_tree->evsub;
#endif
int res;
if (allocate_tdata_buffer(tdata)) {
ast_log(LOG_ERROR, "SIP request %s is too large to send.\n", tdata->info);
pjsip_tx_data_dec_ref(tdata);
return -1;
}
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
res = pjsip_evsub_send_request(sub_tree->evsub, tdata);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
subscription_persistence_update(sub_tree, NULL, SUBSCRIPTION_PERSISTENCE_SEND_REQUEST);
ast_test_suite_event_notify("SUBSCRIPTION_STATE_SET",
"StateText: %s\r\n"
"Endpoint: %s\r\n",
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
pjsip_evsub_get_state_name(evsub),
ast_sorcery_object_get_id(endpoint));
return (res == PJ_SUCCESS ? 0 : -1);
}
/*!
* \brief Add a resource XML element to an RLMI body
*
* Each resource element represents a subscribed resource in the list. This function currently
* will unconditionally add an instance element to each created resource element. Instance
* elements refer to later parts in the multipart body.
*
* \param pool PJLIB allocation pool
* \param rlmi
* \param cid Content-ID header of the resource
* \param resource_name Name of the resource
* \param resource_uri URI of the resource
* \param state State of the subscribed resource
*/
static void add_rlmi_resource(pj_pool_t *pool, pj_xml_node *rlmi, const pjsip_generic_string_hdr *cid,
const char *resource_name, const pjsip_sip_uri *resource_uri, pjsip_evsub_state state)
{
static pj_str_t cid_name = { "cid", 3 };
pj_xml_node *resource;
pj_xml_node *name;
pj_xml_node *instance;
pj_xml_attr *cid_attr;
char id[6];
char uri[PJSIP_MAX_URL_SIZE];
char name_sanitized[PJSIP_MAX_URL_SIZE];
/* This creates a string representing the Content-ID without the enclosing < > */
const pj_str_t cid_stripped = {
.ptr = cid->hvalue.ptr + 1,
.slen = cid->hvalue.slen - 2,
};
resource = ast_sip_presence_xml_create_node(pool, rlmi, "resource");
name = ast_sip_presence_xml_create_node(pool, resource, "name");
instance = ast_sip_presence_xml_create_node(pool, resource, "instance");
pjsip_uri_print(PJSIP_URI_IN_CONTACT_HDR, resource_uri, uri, sizeof(uri));
ast_sip_presence_xml_create_attr(pool, resource, "uri", uri);
ast_sip_sanitize_xml(resource_name, name_sanitized, sizeof(name_sanitized));
pj_strdup2(pool, &name->content, name_sanitized);
ast_generate_random_string(id, sizeof(id));
ast_sip_presence_xml_create_attr(pool, instance, "id", id);
ast_sip_presence_xml_create_attr(pool, instance, "state",
state == PJSIP_EVSUB_STATE_TERMINATED ? "terminated" : "active");
/* Use the PJLIB-util XML library directly here since we are using a
* pj_str_t
*/
cid_attr = pj_xml_attr_new(pool, &cid_name, &cid_stripped);
pj_xml_add_attr(instance, cid_attr);
}
/*!
* \brief A multipart body part and meta-information
*
* When creating a multipart body part, the end result (the
* pjsip_multipart_part) is hard to inspect without undoing
* a lot of what was done to create it. Therefore, we use this
* structure to store meta-information about the body part.
*
* The main consumer of this is the creator of the RLMI body
* part of a multipart resource list body.
*/
struct body_part {
/*! Content-ID header for the body part */
pjsip_generic_string_hdr *cid;
/*! Subscribed resource represented in the body part */
const char *resource;
/*! URI for the subscribed body part */
pjsip_sip_uri *uri;
/*! Subscription state of the resource represented in the body part */
pjsip_evsub_state state;
/*! The actual body part that will be present in the multipart body */
pjsip_multipart_part *part;
/*! Display name for the resource */
const char *display_name;
};
/*!
* \brief Type declaration for container of body part structures
*/
AST_VECTOR(body_part_list, struct body_part *);
/*!
* \brief Create a Content-ID header
*
* Content-ID headers are required by RFC2387 for multipart/related
* bodies. They serve as identifiers for each part of the multipart body.
*
* \param pool PJLIB allocation pool
* \param sub Subscription to a resource
*/
static pjsip_generic_string_hdr *generate_content_id_hdr(pj_pool_t *pool,
const struct ast_sip_subscription *sub)
{
static const pj_str_t cid_name = { "Content-ID", 10 };
pjsip_generic_string_hdr *cid;
char id[6];
size_t alloc_size;
pj_str_t cid_value;
/* '<' + '@' + '>' = 3. pj_str_t does not require a null-terminator */
alloc_size = sizeof(id) + pj_strlen(&sub->uri->host) + 3;
cid_value.ptr = pj_pool_alloc(pool, alloc_size);
cid_value.slen = sprintf(cid_value.ptr, "<%s@%.*s>",
ast_generate_random_string(id, sizeof(id)),
(int) pj_strlen(&sub->uri->host), pj_strbuf(&sub->uri->host));
cid = pjsip_generic_string_hdr_create(pool, &cid_name, &cid_value);
return cid;
}
static int rlmi_print_body(struct pjsip_msg_body *msg_body, char *buf, pj_size_t size)
{
int num_printed;
pj_xml_node *rlmi = msg_body->data;
num_printed = pj_xml_print(rlmi, buf, size, PJ_TRUE);
if (num_printed <= AST_PJSIP_XML_PROLOG_LEN) {
return -1;
}
return num_printed;
}
static void *rlmi_clone_data(pj_pool_t *pool, const void *data, unsigned len)
{
const pj_xml_node *rlmi = data;
return pj_xml_clone(pool, rlmi);
}
/*!
* \brief Create an RLMI body part for a multipart resource list body
*
* RLMI (Resource list meta information) is a special body type that lists
* the subscribed resources and tells subscribers the number of subscribed
* resources and what other body parts are in the multipart body. The
* RLMI body also has a version number that a subscriber can use to ensure
* that the locally-stored state corresponds to server state.
*
* \param pool The allocation pool
* \param sub The subscription representing the subscribed resource list
* \param body_parts A container of body parts that RLMI will refer to
* \param full_state Indicates whether this is a full or partial state notification
* \return The multipart part representing the RLMI body
*/
static pjsip_multipart_part *build_rlmi_body(pj_pool_t *pool, struct ast_sip_subscription *sub,
struct body_part_list *body_parts, unsigned int full_state)
{
pj_xml_node *rlmi;
pj_xml_node *name;
pjsip_multipart_part *rlmi_part;
char version_str[32];
char uri[PJSIP_MAX_URL_SIZE];
pjsip_generic_string_hdr *cid;
int i;
rlmi = ast_sip_presence_xml_create_node(pool, NULL, "list");
ast_sip_presence_xml_create_attr(pool, rlmi, "xmlns", "urn:ietf:params:xml:ns:rlmi");
ast_sip_subscription_get_local_uri(sub, uri, sizeof(uri));
ast_sip_presence_xml_create_attr(pool, rlmi, "uri", uri);
snprintf(version_str, sizeof(version_str), "%u", sub->version++);
ast_sip_presence_xml_create_attr(pool, rlmi, "version", version_str);
ast_sip_presence_xml_create_attr(pool, rlmi, "fullState", full_state ? "true" : "false");
name = ast_sip_presence_xml_create_node(pool, rlmi, "name");
pj_strdup2(pool, &name->content, ast_sip_subscription_get_resource_name(sub));
for (i = 0; i < AST_VECTOR_SIZE(body_parts); ++i) {
const struct body_part *part = AST_VECTOR_GET(body_parts, i);
add_rlmi_resource(pool, rlmi, part->cid, S_OR(part->display_name, part->resource), part->uri, part->state);
}
rlmi_part = pjsip_multipart_create_part(pool);
rlmi_part->body = PJ_POOL_ZALLOC_T(pool, pjsip_msg_body);
pjsip_media_type_cp(pool, &rlmi_part->body->content_type, &rlmi_media_type);
rlmi_part->body->data = pj_xml_clone(pool, rlmi);
rlmi_part->body->clone_data = rlmi_clone_data;
rlmi_part->body->print_body = rlmi_print_body;
cid = generate_content_id_hdr(pool, sub);
pj_list_insert_before(&rlmi_part->hdr, cid);
return rlmi_part;
}
static pjsip_msg_body *generate_notify_body(pj_pool_t *pool, struct ast_sip_subscription *root,
unsigned int force_full_state);
/*!
* \brief Destroy a list of body parts
*
* \param parts The container of parts to destroy
*/
static void free_body_parts(struct body_part_list *parts)
{
int i;
for (i = 0; i < AST_VECTOR_SIZE(parts); ++i) {
struct body_part *part = AST_VECTOR_GET(parts, i);
ast_free(part);
}
AST_VECTOR_FREE(parts);
}
/*!
* \brief Allocate and initialize a body part structure
*
* \param pool PJLIB allocation pool
* \param sub Subscription representing a subscribed resource
*/
static struct body_part *allocate_body_part(pj_pool_t *pool, const struct ast_sip_subscription *sub)
{
struct body_part *bp;
bp = ast_calloc(1, sizeof(*bp));
if (!bp) {
return NULL;
}
bp->cid = generate_content_id_hdr(pool, sub);
bp->resource = sub->resource;
bp->state = sub->subscription_state;
bp->uri = sub->uri;
bp->display_name = sub->display_name;
return bp;
}
/*!
* \brief Create a multipart body part for a subscribed resource
*
* \param pool PJLIB allocation pool
* \param sub The subscription representing a subscribed resource
* \param parts A vector of parts to append the created part to.
* \param use_full_state Unused locally, but may be passed to other functions
*/
static void build_body_part(pj_pool_t *pool, struct ast_sip_subscription *sub,
struct body_part_list *parts, unsigned int use_full_state)
{
struct body_part *bp;
pjsip_msg_body *body;
bp = allocate_body_part(pool, sub);
if (!bp) {
return;
}
body = generate_notify_body(pool, sub, use_full_state);
if (!body) {
/* Partial state was requested and the resource has not changed state */
ast_free(bp);
return;
}
bp->part = pjsip_multipart_create_part(pool);
bp->part->body = body;
pj_list_insert_before(&bp->part->hdr, bp->cid);
if (AST_VECTOR_APPEND(parts, bp)) {
ast_free(bp);
}
}
/*!
* \brief Create and initialize the PJSIP multipart body structure for a resource list subscription
*
* \param pool
* \return The multipart message body
*/
static pjsip_msg_body *create_multipart_body(pj_pool_t *pool)
{
pjsip_media_type media_type;
pjsip_param *media_type_param;
char boundary[6];
pj_str_t pj_boundary;
pjsip_media_type_init2(&media_type, "multipart", "related");
media_type_param = pj_pool_alloc(pool, sizeof(*media_type_param));
pj_list_init(media_type_param);
pj_strdup2(pool, &media_type_param->name, "type");
pj_strdup2(pool, &media_type_param->value, "\"application/rlmi+xml\"");
pj_list_insert_before(&media_type.param, media_type_param);
pj_cstr(&pj_boundary, ast_generate_random_string(boundary, sizeof(boundary)));
return pjsip_multipart_create(pool, &media_type, &pj_boundary);
}
/*!
* \brief Create a resource list body for NOTIFY requests
*
* Resource list bodies are multipart/related bodies. The first part of the multipart body
* is an RLMI body that describes the rest of the parts to come. The other parts of the body
* convey state of individual subscribed resources.
*
* \param pool PJLIB allocation pool
* \param sub Subscription details from which to generate body
* \param force_full_state If true, ignore resource list settings and send a full state notification
* \return The generated multipart/related body
*/
static pjsip_msg_body *generate_list_body(pj_pool_t *pool, struct ast_sip_subscription *sub,
unsigned int force_full_state)
{
int i;
pjsip_multipart_part *rlmi_part;
pjsip_msg_body *multipart;
struct body_part_list body_parts;
unsigned int use_full_state = force_full_state ? 1 : sub->full_state;
if (AST_VECTOR_INIT(&body_parts, AST_VECTOR_SIZE(&sub->children))) {
return NULL;
}
for (i = 0; i < AST_VECTOR_SIZE(&sub->children); ++i) {
build_body_part(pool, AST_VECTOR_GET(&sub->children, i), &body_parts, use_full_state);
}
/* This can happen if issuing partial state and no children of the list have changed state */
if (AST_VECTOR_SIZE(&body_parts) == 0) {
free_body_parts(&body_parts);
return NULL;
}
multipart = create_multipart_body(pool);
rlmi_part = build_rlmi_body(pool, sub, &body_parts, use_full_state);
if (!rlmi_part) {
free_body_parts(&body_parts);
return NULL;
}
pjsip_multipart_add_part(pool, multipart, rlmi_part);
for (i = 0; i < AST_VECTOR_SIZE(&body_parts); ++i) {
pjsip_multipart_add_part(pool, multipart, AST_VECTOR_GET(&body_parts, i)->part);
}
free_body_parts(&body_parts);
return multipart;
}
/*!
* \brief Create the body for a NOTIFY request.
*
* \param pool The pool used for allocations
* \param root The root of the subscription tree
* \param force_full_state If true, ignore resource list settings and send a full state notification
*/
static pjsip_msg_body *generate_notify_body(pj_pool_t *pool, struct ast_sip_subscription *root,
unsigned int force_full_state)
{
pjsip_msg_body *body;
if (AST_VECTOR_SIZE(&root->children) == 0) {
if (force_full_state || root->body_changed) {
/* Not a list. We've already generated the body and saved it on the subscription.
* Use that directly.
*/
pj_str_t type;
pj_str_t subtype;
pj_str_t text;
pj_cstr(&type, ast_sip_subscription_get_body_type(root));
pj_cstr(&subtype, ast_sip_subscription_get_body_subtype(root));
pj_cstr(&text, ast_str_buffer(root->body_text));
body = pjsip_msg_body_create(pool, &type, &subtype, &text);
root->body_changed = 0;
} else {
body = NULL;
}
} else {
body = generate_list_body(pool, root, force_full_state);
}
return body;
}
/*!
* \brief Shortcut method to create a Require: eventlist header
*/
static pjsip_require_hdr *create_require_eventlist(pj_pool_t *pool)
{
pjsip_require_hdr *require;
require = pjsip_require_hdr_create(pool);
pj_strdup2(pool, &require->values[0], "eventlist");
require->count = 1;
return require;
}
/*!
* \brief Send a NOTIFY request to a subscriber
*
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
* \pre sub_tree->dlg is locked
*
* \param sub_tree The subscription tree representing the subscription
* \param force_full_state If true, ignore resource list settings and send full resource list state.
* \retval 0 Success
* \retval non-zero Failure
*/
static int send_notify(struct sip_subscription_tree *sub_tree, unsigned int force_full_state)
{
pjsip_evsub *evsub = sub_tree->evsub;
pjsip_tx_data *tdata;
HTTP: Stop accepting requests on final system shutdown. There are three CLI commands to stop and restart Asterisk each. 1) core stop/restart now - Hangup all calls and stop or restart Asterisk. New channels are prevented while the shutdown request is pending. 2) core stop/restart gracefully - Stop or restart Asterisk when there are no calls remaining in the system. New channels are prevented while the shutdown request is pending. 3) core stop/restart when convenient - Stop or restart Asterisk when there are no calls in the system. New calls are not prevented while the shutdown request is pending. ARI has made stopping/restarting Asterisk more problematic. While a shutdown request is pending it is desirable to continue to process ARI HTTP requests for current calls. To handle the current calls while a shutdown request is pending, a new committed to shutdown phase is needed so ARI applications can deal with the calls until the system is fully committed to shutdown. * Added a new shutdown committed phase so ARI applications can deal with calls until the final committed to shutdown phase is reached. * Made refuse new HTTP requests when the system has reached the final system shutdown phase. Starting anything while the system is actively releasing resources and unloading modules is not a good thing. * Split the bridging framework shutdown to not cleanup the global bridging containers when shutting down in a hurry. This is similar to how other modules prevent crashes on rapid system shutdown. * Moved ast_begin_shutdown(), ast_cancel_shutdown(), and ast_shutting_down(). You should not have to include channel.h just to access these system functions. ASTERISK-24752 #close Reported by: Matthew Jordan Review: https://reviewboard.asterisk.org/r/4399/ ........ Merged revisions 431692 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@431694 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-02-11 17:39:13 +00:00
if (ast_shutdown_final()
&& sub_tree->root->subscription_state == PJSIP_EVSUB_STATE_TERMINATED
&& sub_tree->persistence) {
return 0;
}
if (pjsip_evsub_notify(evsub, sub_tree->root->subscription_state,
NULL, NULL, &tdata) != PJ_SUCCESS) {
return -1;
}
tdata->msg->body = generate_notify_body(tdata->pool, sub_tree->root, force_full_state);
if (!tdata->msg->body) {
pjsip_tx_data_dec_ref(tdata);
return -1;
}
if (sub_tree->is_list) {
pjsip_require_hdr *require = create_require_eventlist(tdata->pool);
pjsip_msg_add_hdr(tdata->msg, (pjsip_hdr *) require);
}
if (sip_subscription_send_request(sub_tree, tdata)) {
/* do not call pjsip_tx_data_dec_ref(tdata). The pjsip_dlg_send_request deletes the message on error */
return -1;
}
sub_tree->send_scheduled_notify = 0;
return 0;
}
static int serialized_send_notify(void *userdata)
{
struct sip_subscription_tree *sub_tree = userdata;
pjsip_dialog *dlg = sub_tree->dlg;
pjsip_dlg_inc_lock(dlg);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
sub_tree->notify_sched_id = -1;
/* It's possible that between when the notification was scheduled
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
* and now a new SUBSCRIBE arrived requiring full state to be
* sent out in an immediate NOTIFY. It's also possible that we're
* already processing a terminate. If that has happened, we need to
* bail out here instead of sending the batched NOTIFY.
