Prior to this change, a common method for determining if a timeout
was reached was to call a function such as ast_waitfor_n() and inspect
the out parameter that told how many milliseconds were left, then use
that as the input to ast_waitfor_n() on the next go-around.
The problem with this is that in some cases, submillisecond timeouts
can occur, resulting in the out parameter not decreasing any. When this
happens thousands of times, the result is that the timeout takes much
longer than intended to be reached. As an example, I had a situation where
a 3 second timeout took multiple days to finally end since most wakeups
from ast_waitfor_n() were under a millisecond.
This patch seeks to fix this pattern throughout the code. Now we log the
time when an operation began and find the difference in wall clock time
between now and when the event started. This means that sub-millisecond timeouts
now cannot play havoc when trying to determine if something has timed out.
Part of this fix also includes changing the function ast_waitfor() so that it
is possible for it to return less than zero when a negative timeout is given
to it. This makes it actually possible to detect errors in ast_waitfor() when
there is no timeout.
(closes issue ASTERISK-20414)
reported by David M. Lee
Refactor ast_timer_ack to return an error and handle the error in timer users
Currently, if an acknowledgement of a timer fails Asterisk will not realize
that a serious error occurred and will continue attempting to use the timer's
file descriptor. This can lead to situations where errors stream to the
CLI/log file. This consumes significant resources, masks the actual problem
that occurred (whatever caused the timer to fail in the first place), and
can leave channels in odd states.
This patch propagates the errors in the timing resource modules up through
the timer core, and makes users of these timers handle acknowledgement
failures. It also adds some defensive coding around the use of timers
to prevent using bad file descriptors in off nominal code paths.
Note that the patch created by the issue reporter was modified slightly for
this commit and backported to 1.8, as it was originally written for
Asterisk 10.
This patch does two things:
1) It properly unregisters the manager CLI commands
2) It cleans up AMI users on exit. Prior to this patch, the AMI users
were not being disposed of properly, resulting in a memory leak.
Merged revisions 375793 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
r375797 | mjordan | 2012-11-03 21:42:43 -0500 (Sat, 03 Nov 2012) | 9 lines
Only deref a reserved gateway session if we actually reserved one
Its perfectly acceptable to have a gateway session unreserved when we go to
first allocate one. Unreffing the reserved gateway session - when its NULL -
will result in an assertion error.
This problem was caught by the Asterisk Test Suite (once we had enough of the
debugging flags enabled)
................
r375801 | mjordan | 2012-11-03 22:08:12 -0500 (Sat, 03 Nov 2012) | 17 lines
Don't attempt to purge sessions when no sessions exist
Manager's tcp/tls objects have a periodic function that purge old manager
sessions periodically. During shutdown, the underlying container holding
those sessions can be disposed of and set to NULL before the tcp/tls periodic
function is stopped. If the periodic function fires, it will attempt to
iterate over a NULL container.
This patch checks for whether or not the sessions container exists before
attempting to purge sessions out of it. If the sessions container is NULL,
we simply return.
Note that this error was also caught by the Asterisk Test Suite.
........
Merged revisions 375800 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
This patch is a modified version of a patch originally committed for the
Asterisk 11 branch in r375756. A portion of that patch, that fixed the
memory leak during unloading XML documentation, applies to branches 1.8
and 10 as well.
The patch for this issue was modified for these two branches.
Merged revisions 375758 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
r375761 | mjordan | 2012-11-03 20:13:37 -0500 (Sat, 03 Nov 2012) | 15 lines
Properly finalize prepared SQLite3 statements to prevent memory leak
The AstDB uses prepared SQLite3 statements to retrieve data from the SQLite3
database. These statements should be finalized during Asterisk shutdown so
that the SQLite3 database can be properly closed. Failure to finalize the
statements results in a memory leak and a failure when closing the database.
This patch fixes those issues by ensuring that all prepared statements are
properly finalized at shutdown.
Prevent multiple CDR batches from conflicting when scheduling the CDR write
The Asterisk Test Suite caught an error condition where a scheduled CDR batch
write can be deleted twice if two channels attempt to post their CDRs at the
same time. The batch CDR mutex is locked while the CDRs are appended to the
current batch list; however, it is unlocked prior to actually scheduling the
CDR write. As such, two threads can attempt to remove the currently scheduled
batch write at the same time, resulting in an assertion error.
This patch extends the time that the mutex is locked to encompass actually
scheduling the write. This prevents two threads from unscheduling the
currently scheduled write at the same time.
........
Merged revisions 375727 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Replace links to missing text files removed in the 1.6.x series with links to the wiki. Doxygen can handle URLs fine, don't atempt to quote them. Also update the wiki link in the Readme to get everyone on the same page.
(issue ASTERISK-20259)
........
Merged revisions 375698 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
chan_misdn: Timer primitives must be handled first.
The frm->addr is a different "address space" than the stack/instance
address of other Lx primitives. The test for B channel instance address
could fail.
Patches:
patch01_timers.diff (license #6372) patch uploaded by Guenther Kelleter
* An NT-PTMP cannot de/establish L2 since it doesn't know the TEIs.
* On NT-PTP L2 is started when L1 is finally active in handle_l1.
* L2 deactivation logging cleanup.
* L2 aggregate link status is unknown for NT-PTMP, show as "UNKN".
* Removed unused functions and code for L2 handling.
Patches:
patch03_L2estab.diff (license #6372) patch uploaded by Guenther Kelleter
Modified
Sending PH prim via lower_id layer (3 or 1) simply does not work. For TE
(3) it returns an error (len=-6) which is not evaluated by handle_l1(), so
the L1 layer status ends up wrong. Instead PH must be sent via L4, only
then does it reach L1 without an error message.
