Mark Michelson [Wed, 16 Jan 2008 17:20:11 +0000 (17:20 +0000)]
Fix a deadlock in chan_local in local_hangup. There was contention because
the local_pvt was held and it was attempting to lock a channel, which is the
incorrect locking order.
Joshua Colp [Wed, 16 Jan 2008 15:08:24 +0000 (15:08 +0000)]
Introduce a lock into the dialing API that protects it when destroying the structure.
(closes issue #11687)
Reported by: callguy
Patches:
11687.diff uploaded by file (license 11)
Joshua Colp [Wed, 16 Jan 2008 03:07:24 +0000 (03:07 +0000)]
Don't drop the old record route information when dealing with packets related to a reinvite.
(closes issue #11545)
Reported by: kebl0155
Patches:
reinvite-patch.txt uploaded by kebl0155 (license 356)
Joshua Colp [Wed, 16 Jan 2008 01:13:27 +0000 (01:13 +0000)]
Add autoconf logic for speexdsp. Later versions use a separate library for some things so we need to use it if present in codec_speex.
(closes issue #11693)
Reported by: yzg
Russell Bryant [Tue, 15 Jan 2008 23:50:10 +0000 (23:50 +0000)]
Change a buffer in check_auth() to be a thread local dynamically allocated
buffer, instead of a massive buffer on the stack. This fixes a crash reported
by Qwell due to running out of stack space when building with LOW_MEMORY defined.
On a very related note, the usage of BUFSIZ in various places in chan_sip is
arbitrary and careless. BUFSIZ is a system specific define. On my machine,
it is 8192, but by definition (according to google) could be as small as 256.
So, this buffer in check_auth was 16 kB. We don't even support SIP messages
larger than 4 kB! Further usage of this define should be avoided, unless it
is used in the proper context.
Russell Bryant [Tue, 15 Jan 2008 23:26:52 +0000 (23:26 +0000)]
Commit a fix for some memory access errors pointed out by the valgrind2.txt
output on issue #11698.
The issue here is that it is possible for an instance of a translator to get
destroyed while the frame allocated as a part of the translator is still being
processed. Specifically, this is possible anywhere between a call to ast_read()
and ast_frame_free(), which is _a lot_ of places in the code. The reason this
happens is that the channel might get masqueraded during this time. During a
masquerade, existing translation paths get destroyed.
So, this patch fixes the issue in an API and ABI compatible way. (This one is
for you, paravoid!)
It changes an int in ast_frame to be used as flag bits. The 1 bit is still used
to indicate that the frame contains timing information. Also, a second flag has
been added to indicate that the frame came from a translator. When a frame with
this flag gets released and has this flag, a function is called in translate.c to
let it know that this frame is doing being processed. At this point, the flag gets
cleared. Also, if the translator was requested to be destroyed while its internal
frame still had this flag set, its destruction has been deffered until it finds out
that the frame is no longer being processed.
Admittedly, this feels like a hack. But, it does fix the issue, and I was not able
to think of a better solution ...
Mark Michelson [Mon, 14 Jan 2008 16:21:28 +0000 (16:21 +0000)]
Adding explicit defaults for missing options to init_queue. This is necessary because
if a user either removes or comments one of these options and reloads their queues, the
option will not reset to its default, instead maintaining the value from prior to the
reload.
Thanks to John Bigelow for pointing this error out to me.
Tilghman Lesher [Sat, 12 Jan 2008 00:05:08 +0000 (00:05 +0000)]
Add a connection timeout attribute, as that was what was intended with the
login timeout, but ODBC divides it up into 2 different timeouts.
(Closes issue #11745)
Joshua Colp [Fri, 11 Jan 2008 19:51:10 +0000 (19:51 +0000)]
If the incoming RTP stream changes codec force the bridge to break if the other side does not support it.
(closes issue #11729)
Reported by: tsearle
Patches:
new_codec_patch_udiff.patch uploaded by tsearle (license 373)
Joshua Colp [Fri, 11 Jan 2008 19:28:30 +0000 (19:28 +0000)]
If the channel is hungup during RECORD FILE send a result code of -1 to be uniform with everything else.
(closes issue #11743)
Reported by: davevg
Patches:
res_agi.diff uploaded by davevg (license 209)
Russell Bryant [Fri, 11 Jan 2008 18:25:30 +0000 (18:25 +0000)]
Backport the ability to set the ToS bits on Linux when not running as root.
