Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*
|
|
|
|
* astobj2 - replacement containers for asterisk data structures.
|
|
|
|
*
|
|
|
|
* Copyright (C) 2006 Marta Carbone, Luigi Rizzo - Univ. di Pisa, Italy
|
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Function implementing astobj2 objects.
|
|
|
|
*/
|
|
|
|
#include "asterisk.h"
|
|
|
|
|
|
|
|
ASTERISK_FILE_VERSION(__FILE__, "$Revision$")
|
|
|
|
|
2007-11-20 22:18:21 +00:00
|
|
|
#include "asterisk/_private.h"
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
#include "asterisk/astobj2.h"
|
|
|
|
#include "asterisk/utils.h"
|
|
|
|
#include "asterisk/cli.h"
|
2008-04-16 17:45:28 +00:00
|
|
|
#define REF_FILE "/tmp/refs"
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
/*!
|
2007-09-06 15:43:49 +00:00
|
|
|
* astobj2 objects are always preceded by this data structure,
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
* which contains a lock, a reference counter,
|
|
|
|
* the flags and a pointer to a destructor.
|
|
|
|
* The refcount is used to decide when it is time to
|
|
|
|
* invoke the destructor.
|
|
|
|
* The magic number is used for consistency check.
|
|
|
|
* XXX the lock is not always needed, and its initialization may be
|
|
|
|
* expensive. Consider making it external.
|
|
|
|
*/
|
|
|
|
struct __priv_data {
|
|
|
|
ast_mutex_t lock;
|
|
|
|
int ref_counter;
|
|
|
|
ao2_destructor_fn destructor_fn;
|
|
|
|
/*! for stats */
|
|
|
|
size_t data_size;
|
|
|
|
/*! magic number. This is used to verify that a pointer passed in is a
|
|
|
|
* valid astobj2 */
|
|
|
|
uint32_t magic;
|
|
|
|
};
|
|
|
|
|
|
|
|
#define AO2_MAGIC 0xa570b123
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* What an astobj2 object looks like: fixed-size private data
|
|
|
|
* followed by variable-size user data.
|
|
|
|
*/
|
|
|
|
struct astobj2 {
|
|
|
|
struct __priv_data priv_data;
|
|
|
|
void *user_data[0];
|
|
|
|
};
|
|
|
|
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AST_DEVMODE
|
|
|
|
#define AO2_DEBUG 1
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
struct ao2_stats {
|
|
|
|
volatile int total_objects;
|
|
|
|
volatile int total_mem;
|
|
|
|
volatile int total_containers;
|
|
|
|
volatile int total_refs;
|
|
|
|
volatile int total_locked;
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct ao2_stats ao2;
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
#ifndef HAVE_BKTR /* backtrace support */
|
|
|
|
void ao2_bt(void) {}
|
|
|
|
#else
|
|
|
|
#include <execinfo.h> /* for backtrace */
|
|
|
|
|
|
|
|
void ao2_bt(void)
|
|
|
|
{
|
2008-03-04 23:04:29 +00:00
|
|
|
int c, i;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
#define N1 20
|
2008-03-04 23:04:29 +00:00
|
|
|
void *addresses[N1];
|
|
|
|
char **strings;
|
|
|
|
|
|
|
|
c = backtrace(addresses, N1);
|
2010-12-18 00:08:13 +00:00
|
|
|
strings = ast_bt_get_symbols(addresses,c);
|
2008-03-04 23:04:29 +00:00
|
|
|
ast_verbose("backtrace returned: %d\n", c);
|
|
|
|
for(i = 0; i < c; i++) {
|
|
|
|
ast_verbose("%d: %p %s\n", i, addresses[i], strings[i]);
|
|
|
|
}
|
|
|
|
free(strings);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* \brief convert from a pointer _p to a user-defined object
|
|
|
|
*
|
|
|
|
* \return the pointer to the astobj2 structure
|
|
|
|
*/
|
|
|
|
static inline struct astobj2 *INTERNAL_OBJ(void *user_data)
|
|
|
|
{
|
|
|
|
struct astobj2 *p;
|
|
|
|
|
|
|
|
if (!user_data) {
|
|
|
|
ast_log(LOG_ERROR, "user_data is NULL\n");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
p = (struct astobj2 *) ((char *) user_data - sizeof(*p));
|
|
|
|
if (AO2_MAGIC != (p->priv_data.magic) ) {
|
|
|
|
ast_log(LOG_ERROR, "bad magic number 0x%x for %p\n", p->priv_data.magic, p);
|
|
|
|
p = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
2008-11-25 01:01:49 +00:00
|
|
|
enum ao2_callback_type {
|
|
|
|
DEFAULT,
|
|
|
|
WITH_DATA,
|
|
|
|
};
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*!
|
|
|
|
* \brief convert from a pointer _p to an astobj2 object
|
|
|
|
*
|
|
|
|
* \return the pointer to the user-defined portion.
|
|
|
|
*/
|
|
|
|
#define EXTERNAL_OBJ(_p) ((_p) == NULL ? NULL : (_p)->user_data)
|
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
/* the underlying functions common to debug and non-debug versions */
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
static int internal_ao2_ref(void *user_data, const int delta);
|
|
|
|
static struct ao2_container *internal_ao2_container_alloc(struct ao2_container *c, const uint n_buckets, ao2_hash_fn *hash_fn,
|
|
|
|
ao2_callback_fn *cmp_fn);
|
2010-05-16 11:14:37 +00:00
|
|
|
static struct bucket_entry *internal_ao2_link(struct ao2_container *c, void *user_data, const char *file, int line, const char *func);
|
2009-03-18 02:21:23 +00:00
|
|
|
static void *internal_ao2_callback(struct ao2_container *c,
|
|
|
|
const enum search_flags flags, void *cb_fn, void *arg, void *data, enum ao2_callback_type type,
|
|
|
|
char *tag, char *file, int line, const char *funcname);
|
2010-05-16 11:14:37 +00:00
|
|
|
static void *internal_ao2_iterator_next(struct ao2_iterator *a, struct bucket_entry **q);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
int __ao2_lock(void *user_data, const char *file, const char *func, int line, const char *var)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *p = INTERNAL_OBJ(user_data);
|
|
|
|
|
|
|
|
if (p == NULL)
|
|
|
|
return -1;
|
|
|
|
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_locked, 1);
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2008-11-21 22:12:37 +00:00
|
|
|
return __ast_pthread_mutex_lock(file, line, func, var, &p->priv_data.lock);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
int __ao2_unlock(void *user_data, const char *file, const char *func, int line, const char *var)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *p = INTERNAL_OBJ(user_data);
|
|
|
|
|
|
|
|
if (p == NULL)
|
|
|
|
return -1;
|
|
|
|
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_locked, -1);
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2008-11-21 22:12:37 +00:00
|
|
|
return __ast_pthread_mutex_unlock(file, line, func, var, &p->priv_data.lock);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
int __ao2_trylock(void *user_data, const char *file, const char *func, int line, const char *var)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *p = INTERNAL_OBJ(user_data);
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (p == NULL)
|
|
|
|
return -1;
|
2009-02-12 16:28:06 +00:00
|
|
|
ret = __ast_pthread_mutex_trylock(file, line, func, var, &p->priv_data.lock);
|
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
#ifdef AO2_DEBUG
|
|
|
|
if (!ret)
|
|
|
|
ast_atomic_fetchadd_int(&ao2.total_locked, 1);
|
|
|
|
#endif
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
void *ao2_object_get_lockaddr(void *obj)
|
|
|
|
{
|
|
|
|
struct astobj2 *p = INTERNAL_OBJ(obj);
|
|
|
|
|
|
|
|
if (p == NULL)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return &p->priv_data.lock;
|
|
|
|
}
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*
|
|
|
|
* The argument is a pointer to the user portion.
