Mohamed Chibani authored 4 years ago and Sébastien Blin committed 4 years ago

SipVoipLink may never be intialized in Manager instance (Manager::init())
leading to a crash when the Manager calls finish(). Added a check
to prevent the crash.

Gitlab: #481

Change-Id: I817a7795ce4341e6214a44e46037b0183fde53c5

Pierre Lespagnol authored 4 years ago and Sébastien Blin committed 4 years ago

Gitlab: #479
Change-Id: I19c6286311936d61c8a850f99b4ed3008b675b48

Mohamed Chibani authored 4 years ago and Sébastien Blin committed 4 years ago

This patch adds many improvements and bug fixes:
1- For PUPNP lib, use the synchronous methods for add and remove
requests. The async versions have many issues mainly at shutdown.
If the callbacks do not arrive on time, the memory allocated by
the lib when performing the request will leak.
2- Rework the init and IGD search of PUPNP.
3- Move the termination step (for libupnp in particular) from
the destructor to a newly added method (termiinate()). This will
prevent a race at shutdown between libupnp threads (callbacks) and
the daemon shutdown process.
4- Do not try to initialize libupnp if the host address is invalid.
This will prevent putting the lib in an unstable state.

Gitlab: #472

Change-Id: I14e5ddcd2b735da374a1f09bcef40f6a86a38133

GitLab: #404
Change-Id: I42792e27fb2bdf21870477ccb821f672033fc25d

Instead, use Bash from PATH.  This approach works universally, even in
environments not strictly adhering to the File Hierarchy
Standard (FHS).

Change-Id: I852c5f4bba3a1bee9b895dcb5da24e5738c4186a

The DirectSound host api implementation in PortAudio has multiple
issues:
- requires utf8 support patch
- can't enumerate the default communication devices
- reports incorrect default sample-rates for devices

The WASAPI host api implementation needs to be patched to access
the default comm devices, but functions better out-of-the-box.

Change-Id: Ie27ee3e1418dd6d734ad6ad7685f2dc44db86007
Gitlab: #463

Andreas Traczyk authored 4 years ago and Adrien Béraud committed 4 years ago

- Enables high-pass filter, AEC, AGC, NS.
- Disables system AEC for Windows

Gitlab: #464
Change-Id: Ife6261a815395263abeca482bb78ad2c90133db3

Allows the start/stop of individual streams as well as both input
and output simultaneously. When both streams are to be started,
full duplex mode is attempted first.

Resets device indices to -1 in the case that the host api has
changed, or if the dring audio config is scrapped somehow.

Change-Id: I1655a34d2111222b6add19f1b36b53bc4a2838ec

Andreas Traczyk authored 4 years ago and Adrien Béraud committed 4 years ago

Hide speexdsp's echo cancellation implementation details behind an
EchoCanceller derived class. An AudioLayer may now instantiate
implementations of the EchoCanceller. This may be platform specific
compile-time or swapped out at runtime.

Introduces a null echo canceller for testing. As the speexdsp aec
is not currently functional, the null echo canceller is replaced
as the default implementation.

GitLab: #454
Change-Id: I169f1e9758afbed884fc42d9d78a69ce28d12fe2

Andreas Traczyk authored 4 years ago and Adrien Béraud committed 4 years ago

Change-Id: Ib7396b12f98947b25bc96605381cbc29594aa1b7

Change-Id: I38ffcd79e3392b1eb8a763914fa182e1f17a2685

Regression was caused by 37125acf where the mute button didn't
work anymore.

Change-Id: I7e3d0953e65848818c9dc0e348759dce63274337

Apply patches for CVE-2021-21375 and CVE-2020-15260 affecting
prproject. After a quick look I don't think those issues apply to
us, as GnuTLS manages the TLS part and we do resolution outside
pjsip (the daemon passes ips), but at least our pjproject will
not be affected.

Note, these patches will be removed when we bump pjproject to 2.11.