And NT PH prims only reach L1 when they are sent to layer 2 id.
--> use upper_id to send PH primitives.
* Check for errors in PH_(DE)ACTIVATE | CONFIRM.
* Debug messages are improved.
* The lower_id is now not used for anything, except: Why is lower_id layer
deleted when it wasn't created? I removed this code since it looks very
wrong.
Patches:
patch04_l1activation.diff (license #6372) patch uploaded by Guenther Kelleter
If you make 2 calls out an NT PTMP port which is not connected to any
phone, the B channel associated with that call becomes unusable until
Asterisk is restarted.
The problem is the EVENT_SETUP is queued when L1 is not up in
misdn_lib_send_event(). If L1 cannot be activated the event won't be
dequeued. It gets even worse when the call is hung up. The queued
EVENT_SETUP will be overwritten by an EVENT_DISCONNECT. The reserved B
channel then will never be freed. If later someone connects a phone to
the port, L1 will eventually activate and the queued EVENT_DISCONNECT is
sent down the stack. However, it is ignored because it is the wrong call
state.
The real fix would be that activation and queueing for a new SETUP is done
by the NT stack. But since it doesn't, the workaround must be removed
because it doesn't always work.
Fix: The event is no longer queued but immediately sent to the stack. If
L1 cannot be activated, the L3 state machine that was started by the
EVENT_SETUP will do its work, i.e. a timeout will release the B channel
properly. The SETUP possibly cannot be sent the first time but is resent
by T303 in case L1 could be activated.
Patches:
patch05_bchan-loss.diff (license #6372) patch uploaded by Guenther Kelleter
Modified
Fix Wrong Result In Debug Message For SDP Origin Processing
While looking at some debug logs, I noticed that it was being reported that the
SDP origin line was unsupported or failed. Upon looking into this on my local
machine, I found that I too was getting this debug message yet everything seemed
to be getting processed properly. What was discovered is, that, the variable to
determine what is displayed in the debug message for the SDP line that was
processed, was not being set for the origin line when the result was successful.
This patch fixes this and was tested on local machine.
........
Merged revisions 375594 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Properly extract the Body information of an EWS calendar item
Unlike all other calendar modules, res_calendar_ews fails to extract the Body
information for a calendar item. This is due, in part, to a quirk in the
schema in the XML - not only does a CalendarItem contain a Body element, but
the CalendarItem exists as a descendant of a different Body element. The neon
parser was erroneously skipping all Body elements.
This patch fixes that by bypassing Body elements that are not a child of
CalendarItem, and parsing the Body element out if it is a child.
Note that the original patch by Terry Wilson only needed slight modifications
to make it properly pull the Body information out; as such, while I've linked
to the patch that I uploaded for Dmitry, I've attributed the patch to Terry.
(closes issue ASTERISK-19738)
Reported by: Dmitry Burilov
Tested by: Dmitry Burilov
patches:
calendar_ews_body_2012_10_29.diff uploaded by Terry Wilson (license 6283)
........
Merged revisions 375528 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Ensure that the Queue application tracks busy members in off nominal situations
There are a few code paths where the Queue application fails to count a paused
or in use queue member as being 'busy'. This can cause callers to get stuck
in the Queue until a paused agent unpauses themselves.
Prevent resetting of NATted realtime peer address on reload.
If a "sip reload" is issued for a SIP peer, then his
IP address will be cleared, thus resulting in forgetting the
public IP address. Asterisk will then attempt to route SIP
traffic to the private IP address.
The fix here is to make "sip reload" ignore realtime peers
when "host = dynamic" is spotted. Realtime peers can now only
have their IP address reset if they have gone from being not
dynamic to being dynamic.
(closes issue ASTERISK-18203)
reported by daren ferreira
(closes issue ASTERISK-20572)
reported by JoshE
Patches:
fix_nat_realtime.diff uploaded by JoshE (license #6075)
........
Merged revisions 375415 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
chan_dahdi: Fix segfault dereferencing a NULL tech_pvt.
The tech support customer was using the AMI Redirect action shortly after
a call was placed. While the channel tried to do an ast_read(), the
masquerade resulting from the channel redirect took place. The masquerade
in the middle of the ast_read() resulted in the segfault.
core: Fix a memory leak in app.c from an early return
ast_app_group_match_get_count allocates memory with the regcomp
function and we previously forgot to free it when bailing out
due to a regex compilation failure against category.
app_queue: Make ordering of rrmemory/rrordered persist over add/remove members
Prior to this patch, adding, removing or reloading members to rrmemory would
cause the order to become completely jumbled. Now it behaves more or less like
rrordered other than the fact that it stores the members on a hash table rather
than a linked list. This patch also prevents removal of members and member
reloads from jumbling rrordered queues.
(issue AST-989)
Reported by: Thomas Arimont
Review: https://reviewboard.asterisk.org/r/2164/
........
Merged revisions 375216 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
build_tools: Allow Asterisk to report git SHAs in version string.
Make git more attractive for managing work-in-progress. Especially
convenient when a potential patch set needs to be tested on multiple
platforms since one can use git to keep all the test environments in sync
independent of a subversion server.
Now the Asterisk version will show the exact git SHA5 that was used when
building (still appended by "M" if there are local modifications) from a
git clone of the Asterisk repository so the developer can more easily know
what is actually under test.
This has zero impact for those not using git with the exception of an
extra test in the configure script to gather git's path. This is
necessary to prevent "sudo make install" from failing since git may not be
in the path in make's shell environment.
Ensure Asterisk fails TCP/TLS SIP calls when certificate checking fails
When placing a call to a TCP/TLS SIP endpoint whose certificate is not
signed by a configured CA certificate, Asterisk would issue a warning
and continue to process the call as if there was not an issue with the
certificate. Asterisk now properly fails the call if the certificate
fails verification or if the certificate does not exist when
certificate checking is enabled (the default behavior).
