asterisk/main/sched.c

869 lines
21 KiB
C
Raw Normal View History

/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 1999 - 2010, Digium, Inc.
*
* Mark Spencer <markster@digium.com>
* Russell Bryant <russell@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*! \file
*
* \brief Scheduler Routines (from cheops-NG)
*
* \author Mark Spencer <markster@digium.com>
*/
/*** MODULEINFO
<support_level>core</support_level>
***/
#include "asterisk.h"
#ifdef DEBUG_SCHEDULER
#define DEBUG(a) a
#else
#define DEBUG(a)
#endif
#include <sys/time.h>
#include "asterisk/sched.h"
#include "asterisk/channel.h"
#include "asterisk/lock.h"
#include "asterisk/utils.h"
#include "asterisk/heap.h"
#include "asterisk/threadstorage.h"
/*!
* \brief Max num of schedule structs
*
* \note The max number of schedule structs to keep around
* for use. Undefine to disable schedule structure
* caching. (Only disable this on very low memory
* machines)
*/
#define SCHED_MAX_CACHE 128
AST_THREADSTORAGE(last_del_id);
/*!
* \brief Scheduler ID holder
*
* These form a queue on a scheduler context. When a new
* scheduled item is created, a sched_id is popped off the
* queue and its id is assigned to the new scheduled item.
* When the scheduled task is complete, the sched_id on that
* task is then pushed to the back of the queue to be re-used
* on some future scheduled item.
*/
struct sched_id {
/*! Immutable ID number that is copied onto the scheduled task */
int id;
AST_LIST_ENTRY(sched_id) list;
};
struct sched {
AST_LIST_ENTRY(sched) list;
/*! The ID that has been popped off the scheduler context's queue */
struct sched_id *sched_id;
struct timeval when; /*!< Absolute time event should take place */
/*!
* \brief Tie breaker in case the when is the same for multiple entries.
*
* \note The oldest expiring entry in the scheduler heap goes first.
* This is possible when multiple events are scheduled to expire at
* the same time by internal coding.
*/
unsigned int tie_breaker;
int resched; /*!< When to reschedule */
int variable; /*!< Use return value from callback to reschedule */
const void *data; /*!< Data */
ast_sched_cb callback; /*!< Callback */
ssize_t __heap_index;
/*!
* Used to synchronize between thread running a task and thread
* attempting to delete a task
*/
ast_cond_t cond;
/*! Indication that a running task was deleted. */
unsigned int deleted:1;
/*! Indication that a running task was rescheduled. */
unsigned int rescheduled:1;
};
struct sched_thread {
pthread_t thread;
ast_cond_t cond;
unsigned int stop:1;
};
struct ast_sched_context {
ast_mutex_t lock;
unsigned int eventcnt; /*!< Number of events processed */
unsigned int highwater; /*!< highest count so far */
/*! Next tie breaker in case events expire at the same time. */
unsigned int tie_breaker;
struct ast_heap *sched_heap;
struct sched_thread *sched_thread;
/*! The scheduled task that is currently executing */
struct sched *currently_executing;
sched: Don't allow ast_sched_del to deadlock ast_sched_runq from same thread When fixing ASTERISK~24212, a change was done so a scheduled callback could not be removed while it was running. The caller of ast_sched_del would have to wait. However, when the caller of ast_sched_del is the callback itself (however wrong this might be), this new check would cause a deadlock: it would wait forever for itself. This changeset introduces an additional check: if ast_sched_del is called by the callback itself, it is immediately rejected (along with an ERROR log and a backtrace). Additionally, the AST_SCHED_DEL_UNREF macro is adjusted so the after-ast_sched_del-refcall function is only run if ast_sched_del returned success. This should fix the following spurious race condition found in chan_sip: - thread 1: schedule sip_poke_peer_now (using AST_SCHED_REPLACE) - thread 