Mark Michelson [Fri, 7 Dec 2012 00:30:35 +0000 (00:30 +0000)]
Add new threadpool test and fix some taskprocessor bugs.
The new thread creation test fails because Asterisk locks up
while trying to lock a taskprocessor.
While trying to debug that, I found a race condition during taskprocessor
creation where a default taskprocessor listener could try to operate on
a partially started taskprocessor. This was fixed by adding a new callback
to taskprocessor listeners.
Then while testing that change, I found some bugs in the taskprocessor
tests where I was not properly unlocking when done with a lock. Scoped
locks have spoiled me a bit.
I still have not figured out why the threadpool thread creation test
is locking up.
Fix memory leak in 'manager show event' when command entered incorrectly
When the CLI command 'manager show event' was run incorrectly and its usage
instructions returned, a reference to the event container was leaked. This
would prevent the container from being reclaimed when Asterisk exits. We now
properly decrement the count on the ao2 object using the nifty RAII_VAR macro.
Thanks to Russell for helping me stumble on this, and Terry for writing that
ridiculously helpful macro.
........
Merged revisions 377319 from http://svn.asterisk.org/svn/asterisk/branches/11
................
r377329 | russell | 2012-12-06 09:06:47 -0600 (Thu, 06 Dec 2012) | 7 lines
Add CLI tab completion to 'acl show'.
The 'acl show' CLI command allows you to show the details about a specific
named ACL in acl.conf. This patch adds tab completion to the command.
Review: https://reviewboard.asterisk.org/r/2230/
................
r377330 | russell | 2012-12-06 09:13:37 -0600 (Thu, 06 Dec 2012) | 6 lines
Minor code cleanup in named_acl.c.
This patch makes a few little cleanups to named_acl.c. A couple non-public
functions were made static and an opening brace for a function was moved to
its own line, per the coding guidelines.
................
Russell Bryant [Thu, 6 Dec 2012 15:13:37 +0000 (15:13 +0000)]
Minor code cleanup in named_acl.c.
This patch makes a few little cleanups to named_acl.c. A couple non-public
functions were made static and an opening brace for a function was moved to
its own line, per the coding guidelines.
Matthew Jordan [Thu, 6 Dec 2012 14:26:13 +0000 (14:26 +0000)]
Fix memory leak in 'manager show event' when command entered incorrectly
When the CLI command 'manager show event' was run incorrectly and its usage
instructions returned, a reference to the event container was leaked. This
would prevent the container from being reclaimed when Asterisk exits. We now
properly decrement the count on the ao2 object using the nifty RAII_VAR macro.
Thanks to Russell for helping me stumble on this, and Terry for writing that
ridiculously helpful macro.
........
Merged revisions 377319 from http://svn.asterisk.org/svn/asterisk/branches/11
Fix a SIP request memory leak with TLS connections.
During the TLS re-work in chan_sip some TLS specific code was moved
into a separate function. This function operates on a copy of the
incoming SIP request. This copy was never deinitialized causing a
memory leak for each request processed.
This function is now given a SIP request structure which it can use
to copy the incoming request into. This reduces the amount of memory
allocations done since the internal allocated components are reused
between packets and also ensures the SIP request structure is
deinitialized when the TLS connection is torn down.
(closes issue ASTERISK-20763)
Reported by: deti
........
Merged revisions 377257 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 377258 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 377259 from http://svn.asterisk.org/svn/asterisk/branches/11
................
r377263 | jrose | 2012-12-05 11:17:06 -0600 (Wed, 05 Dec 2012) | 21 lines
res_srtp: Fix a crash caused by srtp_dealloc on an already dealloced session
When srtp_create fails, the session may be dealloced or just not alloced. At
the same time though, the session pointer might not be set to NULL in this
process and attempting to srtp_dealloc it again will cause a segfault. This
patch checks for failure of srtp_create and sets the session pointer to NULL
if it fails.