Fix some potential misuses of ast_str in the code.
Passing an ast_str pointer by value that then calls
ast_str_set(), ast_str_set_va(), ast_str_append(), or
ast_str_append_va() can result in the pointer originally
passed by value being invalidated if the ast_str had
to be reallocated.
This fixes places in the code that do this. Only the
example in ccss.c could result in pointer invalidation
though since the other cases use a stack-allocated ast_str
and cannot be reallocated.
I've also updated the doxygen in strings.h to include
notes about potential misuse of the functions mentioned
previously.
Update config.guess and config.sub to revision fb456b34ef4aa02b95dc6be69aaa66fa94a844fb from the savannah.gnu.org git
repo. Adds support for e.g. aarch64 (ARM 64bit).
Kinsey Moore [Fri, 12 Oct 2012 21:56:59 +0000 (21:56 +0000)]
Avoid a segfault on invalid format names
If a format name was not found by ast_getformatbyname, a NULL pointer
would be passed into ast_format_rate and immediately dereferenced.
This ensures that a valid pointer is used since the structure is
already allocated on the stack.
(closes issue DPH-523) Reported-by: Steve Pitts
git-svn-id: https://origsvn.digium.com/svn/asterisk/branches/10-digiumphones@374931 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Do not use a FILE handle when doing SIP TCP reads.
This is used to solve an issue where a poll on a file
descriptor does not necessarily correspond to the readiness
of a FILE handle to be read.
This change makes it so that for TCP connections, we do a
recv() on the file descriptor instead.
Because TCP does not guarantee that an entire message or even
just one single message will arrive during a read, a loop has
been introduced to ensure that we only attempt to handle a
single message at a time. The tcptls_session_instance structure
has also had an overflow buffer added to it so that if more
than one TCP message arrives in one go, there is a place to
throw the excess.
Huge thanks goes out to Walter Doekes for doing extensive review
on this change and finding edge cases where code could fail.
(closes issue ASTERISK-20212)
reported by Phil Ciccone
Fix incorrect billing duration reported when batch mode is enabled
Similar to r369351, the billing duration can be skewed when batch mode is
enabled. This happened much more rarely than the duration, as it only
occured when the call was answered (thereby indicating an actual answer
time) and immediately hung up on (indicating a billsec of 0). Since
a billing time of '0' can either mean that the call immediately ended
or that the CDR was improperly answered, we have to use additional information
to know whether or not we can trust the CDR billsec value. Prior to this
patch, we looked to see if we had a valid answer time. If we did, and
billsec was zero, we used the current time to calculate what billsec value
we could from the CDR being written. If batch mode is enabled, this will
incorrectly report a billsec value being much greater than the actual
duration of the call.
Instead of relying on the presence of an answer time to know whether or not
we can re-calculate the billsec for the CDR, we now also use the presence
of the CDR's end time to know if we need to re-calculate or whether we can
trust the billsec value that we have. This prevents erroneous jumps in the
billsec value, while still making sure that in the worst case, some billing
time will be calculated.
(closes issue AST-1016)
Reported by: Thomas Arimont
Tested by: Thomas Arimont
........
Merged revisions 374843 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
app_queue: Made pass connected line updates from the caller to ringing queue members.
Party A calls Party B
Party B puts Party A on hold.
Party B calls a queue.
Ringing queue member D sees Party B identification.
Party B transfers Party A to the queue.
Queue member D does not get a connected line update for Party A.
Queue member D answers the call and still sees Party B information.
However, if Party A later transfers the call to Party C then queue member
D gets a connected line update for Party C.
* Made pass connected line updates from the caller to queue members while
the queue members are ringing.
(closes issue AST-1017)
Reported by: Thomas Arimont
(closes issue ABE-2886)
Reported by: Thomas Arimont
Tested by: rmudgett
........
Merged revisions 374801 from https://origsvn.digium.com/svn/asterisk/be/branches/C.3-bier
........
Merged revisions 374802 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Fix execution of 'i' extension due to uninitialized variable.
The fix for ASTERISK-18243 added code that could potentially use
dst_exten[] uninitialized. As a result the 'i' exten may not be executed
when it should.
(closes issue ASTERISK-20455)
Reported by: Richard Miller
Patches:
pbx-1.8.16.0.diff (license #5685) patch uploaded by Richard Miller
Made some cosmetic modifications.
........
Merged revisions 374758 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
If scan_service() cannot open the spool file, it logs a message saying
that it will delete the file and calls remove_from_queue() to do it.
However, remove_from_queue() fails to delete the spool file because struct
outgoing has not yet been fully initialized.
* Merged allocating a new struct outgoing and init_outgoing() into
new_outgoing(). Allocation is initialization.
* Made apply_outgoing() not initialize the spool filename in struct
outgoing.
* Made apply_outgoing() call ast_trim_blanks() and ast_skip_blanks()
rather than manually inlining them.
* Reduced indentation levels in apply_outgoing().
* Fixed a garbled comment in remove_from_queue().
* Reworked scan_service() to simplify it.
(closes issue ASTERISK-17231)
Reported by: David Chappell
Patches:
spool_open_failure.diff (license #4997) patch uploaded by David Chappell
Started with this patch.
........
Merged revisions 374686 from http://svn.asterisk.org/svn/asterisk/branches/1.8
* Fixed some memory leaks on of nominal paths in init_outgoing() when
merging into the new_outgoing() function dealing with o->capabilities.
................
