Russell Bryant [Wed, 26 Dec 2007 17:00:03 +0000 (17:00 +0000)]
List include/asterisk/version.h as a .PHONY target because we want the commands
listed for this target to be executed regardless of whether the file exists or
not. This fixes having the version not up to date when running from svn.
(closes issue #11619, reported by plack, fixed by me)
Tilghman Lesher [Mon, 24 Dec 2007 19:36:59 +0000 (19:36 +0000)]
Race: we need to wait to queue a NewChannel event until after the channel is
inserted into the channel list. The reason is because some manager users
immediately queue requests from the channel when they see that event and are
confused when Asterisk reports no such channel. (Closes issue #11632)
Russell Bryant [Thu, 20 Dec 2007 20:08:42 +0000 (20:08 +0000)]
Fix a deadlock in d-channel handling in chan_zap.
This deadlock was introduced by the fix to ensure that channels are properly
locked when handling channel variables. There were sections of this code where
the channel pvt was locked before the channel lock, when in fact it _must_ be
the other way around.
Mark Michelson [Tue, 18 Dec 2007 17:02:48 +0000 (17:02 +0000)]
Rework deadlock avoidance used in ast_write, since it meant that agent channels which were being monitored
had one audio file recorded and one empty audio file saved.
(closes issue #11529, reported by atis patched by me)
Joshua Colp [Mon, 17 Dec 2007 22:28:09 +0000 (22:28 +0000)]
Do not try to access information about a lock when printing out a trylock attempt. It is possible for the lock that it references to no longer be valid. This would have caused segfaults or deadlocks.
(issue #BE-263)
(closes issue #11080)
Reported by: callguy
(closes issue #11100)
Reported by: callguy
Tilghman Lesher [Mon, 17 Dec 2007 21:12:42 +0000 (21:12 +0000)]
Today is tomorrow's yesterday, and yesterday's tomorrow is today, and
tomorrow's tomorrow is the day after tomorrow, so who cares if you
recycle anyway?
If this confuses you, that's nothing compared to what this fixes. ;-)
Joshua Colp [Mon, 17 Dec 2007 18:05:55 +0000 (18:05 +0000)]
If a call is received with a called number IE containing nothing go to the 's' extension.
(closes issue #9099)
Reported by: kb1_kanobe2
Patches:
20070906__9099.diff.txt uploaded by Corydon76 (license 14)
Kevin P. Fleming [Mon, 17 Dec 2007 06:44:51 +0000 (06:44 +0000)]
In http://lists.digium.com/pipermail/asterisk-dev/2007-December/031145.html,
rizzo brought up some issues related to the way that the metadata required
for menuselect and the rest of the build system is extracted from the source
files. Since I had a few hours to kill on an airplane today, I decided to
improve this situation... so now the system caches the extracted metadata
and uses it to build the menuselect 'tree' as much as it can. The result
of this is that when a single source file is changed, only the metadata for
that file needs to be extracted again, and the rest is used from the cache
files. I also reduced the number of forked processes required to do the
metadata extraction; it was actually possible to do most of what we needed
in the Makefiles themselves without using any shell scripts at all! On my
laptop, these changes resulted in an 80% decrease in the time required
for the 'menuselect.makeopts' automatic check to occur after editing a single
source file.
While doing this work I also cleaned up a few minor things in the Makefiles,
adding a check for 'awk' to the configure script and changed all remaining
places we use 'grep' or 'awk' to use the ones found by the configure script,
and changed the 'prep_tarball' script to build the menuselect metadata so
that tarballs of Asterisk will include it and won't require the user to
wait while it is extracted after unpacking.
Russell Bryant [Fri, 14 Dec 2007 17:36:08 +0000 (17:36 +0000)]
There are a lot of existing systems that #include non-existent files. So, to
make the transition to treating this as an error a bit less painless, just issue
a huge error message for now. Then, later, we can reinstate the code that treats
it as a failure.
