Reduce GObject changes coming in the next commit since we'll need to
change how we access the marshaller anyway. This will make the
following commits easier to review.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
As data is packae in a single piece of memory send it
altogether.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
Including g_clear_pointer() in glib-compat.h and using it to avoid
warnings in odd situations.
Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
When qemu (for example) delivers audio samples to the spice server, it
does so by requesting a buffer from the spice server
(spice_server_playback_get_buffer()), filling them with audio data, and
then calling spice_server_playback_put_samples() to send them to the
client. Between the call to _get_buffer() and the call to
_put_samples(), we need to ensure that the buffer remains valid. Since
this buffer is allocated within PlaybackChannelClient, we did this by
incrementing a ref on the PlaybackChannelClient in _get_buffer(), and
decrementing the ref in _put_samples(). This has the effect of
potentially keeping the PlaybackChannelClient alive after the spice
client has disconnected.
This was not a problem when PlaybackChannelClient was a simple helper
class. But we intend to change PlaybackChannelClient so that it
inherits from RedChannelClient. In that case, the reference taken in
_get_buffer() would result in the RedChannelClient (and associated
RedChannel, etc) being kept alive longer than expected. To avoid this,
we add an additional ref-counted adapter class (AudioFrameContainer)
that owns the allocated audio frames and can outlive the
RedChannelClient if necessary. When the client is freed, the
AudioFrameContainer is just unreferenced.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
Stops using the dummy channel.
Data handling still goes through DummyChannelClient which is why
empty implementation of some required vfuncs is working.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
Try to arrange destruction in the opposite order of the creation
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
The check to limit too low bit rates was setting encoder->bit_rate
instead of bit_rate. However after some lines bit_rate was used
to set encoder->bit_rate basically removing the lower threshold.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Francois Gouget <fgouget@codeweavers.com>
By removing the stream's video encoder we force the stream to send
future frames using the fallback code, that is as regular screen
updates.
However note that we keep the stream object: we have to. Otherwise
future frames would trigger the creation of a new stream object with a
new video encoder which would again try to stream the video and fail
again and again.
Signed-off-by: Francois Gouget <fgouget@codeweavers.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
Reviewed-by: Victor Toso <victortoso@redhat.com>
Make easier to understand that they refer to client and not
all channel.
Specifically:
- RecordChannel -> RecordChannelClient
- PlaybackChannel -> PlaybackChannelClient
- playback_channel -> playback_client
- record_channel -> record_client
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
SndWorker has been historically based on RedChannel, initial git commit
has:
struct SndWorker {
Channel base;
...
};
SndChannel, contrary to what its name may imply is more focused on
marshalling/sending of sound data, which is the responsibility of
RedChannelClient for the other SPICE channels.
This commit and following ones make the naming of these 2 types more
consistent with how the rest of the code is structured.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
SpicePlaybackState and SpiceRecordState have same structures
changing only slightly the behaviour.
Using SndWorker instead allows some minor simplification and
more code reuse.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
Most of the times the check is done externally
so moving inside the function reduce the code.
This is similar to the way free(3) works.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
This function assume there is only one client.
Was used only by some obsolete functions.
Avoid to use such function in the future.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
This function wrongly close the first client.
Wrongly as closing the file descriptor cause a dandling
file descriptor in the object potentially leading
to closing another file descriptor open later.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
This will close all clients and release the channel properly
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
Now that RedStatFile is private there is no need
to include some headers in stat-file.h.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>
These functions are not used since years and are not supporting
multiple clients.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
Use same constants for common commands.
This will allow code reuse between Record and Playback.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Uri Lublin <uril@redhat.com>
Allows to close worker thread.
This will be used to destroy cleanly CursorChannel and
DisplayChannel.
CursorChannel and DisplayChannel are run in a different
thread. However deregistration of channels and different
steps of destruction should be done in the same thread
so this make possible to join again the 2 threads to
avoid race conditions.
For the moment there is no correct cleanup.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
This make compression faster and avoids a warning on newer
lz4 versions.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
Use dash instead of underscores for file names. This is coherent
with rest of file names.
Rename all tests to test-XXX and remove the spice- prefix when
present. This is coherent with most of the tests.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
self is usually used for GObject, avoid confusion.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
The video encoders already have sophisticated bit rate control code that
can react to both stream reports sent by the client, and server frame
drop notifications. Furthermore they can adjust both the frame rate and
the image quality to best match the network conditions.
But if the client cannot send stream reports all this is bypassed and
instead the streaming code performs its own primitive bit rate control
that can only adjust the frame rate.
So this patch removes the code duplication and lets the video encoders
do their job.
Signed-off-by: Francois Gouget <fgouget@codeweavers.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
The wrong type (RED_TYPE_CHANNEL_CLIENT rather than
TYPE_SMARTCARD_CHANNEL_CLIENT) was used when creating a
SmartcardChannelClient, which meant the _init function was never called,
and SmartcardChannelClient::priv was never initialized.
This caused a crash when trying to connect with --spice-smartcard to a
VM with a smartcard channel configured.
This leaks happen for every connection. Potentially the timer can
be called after the client is closed causing an use after free.
Recently RED_STATISTICS was switched off by default but previous
version have this issue.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Pavel Grunt <pgrunt@redhat.com>
This reverts commit c6881ad1a0.
This patch cause the replay utility run at full speed
to slow down a lot and in some cases getting stuck.
I don't understand the reason and when I tested was working
but as we are going to release would be a pity if this test utility
won't work as useful to get feedback and reports.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>
The following build error occurs when building outside of the source
directory:
glib-mkenums --template spice-server-enums.tmpl.c ../../server/spice-server.h > spice-server-enums.c
Can't open spice-server-enums.tmpl.c: No such file or directory
Makefile:1111: recipe for target 'spice-server-enums.c' failed
Make/Automake uses VPATH to determine that the spice-server-enums.tmpl.c
file listed in the prerequisites for the rule is located in the srcdir.
When we use an automatic variable (e.g. $<), the full path to the
resolved file is used. But when we use the literal filename directly
within the rule definition, this won't happen. So we need to explicitly
specify that the input template file is located in srcdir.
Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>