Resolve issues in ConfBridge regarding marked, waitmarked, and unmarked users
Thank's to Neil Tallim (flan)'s tireless testing, issue reporting, and patches
it became clear that app_confbridge had some complex logic in how it handled
interactions between marked, waitmarked, and unmarked users. In particular,
there were some areas in which the interactions between the users resulted
in inconsistent behavior, and app_confbridge was missing logic in how to handle
some corner cases. Some areas included:
* Poor handling of mixing unmarked and waitmarked users
* Inconsistencies in how MOH and muting was applied to various users
* Handling of various announcements for different user profile options
flan's patches seem to fix the various issues, but highlighted how hard the
code could be to maintain. In an attempt to make things easier to maintain and
to more fully enumerate the various cases that exist, this patch breaks up the
logic into a state machine-like setup.
Please note that the various state transitioned are documented on the Asterisk
wiki:
Note that for the following issues, mjordan uploaded the patch, although it
was written by twilson. Any contributor license discrepency is due to that.
(closes issue ASTERISK-20181)
Reported by: Jonathan White
Tested by: Jonathan White
patches:
bugASTERISK-19562_ASTERISK-19726_ASTERISK-20181.patch uploaded by twilson (license 6283)
........
In AMI's parser, when it receives a long line (> 1024 characters), it discards
that line, but continues to process the message normally.
Typically, this is not a problem because a) who has lines that long and b)
usually a discarded line results in an invalid message. But if that line is
specifying an optional field, then the message will be processed, you get a
'Response: Success', but things don't work the way you expected them to.
This patch changes the behavior when a line-too-long parse error occurs.
* Changes the log message to avoid way-too-long (and truncated anyways) log
messages
* Adds a 'parsing' status flag to Response: Success
* Sets parsing = MESSAGE_LINE_TOO_LONG if, well, a line is too long
* Responds with an appropriate error if parsing != MESSAGE_OKAY
(closes issue AST-961)
Reported by: John Bigelow
Review: https://reviewboard.asterisk.org/r/2142/
........
r374581 | dlee | 2012-10-05 15:20:28 -0500 (Fri, 05 Oct 2012) | 1 line
I've committed too much. Reverting part of r374570.
........
Merged revisions 374570,374581 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
In handle_frm_te() after calling misdn_lib_send_event(bc,
EVENT_RELEASE_COMPLETE) bc is emptied, cleaned and set not in use,
although misdn_lib_send_event() already did the same. This is bad. When
it's not in use we are not allowed to touch it.
* Moved log message in front of the resulting actions and fixed it to
match the case.
Patches:
patch5_bccleanup.diff (license #6372) patch uploaded by Guenther Kelleter
chan_misdn: We must initialize cause on sending a DISCONNECT.
We must initialize cause on sending a DISCONNECT, so it is later correctly
indicated to ast_channel in case the answer (RELEASE/RELEASE_COMPLETE)
does not include one.
Patches:
patch7_hangupcause.diff (license #6372) patch uploaded by Guenther Kelleter
chan_misdn: setup_bc() is called too early for an incoming SETUP on TE.
This prevents the B channel from being setup for HDLC mode when requested
by the bearer capability and config option hdlc=yes. It violates
ETS300102 Ch.5.2.3.2: "The user, in any case, must not connect to the
channel until a CONNECT ACKNOWLEDGE message has been received."
* Call setup_bc() on receipt of CONNECT_ACKNOWLEGDE for PTMP, and on first
response to SETUP for PTP.
Patches:
abe-2881-2.diff (license #6372) patch uploaded by Guenther Kelleter
Modified.
Fix a regression from direct media ACLs where the directrtpsetup option no longer works.
A check was added for direct media ACLs that immediately forbid remote bridging if there
was no bridged channel. This caused directrtpsetup to no longer function as it needs this
information before bridging actually occurs.
Logic has now been adjusted so if there is no bridged channel a remote bridge will still
be attempted.
The AMI DBDelTree command will return Success/Key tree deleted successfully even
if the given key does not exist. The CLI command 'database deltree' had a
similar problem, but was saved because it actually responded with '0 database
entries removed'. AGI had a slightly different error, where it would return
success if the database was unavailable.
This came from confusion about the ast_db_deltree retval, which is -1 in the
event of a database error, or number of entries deleted (including 0 for
deleting nothing).
* Changed some poorly named res variables to num_deleted
* Specified specific errors when calling ast_db_deltree (database unavailable
vs. entry not found vs. success)
* Fixed similar bug in AGI database deltree, where 'Database unavailable'
results in successful result
(closes issue AST-967)
Reported by: John Bigelow
Review: https://reviewboard.asterisk.org/r/2138/
........
Merged revisions 374426 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
dsp.c User configuration of DTMF_NORMAL_TWIST and DTMF_REVERSE_TWIST values
Asterisk's DTMF Specifications are based on AT&T specs, which may not be compatible in other countries.
Various countries have different specifications for the maximum power level differences
between the DTMF low group and high group of frequencies.
Power level difference between frequencies for different Administrations/RPOAs
NTT = Max. 5 dB
AT&T = 4dB(reverse) to 8dB(normal)
Danish = Max. 6 dB
Australian = Max. 10 dB
Brazilian = Max. 9 dB
ETSI = Max. 6 dB from ETSI ES 201 235-3 V1.3.1 (2006-03)
Now allow 4 variables to be individually configured in dsp.conf, with reasonable min/max of 2dB to 20dB.
Default is AT&T specifications
Add's the following variables to dsp.conf
;dtmf_normal_twist=6.31
;dtmf_reverse_twist=2.51
;relax_dtmf_normal_twist=6.31
;relax_dtmf_reverse_twist=3.98
Check for presence of buddy in info/dinfo handlers
The res_jabber resource module uses the ASTOBJ library for managing its ref
counted objects. After calling ASTOBJ_CONTAINER_FIND to locate a buddy object,
the pointer to the object has to be checked to see if the buddy existed.
