Sébastien Blin authored 3 years ago and Adrien Béraud committed 3 years ago

verifyOcsp takes a callback with arguments passed by reference.
However, if the method finishes (verifyCertificateWrapper) cert
will be destroyed and references potentially invalidated.
So, instead of waiting on a condition variable use a future to
be sure that the cert is used before destruction.
The callback is always called by sendOcspRequest with a similar
timeout so also remove the useless timeout.

Change-Id: If1a8eba70e13d2d613b758ad24629efa4fe57bb3
GitLab: #577

Dolt doesn't provide the performance benefits it promises, so drop it
to simplify the moving parts of the Autotools build system.

Also remove unused M4 modules.

* m4/dolt.m4: Delete file.
* m4/ax_boost_base.m4: Likewise.
* configure.ac: Do not call DOLT.

Change-Id: I58d286457005ad82129e089679c52933989c59a8

Adrien Béraud authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: Ia395103ba2ec2ec85400cf710a07c40888568d68

For whatever reason, a conversation can be malformed. This should
be detected and the conversation not loaded. The conversation
will be removed and re-synced if necessary.

Change-Id: I40e78564559e7f09159ceba4d263a4ea75b25794

Maxim Cournoyer authored 3 years ago and Maxim Cournoyer committed 3 years ago

Automated using the following commands:

  $ mv src/{dring,jami}
  $ git grep -l src/dring | xargs sed -i 's,src/dring,src/jami,g'
  $ git grep -l '#include "dring/' | \
    xargs sed -i 's,#include "dring/,#include "jami/,g'
  $ git grep -l 'dring.h' | xargs sed -i 's,dring.h,jami.h,g'

And finally,

  $ git grep -l 'dring' | xargs sed -i 's,dring,jami,g'

  $ files=$(find -name '*dring*' | sort)
  $ for f in $files; do mkdir -p "$(dirname "$f")"; \
      mv "$f" "$(echo $f | sed 's/dring/jami/g')"; done

To resolve a bad renaming favorably:

  $ git grep -l -i AlsaCarjami | \
    xargs sed -i -E 's/([Aa])lsaCarjami/\1lsaCardRingtone/g'

The above renaming command is not perfect, so some hand-tuning was
required to complete it.

* src/manager.cpp (Manager::ManagerPimpl::retrieveConfigPath):
Preserve the dring.yml configuration file name, until we add something
to migrate (rename) it to jami.yml.
* man/dring.pod: Delete.
* bin/dbus/jamid.pod: Move to ...
* man/jamid.pod: here.
* bin/dbus/meson.build (jamid_targets): Normalize man section to the
pre-existing 1 and adjust accordingly.
* src/jami/def.h (dring_EXPORTS): Rename to ...
(jami_EXPORTS): ... this.

change-Id: I9828be6da9c711ab2f22c4d1b9539fea89d7b6fb

This logic was added via 9d9a6b3d,
but this will give potential deadlocks, because pj's group lock
will be locked at this point. So we will have:
group-lock->callMutex_
and other methods like SIPCall::hangup will try to lock:
callMutex_->group-lock
causing a deadlock.

The original issue was adding this line due to a use after free,
however since then we fixed the ref counter of inv, so it should
be a valid fix.

Change-Id: I2fdc236ca63487634ebf2659ecdebb8e0c3234fe
GitLab: #586

Change-Id: I33c732784833808d092cefa402669a0bbdc0561e
GitLab: #582

Gitlab: #583

Change-Id: Iddd1b582595daee498b383a11c6323546d14bef4

If `monitorLog` is set to true in between the copy of variadic arguments and
branch to `vnsprintf`, it results in SEGFAULT.

Change-Id: I610b950cec22e9b7123cca709e2f145b1bc1be16

Gitlab: #570

Change-Id: I87f476082f0f04b342e737e523eae976b0717bca

Olivier Dion authored 3 years ago and Sébastien Blin committed 3 years ago

Increment of the invite session has to be done before SIPCall takes a reference
to it.

Otherwise, `pjsip_inv_dec_ref()` will be called internally by PJSIP when the
reference counter has reached 0.  This happen when `SIPCall::setInviteSession()`
is called in `invite_session_state_changed_cb()`, resulting in the deletion of
the invitation.

Change-Id: Icba0985ea26ce5dec462bd124cca4dd2187d6a47

Change-Id: I0f0a926996d34ae22051f0247e36575ab755d0ba

Rework the test to use two audio streams instead of audio+video.
This is needed to run the test on environments where the camera
is not available (e.g. Docker). For this specific test, the media
type is irrelevant, so using two audio media is acceptable.

Gitlab: #516

Change-Id: Idc4911d5aea5ffdd931bd7d63516ed484344d41c

Olivier Dion authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I1acac4dd8e87d2ffe1c63638d03fd28c6749adef

Change-Id: I702ab3422c3fd6bd528fe65a5541c98150a5722d

Change-Id: I68e954bc8de9548a9da0026c014f27141e0b7ec0

Change-Id: Ice706b3c5384cb8bec7becfc6209658a78b9ac9a

Olivier Dion authored 3 years ago and Sébastien Blin committed 3 years ago

`last_REMB_*` can be at `time_point::min()` when `delayMonitor()` is called.  This
results in a signed integer overflow, which UBASAN doesn't like.

Fix this by setting `last_REMB_*` to `clock::now()` before setting callback.