2: run sip_poke_peer_now - thread 2: blank out sched-ID (too soon!) - thread 1: set sched-ID (too late!) - thread 2: try to delete the currently running sched-ID After this fix, an ERROR would be logged, but no deadlocks (in do_monitor) nor excess calls to sip_unref_peer(peer) (causing double frees of rtp_instances and other madness) should occur. (Thanks Richard Mudgett for reviewing/improving this "scary" change.) Note that this change does not fix the observed race condition: unlocked access to peer->pokeexpire (and potentially other scheduled items in chan_sip), causing AST_SCHED_DEL_UNREF to look at a changing id. But it will make the deadlock go away. And in the observed case, it will not have adverse affects (like memory leaks) because the scheduled item is removed through a different path. ASTERISK-28282 Change-Id: Ic26777fa0732725e6ca7010df17af77a012aa856
2019-07-17 13:06:12 +00:00
/*! Valid while currently_executing is not NULL */
pthread_t executing_thread_id;
#ifdef SCHED_MAX_CACHE
AST_LIST_HEAD_NOLOCK(, sched) schedc; /*!< Cache of unused schedule structures and how many */
unsigned int schedccnt;
#endif
/*! Queue of scheduler task IDs to assign */
AST_LIST_HEAD_NOLOCK(, sched_id) id_queue;
/*! The number of IDs in the id_queue */
int id_queue_size;
};
static void *sched_run(void *data)
{
struct ast_sched_context *con = data;
while (!con->sched_thread->stop) {
int ms;
struct timespec ts = {
.tv_sec = 0,
};
ast_mutex_lock(&con->lock);
if (con->sched_thread->stop) {
ast_mutex_unlock(&con->lock);
return NULL;
}
ms = ast_sched_wait(con);
if (ms == -1) {
ast_cond_wait(&con->sched_thread->cond, &con->lock);
} else {
struct timeval tv;
tv = ast_tvadd(ast_tvnow(), ast_samp2tv(ms, 1000));
ts.tv_sec = tv.tv_sec;
ts.tv_nsec = tv.tv_usec * 1000;
ast_cond_timedwait(&con->sched_thread->cond, &con->lock, &ts);
}
ast_mutex_unlock(&con->lock);
if (con->sched_thread->stop) {
return NULL;
}
ast_sched_runq(con);
}
return NULL;
}
static void sched_thread_destroy(struct ast_sched_context *con)
{
if (!con->sched_thread) {
return;
}
if (con->sched_thread->thread != AST_PTHREADT_NULL) {
ast_mutex_lock(&con->lock);
con->sched_thread->stop = 1;
ast_cond_signal(&con->sched_thread->cond);
ast_mutex_unlock(&con->lock);
pthread_join(con->sched_thread->thread, NULL);
con->sched_thread->thread = AST_PTHREADT_NULL;
}
ast_cond_destroy(&con->sched_thread->cond);
ast_free(con->sched_thread);
con->sched_thread = NULL;
}
int ast_sched_start_thread(struct ast_sched_context *con)
{
struct sched_thread *st;
if (con->sched_thread) {
ast_log(LOG_ERROR, "Thread already started on this scheduler context\n");
return -1;
}
if (!(st = ast_calloc(1, sizeof(*st)))) {
return -1;
}
ast_cond_init(&st->cond, NULL);
st->thread = AST_PTHREADT_NULL;
con->sched_thread = st;
if (ast_pthread_create_background(&st->thread, NULL, sched_run, con)) {
ast_log(LOG_ERROR, "Failed to create scheduler thread\n");
sched_thread_destroy(con);
return -1;
}
return 0;
}
static int sched_time_cmp(void *va, void *vb)
{
struct sched *a = va;
struct sched *b = vb;
int cmp;
cmp = ast_tvcmp(b->when, a->when);
if (!cmp) {
cmp = b->tie_breaker - a->tie_breaker;
}
return cmp;
}
struct ast_sched_context *ast_sched_context_create(void)
{
struct ast_sched_context *tmp;
if (!(tmp = ast_calloc(1, sizeof(*tmp)))) {
return NULL;
}
ast_mutex_init(&tmp->lock);
tmp->eventcnt = 1;
AST_LIST_HEAD_INIT_NOLOCK(&tmp->id_queue);
if (!(tmp->sched_heap = ast_heap_create(8, sched_time_cmp,
offsetof(struct sched, __heap_index)))) {
ast_sched_context_destroy(tmp);
return NULL;
}
return tmp;
}
static void sched_free(struct sched *task)
{
/* task->sched_id will be NULL most of the time, but when the
* scheduler context shuts down, it will free all scheduled
* tasks, and in that case, the task->sched_id will be non-NULL