Prior to this patch, the buddy object was not checked for NULL; with this patch
in both aji_client_info_handler and aji_dinfo_handler the pointer is checked
before used and, if no buddy object was found, the handlers return an error
code.
This patch does not take the approach that our JID can be used to log in from
another resource. If that approach is desired, an improvement could be made to
this patch to create the buddy on the fly. This patch seeks only to prevent
Asterisk from crashing.
Note that multiple people have proposed patches for this issue; the patch being
committed here is based on those.
(closes issue ASTERISK-19532)
Reported by: Karsten Wemheuer
Tested by: Byron Clark
patches:
fix-jabber uploaded by Karsten Wemheuer (license #5930)
xmpp_no_crash_with_ejabberd.patch uploaded by Byron Clark (license #6157)
........
Destroy the generic_monitors container after the core_instances in ccss
For each item in core_instances disposed of in the shutdown of ccss, any
generic monitor instances referenced by the objects will be removed from
generic_monitors during their destruction. Hilarity ensues if
generic_monitors no longer exists.
Thanks to the Asterisk Test Suite's generic_ccss test for complaining loudly
when it ran into this.
................
Destroy the generic_monitors container after the core_instances in ccss
For each item in core_instances disposed of in the shutdown of ccss, any
generic monitor instances referenced by the objects will be removed from
generic_monitors during their destruction. Hilarity ensues if
generic_monitors no longer exists.
Thanks to the Asterisk Test Suite's generic_ccss test for complaining loudly
when it ran into this.
........
Ensure Shutdown AMI event is still fired during Asterisk shutdown
Richard pointed out that having the manager dispose of itself gracefully
during shutdown meant that the Shutdown event will no longer get fired.
This patch moves the AMI event just prior to running the atexit callbacks.
........
Merged revisions 374230 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Richard pointed out two problems with the check-in from r374177:
* The ast_msg_shutdown function declaration doesn't match the prototype
in main/message.c.
* The ref/alloc function usage in astobj2 (in 11+) can use the ao2_t_* variants
of the functions to allow the REF_DEBUG flag to enable/disable their debug
counterparts.
........
Jason Parker [Tue, 2 Oct 2012 16:41:11 +0000 (16:41 +0000)]
Fix a variety of ref counting issues
This patch resolves a number of ref leaks that occur primarily on Asterisk
shutdown. It adds a variety of shutdown routines to core portions of
Asterisk such that they can reclaim resources allocate duringd initialization.
Use ast_copy_string instead of strncpy to guarantee a NUL terminated string.
................
r374135 | seanbright | 2012-10-01 12:52:38 -0500 (Mon, 01 Oct 2012) | 23 lines
app_queue: Support persisting and loading of long member lists.
Greenlight in #asterisk brought up that he was receiving an error message "Could
not create persistent member string, out of space" when running app_queue in
Asterisk 10. dump_queue_members() made an assumption that 8K would be enough to
store the generated string, but with queues that have large member lists this is
not always the case. This patch removes the limitation and uses ast_str instead
of a fixed sized buffer.
The complicating factor comes from the fact that ast_db_get requires a buffer
and buffer size argument, which doesn't let us pull back more than what we pass
in, so I introduced a new ast_db_get_allocated() which returns an ast_strdup()'d
copy of the value from astdb.
As an aside, I did some testing on the maximum size of data that we can store in
the BDB library we distribute and was able to store a 10MB string and retrieve
it with no problems, so I feel this is a safe patch.
The opinion of development was that it is both improper to have Matt's
personal email address used in the source and that the command wouldn't
be useful without it.
Fix SendDTMF crash and channel reference leak using channel name parameter.
The SendDTMF channel name parameter has two issues.
1) Crashes if the channel name does not exist.
2) Leaks a channel reference if the channel is the current channel.
Problem introduced by ASTERISK-15956.
* Updated SendDTMF documentation.
* Renamed app to senddtmf_name and tweaked the type.
........
Merged revisions 373945 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
loader: Ensure dependent modules are properly initialized.
If an Asterisk module specifies a dependency in ast_module_info.nonoptreq, it
is possible for Asterisk to skip calling the modules's .load function.
Asterisk was loading and linking the module via load_dynamic_module() but was
not adding the module to the resource_heap. Therefore the module was not
initialized based on it's priority along with the other modules in the heap.
Now use load_resource() instead of load_dynamic_module() for non-optional
requirement. This will add the module to the resource_heap so the module can
be properly initialized in the correct order.
This is required if there are any module global data structures initialized in
the .load() callback for the module on platforms which do not support weak
references.
Fix an issue where Local channels dialed by app_queue are considered in use immediately.
The chan_local channel driver returns a device state of in use even if a created Local
channel has not yet been dialed. This fix changes the logic to return a state of not
in use until the channel itself has been dialed.
Fixed meetme tab completion and command documentation.
* Removed unnecessary case sensitivity in meetme list, lock, unlock, mute,
unmute, and kick commands.
* Separated meetme lock/unlock, mute/unmute, and kick commands into their
own registered commands to simplify tab completion and parameter checking.
meetme_lock_cmd(), meetme_mute_cmd(), and meetme_kick_cmd()
Fix error where improper IMAP greetings would be deleted.
(closes issue ASTERISK-20435)
Reported by: fhackenberger
Patches:
asterisk-20435-imap-del-greeting.diff uploaded by Michael L. Young (License #5026)
(with suggested modification made by me)
........
Merged revisions 373735 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Fix T.38 support when used with chan_local in between.