Normally, we would not backport features into 1.4, but, I was convinced by the
justification supplied by the supplier of this patch. He pointed out that this
patch removes a requirement for running as root, thus reducing the potential
impacts of security issues.
Joshua Colp [Fri, 11 Jan 2008 17:22:53 +0000 (17:22 +0000)]
Ensure the return value of ast_bridge_call is passed back up as the application return value. This is needed for transfers to function so the PBX core knows to continue execution.
(closes issue #10327)
Reported by: kkiely
Tilghman Lesher [Thu, 10 Jan 2008 23:08:36 +0000 (23:08 +0000)]
1) When we get a translated frame out, clone it, because if the
translator pvt is freed before we use the frame, bad things happen.
2) Getting a failure from ast_sched_delete means that the schedule
ID is currently running. Don't just ignore it.
(Closes issue #11698)
Steve Murphy [Thu, 10 Jan 2008 21:37:10 +0000 (21:37 +0000)]
Applied the same fixes for ael.flex as was done in 97849 for ast_expr2.fl; overrode the normally generate yyfree func with our own version that checks the pointer for non-null before passing to free(). Also takes care of a little problem with 2.5.33 and the use of the __STDC_VERSION__ macro.
Steve Murphy [Thu, 10 Jan 2008 20:21:27 +0000 (20:21 +0000)]
This is a fix for 2 things: a problem Terry was having in OSX with null pointers, which was my fault, as I probably forgot to run the sed script last time I made mods. So, I moved the fix into the flex input itself. Then, I found when I used flex 2.5.33, that it was using __STDC_VERSION__, and that's not real good; so I added back in a DIFFERENT sed script to fix that little mess. Tested everything, a couple different ways. Hope I did no harm, at the least.
Joshua Colp [Thu, 10 Jan 2008 15:07:12 +0000 (15:07 +0000)]
Don't try to copy the category from the group if no category exists.
(closes issue #11724)
Reported by: IgorG
Patches:
group_count.v1.patch uploaded by IgorG (license 20)
Philippe Sultan [Wed, 9 Jan 2008 16:44:24 +0000 (16:44 +0000)]
Set the caller id within the gtalk_alloc function.
As underlined in issue #10437 by Josh, we need to prevent a possible
memory leak. We only set the name part of the caller id, the number
part is not relevant when dealing with JIDs.
Joshua Colp [Wed, 9 Jan 2008 16:11:17 +0000 (16:11 +0000)]
Don't do conferencing totally in Zaptel if Monitor is running on the channel.
(closes issue #11709)
Reported by: BigJimmy
Patches:
patch-meetmerec uploaded by BigJimmy (license 371)
Tilghman Lesher [Wed, 9 Jan 2008 00:44:14 +0000 (00:44 +0000)]
Allow filename completion on zero-length modules, remove a memory leak, remove
a file descriptor leak, and make filename completion thread-safe.
Patched and tested by tilghman.
(Closes issue #11681)
Mark Michelson [Tue, 8 Jan 2008 23:49:11 +0000 (23:49 +0000)]
Part 1 of N of adding doxygen comments to app_queue. I picked some of the most common functions
used (which also happen to be some the biggest/ugliest functions too) to document first. I'm pretty
new to doxygen so criticism is welcome.
Mark Michelson [Tue, 8 Jan 2008 20:42:07 +0000 (20:42 +0000)]
Making some changes designed to not allow for a corrupted mailstream for a vm_state.
1. Add locking to the vm_state retrieval functions so that no linked list corruption occurs.
2. Make sure to always grab the persistent vm_state when mailstream access is necessary.
3. Correct an incorrect return value in the init_mailstream function.
Joshua Colp [Tue, 8 Jan 2008 19:53:52 +0000 (19:53 +0000)]
If no group has been provided to the GROUP_COUNT dialplan function then use the first one specific to the channel.
(closes issue #11077)
Reported by: m4him
Tilghman Lesher [Fri, 4 Jan 2008 23:03:40 +0000 (23:03 +0000)]
Fix the problem of notification of a device state change to a device with a '-'
in the name. Could probably do with a better fix in trunk, but this bug has
been open way too long without a better solution.
Reported by: stevedavies
Patch by: tilghman
(Closes issue #9668)
Mark Michelson [Wed, 2 Jan 2008 23:46:02 +0000 (23:46 +0000)]
We need to reset the membername to NULL on each iteration of this loop, otherwise the result is that
multiple members can have the same name, since the variable was not reset on each iteration of the loop.