|
|
|
|
*/
|
2008-04-16 17:45:28 +00:00
|
|
|
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
int __ao2_ref_debug(void *user_data, const int delta, char *tag, char *file, int line, const char *funcname)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *obj = INTERNAL_OBJ(user_data);
|
|
|
|
|
|
|
|
if (obj == NULL)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
if (delta != 0) {
|
|
|
|
FILE *refo = fopen(REF_FILE,"a");
|
2009-04-23 20:36:35 +00:00
|
|
|
fprintf(refo, "%p %s%d %s:%d:%s (%s) [@%d]\n", user_data, (delta<0? "":"+"), delta, file, line, funcname, tag, obj ? obj->priv_data.ref_counter : -1);
|
2008-04-16 17:45:28 +00:00
|
|
|
fclose(refo);
|
|
|
|
}
|
|
|
|
if (obj->priv_data.ref_counter + delta == 0 && obj->priv_data.destructor_fn != NULL) { /* this isn't protected with lock; just for o/p */
|
|
|
|
FILE *refo = fopen(REF_FILE,"a");
|
|
|
|
fprintf(refo, "%p **call destructor** %s:%d:%s (%s)\n", user_data, file, line, funcname, tag);
|
|
|
|
fclose(refo);
|
|
|
|
}
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_ref(user_data, delta);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
int __ao2_ref(void *user_data, const int delta)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *obj = INTERNAL_OBJ(user_data);
|
|
|
|
|
|
|
|
if (obj == NULL)
|
|
|
|
return -1;
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_ref(user_data, delta);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
static int internal_ao2_ref(void *user_data, const int delta)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
struct astobj2 *obj = INTERNAL_OBJ(user_data);
|
|
|
|
int current_value;
|
|
|
|
int ret;
|
|
|
|
|
2010-03-19 07:37:00 +00:00
|
|
|
if (obj == NULL)
|
|
|
|
return -1;
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* if delta is 0, just return the refcount */
|
|
|
|
if (delta == 0)
|
|
|
|
return (obj->priv_data.ref_counter);
|
|
|
|
|
|
|
|
/* we modify with an atomic operation the reference counter */
|
|
|
|
ret = ast_atomic_fetchadd_int(&obj->priv_data.ref_counter, delta);
|
|
|
|
current_value = ret + delta;
|
2007-09-13 18:52:35 +00:00
|
|
|
|
|
|
|
#ifdef AO2_DEBUG
|
|
|
|
ast_atomic_fetchadd_int(&ao2.total_refs, delta);
|
|
|
|
#endif
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* this case must never happen */
|
|
|
|
if (current_value < 0)
|
|
|
|
ast_log(LOG_ERROR, "refcount %d on object %p\n", current_value, user_data);
|
|
|
|
|
|
|
|
if (current_value <= 0) { /* last reference, destroy the object */
|
2008-04-16 17:45:28 +00:00
|
|
|
if (obj->priv_data.destructor_fn != NULL) {
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
obj->priv_data.destructor_fn(user_data);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
ast_mutex_destroy(&obj->priv_data.lock);
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_mem, - obj->priv_data.data_size);
|
2007-09-13 18:52:35 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_objects, -1);
|
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* for safety, zero-out the astobj2 header and also the
|
|
|
|
* first word of the user-data, which we make sure is always
|
|
|
|
* allocated. */
|
2008-10-09 14:17:33 +00:00
|
|
|
memset(obj, '\0', sizeof(struct astobj2 *) + sizeof(void *) );
|
2009-10-21 21:08:47 +00:00
|
|
|
ast_free(obj);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We always alloc at least the size of a void *,
|
|
|
|
* for debugging purposes.
|
|
|
|
*/
|
2009-05-05 10:34:19 +00:00
|
|
|
static void *internal_ao2_alloc(size_t data_size, ao2_destructor_fn destructor_fn, const char *file, int line, const char *funcname)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
/* allocation */
|
|
|
|
struct astobj2 *obj;
|
|
|
|
|
|
|
|
if (data_size < sizeof(void *))
|
|
|
|
data_size = sizeof(void *);
|
|
|
|
|
2009-05-04 16:24:16 +00:00
|
|
|
#if defined(__AST_DEBUG_MALLOC)
|
|
|
|
obj = __ast_calloc(1, sizeof(*obj) + data_size, file, line, funcname);
|
|
|
|
#else
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
obj = ast_calloc(1, sizeof(*obj) + data_size);
|
2009-05-04 16:24:16 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
if (obj == NULL)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
ast_mutex_init(&obj->priv_data.lock);
|
|
|
|
obj->priv_data.magic = AO2_MAGIC;
|
|
|
|
obj->priv_data.data_size = data_size;
|
|
|
|
obj->priv_data.ref_counter = 1;
|
|
|
|
obj->priv_data.destructor_fn = destructor_fn; /* can be NULL */
|
2007-09-13 18:52:35 +00:00
|
|
|
|
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_objects, 1);
|
|
|
|
ast_atomic_fetchadd_int(&ao2.total_mem, data_size);
|
|
|
|
ast_atomic_fetchadd_int(&ao2.total_refs, 1);
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
/* return a pointer to the user data */
|
|
|
|
return EXTERNAL_OBJ(obj);
|
|
|
|
}
|
|
|
|
|
2009-05-05 10:34:19 +00:00
|
|
|
void *__ao2_alloc_debug(size_t data_size, ao2_destructor_fn destructor_fn, char *tag,
|
|
|
|
const char *file, int line, const char *funcname, int ref_debug)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
/* allocation */
|
|
|
|
void *obj;
|
2009-05-05 10:34:19 +00:00
|
|
|
FILE *refo = ref_debug ? fopen(REF_FILE,"a") : NULL;
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2009-05-28 15:35:23 +00:00
|
|
|
if ((obj = internal_ao2_alloc(data_size, destructor_fn, file, line, funcname)) == NULL) {
|
|
|
|
fclose(refo);
|
2008-04-16 17:45:28 +00:00
|
|
|
return NULL;
|
2009-05-28 15:35:23 +00:00
|
|
|
}
|
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
if (refo) {
|
|
|
|
fprintf(refo, "%p =1 %s:%d:%s (%s)\n", obj, file, line, funcname, tag);
|
|
|
|
fclose(refo);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* return a pointer to the user data */
|
|
|
|
return obj;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_alloc(size_t data_size, ao2_destructor_fn destructor_fn)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2009-05-05 13:18:21 +00:00
|
|
|
return internal_ao2_alloc(data_size, destructor_fn, __FILE__, __LINE__, __FUNCTION__);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* internal callback to destroy a container. */
|
|
|
|
static void container_destruct(void *c);
|
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
/* internal callback to destroy a container. */
|
|
|
|
static void container_destruct_debug(void *c);
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* each bucket in the container is a tailq. */
|
2010-05-16 11:14:37 +00:00
|
|
|
AST_LIST_HEAD_NOLOCK(bucket, bucket_entry);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
/*!