Users of the T.38 API can indicate AST_T38_REQUEST_PARMS on a channel to request that the
channel indicate a T.38 negotiation with the parameters present on the channel. The return
value of this indication is expected to be AST_T38_REQUEST_PARMS upon success but with
chan_local involved this could never occur.
This fix changes chan_local to always return AST_T38_REQUEST_PARMS for this situation. If
the underlying channel technology on the other side does not support T.38 this would have
been determined ahead of time using ast_channel_get_t38_state and an indication would
not occur.
Properly handle UAC/UAS roles for SIP session timers
The SIP session timer mechanism contains a mandatory 'refresher' parameter
(included in the Session-Expires header) which is used in the session timer
offer/answer signaling within a SIP Invite dialog. It looks like asterisk is
interpreting the uac resp. uas role only as the initial role of client and
server (caller is uac, callee is uas). The standard rfc 4028 however assigns
the client role to the ((RE)-Invite) requester, the server role to the
((RE)-Invite) responder.
This patch has Asterisk track the actual refresher as "us" or "them" as opposed
to relying on just the configured "uas" or "uac" properties.
(closes issue AST-922)
Reported by: Thomas Airmont
"show" completion option for "queue" shouldn't appear twice
When tab-completing CLI commands starting with "queue", "show" appeared
twice in the list due to the way that Asterisk's tab completion
functions and the order in which the commands were registered. The
registration order has been altered to resolve this issue.
(closes issue AST-940) Reported-by: Steve Pitts
........
Merged revisions 373666 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Fix an issue where a caller to ast_write on a MulticastRTP channel would determine it failed when in reality it did not.
When sending RTP packets via multicast the amount of data sent is stored in a variable and returned
from the write function. This is incorrect as any non-zero value returned is considered a failure while
a return value of 0 is success. For callers (such as ast_streamfile) that checked the return value
they would have considered it a failure when in reality nothing went wrong and it was actually a success.
The write function for the multicast RTP engine now returns -1 on failure and 0 on success, as it should.
Be consistent, send From: "Anonymous" <sip:anonymous@anonymous.invalid>
When setting CALLERID(pres)=unavailable in the dialplan, the From header
in the SIP message contains "Anonymous" <sip:Anonymous@anonymous.invalid>.
For consistency, Asterisk should use a lowercase a in the userpart of the
URI.
* Make the From header use a lowercase A in the userpart of the anonymous
URI.
(closes issue ASTERISK-19838)
Reported by: Antti Yrjola
Patches:
chan_sip_patch_ASTERISK-19838.patch (license #6383) patch uploaded by Antti Yrjola
........
Revert change to res_rtp_asterisk committed in r373236 (1.8)
The change committed in r373236 attempted to account for endpoints that
increased their RTP timestamp in DTMF end of event re-transmissions. This
change attempted to make Asterisk continue to work with endpoints that
failed to follow the RFC while maintaining the fix that allowed for out of
order DTMF to be handled. Unfortunately, there is no free lunch, and this
patch broke any system that sent DTMF immediately after an RTP session was
established or when an SSRC is updated. As such, that patch is being
reverted for the previous behavior.
Endpoints that erroneously increase the RTP timestamp in DTMF end of event
packets will not work properly with Asterisk.
(issue ASTERISK-20424)
........
Merged revisions 373504 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Asterisk v1.8 and later was not as vulnerable to this issue.
* Made find_call() lock each private as it processes the found dialogs.
(Primary cause of ABE-2876)
* Made the other functions that traverse the dialogs container lock each
private as it examines them.
* Fix race condition in sip_call() if the thread that sent the INVITE is
held up long enough for a response to be processed. The p->initid for the
INVITE retransmission could be added after it was canceled by the response
processing.
* Made __sip_destroy() clean up resource pointers after freeing. This is
primarily defensive in case someone has a stale private pointer.
* Removed redundant memset() in reqprep(). The call to init_req() already
does the memset() and is the first reference to req in reqprep().
* Removed useless set of req.method in transmit_invite(). The calls to
initreqprep() and reqprep() have to do this because they memset() the req.
JIRA ABE-2876
..........
Merged -r373423 from https://origsvn.digium.com/svn/asterisk/be/branches/C.3-bier
........
func_audiohookinherit: Document some missed sources.
This patch also mentions that AUDIOHOOK_INHERIT can be used to
transfer MixMonitor audiohooks. There is also wiki that addresses
audiohooks and the use of AUDIOHOOK_INHERIT at the following link:
https://wiki.asterisk.org/wiki/display/AST/Audiohooks
(closes issue ASTERISK-18220)
Reported by: Ishfaq Malik
........
Merged revisions 373467 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Jason Parker [Mon, 24 Sep 2012 21:22:34 +0000 (21:22 +0000)]
Fix a deadlock caused by a race condition between removing a hint and reloading the dialplan and subscribing to the removed hint.
If conditions were right it was possible for both the PBX core and chan_sip to deadlock by both having a lock that the other
wants. In the case of the PBX core it had the contexts lock and wanted a SIP dialog lock, while in the case of chan_sip it
had the SIP dialog lock and wanted the contexts lock.
This fix unlocks the SIP dialog before getting the extension state so that the other thread will not block on trying to lock
it. Once the extension state is retrieved the SIP dialog is locked again and life carries on.
As the SIP dialog is reference counted it is not possible for it to go away after unlocking.
app_queue: Make queue reload members and variants of that work
Prior to this patch, 'queue reload members' cli command did not
work at all. This also affects the manager function 'QueueReload'
when supplied with the 'members: yes' field.
(closes issue AST-956)
Reported by: John Bigelow
........
Merged revisions 373298 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
When processing RFC 2833 DTMF, accomodate increasing timestamps in End events
While endpoints should not be changing the source timestamp between DTMF event
packets, the fact is there exists those endpoints that do exactly that. To
work around this, we absorb timestamps within the expected re-transmit period.