Mark Michelson [Fri, 14 Dec 2007 01:24:06 +0000 (01:24 +0000)]
When compiling with DETECT_DEADLOCKS, don't spam the CLI with messages
about possible deadlocks. Instead just print the intended single message every
five seconds.
(closes issue 11537, reported and patched by dimas)
Mark Michelson [Wed, 12 Dec 2007 17:46:14 +0000 (17:46 +0000)]
Correctly detect where a dynamic feature was activated. Before this patch,
the channel which initiated the bridge was always assumed to have been the one
which activated the dynamic feature. This patch corrects this.
(closes issue #11529, reported and patched by nic_bellamy)
Joshua Colp [Tue, 11 Dec 2007 19:51:40 +0000 (19:51 +0000)]
Fix potential memory leak with the dialed interfaces list if another memory allocation fails.
(closes issue #11507)
Reported by: eliel
Patches:
global_datastores.c.patch uploaded by eliel (license 64)
Mark Michelson [Tue, 11 Dec 2007 17:42:25 +0000 (17:42 +0000)]
Fixing autofill to be more accurate. Specifically, if calls ahead of the current
caller were ringing members (but not yet bridged) there could be available members
and waiting callers who would not get matched up. The member availability checker
was correctly determining the number of available members in this scenario, but
the queue itself did not parallelly reflect this status on the pending calls. This
commit corrects the issue.
(closes issue #11459, reported by equissoftware, patched by me)
Joshua Colp [Mon, 10 Dec 2007 16:36:15 +0000 (16:36 +0000)]
Add G729A as another possible payload name for G729. Some devices use this instead of G729, which is perfectly normal since the payload number itself is defined and can't be used by anything else so the name doesn't matter that much.
(closes issue #11483)
Reported by: revolution
Patches:
rtp.diff uploaded by revolution (license 346)
Mark Michelson [Mon, 10 Dec 2007 16:29:44 +0000 (16:29 +0000)]
If there are no members in a queue, then the loop where the datastore for detecting
duplicate dialed numbers will be skipped, meaning the datastore isn't created. This means
that when we try to free it, there's a crash. This stops that crash from occurring.
(closes issue #11499, reported by slavon, patched by eliel)
Olle Johansson [Mon, 10 Dec 2007 14:04:44 +0000 (14:04 +0000)]
Avoid reinvite race situations with two Asterisks trying
to reinvite each other in 1.4 and trunk.
This patch implements support for the 491 error code that
Asterisk 1.4 generates on situations where we get an
incoming INVITE and already has one in progress.
Thanks to mavetju for reporting and to Raj Jain for an
excellent explanation of the problem.
Patch by myself. Tested with 8 Asterisk servers connected
to each other in a training network.
even though Asterisk explicitly requests that endpoints using G.729 do *not* use Annex B (silence detection and comfort noise generation) some do anyway; the transcoder card interface does not currently work properly with CNG frames, so trim off the CNG before sending the data
Russell Bryant [Fri, 7 Dec 2007 21:24:33 +0000 (21:24 +0000)]
Make the lock protecting each thread's list of locks it currently holds
recursive. I think that this will fix the situation where some people have
said that "core show locks" locks up the CLI.
(related to issue #11080)
Russell Bryant [Fri, 7 Dec 2007 21:17:24 +0000 (21:17 +0000)]
Fix another bug in the DEBUG_THREADS code. The ast_mutex_init() function had
the mutex attribute object marked as static. This means that multiple threads
initializing locks at the same time could step on each other and end up with
improperly initialized locks.
(found when tracking down locking issues related to issue #11080)
Russell Bryant [Fri, 7 Dec 2007 21:11:08 +0000 (21:11 +0000)]
I love fixing lock related errors in the lock debugging code. That's about as
ironic as it gets in Asterisk programming land. Anyway, I spotted this bug while
trying to track down why systems are locking up and acting weird in issue #11080.
The mutex attribute object was marked as static in this function when it should
not have been.
Russell Bryant [Fri, 7 Dec 2007 16:38:48 +0000 (16:38 +0000)]
* Add channel locking around datastore operations that expect the channel
to be locked.