|
|
|
|
* A container; stores the hash and callback functions, information on
|
|
|
|
* the size, the hash bucket heads, and a version number, starting at 0
|
|
|
|
* (for a newly created, empty container)
|
|
|
|
* and incremented every time an object is inserted or deleted.
|
|
|
|
* The assumption is that an object is never moved in a container,
|
|
|
|
* but removed and readded with the new number.
|
|
|
|
* The version number is especially useful when implementing iterators.
|
|
|
|
* In fact, we can associate a unique, monotonically increasing number to
|
|
|
|
* each object, which means that, within an iterator, we can store the
|
|
|
|
* version number of the current object, and easily look for the next one,
|
|
|
|
* which is the next one in the list with a higher number.
|
|
|
|
* Since all objects have a version >0, we can use 0 as a marker for
|
|
|
|
* 'we need the first object in the bucket'.
|
|
|
|
*
|
|
|
|
* \todo Linking and unlink objects is typically expensive, as it
|
|
|
|
* involves a malloc() of a small object which is very inefficient.
|
2010-05-16 11:14:37 +00:00
|
|
|
* To optimize this, we allocate larger arrays of bucket_entry's
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
* when we run out of them, and then manage our own freelist.
|
|
|
|
* This will be more efficient as we can do the freelist management while
|
|
|
|
* we hold the lock (that we need anyways).
|
|
|
|
*/
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_container {
|
2007-09-06 15:43:49 +00:00
|
|
|
ao2_hash_fn *hash_fn;
|
|
|
|
ao2_callback_fn *cmp_fn;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
int n_buckets;
|
|
|
|
/*! Number of elements in the container */
|
|
|
|
int elements;
|
|
|
|
/*! described above */
|
|
|
|
int version;
|
|
|
|
/*! variable size */
|
|
|
|
struct bucket buckets[0];
|
|
|
|
};
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* \brief always zero hash function
|
|
|
|
*
|
|
|
|
* it is convenient to have a hash function that always returns 0.
|
|
|
|
* This is basically used when we want to have a container that is
|
|
|
|
* a simple linked list.
|
|
|
|
*
|
|
|
|
* \returns 0
|
|
|
|
*/
|
|
|
|
static int hash_zero(const void *user_obj, const int flags)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* A container is just an object, after all!
|
|
|
|
*/
|
2009-03-18 02:21:23 +00:00
|
|
|
static struct ao2_container *internal_ao2_container_alloc(struct ao2_container *c, const unsigned int n_buckets, ao2_hash_fn *hash_fn,
|
|
|
|
ao2_callback_fn *cmp_fn)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
/* XXX maybe consistency check on arguments ? */
|
|
|
|
/* compute the container size */
|
|
|
|
|
|
|
|
if (!c)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
c->version = 1; /* 0 is a reserved value here */
|
2009-04-23 20:36:35 +00:00
|
|
|
c->n_buckets = hash_fn ? n_buckets : 1;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
c->hash_fn = hash_fn ? hash_fn : hash_zero;
|
|
|
|
c->cmp_fn = cmp_fn;
|
2007-09-13 18:52:35 +00:00
|
|
|
|
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_containers, 1);
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return c;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
struct ao2_container *__ao2_container_alloc_debug(const unsigned int n_buckets, ao2_hash_fn *hash_fn,
|
2009-05-05 10:34:19 +00:00
|
|
|
ao2_callback_fn *cmp_fn, char *tag, char *file, int line,
|
|
|
|
const char *funcname, int ref_debug)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
/* XXX maybe consistency check on arguments ? */
|
|
|
|
/* compute the container size */
|
2009-04-23 20:36:35 +00:00
|
|
|
const unsigned int num_buckets = hash_fn ? n_buckets : 1;
|
|
|
|
size_t container_size = sizeof(struct ao2_container) + num_buckets * sizeof(struct bucket);
|
2009-05-05 10:34:19 +00:00
|
|
|
struct ao2_container *c = __ao2_alloc_debug(container_size, container_destruct_debug, tag, file, line, funcname, ref_debug);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2009-04-23 20:36:35 +00:00
|
|
|
return internal_ao2_container_alloc(c, num_buckets, hash_fn, cmp_fn);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
struct ao2_container *__ao2_container_alloc(const unsigned int n_buckets, ao2_hash_fn *hash_fn,
|
|
|
|
ao2_callback_fn *cmp_fn)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
/* XXX maybe consistency check on arguments ? */
|
|
|
|
/* compute the container size */
|
|
|
|
|
2009-04-23 20:36:35 +00:00
|
|
|
const unsigned int num_buckets = hash_fn ? n_buckets : 1;
|
|
|
|
size_t container_size = sizeof(struct ao2_container) + num_buckets * sizeof(struct bucket);
|
2009-03-18 02:21:23 +00:00
|
|
|
struct ao2_container *c = __ao2_alloc(container_size, container_destruct);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2009-04-23 20:36:35 +00:00
|
|
|
return internal_ao2_container_alloc(c, num_buckets, hash_fn, cmp_fn);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*!
|
|
|
|
* return the number of elements in the container
|
|
|
|
*/
|
2007-09-04 18:40:07 +00:00
|
|
|
int ao2_container_count(struct ao2_container *c)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
return c->elements;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* A structure to create a linked list of entries,
|
|
|
|
* used within a bucket.
