Timo Teräs [Thu, 17 Nov 2016 14:25:41 +0000 (16:25 +0200)]
codec_dahdi: Fix poll.h include.
POSIX defines poll.h. sys/poll.h should not be used as it is c-library
internal header which may or may not exist. Notably in musl including
sys/poll.h generates warning of being incorrect.
gestoip2 [Fri, 11 Nov 2016 14:16:50 +0000 (14:16 +0000)]
res_rtp_asterisk: RTT miscalculation in RTCP
When retrieving RTCP stats for PJSIP channels, RTT values are unreliable.
RTT calculation is correct, but the data representation isn't. RTT is
represented by a 32-bit fixed-point number with the integer part in the
first 16 bits and the fractional part in the last 16 bits. In order to
get the RTT value, the fractional part is miscalculated, there is an
unnecessary 16 bit shift that causes overflow. Besides this there is
another mistake, when transforming the integer value to the fixed point
fractional part via bitwise operation, that loses precision.
* RTT fractional part is no longer shifted, avoiding overflow.
* RTT fractional part is transformed to its fixed-point value more
precisely.
* Fixed timeval2ntp() and ntp2timeval() second fraction conversions.
* Fixed NTP timestamp report logging. The usec was inexplicably
multiplied by 4096.
ASTERISK-26566 #close
Reported by Hector Royo Concepcion
Timo Teräs [Fri, 11 Nov 2016 06:29:40 +0000 (08:29 +0200)]
addons/chan_mobile: do not use strerror_r
The two reasons why it might be used are that some systems do not
implement strerror in thread safe manner, and that strerror_r returns
the error code in the string in case there's no error message.
However, all of asterisk elsewhere uses strerror() and assumes it
to be thread safe. And in chan_mobile the errno is also explicitly
printed so neither of the above reasons are valid.
The reasoning to remove usage is that there are actually two versions
of strerror_r: XSI and GNU. They are incompatible in their return
value, and there's no easy way to figure out which one is being
used. glibc gives you the GNU version if _GNU_SOURCE is defined,
but the same feature test macro is needed for other symbols. On
all other systems you assumedly get XSI symbol, and compilation warnings
as well as non-working error printing.
Thus the easiest solution is to just remove strerror_r and use
strerror as rest of the code. Alternative is to introduce ast_strerror
in separate translation unit so it can request the XSI symbol in
glibc case, and replace all usage of strerror.
George Joseph [Mon, 21 Nov 2016 15:40:59 +0000 (08:40 -0700)]
build: Backport addition of librt check to configure.ac
A while back, a master-only change was made to check for librt which
should probably have been cherry-picked to 13 at that time. Sometime
between then and now, part of that change did make it into 13 but it
was incomplete and non-functional. This patch backports the rest
of the librt check and allows the link of libasteriskpj to use the
results.
George Joseph [Wed, 16 Nov 2016 18:05:43 +0000 (11:05 -0700)]
pjproject_bundled: Improve reliability of pjproject download
The download process now has a timeout which will cause wget to retry
if it stops retrieving data for 5 seconds and fetch and curl to timeout
if the whole retrieval take smore than 30 seconds.
If the tarball retrieval works, the MD5SUM file is retrieved from
the downloads site and the md5 checksum is verified.
If either the tarball retrieval or MD5SUM retrieval fails, or the
checksums don't match, the entire process is retried once. If it
fails again, any incomplete tarball is deleted.
.DELETE_ON_ERROR: was also added to the Makefile. Not only does
this delete the tarball on failure, it till also delete corrupted
library files from the pjproject source directory should they
fail to build correctly.
Tested all the way back to FreeBSD 9, CentOS 6, Debian 6 and
Ubuntu 14.
George Joseph [Thu, 17 Nov 2016 02:24:08 +0000 (19:24 -0700)]
build: Various OpenBSD issues
OpenBSD's 'find' doesn't take the -delete argument so you have to pipe
through 'xargs rm -rf'.
'echo -e' doesn't like \t starting a line. It just prints 't' which
causes the libasteriskpj.exports file to be garbage. They were just
cosmetic so they were removed.
librt doesn't exist so the link of libasteriskpj.so fails. It's not
actually needed for linux anyway so -lrt was removed from the link.
res_rtp_asterisk was failing to load because of an undefined
DTLS_method. '|| defined(LIBRESSL_VERSION_NUMBER)' was added to the #if
so DTLSv1_method is used instead.