*/
ast_free(task->sched_id);
ast_cond_destroy(&task->cond);
ast_free(task);
}
void ast_sched_context_destroy(struct ast_sched_context *con)
{
struct sched *s;
struct sched_id *sid;
sched_thread_destroy(con);
con->sched_thread = NULL;
ast_mutex_lock(&con->lock);
#ifdef SCHED_MAX_CACHE
while ((s = AST_LIST_REMOVE_HEAD(&con->schedc, list))) {
sched_free(s);
}
#endif
if (con->sched_heap) {
while ((s = ast_heap_pop(con->sched_heap))) {
sched_free(s);
}
ast_heap_destroy(con->sched_heap);
con->sched_heap = NULL;
}
while ((sid = AST_LIST_REMOVE_HEAD(&con->id_queue, list))) {
ast_free(sid);
}
ast_mutex_unlock(&con->lock);
ast_mutex_destroy(&con->lock);
ast_free(con);
}
#define ID_QUEUE_INCREMENT 16
/*!
* \brief Add new scheduler IDs to the queue.
*
* \retval The number of IDs added to the queue
*/
static int add_ids(struct ast_sched_context *con)
{
int new_size;
int original_size;
int i;
original_size = con->id_queue_size;
/* So we don't go overboard with the mallocs here, we'll just up
* the size of the list by a fixed amount each time instead of
* multiplying the size by any particular factor
*/
new_size = original_size + ID_QUEUE_INCREMENT;
if (new_size < 0) {
/* Overflow. Cap it at INT_MAX. */
new_size = INT_MAX;
}
for (i = original_size; i < new_size; ++i) {
struct sched_id *new_id;
new_id = ast_calloc(1, sizeof(*new_id));
if (!new_id) {
break;
}
/*
* According to the API doxygen a sched ID of 0 is valid.
* Unfortunately, 0 was never returned historically and
* several users incorrectly coded usage of the returned
* sched ID assuming that 0 was invalid.
*/
new_id->id = ++con->id_queue_size;
AST_LIST_INSERT_TAIL(&con->id_queue, new_id, list);
}
return con->id_queue_size - original_size;
}
static int set_sched_id(struct ast_sched_context *con, struct sched *new_sched)
{
if (AST_LIST_EMPTY(&con->id_queue) && (add_ids(con) == 0)) {
return -1;
}
new_sched->sched_id = AST_LIST_REMOVE_HEAD(&con->id_queue, list);
return 0;
}
static void sched_release(struct ast_sched_context *con, struct sched *tmp)
{
if (tmp->sched_id) {
AST_LIST_INSERT_TAIL(&con->id_queue, tmp->sched_id, list);
tmp->sched_id = NULL;
}
/*
* Add to the cache, or just free() if we
* already have too many cache entries
*/
#ifdef SCHED_MAX_CACHE
if (con->schedccnt < SCHED_MAX_CACHE) {
AST_LIST_INSERT_HEAD(&con->schedc, tmp, list);
con->schedccnt++;
} else
#endif
sched_free(tmp);
}
static struct sched *sched_alloc(struct ast_sched_context *con)
{
struct sched *tmp;
/*
* We keep a small cache of schedule entries
* to minimize the number of necessary malloc()'s
*/
#ifdef SCHED_MAX_CACHE
if ((tmp = AST_LIST_REMOVE_HEAD(&con->schedc, list))) {
con->schedccnt--;
} else
#endif
{
tmp = ast_calloc(1, sizeof(*tmp));
if (!tmp) {
return NULL;
}
ast_cond_init(&tmp->cond, NULL);
}
if (set_sched_id(con, tmp)) {
sched_release(con, tmp);
return NULL;
}
return tmp;
}
void ast_sched_clean_by_callback(struct ast_sched_context *con, ast_sched_cb match, ast_sched_cb cleanup_cb)
{
int i = 1;
struct sched *current;
ast_mutex_lock(&con->lock);
while ((current = ast_heap_peek(con->sched_heap, i))) {
if (current->callback != match) {
i++;
continue;
}
ast_heap_remove(con->sched_heap, current);
cleanup_cb(current->data);
sched_release(con, current);
}
ast_mutex_unlock(&con->lock);
}
/*! \brief
* Return the number of milliseconds
* until the next scheduled event
*/
int ast_sched_wait(struct ast_sched_context *con)
{
int ms;
struct sched *s;
DEBUG(ast_debug(1, "ast_sched_wait()\n"));
ast_mutex_lock(&con->lock);
if ((s = ast_heap_peek(con->sched_heap, 1))) {
ms = ast_tvdiff_ms(s->when, ast_tvnow());
if (ms < 0) {
ms = 0;
}
} else {
ms = -1;
}
ast_mutex_unlock(&con->lock);
return ms;
}
/*! \brief
* Take a sched structure and put it in the
* queue, such that the soonest event is
* first in the list.
