Gregory Nietsky [Mon, 3 Oct 2011 09:37:59 +0000 (09:37 +0000)]
Fixup a race condition in res_fax.c where FAXOPT(gateway)=no will
turn off the gateway but the framehook is not destroyed.
this problem happens when a gateway is attempted in the dialplan and
the device is not available i may want to do fax to mail in the server
it will not be allowed.
instead of checking only AST_FAX_TECH_GATEWAY also check gateway_id
Fix segfault in analog_ss_thread() not checking ast_read() for NULL.
NOTE: The problem was reported against v1.6.2. It is unlikely to ever
happen on v1.8 and above since chan_dahdi.c:analog_ss_thread() is unlikely
to be used. The version in sig_analog.c has largely replaced it.
(closes issue ASTERISK-18648)
Reported by: Stephan Bosch
Patches:
jira_asterisk_18648_v1.8.patch (license #5621) patch uploaded by rmudgett
Tested by: Stephan Bosch
........
The SIP_HEADER function only works on the the initial SIP INVITE. The documentation was updated
in trunk, but not in 1.8 or 10, so I'm making them match.
Dummy channels created by ast_dummy_channel_alloc() should be destoyed by
ast_channel_unref(). Using ast_channel_release() needlessly grabs the
channel container lock and can cause a deadlock as a result.
* Analyzed use of ast_dummy_channel_alloc() and made use
ast_channel_unref() when done with the dummy channel. (Primary reason for
the reported deadlock.)
* Made app_dial.c:dial_exec_full() not call ast_call() holding any channel
locks. Chan_local could not perform deadlock avoidance correctly.
(Potential deadlock exposed by this issue. Secondary reason for the
reported deadlock since the held lock was part of the deadlock chain.)
* Fixed some uses of ast_dummy_channel_alloc() not checking the returned
channel pointer for failure.
* Fixed some potential chan=NULL pointer usage in func_odbc.c. Protected
by testing the bogus_chan value.
* Fixed needlessly clearing a 1024 char auto array when setting the first
char to zero is enough in manager.c:action_getvar().
(closes issue ASTERISK-18613)
Reported by: Thomas Arimont
Patches:
jira_asterisk_18613_v1.8.patch (license #5621) patch uploaded by rmudgett
Tested by: Thomas Arimont
........
With these options enabled, they can cause Asterisk to freak out by
SYN flooding a network and eating the CPU. Obviously it would be good to
fix the code so that this can't happen, but we can at least change the default
configuration so it doesn't happen.
This was reported downstream to the Fedora issue tracker:
Made ISDN not add numbering plan prefix strings to empty numbers.
When the Caller-ID is restricted, the expected behavior is for the
Caller-ID to be blank. In chan_dahdi, the national prefix is placed onto
the Caller-ID number even if it is restricted (empty) causing the
Caller-ID to be the national prefix rather than blank.
This behavior was lost when sig_pri was extracted from chan_dahdi.
* Made not add prefix strings to empty connected line, calling, and ANI
number strings.
Jonathan Rose [Thu, 22 Sep 2011 15:47:05 +0000 (15:47 +0000)]
Forgot to svn add new files to r337595
Part of Generating security events for chan_sip
(issue ASTERISK-18264)
Reported by: Michael L. Young
Patches:
security_events_chan_sip_v4.patch (License #5026) by Michael L. Young
Reviewboard: https://reviewboard.asterisk.org/r/1362/
Jonathan Rose [Thu, 22 Sep 2011 15:35:50 +0000 (15:35 +0000)]
Generate Security events in chan_sip using new Security Events Framework
Security Events Framework was added in 1.8 and support was added for AMI to generate
events at that time. This patch adds support for chan_sip to generate security events.
(closes issue ASTERISK-18264)
Reported by: Michael L. Young
Patches:
security_events_chan_sip_v4.patch (license #5026) by Michael L. Young
Review: https://reviewboard.asterisk.org/r/1362/
Olle Johansson [Wed, 21 Sep 2011 09:32:50 +0000 (09:32 +0000)]
Make ast_pbx_run() not default to s@default if extension is not found
Review: https://reviewboard.asterisk.org/r/1446/
This is a bug - or architecture mistake - that has been in Asterisk for a
very long time. It was exposed by the AMI originate action and possibly
some other applications. Most channel drivers checks if an extension
exists BEFORE starting a pbx on an inbound call, so most calls will
not depend on this issue.
Thanks everyone involved in the review and on IRC and the mailing list
for a quick review and all the feedback.
