Bug 1568101/D40532 captures the input timestamp and uses it in the resolve
function. However, there could be multiple encoded frames returned (e.g.,
for Drain()), and all of them will get the same captured timestamp. Read
back the timestamp from encoded data makes sure every frame gets the right
value. Also convert it for EncodedImage::_timeStamp uses different unit.
Differential Revision: https://phabricator.services.mozilla.com/D45774
--HG--
extra : moz-landing-system : lando
Freeing the conduits when the transceiver is stopped releases all of the
associated webrtc.org objects which results in substantial memory savings.
On my system, with an opt+debug build and 200 stopped transceivers, I see
373.17 MB of memory use in the content process without this patch, and
158.93 MB of memory use with this patch applied.
Going further and calling Shutdown_m as part of Stop() reduces the memory
use to 157.98 MB, which seems like a marginal improvement at the cost of
a much larger risk of introducing bugs.
Differential Revision: https://phabricator.services.mozilla.com/D45845
--HG--
extra : moz-landing-system : lando
With NSS now being used to do the decryption in the Clearkey CDM, remove the
source and licensing information related to OpenAES.
Differential Revision: https://phabricator.services.mozilla.com/D45205
--HG--
extra : moz-landing-system : lando
I just copied all the files from the upstream repository into the
rsdparsa directory
Differential Revision: https://phabricator.services.mozilla.com/D45717
--HG--
extra : moz-landing-system : lando
Having a full VPATH for the srcdir sometimes causes make to grab the
wrong prerequisite for a rule, in particular if we have a file in the
srcdir and also generate a file of the same name in the objdir. We don't
really need VPATH anymore though, since most of the information comes
from mozbuild, where we can explicitly list the path to the srcdir or
objdir as necessary.
Differential Revision: https://phabricator.services.mozilla.com/D42968
--HG--
extra : moz-landing-system : lando
Clearkey previously relied on OpenAES to do its encryption. In order to
facilitate future changes and the need for CBC support, switch to NSS, which
should be more flexible and actively maintained.
Differential Revision: https://phabricator.services.mozilla.com/D41993
--HG--
extra : moz-landing-system : lando
This adds a mObfuscateHostAddresses member and uses it to control whether or
not peer reflex candidates are redacted. It is only necessary to keep track of
signaled addresses to support redacting peer reflex candidates, so there is no
need to store them if host address obfuscation is disabled.
Differential Revision: https://phabricator.services.mozilla.com/D44531
--HG--
extra : moz-landing-system : lando
To accomodate this, MediaStreamTrackListener::NotifyEnded/Removed get an
additional aGraph argument. NotifyEnabledStateChanged is getting it too, for
completion.
Differential Revision: https://phabricator.services.mozilla.com/D43667
--HG--
extra : moz-landing-system : lando
The Rust get_if_addrs library previously used does not build on Android with
our build system. Since we're already using nICEr to determine the local
interface addresses, rather than fix the Rust library, we can use those
addresses to set the interface on which mdns_service listens.
Differential Revision: https://phabricator.services.mozilla.com/D42760
--HG--
extra : moz-landing-system : lando
Checking this assertion outside of the if statement can result in
a use-after-free in debug builds.
Differential Revision: https://phabricator.services.mozilla.com/D42152
--HG--
extra : moz-landing-system : lando
The current code causes one mDNS service to be created for each PeerConnection.
Due to Bug 1569311, the services persist until shutdown, which can lead to a
lot of mDNS threads running on sites which use WebRTC for fingerprinting. This
change makes it so we start at most one mDNS service.
I've filed Bug 1569955 to look at shutting down the mDNS service after the
last hostname is unregistered. As an alternative, if we fix Bug 1569311, we
could also use refcounting and stop the mDNS service after the last
StunAddrsRequestParent is freed.
Differential Revision: https://phabricator.services.mozilla.com/D42151
--HG--
extra : moz-landing-system : lando
With the move to the socket process, the UUID service is no longer available
in nricectx. This adds a pair of helper functions to mdns_service to generate
UUIDs and uses them to generate hostnames inside nricectx.
Differential Revision: https://phabricator.services.mozilla.com/D42150
--HG--
extra : moz-landing-system : lando
This only enables mDNS on OS X for now. Some versions of Windows lack mDNS
support, there are some oddities with resolving IPv6 addresses on Linux, and
Android has not yet been tested. All of these will be addressed in follow on
bugs.
Differential Revision: https://phabricator.services.mozilla.com/D38496
--HG--
extra : moz-landing-system : lando
This adds a mdns_service to mtransport to handle responding to mDNS queries.