Mark Michelson [Wed, 16 Nov 2016 21:42:39 +0000 (15:42 -0600)]
res_format_attr_opus: Fix fmtp generation.
res_format_attr_opus assumed that the string being passed into it was
empty. It tried to determine if the only thing it had written was
a=fmtp:<num>
And if it had, it would reset the string. Its calculation was off when
working with chan_sip, though. chan_sip passes the entire built SDP
rather than an empty string. This resulted in always putting an empty
fmtp line in the SDP.
Richard Mudgett [Tue, 15 Nov 2016 22:23:35 +0000 (16:23 -0600)]
codec_opus: Fix warning when Opus negotiated but codec_opus not loaded.
When Opus is negotiated but not loaded, the log is spammed with messages
because the system does not know how to calculate the number of samples in
a frame.
* Suppress the warning by supplying a function that assumes 20ms of
samples in the frame. For pass through support it doesn't really seem to
matter what number of samples is returned anyway.
Richard Mudgett [Mon, 14 Nov 2016 20:36:52 +0000 (14:36 -0600)]
res_pjsip_outbound_authenticator_digest.c: Fix memory pool leak.
Responding to authentication challenges leaks PJSIP memory pools.
The leak was introduced with a pjproject 2.5.5 API change.
https://trac.pjsip.org/repos/ticket/1929 changed the API usage of
pjsip_auth_clt_init() to require the new API pjsip_auth_clt_deinit() to
clean up cached authentication allocations that get allocated with
pjsip_auth_clt_reinit_req().
George Joseph [Tue, 15 Nov 2016 18:01:04 +0000 (11:01 -0700)]
file.c/__ast_file_read_dirs: Fix issues on filesystems without d_type
One of the code paths in __ast_file_read_dirs will only get executed if
the OS doesn't support dirent->d_type OR if the filesystem the
particular file is on doesn't support it. So, while standard Linux
systems support the field, some filesystems like XFS do not. In this
case, we need to call stat() to determine whether the directory entry
is a file or directory so we append the filename to the supplied
directory path and call stat. We forgot to truncate path back to just
the directory afterwards though so we were passing a complete file name
to the callback in the dir_name parameter instead of just the directory
name.
The logic has been re-written to only create a full_path if we need to
call stat() or if we need to descend into another directory.
Matt Jordan [Mon, 14 Nov 2016 21:57:08 +0000 (15:57 -0600)]
pjproject: Use a much higher limit for PJ_ICE_MAX_CHECKS
The PJ_ICE_MAX_CHECKS constant is used by pjproject to determine how
many pairs of local/remote candidates will be made. If for some reason
we reach this upper bound, ICE will generally fail and no media will
flow between the browser and Asterisk.
This patch makes PJ_ICE_MAX_CHECKS set to the total possible number of
pairs of candidates we'd theoretically allow, which is
PJ_ICE_MAX_CAND^2. Prior to this patch, we simply multiplied
PJ_ICE_MAX_CAND by two; on systems with multiple interfaces (I blame
Docker), this is far too low to allow WebRTC calls to succeed.
Setting this to be PJ_ICE_MAX_CAND^2 allowed WebRTC calls to succeed
even when the system Asterisk was running on had quite a few virtual
interfaces.
Matt Jordan [Mon, 14 Nov 2016 21:32:14 +0000 (15:32 -0600)]
apps/app_echo: Only relay a single video source change frame
In 9785e8d0, app_echo was updated to relay video source updates to the
channel for the purposes of displaying video in WebRTC tests.
Unfortunately, this can cause a Kafkaesque nightmare if two or more
Local channels are in a bridge together where their ends are in
app_echo. When this situation occurs, a video update sent into app_echo
will cause the video update to be relayed to the other Local channels,
causing another round of video updates, etc. In not much time at all,
the channel length queues will be overwhelmed, channel alert pipes will
fail, and all hell will break loose as Asterisk merrily continues to
throw more video update requests onto the channels.
This patch updates app_echo to *only* relay a single video update. Once
a video update has been made, all further video updates are dropped.
This meets the intended purpose of the original patch: if we get a video
update and we're in app_echo, go ahead and ask the sender to update
themselves. However, once we've got that video stream sync'd up, don't
keep spamming the world.