Jonathan Rose [Wed, 5 Dec 2012 17:17:06 +0000 (17:17 +0000)]
res_srtp: Fix a crash caused by srtp_dealloc on an already dealloced session
When srtp_create fails, the session may be dealloced or just not alloced. At
the same time though, the session pointer might not be set to NULL in this
process and attempting to srtp_dealloc it again will cause a segfault. This
patch checks for failure of srtp_create and sets the session pointer to NULL
if it fails.
Joshua Colp [Wed, 5 Dec 2012 16:51:58 +0000 (16:51 +0000)]
Fix a SIP request memory leak with TLS connections.
During the TLS re-work in chan_sip some TLS specific code was moved
into a separate function. This function operates on a copy of the
incoming SIP request. This copy was never deinitialized causing a
memory leak for each request processed.
This function is now given a SIP request structure which it can use
to copy the incoming request into. This reduces the amount of memory
allocations done since the internal allocated components are reused
between packets and also ensures the SIP request structure is
deinitialized when the TLS connection is torn down.
(closes issue ASTERISK-20763)
Reported by: deti
........
Merged revisions 377257 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 377258 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 377259 from http://svn.asterisk.org/svn/asterisk/branches/11
* Made func_confbridge_helper() allow an empty value when setting options.
You previously could not Set(CONFBRIDGE(user,pin)=) and clear the
configured pin from the dialplan.
* Made func_confbridge_helper() handle its datastore better if multiple
threads attempt to set the first CONFBRIDGE option value on the channel.
* Made the func_confbridge_helper() only output one diagnostic message
concerning the option.
* Made the bridge video_mode able to repeatedly change in the config file
and CONFBRIDGE dialplan function. The video_mode option values are an
enum and not independent of each other.
* Made handle_cli_confbridge_show_bridge_profile() better handle the
video_mode option.
* Simplified datastore handling code in conf_find_user_profile() and
conf_find_bridge_profile().
Richard Mudgett [Wed, 5 Dec 2012 01:11:26 +0000 (01:11 +0000)]
confbridge: Fix several small issues.
* Made func_confbridge_helper() allow an empty value when setting options.
You previously could not Set(CONFBRIDGE(user,pin)=) and clear the
configured pin from the dialplan.
* Made func_confbridge_helper() handle its datastore better if multiple
threads attempt to set the first CONFBRIDGE option value on the channel.
* Made the func_confbridge_helper() only output one diagnostic message
concerning the option.
* Made the bridge video_mode able to repeatedly change in the config file
and CONFBRIDGE dialplan function. The video_mode option values are an
enum and not independent of each other.
* Made handle_cli_confbridge_show_bridge_profile() better handle the
video_mode option.
* Simplified datastore handling code in conf_find_user_profile() and
conf_find_bridge_profile().
Mark Michelson [Tue, 4 Dec 2012 22:11:31 +0000 (22:11 +0000)]
Remove zombie state from threadpool altogether.
After giving it some consideration, there's no real
use for zombie threads. Listeners can't really use the
current number of zombie threads as a way of gauging activity,
zombifying threads is just an extra step before they die that
really serves no purpose, and since there's no way to re-animate
zombies, the operation does not need to be around.
I also fixed up some miscellaneous compilation errors that
were lingering from some past revisions.
Mark Michelson [Tue, 4 Dec 2012 21:21:15 +0000 (21:21 +0000)]
Simplify threadpool refcounting a bit.
Since threadpool shutdown is very strictly controlled,
there is no need to be so precise with reference counts
in queued operations. Since the threadpool shuts down its
own control taskprocessor before doing anything else destructive,
it can be guaranteed that all queued tasks will have a valid
pointer to the pool. This meant that some destructor functions
for helper structs could be removed entirely.
* Convert atexits list to a mutex instead of a rd/wr lock. The lock is
only write locked.
* Move CLI verbose Asterisk ending message to where AMI message is output
in really_quit() to avoid further surprises about using stuff already
shutdown.
Richard Mudgett [Mon, 3 Dec 2012 23:00:08 +0000 (23:00 +0000)]
Cleanup ast_run_atexits() atexits list.
* Convert atexits list to a mutex instead of a rd/wr lock. The lock is
only write locked.