*/
static void schedule(struct ast_sched_context *con, struct sched *s)
{
size_t size;
size = ast_heap_size(con->sched_heap);
/* Record the largest the scheduler heap became for reporting purposes. */
if (con->highwater <= size) {
con->highwater = size + 1;
}
/* Determine the tie breaker value for the new entry. */
if (size) {
++con->tie_breaker;
} else {
/*
* Restart the sequence for the first entry to make integer
* roll over more unlikely.
*/
con->tie_breaker = 0;
}
s->tie_breaker = con->tie_breaker;
ast_heap_push(con->sched_heap, s);
}
/*! \brief
* given the last event *tv and the offset in milliseconds 'when',
* computes the next value,
*/
static void sched_settime(struct timeval *t, int when)
{
struct timeval now = ast_tvnow();
if (when < 0) {
/*
* A negative when value is likely a bug as it
* represents a VERY large timeout time.
*/
ast_log(LOG_WARNING,
"Bug likely: Negative time interval %d (interpreted as %u ms) requested!\n",
when, (unsigned int) when);
ast_assert(0);
}
/*ast_debug(1, "TV -> %lu,%lu\n", tv->tv_sec, tv->tv_usec);*/
if (ast_tvzero(*t)) /* not supplied, default to now */
*t = now;
*t = ast_tvadd(*t, ast_samp2tv(when, 1000));
if (ast_tvcmp(*t, now) < 0) {
*t = now;
}
}
int ast_sched_replace_variable(int old_id, struct ast_sched_context *con, int when, ast_sched_cb callback, const void *data, int variable)
{
/* 0 means the schedule item is new; do not delete */
if (old_id > 0) {
AST_SCHED_DEL(con, old_id);
}
return ast_sched_add_variable(con, when, callback, data, variable);
}
/*! \brief
* Schedule callback(data) to happen when ms into the future
*/
int ast_sched_add_variable(struct ast_sched_context *con, int when, ast_sched_cb callback, const void *data, int variable)
{
struct sched *tmp;
int res = -1;
DEBUG(ast_debug(1, "ast_sched_add()\n"));
ast_mutex_lock(&con->lock);
if ((tmp = sched_alloc(con))) {
con->eventcnt++;
tmp->callback = callback;
tmp->data = data;
tmp->resched = when;
tmp->variable = variable;
tmp->when = ast_tv(0, 0);
tmp->deleted = 0;
sched_settime(&tmp->when, when);
schedule(con, tmp);
res = tmp->sched_id->id;
}
#ifdef DUMP_SCHEDULER
/* Dump contents of the context while we have the lock so nothing gets screwed up by accident. */
ast_sched_dump(con);
#endif
if (con->sched_thread) {
ast_cond_signal(&con->sched_thread->cond);
}
ast_mutex_unlock(&con->lock);
return res;
}
int ast_sched_replace(int old_id, struct ast_sched_context *con, int when, ast_sched_cb callback, const void *data)
{
if (old_id > -1) {
AST_SCHED_DEL(con, old_id);
}
return ast_sched_add(con, when, callback, data);
}
int ast_sched_add(struct ast_sched_context *con, int when, ast_sched_cb callback, const void *data)
{
return ast_sched_add_variable(con, when, callback, data, 0);
}
static struct sched *sched_find(struct ast_sched_context *con, int id)
{
int x;
size_t heap_size;
heap_size = ast_heap_size(con->sched_heap);
for (x = 1; x <= heap_size; x++) {
struct sched *cur = ast_heap_peek(con->sched_heap, x);
if (cur->sched_id->id == id) {
return cur;
}
}
return NULL;
}
const void *ast_sched_find_data(struct ast_sched_context *con, int id)
{
struct sched *s;
const void *data = NULL;
ast_mutex_lock(&con->lock);
s = sched_find(con, id);
if (s) {
data = s->data;
}
ast_mutex_unlock(&con->lock);
return data;
}
/*! \brief
* Delete the schedule entry with number
* "id". It's nearly impossible that there
* would be two or more in the list with that
* id.