Note that this period only affects End of Event packets, so it should not
prevent the detection of new DTMF digits that happen to arrive right on top
of each other.
(closes issue ASTERISK-20424)
Reported by: Vladimir Mikhelson
Tested by: mjordan, Vladimir Mikhelson
Fix incorrect MeetME conference bridge reference count decrementing and sometimes premature destruction.
When using the 'e' or 'E' option to MeetMe the configured conference bridges are loaded and examined to see
if any are empty. If no conference bridges are empty the caller is prompted to enter the number of one.
This operation left around a pointer to the last created conference bridge still containing participants.
When the caller that was not able to find any empty conference bridge hung up this pointer was disposed of
and the reference count of the conference bridge decremented. If there was only a single participant in the
conference bridge it was ultimately destroyed prematurely.
(closes issue AST-994)
Reported by: John Bigelow
........
Merged revisions 373242 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Ensure that all ConfBridge sounds can be set using CONFBRIDGE function
The CONFBRIDGE function can be used to set the sounds in a ConfBridge
bridge profile. Unfortunately, three sounds were missed in the portion
of the code that applies the settings passed in from the function:
sound_only_one, join, and leave. This patch makes sure that the sounds
passed from the function are applied to the bridge profile.
Fix a regression where direct media was not permitted for calls using SIP INFO DTMF.
A change was committed to fix direct media ACL support. This change wrongly assumed that
only a single channel technology structure exists for chan_sip. This is in fact false as
a second exists for calls using SIP INFO DTMF. The code which performs direct media ACL
checking now checks for both the non-INFO DTMF and INFO DTMF channel technology structures.
Don't crash when passing a NULL message to __astman_get_header.
Before this commit, __astman_get_header would blindly dereference the passed in
'struct message *' to traverse the header list. There are cases, however, such
as '*CLI> sip qualify peer foo' where the message pointer is NULL, so we need
to check for that.
........
Merged revisions 373131 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Made companding law for SS7 calls only determined by SS7 signaling type.
For SS7, the companding law for a call was chosen inconsistently depending
upon ss7type (ITU vs ANSI) and the DAHDI companding default (T1 vs E1).
For incoming calls, the companding law was determined by ss7type. For
outgoing calls, the companding law was determined by the DAHDI default.
With the wrong combination you would get A-law/u-law conflicts. An
A-law/u-law conflict sounds like bad static on the line.
SS7 ITU signaling with E1 line: ok
SS7 ITU signaling with T1 line: noise
SS7 ANSI signaling with E1 line: noise
SS7 ANSI signaling with T1 line: ok
* Fix the companding law used to be determined by the SS7 signaling type
only.
........
Merged revisions 373090 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
When REF_DEBUG is enabled in certain files - most notably ccss.c - the 'tag'
parameter passed to __ao2_ref_debug will be a const char *. The function
currently expects that parameter to not be const. This causes a warning
when compiling, as the const qualifier is being discarded. With dev-mode
enabled, this prevents compiling Asterisk.
This patch makes __ao2_ref_debug's tag and file parameters const.
Resolve memory leaks in TLS initialization and TLS client connections
This patch resolves two sources of memory leaks when using TLS in Asterisk:
1) It removes improper initialization (and multiple re-initializations) of
portions of the SSL library. Asterisk calls SSL_library_init and
SSL_load_error_strings during SSL initialization; collectively this
obviates the need for calling any of the following during initialization
or client connection handling:
* ERR_load_crypto_strings (handled by SSL_load_error_strings)
* OpenSSL_add_all_algorithms (synonym for SSL_library_init)
* SSLeay_add_ssl_algorithms (synonym for SSL_library_init)
2) Failure to completely clean up all memory allocated by Asterisk and by
the SSL library for TLS clients. This included not freeing the SSL_CTX
object in the SIP channel driver, as well as not clearing the error
stack when the TLS client exited.
Note that these memory leaks were found by Thomas Arimont, and this patch
was essentially written by him with some minor tweaks.
(closes issue AST-889)
Reported by: Thomas Arimont
Tested by: Thomas Arimont
patches:
(bugAST-889.patch) by Thomas Arimont (license 5525)
ast_waitfordigit_full would simply pass its timeout to ast_waitfor_nandfds,
expecting it to decrement the timeout by however many milliseconds were
waited. This is a problem if it consistently waits less than 1ms. The timeout
will never be decremented, and we wait... FOREVER!
This patch makes ast_waitfordigit_full manage the timeout itself. It maintains
the previously undocumented behavior that negative timeouts wait forever.
(closes issue ASTERISK-20375)
Reported by: Mark Michelson
Tested by: Mark Michelson
Review: https://reviewboard.asterisk.org/r/2109/
........
Merged revisions 373024 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Add channel name to a warning to make debugging easier.
The "autodestruct with owner in place" message is typically
indicative of a channel reference leak. Printing out the name
of the channel in the message may be helpful when trying to
debug the issue.
........
Merged revisions 372932 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
chan_local: Switch from using a random 4 digit hex identifier to unique id
Changes chan_local channels to use an 8 digit hex identifier generated
atomically and sequentially in order to eliminate the chance of having
multiple channels with the same name during high call volume situations.
(issue ASTERISK-20318)
Reported by: Dan Cropp
Review: https://reviewboard.asterisk.org/r/2104/
........
Merged revisions 372902 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Fix inability to shutdown gracefully due to an unending channel reference.
message.c makes use of a special message queue channel that exists
in thread storage. This channel never goes away due to the fact that
the taskprocessor used by message.c does not get shut down, meaning
that it never ends the thread that stores the channel.