Change-Id: Ie9029f5501096c7148416cbb803a40f1c053a3d0

Change-Id: Ia9cf5638678c0e92127ec9bf3631235a2827d49d

Pierre Lespagnol authored 4 years ago and Adrien Béraud committed 4 years ago

During the detach, the audio input need to be unbinded from
subcalls composing the conference, not from all calls.
For the removeConference, the leaving call is unbinded in
removeAudio and the remaining call is not unbinded (no need
to unbind all and re-bind remaining call)

Gitlab: #414

Change-Id: I900acac1293f0f8c6e13a74309167d7356015089

Aline Gondim Santos authored 4 years ago and Sébastien Blin committed 4 years ago

GitLab: #422
Change-Id: Id94eb245d9675d305a0d05da76e62bcd245525ec

- adds a missing space in the configure command line
- removes some python3 warnings

Change-Id: I79458a408a6f3198cb329072abf530ed8f23986a

Change-Id: Iff69c5888acba0a9a165bce4f4106a217dc78134

This allows the rendez-vous to be considered as a
conference and not as participant for the other
conference.

Change-Id: I9053533a605f1928da93ed3d3e9235db8b764591

Audio sender doesn't need audioInput, audio rtp session already has one.
(Consistency with video rtp session)
muteState is not used in audio sender.

Change-Id: I7c798c94050fa6533ff9b726914f2f880e426a52

Pierre Lespagnol authored 4 years ago and Sébastien Blin committed 4 years ago

Gitlab: #453

Change-Id: I9a870b2f930c69fe26432f1f15169d13365e702d

Sébastien Blin authored 4 years ago and Adrien Béraud committed 4 years ago

On some very slow network, the process to connect to a relay
candidates can take a lot of time. Cause this timer includes
the time between the ConnectionAttempt and sending the
Binding response. With some packet loss it can quickly increase
this delay.

Example:
Classic nego (Successful ICE):

10:36:28.126 .RX 144 bytes STUN message from 51.222.10.40:3478:
--- begin STUN message ---
STUN ConnectionAttempt indication (...)

10:36:29.378 .TX 84 bytes STUN message to IP:45320:
--- begin STUN message ---
STUN Binding success response

Problematic case on a bad network:

// Bad network peer
09:39:10.910 .RX 288 bytes STUN message from 192.252.140.236:3478:
--- begin STUN message ---
STUN ConnectionAttempt indication

// other PEER:
09:38:51.855 192.252.140.236:11846 (nominated): connectivity check FAILED:
STUN transaction has timed out (PJNATH_ESTUNTIMEDOUT)
=> The other peer close the socket

// Bad network peer
09:38:52.258 .RX 76 bytes STUN message from IP:36436:
--- begin STUN message ---
STUN ConnectionBind success response

9:38:53.756   tcprel0x7f7098178460  .socket send(): Not found (PJ_ENOTFOUND)

The value of 15 secs is chosen to follow ice_inactive_timeout in
the webrtc's sources (CONNECTION_WRITE_TIMEOUT)

Change-Id: Idc66f6cb2018fbf03e34059640fa91601535dcb1
GitLab: #1201

Pierre Lespagnol authored 4 years ago and Adrien Béraud committed 4 years ago

Gitlab: #409

Change-Id: Icaf85305227bc5fdbfc2a23966c1080b7e05f0fc

Pierre Lespagnol authored 4 years ago and Adrien Béraud committed 4 years ago

Gitlab: #447

Change-Id: I249fb7034a12f117298187b7b42977760be3ef83

Rework shutdown of the UPnPContext to cleanly remove allocated
mappings. Also apply many code improvements and clean-up.
This will also prevent some crashes seen at shutdown.