* Deprecated in favor of ast_sched_del_nonrunning
* which checks running event status.
*/
int ast_sched_del(struct ast_sched_context *con, int id)
{
return ast_sched_del_nonrunning(con, id) ? -1 : 0;
}
/*! \brief
* Delete the schedule entry with number "id".
* If running, wait for the task to complete,
* check to see if it is rescheduled then
* schedule the release.
* It's nearly impossible that there would be
* two or more in the list with that id.
*/
int ast_sched_del_nonrunning(struct ast_sched_context *con, int id)
{
struct sched *s = NULL;
int *last_id = ast_threadstorage_get(&last_del_id, sizeof(int));
int res = 0;
DEBUG(ast_debug(1, "ast_sched_del(%d)\n", id));
if (id < 0) {
return 0;
}
ast_mutex_lock(&con->lock);
s = sched_find(con, id);
if (s) {
if (!ast_heap_remove(con->sched_heap, s)) {
ast_log(LOG_WARNING,"sched entry %d not in the sched heap?\n", s->sched_id->id);
}
sched_release(con, s);
} else if (con->currently_executing && (id == con->currently_executing->sched_id->id)) {
sched: Don't allow ast_sched_del to deadlock ast_sched_runq from same thread When fixing ASTERISK~24212, a change was done so a scheduled callback could not be removed while it was running. The caller of ast_sched_del would have to wait. However, when the caller of ast_sched_del is the callback itself (however wrong this might be), this new check would cause a deadlock: it would wait forever for itself. This changeset introduces an additional check: if ast_sched_del is called by the callback itself, it is immediately rejected (along with an ERROR log and a backtrace). Additionally, the AST_SCHED_DEL_UNREF macro is adjusted so the after-ast_sched_del-refcall function is only run if ast_sched_del returned success. This should fix the following spurious race condition found in chan_sip: - thread 1: schedule sip_poke_peer_now (using AST_SCHED_REPLACE) - thread 2: run sip_poke_peer_now - thread 2: blank out sched-ID (too soon!) - thread 1: set sched-ID (too late!) - thread 2: try to delete the currently running sched-ID After this fix, an ERROR would be logged, but no deadlocks (in do_monitor) nor excess calls to sip_unref_peer(peer) (causing double frees of rtp_instances and other madness) should occur. (Thanks Richard Mudgett for reviewing/improving this "scary" change.) Note that this change does not fix the observed race condition: unlocked access to peer->pokeexpire (and potentially other scheduled items in chan_sip), causing AST_SCHED_DEL_UNREF to look at a changing id. But it will make the deadlock go away. And in the observed case, it will not have adverse affects (like memory leaks) because the scheduled item is removed through a different path. ASTERISK-28282 Change-Id: Ic26777fa0732725e6ca7010df17af77a012aa856
2019-07-17 13:06:12 +00:00
if (con->executing_thread_id == pthread_self()) {
/* The scheduled callback is trying to delete itself.
* Not good as that is a deadlock. */
ast_log(LOG_ERROR,
"BUG! Trying to delete sched %d from within the callback %p. "
"Ignoring so we don't deadlock\n",
id, con->currently_executing->callback);
ast_log_backtrace();
/* We'll return -1 below because s is NULL.