*/
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (sub_tree->state >= SIP_SUB_TREE_TERMINATE_IN_PROGRESS
|| !sub_tree->send_scheduled_notify) {
pjsip_dlg_dec_lock(dlg);
ao2_cleanup(sub_tree);
return 0;
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (sub_tree->root->subscription_state == PJSIP_EVSUB_STATE_TERMINATED) {
sub_tree->state = SIP_SUB_TREE_TERMINATE_IN_PROGRESS;
}
send_notify(sub_tree, 0);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_test_suite_event_notify(
sub_tree->state == SIP_SUB_TREE_TERMINATED
? "SUBSCRIPTION_TERMINATED" : "SUBSCRIPTION_STATE_CHANGED",
"Resource: %s", sub_tree->root->resource);
pjsip_dlg_dec_lock(dlg);
ao2_cleanup(sub_tree);
return 0;
}
static int sched_cb(const void *data)
{
struct sip_subscription_tree *sub_tree = (struct sip_subscription_tree *) data;
/* We don't need to bump the refcount of sub_tree since we bumped it when scheduling this task */
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (ast_sip_push_task(sub_tree->serializer, serialized_send_notify, sub_tree)) {
ao2_cleanup(sub_tree);
}
return 0;
}
static int schedule_notification(struct sip_subscription_tree *sub_tree)
{
/* There's already a notification scheduled */
if (sub_tree->notify_sched_id > -1) {
return 0;
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
sub_tree->send_scheduled_notify = 1;
sub_tree->notify_sched_id = ast_sched_add(sched, sub_tree->notification_batch_interval, sched_cb, ao2_bump(sub_tree));
if (sub_tree->notify_sched_id < 0) {
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ao2_cleanup(sub_tree);
return -1;
}
return 0;
}
int ast_sip_subscription_notify(struct ast_sip_subscription *sub, struct ast_sip_body_data *notify_data,
int terminate)
{
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
int res;
pjsip_dialog *dlg = sub->tree->dlg;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
pjsip_dlg_inc_lock(dlg);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (sub->tree->state != SIP_SUB_TREE_NORMAL) {
pjsip_dlg_dec_lock(dlg);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
return 0;
}
if (ast_sip_pubsub_generate_body_content(ast_sip_subscription_get_body_type(sub),
ast_sip_subscription_get_body_subtype(sub), notify_data, &sub->body_text)) {
pjsip_dlg_dec_lock(dlg);
return -1;
}
sub->body_changed = 1;
if (terminate) {
sub->subscription_state = PJSIP_EVSUB_STATE_TERMINATED;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
sub->tree->state = SIP_SUB_TREE_TERMINATE_PENDING;
}
if (sub->tree->notification_batch_interval) {
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
res = schedule_notification(sub->tree);
} else {
/* See the note in pubsub_on_rx_refresh() for why sub->tree is refbumped here */
ao2_ref(sub->tree, +1);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (terminate) {
sub->tree->state = SIP_SUB_TREE_TERMINATE_IN_PROGRESS;
}
res = send_notify(sub->tree, 0);
ast_test_suite_event_notify(terminate ? "SUBSCRIPTION_TERMINATED" : "SUBSCRIPTION_STATE_CHANGED",
"Resource: %s",
sub->tree->root->resource);
ao2_ref(sub->tree, -1);
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
pjsip_dlg_dec_lock(dlg);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
return res;
}
res_pjsip_mwi: Add voicemail extension and mwi_subscribe_replaces_unsolicited res_pjsip_mwi was missing the chan_sip "vmexten" functionality which adds the Message-Account header to the MWI NOTIFY. Also, specifying mailboxes on endpoints for unsolicited mwi and on aors for subscriptions required that the admin know in advance which the client wanted. If you specified mailboxes on the endpoint, subscriptions were rejected even if you also specified mailboxes on the aor. Voicemail extension: * Added a global default_voicemail_extension which defaults to "". * Added voicemail_extension to both endpoint and aor. * Added ast_sip_subscription_get_dialog for support. * Added ast_sip_subscription_get_sip_uri for support. When an unsolicited NOTIFY is constructed, the From header is parsed, the voicemail extension from the endpoint is substituted for the user, and the result placed in the Message-Account field in the body. When a subscribed NOTIFY is constructed, the subscription dialog local uri is parsed, the voicemail_extension from the aor (looked up from the subscription resource name) is substituted for the user, and the result placed in the Message-Account field in the body. If no voicemail extension was defined, the Message-Account field is not added to the NOTIFY body. mwi_subscribe_replaces_unsolicited: * Added mwi_subscribe_replaces_unsolicited to endpoint. The previous behavior was to reject a subscribe if a previous internal subscription for unsolicited MWI was found for the mailbox. That remains the default. However, if there are mailboxes also set on the aor and the client subscribes and mwi_subscribe_replaces_unsolicited is set, the existing internal subscription is removed and replaced with the external subscription. This allows an admin to configure mailboxes on both the endpoint and aor and allows the client to select which to use. ASTERISK-25865 #close Reported-by: Ross Beer Change-Id: Ic15a9415091760539c7134a5ba3dc4a6a1217cea
2016-03-25 03:55:03 +00:00
pjsip_sip_uri *ast_sip_subscription_get_sip_uri(struct ast_sip_subscription *sub)
{
return sub->uri;
}
void ast_sip_subscription_get_local_uri(struct ast_sip_subscription *sub, char *buf, size_t size)
{
pjsip_uri_print(PJSIP_URI_IN_CONTACT_HDR, sub->uri, buf, size);
}
void ast_sip_subscription_get_remote_uri(struct ast_sip_subscription *sub, char *buf, size_t size)
{
pjsip_dialog *dlg;
pjsip_sip_uri *uri;
dlg = sub->tree->dlg;
uri = pjsip_uri_get_uri(dlg->remote.info->uri);
if (pjsip_uri_print(PJSIP_URI_IN_FROMTO_HDR, uri, buf, size) < 0) {
*buf = '\0';
}
}
const char *ast_sip_subscription_get_resource_name(struct ast_sip_subscription *sub)
{
return sub->resource;
}
int ast_sip_subscription_is_terminated(const struct ast_sip_subscription *sub)
{
return sub->subscription_state == PJSIP_EVSUB_STATE_TERMINATED ? 1 : 0;
}
static int sip_subscription_accept(struct sip_subscription_tree *sub_tree, pjsip_rx_data *rdata, int response)
{
pjsip_hdr res_hdr;
/* If this is a persistence recreation the subscription has already been accepted */
if (ast_sip_mod_data_get(rdata->endpt_info.mod_data, pubsub_module.id, MOD_DATA_PERSISTENCE)) {
return 0;
}
pj_list_init(&res_hdr);
if (sub_tree->is_list) {
/* If subscribing to a list, our response has to have a Require: eventlist header in it */
pj_list_insert_before(&res_hdr, create_require_eventlist(rdata->tp_info.pool));
}
return pjsip_evsub_accept(sub_tree->evsub, rdata, response, &res_hdr) == PJ_SUCCESS ? 0 : -1;
}
struct ast_datastore *ast_sip_subscription_alloc_datastore(const struct ast_datastore_info *info, const char *uid)
{
return ast_datastores_alloc_datastore(info, uid);
}
int ast_sip_subscription_add_datastore(struct ast_sip_subscription *subscription, struct ast_datastore *datastore)
{
return ast_datastores_add(subscription->datastores, datastore);
}
struct ast_datastore *ast_sip_subscription_get_datastore(struct ast_sip_subscription *subscription, const char *name)
{
return ast_datastores_find(subscription->datastores, name);
}
void ast_sip_subscription_remove_datastore(struct ast_sip_subscription *subscription, const char *name)
{
ast_datastores_remove(subscription->datastores, name);
}
struct ao2_container *ast_sip_subscription_get_datastores(const struct ast_sip_subscription *subscription)
{
return subscription->datastores;
}
int ast_sip_publication_add_datastore(struct ast_sip_publication *publication, struct ast_datastore *datastore)
{
return ast_datastores_add(publication->datastores, datastore);
}
struct ast_datastore *ast_sip_publication_get_datastore(struct ast_sip_publication *publication, const char *name)
{
return ast_datastores_find(publication->datastores, name);
}
void ast_sip_publication_remove_datastore(struct ast_sip_publication *publication, const char *name)
{
ast_datastores_remove(publication->datastores, name);
}
struct ao2_container *ast_sip_publication_get_datastores(const struct ast_sip_publication *publication)
{
return publication->datastores;
}
void ast_sip_subscription_set_persistence_data(struct ast_sip_subscription *subscription, struct ast_json *persistence_data)
{
ast_json_unref(subscription->persistence_data);
subscription->persistence_data = persistence_data;
if (subscription->tree->persistence) {
if (!subscription->tree->persistence->generator_data) {
subscription->tree->persistence->generator_data = ast_json_object_create();
if (!subscription->tree->persistence->generator_data) {
return;
}
}
ast_json_object_set(subscription->tree->persistence->generator_data, subscription->resource,
ast_json_ref(persistence_data));
}
}
const struct ast_json *ast_sip_subscription_get_persistence_data(const struct ast_sip_subscription *subscription)
{
return subscription->persistence_data;
}
AST_RWLIST_HEAD_STATIC(publish_handlers, ast_sip_publish_handler);
static int publication_hash_fn(const void *obj, const int flags)
{
const struct ast_sip_publication *publication = obj;
const int *entity_tag = obj;
return flags & OBJ_KEY ? *entity_tag : publication->entity_tag;
}
static int publication_cmp_fn(void *obj, void *arg, int flags)
{
const struct ast_sip_publication *publication1 = obj;
const struct ast_sip_publication *publication2 = arg;
const int *entity_tag = arg;
return (publication1->entity_tag == (flags & OBJ_KEY ? *entity_tag : publication2->entity_tag) ?
CMP_MATCH | CMP_STOP : 0);
}
static void publish_add_handler(struct ast_sip_publish_handler *handler)
{
AST_RWLIST_WRLOCK(&publish_handlers);
AST_RWLIST_INSERT_TAIL(&publish_handlers, handler, next);
AST_RWLIST_UNLOCK(&publish_handlers);
}
int ast_sip_register_publish_handler(struct ast_sip_publish_handler *handler)
{
if (ast_strlen_zero(handler->event_name)) {
ast_log(LOG_ERROR, "No event package specified for publish handler. Cannot register\n");
return -1;
}
handler->publications = ao2_container_alloc_hash(AO2_ALLOC_OPT_LOCK_MUTEX, 0,
PUBLICATIONS_BUCKETS, publication_hash_fn, NULL, publication_cmp_fn);
if (!handler->publications) {
ast_log(LOG_ERROR, "Could not allocate publications container for event '%s'\n",
handler->event_name);
return -1;
}
publish_add_handler(handler);
return 0;
}
void ast_sip_unregister_publish_handler(struct ast_sip_publish_handler *handler)
{
struct ast_sip_publish_handler *iter;
AST_RWLIST_WRLOCK(&publish_handlers);
AST_RWLIST_TRAVERSE_SAFE_BEGIN(&publish_handlers, iter, next) {
if (handler == iter) {
AST_RWLIST_REMOVE_CURRENT(next);
ao2_cleanup(handler->publications);
break;
}
}
AST_RWLIST_TRAVERSE_SAFE_END;
AST_RWLIST_UNLOCK(&publish_handlers);
}
AST_RWLIST_HEAD_STATIC(subscription_handlers, ast_sip_subscription_handler);
static void sub_add_handler(struct ast_sip_subscription_handler *handler)
{
AST_RWLIST_WRLOCK(&subscription_handlers);
AST_RWLIST_INSERT_TAIL(&subscription_handlers, handler, next);
AST_RWLIST_UNLOCK(&subscription_handlers);
}
static struct ast_sip_subscription_handler *find_sub_handler_for_event_name(const char *event_name)
{
struct ast_sip_subscription_handler *iter;
AST_RWLIST_RDLOCK(&subscription_handlers);
AST_RWLIST_TRAVERSE(&subscription_handlers, iter, next) {
if (!strcmp(iter->event_name, event_name)) {
break;
}
}
AST_RWLIST_UNLOCK(&subscription_handlers);
return iter;
}
int ast_sip_register_subscription_handler(struct ast_sip_subscription_handler *handler)
{
pj_str_t event;
pj_str_t accept[AST_SIP_MAX_ACCEPT] = { {0, }, };
struct ast_sip_subscription_handler *existing;
int i = 0;
if (ast_strlen_zero(handler->event_name)) {
ast_log(LOG_ERROR, "No event package specified for subscription handler. Cannot register\n");
return -1;
}
existing = find_sub_handler_for_event_name(handler->event_name);
if (existing) {
ast_log(LOG_ERROR,
"Unable to register subscription handler for event %s. A handler is already registered\n",
handler->event_name);
return -1;
}
for (i = 0; i < AST_SIP_MAX_ACCEPT && !ast_strlen_zero(handler->accept[i]); ++i) {
pj_cstr(&accept[i], handler->accept[i]);
}
pj_cstr(&event, handler->event_name);
pjsip_evsub_register_pkg(&pubsub_module, &event, DEFAULT_EXPIRES, i, accept);
sub_add_handler(handler);
return 0;
}
void ast_sip_unregister_subscription_handler(struct ast_sip_subscription_handler *handler)
{
struct ast_sip_subscription_handler *iter;
AST_RWLIST_WRLOCK(&subscription_handlers);
AST_RWLIST_TRAVERSE_SAFE_BEGIN(&subscription_handlers, iter, next) {
if (handler == iter) {
AST_RWLIST_REMOVE_CURRENT(next);
break;
}
}
AST_RWLIST_TRAVERSE_SAFE_END;
AST_RWLIST_UNLOCK(&subscription_handlers);
}
static struct ast_sip_pubsub_body_generator *find_body_generator_type_subtype_nolock(const char *type, const char *subtype)
{
struct ast_sip_pubsub_body_generator *gen;
AST_LIST_TRAVERSE(&body_generators, gen, list) {
if (!strcmp(gen->type, type)
&& !strcmp(gen->subtype, subtype)) {
break;
}
}
return gen;
}
static struct ast_sip_pubsub_body_generator *find_body_generator_type_subtype(const char *type, const char *subtype)
{
struct ast_sip_pubsub_body_generator *gen;
AST_RWLIST_RDLOCK(&body_generators);
gen = find_body_generator_type_subtype_nolock(type, subtype);
AST_RWLIST_UNLOCK(&body_generators);
return gen;
}
static struct ast_sip_pubsub_body_generator *find_body_generator_accept(const char *accept)
{
char *accept_copy = ast_strdupa(accept);
char *subtype = accept_copy;
char *type = strsep(&subtype, "/");
if (ast_strlen_zero(type) || ast_strlen_zero(subtype)) {
return NULL;
}
return find_body_generator_type_subtype(type, subtype);
}
static struct ast_sip_pubsub_body_generator *find_body_generator(char accept[AST_SIP_MAX_ACCEPT][64],
size_t num_accept, const char *body_type)
{
int i;
struct ast_sip_pubsub_body_generator *generator = NULL;
for (i = 0; i < num_accept; ++i) {
generator = find_body_generator_accept(accept[i]);
if (generator) {
ast_debug(3, "Body generator %p found for accept type %s\n", generator, accept[i]);
if (strcmp(generator->body_type, body_type)) {
ast_log(LOG_WARNING, "Body generator '%s/%s'(%p) does not accept the type of data this event generates\n",
generator->type, generator->subtype, generator);
generator = NULL;
continue;
}
break;
} else {
ast_debug(3, "No body generator found for accept type %s\n", accept[i]);
}
}
return generator;
}
static int generate_initial_notify(struct ast_sip_subscription *sub)
{
void *notify_data;
int res;
struct ast_sip_body_data data = {
.body_type = sub->handler->body_type,
};
if (AST_VECTOR_SIZE(&sub->children) > 0) {
int i;
for (i = 0; i < AST_VECTOR_SIZE(&sub->children); ++i) {
if (generate_initial_notify(AST_VECTOR_GET(&sub->children, i))) {
return -1;
}
}
return 0;
}
/* We notify subscription establishment only on the tree leaves. */
if (sub->handler->notifier->subscription_established(sub)) {
return -1;
}
notify_data = sub->handler->notifier->get_notify_data(sub);
if (!notify_data) {
return -1;
}
data.body_data = notify_data;
res = ast_sip_pubsub_generate_body_content(ast_sip_subscription_get_body_type(sub),
ast_sip_subscription_get_body_subtype(sub), &data, &sub->body_text);
ao2_cleanup(notify_data);
return res;
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
static int pubsub_on_refresh_timeout(void *userdata);
static int initial_notify_task(void * obj)
{
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
struct initial_notify_data *ind = obj;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
if (generate_initial_notify(ind->sub_tree->root)) {
pjsip_evsub_terminate(ind->sub_tree->evsub, PJ_TRUE);
} else {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
send_notify(ind->sub_tree, 1);
ast_test_suite_event_notify("SUBSCRIPTION_ESTABLISHED",
"Resource: %s",
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ind->sub_tree->root->resource);
}
if (ind->expires != PJSIP_EXPIRES_NOT_SPECIFIED) {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
char *name = ast_alloca(strlen("->/ ") +
strlen(ind->sub_tree->persistence->endpoint) +
strlen(ind->sub_tree->root->resource) +
strlen(ind->sub_tree->root->handler->event_name) +
ind->sub_tree->dlg->call_id->id.slen + 1);
sprintf(name, "%s->%s/%s %.*s", ind->sub_tree->persistence->endpoint,
ind->sub_tree->root->resource, ind->sub_tree->root->handler->event_name,
(int)ind->sub_tree->dlg->call_id->id.slen, ind->sub_tree->dlg->call_id->id.ptr);
ast_debug(3, "Scheduling timer: %s\n", name);
ind->sub_tree->expiration_task = ast_sip_schedule_task(ind->sub_tree->serializer,
ind->expires * 1000, pubsub_on_refresh_timeout, name,
ind->sub_tree, AST_SIP_SCHED_TASK_FIXED | AST_SIP_SCHED_TASK_DATA_AO2);
if (!ind->sub_tree->expiration_task) {
ast_log(LOG_ERROR, "Unable to create expiration timer of %d seconds for %s\n",
ind->expires, name);
}
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ao2_ref(ind->sub_tree, -1);
ast_free(ind);
return 0;
}
static pj_bool_t pubsub_on_rx_subscribe_request(pjsip_rx_data *rdata)
{
pjsip_expires_hdr *expires_header;
struct ast_sip_subscription_handler *handler;
RAII_VAR(struct ast_sip_endpoint *, endpoint, NULL, ao2_cleanup);
struct sip_subscription_tree *sub_tree;
struct ast_sip_pubsub_body_generator *generator;
char *resource;
pjsip_uri *request_uri;
size_t resource_size;
int resp;
struct resource_tree tree;
pj_status_t dlg_status;
const pj_str_t *user;
endpoint = ast_pjsip_rdata_get_endpoint(rdata);
ast_assert(endpoint != NULL);
if (!endpoint->subscription.allow) {
ast_log(LOG_WARNING, "Subscriptions not permitted for endpoint %s.\n", ast_sorcery_object_get_id(endpoint));
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 603, NULL, NULL, NULL);
return PJ_TRUE;
}
request_uri = rdata->msg_info.msg->line.req.uri;
if (!ast_sip_is_uri_sip_sips(request_uri)) {
char uri_str[PJSIP_MAX_URL_SIZE];
pjsip_uri_print(PJSIP_URI_IN_REQ_URI, request_uri, uri_str, sizeof(uri_str));
ast_log(LOG_WARNING, "Request URI '%s' is not a sip: or sips: URI.\n", uri_str);
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 416, NULL, NULL, NULL);
return PJ_TRUE;
}
user = ast_sip_pjsip_uri_get_username(request_uri);
resource_size = pj_strlen(user) + 1;
resource = ast_alloca(resource_size);
ast_copy_pj_str(resource, user, resource_size);
/*
* We may want to match without any user options getting
* in the way.