Fix for incorrect voicemail duration in external notifications
This patch fixes an issue where the voicemail duration was being reported
with a duration significantly less than the actual sound file duration.
Voicemails that contained mostly silence were reporting the duration of
only the sound in the file, as opposed to the duration of the file with
the silence. This patch fixes this by having two durations reported in
the __ast_play_and_record family of functions - the sound_duration and the
actual duration of the file. The sound_duration, which is optional, now
reports the duration of the sound in the file, while the actual full duration
of the file is reported in the duration parameter. This allows the voicemail
applications to use the sound_duration for minimum duration checking, while
reporting the full duration to external parties if the voicemail is kept.
(issue ASTERISK-2234)
(closes issue ASTERISK-16981)
Reported by: Mary Ciuciu, Byron Clark, Brad House, Karsten Wemheuer, KevinH
Tested by: Matt Jordan
Richard Mudgett [Tue, 20 Sep 2011 22:47:45 +0000 (22:47 +0000)]
Fix crash with STRREPLACE function.
The ast_func_read() function calls the .read2 callback with the len
parameter set to zero indicating no size restrictions on the supplied
ast_str buffer. The value was used to dimension a local starts[] array
with the array subsequently used.
* Reworked the strreplace() function to perform the string replacement in
a straight forward manner. Eliminated the need for the starts[] array.
Make CANMATCH with the new pattern match engine behave more like the old one
When checking an extension for E_CANMATCH using the new extension matching
algorithm, an exact match was not returned as a possible match resulting in the
queue failing to allow a caller to exit on DTMF. This removes the requirement
that an extension be longer than acquired digits for an E_CANMATCH operation
to succeed.
Check if a channel was created before using the pointer in sig_ss7_new_ast_channel().
Fixes the crash in ASTERISK-17955 gdb-11918.txt backtrace.
* Added some missing libss7 access lock protection.
* Prevent cancelling the ss7_linkset() thread at inoportune times just
like the pri_dchannel() thread.
(issue ASTERISK-17955)
Reported by: Ian M Sherman
Patches:
jira_asterisk_17955_v1.8.patch (license #5621) patch uploaded by rmudgett
(attached to related ASTERISK-17966)
........
This patch addresses crashes related to RTCP handling. The backtraces just
show a crash in ast_rtcp_write() where it appears that the RTP instance is no
longer valid. There is a race condition with scheduled RTCP transmissions and
the destruction of the RTP instance. This patch utilizes the fact that
ast_rtp_instance is a reference counted object and ensures that it will not get
destroyed while a reference is still around due to scheduled RTCP
transmissions.
RTCP transmissions are scheduled and executed from the chan_sip scheduler
context. This scheduler context is processed in the SIP monitor thread. The
destruction of an RTP instance occurs when the associated sip_pvt gets
destroyed (which happens when the sip_pvt reference count reaches 0). However,
the SIP monitor thread is not the only thread that can cause a sip_pvt to get
destroyed. The sip_hangup function, executed from a channel thread, also
decrements the reference count on a sip_pvt and could cause it to get
destroyed.
While this is being changed anyway, the patch also removes calling
ast_sched_del() from within the RTCP scheduler callback. It's not helpful.
Simply returning 0 prevents the callback from being rescheduled.
(closes issue ASTERISK-18570)
Related issues that look like they are the same problem:
Various changes to allow 1.8 to compile on Mac OS X Lion (10.7)
* Makefile workaround for 10.6 extended to work on 10.7 and later.
* Now uses the 'weak' symbol for Lion systems, which no longer support
'weak_import'
Document applications that play audio and do not answer unanswered calls.
This patch is part of an effort to document early media and its usage. If you are
interested in contributing to this documentation effort, there are probably other
applications worth documenting as well as an Asterisk wiki article at
https://wiki.asterisk.org/wiki/display/AST/Early+Media+and+the+Progress+Application
........
Made Dial d and H options no longer immediately auto-answer the calling leg.
The Dial d and H options break DTMF attended transfer atxferdropcall
option.
1) Party A calls party B.
2) Party B does a DTMF attended transfer to Party C.
If the dialplan uses the Dial d or H options to call Party C then the Dial
application answers the call immediately before initiating the call leg to
Party C. The premature answer causes the transfer code to not invoke the
atxferdropcall=no behavior for a blonde transfer since Party C has
"answered". The transfer code thinks that Party B has "consulted" with
Party C when Party B hangs up and completes the transfer to Party A.
Party A now hears ringback until Party C actually answers.
ASTERISK-13294 Dial d option.