|
|
|
|
* XXX \todo this should be private to the container code
|
|
|
|
*/
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry {
|
|
|
|
AST_LIST_ENTRY(bucket_entry) entry;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
int version;
|
|
|
|
struct astobj2 *astobj; /* pointer to internal data */
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* link an object to a container
|
|
|
|
*/
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2010-05-16 11:14:37 +00:00
|
|
|
static struct bucket_entry *internal_ao2_link(struct ao2_container *c, void *user_data, const char *file, int line, const char *func)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
/* create a new list entry */
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
struct astobj2 *obj = INTERNAL_OBJ(user_data);
|
2009-03-11 00:29:59 +00:00
|
|
|
|
2010-03-19 07:37:00 +00:00
|
|
|
if (obj == NULL)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (INTERNAL_OBJ(c) == NULL)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
p = ast_calloc(1, sizeof(*p));
|
|
|
|
if (!p)
|
|
|
|
return NULL;
|
|
|
|
|
2009-10-02 16:59:57 +00:00
|
|
|
i = abs(c->hash_fn(user_data, OBJ_POINTER));
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
ao2_lock(c);
|
|
|
|
i %= c->n_buckets;
|
|
|
|
p->astobj = obj;
|
|
|
|
p->version = ast_atomic_fetchadd_int(&c->version, 1);
|
2007-08-22 22:43:12 +00:00
|
|
|
AST_LIST_INSERT_TAIL(&c->buckets[i], p, entry);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&c->elements, 1);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
|
|
|
/* the last two operations (ao2_ref, ao2_unlock) must be done by the calling func */
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_link_debug(struct ao2_container *c, void *user_data, char *tag, char *file, int line, const char *funcname)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p = internal_ao2_link(c, user_data, file, line, funcname);
|
2009-03-11 00:29:59 +00:00
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
if (p) {
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref_debug(user_data, +1, tag, file, line, funcname);
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_unlock(c);
|
|
|
|
}
|
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_link(struct ao2_container *c, void *user_data)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p = internal_ao2_link(c, user_data, __FILE__, __LINE__, __PRETTY_FUNCTION__);
|
2009-03-11 00:29:59 +00:00
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
if (p) {
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref(user_data, +1);
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_unlock(c);
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return p;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* \brief another convenience function is a callback that matches on address
|
|
|
|
*/
|
2008-11-25 01:01:49 +00:00
|
|
|
int ao2_match_by_addr(void *user_data, void *arg, int flags)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
return (user_data == arg) ? (CMP_MATCH | CMP_STOP) : 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Unlink an object from the container
|
2010-05-16 11:14:37 +00:00
|
|
|
* and destroy the associated * bucket_entry structure.
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
*/
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_unlink_debug(struct ao2_container *c, void *user_data, char *tag,
|
|
|
|
char *file, int line, const char *funcname)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
if (INTERNAL_OBJ(user_data) == NULL) /* safety check on the argument */
|
|
|
|
return NULL;
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_callback_debug(c, OBJ_UNLINK | OBJ_POINTER | OBJ_NODATA, ao2_match_by_addr, user_data, tag, file, line, funcname);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_unlink(struct ao2_container *c, void *user_data)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
|
|
|
if (INTERNAL_OBJ(user_data) == NULL) /* safety check on the argument */
|
|
|
|
return NULL;
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_callback(c, OBJ_UNLINK | OBJ_POINTER | OBJ_NODATA, ao2_match_by_addr, user_data);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* \brief special callback that matches all
|
|
|
|
*/
|
2008-11-25 01:01:49 +00:00
|
|
|
static int cb_true(void *user_data, void *arg, int flags)
|
|
|
|
{
|
|
|
|
return CMP_MATCH;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* \brief similar to cb_true, but is an ao2_callback_data_fn instead
|
|
|
|
*/
|
|
|
|
static int cb_true_data(void *user_data, void *arg, void *data, int flags)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
return CMP_MATCH;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* Browse the container using different stategies accoding the flags.
|
|
|
|
* \return Is a pointer to an object or to a list of object if OBJ_MULTIPLE is
|
|
|
|
* specified.
|
2008-04-16 17:45:28 +00:00
|
|
|
* Luckily, for debug purposes, the added args (tag, file, line, funcname)
|
|
|
|
* aren't an excessive load to the system, as the callback should not be
|
|
|
|
* called as often as, say, the ao2_ref func is called.
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
*/
|
2009-03-18 02:21:23 +00:00
|
|
|
static void *internal_ao2_callback(struct ao2_container *c,
|
|
|
|
const enum search_flags flags, void *cb_fn, void *arg, void *data, enum ao2_callback_type type,
|
|
|
|
char *tag, char *file, int line, const char *funcname)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2009-09-03 16:31:54 +00:00
|
|
|
int i, start, last; /* search boundaries */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
void *ret = NULL;
|
2008-11-25 01:01:49 +00:00
|
|
|
ao2_callback_fn *cb_default = NULL;
|
|
|
|
ao2_callback_data_fn *cb_withdata = NULL;
|
2009-10-21 21:08:47 +00:00
|
|
|
struct ao2_container *multi_container = NULL;
|
|
|
|
struct ao2_iterator *multi_iterator = NULL;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
if (INTERNAL_OBJ(c) == NULL) /* safety check on the argument */
|
|
|
|
return NULL;
|
|
|
|
|
2010-02-05 21:21:05 +00:00
|
|
|
/*
|
|
|
|
* This logic is used so we can support OBJ_MULTIPLE with OBJ_NODATA
|
|
|
|
* turned off. This if statement checks for the special condition
|
|
|
|
* where multiple items may need to be returned.
|
|
|
|
*/
|
|
|
|
if ((flags & (OBJ_MULTIPLE | OBJ_NODATA)) == OBJ_MULTIPLE) {
|
2009-10-21 21:08:47 +00:00
|
|
|
/* we need to return an ao2_iterator with the results,
|
|
|
|
* as there could be more than one. the iterator will
|
|
|
|
* hold the only reference to a container that has all the
|
|
|
|
* matching objects linked into it, so when the iterator
|
|
|
|
* is destroyed, the container will be automatically
|
|
|
|
* destroyed as well.