Matt Jordan [Tue, 8 Nov 2016 16:11:41 +0000 (10:11 -0600)]
res/ari/resource_bridges: Add the ability to manipulate the video source
In multi-party bridges, Asterisk currently supports two video modes:
* Follow the talker, in which the speaker with the most energy is shown
to all participants but the speaker, and the speaker sees the
previous video source
* Explicitly set video sources, in which all participants see a locked
video source
Prior to this patch, ARI had no ability to manipulate the video source.
This isn't important for two-party bridges, in which Asterisk merely
relays the video between the participants. However, in a multi-party
bridge, it can be advantageous to allow an external application to
manipulate the video source.
This patch provides two new routes to accomplish this:
(1) setVideoSource: POST /bridges/{bridgeId}/videoSource/{channelId}
Sets a video source to an explicit channel
(2) clearVideoSource: DELETE /bridges/{bridgeId}/videoSource
Removes any explicit video source, and sets the video mode to talk
detection
George Joseph [Mon, 14 Nov 2016 18:16:03 +0000 (11:16 -0700)]
cli: Fix ast_el_read_char to work with libedit >= 3.1
Libedit 3.1 is not build with unicode on as a default and so the
prototype for the el_gets callback changed from expecting a char buffer
to accepting a wchar buffer. If ast_el_read_char isn't changed,
the cli reads garbage from teh terminal.
Added a configure test for (*el_rfunc_t)(EditLine *, wchar_t *) and
updated ast_el_read_char to use the HAVE_ define to detemrine whether
to use char or wchar.
* Don't hold the req_wrapper lock too long in endpt_send_request(). We
could block the PJSIP monitor thread if the timeout timer expires.
sip_get_tpselector_from_endpoint() does a sorcery access that could take
awhile accessing a database. pjsip_endpt_send_request() might take awhile
if selecting a transport.
* Shorten the time that the req_wrapper lock is held in the callback
functions.
George Joseph [Thu, 10 Nov 2016 00:18:00 +0000 (17:18 -0700)]
build: Fix default values for some SANITIZER options
2 of the sanitizers didn't have default values so in systems that
don't support sanitizers menuselect would spit out warnings. They
were harmless but confusing. They've now been set to "0".
Mark Michelson [Tue, 8 Nov 2016 16:48:32 +0000 (10:48 -0600)]
res_pjsip_session: Do not call session supplements when it's too late.
res_pjsip_sesssion was hooking into transaction and invite state
changes. One of the reasons for doing so was due to the
PJSIP_EVENT_TX_MSG event. The idea was that we were hooking into the
message sending process, and so we should call session supplements to
alter the outgoing message.
In reality, this event was meant to indicate that the message either
a) had already been sent, or
b) required a DNS lookup and would be sent when the DNS query
completed.
In case (a), this meant we were altering an already-sent
request/response for no reason. In case (b), this potentially meant we
could be trying to alter a request/response at the same time that the
DNS resolution completed. In this case, it meant we might be stomping on
memory being used by the thread actually sending the message. This
caused potential crashes and memory corruption.
This patch removes the calls to session supplements from the case where
the PJSIP_EVENT_TX_MSG event occurs. In all of these cases, trying to
alter the message at this point is too late, and it can cause nothing
but harm to try to do it. Because there were no longer any calls to the
handle_outgoing() function, it has been removed.
Mark Michelson [Thu, 3 Nov 2016 21:42:40 +0000 (16:42 -0500)]
AGI: Only defer frames when in an interception routine.
AGI recently was modified to defer important frames. This was because
when AGI was used in a connected line interception routine, the
resulting connected line frame would end up getting discarded by the
AGI.
However, this caused bad behavior in other cases. Specifically, during a
transfer, if someone attempted to manually set the Caller ID on a
channel in an AGI, the deferred connected line frame would end up
overwriting what had been manually set in the AGI.
Since the initial issue was specific to interception routines, this
change removes the manual frame deferral from AGI and instead uses the
new frame deferral API in interception routines.
Michael Kuron [Sun, 6 Nov 2016 09:46:30 +0000 (10:46 +0100)]
automon: restore mixing of the both channels after recording stops
This is a regression over Asterisk 11, introduced by 2dc8a060064f359a17f5ebcd515d85fe5203c019. Previously, recordings started via
the automon DTMF code would automatically be mixed together using sox because
app_monitor would be called with the m option. This commit restores this
behavior.
