Add support for different suspend modes. The stream is suspended right after
producing the SDP and after PAUSE. Different suspend modes are available that
affect the state of the pipeline. NONE leaves the pipeline state unchanged and
is the current and old behaviour, PAUSE will set the pipeline to the PAUSED
state and RESET will bring the pipeline to the NULL state.
A stream is also unsuspended when it goes back to PLAYING, for RESET streams,
this means that the pipeline needs to be prerolled again.
Base on patches by Ognyan Tonchev <ognyan@axis.com>
See https://bugzilla.gnome.org/show_bug.cgi?id=711257
Start live streams in the blocked state and make them preroll using the
messages. This ensure that no data is played by the sink until we explicitly
unblock the stream right before going to PLAYING.
See https://bugzilla.gnome.org/show_bug.cgi?id=711257
* gst/rtsp-server/rtsp-media.c (gst_rtsp_media_set_state): if we are
trying to change to GST_STATE_NULL and media is in error status, we
remove all transports.
Fixes https://bugzilla.gnome.org/show_bug.cgi?id=712776
* gst/rtsp-server/rtsp-stream.c: add new method gst_rtsp_stream_get_pt.
* gst/rtsp-server/rtsp-media.c (pad_added_cb): find real payloader
element and create the stream with this one instead of the dynpay%d
element.
https://bugzilla.gnome.org/show_bug.cgi?id=712396
* rtsp-auth: Refer to GstRTSPToken, not GstRTSPtoken
* rtsp-auth: Refer to part of constant name as text
* rtsp-auth/-permissions/-token: Refer to Permissions not Permission
* rtsp-session-media: Fix GstRTSPSessionMedia typo
* rtsp-stream: Fix typo when refering to GstBin
https://bugzilla.gnome.org/show_bug.cgi?id=714988
Use a guint instead of guint8 to increment the address. It's still not
completely correct because a guint might not be able to hold the complete
address range, but that's an enhacement for later.
Add unit test to test improved behaviour.
https://bugzilla.gnome.org/show_bug.cgi?id=708237
If we call g_main_loop_quit before the thread has entered g_main_loop_run, we
don't actually stop the mainloop ever. Solve this race by adding an idle source
to the mainloop that calls the _quit. This way we immediately exit the mainloop
if quit was called before we started it.
Previously a role that was removed was unreffed twice, and when
replacing an existing role the replaced role was freed while still being
referenced. Both bugs are now fixed.
See https://bugzilla.gnome.org/show_bug.cgi?id=710202
If the SETUP url contains a query it must be appended to the control
path so that it matches any already created stream in the media. The
query will also be appended to the session media path.
We should not depend on whether or not the pipeline state change
returned NO_PREROLL or not. A media could dynamically change its
element and switch from seekable to non seekable so it's best to test
the seekable nature of the pipeline dynamically when we try to do a seek.
Only map the request url to a path in the DESCRIBE method. The SDP then
contains the base and control urls that should be used to SETUP/PAUSE/
PLAY/TEARDOWN the media.
This reverts commit e3fded2cec.
This is not correct, we only remap the URL to a path in DESCRIBE, the SDP then
contains the base and control urls which are used in the SETUP, PLAY,
PAUSE and TEARDOWN requests.
Make a vmethod to transform an url into a path. The path is then used to lookup
the factory. This makes it possible to also use other bits of the url, such as
the query parameters, to locate the factory.
Also fix race condition if two threads are asking for the first
thread from the thread pool at once. This would case two internal
GThreadPools to be created.
https://bugzilla.gnome.org/show_bug.cgi?id=707753
When no auth module is specified, use our table of defaults to look up the
default value of the check instead of always allowing everything. This was
we can disallow client settings by default.
If we try to reuse a thread right after we made it stop, we end up using a
stopped thread. Catch this case and only reuse threads that are not stopping.
Don't authorize on methods anymore but on the resources that we
try to access, this is more flexible.
Move the authorization checks to where they are needed and let the
check return the response on error.
Avoid passing GstStructure in the add_role method, use varargs instead
to construct the structure behind the scenes. We can then also use the
structure name as the role and simplify some more logic.
Move handling of the unauthorized response to the auth module, it can add
the appropriate headers to request authorization for the required method
much better than the client.