* Move CLI verbose Asterisk ending message to where AMI message is output
in really_quit() to avoid further surprises about using stuff already
shutdown.
Mark Michelson [Mon, 3 Dec 2012 17:24:11 +0000 (17:24 +0000)]
Fix a few miscellaneous things
* Renamed some taskprocessor listener callbacks
* Fixed what would become a refcounting issue for pools
* Removed some XXX comments from allocation failure points
Move functions to AFTER the block of forward declarations of functions.
It was a mess. The first part of chan_sip.c is constants, declarations, structures and stuff,
then forward declarations and then actual code. It's still a mess, but a bit less messy ;-)
................
r377022 | file | 2012-12-03 08:56:36 -0600 (Mon, 03 Dec 2012) | 13 lines
Fix an RTP instance reference count leak in chan_motif.
When setting up an RTP instance the RTCP portion of the instance
keeps a reference to the instance itself. In order to release this
reference and stop RTCP the stop API call must be called before
destroying the instance.
Joshua Colp [Mon, 3 Dec 2012 14:56:36 +0000 (14:56 +0000)]
Fix an RTP instance reference count leak in chan_motif.
When setting up an RTP instance the RTCP portion of the instance
keeps a reference to the instance itself. In order to release this
reference and stop RTCP the stop API call must be called before
destroying the instance.
Olle Johansson [Mon, 3 Dec 2012 14:46:02 +0000 (14:46 +0000)]
Move functions to AFTER the block of forward declarations of functions.
It was a mess. The first part of chan_sip.c is constants, declarations, structures and stuff,
then forward declarations and then actual code. It's still a mess, but a bit less messy ;-)
Tweak extension used for incoming calls received on Motif.
Based on feedback from numerous individuals this patch tweaks incoming calls
to first look for an extension with the name of the endpoint. If no such extension
exists the call will silently fall back to the "s" extension as it previously
did.
........
Merged revisions 376983 from http://svn.asterisk.org/svn/asterisk/branches/11
................
Joshua Colp [Sat, 1 Dec 2012 00:47:42 +0000 (00:47 +0000)]
Tweak extension used for incoming calls received on Motif.
Based on feedback from numerous individuals this patch tweaks incoming calls
to first look for an extension with the name of the endpoint. If no such extension
exists the call will silently fall back to the "s" extension as it previously
did.
........
Merged revisions 376983 from http://svn.asterisk.org/svn/asterisk/branches/11
Fix potential crashes during SIP attended transfers.
The principal behind this patch is simple. During a transfer,
we manipulate channels that are owned by a separate thread than
the one we currently are running in, so it makes sense that we
need to grab a reference to the channels so that they cannot
disappear out from under us.
In the wild, crashes were sometimes seen when the transferring
party would hang up the call before the transfer target answered
the call. The most common place to see the crash occur was when
attempting to send a connected line update to the transferer
channel.
(closes issue ASTERISK-20226)
Reported by Jared Smith
Patches:
ASTERISK-20226.patch uploaded by Mark Michelson (License #5049)
Tested by: Jared Smith
........
Merged revisions 376901 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376916 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376917 from http://svn.asterisk.org/svn/asterisk/branches/11
................
r376922 | seanbright | 2012-11-30 11:08:41 -0600 (Fri, 30 Nov 2012) | 11 lines
Minor spelling fix to the VOLUME documentation.
........
Merged revisions 376919 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376920 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376921 from http://svn.asterisk.org/svn/asterisk/branches/11
................
Mark Michelson [Fri, 30 Nov 2012 16:56:53 +0000 (16:56 +0000)]
Fix potential crashes during SIP attended transfers.
The principal behind this patch is simple. During a transfer,
we manipulate channels that are owned by a separate thread than
the one we currently are running in, so it makes sense that we
need to grab a reference to the channels so that they cannot
disappear out from under us.
In the wild, crashes were sometimes seen when the transferring
party would hang up the call before the transfer target answered
the call. The most common place to see the crash occur was when
attempting to send a connected line update to the transferer
channel.