*/
AST_SIP_USER_OPTIONS_TRUNCATE_CHECK(resource);
expires_header = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_EXPIRES, rdata->msg_info.msg->hdr.next);
if (expires_header) {
if (expires_header->ivalue == 0) {
ast_debug(1, "Subscription request from endpoint %s rejected. Expiration of 0 is invalid\n",
ast_sorcery_object_get_id(endpoint));
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 400, NULL, NULL, NULL);
return PJ_TRUE;
}
if (expires_header->ivalue < endpoint->subscription.minexpiry) {
ast_log(LOG_WARNING, "Subscription expiration %d is too brief for endpoint %s. Minimum is %u\n",
expires_header->ivalue, ast_sorcery_object_get_id(endpoint), endpoint->subscription.minexpiry);
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 423, NULL, NULL, NULL);
return PJ_TRUE;
}
}
handler = subscription_get_handler_from_rdata(rdata, ast_sorcery_object_get_id(endpoint));
if (!handler) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 489, NULL, NULL, NULL);
return PJ_TRUE;
}
generator = subscription_get_generator_from_rdata(rdata, handler);
if (!generator) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 489, NULL, NULL, NULL);
return PJ_TRUE;
}
memset(&tree, 0, sizeof(tree));
resp = build_resource_tree(endpoint, handler, resource, &tree,
ast_sip_pubsub_has_eventlist_support(rdata));
if (!PJSIP_IS_STATUS_IN_CLASS(resp, 200)) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, resp, NULL, NULL, NULL);
resource_tree_destroy(&tree);
return PJ_TRUE;
}
sub_tree = create_subscription_tree(handler, endpoint, rdata, resource, generator, &tree, &dlg_status, NULL);
if (!sub_tree) {
if (dlg_status != PJ_EEXISTS) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 500, NULL, NULL, NULL);
}
} else {
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
struct initial_notify_data *ind = ast_malloc(sizeof(*ind));
if (!ind) {
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
resource_tree_destroy(&tree);
return PJ_TRUE;
}
ind->sub_tree = ao2_bump(sub_tree);
/* Since this is a normal subscribe, pjproject takes care of the timer */
ind->expires = PJSIP_EXPIRES_NOT_SPECIFIED;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
sub_tree->persistence = subscription_persistence_create(sub_tree);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
subscription_persistence_update(sub_tree, rdata, SUBSCRIPTION_PERSISTENCE_CREATED);
sip_subscription_accept(sub_tree, rdata, resp);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
if (ast_sip_push_task(sub_tree->serializer, initial_notify_task, ind)) {
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
ao2_ref(sub_tree, -1);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_free(ind);
}
}
resource_tree_destroy(&tree);
return PJ_TRUE;
}
static struct ast_sip_publish_handler *find_pub_handler(const char *event)
{
struct ast_sip_publish_handler *iter = NULL;
AST_RWLIST_RDLOCK(&publish_handlers);
AST_RWLIST_TRAVERSE(&publish_handlers, iter, next) {
if (strcmp(event, iter->event_name)) {
ast_debug(3, "Event %s does not match %s\n", event, iter->event_name);
continue;
}
ast_debug(3, "Event name match: %s = %s\n", event, iter->event_name);
break;
}
AST_RWLIST_UNLOCK(&publish_handlers);
return iter;
}
static enum sip_publish_type determine_sip_publish_type(pjsip_rx_data *rdata,
pjsip_generic_string_hdr *etag_hdr, unsigned int *expires, int *entity_id)
{
pjsip_expires_hdr *expires_hdr = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_EXPIRES, NULL);
if (etag_hdr) {
char etag[pj_strlen(&etag_hdr->hvalue) + 1];
ast_copy_pj_str(etag, &etag_hdr->hvalue, sizeof(etag));
if (sscanf(etag, "%30d", entity_id) != 1) {
return SIP_PUBLISH_UNKNOWN;
}
}
*expires = expires_hdr ? expires_hdr->ivalue : DEFAULT_PUBLISH_EXPIRES;
if (!(*expires)) {
return SIP_PUBLISH_REMOVE;
} else if (!etag_hdr && rdata->msg_info.msg->body) {
return SIP_PUBLISH_INITIAL;
} else if (etag_hdr && !rdata->msg_info.msg->body) {
return SIP_PUBLISH_REFRESH;
} else if (etag_hdr && rdata->msg_info.msg->body) {
return SIP_PUBLISH_MODIFY;
}
return SIP_PUBLISH_UNKNOWN;
}
/*! \brief Internal destructor for publications */
static void publication_destroy_fn(void *obj)
{
struct ast_sip_publication *publication = obj;
ast_debug(3, "Destroying SIP publication\n");
ao2_cleanup(publication->datastores);
ao2_cleanup(publication->endpoint);
ast_module_unref(ast_module_info->self);
}
static struct ast_sip_publication *sip_create_publication(struct ast_sip_endpoint *endpoint, pjsip_rx_data *rdata,
const char *resource, const char *event_configuration_name)
{
struct ast_sip_publication *publication;
pjsip_expires_hdr *expires_hdr = pjsip_msg_find_hdr(rdata->msg_info.msg, PJSIP_H_EXPIRES, NULL);
size_t resource_len = strlen(resource) + 1, event_configuration_name_len = strlen(event_configuration_name) + 1;
char *dst;
ast_assert(endpoint != NULL);
if (!(publication = ao2_alloc(sizeof(*publication) + resource_len + event_configuration_name_len, publication_destroy_fn))) {
return NULL;
}
ast_module_ref(ast_module_info->self);
if (!(publication->datastores = ast_datastores_alloc())) {
ao2_ref(publication, -1);
return NULL;
}
publication->entity_tag = ast_atomic_fetchadd_int(&esc_etag_counter, +1);
ao2_ref(endpoint, +1);
publication->endpoint = endpoint;
publication->expires = expires_hdr ? expires_hdr->ivalue : DEFAULT_PUBLISH_EXPIRES;
publication->sched_id = -1;
dst = publication->data;
publication->resource = strcpy(dst, resource);
dst += resource_len;
publication->event_configuration_name = strcpy(dst, event_configuration_name);
return publication;
}
static int sip_publication_respond(struct ast_sip_publication *pub, int status_code,
pjsip_rx_data *rdata)
{
pjsip_tx_data *tdata;
pjsip_transaction *tsx;
if (pjsip_endpt_create_response(ast_sip_get_pjsip_endpoint(), rdata, status_code, NULL, &tdata) != PJ_SUCCESS) {
return -1;
}
if (PJSIP_IS_STATUS_IN_CLASS(status_code, 200)) {
char buf[30];
snprintf(buf, sizeof(buf), "%d", pub->entity_tag);
ast_sip_add_header(tdata, "SIP-ETag", buf);
snprintf(buf, sizeof(buf), "%d", pub->expires);
ast_sip_add_header(tdata, "Expires", buf);
}
if (pjsip_tsx_create_uas(&pubsub_module, rdata, &tsx) != PJ_SUCCESS) {
pjsip_tx_data_dec_ref(tdata);
return -1;
}
pjsip_tsx_recv_msg(tsx, rdata);
if (pjsip_tsx_send_msg(tsx, tdata) != PJ_SUCCESS) {
pjsip_tx_data_dec_ref(tdata);
return -1;
}
return 0;
}
static struct ast_sip_publication *publish_request_initial(struct ast_sip_endpoint *endpoint, pjsip_rx_data *rdata,
struct ast_sip_publish_handler *handler)
{
struct ast_sip_publication *publication;
char *resource_name;
size_t resource_size;
RAII_VAR(struct ast_sip_publication_resource *, resource, NULL, ao2_cleanup);
struct ast_variable *event_configuration_name = NULL;
pjsip_uri *request_uri;
int resp;
const pj_str_t *user;
request_uri = rdata->msg_info.msg->line.req.uri;
if (!ast_sip_is_uri_sip_sips(request_uri)) {
char uri_str[PJSIP_MAX_URL_SIZE];
pjsip_uri_print(PJSIP_URI_IN_REQ_URI, request_uri, uri_str, sizeof(uri_str));
ast_log(LOG_WARNING, "Request URI '%s' is not a sip: or sips: URI.\n", uri_str);
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 416, NULL, NULL, NULL);
return NULL;
}
user = ast_sip_pjsip_uri_get_username(request_uri);
resource_size = pj_strlen(user) + 1;
resource_name = ast_alloca(resource_size);
ast_copy_pj_str(resource_name, user, resource_size);
/*
* We may want to match without any user options getting
* in the way.
*/
AST_SIP_USER_OPTIONS_TRUNCATE_CHECK(resource_name);
resource = ast_sorcery_retrieve_by_id(ast_sip_get_sorcery(), "inbound-publication", resource_name);
if (!resource) {
ast_debug(1, "No 'inbound-publication' defined for resource '%s'\n", resource_name);
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 404, NULL, NULL, NULL);
return NULL;
}
if (!ast_strlen_zero(resource->endpoint) && strcmp(resource->endpoint, ast_sorcery_object_get_id(endpoint))) {
ast_debug(1, "Resource %s has a defined endpoint '%s', but does not match endpoint '%s' that received the request\n",
resource_name, resource->endpoint, ast_sorcery_object_get_id(endpoint));
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 403, NULL, NULL, NULL);
return NULL;
}
for (event_configuration_name = resource->events; event_configuration_name; event_configuration_name = event_configuration_name->next) {
if (!strcmp(event_configuration_name->name, handler->event_name)) {
break;
}
}
if (!event_configuration_name) {
ast_debug(1, "Event '%s' is not configured for '%s'\n", handler->event_name, resource_name);
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 404, NULL, NULL, NULL);
return NULL;
}
resp = handler->new_publication(endpoint, resource_name, event_configuration_name->value);
if (!PJSIP_IS_STATUS_IN_CLASS(resp, 200)) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, resp, NULL, NULL, NULL);
return NULL;
}
publication = sip_create_publication(endpoint, rdata, S_OR(resource_name, ""), event_configuration_name->value);
if (!publication) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 503, NULL, NULL, NULL);
return NULL;
}
publication->handler = handler;
if (publication->handler->publication_state_change(publication, rdata->msg_info.msg->body,
AST_SIP_PUBLISH_STATE_INITIALIZED)) {
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 500, NULL, NULL, NULL);
ao2_cleanup(publication);
return NULL;
}
sip_publication_respond(publication, resp, rdata);
return publication;
}
static int publish_expire_callback(void *data)
{
RAII_VAR(struct ast_sip_publication *, publication, data, ao2_cleanup);
if (publication->handler->publish_expire) {
publication->handler->publish_expire(publication);
}
return 0;
}
static int publish_expire(const void *data)
{
struct ast_sip_publication *publication = (struct ast_sip_publication*)data;
ao2_unlink(publication->handler->publications, publication);
publication->sched_id = -1;
if (ast_sip_push_task(NULL, publish_expire_callback, publication)) {
ao2_cleanup(publication);
}
return 0;
}
static pj_bool_t pubsub_on_rx_publish_request(pjsip_rx_data *rdata)
{
pjsip_event_hdr *event_header;
struct ast_sip_publish_handler *handler;
RAII_VAR(struct ast_sip_endpoint *, endpoint, NULL, ao2_cleanup);
char event[32];
static const pj_str_t str_sip_if_match = { "SIP-If-Match", 12 };
pjsip_generic_string_hdr *etag_hdr = pjsip_msg_find_hdr_by_name(rdata->msg_info.msg, &str_sip_if_match, NULL);
enum sip_publish_type publish_type;
RAII_VAR(struct ast_sip_publication *, publication, NULL, ao2_cleanup);
unsigned int expires = 0;
int entity_id, response = 0;
endpoint = ast_pjsip_rdata_get_endpoint(rdata);
ast_assert(endpoint != NULL);
event_header = pjsip_msg_find_hdr_by_name(rdata->msg_info.msg, &str_event_name, rdata->msg_info.msg->hdr.next);
if (!event_header) {
ast_log(LOG_WARNING, "Incoming PUBLISH request from %s with no Event header\n",
ast_sorcery_object_get_id(endpoint));
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 489, NULL, NULL, NULL);
return PJ_TRUE;
}
ast_copy_pj_str(event, &event_header->event_type, sizeof(event));
handler = find_pub_handler(event);
if (!handler) {
ast_log(LOG_WARNING, "No registered publish handler for event %s from %s\n", event,
ast_sorcery_object_get_id(endpoint));
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 489, NULL, NULL, NULL);
return PJ_TRUE;
}
publish_type = determine_sip_publish_type(rdata, etag_hdr, &expires, &entity_id);
/* If this is not an initial publish ensure that a publication is present */
if ((publish_type != SIP_PUBLISH_INITIAL) && (publish_type != SIP_PUBLISH_UNKNOWN)) {
if (!(publication = ao2_find(handler->publications, &entity_id, OBJ_KEY | OBJ_UNLINK))) {
static const pj_str_t str_conditional_request_failed = { "Conditional Request Failed", 26 };
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 412, &str_conditional_request_failed,
NULL, NULL);
return PJ_TRUE;
}
/* Per the RFC every response has to have a new entity tag */
publication->entity_tag = ast_atomic_fetchadd_int(&esc_etag_counter, +1);
/* Update the expires here so that the created responses will contain the correct value */
publication->expires = expires;
}
switch (publish_type) {
case SIP_PUBLISH_INITIAL:
publication = publish_request_initial(endpoint, rdata, handler);
break;
case SIP_PUBLISH_REFRESH:
case SIP_PUBLISH_MODIFY:
if (handler->publication_state_change(publication, rdata->msg_info.msg->body,
AST_SIP_PUBLISH_STATE_ACTIVE)) {
/* If an error occurs we want to terminate the publication */
expires = 0;
}
response = 200;
break;
case SIP_PUBLISH_REMOVE:
handler->publication_state_change(publication, rdata->msg_info.msg->body,
AST_SIP_PUBLISH_STATE_TERMINATED);
response = 200;
break;
case SIP_PUBLISH_UNKNOWN:
default:
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, 400, NULL, NULL, NULL);
break;
}
if (publication) {
if (expires) {
ao2_link(handler->publications, publication);
AST_SCHED_REPLACE_UNREF(publication->sched_id, sched, expires * 1000, publish_expire, publication,
ao2_ref(_data, -1), ao2_ref(publication, -1), ao2_ref(publication, +1));
} else {
AST_SCHED_DEL_UNREF(sched, publication->sched_id, ao2_ref(publication, -1));
}
}
if (response) {
sip_publication_respond(publication, response, rdata);
}
return PJ_TRUE;
}
struct ast_sip_endpoint *ast_sip_publication_get_endpoint(struct ast_sip_publication *pub)
{
return pub->endpoint;
}
const char *ast_sip_publication_get_resource(const struct ast_sip_publication *pub)
{
return pub->resource;
}
const char *ast_sip_publication_get_event_configuration(const struct ast_sip_publication *pub)
{
return pub->event_configuration_name;
}
int ast_sip_pubsub_is_body_generator_registered(const char *type, const char *subtype)
{
return !!find_body_generator_type_subtype(type, subtype);
}
int ast_sip_pubsub_register_body_generator(struct ast_sip_pubsub_body_generator *generator)
{
struct ast_sip_pubsub_body_generator *existing;
pj_str_t accept;
pj_size_t accept_len;
AST_RWLIST_WRLOCK(&body_generators);
existing = find_body_generator_type_subtype_nolock(generator->type, generator->subtype);
if (existing) {
AST_RWLIST_UNLOCK(&body_generators);
ast_log(LOG_WARNING, "A body generator for %s/%s is already registered.\n",
generator->type, generator->subtype);
return -1;
}
AST_LIST_INSERT_HEAD(&body_generators, generator, list);
AST_RWLIST_UNLOCK(&body_generators);
/* Lengths of type and subtype plus a slash. */
accept_len = strlen(generator->type) + strlen(generator->subtype) + 1;
/* Add room for null terminator that sprintf() will set. */
pj_strset(&accept, ast_alloca(accept_len + 1), accept_len);
sprintf((char *) pj_strbuf(&accept), "%s/%s", generator->type, generator->subtype);/* Safe */
pjsip_endpt_add_capability(ast_sip_get_pjsip_endpoint(), &pubsub_module,
PJSIP_H_ACCEPT, NULL, 1, &accept);
return 0;
}
void ast_sip_pubsub_unregister_body_generator(struct ast_sip_pubsub_body_generator *generator)
{
struct ast_sip_pubsub_body_generator *iter;
AST_RWLIST_WRLOCK(&body_generators);
AST_RWLIST_TRAVERSE_SAFE_BEGIN(&body_generators, iter, list) {
if (iter == generator) {
AST_LIST_REMOVE_CURRENT(list);
break;
}
}
AST_RWLIST_TRAVERSE_SAFE_END;
AST_RWLIST_UNLOCK(&body_generators);
}
int ast_sip_pubsub_register_body_supplement(struct ast_sip_pubsub_body_supplement *supplement)
{
AST_RWLIST_WRLOCK(&body_supplements);
AST_RWLIST_INSERT_TAIL(&body_supplements, supplement, list);
AST_RWLIST_UNLOCK(&body_supplements);
return 0;
}
void ast_sip_pubsub_unregister_body_supplement(struct ast_sip_pubsub_body_supplement *supplement)
{
struct ast_sip_pubsub_body_supplement *iter;
AST_RWLIST_WRLOCK(&body_supplements);
AST_RWLIST_TRAVERSE_SAFE_BEGIN(&body_supplements, iter, list) {
if (iter == supplement) {
AST_LIST_REMOVE_CURRENT(list);
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
break;
}
}
AST_RWLIST_TRAVERSE_SAFE_END;
AST_RWLIST_UNLOCK(&body_supplements);
}
const char *ast_sip_subscription_get_body_type(struct ast_sip_subscription *sub)
{
return sub->body_generator->type;
}
const char *ast_sip_subscription_get_body_subtype(struct ast_sip_subscription *sub)
{
return sub->body_generator->subtype;
}
int ast_sip_pubsub_generate_body_content(const char *type, const char *subtype,
struct ast_sip_body_data *data, struct ast_str **str)
{
struct ast_sip_pubsub_body_supplement *supplement;
struct ast_sip_pubsub_body_generator *generator;
int res = 0;
void *body;
generator = find_body_generator_type_subtype(type, subtype);
if (!generator) {
ast_log(LOG_WARNING, "Unable to find a body generator for %s/%s\n",
type, subtype);
return -1;
}
if (strcmp(data->body_type, generator->body_type)) {
ast_log(LOG_WARNING, "%s/%s body generator does not accept the type of data provided\n",
type, subtype);