ASTERISK-11067 Dial H option to disconnect before answer.
The referenced issues made Dial answer with the d and H options because
many SIP and ISDN phones cannot send DTMF before the call is connected.
* Made require the dialplan to control when or if the call needs to be
answered to use the Dial application d and H options. (The call is no
longer surprise answered when using the Dial d or H options.)
Recently iLBC support in Asterisk has changed after the acquisition of GIPS
by Google. More information about how this may affect you is available in a
blog post at:
Fix bad RTP media bridges in directmedia calls on peers separated by multiple Asterisk nodes.
In a situation involving devices on separate Asterisk trunks, the remote RTP bridge would
break when starting a call with directmedia. This patch queues a new type of control frame
so that our RTP bridge loop can properly detect when these situations occur and check to see
if peers need to be updated in order to send their media to the proper location.
a channel lock must never be held with the queues container lock held.
the deadlock occured on masquerade.
the queues container lock is a relic of the past the old queue module lock.
with ao2 there is no need to hold this lock when dealing with members this
patch removes unneeded locks.
(closes issue ASTERISK-18101)
(closes issue ASTERISK-18487)
Reported by: Paul Rolfe, Jason Legault
Tested by: irroot, Jason Legault, Paul Rolfe
Reviewed by: Matthew Nicholson
Olle Johansson [Thu, 15 Sep 2011 12:46:38 +0000 (12:46 +0000)]
Meetme: Introducing a new option "k" to kill a conference if there's only a single member left.
When using Meetme as a modular call bridge from third party applications, it's handy to make
it behave like a normal call bridge. When the second to last person exists, the last person
will be kicked out of the conference when this option is enabled.
Remove unnecessary libpri dependency checks in the configure script.
Using the --with-pri option with the configure script generated an error
about not having PRI_L2_PERSISTENCE if you did not have the absolute
latest libpri SVN checkout installed.
The AST_EXT_LIB_SETUP_DEPENDENT macro in the configure.ac script seems to
be for libraries that are dependent upon other libraries and not
necessarily for optional/added features within a library.
(closes issue ASTERISK-18535)
Reported by: Michael Keuter
........
Do parse the option "defaultlanguage" from the [options] section of
asterisk.conf, as in the sample config file. Otherwise the build-time
default language (normally "en") is always the default one.
This patch resolves a crash observed in a load testing environment that
involved the use of the res_ais module. I observed some crashes where
the event delivery callback would get called, but the length parameter
incidcating how much data there was to read was 0. The code assumed
(with good reason I would think) that if this callback got called, there
was an event available to read. However, if the rare case that there's
nothing there, catch it and return instead of blowing up.
More specifically, the change always ensure that the size of the received
event in the cluster is always big enough to be a real ast_event.
Ensure frames are not written to dialed channel if ringback is requested
When a single channel was dialed and there was media to be forwarded to the
calling channel, the media was written without regard for ringback causing
silence to be heard in some circumstances. This regression was introduced
when the meaning of "single" changed to mean only the number of channels
dialed.
Lock the peer->mvipvt to avoid crashes with SIP history enabled
After the launch of 1.6 event-based MWI we have two threads handling the peer->mwipvt,
which cause issues with SIP history additions in combination with the max limit for
number of history entries.
Review: https://reviewboard.asterisk.org/r/1373/
(closes issue ASTERISK-18288)
Thanks to irrot for peer review. Work with this bug funded by IPvision AS
........
IAX2 does not support IPv6 and getting such addresses from DNS can cause error
messages on the remote end involving bad IPv4 address casts in the presence of
IPv6/IPv4 tunnels. This patch ensures that IAX2 will not encounter IPv6
addresses via DNS queries.
build_peer doesnt unlink a peer object from peers_by_ip container which leads to a wrong refcounter value.
adding an ao2_unlink from the peers_by_ip container fix it.
Updated SIP 484 handling; added Incomplete control frame
When a SIP phone uses the dial application and receives a 484 Address
Incomplete response, if overlapped dialing is enabled for SIP, then
the 484 Address Incomplete is forwarded back to the SIP phone and the
HANGUPCAUSE channel variable is set to 28. Previously, the Incomplete
application dialplan logic was automatically triggered; now, explicit
dialplan usage of the application is required.
Additionally, this patch adds a new AST_CONTOL_FRAME type called
AST_CONTROL_INCOMPLETE. If a channel driver receives this control frame,
it is an indication that the dialplan expects more digits back from the
device. If the device supports overlap dialing it should attempt to
notify the device that the dialplan is waiting for more digits; otherwise,
it can handle the frame in a manner appropriate to the channel driver.