* Document why we don't record Local channels in the dialed interfaces list.
* Remove the dialed variable as it isn't needed.
* Restructure some code for clarity and coding guidelines stuff
Russell Bryant [Fri, 7 Dec 2007 16:25:25 +0000 (16:25 +0000)]
* Add channel locking around datastore operations that expect the channel
to be locked.
* Document why we don't record Local channels in the dialed interfaces list.
* Handle memory allocation failure.
* Remove the dialed variable, as it wasn't actually needed.
* Tweak some formatting to conform to coding guidelines.
Russell Bryant [Fri, 7 Dec 2007 16:08:35 +0000 (16:08 +0000)]
* Add a bit more of a verbose comment as to why a hangup frame needs to be
queued up if autoservice gets a NULL return from ast_read().
* Make the process of queueing the hangup frame more efficient by putting the
frame where it is going to end up and avoiding some locking and extra memory
allocations and freeing.
Mark Michelson [Fri, 7 Dec 2007 15:39:58 +0000 (15:39 +0000)]
Hangups that happen during autoservice were not processed appropriately. This is
because a hangup actually causes a NULL frame to be received, not a hangup frame.
Queueing a hangup if we receive a NULL frame during autoservice corrects this problem
(closes issue #11467, reported by jmls, patched by me)
Russell Bryant [Fri, 7 Dec 2007 02:38:40 +0000 (02:38 +0000)]
Allow dialing local channels from Queue() and Dial() again. There was a slight
flaw in the code to prevent call forwards from looping that caused this problem.
(related to issue #11486)
Russell Bryant [Fri, 7 Dec 2007 02:19:45 +0000 (02:19 +0000)]
Fix in an issue in the call forwarding handling code that was causing crashes
on every call into a queue. I'm not entirely sure about the logic in this part
of the code, so I want to look at it some more tomorrow. However, this makes
it safe and keeps it from crashing.
(closes issue #11486, reported by adamg, patched by me)
Tilghman Lesher [Fri, 7 Dec 2007 00:52:17 +0000 (00:52 +0000)]
At the end of a call, when we're reporting, RTCP may already be partially torn down, so check for NULL dereference
Reported by: blitzrage
Patch by: tilghman
(Closes issue #11450)
Mark Michelson [Thu, 6 Dec 2007 20:25:11 +0000 (20:25 +0000)]
IMAP storage did not honor the maxmsg setting in voicemail.conf, and it also had the possibility
of crashing if a user had more than 256 messages in their voicemail. This patch kills two birds with
one stone by adding maxmsg support and also setting a hard limit on the number of messages at 255 so
that the crashes cannot happen.
(closes issue #11101, reported by Skavin, patched by me)
Russell Bryant [Thu, 6 Dec 2007 19:11:35 +0000 (19:11 +0000)]
Add a new module flag to indicate that a build sum is present. Modules built
against older Asterisk 1.4 headers will now load properly with just a warning
indicating that they are old and may cause problems.
(patch by paravoid)
Joshua Colp [Thu, 6 Dec 2007 16:49:42 +0000 (16:49 +0000)]
Fix various in the udptl implementation. It could return empty modem frames, have an incorrect sequence number on packets, and display the wrong sequence number in the debug messages.
(closes issue #11228)
Reported by: Cache
Patches:
udptl-4.patch uploaded by dimas (license 88)
Mark Michelson [Wed, 5 Dec 2007 22:35:52 +0000 (22:35 +0000)]
The 'G' option for Dial() did not properly handle the case where only a label was
provided. This was due to the fact that the answering channel did not have an extension
set, so ast_parseable_goto would fail. This fix eliminates the call to ast_parseable_goto
on the answering channel since it is a wasteful call. The answering channel and the calling
channel are both directed to the same extension and context, just different priorities, so
we can just copy the values from the calling channel to the answering channel and increment
the answering channel's priority.
