When tunneling RTP over RTSP the stream transports are stored in a hash
table in the GstRTSPClientPrivate struct. They are used for, among other
things, mapping channel id to stream transports when receiving data from
the client. The stream tranports are created and added to the hash table
in handle_setup_request(), but unfortuately they are not removed in
handle_teardown_request(). This means that if the client sends data on
the RTSP connection after it has sent the TEARDOWN, which is often the
case when audio backchannel is enabled, handle_data() will still be able
to map the channel to a session transport and pass the data along to it.
Which eventually leads to a failing assert in gst_rtsp_stream_recv_rtp()
because the stream is no longer joined to a bin.
We avoid this by removing the stream transports from the hash table when
we handle the TEARDOWN request.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/184>
As far as I can tell, this is neither explicitly allowed nor
forbidden by RFC 7826.
Meanwhile, URLs such as rtsp://<IP>:554 or rtsp://<IP>:554/ are in
use in the wild (presumably with non-GStreamer servers).
GStreamer's prior behavior was confusing, in that
gst_rtsp_mount_points_add_factory() would appear to accept a mount
path of "" or "/", but later connection attempts would fail with a
"media not found" error.
This commit makes a mount path of "/" work for either form of URL,
while an empty mount path ("") is rejected and logs a warning.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/168>
To prevent cases with prerolling when the inactive stream prerolls first
and the server proceeds without waiting for the active stream, we will
ignore GstRTSPStreamBlocking messages from incomplete streams. When
there are no complete streams (during DESCRIBE), we will listen to all
streams.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/167>
Add another seek_one_active_stream test but with a demuxer. The demuxer
will flush both streams in opposed to the existing test which only
flushes the active stream. This will help exposing problems with the
prerolling process after a flushing seek.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/167>
When play a media with both sender and receiver stream, like ONVIF
back channel audio in, gst_rtsp_media_get_rates call
gst_rtsp_stream_get_rates for each stream to set the rates. But
gst_rtsp_stream_get_rates return false for the receiver steam, which
lead a g_assert crash.
Instead to get rates on all streams, now just get rates on sender
streams.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/150>
ulpfec correction is obviously useless when receiving a stream
over TCP, and in TCP modes the rtp storage receives non
timestamped buffers, causing it to queue buffers indefinitely,
until the queue grows so large that sanity checks kick in and
warnings start to get emitted.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/149>
rtsp_media_unsuspend() is called from handle_play_request()
before sending the play response. Unblocking the streams here
was causing data to be sent out before the client was ready
to handle it, with obvious side effects such as initial packets
getting discarded, causing decoding errors.
Instead we can simply let the media streams be unblocked when
the state of the media is set to PLAYING, which occurs after
sending the play response.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/147>
clang 10 is complaining about incompatible types due to the
glib typesystem.
```
../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-thread-pool.c:534:10: error: incompatible pointer types passing 'typeof ((((void *)0))) *' (aka 'void **') to parameter of type 'GThreadPool **' (aka 'struct _GThreadPool **') [-Werror,-Wincompatible-pointer-types]
```
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/145>
clang 10 is complaining about incompatible types due to the
glib typesystem.
```
../subprojects/gst-rtsp-server/gst/rtsp-server/rtsp-thread-pool.c:534:10: error: incompatible pointer types passing 'typeof ((((void *)0))) *' (aka 'void **') to parameter of type 'GThreadPool **' (aka 'struct _GThreadPool **') [-Werror,-Wincompatible-pointer-types]
```
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/145>
This causes them to send caps events before data flow, which is
usually a pretty correct thing to do!
Not doing so manifested in a bug where ssrcdemux wouldn't forward
the caps it had received with an extra ssrc field, as it hadn't
received any caps event.
Fixes#85
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/141>
Otherwise the transports are not set up yet during the PLAY request
handling when unsuspending (and thus unblocking) the media.
In case of live pipelines this then causes the first few packets to go
to the sinks before they know what to do with them, and they simply
discard them which is rather suboptimal in case of keyframes.
For non-live pipelines this is not a problem because the sink will still
be PAUSED and as such not send out the data yet but wait until it goes
to PLAYING, which is late enough.
Adding the transports multiple times is not a problem: if the transport
is already added it won't be added another time and TRUE will be
returned.
This fixes a regression introduced by a7732a68e8
before 1.14.0.
Fixes https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/issues/107
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-rtsp-server/-/merge_requests/135>