(closes issue ASTERISK-17288)
Reported by: Mikael Carlsson
Tested by: Matthew Jordan
Fix crash with res_fax when MALLOC_DEBUG and "core stop gracefully" are used.
Asterisk crashes if MALLOC_DEBUG is enabled when res_fax tries to
unregister its logger level.
* Make ast_logger_unregister_level() use ast_free() instead of free().
When MALLOC_DEBUG is enabled, ast_free() does not degenerate into a call
to free(). Therefore, if you allocated memory with a form of ast_malloc
you must free it with ast_free.
........
* Made AMI action Park not say anything to the parker channel (AMI header
Channel2) since the AMI action is a third party parking the call. (This
is a change in behavior that cannot be preserved without a lot of effort.)
* Made not play pbx-parkingfailed if the Park 's' option is used.
Adding the Feature to sent a Reason Header in a SIP Cancel message by set the flag AST_FLAG_ANSWERED_ELSEWHERE before doing a masquerade in the pickup function.
........
Stefan Schmidt [Wed, 7 Sep 2011 13:26:50 +0000 (13:26 +0000)]
Adding the Feature to sent a Reason Header in a SIP Cancel message by set the flag AST_FLAG_ANSWERED_ELSEWHERE before doing a masquerade in the pickup function.
MusicOnHold has extra unref which may lead to memory corruption and crash.
The problem happens when a call is disconnected and you had started a MOH
class that does not use the files mode. If you define REF_DEBUG and
recreate the problem, it will announce itself with the following warning:
Attempt to unref mohclass 0xb70722e0 (default) when only 1 ref remained,
and class is still in a container!
* Fixed moh_alloc() and moh_release() functions not handling the
state->class reference consistently.
(closes issue ASTERISK-18346)
Reported by: Mark Murawski
Patches:
jira_asterisk_18346_v1.8.patch (license #5621) patch uploaded by rmudgett
Tested by: rmudgett, Mark Murawski
Fix potential memory allocation failure crashes in config.c.
* Added required checks to the returned memory allocation pointers to
prevent crashes.
* Made ast_include_rename() create a replacement ast_variable list node if
the new filename is longer than the available space. Fixes potential
crash and memory leak.
* Factored out ast_variable_move() from ast_variable_update() so
ast_include_rename() can also use it when creating a replacement
ast_variable list node.
* Made the filename stuffed at the end of the struct a minimum allocated
size in ast_variable_new() in case ast_include_rename() changes the stored
filename.
* Constify struct char pointers pointing to strings stuffed at the end of
the struct for: ast_variable, cache_file_mtime, and ast_config_map.
* Factored out cfmtime_new() to remove inlined code and allow some struct
pointers to become const.
* Removed the list lock from struct cache_file_mtime that was never used.
* Added doxygen comments to several structure elements and better
documented what strings are stuffed at the struct end char array.
* Reworked ast_config_text_file_save() and set_fn() to handle allocation
failure of the include file scratch pad object tracking blank lines.
* Made ast_config_text_file_save() fn[] declared with PATH_MAX to ensure
it is long enough for any filename with path. Also reduced the number of
container fileset buckets from a rediculus 180,000 to 1023.
As a function pointer, the reference has to be resolved at load time
irrespective of the RTLD_LAZY flag. Creating a local alias solves
this problem, because the structure is initialized with that local
function pointer, while the actual function can remain lazily linked
until runtime.
The reason why this is important is because we lazily load function
references during the module loading process, in order to obtain
priority values for each module, ensuring that modules are loaded in
the correct order. Previous to this change, when this module was
initially loaded, the module loader would emit a symbol resolution
error, because of the above requirement.
Closes ASTERISK-18399 (reported by Mikael Carlsson, fix suggested by
Walter Doekes, patch by me)
........
No DAHDI channel available for conference, user introduction disabled.
The following error will consistently occur when trying to dial into a
MeetMe conference when the server does not have DAHDI hardware installed:
app_meetme.c: No DAHDI channel available for conference, user introduction
disabled (is chan_dahdi loaded?)
While chan_dahdi is loaded correctly during compilation and install of
Asterisk/Dahdi, including associated modules, etc., a chan_dahdi.conf
configuration file in /etc/asterisk is not created by FreePBX if hardware
does not exist, causing MeetMe to be unable to open a DAHDI pseudo
channel.
* Allow chan_dahdi to create a pseudo channel when there is no
chan_dahdi.conf file to load.
projects / thirdparty / asterisk.git / log