(closes issue #11382, reported by jon, patch by me with correction by jon)
Russell Bryant [Wed, 5 Dec 2007 00:48:47 +0000 (00:48 +0000)]
When DEBUG_THREADS is enabled, we only have the details about who is holding
a lock that we are waiting on for a mutex, not rwlocks. This should fix the
problem where people have reported "core show locks" crashing sometimes.
Russell Bryant [Wed, 5 Dec 2007 00:35:31 +0000 (00:35 +0000)]
Fix some crashes in chan_iax2 that were reported as happening on Mac systems.
It turns out that the problem was the Mac version of the ast_atomic_fetchadd_int()
function. The Mac atomic add function returns the _new_ value, while this function
is supposed to return the old value. So, the crashes happened on unreferencing
objects. If the reference count was decreased to 1, ao2_ref() thought that it
had been decreased to zero, and called the destructor. However, there was still
an outstanding reference around.
Russell Bryant [Tue, 4 Dec 2007 23:46:40 +0000 (23:46 +0000)]
Modify file.h to maintain API compatibility with earlier versions. If a recent
compiler is being used, then a warning will show up for any modules still using
the old name "private" instead of "_private".
(patch suggested by paravoid)
Russell Bryant [Tue, 4 Dec 2007 19:57:39 +0000 (19:57 +0000)]
Make some changes to some additions I made recently for doing channel autoservice
when looking up extensions. This code was added to handle the case where a
dialplan switch was in use that could block for a long time. However, the way
that I added it, it did this for all extension lookups. However, lookups in the
in-memory tree of extensions should _not_ take long enough to matter. So, move
the autoservice stuff to be only around executing a switch.
Mark Michelson [Mon, 3 Dec 2007 23:12:17 +0000 (23:12 +0000)]
A big one...
This is the merge of the forward-loop branch. The main change here is that call-forwards can no longer loop.
This is accomplished by creating a datastore on the calling channel which has a linked list of all devices
dialed. If a forward happens, then the local channel which is created inherits the datastore. If, through this
progression of forwards and datastore inheritance, a device is attempted to be dialed a second time, it will simply
be skipped and a warning message will be printed to the CLI. After the dialing has been completed, the datastore
is detached from the channel and destroyed.
This change also introduces some side effects to the code which I shall enumerate here:
1. Datastore inheritance has been backported from trunk into 1.4
2. A large chunk of code has been removed from app_dial. This chunk is the section of code
which handles the call forward case after the channel has been requested but before it has
been called. This was removed because call-forwarding still works fine without it, it makes the
code less error-prone should it need changing, and it made this set of changes much less painful
to just have the forwarding handled in one place in each module.
3. Two new files, global_datastores.h and .c have been added. These are necessary since the datastore
which is attached to the channel may be created and attached in either app_dial or app_queue, so they
need a common place to find the datastore info. This approach was taken in case similar datastores are
needed in the future, there will be a common place to add them.
Russell Bryant [Sun, 2 Dec 2007 18:18:52 +0000 (18:18 +0000)]
The other day when I went through making changes as a result of the ao2_link()
change, I added some code to set pointers to NULL after they were unreferenced.
This pointed out that in this place, the object was unreferenced before the
code was done using it. So, move the unref down a little bit.
(crash reported by jmls on IRC)
Tilghman Lesher [Sun, 2 Dec 2007 09:34:23 +0000 (09:34 +0000)]
Clarify the return value on autoservice. Specifically, if you started
autoservice and autoservice was already on, it would erroneously return an
error.
Reported by: adiemus
Patch by: dimas
(Closes issue #11433)
Russell Bryant [Fri, 30 Nov 2007 19:26:04 +0000 (19:26 +0000)]
Change the behavior of ao2_link(). Previously, in inherited a reference.
Now, it automatically increases the reference count to reflect the reference
that is now held by the container.
This was done to be more consistent with ao2_unlink(), which automatically
releases the reference held by the container. It also makes it so it is
no longer possible for a pointer to be invalid after ao2_link() returns.
projects / thirdparty / asterisk.git / log