All hostnames will be generated from UUIDs, so the responder assumes that it
is the only responder for a hostname which is registered with it. Because of
this, the responder does not first make a DNS query itself to see if any other
responder is handling a hostname, and does not wait a random amount of time
before replying, both of which are required by the specification to avoid
collisions with other responders.
Differential Revision: https://phabricator.services.mozilla.com/D38489
--HG--
extra : moz-landing-system : lando
The Rust get_if_addrs library previously used does not build on Android with
our build system. Since we're already using nICEr to determine the local
interface addresses, rather than fix the Rust library, we can use those
addresses to set the interface on which mdns_service listens.
Differential Revision: https://phabricator.services.mozilla.com/D42760
--HG--
extra : moz-landing-system : lando
Checking this assertion outside of the if statement can result in
a use-after-free in debug builds.
Differential Revision: https://phabricator.services.mozilla.com/D42152
--HG--
extra : moz-landing-system : lando
The current code causes one mDNS service to be created for each PeerConnection.
Due to Bug 1569311, the services persist until shutdown, which can lead to a
lot of mDNS threads running on sites which use WebRTC for fingerprinting. This
change makes it so we start at most one mDNS service.
I've filed Bug 1569955 to look at shutting down the mDNS service after the
last hostname is unregistered. As an alternative, if we fix Bug 1569311, we
could also use refcounting and stop the mDNS service after the last
StunAddrsRequestParent is freed.
Differential Revision: https://phabricator.services.mozilla.com/D42151
--HG--
extra : moz-landing-system : lando
With the move to the socket process, the UUID service is no longer available
in nricectx. This adds a pair of helper functions to mdns_service to generate
UUIDs and uses them to generate hostnames inside nricectx.
Differential Revision: https://phabricator.services.mozilla.com/D42150
--HG--
extra : moz-landing-system : lando
This only enables mDNS on OS X for now. Some versions of Windows lack mDNS
support, there are some oddities with resolving IPv6 addresses on Linux, and
Android has not yet been tested. All of these will be addressed in follow on
bugs.
Differential Revision: https://phabricator.services.mozilla.com/D38496
--HG--
extra : moz-landing-system : lando
This adds a mdns_service to mtransport to handle responding to mDNS queries.
All hostnames will be generated from UUIDs, so the responder assumes that it
is the only responder for a hostname which is registered with it. Because of
this, the responder does not first make a DNS query itself to see if any other
responder is handling a hostname, and does not wait a random amount of time
before replying, both of which are required by the specification to avoid
collisions with other responders.
Differential Revision: https://phabricator.services.mozilla.com/D38489
--HG--
extra : moz-landing-system : lando
This changes the hard-limit of `RLIMIT_RTTIME` to be the maximum available
(200ms on my system), and keep the soft limit to the same number.
Having different numbers allow catching SIGXCPU before getting SIGKILL.
Differential Revision: https://phabricator.services.mozilla.com/D43402
--HG--
extra : moz-landing-system : lando
In InitDecode it is possible for mDecoder to be null if the CreateDecoder call
fails, in which case an error code is returned. Release is called by
GenericDecoder::~VCMGenericDecoder regardless of whether InitDecode succeeds, so
we need to check if mDecoder is null in Release.
Differential Revision: https://phabricator.services.mozilla.com/D43469
--HG--
extra : moz-landing-system : lando
This changes the hard-limit of `RLIMIT_RTTIME` to be the maximum available
(200ms on my system), and keep the soft limit to the same number.
Having different numbers allow catching SIGXCPU before getting SIGKILL.
Differential Revision: https://phabricator.services.mozilla.com/D43402
--HG--
extra : moz-landing-system : lando
In OnIncomingFrame() we check if the capture has started by calling
CaptureStarted() which holds a lock. If StopCapture() is called on the
control thread after we check CaptureStarted() it is possible that
_captureCapability will be zeroed out prior to frame being delivered because
the lock has been released. This would result in an unknown video type in
CalcBufferSize() and trigger an assertion there. This creates a local copy of
_captureCapability with the lock held to ensure it is not zeroed out when the
frame is delivered.
Differential Revision: https://phabricator.services.mozilla.com/D42338
--HG--
extra : moz-landing-system : lando
John Lin
Byron Campen [:bwc]
Dan Minor
Ryan Alderete
Bastien Orivel
Boris Zbarsky
Mike Shal
Mihai Alexandru Michis
Ryan VanderMeulen
Christian Holler
Andreas Pehrson
Eric Rahm
Cameron Kaiser
Gurzau Raul
Kershaw Chang
Sylvestre Ledru
Paul Adenot
Coroiu Cristina