* The caller will rightly assume that the unscheduling failed. */
} else {
s = con->currently_executing;
s->deleted = 1;
/* Wait for executing task to complete so that the caller of
* ast_sched_del() does not free memory out from under the task. */
while (con->currently_executing && (id == con->currently_executing->sched_id->id)) {
ast_cond_wait(&s->cond, &con->lock);
}
/* This is not rescheduled so the caller of ast_sched_del_nonrunning needs to know
* that it was still deleted
*/
if (!s->rescheduled) {
res = -2;
}
/* ast_sched_runq knows we are waiting on this item and is passing responsibility for
* its destruction to us
*/
sched_release(con, s);
s = NULL;
}
}
#ifdef DUMP_SCHEDULER
/* Dump contents of the context while we have the lock so nothing gets screwed up by accident. */
ast_sched_dump(con);
#endif
if (con->sched_thread) {
ast_cond_signal(&con->sched_thread->cond);
}
ast_mutex_unlock(&con->lock);
if(res == -2){
return res;
}
else if (!s && *last_id != id) {
ast_debug(1, "Attempted to delete nonexistent schedule entry %d!\n", id);
/* Removing nonexistent schedule entry shouldn't trigger assert (it was enabled in DEV_MODE);
* because in many places entries is deleted without having valid id. */
*last_id = id;
return -1;
} else if (!s) {
return -1;
}
return res;
}
void ast_sched_report(struct ast_sched_context *con, struct ast_str **buf, struct ast_cb_names *cbnames)
{
int i, x;
struct sched *cur;
int countlist[cbnames->numassocs + 1];
size_t heap_size;
memset(countlist, 0, sizeof(countlist));
ast_str_set(buf, 0, " Highwater = %u\n schedcnt = %zu\n", con->highwater, ast_heap_size(con->sched_heap));
ast_mutex_lock(&con->lock);
heap_size = ast_heap_size(con->sched_heap);
for (x = 1; x <= heap_size; x++) {
cur = ast_heap_peek(con->sched_heap, x);
/* match the callback to the cblist */
for (i = 0; i < cbnames->numassocs; i++) {
if (cur->callback == cbnames->cblist[i]) {
break;
}
}
if (i < cbnames->numassocs) {
countlist[i]++;
} else {
countlist[cbnames->numassocs]++;
}
}
ast_mutex_unlock(&con->lock);
for (i = 0; i < cbnames->numassocs; i++) {
ast_str_append(buf, 0, " %s : %d\n", cbnames->list[i], countlist[i]);
}
ast_str_append(buf, 0, " <unknown> : %d\n", countlist[cbnames->numassocs]);
}
/*! \brief Dump the contents of the scheduler to LOG_DEBUG */
void ast_sched_dump(struct ast_sched_context *con)
{
struct sched *q;
struct timeval when;
int x;
size_t heap_size;
if (!DEBUG_ATLEAST(1)) {
return;
}
when = ast_tvnow();
#ifdef SCHED_MAX_CACHE
ast_log(LOG_DEBUG, "Asterisk Schedule Dump (%zu in Q, %u Total, %u Cache, %u high-water)\n",
ast_heap_size(con->sched_heap), con->eventcnt - 1, con->schedccnt, con->highwater);
#else
ast_log(LOG_DEBUG, "Asterisk Schedule Dump (%zu in Q, %u Total, %u high-water)\n",
ast_heap_size(con->sched_heap), con->eventcnt - 1, con->highwater);
#endif
ast_log(LOG_DEBUG, "=============================================================\n");
ast_log(LOG_DEBUG, "|ID Callback Data Time (sec:ms) |\n");
ast_log(LOG_DEBUG, "+-----+-----------------+-----------------+-----------------+\n");
ast_mutex_lock(&con->lock);
heap_size = ast_heap_size(con->sched_heap);
for (x = 1; x <= heap_size; x++) {
struct timeval delta;
q = ast_heap_peek(con->sched_heap, x);
delta = ast_tvsub(q->when, when);
ast_log(LOG_DEBUG, "|%.4d | %-15p | %-15p | %.6ld : %.6ld |\n",
q->sched_id->id,
q->callback,
q->data,
(long)delta.tv_sec,
(long int)delta.tv_usec);
}
ast_mutex_unlock(&con->lock);
ast_log(LOG_DEBUG, "=============================================================\n");
}
/*! \brief
* Launch all events which need to be run at this time.
