It is entirely possible for the cancellable to be cancelled (and freed)
in gst_rtsp_connection_flush() while there may be an ongoing read/write
operation.
Nothing prevents gst_rtsp_connection_flush() from waiting for the
outstanding read/writes.
This could lead to a crash like (where cancellable has been freed
within gst_rtsp_connection_flush()):
#0 0x00007ffff4351096 in g_output_stream_writev (stream=stream@entry=0x7fff30002950, vectors=vectors@entry=0x7ffe2c6afa80, n_vectors=n_vectors@entry=3, bytes_written=bytes_written@entry=0x7ffe2c6af950, cancellable=cancellable@entry=0x7fff300288a0, error=error@entry=0x7ffe2c6af958) at ../subprojects/glib/gio/goutputstream.c:377
#1 0x00007ffff44b2c38 in writev_bytes (stream=0x7fff30002950, vectors=vectors@entry=0x7ffe2c6afa80, n_vectors=n_vectors@entry=3, bytes_written=bytes_written@entry=0x7ffe2c6afb90, block=block@entry=1, cancellable=0x7fff300288a0) at ../subprojects/gst-plugins-base/gst-libs/gst/rtsp/gstrtspconnection.c:1320
#2 0x00007ffff44b583e in gst_rtsp_connection_send_messages_usec (conn=0x7fff30001370, messages=messages@entry=0x7ffe2c6afcc0, n_messages=n_messages@entry=1, timeout=timeout@entry=3000000) at ../subprojects/gst-plugins-base/gst-libs/gst/rtsp/gstrtspconnection.c:2056
#3 0x00007ffff44d2669 in gst_rtsp_client_sink_connection_send_messages (sink=0x7fffac0192c0, timeout=3000000, n_messages=1, messages=0x7ffe2c6afcc0, conninfo=0x7fffac019610) at ../subprojects/gst-rtsp-server/gst/rtsp-sink/gstrtspclientsink.c:1929
#4 gst_rtsp_client_sink_try_send (sink=sink@entry=0x7fffac0192c0, conninfo=conninfo@entry=0x7fffac019610, requests=requests@entry=0x7ffe2c6afcc0, n_requests=n_requests@entry=1, response=response@entry=0x0, code=code@entry=0x0) at ../subprojects/gst-rtsp-server/gst/rtsp-sink/gstrtspclientsink.c:2845
#5 0x00007ffff44d3077 in do_send_data (buffer=0x7fff38075c60, channel=<optimized out>, context=0x7fffac042640) at ../subprojects/gst-rtsp-server/gst/rtsp-sink/gstrtspclientsink.c:3896
#6 0x00007ffff4281cc6 in gst_rtsp_stream_transport_send_rtp (trans=trans@entry=0x7fff20061f80, buffer=<optimized out>) at ../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-stream-transport.c:632
#7 0x00007ffff4278e9b in push_data (stream=0x7fff40019bf0, is_rtp=<optimized out>, buffer_list=0x0, buffer=<optimized out>, trans=0x7fff20061f80) at ../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-stream.c:2586
#8 check_transport_backlog (stream=0x7fff40019bf0, trans=0x7fff20061f80) at ../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-stream.c:2645
#9 0x00007ffff42793b3 in send_tcp_message (idx=<optimized out>, stream=0x7fff40019bf0) at ../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-stream.c:2741
#10 send_func (stream=0x7fff40019bf0) at ../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-stream.c:2776
#11 0x00007ffff7d59fad in g_thread_proxy (data=0x7fffbc062920) at ../subprojects/glib/glib/gthread.c:827
#12 0x00007ffff7a8ce2d in start_thread () from /lib64/libc.so.6
#13 0x00007ffff7b12620 in clone3 () from /lib64/libc.so.6
Fix by adding a cancellable lock and returning an extra reference used
across all read/write operations. gst_rtsp_connection_flush() can free
the in-use cancellable and it will no longer affect any in progress
read/write.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2799>
1.21.0.1 should not satisfy a check for 1.22.0.
If someone needs more control they should do a feature check for
the symbol in the headers or lib.
Based on a similar patch by Tim-Philipp Müller for libnice.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2501>
The caps negotiation should respect the selected method to the test pipeline below works properly.
gst-launch-1.0 videotestsrc ! video/x-raw,width=320,height=600 ! videoflip method=clockwise ! video/x-raw,width=600,height=320 ! fakesink
Signed-off-by: Adrian Fiergolski <adrian.fiergolski@fastree3d.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2803>
In the trick mode, driver may queue a valid buffer follow by an
empty buffer which has no valid data to indicate EOS.For the empty
buffer whose memory is multi-plane, need to resize it before
unreference it.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2731>
Depending on device feature level, d3d11 runtime can support
ID3D11Fence which is equivalent to ID3D12Fence.
Waiting using fence has performance-wise benefit over pulling
ID3D11Query status. If ID3D11Fence is not supported by device,
then ID3D11Query will be used instead.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2790>
4x downscaling of chroma with co-sited chroma has never worked
it seems.
Fixes incorrect videotestsrc output and videoconvert conversions
to Y41B, YUV9, YVU9 and IYU9 with co-sited chroma.
e.g.
gst-launch-1.0 videotestsrc ! video/x-raw,format=Y41B,width=1280,height=720 ! \
videoconvert ! autovideosink
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2789>
It may happens that bitstream doesn't provided SPS in decoding order
(like in VPSSPSPPS_A_MainConcept_1 conformance test file).
To be sure that the decoder got the correct SPS parameters process
SPS just before start decoding the frame.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2575>
While possible defer computataion of pps and sps fields until
slice parsing since it may happens that bitstreams don't encoded
them in expected order.
A example weird ordered bitstreams is VPSSPSPPS_A_MainConcept_1
conformance test.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2575>
The function g_array_sized_new() leaves the len to 0, but the slice
implementation assumes it would be set to 4. Sending multiple slices is
not yet support for H.264 as no driver needed it yet, but if that code
was to be used it would have overflowed as the array would never grow as
multiple 0 by 2 always results in 0.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1079>
And also don't assert that there are no buffers queued up when handling
an EOS event. The pad's streaming thread might've already received a new
stream-start event and queued up a buffer in the meantime.
This still leaves a race condition where the srcpad task sees all pads
in EOS state and finishes the stream, while shortly afterwards a pad
might receive a stream-start event again, but this doesn't seem to be
solveable with the current aggregator design.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2769>