(closes issue ASTERISK-20226)
Reported by Jared Smith
Patches:
ASTERISK-20226.patch uploaded by Mark Michelson (License #5049)
Tested by: Jared Smith
........
Merged revisions 376901 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376916 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376917 from http://svn.asterisk.org/svn/asterisk/branches/11
Improve Code Readability And Fix Setting natdetected Flag
For 1.8, 10, 11 and trunk we are are improving the code readability.
For 11 and trunk, auto nat detection was added. The natdetected flag was being
set to 1 when the host address in the VIA header did not specifiy a port. This
patch fixes this by setting the port on the temporary sock address used to
SIP_STANDARD_PORT in order for the sock address comparison to work properly.
(closes issue ASTERISK-20724)
Reported by: Michael L. Young
Patches:
asterisk-20724-set-port-v2.diff uploaded by Michael L. Young (license 5026)
Michael L. Young [Thu, 29 Nov 2012 21:58:41 +0000 (21:58 +0000)]
Improve Code Readability And Fix Setting natdetected Flag
For 1.8, 10, 11 and trunk we are are improving the code readability.
For 11 and trunk, auto nat detection was added. The natdetected flag was being
set to 1 when the host address in the VIA header did not specifiy a port. This
patch fixes this by setting the port on the temporary sock address used to
SIP_STANDARD_PORT in order for the sock address comparison to work properly.
(closes issue ASTERISK-20724)
Reported by: Michael L. Young
Patches:
asterisk-20724-set-port-v2.diff uploaded by Michael L. Young (license 5026)
Websocket by default doesn't return an ast_str for the payload received. When
converting it to an ast_str on chan_sip the last character was being omitted,
because ast_str functions expects that the given length includes the trailing
0x00. payload_len only has the actual string length without counting the
trailing zero.
For most cases this passed unnoticed as most of SIP messages ends with \r\n.
Pedro Kiefer [Thu, 29 Nov 2012 16:44:42 +0000 (16:44 +0000)]
Fix chan_sip websocket payload handling
Websocket by default doesn't return an ast_str for the payload received. When
converting it to an ast_str on chan_sip the last character was being omitted,
because ast_str functions expects that the given length includes the trailing
0x00. payload_len only has the actual string length without counting the
trailing zero.
For most cases this passed unnoticed as most of SIP messages ends with \r\n.
* Adds the following CLI commands to control MALLOC_DEBUG reporting of
unreleased malloc memory when Asterisk is shut down.
memory atexit list on
memory atexit list off
memory atexit summary byline
memory atexit summary byfunc
memory atexit summary byfile
memory atexit summary off
* Made check all remaining allocated region blocks atexit for fence
violations.
* Increased the allocated region hash table size by about three times. It
still isn't large enough considering the number of malloced blocks
Asterisk uses.
* Made CLI "memory show allocations anomalies" use
regions_check_all_fences().
* Adds the following CLI commands to control MALLOC_DEBUG reporting of
unreleased malloc memory when Asterisk is shut down.
memory atexit list on
memory atexit list off
memory atexit summary byline
memory atexit summary byfunc
memory atexit summary byfile
memory atexit summary off
* Made check all remaining allocated region blocks atexit for fence
violations.
* Increased the allocated region hash table size by about three times. It
still isn't large enough considering the number of malloced blocks
Asterisk uses.
* Made CLI "memory show allocations anomalies" use
regions_check_all_fences().
manager: Make challenge work with allowmultiplelogin=no
Prior to this patch, challenge would yield a multiple logins error if used
without providing the username (which isn't really supposed to be an argument
to challenge) if allowmultiplelogin was set to no because allowmultiplelogin
finds a user with a zero length login name. This check is simply disabled for
the challenge action when the username is empty by this patch.
(closes issue ASTERISK-20677)
Reported by: Vladimir
Patches:
challenge_action_nomultiplelogin.diff uploaded by Jonathan Rose (license 6182)
........
Merged revisions 376725 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376726 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376727 from http://svn.asterisk.org/svn/asterisk/branches/11
................