Matt Jordan [Fri, 4 Nov 2016 20:42:09 +0000 (15:42 -0500)]
res_http_websocket: Increase the buffer size for non-LOW_MEMORY systems
Not surprisingly, using Respoke (and possibly other systems) it is
possible to blow past the 16k limit for a WebSocket packet size. This
patch bumps it up to 32k, which, at least for Respoke, is sufficient.
For now.
Because 32k is laughable on a LOW_MEMORY system (as is 16k, for that
matter), this patch adds a LOW_MEMORY directive that sets the buffer to
8k for systems who have asked for their reduced memory availability to
be considered.
Matt Jordan [Fri, 4 Nov 2016 20:40:58 +0000 (15:40 -0500)]
res_stasis: Set a video source mode on Stasis created bridges
When a bridge is created via ARI (through res_stasis), no video source
mode is set by default. As a result, any endpoint sending video media
won't ever see any video reflected back to it.
This patch defaults a bridge to a 'follow the talker' video mode.
Further work can be done to add routes that allow for the video mode to
be controlled through the /bridges resource.
Matt Jordan [Fri, 4 Nov 2016 20:37:57 +0000 (15:37 -0500)]
main/bridge_channel: Fix channel reference leak on video source
When a channel is made the video source, the bridge holds a reference to
it. Whenever the video source changes, that reference is released.
However, a ref leak does occur if the channel leaves the bridge (such as
being hung up) while it is the video source, as the bridge never
releases the ref in such a case.
This patch adds a line to the bridge_channel_internal_join routine such
that, when a channel finishes its time in the bridge, it notifies the
bridge via ast_bridge_remove_video_src that if it is a video source its
reference should be released.
Matt Jordan [Fri, 4 Nov 2016 20:36:42 +0000 (15:36 -0500)]
main/bridge: Add some verbose logging for video source changes
It's actually quite useful to see the source of a video stream change.
This doesn't happen terribly often, even with talk detection - but when
it does, it's nice to know which channel is now providing your video
stream.
As a verbose 5 level message, it shouldn't be terribly spammy or costly
to have, and is 'lower level' then most other verbose messages that the
bridge system emits.
Matt Jordan [Fri, 4 Nov 2016 20:33:35 +0000 (15:33 -0500)]
bridges/bridge_softmix: Remove SSRC changes on join/leave; update video source
WebRTC clients really, really want to know the SSRC of the media they're
getting. Changing the SSRC is generally not a good thing.
bridge_softmix, starting in Asterisk 12, started changing the SSRC of
parties as they joined or left the bridge. With most phones, this isn't
a problem: phones just play back the stream they're getting. With WebRTC
clients, however, the SSRC is tied to a media stream that may be
negotiated. When a new SSRC just shows up, the media can be dropped.
As it turns out, the SSRC change shouldn't even be necessary. From the
perspective of the client, it's still talking to Asterisk with the same
media stream: why indicate that the far party has suddenly changed to a
different source of media?
This patch opts to just remove the SSRC changes. With this patch, video
clients that join/leave a softmix bridge actually get the video stream
instead of freaking out.
Kevin Harwell [Fri, 28 Oct 2016 20:11:35 +0000 (15:11 -0500)]
stasis_recording/stored: remove calls to deprecated readdir_r function.
The readdir_r function has been deprecated and should no longer be used. This
patch removes the readdir_r dependency (replaced it with readdir) and also moves
the directory search code to a more centralized spot (file.c)
Also removed a strict dependency on the dirent structure's d_type field as it
is not portable. The code now checks to see if the value is available. If so,
it tries to use it, but defaults back to using the stats function if necessary.
Lastly, for most implementations of readdir it *should* be thread-safe to make
concurrent calls to it as long as different directory streams are specified.
glibc falls into this category. However, since it is possible that there exist
some implementations that are not safe, locking has been added for those other
than glibc.
Joshua Colp [Wed, 2 Nov 2016 15:52:13 +0000 (15:52 +0000)]
res_stasis: Don't unsubscribe from a NULL bridge.
A NULL bridge has special meaning in res_stasis for
unsubscribing. It means that a subscription to ALL
bridges should be removed. This should not be done
as part of the normal subscription management in
the res_stasis channel loop.
projects / thirdparty / asterisk.git / log