return -1;
}
body = generator->allocate_body(data->body_data);
if (!body) {
ast_log(LOG_WARNING, "%s/%s body generator could not to allocate a body\n",
type, subtype);
return -1;
}
if (generator->generate_body_content(body, data->body_data)) {
res = -1;
goto end;
}
AST_RWLIST_RDLOCK(&body_supplements);
AST_RWLIST_TRAVERSE(&body_supplements, supplement, list) {
if (!strcmp(generator->type, supplement->type) &&
!strcmp(generator->subtype, supplement->subtype)) {
res = supplement->supplement_body(body, data->body_data);
if (res) {
break;
}
}
}
AST_RWLIST_UNLOCK(&body_supplements);
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
if (!res) {
generator->to_string(body, str);
}
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
end:
if (generator->destroy_body) {
generator->destroy_body(body);
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
return res;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
struct simple_message_summary {
int messages_waiting;
int voice_messages_new;
int voice_messages_old;
int voice_messages_urgent_new;
int voice_messages_urgent_old;
char message_account[PJSIP_MAX_URL_SIZE];
};
static int parse_simple_message_summary(char *body,
struct simple_message_summary *summary)
{
char *line;
char *buffer;
int found_counts = 0;
if (ast_strlen_zero(body) || !summary) {
return -1;
}
buffer = ast_strdupa(body);
memset(summary, 0, sizeof(*summary));
while ((line = ast_read_line_from_buffer(&buffer))) {
line = ast_str_to_lower(line);
if (sscanf(line, "voice-message: %d/%d (%d/%d)",
&summary->voice_messages_new, &summary->voice_messages_old,
&summary->voice_messages_urgent_new, &summary->voice_messages_urgent_old)) {
found_counts = 1;
} else {
sscanf(line, "message-account: %s", summary->message_account);
}
}
return !found_counts;
}
static pj_bool_t pubsub_on_rx_mwi_notify_request(pjsip_rx_data *rdata)
{
RAII_VAR(struct ast_sip_endpoint *, endpoint, NULL, ao2_cleanup);
struct simple_message_summary summary;
const char *endpoint_name;
char *atsign;
char *context;
char *body;
char *mailbox;
int rc;
endpoint = ast_pjsip_rdata_get_endpoint(rdata);
if (!endpoint) {
ast_debug(1, "Incoming MWI: Endpoint not found in rdata (%p)\n", rdata);
rc = 404;
goto error;
}
endpoint_name = ast_sorcery_object_get_id(endpoint);
ast_debug(1, "Incoming MWI: Found endpoint: %s\n", endpoint_name);
if (ast_strlen_zero(endpoint->incoming_mwi_mailbox)) {
ast_debug(1, "Incoming MWI: No incoming mailbox specified for endpoint '%s'\n", endpoint_name);
ast_test_suite_event_notify("PUBSUB_NO_INCOMING_MWI_MAILBOX",
"Endpoint: %s", endpoint_name);
rc = 404;
goto error;
}
mailbox = ast_strdupa(endpoint->incoming_mwi_mailbox);
atsign = strchr(mailbox, '@');
if (!atsign) {
ast_debug(1, "Incoming MWI: No '@' found in endpoint %s's incoming mailbox '%s'. Can't parse context\n",
endpoint_name, endpoint->incoming_mwi_mailbox);
rc = 404;
goto error;
}
*atsign = '\0';
context = atsign + 1;
body = ast_alloca(rdata->msg_info.msg->body->len + 1);
rdata->msg_info.msg->body->print_body(rdata->msg_info.msg->body, body,
rdata->msg_info.msg->body->len + 1);
if (parse_simple_message_summary(body, &summary) != 0) {
ast_debug(1, "Incoming MWI: Endpoint: '%s' There was an issue getting message info from body '%s'\n",
ast_sorcery_object_get_id(endpoint), body);
rc = 404;
goto error;
}
if (ast_publish_mwi_state(mailbox, context,
summary.voice_messages_new, summary.voice_messages_old)) {
ast_log(LOG_ERROR, "Incoming MWI: Endpoint: '%s' Could not publish MWI to stasis. "
"Mailbox: %s Message-Account: %s Voice-Messages: %d/%d (%d/%d)\n",
endpoint_name, endpoint->incoming_mwi_mailbox, summary.message_account,
summary.voice_messages_new, summary.voice_messages_old,
summary.voice_messages_urgent_new, summary.voice_messages_urgent_old);
rc = 404;
} else {
ast_debug(1, "Incoming MWI: Endpoint: '%s' Mailbox: %s Message-Account: %s Voice-Messages: %d/%d (%d/%d)\n",
endpoint_name, endpoint->incoming_mwi_mailbox, summary.message_account,
summary.voice_messages_new, summary.voice_messages_old,
summary.voice_messages_urgent_new, summary.voice_messages_urgent_old);
ast_test_suite_event_notify("PUBSUB_INCOMING_MWI_PUBLISH",
"Endpoint: %s\r\n"
"Mailbox: %s\r\n"
"MessageAccount: %s\r\n"
"VoiceMessagesNew: %d\r\n"
"VoiceMessagesOld: %d\r\n"
"VoiceMessagesUrgentNew: %d\r\n"
"VoiceMessagesUrgentOld: %d",
endpoint_name, endpoint->incoming_mwi_mailbox, summary.message_account,
summary.voice_messages_new, summary.voice_messages_old,
summary.voice_messages_urgent_new, summary.voice_messages_urgent_old);
rc = 200;
}
error:
pjsip_endpt_respond_stateless(ast_sip_get_pjsip_endpoint(), rdata, rc, NULL, NULL, NULL);
return PJ_TRUE;
}
static pj_bool_t pubsub_on_rx_notify_request(pjsip_rx_data *rdata)
{
if (rdata->msg_info.msg->body &&
ast_sip_is_content_type(&rdata->msg_info.msg->body->content_type,
"application", "simple-message-summary")) {
return pubsub_on_rx_mwi_notify_request(rdata);
}
return PJ_FALSE;
}
static pj_bool_t pubsub_on_rx_request(pjsip_rx_data *rdata)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
if (!pjsip_method_cmp(&rdata->msg_info.msg->line.req.method, pjsip_get_subscribe_method())) {
return pubsub_on_rx_subscribe_request(rdata);
} else if (!pjsip_method_cmp(&rdata->msg_info.msg->line.req.method, &pjsip_publish_method)) {
return pubsub_on_rx_publish_request(rdata);
} else if (!pjsip_method_cmp(&rdata->msg_info.msg->line.req.method, &pjsip_notify_method)) {
return pubsub_on_rx_notify_request(rdata);
}
return PJ_FALSE;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
static void set_state_terminated(struct ast_sip_subscription *sub)
{
int i;
sub->subscription_state = PJSIP_EVSUB_STATE_TERMINATED;
for (i = 0; i < AST_VECTOR_SIZE(&sub->children); ++i) {
set_state_terminated(AST_VECTOR_GET(&sub->children, i));
}
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
/*!
* \brief Callback sequence for subscription terminate:
*
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
* * Please note that the descriptions below represent pjproject behavior on versions
* >= 2.13.
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
* * Client initiated:
* pjproject receives SUBSCRIBE on the subscription's serializer thread
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
* calls pubsub_evsub_set_state with state = TERMINATED
* pubsub_on_evsub_state checks the event and finds it is due to a received
* SUBSCRIBE with an expires of 0 and so does nothing.
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
* calls pubsub_on_rx_refresh with dialog locked
* pubsub_on_rx_refresh sets TERMINATE_PENDING
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
* calls pubsub_on_refresh_timeout to push final NOTIFY to pjproject
* checks state == TERMINATE_PENDING
* sets TERMINATE_IN_PROGRESS
* calls send_notify (2)
* send_notify ultimately calls pjsip_evsub_send_request
* pjsip_evsub_send_request calls evsub's set_state
* set_state calls pubsub_evsub_set_state
* pubsub_on_evsub_state checks state == TERMINATE_IN_PROGRESS
* removes the subscriptions
* cleans up references to evsub
* sets state = TERMINATED
* pubsub_on_refresh_timeout unlocks dialog
* returns to pjproject
* pjproject unlocks dialog
*
* * Subscription timer expires:
* pjproject timer expires
* locks dialog
* calls pubsub_on_server_timeout
* pubsub_on_server_timeout checks state == NORMAL
* sets TERMINATE_PENDING
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
* pushes serialized_pubsub_on_refresh_timeout
* returns to pjproject
* pjproject unlocks dialog
* serialized_pubsub_on_refresh_timeout starts (1)
* locks dialog
* checks state == TERMINATE_PENDING
* sets TERMINATE_IN_PROGRESS
* calls send_notify (2)
* send_notify ultimately calls pjsip_evsub_send_request
* pjsip_evsub_send_request calls evsub's set_state
* set_state calls pubsub_evsub_set_state
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
* pubsub_on_evsub_state checks state == TERMINATE_IN_PROGRESS
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
* checks that the event is not due to un-SUBSCRIBE
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
* removes the subscriptions
* cleans up references to evsub
* sets state = TERMINATED
* serialized_pubsub_on_refresh_timeout unlocks dialog
*
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
* * Transmission failure sending NOTIFY or error response from client
* pjproject transaction timer expires or non OK response
* pjproject locks dialog
* calls pubsub_on_evsub_state with event TSX_STATE
* pubsub_on_evsub_state checks event == TSX_STATE
* removes the subscriptions
* cleans up references to evsub
* sets state = TERMINATED
* pjproject unlocks dialog
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
*
* * ast_sip_subscription_notify is called
* checks state == NORMAL
* if not batched...
* sets TERMINATE_IN_PROGRESS (if terminate is requested)
* calls send_notify
* See (2) Above
* if batched...
* sets TERMINATE_PENDING
* schedules task
* scheduler runs sched_task
* sched_task pushes serialized_send_notify
* serialized_send_notify starts
* checks state <= TERMINATE_PENDING
* if state == TERMINATE_PENDING set state = TERMINATE_IN_PROGRESS
* call send_notify
* See (2) Above
*
*/
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
/* The code in this function was previously in pubsub_on_evsub_state. */
static void clean_sub_tree(pjsip_evsub *evsub){
struct sip_subscription_tree *sub_tree;
sub_tree = pjsip_evsub_get_mod_data(evsub, pubsub_module.id);
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
ast_debug(3, "Cleaning subscription %p\n", evsub);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
if (sub_tree->expiration_task) {
char task_name[256];
ast_sip_sched_task_get_name(sub_tree->expiration_task, task_name, sizeof(task_name));
ast_debug(3, "Cancelling timer: %s\n", task_name);
ast_sip_sched_task_cancel(sub_tree->expiration_task);
ao2_cleanup(sub_tree->expiration_task);
sub_tree->expiration_task = NULL;
}
remove_subscription(sub_tree);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
pjsip_evsub_set_mod_data(evsub, pubsub_module.id, NULL);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
#ifdef HAVE_PJSIP_EVSUB_GRP_LOCK
pjsip_evsub_dec_ref(sub_tree->evsub);
#endif
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
sub_tree->evsub = NULL;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ast_sip_dialog_set_serializer(sub_tree->dlg, NULL);
ast_sip_dialog_set_endpoint(sub_tree->dlg, NULL);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
subscription_persistence_remove(sub_tree);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
shutdown_subscriptions(sub_tree->root);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
sub_tree->state = SIP_SUB_TREE_TERMINATED;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
/* Remove evsub's reference to the sub_tree */
ao2_ref(sub_tree, -1);
}
/* This functionality appeared in pjsip 2.13 */
#if PJ_VERSION_NUM >= 0x020D0000
# define HAVE_PJSIP_EVSUB_PENDING_NOTIFY 1
#endif
/*!
* \brief PJSIP callback when underlying SIP subscription changes state
*
* Although this function is called for every state change, we only care
* about the TERMINATED state, and only when we're actually processing the final
* notify (SIP_SUB_TREE_TERMINATE_IN_PROGRESS) OR when a transmission failure
* occurs (PJSIP_EVENT_TSX_STATE). In this case, we do all the subscription tree
* cleanup tasks and decrement the evsub reference.
*/
static void pubsub_on_evsub_state(pjsip_evsub *evsub, pjsip_event *event)
{
struct sip_subscription_tree *sub_tree =
pjsip_evsub_get_mod_data(evsub, pubsub_module.id);
ast_debug(3, "evsub %p state %s event %s sub_tree %p sub_tree state %s\n", evsub,
pjsip_evsub_get_state_name(evsub), pjsip_event_str(event->type), sub_tree,
(sub_tree ? sub_tree_state_description[sub_tree->state] : "UNKNOWN"));
if (!sub_tree || pjsip_evsub_get_state(evsub) != PJSIP_EVSUB_STATE_TERMINATED) {
return;
}
/* It's easier to write this as what we WANT to process, then negate it. */
if (!(sub_tree->state == SIP_SUB_TREE_TERMINATE_IN_PROGRESS
|| (event->type == PJSIP_EVENT_TSX_STATE && sub_tree->state == SIP_SUB_TREE_NORMAL)
)) {
ast_debug(3, "Do nothing.\n");
return;
}
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
#ifdef HAVE_PJSIP_EVSUB_PENDING_NOTIFY
/* This check looks for re-subscribes with an expires of 0. If we receive one of those,
we don't want to clean the evsub because we still need it to send the final NOTIFY.
This was previously handled by pubsub_on_rx_refresh setting:
'sub_tree->state = SIP_SUB_TREE_TERMINATE_PENDING' */
if (event->body.tsx_state.type == PJSIP_EVENT_RX_MSG &&
!pjsip_method_cmp(&event->body.tsx_state.tsx->method, &pjsip_subscribe_method) &&
pjsip_evsub_get_expires(evsub) == 0) {
ast_debug(3, "Subscription ending, do nothing.\n");
return;
}
#endif
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
/* If we made it this far, we want to clean the sub tree. For pjproject <2.13, the sub_tree
state check makes sure the evsub is not cleaned at the wrong time */
clean_sub_tree(evsub);
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
static int pubsub_on_refresh_timeout(void *userdata)
{
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
struct sip_subscription_tree *sub_tree = userdata;
pjsip_dialog *dlg = sub_tree->dlg;
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "sub_tree %p sub_tree state %s\n", sub_tree,
(sub_tree ? sub_tree_state_description[sub_tree->state] : "UNKNOWN"));
pjsip_dlg_inc_lock(dlg);
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (sub_tree->state >= SIP_SUB_TREE_TERMINATE_IN_PROGRESS) {
pjsip_dlg_dec_lock(dlg);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
return 0;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (sub_tree->state == SIP_SUB_TREE_TERMINATE_PENDING) {
sub_tree->state = SIP_SUB_TREE_TERMINATE_IN_PROGRESS;
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
set_state_terminated(sub_tree->root);
}
if (sub_tree->generate_initial_notify) {
sub_tree->generate_initial_notify = 0;
if (generate_initial_notify(sub_tree->root)) {
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
pjsip_dlg_dec_lock(dlg);
return 0;
}
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
send_notify(sub_tree, 1);
ast_test_suite_event_notify(sub_tree->root->subscription_state == PJSIP_EVSUB_STATE_TERMINATED ?
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
"SUBSCRIPTION_TERMINATED" : "SUBSCRIPTION_REFRESHED",
"Resource: %s", sub_tree->root->resource);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
pjsip_dlg_dec_lock(dlg);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
return 0;
}
static int serialized_pubsub_on_refresh_timeout(void *userdata)
{
struct sip_subscription_tree *sub_tree = userdata;
ast_debug(3, "sub_tree %p sub_tree state %s\n", sub_tree,
(sub_tree ? sub_tree_state_description[sub_tree->state] : "UNKNOWN"));
pubsub_on_refresh_timeout(userdata);
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
ao2_cleanup(sub_tree);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
return 0;
}
/*!
* \brief Compare strings for equality checking for NULL.
*
* This function considers NULL values as empty strings.
* This means NULL or empty strings are equal.
*
* \retval 0 The strings are equal
* \retval 1 The strings are not equal
*/
static int cmp_strings(char *s1, char *s2)
{
if (!ast_strlen_zero(s1) && !ast_strlen_zero(s2)) {
return strcmp(s1, s2);
}
return ast_strlen_zero(s1) == ast_strlen_zero(s2) ? 0 : 1;
}
/*!
* \brief compares the childrens of two ast_sip_subscription s1 and s2
*
* \retval 0 The s1 childrens match the s2 childrens
* \retval 1 The s1 childrens do not match the s2 childrens
*/
static int cmp_subscription_childrens(struct ast_sip_subscription *s1, struct ast_sip_subscription *s2)
{
int i;
if (AST_VECTOR_SIZE(&s1->children) != AST_VECTOR_SIZE(&s2->children)) {
return 1;
}
for (i = 0; i < AST_VECTOR_SIZE(&s1->children); ++i) {
struct ast_sip_subscription *c1 = AST_VECTOR_GET(&s1->children, i);
struct ast_sip_subscription *c2 = AST_VECTOR_GET(&s2->children, i);
if (cmp_strings(c1->resource, c2->resource)
|| cmp_strings(c1->display_name, c2->display_name)) {
return 1;
}
}
return 0;
}
static int destroy_subscriptions_task(void *obj)
{
struct ast_sip_subscription *sub = (struct ast_sip_subscription *) obj;
destroy_subscriptions(sub);
return 0;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
/*!
* \brief Called whenever an in-dialog SUBSCRIBE is received
*
* This includes both SUBSCRIBE requests that actually refresh the subscription
* as well as SUBSCRIBE requests that end the subscription.