Jonathan Rose [Wed, 28 Nov 2012 16:47:44 +0000 (16:47 +0000)]
manager: Make challenge work with allowmultiplelogin=no
Prior to this patch, challenge would yield a multiple logins error if used
without providing the username (which isn't really supposed to be an argument
to challenge) if allowmultiplelogin was set to no because allowmultiplelogin
finds a user with a zero length login name. This check is simply disabled for
the challenge action when the username is empty by this patch.
(closes issue ASTERISK-20677)
Reported by: Vladimir
Patches:
challenge_action_nomultiplelogin.diff uploaded by Jonathan Rose (license 6182)
........
Merged revisions 376725 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376726 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376727 from http://svn.asterisk.org/svn/asterisk/branches/11
The '-' char is supposed to be ignored by the dialplan extension matching.
Unfortunately, it's treatment is not handled consistently throughout the
extension matching code.
* Made the old exten matching code consistently ignore '-' chars.
* Made the old exten matching code consistently handle case in the
matching.
* Made ignore empty character sets.
* Fixed ast_extension_cmp() to return -1, 0, or 1 as documented. The only
user of it in pbx_lua.c was testing for -1. It was originally returning
the strcmp() value for less than which is not usually going to be -1.
* Fix character set sorting if the sets have the same number of characters
and start with the same character. Character set [0-9] now sorts before
[02-9a] as originally intended.
* Updated some extension label and priority already in use warnings to
also indicate if the extension is aliased.
Richard Mudgett [Wed, 28 Nov 2012 00:13:10 +0000 (00:13 +0000)]
Fix extension matching with the '-' char.
The '-' char is supposed to be ignored by the dialplan extension matching.
Unfortunately, it's treatment is not handled consistently throughout the
extension matching code.
* Made the old exten matching code consistently ignore '-' chars.
* Made the old exten matching code consistently handle case in the
matching.
* Made ignore empty character sets.
* Fixed ast_extension_cmp() to return -1, 0, or 1 as documented. The only
user of it in pbx_lua.c was testing for -1. It was originally returning
the strcmp() value for less than which is not usually going to be -1.
* Fix character set sorting if the sets have the same number of characters
and start with the same character. Character set [0-9] now sorts before
[02-9a] as originally intended.
* Updated some extension label and priority already in use warnings to
also indicate if the extension is aliased.
* Removed call to ast_module_user_hangup_all() in res_config_mysql.c since
it is effectively a noop. No channels can attach a reference to that
module.
* Removed call to ast_module_user_hangup_all() in app_celgenuserevent.c.
The caller of unload_module() has already called it.
* Removed redundant channel module references in pbx_dundi.c. The
registered dialplan function callback dispatchers for the read/read2/write
callbacks already reference the module before calling.
* pbx_dundi: Moved unregistering CLI commands, DUNDi switch, and dialplan
functions to the first thing the unload_module() does. This will reduce
the chance of new channels using DUNDi services while the module is being
torn down.
........
Merged revisions 376657 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376658 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376659 from http://svn.asterisk.org/svn/asterisk/branches/11
................
Richard Mudgett [Tue, 27 Nov 2012 20:39:51 +0000 (20:39 +0000)]
Remove unnecessary channel module references.
* Removed call to ast_module_user_hangup_all() in res_config_mysql.c since
it is effectively a noop. No channels can attach a reference to that
module.
* Removed call to ast_module_user_hangup_all() in app_celgenuserevent.c.
The caller of unload_module() has already called it.
* Removed redundant channel module references in pbx_dundi.c. The
registered dialplan function callback dispatchers for the read/read2/write
callbacks already reference the module before calling.
* pbx_dundi: Moved unregistering CLI commands, DUNDi switch, and dialplan
functions to the first thing the unload_module() does. This will reduce
the chance of new channels using DUNDi services while the module is being
torn down.
........