|
|
|
|
*/
|
|
|
|
if (!(multi_container = __ao2_container_alloc(1, NULL, NULL))) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
if (!(multi_iterator = ast_calloc(1, sizeof(*multi_iterator)))) {
|
|
|
|
ao2_ref(multi_container, -1);
|
|
|
|
return NULL;
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* override the match function if necessary */
|
2008-11-25 01:01:49 +00:00
|
|
|
if (cb_fn == NULL) { /* if NULL, match everything */
|
|
|
|
if (type == WITH_DATA) {
|
|
|
|
cb_withdata = cb_true_data;
|
|
|
|
} else {
|
|
|
|
cb_default = cb_true;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* We do this here to avoid the per object casting penalty (even though
|
2009-10-21 21:08:47 +00:00
|
|
|
that is probably optimized away anyway). */
|
2008-11-25 01:01:49 +00:00
|
|
|
if (type == WITH_DATA) {
|
|
|
|
cb_withdata = cb_fn;
|
|
|
|
} else {
|
|
|
|
cb_default = cb_fn;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*
|
|
|
|
* XXX this can be optimized.
|
|
|
|
* If we have a hash function and lookup by pointer,
|
|
|
|
* run the hash function. Otherwise, scan the whole container
|
|
|
|
* (this only for the time being. We need to optimize this.)
|
|
|
|
*/
|
|
|
|
if ((flags & OBJ_POINTER)) /* we know hash can handle this case */
|
2009-09-03 16:31:54 +00:00
|
|
|
start = i = c->hash_fn(arg, flags & OBJ_POINTER) % c->n_buckets;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
else /* don't know, let's scan all buckets */
|
2009-09-04 06:08:33 +00:00
|
|
|
start = i = -1; /* XXX this must be fixed later. */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
/* determine the search boundaries: i..last-1 */
|
|
|
|
if (i < 0) {
|
2009-09-03 16:31:54 +00:00
|
|
|
start = i = 0;
|
|
|
|
last = c->n_buckets;
|
|
|
|
} else if ((flags & OBJ_CONTINUE)) {
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
last = c->n_buckets;
|
|
|
|
} else {
|
|
|
|
last = i + 1;
|
|
|
|
}
|
|
|
|
|
2010-12-20 18:03:09 +00:00
|
|
|
|
|
|
|
if (!(flags & OBJ_NOLOCK)) {
|
|
|
|
ao2_lock(c); /* avoid modifications to the content */
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
for (; i < last ; i++) {
|
|
|
|
/* scan the list with prev-cur pointers */
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *cur;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
AST_LIST_TRAVERSE_SAFE_BEGIN(&c->buckets[i], cur, entry) {
|
2008-11-25 01:01:49 +00:00
|
|
|
int match = (CMP_MATCH | CMP_STOP);
|
|
|
|
|
|
|
|
if (type == WITH_DATA) {
|
|
|
|
match &= cb_withdata(EXTERNAL_OBJ(cur->astobj), arg, data, flags);
|
|
|
|
} else {
|
|
|
|
match &= cb_default(EXTERNAL_OBJ(cur->astobj), arg, flags);
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2009-10-26 22:03:29 +00:00
|
|
|
/* we found the object, performing operations according flags */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
if (match == 0) { /* no match, no stop, continue */
|
|
|
|
continue;
|
|
|
|
} else if (match == CMP_STOP) { /* no match but stop, we are done */
|
|
|
|
i = last;
|
|
|
|
break;
|
|
|
|
}
|
2009-10-21 21:08:47 +00:00
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* we have a match (CMP_MATCH) here */
|
|
|
|
if (!(flags & OBJ_NODATA)) { /* if must return the object, record the value */
|
|
|
|
/* it is important to handle this case before the unlink */
|
|
|
|
ret = EXTERNAL_OBJ(cur->astobj);
|
2009-10-21 21:08:47 +00:00
|
|
|
if (!(flags & (OBJ_UNLINK | OBJ_MULTIPLE))) {
|
|
|
|
if (tag)
|
|
|
|
__ao2_ref_debug(ret, 1, tag, file, line, funcname);
|
|
|
|
else
|
|
|
|
__ao2_ref(ret, 1);
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2010-02-05 21:21:05 +00:00
|
|
|
/* If we are in OBJ_MULTIPLE mode and OBJ_NODATE is off,
|
|
|
|
* link the object into the container that will hold the results.
|
2009-10-21 21:08:47 +00:00
|
|
|
*/
|
2010-02-05 21:21:05 +00:00
|
|
|
if (ret && (multi_container != NULL)) {
|
2009-10-21 21:08:47 +00:00
|
|
|
__ao2_link(multi_container, ret);
|
|
|
|
ret = NULL;
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2009-10-21 21:08:47 +00:00
|
|
|
if (flags & OBJ_UNLINK) { /* must unlink */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/* we are going to modify the container, so update version */
|
|
|
|
ast_atomic_fetchadd_int(&c->version, 1);
|
2007-11-08 05:28:47 +00:00
|
|
|
AST_LIST_REMOVE_CURRENT(entry);
|
2009-10-21 21:08:47 +00:00
|
|
|
/* update number of elements */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&c->elements, -1);
|
2010-02-11 18:42:25 +00:00
|
|
|
|
|
|
|
/* - When unlinking and not returning the result, (OBJ_NODATA), the ref from the container
|
|
|
|
* must be decremented.