Joshua Colp [Wed, 2 Jan 2008 20:24:09 +0000 (20:24 +0000)]
Allocate a SIP refer structure when performing a transfer using BYE with Also so that the transfer information is properly stored. (AST-2008-001)
(closes issue #11637)
Reported by: greyvoip
Mark Michelson [Wed, 2 Jan 2008 17:51:22 +0000 (17:51 +0000)]
A change to improve the accuracy of queue logging in the case where a member does not
answer during the specified timeout period. Prior to this change, there was a small chance
that the member name recorded in this case would be blank. Also prior to this change, if using
the ringall strategy, if no one answered the call during the specified timeout, the member name
listed in the queue log would randomly be one of the members that was rung.
(closes issue #11498, reported and tested by hloubser, patched by me)
Mark Michelson [Mon, 31 Dec 2007 23:43:13 +0000 (23:43 +0000)]
Avoiding a potentially bad locking situation. ast_merge_contexts_and_delete writelocks the conlock, then
calls ast_hint_extension, which attempts to readlock the same lock. Recursion with read-write locks is
dangerous, so the inner lock needs to be removed. I did this by copying the "guts" of ast_hint_extension
into ast_merge_contexts_and_delete (sans the extra lock).
(this change is inspired by the locking problems seen in issue #11080, but I have no idea if this is the
problematic area experienced by the reporters of that issue)
Mark Michelson [Fri, 28 Dec 2007 00:16:15 +0000 (00:16 +0000)]
I found a bug while browsing the queue code and managed to reproduce it in a small setup.
If a queue uses the ringall strategy, it was possible through unfortunate coincidence for a single member at a given penalty level to
make app_queue think that all members at that penalty level were unavailable and cause the members at the
next penalty level to be rung. With this patch, we will only move to the next penalty level if ALL the members
at a given penalty level are unreachable.
Joshua Colp [Thu, 27 Dec 2007 17:32:15 +0000 (17:32 +0000)]
Include types.h in chan_h323 as without it it can not be compiled on some operating systems like FreeBSD to name one.
(closes issue #11585)
Reported by: sobomax
Patches:
chan_h323.c.diff uploaded by sobomax (license 359)
Joshua Colp [Thu, 27 Dec 2007 17:27:11 +0000 (17:27 +0000)]
Use ast_strlen_zero to see if our_contact is set or not on the dialog. It is possible for it to be a pointer to NULL.
(closes issue #11557)
Reported by: FuriousGeorge
Russell Bryant [Thu, 27 Dec 2007 15:16:56 +0000 (15:16 +0000)]
Now that the contexts lock is a read/write lock, it should not be locked here
in ast_hint_state_changed(). This makes it get locked recursively which now
causes a deadlock.
(closes issue #11080, thanks to callguy for the access to a deadlocked machine)
Russell Bryant [Thu, 27 Dec 2007 14:33:21 +0000 (14:33 +0000)]
Change ast_translator_best_choice() to only pay attention to audio formats.
This fixes a problem where Asterisk claims that a translation path can not be
found for channels involving video.
(closes issue #11638)
Reported by: cwhuang
Tested by: cwhuang
Patch suggested by cwhuang, with some additional changes by me.
Tilghman Lesher [Wed, 26 Dec 2007 20:43:38 +0000 (20:43 +0000)]
Workaround for what is probably a glibc bug (but we'll see this crop up again
and again, if we don't add the workaround).
Reported by: rolek
Patch by: tilghman
(Closes issue #11601, closes issue #11426)
Russell Bryant [Wed, 26 Dec 2007 19:04:31 +0000 (19:04 +0000)]
Just in case the AST_FLAG_END_DTMF_ONLY flag was already set before starting
autoservice, remember it and ensure that the channel has the same setting when
autoservice gets stopped. (pointed out by d1mas, patched up by me)
Russell Bryant [Wed, 26 Dec 2007 18:46:39 +0000 (18:46 +0000)]
When a channel is in autoservice, mark a flag on the channel that says that
we only care about the END of a digit. That way, no magic digit emulation stuff
will happen when all we're doing is queueing up END frames.
Russell Bryant [Wed, 26 Dec 2007 17:06:26 +0000 (17:06 +0000)]
Don't store DTMF BEGIN frames while a channel is in autoservice. It's just
going to make ast_read() do a lot of extra work when the channel comes back
out of autoservice.
(closes issue #11628, patched by me)
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)
projects / thirdparty / asterisk.git / log