Change-Id: I0b24d759927c2eb7e23253a16333ba54dc3f9edf

Aline Gondim Santos authored 3 years ago and Kateryna Kostiuk committed 3 years ago

Change-Id: I13b63eff7f7b729ece6ac3d16cd4d36767a20464

Change-Id: Ib943587e49bcbf16104cf3bbc206e30bc3448f82

Mohamed Chibani authored 3 years ago and Sébastien Blin committed 3 years ago

Currently, when an empty offer is received, a media offer is
generated with audio, and video if enabled in the account.
Now, the generated offer will include only audio. Both
participants can add video to the call if they wish.
Also some code clean-up

Gitlab: #556

Change-Id: Id51bf44b2b84fd9341de4b67bf53963614088ef8

Olivier Dion authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I67825b04f1fead3d4267564e4c366cb54037c593

This seems to avoid a heap buffer overflow. Anyway, we only add data at the
end of the buffer so the deque has no need.

Change-Id: I8f3932599a065f22b11adc701ededfb5ebedfff8

Change-Id: I62f93b4b0706863debc9595b0b4646c3b0974b52

This can re-generate MessageReceived for already received messages. If we
are not behind from, this means we are up-to-date and no-need to re-announce.

Change-Id: I890ec20117235aeec56e4a7d3f720c1193e9633c

avoid any possible race condition. shutdown the sock when onShutdown is set

Change-Id: I0e7a5e940267154aec4417c8b653d19b1f54430c

Adrien Béraud authored 3 years ago and Sébastien Blin committed 3 years ago

If we receive EOF, didn't the peer already closed the socket ?

Change-Id: I06ae908b49ddd4ba0a5eb43ff25f51da1f15cc21

Adrien Béraud authored 3 years ago and Sébastien Blin committed 3 years ago

Change-Id: I34eeff81cb639bab901e939dd6436f6645c32b4d

Adrien Béraud authored 3 years ago and Sébastien Blin committed 3 years ago

Refactor structures to improve memory consistency
and reduce concurrency on the main mutex.

Change-Id: I3dcf1b94d96d51bddff4446b7011605821c4f1b6

This is a follow up commit to
d5021c78.

* bin/dbus/cx.ring.Ring.service.in (Exec): Replace @libexec@ by
@libexecdir@.

Change-Id: Id72503d45f24bf1ff4341d025b9c7af51cb52479

Windows fixup post https://review.jami.net/c/jami-client-qt/+/18383

Change-Id: Id3b4b11ad49f09bfc0be22155145080b661aceeb

setOnRecv must lock socketsMutex. Generally it,s the case because onAccept
will be ready right away and still locked, however if onChannelReady_ retrigger
the callback, setOnRecv will be used without socketsMutex locked.

Retrigger in a thread and always lock socketsMutex to be sure.

Change-Id: If44d240e113585a5ce3320b8f17feaaaa8be8515

When starting an outgoing call, if the ICE initialization is too fast:
+ SIPCall::initIceMediaTransport will init the ICE transport
+ onInitDone will be called before tmpMediaTransport_ initialized
+ JamiAccount::SIPStartCall will be called and add ICE candidates
BUT because transportMtx_ is not locked, tmpMediaTransport_ will be null
and no ICE transport will be added.

This is why createUTransport should be done while locked. Anyway the method is
async so should not block.

GitLab: #574
Change-Id: Id88dc92f6d77c18de3ea4614e2b5fd79d3429ff0

* src/meson.build (libjami_dep): Sort headers.
* src/Makefile.am (nobase_include_HEADERS): Sort and add
'jami/conversation_interface.h'.

Change-Id: Ib66e4c9904906dcca4722cc81b6487aa43086e2e

Maxim Cournoyer authored 3 years ago and Sébastien Blin committed 3 years ago

This avoids the error:

   ld: /home/mcournoyer/src/ring-project/daemon/contrib/x86_64-unknown-linux-gnu/lib/libswscale.a(swscale.o): relocation R_X86_64_PC32 against symbol `ff_M24A' can not be used when making a shared object; recompile with -fPIC
   ld: final link failed: bad value

caused by the missing '-Wl,-Bsymbolic' linker directive (which is
documented as required by ffmpeg when a shared library is built
against a static ffmpeg object).

* src/Makefile.am (libring_la_LDFLAGS): Set to AM_LDFLAGS.
[HAVE_WIN32]: Append using the += operator.
[HAVE_OSX]: Likewise.
[HAVE_LINUX]: Remove; it was not necessary.

Change-Id: Id2af9ea5812b2d1fd3b4f7f893108bb94671ffc6
GitLab: jami-packaging#101

Ming Rui Zhang authored 3 years ago and Adrien Béraud committed 3 years ago

Gitlab: #567

Change-Id: I9fdb1601f407759a8b6ef73bccc32769e5693b5b

Maxim Cournoyer authored 3 years ago and Sébastien Blin committed 3 years ago

This is a follow up to commit 2afa99fd (Install dring to libexec and
edit templates at install time).

* bin/dbus/Makefile.am (do_subst): Replace bindir by libexecdir.

Change-Id: I380e61467ef4f3c4873698c9e2061b9c61370d44

Olivier Dion authored 3 years ago and Adrien Béraud committed 3 years ago

Change-Id: Ia9ce297062ac28dcf7f03ee28247cad4411d1b0f

Olivier Dion authored 3 years ago and Adrien Béraud committed 3 years ago

Change-Id: I92e57a1f62e3bae67c8ae4ca6d7654dfe427c361

Olivier Dion authored 3 years ago and Adrien Béraud committed 3 years ago

Change-Id: I31128ff65ca03c62971165ed61683cbc84b0b5d4