|
|
|
|
* - When unlinking with OBJ_MULTIPLE the ref from the original container
|
|
|
|
* must be decremented regardless if OBJ_NODATA is used. This is because the result is
|
|
|
|
* returned in a new container that already holds its own ref for the object. If the ref
|
|
|
|
* from the original container is not accounted for here a memory leak occurs. */
|
|
|
|
if (flags & (OBJ_NODATA | OBJ_MULTIPLE)) {
|
2009-10-26 22:03:29 +00:00
|
|
|
if (tag)
|
|
|
|
__ao2_ref_debug(EXTERNAL_OBJ(cur->astobj), -1, tag, file, line, funcname);
|
|
|
|
else
|
|
|
|
__ao2_ref(EXTERNAL_OBJ(cur->astobj), -1);
|
|
|
|
}
|
2009-10-21 21:08:47 +00:00
|
|
|
ast_free(cur); /* free the link record */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2010-02-05 21:21:05 +00:00
|
|
|
if ((match & CMP_STOP) || !(flags & OBJ_MULTIPLE)) {
|
2010-01-11 16:40:23 +00:00
|
|
|
/* We found our only (or last) match, so force an exit from
|
|
|
|
the outside loop. */
|
|
|
|
i = last;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2007-11-08 05:28:47 +00:00
|
|
|
AST_LIST_TRAVERSE_SAFE_END;
|
2009-09-03 16:31:54 +00:00
|
|
|
|
|
|
|
if (ret) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (i == c->n_buckets - 1 && (flags & OBJ_POINTER) && (flags & OBJ_CONTINUE)) {
|
|
|
|
/* Move to the beginning to ensure we check every bucket */
|
|
|
|
i = -1;
|
|
|
|
last = start;
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
2010-12-20 18:03:09 +00:00
|
|
|
|
|
|
|
if (!(flags & OBJ_NOLOCK)) {
|
|
|
|
ao2_unlock(c);
|
|
|
|
}
|
2010-02-05 21:21:05 +00:00
|
|
|
|
|
|
|
/* if multi_container was created, we are returning multiple objects */
|
|
|
|
if (multi_container != NULL) {
|
2009-10-21 21:08:47 +00:00
|
|
|
*multi_iterator = ao2_iterator_init(multi_container,
|
|
|
|
AO2_ITERATOR_DONTLOCK | AO2_ITERATOR_UNLINK | AO2_ITERATOR_MALLOCD);
|
|
|
|
ao2_ref(multi_container, -1);
|
|
|
|
return multi_iterator;
|
|
|
|
} else {
|
|
|
|
return ret;
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_callback_debug(struct ao2_container *c,
|
|
|
|
const enum search_flags flags,
|
|
|
|
ao2_callback_fn *cb_fn, void *arg,
|
|
|
|
char *tag, char *file, int line, const char *funcname)
|
2008-11-25 01:01:49 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_callback(c,flags, cb_fn, arg, NULL, DEFAULT, tag, file, line, funcname);
|
2008-11-25 01:01:49 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_callback(struct ao2_container *c, const enum search_flags flags,
|
|
|
|
ao2_callback_fn *cb_fn, void *arg)
|
2008-11-25 01:01:49 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_callback(c,flags, cb_fn, arg, NULL, DEFAULT, NULL, NULL, 0, NULL);
|
2008-11-25 01:01:49 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_callback_data_debug(struct ao2_container *c,
|
|
|
|
const enum search_flags flags,
|
|
|
|
ao2_callback_data_fn *cb_fn, void *arg, void *data,
|
|
|
|
char *tag, char *file, int line, const char *funcname)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_callback(c, flags, cb_fn, arg, data, WITH_DATA, tag, file, line, funcname);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_callback_data(struct ao2_container *c, const enum search_flags flags,
|
|
|
|
ao2_callback_data_fn *cb_fn, void *arg, void *data)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return internal_ao2_callback(c, flags, cb_fn, arg, data, WITH_DATA, NULL, NULL, 0, NULL);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*!
|
|
|
|
* the find function just invokes the default callback with some reasonable flags.
|
|
|
|
*/
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_find_debug(struct ao2_container *c, void *arg, enum search_flags flags, char *tag, char *file, int line, const char *funcname)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return __ao2_callback_debug(c, flags, c->cmp_fn, arg, tag, file, line, funcname);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_find(struct ao2_container *c, void *arg, enum search_flags flags)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
return __ao2_callback(c, flags, c->cmp_fn, arg);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*!
|
|
|
|
* initialize an iterator so we start from the first object
|
|
|
|
*/
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_iterator ao2_iterator_init(struct ao2_container *c, int flags)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_iterator a = {
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
.c = c,
|
|
|
|
.flags = flags
|
|
|
|
};
|
2009-10-06 01:24:24 +00:00
|
|
|
|
|
|
|
ao2_ref(c, +1);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
return a;
|
|
|
|
}
|
|
|
|
|
2009-10-06 01:24:24 +00:00
|
|
|
/*!
|
|
|
|
* destroy an iterator
|
|
|
|
*/
|
|
|
|
void ao2_iterator_destroy(struct ao2_iterator *i)
|
|
|
|
{
|
|
|
|
ao2_ref(i->c, -1);
|
2009-10-21 21:08:47 +00:00
|
|
|
if (i->flags & AO2_ITERATOR_MALLOCD) {
|
|
|
|
ast_free(i);
|
|
|
|
} else {
|
|
|
|
i->c = NULL;
|
|
|
|
}
|
2009-10-06 01:24:24 +00:00
|
|
|
}
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*
|
|
|
|
* move to the next element in the container.
|
|
|
|
*/
|
2010-05-16 11:14:37 +00:00
|
|
|
static void *internal_ao2_iterator_next(struct ao2_iterator *a, struct bucket_entry **q)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
|
|
|
int lim;
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p = NULL;
|
2007-09-10 16:19:45 +00:00
|
|
|
void *ret = NULL;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
*q = NULL;
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
if (INTERNAL_OBJ(a->c) == NULL)
|
|
|
|
return NULL;
|
|
|
|
|
2009-10-06 01:24:24 +00:00
|
|
|
if (!(a->flags & AO2_ITERATOR_DONTLOCK))
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ao2_lock(a->c);
|
|
|
|
|
|
|
|
/* optimization. If the container is unchanged and
|
|
|
|
* we have a pointer, try follow it
|
|
|
|
*/
|
2009-10-21 21:08:47 +00:00
|
|
|
if (a->c->version == a->c_version && (p = a->obj)) {
|
|
|
|
if ((p = AST_LIST_NEXT(p, entry)))
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
goto found;
|
|
|
|
/* nope, start from the next bucket */
|
|
|
|
a->bucket++;
|
|
|
|
a->version = 0;
|
|
|
|
a->obj = NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
lim = a->c->n_buckets;
|
|
|
|
|
|
|
|
/* Browse the buckets array, moving to the next
|
|
|
|
* buckets if we don't find the entry in the current one.
|
|
|
|
* Stop when we find an element with version number greater
|
|
|
|
* than the current one (we reset the version to 0 when we
|
|
|
|
* switch buckets).
