While testing the [implementation for insertable streams] in `webrtcsink` &
`webrtcsrc`, I encountered critical warnings, which turned out to result from
two race conditions in `rtpsession`. Both race conditions produce:
> GLib-CRITICAL: g_hash_table_foreach:
> assertion 'version == hash_table->version' failed
This commit fixes one of the race conditions observed.
In its simplest form, the test consists in 2 pipelines and a Signalling server:
* pipelines_sink: audiotestsrc ! webrtcsink
* pipelines_src: webrtcsrc ! appsrc
1. Set `pipelines_sink` to `Playing`.
2. The Signalling server delivers the `producer_id`.
3. Initialize `pipelines_src` to establish a session with `producer_id`.
4. Set `pipelines_src` to `Playing`.
5. Wait for a buffer to be received by the `appsrc`.
6. Set `pipelines_src` to `Null`.
7. Set `pipelines_sink` to `Null`.
The race condition happens in the following sequence:
* `webrtcsink` runs a task to periodically retrieve statistics from `webrtcbin`.
This transitively ends up executing `rtp_session_create_stats`.
* `pipelines_sink` is set to `Null`.
* In `Paused` to `Ready`, `gst_rtp_session_change_state()` calls
`rtp_session_reset()`.
* The assertion failure occurs when `rtp_session_reset` is called while
`rtp_session_create_stats` is executing.
This is because `rtp_session_create_stats` acquires the lock on `session` prior
to calling `g_hash_table_foreach`, but `rtp_session_reset` doesn't acquire the
lock before calling `g_hash_table_remove_all`.
Acquiring the lock in `rtp_session_reset` fixes the issue.
[implementing insertable streams support]: https://gitlab.freedesktop.org/gstreamer/gst-plugins-rs/-/merge_requests/1176
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4528>
Unfortunately streamoff does not flush the events, and this can cause all
sort of issues. Flush events on capture queue. We also return
GST_V4L2_FLOW_RESOLUTION_CHANGE in case a resolution change was seen.
This allow skipping streamon(capture) on flush, which could lead to a
configuration miss-match, or failure if the buffers aren't of the right
size.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4437>
Let the driver detects the change and reconfigure the capture side
transparently from there. This avoid reallocation of the output buffers,
and eliminates the need to stop and restart the capture task. This is
only happening if the driver have support for this, otherwise the old
behaviour is maintained.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4437>
Stop doing capture buffer allocation based on guesses
and wait for the source change event when available.
Unlike stateless decoder, the stateful decoder is not aware of
the coded resolution, and this may lead to the wrong result
even when using TRY_FMT.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4437>
In previous implementation that job was split between handle_frame and
the processing loop and it wasn't clear if this mechanism was race
free. The capture setup would also be tried for every buffer, which was
not necessary.
This also simplify the handling of SRC_CH event, dropping the unneeded
atomic boolean.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4437>
When seek flush, gst v4l2 buffer pool flush is not atomic which will
lead double enqueue buffer (qbuf) issue, and v4l2 buffer pool qbuf is
also not atomic which will lead no free buffer found in the pool.
1. add lock for calculate enqueue number in streamon function
2. add lock for v4l2 capture end streamoff in pool flush function
3. lock the whole funciton of v4l2 buffer pool qbuf, then the buffer
pool index and qbuf operation are atomic
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4465>
when regotiation happens, v4l2src will check if it can reuse current caps,
but we need check if current caps is subset of all query caps from downstream
instead of check it with query caps one by one.
Assuming that the current caps is not the subset of first caps from query caps,
it will go to try fmt. when try fmt success, v4l2src will make pending_set_fmt
to TRUE and going to reset.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4500>
This reverts commit f29c19be58. If this is
called for the reference context then we would run into an infinite
loop, which is not really better than an assertion.
By fixing up DTS to never be ahead of the PTS in the previous commit
this situation should be impossible to hit now.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4498>
Fix the following use:
- upstream sends a video with a rotation tag, say 90°
- upstream switches to another video without rotation
- the second video was still rotated by videoflip
Fix this by resetting the orientation when receiving STREAM_START.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4377>
In order to provide build and provide the jack plugin with the prebuilt
binaries of gstreamer we distribute with releases, we can not depend
on an external dependency nor can we ship plugins linking to libraries
we don't provide.