*
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
* In either case we push an appropriate NOTIFY via pubsub_on_refresh_timeout.
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
*/
static void pubsub_on_rx_refresh(pjsip_evsub *evsub, pjsip_rx_data *rdata,
int *p_st_code, pj_str_t **p_st_text, pjsip_hdr *res_hdr, pjsip_msg_body **p_body)
{
struct sip_subscription_tree *sub_tree;
sub_tree = pjsip_evsub_get_mod_data(evsub, pubsub_module.id);
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
ast_debug(3, "evsub %p sub_tree %p sub_tree state %s\n", evsub, sub_tree,
(sub_tree ? sub_tree_state_description[sub_tree->state] : "UNKNOWN"));
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (!sub_tree || sub_tree->state != SIP_SUB_TREE_NORMAL) {
return;
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
if (sub_tree->expiration_task) {
char task_name[256];
ast_sip_sched_task_get_name(sub_tree->expiration_task, task_name, sizeof(task_name));
ast_debug(3, "Cancelling timer: %s\n", task_name);
ast_sip_sched_task_cancel(sub_tree->expiration_task);
ao2_cleanup(sub_tree->expiration_task);
sub_tree->expiration_task = NULL;
}
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
/* PJSIP will set the evsub's state to terminated before calling into this function
* if the Expires value of the incoming SUBSCRIBE is 0.
*/
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (pjsip_evsub_get_state(sub_tree->evsub) == PJSIP_EVSUB_STATE_TERMINATED) {
sub_tree->state = SIP_SUB_TREE_TERMINATE_PENDING;
}
if (sub_tree->state == SIP_SUB_TREE_NORMAL && sub_tree->is_list) {
/* update RLS */
const char *resource = sub_tree->root->resource;
struct ast_sip_subscription *old_root = sub_tree->root;
struct ast_sip_subscription *new_root = NULL;
RAII_VAR(struct ast_sip_endpoint *, endpoint, NULL, ao2_cleanup);
struct ast_sip_subscription_handler *handler = NULL;
struct ast_sip_pubsub_body_generator *generator = NULL;
if ((endpoint = ast_pjsip_rdata_get_endpoint(rdata))
&& (handler = subscription_get_handler_from_rdata(rdata, ast_sorcery_object_get_id(endpoint)))
&& (generator = subscription_get_generator_from_rdata(rdata, handler))) {
struct resource_tree tree;
int resp;
memset(&tree, 0, sizeof(tree));
resp = build_resource_tree(endpoint, handler, resource, &tree,
ast_sip_pubsub_has_eventlist_support(rdata));
if (PJSIP_IS_STATUS_IN_CLASS(resp, 200)) {
new_root = create_virtual_subscriptions(handler, resource, generator, sub_tree, tree.root);
if (new_root) {
if (cmp_subscription_childrens(old_root, new_root)) {
ast_debug(1, "RLS '%s->%s' was modified, regenerate it\n", ast_sorcery_object_get_id(endpoint), old_root->resource);
new_root->version = old_root->version;
sub_tree->root = new_root;
sub_tree->generate_initial_notify = 1;
/* If there is scheduled notification need to delete it to avoid use old subscriptions */
if (sub_tree->notify_sched_id > -1) {
AST_SCHED_DEL_UNREF(sched, sub_tree->notify_sched_id, ao2_ref(sub_tree, -1));
sub_tree->send_scheduled_notify = 0;
}
/* Terminate old subscriptions to stop sending NOTIFY messages on exten/device state changes */
set_state_terminated(old_root);
/* Shutdown old subscriptions to remove exten/device state change callbacks
that can queue tasks for old subscriptions */
shutdown_subscriptions(old_root);
/* Postpone destruction until all already queued tasks that may be using old subscriptions have completed */
if (ast_sip_push_task(sub_tree->serializer, destroy_subscriptions_task, old_root)) {
ast_log(LOG_ERROR, "Failed to push task to destroy old subscriptions for RLS '%s->%s'.\n",
ast_sorcery_object_get_id(endpoint), old_root->resource);
}
} else {
destroy_subscriptions(new_root);
}
}
} else {
sub_tree->state = SIP_SUB_TREE_TERMINATE_PENDING;
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
}
resource_tree_destroy(&tree);
}
}
res_pjsip_pubsub: Correctly implement persisted subscriptions This patch fixes 2 original issues and more that those 2 exposed. * When we send a NOTIFY, and the client either doesn't respond or responds with a non OK, pjproject only calls our pubsub_on_evsub_state callback, no others. Since pubsub_on_evsub_state (which does the sub_tree cleanup) does not expect to be called back without the other callbacks being called first, it just returns leaving the sub_tree orphaned. Now pubsub_on_evsub_state checks the event for PJSIP_EVENT_TSX_STATE which is what pjproject will set to tell us that it was the transaction that timed out or failed and not the subscription itself timing our or being terminated by the client. If is TSX_STATE, pubsub_on_evsub_state now does the proper cleanup regardless of the state of the subscription. * When a client renews a subscription, we don't update the persisted subscription with the new expires timestamp. This causes subscription_persistence_recreate to prune the subscription if/when asterisk restarts. Now, pubsub_on_rx_refresh calls subscription_persistence_update to apply the new expires timestamp. This exposed other issues however... * When creating a dialog from rdata (which sub_persistence_recreate does from the packet buffer) there must NOT be a tag on the To header (which there will be when a client refreshes a subscription). If there is one, pjsip_dlg_create_uas will fail. To address this, subscription_persistence_update now accepts a flag that indicates that the original packet buffer must not be updated. New subscribes don't set the flag and renews do. This makes sure that when the rdata is recreated on asterisk startup, it's done from the original subscribe packet which won't have the tag on To. * When creating a dialog from rdata, we were setting the dialog's remote (SUBSCRIBE) cseq to be the same as the local (NOTIFY) cseq. When the client tried to resubscribe after a restart with the correct cseq, we'd reject the request with an Invalid CSeq error. * The acts of creating a dialog and evsub by themselves when recreating a subscription does NOT restart pjproject's subscription timer. The result was that even if we did correctly recreate the subscription, we never removed it if the client happened to go away or send a non-OK response to a NOTIFY. However, there is no pjproject function exposed to just set the timer on an evsub that wasn't created by an incoming subscribe request. To address this, we create our own timer using ast_sip_schedule_task. This timer is used only for re-establishing subscriptions after a restart. An earlier approach was to add support for setting pjproject's timer (via a pjproject patch) and while that patch is still included here, we don't use that call at the moment. While addressing these issues, additional debugging was added and some existing messages made more useful. A few formatting changes were also made to 'pjsip show scheduled tasks' to make displaying the subscription timers a little more friendly. ASTERISK-26696 ASTERISK-26756 Change-Id: I8c605fc1e3923f466a74db087d5ab6f90abce68e
2017-02-07 19:17:12 +00:00
subscription_persistence_update(sub_tree, rdata, SUBSCRIPTION_PERSISTENCE_REFRESHED);
#ifdef HAVE_PJSIP_EVSUB_PENDING_NOTIFY
/* As of pjsip 2.13, the NOTIFY has to be sent within this function as pjproject now
requires it. Previously this would have caused an early NOTIFY to go out before the
SUBSCRIBE's 200 OK. The previous solution was to push the NOTIFY, but now pjproject
res_pjsip_pubsub: subscription cleanup changes There are two main parts of the change associated with this commit. These are driven by the change in call order of pubsub_on_rx_refresh and pubsub_on_evsub_state by pjproject when an in-dialog SUBSCRIBE is received. First, the previous behavior was for pjproject to call pubsub_on_rx_refresh before calling pubsub_on_evsub_state when an in-dialog SUBSCRIBE was received that changes the subscription state. If that change was a termination due to a re-SUBSCRIBE with an expires of 0, we used to use the call to pubsub_on_rx_refresh to set the substate of the evsub to TERMINATE_PENDING before pjproject could call pubsub_on_evsub_state. This substate let pubsub_on_evsub_state know that the subscription TERMINATED event could be ignored as there was still a subsequent NOTIFY that needed to be generated and another call to pubsub_on_evsub_state to come with it. That NOTIFY was sent via serialized_pubsub_on_refresh_timeout which would see the TERMINATE_PENDING state and transition it to TERMINATE_IN_PROGRESS before triggering another call to pubsub_on_evsub_state (which now would clean up the evsub.) The new pjproject behavior is to call pubsub_on_evsub_state before pubsub_on_rx_refresh. This means we no longer can set the state to TERMINATE_PENDING to tell pubsub_on_evsub_state that it can ignore the first TERMINATED event. To handle this, we now look directly at the event type, method type and the expires value to determine whether we want to ignore the event or use it to trigger the evsub cleanup. Second, pjproject now expects the NOTIFY to actually be sent during pubsub_on_rx_refresh and avoids the protocol violation inherent in sending a NOTIFY before the SUBSCRIBE is acknowledged by caching the sent NOTIFY then sending it after responding to the SUBSCRIBE. This requires we send the NOTIFY using the non-serialized pubsub_on_refresh_timeout directly and let pjproject handle the protocol violation. ASTERISK-30469 Change-Id: I05c1d91a44fe28244ae93faa4a2268a3332b5fd7
2023-03-29 18:49:51 +00:00
looks for the NOTIFY to be sent from this function and caches it to send after it
auto-replies to the SUBSCRIBE. */
pubsub_on_refresh_timeout(sub_tree);
#else
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (ast_sip_push_task(sub_tree->serializer, serialized_pubsub_on_refresh_timeout, ao2_bump(sub_tree))) {
/* If we can't push the NOTIFY refreshing task...we'll just go with it. */
ast_log(LOG_ERROR, "Failed to push task to send NOTIFY.\n");
sub_tree->state = SIP_SUB_TREE_NORMAL;
ao2_ref(sub_tree, -1);
}
#endif
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
if (sub_tree->is_list) {
pj_list_insert_before(res_hdr, create_require_eventlist(rdata->tp_info.pool));
}
}
static void pubsub_on_rx_notify(pjsip_evsub *evsub, pjsip_rx_data *rdata, int *p_st_code,
pj_str_t **p_st_text, pjsip_hdr *res_hdr, pjsip_msg_body **p_body)
{
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
struct ast_sip_subscription *sub;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (!(sub = pjsip_evsub_get_mod_data(evsub, pubsub_module.id))) {
return;
}
sub->handler->subscriber->state_change(sub, rdata->msg_info.msg->body,
pjsip_evsub_get_state(evsub));
}
static int serialized_pubsub_on_client_refresh(void *userdata)
{
struct sip_subscription_tree *sub_tree = userdata;
pjsip_tx_data *tdata;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (!sub_tree->evsub) {
ao2_cleanup(sub_tree);
return 0;
}
if (pjsip_evsub_initiate(sub_tree->evsub, NULL, -1, &tdata) == PJ_SUCCESS) {
pjsip_evsub_send_request(sub_tree->evsub, tdata);
} else {
pjsip_evsub_terminate(sub_tree->evsub, PJ_TRUE);
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
ao2_cleanup(sub_tree);
return 0;
}
static void pubsub_on_client_refresh(pjsip_evsub *evsub)
{
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
struct sip_subscription_tree *sub_tree;
if (!(sub_tree = pjsip_evsub_get_mod_data(evsub, pubsub_module.id))) {
return;
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
if (ast_sip_push_task(sub_tree->serializer, serialized_pubsub_on_client_refresh, ao2_bump(sub_tree))) {
ao2_cleanup(sub_tree);
}
}
static void pubsub_on_server_timeout(pjsip_evsub *evsub)
{
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
struct sip_subscription_tree *sub_tree;
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
/* PJSIP does not terminate the server timeout timer when a SUBSCRIBE
* with Expires: 0 arrives to end a subscription, nor does it terminate
* this timer when we send a NOTIFY request in response to receiving such
* a SUBSCRIBE. PJSIP does not stop the server timeout timer until the
* NOTIFY transaction has finished (either through receiving a response
* or through a transaction timeout).
*
* Therefore, it is possible that we can be told that a server timeout
* occurred after we already thought that the subscription had been
* terminated. In such a case, we will have already removed the sub_tree
* from the evsub's mod_data array.
*/
sub_tree = pjsip_evsub_get_mod_data(evsub, pubsub_module.id);
if (!sub_tree || sub_tree->state != SIP_SUB_TREE_NORMAL) {
res_pjsip_pubsub: Solidify lifetime and ownership of objects. There have been crashes and general instability seen in the pubsub code, so this patch introduces three changes to increase the stability. First, the ownership model for subscriptions has been modified. Due to RLS, subscriptions are stored in memory as a tree structure. Prior to my patch, the PJSIP subscription was the owner of the subscription tree. When the PJSIP subscription told us that it was terminating, we started destroying the subscription tree along with all of the individual leaf subscriptions that belong to the tree. The problem with this model is that the two actors in play here, the PJSIP subscription and the individual leaf subscriptions, need to have joint ownership of the subscription tree. So now, the PJSIP subscription and the individual leaf subscriptions each have a reference to the subscription tree. This way, we will not actually free memory until no players are left that care. The PJSIP subscription is a bigger stakeholder, in that if the PJSIP subscription's reference to the subscription tree is removed, the subscription tree instructs the leaf subscriptions to shut down and drop their references to the subscription tree when possible. The individual leaf subscriptions, upon being told to shut down, can drop their stasis subscriptions or whatever they use to learn of new state, and then drop their reference to the subscription tree once they are ready to die. Second, the lifetime of a PJSIP subscription's reference to our subscription tree has been altered. As I learned from doing a deep dive, the PJSIP evsub code can tell Asterisk multiple times that the subscription has been terminated, and not all of these times are especially helpful. I have altered the message flow that we use for SIP subscriptions such that we will always drop the PJSIP subscription's reference to the subscription tree when we send the NOTIFY that terminates a SIP subscription. This also means that we will now queue NOTIFY requests to be sent after responding to incoming SUBSCRIBEs so that we can have predictable state changes from the PJSIP evsub code. Third, the synchronization of operations has been improved. PJSIP can call into our code from a serializer thread (e.g. upon receiving an incoming request) or from the monitor thread (e.g. when a subscription times out). Because of this, there is the possibility of competing threads stepping on each other. PJSIP attempts to do some synchronization on its own by always keeping the dialog lock held when it calls into us. However, since we end up pushing tasks into the serializer, the result was that serialized operations were not grabbing the dialog lock and could, as a result, step on something that was being attempted by a different thread. Now we ensure that serialized operations grab the dialog lock, then check for extenuating circumstances, then proceed with their operation if they can. Change-Id: Iff2990c40178dad9cc5f6a5c7f76932ec644b2e5
2015-09-01 20:47:19 +00:00
return;
}
res_pjsip_pubsub: Address SEGV when attempting to terminate a subscription Occasionally under load we'll attempt to send a final NOTIFY on a subscription that's already been terminated and a SEGV will occur down in pjproject's evsub_destroy function. This is a result of a race condition between all the paths that can generate a notify and/or destroy the underlying pjproject evsub object: * The client can send a SUBSCRIBE with Expires: 0. * The client can send a SUBSCRIBE/refresh. * The subscription timer can expire. * An extension state can change. * An MWI event can be generated. * The pjproject transaction timer (timer_b) can expire. Normally when our pubsub_on_evsub_state is called with a terminate, we push a task to the serializer and return at which point the dialog is unlocked. This is usually not a problem because the task runs immediately and locks the dialog again. When the system is heavily loaded though, there may be a delay between the unlock and relock during which another event may occur such as the subscription timer or timer_b expiring, an extension state change, etc. These may also cause a terminate to be processed and if so, we could cause pjproject to try to destroy the evsub structure twice. There's no way for us to tell that the evsub was already destroyed and the evsub's group lock can't tolerate this and SEGVs. The remedy is twofold. * A patch has been submitted to Teluu and added to the bundled pjproject which adds add/decrement operations on evsub's group lock. * In res_pjsip_pubsub: * configure.ac and pjproject-bundled's configure.m4 were updated to check for the new evsub group lock APIs. * We now add a reference to the evsub group lock when we create the subscription and remove the reference when we clean up the subscription. This prevents evsub from being destroyed before we're done with it. * A state has been added to the subscription tree structure so termination progress can be tracked through the asyncronous tasks. * The pubsub_on_evsub_state callback has been split so it's not doing double duty. It now only handles the final cleanup of the subscription tree. pubsub_on_rx_refresh now handles both client refreshes and client terminates. It was always being called for both anyway. * The serialized_on_server_timeout task was removed since serialized_pubsub_on_rx_refresh was almost identical. * Missing state checks and ao2_cleanups were added. * Some debug levels were adjusted to make seeing only off-nominal things at level 1 and nominal or progress things at level 2+. ASTERISK-26099 #close Reported-by: Ross Beer. Change-Id: I779d11802cf672a51392e62a74a1216596075ba1
2016-06-12 16:19:27 +00:00
sub_tree->state = SIP_SUB_TREE_TERMINATE_PENDING;
if (ast_sip_push_task(sub_tree->serializer, serialized_pubsub_on_refresh_timeout, ao2_bump(sub_tree))) {
sub_tree->state = SIP_SUB_TREE_NORMAL;
ao2_cleanup(sub_tree);
}
}
static int ami_subscription_detail(struct sip_subscription_tree *sub_tree,
struct ast_sip_ami *ami,
const char *event)
{
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
struct ast_str *buf;
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
buf = ast_sip_create_ami_event(event, ami);
if (!buf) {
return -1;
}
sip_subscription_to_ami(sub_tree, &buf);
astman_append(ami->s, "%s\r\n", ast_str_buffer(buf));
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