This patch fixes the problem by shutting down the taskprocessor when
Asterisk is shut down. In addition, the thread storage has a destructor
that will release the channel reference when the taskprocessor is destroyed.
(closes issue AST-937)
Reported by Jason Parker
Patches:
AST-937.patch uploaded by Mark Michelson (License #5049)
Tested by Jason Parker
........
When channels get bridged due to an AMI bridge action
or a DTMF attended transfer, the two channels that
get bridged have their application data pointing to
the other channel's name. This means that if one channel
is hung up but the other moves on, it means that the
channel that moves on will have its application data
pointing at freed memory.
(issue ASTERISK-20335)
........
Merged revisions 372840 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Ensure iax2 debug output is displayed when expected
When IAX2 debug was changed from iax_showframe to iax_outputframe,
some instances were missed (or added afterward). This was causing
debug output to not be displayed when expected.
(closes issue ASTERISK-20338) Reported-by: John Covert Patch-by: John Covert
........
Merged revisions 372804 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
This adds a CLI warning when a SDP offer is rejected due to UDPTL
initialization failure. Previously, there was no indication of the
reason for offer rejection in this case.
(closes issue ASTERISK-20357) Reported-by: Francesco Usseglio Gaudi
........
Masquerade: Retain parkinglot settings made by CHANNEL function.
Prior to this patch, the user would have a parkinglot set on a channel that
was parked and when the channel was retrieved, any attempt by that channel
to park would simply use the default. This patch makes parkinglot values
set in this way be retained through the masquerade.
(closes issue AST-990)
Reported by: Nick Huskinson
Patches:
masquerade_parkinglot_patch.diff Uploaded by Jonathan Rose (license 6182)
........
Merged revisions 372736 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
In r356604, SRTP handling was fixed to accomodate multiple crypto keys in an
SDP offer and the ability to re-create an SRTP session when the crypto keys
changed. In certain circumstances - most notably when a phone is put on
hold after having been bridged for a significant amount of time - the act
of re-creating the SRTP session causes problems for certain models of phones.
The patch committed in r356604 always re-created the SRTP session regardless
of whether or not the cryptographic keys changed. Since this is technically
not necessary, this patch modifies the behavior to only re-create the SRTP
session if Asterisk detects that the remote key has changed. This allows
models of phones that do not handle the SRTP session changing to continue
to work, while also providing the behavior needed for those phones that do
re-negotiate cryptographic keys.
Add OPENSSL_INCLUDE to the CFLAGS for ssl.c and tcptls.c.
Without this flag, those files will compile with the system installed
OpenSSL headers (if they exist). This is a real bummer if a different
path was specified using --with-ssl=
(closes issue ASTERISK-20392)
........
Merged revisions 372682 from http://svn.asterisk.org/svn/asterisk/branches/1.8
................
Fix VoicemailUserEntry event headers ServerEmail and MailCommand reported values.
The AMI action VoicemailUsersList VoicemailUserEntry event headers
ServerEmail and MailCommand did not report the global values if they were
not overridden. The VoicemailUserEntry event header ServerEmail was not
populated with the global value if the voicemail user did not override it.
The VoicemailUserEntry event header MailCommand was never populated with a
value.
* Removed unused struct ast_vm_user member mailcmd[].
Free ast_str objects when temp file fails to be created in MiniVM
The previous commit (r372554) was from a patch that was written before
r366880, which ensured that ast_str objects allocated in the sendmail
routine were free'd in off nominal paths. This commit frees the
string objects in the off nominal path introduced in r372554.
Fix file descriptor leak and pointer scope issue in MiniVM when sending mail
When MiniVM sends an e-mail and it has the volgain option set, it will spawn
sox in a separate process to handle the manipulation of the sound file. In
doing so, it creates a temporary file. There are two problems here:
1) The file descriptor returned from mkstemp is leaked
2) The finalfilename character pointer points to a buffer that loses scope
once volgain processing is finished.
Note that in r316265, Russell fixed some gcc warnings by using the return
value of the mkstemp call. A warning was placed in minivm that the file
descriptor was going to be leaked. This patch reverts that change, as it
handles the leak and 'uses' the file descriptor returned from mkstemp.
Ensure listed queues are not offered for completion
When using tab-completion for the list of queues on "queue reset stats"
or "queue reload {all|members|parameters|rules}", the tab-completion
listing for further queues erroneously listed queues that had already
been added to the list. The tab-completion listing now only displays
queues that are not already in the list.
(closes issue AST-963) Reported-by: John Bigelow
........
Fix loss of MOH on an ISDN channel when parking a call for the second time.
Using the AMI redirect action to take an ISDN call out of a parking lot
causes the MOH state to get confused. The redirect action does not take
the call off of hold. When the call is subsequently parked again, the
call no longer hears MOH.
* Make chan_dahdi/sig_pri restart MOH on repeated AST_CONTROL_HOLD frames
if it is already in a state where it is supposed to be sending MOH. The
MOH may have been stopped by other means. (Such as killing the generator.)
This simple fix is done rather than making the AMI redirect action post an
AST_CONTROL_UNHOLD unconditionally when it redirects a channel and thus
potentially breaking something with an unexpected AST_CONTROL_UNHOLD.
Prior to 1.8, it was not necessary for an explicit "type" to be set for an
asterisk LDAP realtime peer. Now the routine find_peer actually checks the
type field during registration and fails to find the peer if it is not set.
The attached patches make the realtime type equal whatever type is being
searched for if the type is 0 upon return from routine build_peer.
(closes issue ASTERISK-17222)
Reported by: John Covert
Patch by: David Vossel
Tested by: Darren Sessions
projects / thirdparty / asterisk.git / log