Merged revisions 376657 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 376658 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376659 from http://svn.asterisk.org/svn/asterisk/branches/11
Re-initialize logmsgs mutex upon logger initialization to prevent lock errors
Similar to the patch that moved the fork earlier in the startup sequence to
prevent mutex errors in the recursive mutex surrounding the read/write thread
registration lock, this patch re-initializes the logmsgs mutex. Part of the
start up sequence before forking the process into the background includes
reading asterisk.conf; this has to occur prior to the call to daemon in order
to read startup parameters. When reading in a conf file, log statements can
be generated. Since this can't be avoided, the mutex instead is
re-initialized to ensure a reset of any thread tracking information.
This patch also includes some additional debugging to catch errors when
locking or unlocking the recursive mutex that surrounds locks when the
DEBUG_THREADS build option is enabled. DO_CRASH or THREAD_CRASH will
cause an abort() if a mutex error is detected.
Matthew Jordan [Fri, 23 Nov 2012 00:02:23 +0000 (00:02 +0000)]
Re-initialize logmsgs mutex upon logger initialization to prevent lock errors
Similar to the patch that moved the fork earlier in the startup sequence to
prevent mutex errors in the recursive mutex surrounding the read/write thread
registration lock, this patch re-initializes the logmsgs mutex. Part of the
start up sequence before forking the process into the background includes
reading asterisk.conf; this has to occur prior to the call to daemon in order
to read startup parameters. When reading in a conf file, log statements can
be generated. Since this can't be avoided, the mutex instead is
re-initialized to ensure a reset of any thread tracking information.
This patch also includes some additional debugging to catch errors when
locking or unlocking the recursive mutex that surrounds locks when the
DEBUG_THREADS build option is enabled. DO_CRASH or THREAD_CRASH will
cause an abort() if a mutex error is detected.
Reduce CLI spam of "Extension Changed" device state messages.
Asterisk 11 follows RFC3265 that states that after every subscribe or resubscribe a notify should be sent.
Thus the console if filled continuously with the following after every subscribe;
== Extension Changed 8512[phones] new state IDLE for Notify User cisco1
In Asterisk 1.8 only changes would be sent. Thus only when a device state changed was anything emitted to the console.
fix:
Only print to console when device state isn't forced.
Alec L Davis [Tue, 20 Nov 2012 17:39:11 +0000 (17:39 +0000)]
Reduce CLI spam of "Extension Changed" device state messages.
Asterisk 11 follows RFC3265 that states that after every subscribe or resubscribe a notify should be sent.
Thus the console if filled continuously with the following after every subscribe;
== Extension Changed 8512[phones] new state IDLE for Notify User cisco1
In Asterisk 1.8 only changes would be sent. Thus only when a device state changed was anything emitted to the console.
fix:
Only print to console when device state isn't forced.
With some versions of gcc, n_buckets will be flagged as being uninitialized
before use. While its technically impossible (since the switch statement,
even without a default, accounts for all possibilities), we'll initialize the
variable to 0 anyway.
........
Matthew Jordan [Mon, 19 Nov 2012 02:14:54 +0000 (02:14 +0000)]
Fix uninitialized in this function error
With some versions of gcc, n_buckets will be flagged as being uninitialized
before use. While its technically impossible (since the switch statement,
even without a default, accounts for all possibilities), we'll initialize the
variable to 0 anyway.
Reorder startup sequence to prevent lockups when process is sent to background
Although it is very rare and timing dependent, the potential exists for the
call to 'daemon' to cause what appears to be a deadlock in Asterisk during
startup. This can occur when a recursive mutex is obtained prior to the
daemon call executing. Since daemon uses fork to send the process into the
background, any threading primitives are unsafe to re-use after the call.
Implementations of pthread recursive mutexes are highly likely to store the
thread identifier of the thread that previously obtained the mutex. If
the mutex was locked prior to the fork, a subsequent unlock operation will
potentially fail as the thread identifier is no longer valid. Since the
mutex is still locked, all subsequent attempts to grab the mutex by other
threads will block.
This behavior exhibited itself most often when DEBUG_THREADS was enabled, as
this compile time option surrounds the mutexes in Asterisk with another
recursive mutex that protects the storage of thread related information. This
made it much more likely that a recursive mutex would be obtained prior to
daemon and unlocked after the call.