ast_free(buf);
++ami->count;
return 0;
}
static int ami_subscription_detail_inbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_NOTIFIER ? ami_subscription_detail(
sub_tree, arg, "InboundSubscriptionDetail") : 0;
}
static int ami_subscription_detail_outbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_SUBSCRIBER ? ami_subscription_detail(
sub_tree, arg, "OutboundSubscriptionDetail") : 0;
}
static int ami_show_subscriptions_inbound(struct mansession *s, const struct message *m)
{
struct ast_sip_ami ami = { .s = s, .m = m, .action_id = astman_get_header(m, "ActionID"), };
astman_send_listack(s, m, "Following are Events for each inbound Subscription",
"start");
for_each_subscription(ami_subscription_detail_inbound, &ami);
astman_send_list_complete_start(s, m, "InboundSubscriptionDetailComplete", ami.count);
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
astman_send_list_complete_end(s);
return 0;
}
static int ami_show_subscriptions_outbound(struct mansession *s, const struct message *m)
{
struct ast_sip_ami ami = { .s = s, .m = m, .action_id = astman_get_header(m, "ActionID"), };
astman_send_listack(s, m, "Following are Events for each outbound Subscription",
"start");
for_each_subscription(ami_subscription_detail_outbound, &ami);
astman_send_list_complete_start(s, m, "OutboundSubscriptionDetailComplete", ami.count);
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
astman_send_list_complete_end(s);
return 0;
}
static int format_ami_resource_lists(void *obj, void *arg, int flags)
{
struct resource_list *list = obj;
struct ast_sip_ami *ami = arg;
struct ast_str *buf;
buf = ast_sip_create_ami_event("ResourceListDetail", ami);
if (!buf) {
return CMP_STOP;
}
if (ast_sip_sorcery_object_to_ami(list, &buf)) {
ast_free(buf);
return CMP_STOP;
}
astman_append(ami->s, "%s\r\n", ast_str_buffer(buf));
ast_free(buf);
++ami->count;
return 0;
}
static int ami_show_resource_lists(struct mansession *s, const struct message *m)
{
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
struct ast_sip_ami ami = { .s = s, .m = m, .action_id = astman_get_header(m, "ActionID"), };
struct ao2_container *lists;
lists = ast_sorcery_retrieve_by_fields(ast_sip_get_sorcery(), "resource_list",
AST_RETRIEVE_FLAG_MULTIPLE | AST_RETRIEVE_FLAG_ALL, NULL);
if (!lists || !ao2_container_count(lists)) {
astman_send_error(s, m, "No resource lists found\n");
return 0;
}
astman_send_listack(s, m, "A listing of resource lists follows, presented as ResourceListDetail events",
"start");
ao2_callback(lists, OBJ_NODATA, format_ami_resource_lists, &ami);
astman_send_list_complete_start(s, m, "ResourceListDetailComplete", ami.count);
AMI: Make AMI actions that generate event lists consistent. * Made the following AMI actions use list API calls for consistency: Agents BridgeInfo BridgeList BridgeTechnologyList ConfbridgeLIst ConfbridgeLIstRooms CoreShowChannels DAHDIShowChannels DBGet DeviceStateList ExtensionStateList FAXSessions Hangup IAXpeerlist IAXpeers IAXregistry MeetmeList MeetmeListRooms MWIGet ParkedCalls Parkinglots PJSIPShowEndpoint PJSIPShowEndpoints PJSIPShowRegistrationsInbound PJSIPShowRegistrationsOutbound PJSIPShowResourceLists PJSIPShowSubscriptionsInbound PJSIPShowSubscriptionsOutbound PresenceStateList PRIShowSpans QueueStatus QueueSummary ShowDialPlan SIPpeers SIPpeerstatus SIPshowregistry SKINNYdevices SKINNYlines Status VoicemailUsersList * Incremented the AMI version to 2.7.0. * Changed astman_send_listack() to not use the listflag parameter and always set the value to "Start" so the start capitalization is consistent. i.e., The FAXSessions used "Start" while the rest of the system used "start". The corresponding complete event always used "Complete". * Fixed ami_show_resource_lists() "PJSIPShowResourceLists" to output the AMI ActionID for all of its list events. * Fixed off-nominal AMI protocol error in manager_bridge_info(), manager_parking_status_single_lot(), and manager_parking_status_all_lots(). Use of astman_send_error() after responding to the original AMI action request violates the action response pattern by sending two responses. * Fixed minor protocol error in action_getconfig() when no requested categories are found. Each line needs to be formatted as "Header: text". * Fixed off-nominal memory leak in manager_build_parked_call_string(). * Eliminated unnecessary use of RAII_VAR() in ami_subscription_detail(). ASTERISK-24049 #close Reported by: Jonathan Rose Review: https://reviewboard.asterisk.org/r/4315/ ........ Merged revisions 430434 from http://svn.asterisk.org/svn/asterisk/branches/13 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@430435 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2015-01-09 18:16:54 +00:00
astman_send_list_complete_end(s);
return 0;
}
#define AMI_SHOW_SUBSCRIPTIONS_INBOUND "PJSIPShowSubscriptionsInbound"
#define AMI_SHOW_SUBSCRIPTIONS_OUTBOUND "PJSIPShowSubscriptionsOutbound"
#define MAX_REGEX_ERROR_LEN 128
struct cli_sub_parms {
/*! CLI handler entry e parameter */
struct ast_cli_entry *e;
/*! CLI handler entry a parameter */
struct ast_cli_args *a;
/*! CLI subscription entry output line(s) */
struct ast_str *buf;
/*! Compiled regular expression to select if buf is written to CLI when not NULL. */
regex_t *like;
int count;
};
struct cli_sub_complete_parms {
struct ast_cli_args *a;
/*! Found callid for search position */
char *callid;
int wordlen;
int which;
};
static int cli_complete_subscription_common(struct sip_subscription_tree *sub_tree, struct cli_sub_complete_parms *cli)
{
pj_str_t *callid;
if (!sub_tree->dlg) {
return 0;
}
callid = &sub_tree->dlg->call_id->id;
if (cli->wordlen <= pj_strlen(callid)
&& !strncasecmp(cli->a->word, pj_strbuf(callid), cli->wordlen)
&& (++cli->which > cli->a->n)) {
cli->callid = ast_malloc(pj_strlen(callid) + 1);
if (cli->callid) {
ast_copy_pj_str(cli->callid, callid, pj_strlen(callid) + 1);
}
return -1;
}
return 0;
}
static int cli_complete_subscription_inbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_NOTIFIER
? cli_complete_subscription_common(sub_tree, arg) : 0;
}
static int cli_complete_subscription_outbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_SUBSCRIBER
? cli_complete_subscription_common(sub_tree, arg) : 0;
}
static char *cli_complete_subscription_callid(struct ast_cli_args *a)
{
struct cli_sub_complete_parms cli;
on_subscription_t on_subscription;
if (a->pos != 4) {
return NULL;
}
if (!strcasecmp(a->argv[3], "inbound")) {
on_subscription = cli_complete_subscription_inbound;
} else if (!strcasecmp(a->argv[3], "outbound")) {
on_subscription = cli_complete_subscription_outbound;
} else {
/* Should never get here */
ast_assert(0);
return NULL;
}
cli.a = a;
cli.callid = NULL;
cli.wordlen = strlen(a->word);
cli.which = 0;
for_each_subscription(on_subscription, &cli);
return cli.callid;
}
static unsigned int cli_subscription_expiry(struct sip_subscription_tree *sub_tree)
{
int expiry;
expiry = sub_tree->persistence
? ast_tvdiff_ms(sub_tree->persistence->expires, ast_tvnow()) / 1000
: 0;
if (expiry < 0) {
/* Subscription expired */
expiry = 0;
}
return expiry;
}
static int cli_show_subscription_common(struct sip_subscription_tree *sub_tree, struct cli_sub_parms *cli)
{
const char *callid = (const char *) cli->buf;/* Member repurposed to pass in callid */
pj_str_t *sub_callid;
struct ast_str *buf;
char *src;
char *dest;
char *key;
char *value;
char *value_end;
int key_len;
int key_filler_width;
int value_len;
if (!sub_tree->dlg) {
return 0;
}
sub_callid = &sub_tree->dlg->call_id->id;
if (pj_strcmp2(sub_callid, callid)) {
return 0;
}
buf = ast_str_create(512);
if (!buf) {
return -1;
}
ast_cli(cli->a->fd,
"%-20s: %s\n"
"===========================================================================\n",
"ParameterName", "ParameterValue");
ast_str_append(&buf, 0, "Resource: %s\n", sub_tree->root->resource);
ast_str_append(&buf, 0, "Event: %s\n", sub_tree->root->handler->event_name);
ast_str_append(&buf, 0, "Expiry: %u\n", cli_subscription_expiry(sub_tree));
sip_subscription_to_ami(sub_tree, &buf);
/* Convert AMI \r\n to \n line terminators. */
src = strchr(ast_str_buffer(buf), '\r');
if (src) {
dest = src;
++src;
while (*src) {
if (*src == '\r') {
++src;
continue;
}
*dest++ = *src++;
}
*dest = '\0';
ast_str_update(buf);
}
/* Reformat AMI key value pairs to pretty columns */
key = ast_str_buffer(buf);
do {
value = strchr(key, ':');
if (!value) {
break;
}
value_end = strchr(value, '\n');
if (!value_end) {
break;
}
/* Calculate field lengths */
key_len = value - key;
key_filler_width = 20 - key_len;
if (key_filler_width < 0) {
key_filler_width = 0;
}
value_len = value_end - value;
ast_cli(cli->a->fd, "%.*s%*s%.*s\n",
key_len, key, key_filler_width, "",
value_len, value);
key = value_end + 1;
} while (*key);
ast_cli(cli->a->fd, "\n");
ast_free(buf);
return -1;
}
static int cli_show_subscription_inbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_NOTIFIER
? cli_show_subscription_common(sub_tree, arg) : 0;
}
static int cli_show_subscription_outbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_SUBSCRIBER
? cli_show_subscription_common(sub_tree, arg) : 0;
}
static char *cli_show_subscription_inout(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
{
on_subscription_t on_subscription;
struct cli_sub_parms cli;
switch (cmd) {
case CLI_INIT:
e->command = "pjsip show subscription {inbound|outbound}";
e->usage = "Usage:\n"
" pjsip show subscription inbound <call-id>\n"
" pjsip show subscription outbound <call-id>\n"
" Show active subscription with the dialog call-id\n";
return NULL;
case CLI_GENERATE:
return cli_complete_subscription_callid(a);
}
if (a->argc != 5) {
return CLI_SHOWUSAGE;
}
if (!strcasecmp(a->argv[3], "inbound")) {
on_subscription = cli_show_subscription_inbound;
} else if (!strcasecmp(a->argv[3], "outbound")) {
on_subscription = cli_show_subscription_outbound;
} else {
/* Should never get here */
ast_assert(0);
return NULL;
}
/* Find the subscription with the specified call-id */
cli.a = a;
cli.e = e;
cli.buf = (void *) a->argv[4];/* Repurpose the buf member to pass in callid */
for_each_subscription(on_subscription, &cli);
return CLI_SUCCESS;
}
#define CLI_SHOW_SUB_FORMAT_HEADER \
"Endpoint: <Endpoint/Caller-ID.............................................>\n" \
"Resource: <Resource/Event.................................................>\n" \
" Expiry: <Expiry> <Call-id..............................................>\n" \
"===========================================================================\n\n"
#define CLI_SHOW_SUB_FORMAT_ENTRY \
"Endpoint: %s/%s\n" \
"Resource: %s/%s\n" \
" Expiry: %8d %s\n\n"
static int cli_show_subscriptions_detail(struct sip_subscription_tree *sub_tree, struct cli_sub_parms *cli)
{
char caller_id[256];
char callid[256];
ast_callerid_merge(caller_id, sizeof(caller_id),
S_COR(sub_tree->endpoint->id.self.name.valid,
sub_tree->endpoint->id.self.name.str, NULL),
S_COR(sub_tree->endpoint->id.self.number.valid,
sub_tree->endpoint->id.self.number.str, NULL),
"<none>");
/* Call-id */
if (sub_tree->dlg) {
ast_copy_pj_str(callid, &sub_tree->dlg->call_id->id, sizeof(callid));
} else {
ast_copy_string(callid, "<unknown>", sizeof(callid));
}
ast_str_set(&cli->buf, 0, CLI_SHOW_SUB_FORMAT_ENTRY,
ast_sorcery_object_get_id(sub_tree->endpoint), caller_id,
sub_tree->root->resource, sub_tree->root->handler->event_name,
cli_subscription_expiry(sub_tree), callid);
if (cli->like) {
if (regexec(cli->like, ast_str_buffer(cli->buf), 0, NULL, 0)) {
/* Output line did not match the regex */
return 0;
}
}
ast_cli(cli->a->fd, "%s", ast_str_buffer(cli->buf));
++cli->count;
return 0;
}
static int cli_show_subscriptions_inbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_NOTIFIER
? cli_show_subscriptions_detail(sub_tree, arg) : 0;
}
static int cli_show_subscriptions_outbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_SUBSCRIBER
? cli_show_subscriptions_detail(sub_tree, arg) : 0;
}
static char *cli_show_subscriptions_inout(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
{
on_subscription_t on_subscription;
struct cli_sub_parms cli;
regex_t like;
const char *regex;
switch (cmd) {
case CLI_INIT:
e->command = "pjsip show subscriptions {inbound|outbound} [like]";
e->usage = "Usage:\n"
" pjsip show subscriptions inbound [like <regex>]\n"
" Show active inbound subscriptions\n"
" pjsip show subscriptions outbound [like <regex>]\n"
" Show active outbound subscriptions\n"
"\n"
" The regex selects a subscriptions output that matches.\n"
" i.e., All output lines for a subscription are checked\n"
" as a block by the regex.\n";
return NULL;
case CLI_GENERATE:
return NULL;
}
if (a->argc != 4 && a->argc != 6) {
return CLI_SHOWUSAGE;
}
if (!strcasecmp(a->argv[3], "inbound")) {
on_subscription = cli_show_subscriptions_inbound;
} else if (!strcasecmp(a->argv[3], "outbound")) {
on_subscription = cli_show_subscriptions_outbound;
} else {
/* Should never get here */
ast_assert(0);
return CLI_SHOWUSAGE;
}
if (a->argc == 6) {
int rc;
if (strcasecmp(a->argv[4], "like")) {
return CLI_SHOWUSAGE;
}
/* Setup regular expression */
memset(&like, 0, sizeof(like));
cli.like = &like;
regex = a->argv[5];
rc = regcomp(cli.like, regex, REG_EXTENDED | REG_NOSUB);
if (rc) {
char *regerr = ast_alloca(MAX_REGEX_ERROR_LEN);
regerror(rc, cli.like, regerr, MAX_REGEX_ERROR_LEN);
ast_cli(a->fd, "Regular expression '%s' failed to compile: %s\n",
regex, regerr);
return CLI_FAILURE;
}
} else {
cli.like = NULL;
regex = NULL;
}
cli.a = a;
cli.e = e;
cli.count = 0;
cli.buf = ast_str_create(256);
if (!cli.buf) {
if (cli.like) {
regfree(cli.like);
}
return CLI_FAILURE;
}
ast_cli(a->fd, CLI_SHOW_SUB_FORMAT_HEADER);
for_each_subscription(on_subscription, &cli);
ast_cli(a->fd, "%d active subscriptions%s%s%s\n",
cli.count,
regex ? " matched \"" : "",
regex ?: "",
regex ? "\"" : "");
ast_free(cli.buf);
if (cli.like) {
regfree(cli.like);
}
return CLI_SUCCESS;
}
#define CLI_LIST_SUB_FORMAT_HEADER "%-30.30s %-30.30s %6.6s %s\n"
#define CLI_LIST_SUB_FORMAT_ENTRY "%-30.30s %-30.30s %6d %s\n"
static int cli_list_subscriptions_detail(struct sip_subscription_tree *sub_tree, struct cli_sub_parms *cli)
{
char ep_cid_buf[50];
char res_evt_buf[50];
char callid[256];
/* Endpoint/CID column */
snprintf(ep_cid_buf, sizeof(ep_cid_buf), "%s/%s",
ast_sorcery_object_get_id(sub_tree->endpoint),
S_COR(sub_tree->endpoint->id.self.name.valid, sub_tree->endpoint->id.self.name.str,
S_COR(sub_tree->endpoint->id.self.number.valid,
sub_tree->endpoint->id.self.number.str, "<none>")));
/* Resource/Event column */
snprintf(res_evt_buf, sizeof(res_evt_buf), "%s/%s",
sub_tree->root->resource,
sub_tree->root->handler->event_name);
/* Call-id column */
if (sub_tree->dlg) {
ast_copy_pj_str(callid, &sub_tree->dlg->call_id->id, sizeof(callid));
} else {
ast_copy_string(callid, "<unknown>", sizeof(callid));
}
ast_str_set(&cli->buf, 0, CLI_LIST_SUB_FORMAT_ENTRY,
ep_cid_buf,
res_evt_buf,
cli_subscription_expiry(sub_tree),
callid);
if (cli->like) {
if (regexec(cli->like, ast_str_buffer(cli->buf), 0, NULL, 0)) {
/* Output line did not match the regex */
return 0;
}
}
ast_cli(cli->a->fd, "%s", ast_str_buffer(cli->buf));
++cli->count;
return 0;
}
static int cli_list_subscriptions_inbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_NOTIFIER
? cli_list_subscriptions_detail(sub_tree, arg) : 0;
}
static int cli_list_subscriptions_outbound(struct sip_subscription_tree *sub_tree, void *arg)
{
return sub_tree->role == AST_SIP_SUBSCRIBER
? cli_list_subscriptions_detail(sub_tree, arg) : 0;
}
static char *cli_list_subscriptions_inout(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
{
on_subscription_t on_subscription;
struct cli_sub_parms cli;
regex_t like;
const char *regex;
switch (cmd) {
case CLI_INIT:
e->command = "pjsip list subscriptions {inbound|outbound} [like]";
e->usage = "Usage:\n"
" pjsip list subscriptions inbound [like <regex>]\n"
" List active inbound subscriptions\n"
" pjsip list subscriptions outbound [like <regex>]\n"
" List active outbound subscriptions\n"
"\n"
" The regex selects output lines that match.\n";
return NULL;
case CLI_GENERATE:
return NULL;
}
if (a->argc != 4 && a->argc != 6) {
return CLI_SHOWUSAGE;
}
if (!strcasecmp(a->argv[3], "inbound")) {
on_subscription = cli_list_subscriptions_inbound;
} else if (!strcasecmp(a->argv[3], "outbound")) {
on_subscription = cli_list_subscriptions_outbound;
} else {
/* Should never get here */
ast_assert(0);
return CLI_SHOWUSAGE;
}
if (a->argc == 6) {
int rc;
if (strcasecmp(a->argv[4], "like")) {
return CLI_SHOWUSAGE;
}
/* Setup regular expression */
memset(&like, 0, sizeof(like));
cli.like = &like;
regex = a->argv[5];
rc = regcomp(cli.like, regex, REG_EXTENDED | REG_NOSUB);
if (rc) {
char *regerr = ast_alloca(MAX_REGEX_ERROR_LEN);
regerror(rc, cli.like, regerr, MAX_REGEX_ERROR_LEN);
ast_cli(a->fd, "Regular expression '%s' failed to compile: %s\n",
regex, regerr);
return CLI_FAILURE;
}
} else {
cli.like = NULL;
regex = NULL;
}
cli.a = a;
cli.e = e;
cli.count = 0;
cli.buf = ast_str_create(256);
if (!cli.buf) {
if (cli.like) {
regfree(cli.like);
}
return CLI_FAILURE;
}
ast_cli(a->fd, CLI_LIST_SUB_FORMAT_HEADER,
"Endpoint/CLI", "Resource/Event", "Expiry", "Call-id");
for_each_subscription(on_subscription, &cli);
ast_cli(a->fd, "\n%d active subscriptions%s%s%s\n",
cli.count,
regex ? " matched \"" : "",
regex ?: "",
regex ? "\"" : "");
ast_free(cli.buf);
if (cli.like) {
regfree(cli.like);
}
return CLI_SUCCESS;
}
static struct ast_cli_entry cli_commands[] = {
AST_CLI_DEFINE(cli_list_subscriptions_inout, "List active inbound/outbound subscriptions"),
AST_CLI_DEFINE(cli_show_subscription_inout, "Show active subscription details"),
AST_CLI_DEFINE(cli_show_subscriptions_inout, "Show active inbound/outbound subscriptions"),
};
static int persistence_endpoint_str2struct(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct subscription_persistence *persistence = obj;
persistence->endpoint = ast_strdup(var->value);
return 0;
}
static int persistence_endpoint_struct2str(const void *obj, const intptr_t *args, char **buf)
{
const struct subscription_persistence *persistence = obj;
*buf = ast_strdup(persistence->endpoint);
return 0;
}
static int persistence_tag_str2struct(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct subscription_persistence *persistence = obj;
persistence->tag = ast_strdup(var->value);
return 0;
}
static int persistence_tag_struct2str(const void *obj, const intptr_t *args, char **buf)
{
const struct subscription_persistence *persistence = obj;
*buf = ast_strdup(persistence->tag);
return 0;
}
static int persistence_generator_data_str2struct(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct subscription_persistence *persistence = obj;
struct ast_json_error error;
/* We tolerate a failure of the JSON to load and instead start fresh, since this field
* originates from the persistence code and not a user.