|
|
|
|
*/
|
|
|
|
for (; a->bucket < lim; a->bucket++, a->version = 0) {
|
|
|
|
/* scan the current bucket */
|
|
|
|
AST_LIST_TRAVERSE(&a->c->buckets[a->bucket], p, entry) {
|
|
|
|
if (p->version > a->version)
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
found:
|
|
|
|
if (p) {
|
2007-09-10 16:19:45 +00:00
|
|
|
ret = EXTERNAL_OBJ(p->astobj);
|
2009-10-21 21:08:47 +00:00
|
|
|
if (a->flags & AO2_ITERATOR_UNLINK) {
|
|
|
|
/* we are going to modify the container, so update version */
|
|
|
|
ast_atomic_fetchadd_int(&a->c->version, 1);
|
|
|
|
AST_LIST_REMOVE(&a->c->buckets[a->bucket], p, entry);
|
|
|
|
/* update number of elements */
|
|
|
|
ast_atomic_fetchadd_int(&a->c->elements, -1);
|
|
|
|
a->version = 0;
|
|
|
|
a->obj = NULL;
|
|
|
|
a->c_version = a->c->version;
|
|
|
|
ast_free(p);
|
|
|
|
} else {
|
|
|
|
a->version = p->version;
|
|
|
|
a->obj = p;
|
|
|
|
a->c_version = a->c->version;
|
|
|
|
/* inc refcount of returned object */
|
|
|
|
*q = p;
|
|
|
|
}
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_iterator_next_debug(struct ao2_iterator *a, char *tag, char *file, int line, const char *funcname)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p;
|
2008-04-16 17:45:28 +00:00
|
|
|
void *ret = NULL;
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
ret = internal_ao2_iterator_next(a, &p);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
|
|
|
if (p) {
|
|
|
|
/* inc refcount of returned object */
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref_debug(ret, 1, tag, file, line, funcname);
|
2008-04-16 17:45:28 +00:00
|
|
|
}
|
|
|
|
|
2009-10-06 01:24:24 +00:00
|
|
|
if (!(a->flags & AO2_ITERATOR_DONTLOCK))
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_unlock(a->c);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
void *__ao2_iterator_next(struct ao2_iterator *a)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *p = NULL;
|
2008-04-16 17:45:28 +00:00
|
|
|
void *ret = NULL;
|
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
ret = internal_ao2_iterator_next(a, &p);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
|
|
|
if (p) {
|
|
|
|
/* inc refcount of returned object */
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref(ret, 1);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2009-10-06 01:24:24 +00:00
|
|
|
if (!(a->flags & AO2_ITERATOR_DONTLOCK))
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ao2_unlock(a->c);
|
|
|
|
|
2007-09-10 16:19:45 +00:00
|
|
|
return ret;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* callback for destroying container.
|
|
|
|
* we can make it simple as we know what it does
|
|
|
|
*/
|
2008-11-25 01:01:49 +00:00
|
|
|
static int cd_cb(void *obj, void *arg, int flag)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref(obj, -1);
|
2008-04-16 17:45:28 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-11-25 01:01:49 +00:00
|
|
|
static int cd_cb_debug(void *obj, void *arg, int flag)
|
2008-04-16 17:45:28 +00:00
|
|
|
{
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_ref_debug(obj, -1, "deref object via container destroy", __FILE__, __LINE__, __PRETTY_FUNCTION__);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void container_destruct(void *_c)
|
|
|
|
{
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_container *c = _c;
|
2008-06-17 15:52:13 +00:00
|
|
|
int i;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_callback(c, OBJ_UNLINK, cd_cb, NULL);
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2008-06-17 15:52:13 +00:00
|
|
|
for (i = 0; i < c->n_buckets; i++) {
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *current;
|
2008-06-17 15:52:13 +00:00
|
|
|
|
2008-08-10 19:35:50 +00:00
|
|
|
while ((current = AST_LIST_REMOVE_HEAD(&c->buckets[i], entry))) {
|
|
|
|
ast_free(current);
|
2008-06-17 15:52:13 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-04-16 17:45:28 +00:00
|
|
|
#ifdef AO2_DEBUG
|
|
|
|
ast_atomic_fetchadd_int(&ao2.total_containers, -1);
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
static void container_destruct_debug(void *_c)
|
|
|
|
{
|
|
|
|
struct ao2_container *c = _c;
|
2008-06-17 20:19:49 +00:00
|
|
|
int i;
|
2008-04-16 17:45:28 +00:00
|
|
|
|
2009-03-18 02:21:23 +00:00
|
|
|
__ao2_callback_debug(c, OBJ_UNLINK, cd_cb_debug, NULL, "container_destruct_debug called", __FILE__, __LINE__, __PRETTY_FUNCTION__);
|
2007-09-13 18:52:35 +00:00
|
|
|
|
2008-06-17 20:19:49 +00:00
|
|
|
for (i = 0; i < c->n_buckets; i++) {
|
2010-05-16 11:14:37 +00:00
|
|
|
struct bucket_entry *current;
|
2008-06-17 20:19:49 +00:00
|
|
|
|
2008-08-10 19:35:50 +00:00
|
|
|
while ((current = AST_LIST_REMOVE_HEAD(&c->buckets[i], entry))) {
|
|
|
|
ast_free(current);
|
2008-06-17 20:19:49 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_atomic_fetchadd_int(&ao2.total_containers, -1);
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
2007-09-13 19:16:24 +00:00
|
|
|
#ifdef AO2_DEBUG
|
2008-11-25 01:01:49 +00:00
|
|
|
static int print_cb(void *obj, void *arg, int flag)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2009-05-21 21:13:09 +00:00
|
|
|
struct ast_cli_args *a = (struct ast_cli_args *) arg;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
char *s = (char *)obj;
|
|
|
|
|
2009-05-21 21:13:09 +00:00
|
|
|
ast_cli(a->fd, "string <%s>\n", s);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Print stats
|
|
|
|
*/
|
2007-10-11 19:03:06 +00:00
|
|
|
static char *handle_astobj2_stats(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2007-10-11 19:03:06 +00:00
|
|
|
switch (cmd) {
|
|
|
|
case CLI_INIT:
|
2008-09-28 23:32:14 +00:00
|
|
|
e->command = "astobj2 show stats";
|
|
|
|
e->usage = "Usage: astobj2 show stats\n"
|
|
|
|
" Show astobj2 show stats\n";
|
2007-10-11 19:03:06 +00:00
|
|
|
return NULL;
|
|
|
|
case CLI_GENERATE:
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
ast_cli(a->fd, "Objects : %d\n", ao2.total_objects);
|
|
|
|
ast_cli(a->fd, "Containers : %d\n", ao2.total_containers);
|
|
|
|
ast_cli(a->fd, "Memory : %d\n", ao2.total_mem);
|
|
|
|
ast_cli(a->fd, "Locked : %d\n", ao2.total_locked);
|
|
|
|
ast_cli(a->fd, "Refs : %d\n", ao2.total_refs);
|
|
|
|
return CLI_SUCCESS;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This is testing code for astobj
|
|
|
|
*/
|
2007-10-11 19:03:06 +00:00
|
|
|
static char *handle_astobj2_test(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_container *c1;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
int i, lim;
|
|
|
|
char *obj;
|
|
|
|
static int prof_id = -1;
|
2007-10-11 19:03:06 +00:00
|
|
|
struct ast_cli_args fake_args = { a->fd, 0, NULL };
|
|
|
|
|
|
|
|
switch (cmd) {
|
|
|
|
case CLI_INIT:
|
|
|
|
e->command = "astobj2 test";
|
|
|
|
e->usage = "Usage: astobj2 test <num>\n"
|
|
|
|
" Runs astobj2 test. Creates 'num' objects,\n"
|
|
|
|
" and test iterators, callbacks and may be other stuff\n";
|
|
|
|
return NULL;
|
|
|
|
case CLI_GENERATE:
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (a->argc != 3) {
|
|
|
|
return CLI_SHOWUSAGE;
|
|
|
|
}
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
if (prof_id == -1)
|
|
|
|
prof_id = ast_add_profile("ao2_alloc", 0);
|
|
|
|
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "argc %d argv %s %s %s\n", a->argc, a->argv[0], a->argv[1], a->argv[2]);
|
|
|
|
lim = atoi(a->argv[2]);
|
|
|
|
ast_cli(a->fd, "called astobj_test\n");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2007-10-11 19:03:06 +00:00
|
|
|
handle_astobj2_stats(e, CLI_HANDLER, &fake_args);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
/*
|
|
|
|
* allocate a container with no default callback, and no hash function.