We can also not provide jack ourselves, as it would likely cause a
mismatch with the jack daemon on the host.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4350>
The decoder needs to force another enumeration of the format. For
this it was clearing the v4l2object insternal list, leaving a fmtdesc
pointer pointing to freed memory. This patch clears the fmtdesc pointer
that has just been free. It also makes sure the probe function does not
use the cached formats list. The probe function will restore the current
fmtdesc pointer based on the currently configured pixelformat.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4317>
As we don't have anything smart in the fixation process, we may endup with
a format that has a lower bitdepth, even if downstream can handle higher
depth. it is notably the case when negotiating with deinterlace, which places
is non-passthrough caps before its passthrough one. This makes the generic
fixation prefer the formats natively supported by deinterlace element over
the HW 10bit format. As some HW can downscale 10bit to 8bit, this can break
10bit decoding.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4317>
The previous code would only check if two packets in a row were duplicates. If
not (i.e. a packet is a duplicate of a packet received slightly before) the code
would generate completely bogus FCI because it assumes there were no duplicates
present in the array.
In order to be efficient, just store all received packets and remove the
duplicates just before the FCI is generated once the array of observations have
been sorted by seqnum.
Fixes TWCC usage with moderate to high packet duplication.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4328>
Short-circuit parsing and recreating the playlist URI if
no HLS directives are going to be applied to it.
Fixes problems playing some streams (YouTube) that have
unneeded escaped characters in the URI and then complain
when GStreamer removes the escaping
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4335>
With GST_SEEK_FLAG_SNAP_AFTER present, the previous version would
adjust seek time based on the keyframe farthest away from desired_time.
This was incorrect, because we always want the *earliest* suitable keyframe
to seek to, not the last one.
With this fix, in case of the SNAP_AFTER, we now look for the closest keyframe
that can be found after desired_time. Behaviour for SNAP_BEFORE should remain
unchanged.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4183>
Assuming that V4L2 CAPTURE devices always use one buffer per JPEG image, we can
always mark JPEGs provided by a V4L2 element as parsed.
The V4L2 elements require that JPEG images sent to V4L2 OUTPUT devices must
always be parsed.
This is necessary to link a V4L2 CAPTURE device with a V4L2 OUTPUT device
without explicitly marking the stream as parsed or adding a jpegparse into the
pipeline.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4229>
There's no guarantee it will *actually* be the URI which refered to what we are
downloading. It could be a stream URI or anything else.
Instead of putting something wrong, put no (specific) referer as a better choice
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3972>
The flowcombiner and active_streams shouldn't be cleared in the
mse-bytestream variant, only in the mss-fragmented one. Otherwise the
soft reset leaves qtdemux in a state where it still believes that it has
streams, but they've been cleared. In that case, a null pointer
dereference happens and the app crashes.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4199>
The abort() method of SourceBuffer in Media Source Extensions is
expected to flush the demuxer and discard the current fragment,
if any. The configuration of tracks, if any, should be preserved.
qtdemux has different behavior for flush events depending on the
context.
This patch activates the intended behaviour only for streams of the
VARIANT_MSE_BYTESTREAM type, conformant to the ISO BMFF Bytestream
specification[1]. This flush behaviour is the same as the one
already in use for adaptivedemux sources.
[1] https://www.w3.org/TR/mse-byte-stream-format-isobmff/https://bugzilla.gnome.org/show_bug.cgi?id=795424
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/4101>
This patch prevents a possible race condition from taking place between the EOS event handling and rtcp send
function/thread.
The condition starts by getting the GST_EVENT_EOS event on the send_rtp_sink pad, which causes two core things
to happen -- the event gets pushed down to the send_rtp_src pad and all sessions get marked "bye" prior to
completion of the event handler. In another thread the rtp_session_on_timeout function gets called after an
expiration of gst_clock_id_wait in the rtcp_thread function. This results in a call to the
ess->callbacks.send_rtcp(), which is configured as a function pointer to gst_rtp_session_send_rtcp via the
RTPSessionCallbacks structure passed to rtp_session_set_callbacks in the gst_rtp_session_init function.
In the race condition, the call to gst_rtp_session_send_rtcp can have the all_sources_bye boolean set to true
while GST_PAD_IS_EOS(rtpsession->send_rtp_sink) evaluates to false. This is the result of gst_rtp_session_send_rtcp
running before the send_rtp_sink's GST_EVENT_EOS handler completes. The exact point at which this condition occurs
is if there's a context switch to the rtcp_thread right after the call to rtp_session_mark_all_bye in the
GET_EVENT_EOS handler, but before the handler returns.