Change-Id: I5cdc56df9d096d443ccfb2619edbec35bc00a6bc
Gitlab: #436

Change-Id: Ia95bd71499b502f4e6d3df690537a6e1c91422d0

Gitlab: #468

Change-Id: I2c94823bb8efab6fccddd59de1ae96f290745691

Adrien Béraud authored 4 years ago and Sébastien Blin committed 4 years ago

Change-Id: I61ff878c79b4116e28bcdce495367c73029af1e9

Change-Id: I8e98b66bf23a34cb1a4b3d360d8c1027e064c7e5

Sébastien Blin authored 4 years ago and Adrien Béraud committed 4 years ago

getting a TlsSession::shutdown() while doing a gnutls_handshake
provoke a loop (cause the handshake will return GNUTLS_E_INTERRUPTED
and will loop until HANDSHAKE_MAX_RETRY).

Change-Id: Ifabc24c815552a14781b72cb9baf66d987cb25ce

Gitlab: #456
Change-Id: Ic44c9ce0b682bf8352b15ad9eaf8d51a15c4c657

Change-Id: I139b6fcb70986be033c88bbf7e0545af326ff407

Change-Id: I49383922aca110594410b7d86aafde2ece6af4b7

Change-Id: I92a9a61c8d34736af99693ec826dd2385c319bfa

Change-Id: I5efa45a80f05c6ba4106e05234612c6ea5d6fdf9

Maxim Cournoyer authored 4 years ago and Adrien Béraud committed 4 years ago

This brings wget performance up to par with that of curl, which makes
downloading the contrib sources via wget about a ~6X faster.  Wget is
now also made the preferred downloader, as it's a smaller than curl,
and is GNU.

* contrib/src/main.mak (download): Remove the -p option, not suitable
for single file downloads (it's meant for retrieving all the resources
of a web page).  This fixes a warning too.  Remove the --passive
option, which is not documented.  It's probably equivalent to
--passive-ftp, which is the default anyway.  Use --waitretry instead
of -w.  The later was causing the 10 s delay to be inserted on every
HTTP redirection, wasting time.

Change-Id: Ibcd6805c66da3fb0c3f37b7128a34cf6c269f400

Maxim Cournoyer authored 4 years ago and Adrien Béraud committed 4 years ago

It should only worry about its parent directory, not its grandparent.
This makes it possible to bind mount just daemon/contrib in a
container, to run 'make fetch' for example.

* contrib/bootstrap: Adjust relative paths.
* contrib/src/main.mak: Express using the CURDIR variable.
(PATH): Express using the TOPSRC variable.
(pkg_static): Express using the SRC environment variable.

Change-Id: I02cba2249e4971d76ef0af0fa00f4b1af61c5aff

Maxim Cournoyer authored 4 years ago and Adrien Béraud committed 4 years ago

The 'fetch' target relies on the prerequisites of the .sum-$(PKG)
files.  Prior to this change, 'make fetch' would not fetch the
openbsd-libressl tarball.

Change-Id: Ic91cc7032243d0c0f3b3d69fb74cfb3c633d0fc3

Maxim Cournoyer authored 4 years ago and Adrien Béraud committed 4 years ago

This target can be used to print a list of the contrib tarballs
required to build the daemon.  It can be used to copy only the
required tarballs from a cache when building a source release, for
example.

Change-Id: I402c29989b3e6fed8a82418d19dbbacea8c44f50

Maxim Cournoyer authored 4 years ago and Adrien Béraud committed 4 years ago

It was previously possible by overriding the TARBALLS variable on each
make call, but that's inconvenient.  This changes allows specifying it
once, at bootstrap time.  If the TARBALLS environment is defined, it
is also honored (and takes precedence over the value provided at
bootstrap time).

* contrib/src/ffmpeg/rules.mak (ffmpeg): Because the $(TARBALLS)
variable is now made absolute (for clarity in the build output), do
not use a relative path.
* contrib/src/onnx/rules.mak (onnx): Likewise.
* contrib/src/vpx/rules.mak (libvpx): Likewise.
* contrib/src/x264/rules.mak: Likewise.

Change-Id: I07497b095938c7885159f44753ead2814ed2a2fe