*/
int ast_sched_runq(struct ast_sched_context *con)
{
struct sched *current;
struct timeval when;
int numevents;
int res;
DEBUG(ast_debug(1, "ast_sched_runq()\n"));
ast_mutex_lock(&con->lock);
when = ast_tvadd(ast_tvnow(), ast_tv(0, 1000));
for (numevents = 0; (current = ast_heap_peek(con->sched_heap, 1)); numevents++) {
/* schedule all events which are going to expire within 1ms.
* We only care about millisecond accuracy anyway, so this will
* help us get more than one event at one time if they are very
* close together.
*/
if (ast_tvcmp(current->when, when) != -1) {
break;
}
current = ast_heap_pop(con->sched_heap);
/*
* At this point, the schedule queue is still intact. We
* have removed the first event and the rest is still there,
* so it's permissible for the callback to add new events, but
* trying to delete itself won't work because it isn't in
* the schedule queue. If that's what it wants to do, it
* should return 0.
*/
con->currently_executing = current;
sched: Don't allow ast_sched_del to deadlock ast_sched_runq from same thread When fixing ASTERISK~24212, a change was done so a scheduled callback could not be removed while it was running. The caller of ast_sched_del would have to wait. However, when the caller of ast_sched_del is the callback itself (however wrong this might be), this new check would cause a deadlock: it would wait forever for itself. This changeset introduces an additional check: if ast_sched_del is called by the callback itself, it is immediately rejected (along with an ERROR log and a backtrace). Additionally, the AST_SCHED_DEL_UNREF macro is adjusted so the after-ast_sched_del-refcall function is only run if ast_sched_del returned success. This should fix the following spurious race condition found in chan_sip: - thread 1: schedule sip_poke_peer_now (using AST_SCHED_REPLACE) - thread 2: run sip_poke_peer_now - thread 2: blank out sched-ID (too soon!) - thread 1: set sched-ID (too late!) - thread 2: try to delete the currently running sched-ID After this fix, an ERROR would be logged, but no deadlocks (in do_monitor) nor excess calls to sip_unref_peer(peer) (causing double frees of rtp_instances and other madness) should occur. (Thanks Richard Mudgett for reviewing/improving this "scary" change.) Note that this change does not fix the observed race condition: unlocked access to peer->pokeexpire (and potentially other scheduled items in chan_sip), causing AST_SCHED_DEL_UNREF to look at a changing id. But it will make the deadlock go away. And in the observed case, it will not have adverse affects (like memory leaks) because the scheduled item is removed through a different path. ASTERISK-28282 Change-Id: Ic26777fa0732725e6ca7010df17af77a012aa856
2019-07-17 13:06:12 +00:00
con->executing_thread_id = pthread_self();
ast_mutex_unlock(&con->lock);
res = current->callback(current->data);
ast_mutex_lock(&con->lock);
con->currently_executing = NULL;
ast_cond_signal(&current->cond);
if (current->deleted) {
/*
* Another thread is waiting on this scheduled item. That thread
* will be responsible for it's destruction
*/
current->rescheduled = res ? 1 : 0;
} else if (res) {
/*
* If they return non-zero, we should schedule them to be
* run again.
*/
sched_settime(&current->when, current->variable ? res : current->resched);
schedule(con, current);
} else {
/* No longer needed, so release it */
sched_release(con, current);
}
}
ast_mutex_unlock(&con->lock);
return numevents;
}
long ast_sched_when(struct ast_sched_context *con,int id)
{
struct sched *s;
long secs = -1;
DEBUG(ast_debug(1, "ast_sched_when()\n"));
ast_mutex_lock(&con->lock);
s = sched_find(con, id);
if (s) {
struct timeval now = ast_tvnow();
secs = s->when.tv_sec - now.tv_sec;
}
ast_mutex_unlock(&con->lock);
return secs;
}