Normally, this would not be an issue because the rtcp_thread continues to run and indirectly call
gst_rtp_session_send_rtcp. However, the call to rtp_source_reset sets the sent_bye boolean to false, which ends up
causing rtp_session_are_all_sources_bye to return false. This gets passed to gst_rtp_session_send_rtcp and the EOS
event never gets sent.
The race condition results in the EOS event never getting passed to the rtcp_src pad, which prevents the bin and
pipeline from ever completing with EOS.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3798>
Makes "start-bitrate" work without setting "connection-speed" property. Having
another property set as a requirement for this one to work is unexpected.
This commit allows to request some initial bitrate for first segment, then
go into adaptive streaming for the rest of media playback.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3895>
A data offset with an offset smaller than the moof length is wrong
in smooth streaming streams. The samples will not be located and
eventually playback will error out. So compensate assuming data
is in mdat following moof.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3840>
A data offset with an offset smaller than the moof length is wrong
in smooth streaming streams.
The samples will not be located and eventually playback will
error out. So compensate assuming data is in mdat following moof.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3840>
this is an issue seen with musl based linux distros e.g. alpine [1]
musl is not going to change this since it breaks ABI/API interfaces
Newer compilers are stringent ( e.g. clang16 ) which can now detect
signature mismatches in function pointers too, existing code warned but
did not error with older clang
Fixes
gstv4l2object.c:544:23: error: incompatible function pointer types assigning to 'gint (*)(gint, ioctl_req_t, ...)' (aka 'int (*)(int, unsigned long, ...)') from 'int (int, int, ...)' [-Wincompatible-function-pointer-types]
v4l2object->ioctl = ioctl;
^ ~~~~~
[1] https://gitlab.alpinelinux.org/alpine/aports/-/issues/7580
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3950>
The av1C box is optional so dropping parsing does not break anything
fundamentally, and there seems to be no historical record how version 0
even looks like while the comments and the parsing disagreed with each
other.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3882>
When using qtdemux in a pipeline that should only work as a pure demuxer (not
for actual playback), qtdemux shouldn't emit new GstSegments to correct
the start time (jump to the future) to ensure that the user experiences no
playback delay. By doing so, it's generating the wrong segments when an append
of data from the past happens. When that happens, downstream elements such as
parsers (eg: aacparse) may clip those buffers laying before the GstSegment and
create problems on the GStreamer client app (eg: WebKit).
Getting buffers clipped out because of the wrong GstSegments started becoming
a problen when this commit was introduced:
ab6e49e9cc audioparsers: add back segment clipping to parsers that have lost it
This clipping makes test DASH shaka 35 from MVT tests[1] to fail in
WebKitGTK/WPE (at least) and can potentially cause a number of other problems
in the WebKit Media Source Extensions (MSE) code.
Note that this new behaviour of not emitting new GstSegments only makes sense
when qtdemux is being used as a pure demuxer and not as part of a regular
pipeline. This is why the variant field has been added. When equal to
VARIANT_MSE_BYTESTREAM, it will make qtdemux behave differently in push mode,
taking decisions that meet the expectations for an MSE-like processing mode.
This kind of tweaks have been done in the past for MSS streams, for instance.
That code has been refactored to use VARIANT_MSS_FRAGMENTED now, instead of
its own dedicated boolean flag.
Co-authored by: Alicia Boya García <ntrrgc@gmail.com>
...who suggested to use "variant: mse-bytestream" in the caps to identify that
mode, as proposed in her unmerged patch:
https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/issues/467
[1] https://github.com/rdkcentral/mvt
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3867>
The live playlists should be updated at a defined interval. The problem is that
this interval was used *after* the playlist was finally received and processed,
which resulted in a gradual shift happening in playlist updates.
Instead store and use the time at which playlists were requested to determine
when the next one should be downloaded.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3883>
The scanning is done in a reverse order, the proper full checks to do are
therefore:
* If the position is beyond half a "segment duration", it's in the following
segment
* If the position is within the first half of a segment, it's in that one
* If the segment is the first one and the position is within half a duration
backwards, we consider the position as being within that first segment
Also handle the case where a "partial only" segment doesn't have a reliable
duration, and therefore use the playlist target duration instead.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3883>
The implementation wouldn't work with regular HLS streams (i.e. the final
fallback).
Now that the implementation uses time to search for the starting
segment (instead of just the n-th from the end), we can specify the correct
hold_back fallback value from the RFC
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/3883>