This patch does the following:
a) It backports a patch from Asterisk 11 that prevents the spawning of the
localtime monitoring thread. This thread is now spawned after Asterisk has
fully booted.
b) It re-orders the startup sequence to call daemon earlier during Asterisk
startup. This limits the potential of threading primitives being accessed
by initialization calls before daemon is called.
c) It removes calls to ast_verbose/ast_log/etc. prior to daemon being called.
Developers should send error messages directly to stderr prior to daemon,
as calls to ast_log may access recursive mutexes that store thread related
information.
d) It reorganizes when thread local storage is created for storing lock
information during the creation of threads. Prior to this patch, the
read/write lock protecting the list of threads in ast_register_thread would
utilize the lock in the thread local storage prior to it being initialized;
this patch prevents that.
On a very related note, this patch will *greatly* improve the stability of the
Asterisk Test Suite.
Matthew Jordan [Sun, 18 Nov 2012 20:27:45 +0000 (20:27 +0000)]
Reorder startup sequence to prevent lockups when process is sent to background
Although it is very rare and timing dependent, the potential exists for the
call to 'daemon' to cause what appears to be a deadlock in Asterisk during
startup. This can occur when a recursive mutex is obtained prior to the
daemon call executing. Since daemon uses fork to send the process into the
background, any threading primitives are unsafe to re-use after the call.
Implementations of pthread recursive mutexes are highly likely to store the
thread identifier of the thread that previously obtained the mutex. If
the mutex was locked prior to the fork, a subsequent unlock operation will
potentially fail as the thread identifier is no longer valid. Since the
mutex is still locked, all subsequent attempts to grab the mutex by other
threads will block.
This behavior exhibited itself most often when DEBUG_THREADS was enabled, as
this compile time option surrounds the mutexes in Asterisk with another
recursive mutex that protects the storage of thread related information. This
made it much more likely that a recursive mutex would be obtained prior to
daemon and unlocked after the call.
This patch does the following:
a) It backports a patch from Asterisk 11 that prevents the spawning of the
localtime monitoring thread. This thread is now spawned after Asterisk has
fully booted.
b) It re-orders the startup sequence to call daemon earlier during Asterisk
startup. This limits the potential of threading primitives being accessed
by initialization calls before daemon is called.
c) It removes calls to ast_verbose/ast_log/etc. prior to daemon being called.
Developers should send error messages directly to stderr prior to daemon,
as calls to ast_log may access recursive mutexes that store thread related
information.
d) It reorganizes when thread local storage is created for storing lock
information during the creation of threads. Prior to this patch, the
read/write lock protecting the list of threads in ast_register_thread would
utilize the lock in the thread local storage prior to it being initialized;
this patch prevents that.
On a very related note, this patch will *greatly* improve the stability of the
Asterisk Test Suite.
Add a test event that reports changes in ConfBridge state
This patch adds a test event to ConfBridge that reports transitions between
states in ConfBridge. This is used by tests in the Asterisk Test Suite
that verify state changes based on the entering/leaving of conference
participants.
........
Merged revisions 376414 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376415 from http://svn.asterisk.org/svn/asterisk/branches/11
................
Matthew Jordan [Sun, 18 Nov 2012 14:31:32 +0000 (14:31 +0000)]
Add a test event that reports changes in ConfBridge state
This patch adds a test event to ConfBridge that reports transitions between
states in ConfBridge. This is used by tests in the Asterisk Test Suite
that verify state changes based on the entering/leaving of conference
participants.
........
Merged revisions 376414 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 376415 from http://svn.asterisk.org/svn/asterisk/branches/11
Mark Michelson [Sat, 17 Nov 2012 22:27:39 +0000 (22:27 +0000)]
Restructure taskprocessor load test to test integrity of task execution.
This throws a random number to each task and stores a copy locally. After
all tasks have executed, the data is checked to be sure tasks executed
in the correct order.
projects / thirdparty / asterisk.git / log