*/
persistence->generator_data = ast_json_load_string(var->value, &error);
return 0;
}
static int persistence_generator_data_struct2str(const void *obj, const intptr_t *args, char **buf)
{
const struct subscription_persistence *persistence = obj;
char *value;
if (!persistence->generator_data) {
return 0;
}
value = ast_json_dump_string(persistence->generator_data);
if (!value) {
return -1;
}
*buf = ast_strdup(value);
ast_json_free(value);
return 0;
}
static int persistence_expires_str2struct(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct subscription_persistence *persistence = obj;
return ast_get_timeval(var->value, &persistence->expires, ast_tv(0, 0), NULL);
}
static int persistence_expires_struct2str(const void *obj, const intptr_t *args, char **buf)
{
const struct subscription_persistence *persistence = obj;
char secs[AST_TIME_T_LEN];
ast_time_t_to_string(persistence->expires.tv_sec, secs, sizeof(secs));
return (ast_asprintf(buf, "%s", secs) < 0) ? -1 : 0;
}
#define RESOURCE_LIST_INIT_SIZE 4
static void resource_list_destructor(void *obj)
{
struct resource_list *list = obj;
int i;
for (i = 0; i < AST_VECTOR_SIZE(&list->items); ++i) {
ast_free((char *) AST_VECTOR_GET(&list->items, i));
}
AST_VECTOR_FREE(&list->items);
}
static void *resource_list_alloc(const char *name)
{
struct resource_list *list;
list = ast_sorcery_generic_alloc(sizeof(*list), resource_list_destructor);
if (!list) {
return NULL;
}
if (AST_VECTOR_INIT(&list->items, RESOURCE_LIST_INIT_SIZE)) {
ao2_cleanup(list);
return NULL;
}
return list;
}
static int item_in_vector(const struct resource_list *list, const char *item)
{
int i;
for (i = 0; i < AST_VECTOR_SIZE(&list->items); ++i) {
if (!strcmp(item, AST_VECTOR_GET(&list->items, i))) {
return 1;
}
}
return 0;
}
static int list_item_handler(const struct aco_option *opt,
struct ast_variable *var, void *obj)
{
struct resource_list *list = obj;
char *items = ast_strdupa(var->value);
char *item;
while ((item = ast_strip(strsep(&items, ",")))) {
if (ast_strlen_zero(item)) {
continue;
}
if (item_in_vector(list, item)) {
ast_log(LOG_WARNING, "Ignoring duplicated list item '%s'\n", item);
continue;
}
item = ast_strdup(item);
if (!item || AST_VECTOR_APPEND(&list->items, item)) {
ast_free(item);
return -1;
}
}
return 0;
}
static int list_item_to_str(const void *obj, const intptr_t *args, char **buf)
{
const struct resource_list *list = obj;
int i;
struct ast_str *str = ast_str_create(32);
for (i = 0; i < AST_VECTOR_SIZE(&list->items); ++i) {
ast_str_append(&str, 0, "%s,", AST_VECTOR_GET(&list->items, i));
}
/* Chop off trailing comma */
ast_str_truncate(str, -1);
*buf = ast_strdup(ast_str_buffer(str));
ast_free(str);
return 0;
}
static int resource_list_apply_handler(const struct ast_sorcery *sorcery, void *obj)
{
struct resource_list *list = obj;
if (ast_strlen_zero(list->event)) {
ast_log(LOG_WARNING, "Resource list '%s' has no event set\n",
ast_sorcery_object_get_id(list));
return -1;
}
if (AST_VECTOR_SIZE(&list->items) == 0) {
ast_log(LOG_WARNING, "Resource list '%s' has no list items\n",
ast_sorcery_object_get_id(list));
return -1;
}
return 0;
}
static int apply_list_configuration(struct ast_sorcery *sorcery)
{
ast_sorcery_apply_default(sorcery, "resource_list", "config",
"pjsip.conf,criteria=type=resource_list");
if (ast_sorcery_object_register(sorcery, "resource_list", resource_list_alloc,
NULL, resource_list_apply_handler)) {
return -1;
}
ast_sorcery_object_field_register(sorcery, "resource_list", "type", "",
OPT_NOOP_T, 0, 0);
ast_sorcery_object_field_register(sorcery, "resource_list", "event", "",
OPT_CHAR_ARRAY_T, 1, CHARFLDSET(struct resource_list, event));
ast_sorcery_object_field_register(sorcery, "resource_list", "full_state", "no",
OPT_BOOL_T, 1, FLDSET(struct resource_list, full_state));
ast_sorcery_object_field_register(sorcery, "resource_list", "notification_batch_interval",
"0", OPT_UINT_T, 0, FLDSET(struct resource_list, notification_batch_interval));
ast_sorcery_object_field_register_custom(sorcery, "resource_list", "list_item",
"", list_item_handler, list_item_to_str, NULL, 0, 0);
ast_sorcery_object_field_register(sorcery, "resource_list", "resource_display_name", "no",
OPT_BOOL_T, 1, FLDSET(struct resource_list, resource_display_name));
ast_sorcery_reload_object(sorcery, "resource_list");
return 0;
}
#ifdef TEST_FRAMEWORK
/*!
* \brief "bad" resources
*
* These are resources that the test handler will reject subscriptions to.
*/
const char *bad_resources[] = {
"coconut",
"cilantro",
"olive",
"cheese",
};
/*!
* \brief new_subscribe callback for unit tests
*
* Will give a 200 OK response to any resource except the "bad" ones.
*/
static int test_new_subscribe(struct ast_sip_endpoint *endpoint, const char *resource)
{
int i;
for (i = 0; i < ARRAY_LEN(bad_resources); ++i) {
if (!strcmp(resource, bad_resources[i])) {
return 400;
}
}
return 200;
}
/*!
* \brief Subscription notifier for unit tests.
*
* Since unit tests are only concerned with building a resource tree,
* only the new_subscribe callback needs to be defined.
*/
struct ast_sip_notifier test_notifier = {
.new_subscribe = test_new_subscribe,
};
/*!
* \brief Subscription handler for unit tests.
*/
struct ast_sip_subscription_handler test_handler = {
.event_name = "test",
.notifier = &test_notifier,
};
/*!
* \brief Set properties on an allocated resource list
*
* \param list The list to set details on.
* \param event The list's event.
* \param resources Array of resources to add to the list.
* \param num_resources Number of resources in the array.
* \retval 0 Success
* \retval non-zero Failure
*/
static int populate_list(struct resource_list *list, const char *event, const char **resources, size_t num_resources)
{
int i;
ast_copy_string(list->event, event, sizeof(list->event));
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
for (i = 0; i < num_resources; ++i) {
char *resource = ast_strdup(resources[i]);
if (!resource || AST_VECTOR_APPEND(&list->items, resource)) {
ast_free(resource);
return -1;
}
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
return 0;
}
/*!
* \brief RAII callback to destroy a resource list
*/
static void cleanup_resource_list(struct resource_list *list)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
if (!list) {
return;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
ast_sorcery_delete(ast_sip_get_sorcery(), list);
ao2_cleanup(list);
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
/*!
* \brief allocate a resource list, store it in sorcery, and set its details
*
* \param test The unit test. Used for logging status messages.
* \param list_name The name of the list to create.
* \param event The event the list services
* \param resources Array of resources to apply to the list
* \param num_resources The number of resources in the array
* \retval NULL Failed to allocate or populate list
* \retval non-NULL The created list
*/
static struct resource_list *create_resource_list(struct ast_test *test,
const char *list_name, const char *event, const char **resources, size_t num_resources)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
struct resource_list *list;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
list = ast_sorcery_alloc(ast_sip_get_sorcery(), "resource_list", list_name);
if (!list) {
ast_test_status_update(test, "Could not allocate resource list in sorcery\n");
return NULL;
}
if (ast_sorcery_create(ast_sip_get_sorcery(), list)) {
ast_test_status_update(test, "Could not store the resource list in sorcery\n");
ao2_cleanup(list);
return NULL;
}
if (populate_list(list, event, resources, num_resources)) {
ast_test_status_update(test, "Could not add resources to the resource list\n");
cleanup_resource_list(list);
return NULL;
}
return list;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
/*!
* \brief Check the integrity of a tree node against a set of resources.
*
* The tree node's resources must be in the same order as the resources in
* the supplied resources array. Because of this constraint, tests can misrepresent
* the size of the resources array as being smaller than it really is if resources
* at the end of the array should not be present in the tree node.
*
* \param test The unit test. Used for printing status messages.
* \param node The constructed tree node whose integrity is under question.
* \param resources Array of expected resource values
* \param num_resources The number of resources to check in the array.
*/
static int check_node(struct ast_test *test, struct tree_node *node,
const char **resources, size_t num_resources)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
int i;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
if (AST_VECTOR_SIZE(&node->children) != num_resources) {
ast_test_status_update(test, "Unexpected number of resources in tree. Expected %zu, got %zu\n",
num_resources, AST_VECTOR_SIZE(&node->children));
return -1;
}
for (i = 0; i < num_resources; ++i) {
if (strcmp(resources[i], AST_VECTOR_GET(&node->children, i)->resource)) {
ast_test_status_update(test, "Mismatched resources. Expected '%s' but got '%s'\n",
resources[i], AST_VECTOR_GET(&node->children, i)->resource);
return -1;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
}
return 0;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
/*!
* \brief RAII_VAR callback to destroy an allocated resource tree
*/
static void test_resource_tree_destroy(struct resource_tree *tree)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
resource_tree_destroy(tree);
ast_free(tree);
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
static int ineligible_configuration(void)
{
struct ast_config *config;
struct ast_flags flags = {0,};
const char *value;
config = ast_config_load("sorcery.conf", flags);
if (!config) {
return 1;
}
value = ast_variable_retrieve(config, "res_pjsip_pubsub", "resource_list");
if (ast_strlen_zero(value)) {
ast_config_destroy(config);
return 1;
}
if (strcasecmp(value, "memory") && strcasecmp(value, "astdb")) {
ast_config_destroy(config);
return 1;
}
return 0;
}
AST_TEST_DEFINE(resource_tree)
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
{
RAII_VAR(struct resource_list *, list, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources[] = {
"huey",
"dewey",
"louie",
};
int resp;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
switch (cmd) {
case TEST_INIT:
info->name = "resource_tree";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Basic resource tree integrity check";
info->description =
"Create a resource list and ensure that our attempt to build a tree works as expected.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list = create_resource_list(test, "foo", "test", resources, ARRAY_LEN(resources));
if (!list) {
return AST_TEST_FAIL;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
if (check_node(test, tree->root, resources, ARRAY_LEN(resources))) {
return AST_TEST_FAIL;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
return AST_TEST_PASS;
Decouple subscription handling from NOTIFY/PUBLISH body generation. When the PJSIP pubsub framework was created, subscription handlers were required to state what event they handled along with what body types they knew how to generate. While this serves well when implementing a base RFC, it has problems when trying to extend the body to support non-standard or proprietary body elements. The code also was NOTIFY-specific, meaning that when the time comes that we start writing code to send out PUBLISH requests with MWI or presence bodies, we would likely find ourselves duplicating code that had previously been written. This changeset introduces the concept of body generators and body supplements. A body generator is responsible for allocating a native structure for a given body type, providing the primary body content, converting the native structure to a string, and deallocating resources. A body supplement takes the primary body content (the native structure, not a string) generated by the body generator and adds nonstandard elements to the body. With these elements living in their own module, it becomes easy to extend our support for body types and to re-use resources when sending a PUBLISH request. Body generators and body supplements register themselves with the pubsub core, similar to how subscription and publish handlers had done. Now, subscription handlers do not need to know what type of body content they generate, but they still need to inform the pubsub core about what the default body type for a given event package is. The pubsub core keeps track of what body generators and body supplements have been registered. When a SUBSCRIBE arrives, the pubsub core will check that there is a subscription handler for the event in the SUBSCRIBE, then it will check that there is a body generator that can provide the content specified in the Accept header(s). Because of the nature of body generators and supplements, it means res_pjsip_exten_state and res_pjsip_mwi have been completely gutted. They no longer worry about body types, instead calling ast_sip_pubsub_generate_body_content() when they need to generate a NOTIFY body. Review: https://reviewboard.asterisk.org/r/3150 ........ Merged revisions 407016 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407030 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2014-01-31 22:27:07 +00:00
}
AST_TEST_DEFINE(complex_resource_tree)
{
RAII_VAR(struct resource_list *, list_1, NULL, cleanup_resource_list);
RAII_VAR(struct resource_list *, list_2, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources_1[] = {
"huey",
"dewey",
"louie",
"dwarves",
};
const char *resources_2[] = {
"happy",
"grumpy",
"doc",
"bashful",
"dopey",
"sneezy",
"sleepy",
};
int resp;
struct tree_node *node;
switch (cmd) {
case TEST_INIT:
info->name = "complex_resource_tree";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Complex resource tree integrity check";
info->description =
"Create a complex resource list and ensure that our attempt to build a tree works as expected.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list_1 = create_resource_list(test, "foo", "test", resources_1, ARRAY_LEN(resources_1));
if (!list_1) {
return AST_TEST_FAIL;
}
list_2 = create_resource_list(test, "dwarves", "test", resources_2, ARRAY_LEN(resources_2));
if (!list_2) {
return AST_TEST_FAIL;
}
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
}
node = tree->root;
if (check_node(test, node, resources_1, ARRAY_LEN(resources_1))) {
return AST_TEST_FAIL;
}
/* The embedded list is at index 3 in the root node's children */
node = AST_VECTOR_GET(&node->children, 3);
if (check_node(test, node, resources_2, ARRAY_LEN(resources_2))) {
return AST_TEST_FAIL;
}
return AST_TEST_PASS;
}
AST_TEST_DEFINE(bad_resource)
{
RAII_VAR(struct resource_list *, list, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources[] = {
"huey",
"dewey",
"louie",
"coconut", /* A "bad" resource */
};
int resp;
switch (cmd) {
case TEST_INIT:
info->name = "bad_resource";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Ensure bad resources do not end up in the tree";
info->description =
"Create a resource list with a single bad resource. Ensure the bad resource does not end up in the tree.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list = create_resource_list(test, "foo", "test", resources, ARRAY_LEN(resources));
if (!list) {
return AST_TEST_FAIL;
}
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
}
/* We check against all but the final resource since we expect it not to be in the tree */
if (check_node(test, tree->root, resources, ARRAY_LEN(resources) - 1)) {
return AST_TEST_FAIL;
}
return AST_TEST_PASS;
}
AST_TEST_DEFINE(bad_branch)
{
RAII_VAR(struct resource_list *, list_1, NULL, cleanup_resource_list);
RAII_VAR(struct resource_list *, list_2, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources_1[] = {
"huey",
"dewey",
"louie",
"gross",
};
/* This list has nothing but bad resources */
const char *resources_2[] = {
"coconut",
"cilantro",
"olive",
"cheese",
};
int resp;
switch (cmd) {
case TEST_INIT:
info->name = "bad_branch";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Ensure bad branches are pruned from the tree";
info->description =
"Create a resource list that makes a tree with an entire branch of bad resources.\n"
"Ensure the bad branch is pruned from the tree.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list_1 = create_resource_list(test, "foo", "test", resources_1, ARRAY_LEN(resources_1));
if (!list_1) {
return AST_TEST_FAIL;
}
list_2 = create_resource_list(test, "gross", "test", resources_2, ARRAY_LEN(resources_2));
if (!list_2) {
return AST_TEST_FAIL;
}
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
/* We check against all but the final resource of the list since the entire branch should
* be pruned from the tree
*/
if (check_node(test, tree->root, resources_1, ARRAY_LEN(resources_1) - 1)) {
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
return AST_TEST_PASS;
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
}
AST_TEST_DEFINE(duplicate_resource)
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
{
RAII_VAR(struct resource_list *, list_1, NULL, cleanup_resource_list);
RAII_VAR(struct resource_list *, list_2, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources_1[] = {
"huey",
"ducks",
"dewey",
"louie",
};
const char *resources_2[] = {
"donald",
"daisy",
"scrooge",
"dewey",
"louie",
"huey",
};
int resp;
struct tree_node *node;
switch (cmd) {
case TEST_INIT:
info->name = "duplicate_resource";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Ensure duplicated resources do not end up in the tree";
info->description =
"Create a resource list with a single duplicated resource. Ensure the duplicated resource does not end up in the tree.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list_1 = create_resource_list(test, "foo", "test", resources_1, ARRAY_LEN(resources_1));
if (!list_1) {
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
list_2 = create_resource_list(test, "ducks", "test", resources_2, ARRAY_LEN(resources_2));
if (!list_2) {
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
node = tree->root;
/* This node should have "huey" and "ducks". "dewey" and "louie" should not
* be present since they were found in the "ducks" list.