|
|
|
|
* No hash means everything goes in the same bucket.
|
|
|
|
*/
|
2008-04-16 17:45:28 +00:00
|
|
|
c1 = ao2_t_container_alloc(100, NULL /* no callback */, NULL /* no hash */,"test");
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "container allocated as %p\n", c1);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* fill the container with objects.
|
|
|
|
* ao2_alloc() gives us a reference which we pass to the
|
|
|
|
* container when we do the insert.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < lim; i++) {
|
|
|
|
ast_mark(prof_id, 1 /* start */);
|
2008-04-16 17:45:28 +00:00
|
|
|
obj = ao2_t_alloc(80, NULL,"test");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_mark(prof_id, 0 /* stop */);
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "object %d allocated as %p\n", i, obj);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
sprintf(obj, "-- this is obj %d --", i);
|
|
|
|
ao2_link(c1, obj);
|
2008-06-17 22:24:45 +00:00
|
|
|
/* At this point, the refcount on obj is 2 due to the allocation
|
|
|
|
* and linking. We can go ahead and reduce the refcount by 1
|
|
|
|
* right here so that when the container is unreffed later, the
|
|
|
|
* objects will be freed
|
|
|
|
*/
|
|
|
|
ao2_t_ref(obj, -1, "test");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "testing callbacks\n");
|
2009-05-21 21:13:09 +00:00
|
|
|
ao2_t_callback(c1, 0, print_cb, a, "test callback");
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "testing iterators, remove every second object\n");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
{
|
2007-09-04 18:40:07 +00:00
|
|
|
struct ao2_iterator ai;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
int x = 0;
|
|
|
|
|
|
|
|
ai = ao2_iterator_init(c1, 0);
|
2008-04-16 17:45:28 +00:00
|
|
|
while ( (obj = ao2_t_iterator_next(&ai,"test")) ) {
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "iterator on <%s>\n", obj);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
if (x++ & 1)
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_t_unlink(c1, obj,"test");
|
|
|
|
ao2_t_ref(obj, -1,"test");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
2010-02-05 04:22:15 +00:00
|
|
|
ao2_iterator_destroy(&ai);
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "testing iterators again\n");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ai = ao2_iterator_init(c1, 0);
|
2008-04-16 17:45:28 +00:00
|
|
|
while ( (obj = ao2_t_iterator_next(&ai,"test")) ) {
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "iterator on <%s>\n", obj);
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_t_ref(obj, -1,"test");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
2010-02-05 04:22:15 +00:00
|
|
|
ao2_iterator_destroy(&ai);
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "testing callbacks again\n");
|
2009-05-21 21:13:09 +00:00
|
|
|
ao2_t_callback(c1, 0, print_cb, a, "test callback");
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
ast_verbose("now you should see an error message:\n");
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_t_ref(&i, -1, ""); /* i is not a valid object so we print an error here */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
2007-10-11 19:03:06 +00:00
|
|
|
ast_cli(a->fd, "destroy container\n");
|
2008-04-16 17:45:28 +00:00
|
|
|
ao2_t_ref(c1, -1, ""); /* destroy container */
|
2007-10-11 19:03:06 +00:00
|
|
|
handle_astobj2_stats(e, CLI_HANDLER, &fake_args);
|
|
|
|
return CLI_SUCCESS;
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static struct ast_cli_entry cli_astobj2[] = {
|
2007-10-22 20:05:18 +00:00
|
|
|
AST_CLI_DEFINE(handle_astobj2_stats, "Print astobj2 statistics"),
|
|
|
|
AST_CLI_DEFINE(handle_astobj2_test, "Test astobj2"),
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
};
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif /* AO2_DEBUG */
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
|
|
|
|
int astobj2_init(void)
|
|
|
|
{
|
2007-09-13 18:52:35 +00:00
|
|
|
#ifdef AO2_DEBUG
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
ast_cli_register_multiple(cli_astobj2, ARRAY_LEN(cli_astobj2));
|
2007-09-13 18:52:35 +00:00
|
|
|
#endif
|
|
|
|
|
Merged revisions 80362 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r80362 | russell | 2007-08-22 15:21:36 -0500 (Wed, 22 Aug 2007) | 34 lines
Merge changes from team/russell/iax_refcount.
This set of changes fixes problems with the handling of iax2_user and iax2_peer
objects. It was very possible for a thread to still hold a reference to one of
these objects while a reload operation tries to delete them. The fix here is to
ensure that all references to these objects are tracked so that they can't go away
while still in use.
To accomplish this, I used the astobj2 reference counted object model. This
code has been in one of Luigi Rizzo's branches for a long time and was primarily
developed by one of his students, Marta Carbone. I wanted to go ahead and bring
this in to 1.4 because there are other problems similar to the ones fixed by these
changes, so we might as well go ahead and use the new astobj if we're going to go
through all of the work necessary to fix the problems.
As a nice side benefit of these changes, peer and user handling got more efficient.
Using astobj2 lets us not hold the container lock for peers or users nearly as long
while iterating. Also, by changing a define at the top of chan_iax2.c, the objects
will be distributed in a hash table, drastically increasing lookup speed in these
containers, which will have a very big impact on systems that have a large number of
users or peers.
The use of the hash table will be made the default in trunk. It is not the default
in 1.4 because it changes the behavior slightly. Previously, since peers and users
were stored in memory in the same order they were specified in the configuration file,
you could influence peer and user matching order based on the order they are specified
in the configuration. The hash table does not guarantee any order in the container,
so this behavior will be going away. It just means that you have to be a little
more careful ensuring that peers and users are matched explicitly and not forcing
chan_iax2 to have to guess which user is the right one based on secret, host, and
access list settings, instead of simply using the username.
If you have any questions, feel free to ask on the asterisk-dev list.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@80387 65c4cc65-6c06-0410-ace0-fbb531ad65f3
2007-08-22 20:44:23 +00:00
|
|
|
return 0;
|
|
|
|
}
|