*/
if (check_node(test, node, resources_1, ARRAY_LEN(resources_1) - 2)) {
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
/* This node should have "donald", "daisy", "scrooge", "dewey", and "louie".
* "huey" is not here since that was already encountered in the parent list
*/
node = AST_VECTOR_GET(&node->children, 1);
if (check_node(test, node, resources_2, ARRAY_LEN(resources_2) - 1)) {
return AST_TEST_FAIL;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
return AST_TEST_PASS;
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
AST_TEST_DEFINE(loop)
{
RAII_VAR(struct resource_list *, list_1, NULL, cleanup_resource_list);
RAII_VAR(struct resource_list *, list_2, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources_1[] = {
"derp",
};
const char *resources_2[] = {
"herp",
};
int resp;
switch (cmd) {
case TEST_INIT:
info->name = "loop";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Test that loops are properly detected.";
info->description =
"Create two resource lists that refer to each other. Ensure that attempting to build a tree\n"
"results in an empty tree.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list_1 = create_resource_list(test, "herp", "test", resources_1, ARRAY_LEN(resources_1));
if (!list_1) {
return AST_TEST_FAIL;
}
list_2 = create_resource_list(test, "derp", "test", resources_2, ARRAY_LEN(resources_2));
if (!list_2) {
return AST_TEST_FAIL;
}
tree = ast_calloc(1, sizeof(*tree));
resp = build_resource_tree(NULL, &test_handler, "herp", tree, 1);
if (resp == 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
return AST_TEST_PASS;
}
AST_TEST_DEFINE(bad_event)
{
RAII_VAR(struct resource_list *, list, NULL, cleanup_resource_list);
RAII_VAR(struct resource_tree *, tree, NULL, test_resource_tree_destroy);
const char *resources[] = {
"huey",
"dewey",
"louie",
};
int resp;
switch (cmd) {
case TEST_INIT:
info->name = "bad_event";
info->category = "/res/res_pjsip_pubsub/";
info->summary = "Ensure that list with wrong event specified is not retrieved";
info->description =
"Create a simple resource list for event 'tsetse'. Ensure that trying to retrieve the list for event 'test' fails.";
return AST_TEST_NOT_RUN;
case TEST_EXECUTE:
break;
}
if (ineligible_configuration()) {
ast_test_status_update(test, "Ineligible configuration for this test. Please add a "
"'res_pjsip_pubsub' section to sorcery.conf, and set 'resource_list=memory'\n");
return AST_TEST_NOT_RUN;
}
list = create_resource_list(test, "foo", "tsetse", resources, ARRAY_LEN(resources));
if (!list) {
return AST_TEST_FAIL;
}
tree = ast_calloc(1, sizeof(*tree));
/* Since the test_handler is for event "test", this should not build a list, but
* instead result in a single resource being created, called "foo"
*/
resp = build_resource_tree(NULL, &test_handler, "foo", tree, 1);
if (resp != 200) {
ast_test_status_update(test, "Unexpected response %d when building resource tree\n", resp);
return AST_TEST_FAIL;
}
if (!tree->root) {
ast_test_status_update(test, "Resource tree has no root\n");
return AST_TEST_FAIL;
}
if (strcmp(tree->root->resource, "foo")) {
ast_test_status_update(test, "Unexpected resource %s found in tree\n", tree->root->resource);
return AST_TEST_FAIL;
}
return AST_TEST_PASS;
}
#endif
static int resource_endpoint_handler(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct ast_sip_publication_resource *resource = obj;
ast_free(resource->endpoint);
resource->endpoint = ast_strdup(var->value);
return 0;
}
static int resource_event_handler(const struct aco_option *opt, struct ast_variable *var, void *obj)
{
struct ast_sip_publication_resource *resource = obj;
/* The event configuration name starts with 'event_' so skip past it to get the real name */
const char *event = var->name + 6;
struct ast_variable *item;
if (ast_strlen_zero(event) || ast_strlen_zero(var->value)) {
return -1;
}
item = ast_variable_new(event, var->value, "");
if (!item) {
return -1;
}
if (resource->events) {
item->next = resource->events;
}
resource->events = item;
return 0;
}
static int load_module(void)
{
static const pj_str_t str_PUBLISH = { "PUBLISH", 7 };
struct ast_sorcery *sorcery;
sorcery = ast_sip_get_sorcery();
if (!(sched = ast_sched_context_create())) {
ast_log(LOG_ERROR, "Could not create scheduler for publication expiration\n");
return AST_MODULE_LOAD_DECLINE;
}
if (ast_sched_start_thread(sched)) {
ast_log(LOG_ERROR, "Could not start scheduler thread for publication expiration\n");
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
ast_sorcery_apply_config(sorcery, "res_pjsip_pubsub");
ast_sorcery_apply_default(sorcery, "subscription_persistence", "astdb", "subscription_persistence");
if (ast_sorcery_object_register(sorcery, "subscription_persistence", subscription_persistence_alloc,
NULL, NULL)) {
ast_log(LOG_ERROR, "Could not register subscription persistence object support\n");
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "packet", "", OPT_CHAR_ARRAY_T, 0,
CHARFLDSET(struct subscription_persistence, packet));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "src_name", "", OPT_CHAR_ARRAY_T, 0,
CHARFLDSET(struct subscription_persistence, src_name));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "src_port", "0", OPT_UINT_T, 0,
FLDSET(struct subscription_persistence, src_port));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "transport_key", "0", OPT_CHAR_ARRAY_T, 0,
pjsip_transport_events: Fix possible use after free on transport It was possible for a module that registered for transport monitor events to pass in a pjsip_transport that had already been freed. This caused pjsip_transport_events to crash when looking up the monitor for the transport. The fix is a two pronged approach. 1. We now increment the reference count on pjsip_transports when we create monitors for them, then decrement the count when the transport is going to be destroyed. 2. There are now APIs to register and unregister monitor callbacks by "transport key" which is a string concatenation of the remote ip address and port. This way the module needing to monitor the transport doesn't have to hold on to the transport object itself to unregister. It just has to save the transport_key. * Added the pjsip_transport reference increment and decrement. * Changed the internal transport monitor container key from the transport->obj_name (which may not be unique anyway) to the transport_key. * Added a helper macro AST_SIP_MAKE_REMOTE_IPADDR_PORT_STR() that fills a buffer with the transport_key using a passed-in pjsip_transport. * Added the following functions: ast_sip_transport_monitor_register_key ast_sip_transport_monitor_register_replace_key ast_sip_transport_monitor_unregister_key and marked their non-key counterparts as deprecated. * Updated res_pjsip_pubsub and res_pjsip_outbound_register to use the new "key" monitor functions. NOTE: res_pjsip_registrar also uses the transport monitor functionality but doesn't have a persistent object other than contact to store a transport key. At this time, it continues to use the non-key monitor functions. ASTERISK-30244 Change-Id: I1a20baf2a8643c272dcf819871d6c395f148f00b
2022-10-10 14:35:54 +00:00
CHARFLDSET(struct subscription_persistence, transport_type));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "local_name", "", OPT_CHAR_ARRAY_T, 0,
CHARFLDSET(struct subscription_persistence, local_name));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "local_port", "0", OPT_UINT_T, 0,
FLDSET(struct subscription_persistence, local_port));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "cseq", "0", OPT_UINT_T, 0,
FLDSET(struct subscription_persistence, cseq));
ast_sorcery_object_field_register_custom(sorcery, "subscription_persistence", "endpoint", "",
persistence_endpoint_str2struct, persistence_endpoint_struct2str, NULL, 0, 0);
ast_sorcery_object_field_register_custom(sorcery, "subscription_persistence", "tag", "",
persistence_tag_str2struct, persistence_tag_struct2str, NULL, 0, 0);
ast_sorcery_object_field_register_custom(sorcery, "subscription_persistence", "expires", "",
persistence_expires_str2struct, persistence_expires_struct2str, NULL, 0, 0);
res_pjsip: Symmetric transports A new transport parameter 'symmetric_transport' has been added. When a request from a dynamic contact comes in on a transport with this option set to 'yes', the transport name will be saved and used for subsequent outgoing requests like OPTIONS, NOTIFY and INVITE. It's saved as a contact uri parameter named 'x-ast-txp' and will display with the contact uri in CLI, AMI, and ARI output. On the outgoing request, if a transport wasn't explicitly set on the endpoint AND the request URI is not a hostname, the saved transport will be used and the 'x-ast-txp' parameter stripped from the outgoing packet. * config_transport was modified to accept and store the new parameter. * config_transport/transport_apply was updated to store the transport name in the pjsip_transport->info field using the pjsip_transport->pool on UDP transports. * A 'multihomed_on_rx_message' function was added to pjsip_message_ip_updater that, for incoming requests, retrieves the transport name from pjsip_transport->info and retrieves the transport. If transport->symmetric_transport is set, an 'x-ast-txp' uri parameter containing the transport name is added to the incoming Contact header. * An 'ast_sip_get_transport_name' function was added to res_pjsip. It takes an ast_sip_endpoint and a pjsip_sip_uri and returns a transport name if endpoint->transport is set or if there's an 'x-ast-txp' parameter on the uri and the uri host is an ipv4 or ipv6 address. Otherwise it returns NULL. * An 'ast_sip_dlg_set_transport' function was added to res_pjsip which takes an ast_sip_endpoint, a pjsip_dialog, and an optional pjsip_tpselector. It calls ast_sip_get_transport_name() and if a non-NULL is returned, sets the selector and sets the transport on the dialog. If a selector was passed in, it's updated. * res_pjsip/ast_sip_create_dialog_uac and ast_sip_create_dialog_uas were modified to call ast_sip_dlg_set_transport() instead of their original logic. * res_pjsip/create_out_of_dialog_request was modified to call ast_sip_get_transport_name() and pjsip_tx_data_set_transport() instead of its original logic. * Existing transport logic was removed from endpt_send_request since that can only be called after a create_out_of_dialog_request. * res_pjsip/ast_sip_create_rdata was converted to a wrapper around a new 'ast_sip_create_rdata_with_contact' function which allows a contact_uri to be specified in addition to the existing parameters. (See below) * res_pjsip_pubsub/internal_pjsip_evsub_send_request was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac and ast_sip_create_dialog_uas. * 'contact_uri' was added to subscription_persistence. This was necessary because although the parsed rdata contact header has the x-ast-txp parameter added (if appropriate), subscription_persistence_update stores the raw packet which doesn't have it. subscription_persistence_recreate was then updated to call ast_sip_create_rdata_with_contact with the persisted contact_uri so the recreated subscription has the correct transport info to send the NOTIFYs. * res_pjsip_session/internal_pjsip_inv_send_msg was eliminated since all it did was transport selection and that is now done in ast_sip_create_dialog_uac. * pjsip_message_ip_updater/multihomed_on_tx_message was updated to remove all traces of the x-ast-txp parameter from the outgoing headers. NOTE: This change does NOT modify the behavior of permanent contacts specified on an aor. To do so would require that the permanent contact's contact uri be updated with the x-ast-txp parameter and the aor sorcery object updated. If we need to persue this, we need to think about cloning permanent contacts into the same store as the dynamic ones on an aor load so they can be updated without disturbing the originally configured value. You CAN add the x-ast-txp parameter to a permanent contact's uri but it would be much simpler to just set endpoint->transport. Change-Id: I4ee1f51473da32ca54b877cd158523efcef9655f
2017-03-07 14:33:26 +00:00
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "contact_uri", "", OPT_CHAR_ARRAY_T, 0,
CHARFLDSET(struct subscription_persistence, contact_uri));
ast_sorcery_object_field_register(sorcery, "subscription_persistence", "prune_on_boot", "no", OPT_YESNO_T, 1,
FLDSET(struct subscription_persistence, prune_on_boot));
ast_sorcery_object_field_register_custom(sorcery, "subscription_persistence", "generator_data", "",
persistence_generator_data_str2struct, persistence_generator_data_struct2str, NULL, 0, 0);
if (apply_list_configuration(sorcery)) {
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
ast_sorcery_apply_default(sorcery, "inbound-publication", "config", "pjsip.conf,criteria=type=inbound-publication");
if (ast_sorcery_object_register(sorcery, "inbound-publication", publication_resource_alloc,
NULL, NULL)) {
ast_log(LOG_ERROR, "Could not register subscription persistence object support\n");
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
ast_sorcery_object_field_register(sorcery, "inbound-publication", "type", "", OPT_NOOP_T, 0, 0);
ast_sorcery_object_field_register_custom(sorcery, "inbound-publication", "endpoint", "",
resource_endpoint_handler, NULL, NULL, 0, 0);
ast_sorcery_object_fields_register(sorcery, "inbound-publication", "^event_", resource_event_handler, NULL);
ast_sorcery_reload_object(sorcery, "inbound-publication");
if (ast_sip_register_service(&pubsub_module)) {
ast_log(LOG_ERROR, "Could not register pubsub service\n");
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
if (pjsip_evsub_init_module(ast_sip_get_pjsip_endpoint()) != PJ_SUCCESS) {
ast_log(LOG_ERROR, "Could not initialize pjsip evsub module.\n");
ast_sip_unregister_service(&pubsub_module);
ast_sched_context_destroy(sched);
return AST_MODULE_LOAD_DECLINE;
}
/* Once pjsip_evsub_init_module succeeds we cannot unload.
* Keep all module_load errors above this point. */
ast_module_shutdown_ref(ast_module_info->self);
pjsip_media_type_init2(&rlmi_media_type, "application", "rlmi+xml");
pjsip_endpt_add_capability(ast_sip_get_pjsip_endpoint(), NULL, PJSIP_H_ALLOW, NULL, 1, &str_PUBLISH);
if (ast_test_flag(&ast_options, AST_OPT_FLAG_FULLY_BOOTED)) {
ast_sip_push_task(NULL, subscription_persistence_load, NULL);
} else {
struct stasis_subscription *sub;
sub = stasis_subscribe_pool(ast_manager_get_topic(), subscription_persistence_event_cb, NULL);
stasis_subscription_accept_message_type(sub, ast_manager_get_generic_type());
stasis_subscription_set_filter(sub, STASIS_SUBSCRIPTION_FILTER_SELECTIVE);
}
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
ast_manager_register_xml(AMI_SHOW_SUBSCRIPTIONS_INBOUND, EVENT_FLAG_SYSTEM,
ami_show_subscriptions_inbound);
ast_manager_register_xml(AMI_SHOW_SUBSCRIPTIONS_OUTBOUND, EVENT_FLAG_SYSTEM,
ami_show_subscriptions_outbound);
ast_manager_register_xml("PJSIPShowResourceLists", EVENT_FLAG_SYSTEM,
ami_show_resource_lists);
ast_cli_register_multiple(cli_commands, ARRAY_LEN(cli_commands));
AST_TEST_REGISTER(resource_tree);
AST_TEST_REGISTER(complex_resource_tree);
AST_TEST_REGISTER(bad_resource);
AST_TEST_REGISTER(bad_branch);
AST_TEST_REGISTER(duplicate_resource);
AST_TEST_REGISTER(loop);
AST_TEST_REGISTER(bad_event);
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
return AST_MODULE_LOAD_SUCCESS;
}
static int unload_module(void)
{
AST_TEST_UNREGISTER(resource_tree);
AST_TEST_UNREGISTER(complex_resource_tree);
AST_TEST_UNREGISTER(bad_resource);
AST_TEST_UNREGISTER(bad_branch);
AST_TEST_UNREGISTER(duplicate_resource);
AST_TEST_UNREGISTER(loop);
AST_TEST_UNREGISTER(bad_event);
ast_sip_transport_monitor_unregister_all(sub_tree_transport_cb, NULL, NULL);
ast_cli_unregister_multiple(cli_commands, ARRAY_LEN(cli_commands));
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
ast_manager_unregister(AMI_SHOW_SUBSCRIPTIONS_OUTBOUND);
ast_manager_unregister(AMI_SHOW_SUBSCRIPTIONS_INBOUND);
ast_manager_unregister("PJSIPShowResourceLists");
res_pjsip: AMI commands and events. Created the following AMI commands and corresponding events for res_pjsip: PJSIPShowEndpoints - Provides a listing of all pjsip endpoints and a few select attributes on each. Events: EndpointList - for each endpoint a few attributes. EndpointlistComplete - after all endpoints have been listed. PJSIPShowEndpoint - Provides a detail list of attributes for a specified endpoint. Events: EndpointDetail - attributes on an endpoint. AorDetail - raised for each AOR on an endpoint. AuthDetail - raised for each associated inbound and outbound auth TransportDetail - transport attributes. IdentifyDetail - attributes for the identify object associated with the endpoint. EndpointDetailComplete - last event raised after all detail events. PJSIPShowRegistrationsInbound - Provides a detail listing of all inbound registrations. Events: InboundRegistrationDetail - inbound registration attributes for each registration. InboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowRegistrationsOutbound - Provides a detail listing of all outbound registrations. Events: OutboundRegistrationDetail - outbound registration attributes for each registration. OutboundRegistrationDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsInbound - A detail listing of all inbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. PJSIPShowSubscriptionsOutbound - A detail listing of all outboundbound subscriptions and their attributes. Events: SubscriptionDetail - on each subscription detailed attributes SubscriptionDetailComplete - raised after all detail records have been listed. (issue ASTERISK-22609) Reported by: Matt Jordan Review: https://reviewboard.asterisk.org/r/2959/ ........ Merged revisions 403131 from http://svn.asterisk.org/svn/asterisk/branches/12 git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403133 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2013-11-23 17:26:57 +00:00
ast_sip_unregister_service(&pubsub_module);
if (sched) {
ast_sched_context_destroy(sched);
}
return 0;
}
AST_MODULE_INFO(ASTERISK_GPL_KEY, AST_MODFLAG_GLOBAL_SYMBOLS | AST_MODFLAG_LOAD_ORDER, "PJSIP event resource",
.support_level = AST_MODULE_SUPPORT_CORE,
.load = load_module,
.unload = unload_module,
.load_pri = AST_MODPRI_CHANNEL_DEPEND,
.requires = "res_pjsip",
);