For remote (or non-gl) clients, if a valid gl_draw stream exists,
then we first extract the dmabuf fd associated with the scanout and
share it with the encoder along with other key parameters such as
stride, width and height. Once the encoder finishes creating an
encoded buffer (using the dmabuf fd as input), we then send it
over to the client. And, as soon as the encoder notifies that it
is no longer using the dmabuf fd, we send a gl_draw_done async to
the application.
v2: (suggestions and fixups from Frediano)
- Moved the DisplayStreamData initialization code from
red_marshall_stream_data() into a separate function that is reused
when marshalling gl_draw_stream.
- Used new/delete instead of g_new/g_free for creating and destroying
dmabuf_data object
- s/notify_mem_free/free
s/red_gst_mem_free_cb/red_free_fb
- Removed the usage of opaque from red_free_cb
v3:
- Obtain the key params such as fd, stride, etc from the stream instead
of the scanout
- Replace the switch with if in red_marshall_gl_draw_stream() to avoid
printing a warning (and spamming the console) when a frame is dropped
v4:
- Slightly improve the readability of red_marshall_gl_draw_stream() by
adding checks at the start
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
For non-gl/remote clients, if there is no stream associated with
the DisplayChannel, then we create a new stream. Otherwise, we
just update the current stream's timestamp.
v2: (suggestions and fixups from Frediano)
- Moved the gl_draw_stream object from DCC to DC
- Moved the stream initialization code from display_channel_create_stream()
into a separate function that is reused when creating gl_draw_stream
v3:
- Create a new primary surface whenever a new stream gets created
v4:
- Use nullptr instead of NULL and true instead of TRUE (Frediano)
- Create the stream as part of gl scanout instead of gl draw operation
so that if would be easily possible to obtain key params such as
stride, flags, etc
- Store key params such as fd, flags, stride, etc in the stream so that
we do not have to look at scanout again
v5: (Frediano)
- No need to pass scanout object to create_gl_draw_stream as it is
always NULL
- Don't compute the stream's last_time if drawable is valid
- Let the default input_fps be MAX_FPS
- Use uint32_t type for stride
- Make sure that a newly created stream is provided to all connected clients
- When the scanout's drm_dma_buf_fd is associated with a stream, take
an additional reference on the fd to ensure that it is not closed
when the stream might still be using it
v6:
- Add a new helper is_new_stream_needed() to determine when to create new
stream
- Add more checks to see if the encoder is created or not
- Don't share the dmabuf fd with the stream and instead keep the
ownerhip of the fd with the scanout
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
We need to determine if the client is new enough to support multiple
codecs -- which might include any of the Gstreamer based ones.
v2: (suggestions and fixups from Frediano)
- Add is_gl_client() method to DisplayChannelClient instead of a
dcc_is_gl_client() function.
- Avoid the usage of XXX_CAST macro.
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Once it is determined that an Intel GPU is available/active (after
looking into udev's database), we try to see if there is a h/w
based encoder (element) available (in Gstreamer's registry cache)
for the user selected video codec. In other words, if we find that
the Intel Media SDK Gstreamer plugin (libgstmsdk.so) and associated
libraries (such as va or vaapi) are all installed properly, we add
the appropriate h/w based encoder and post-processor/converter
elements to the pipeline (along with any relevant options) instead
of the s/w based elements.
For example, if the user selects h264 as the preferred codec format,
msdkh264enc and vapostproc will be preferred instead of x264enc
and videoconvert.
v2: (addressed some review comments from Frediano)
- Moved the udev helper into spice-common
- Refactored the code to choose plugins in order msdk > va > vaapi
v3: (Frediano)
- Added relevant encoder options for mjpeg and vp9 codecs (Jin Chung)
v4: (Fixups from Frediano)
- Free the encoder when we cannot find vpp
- Change type and find plugins array length using G_N_ELEMENTS
- Fix gstenc_name UAF by freeing it at the end of the function
- Use g_str_has_prefix instead of strstr
- Include the string "_hw_" in function names that deal with
h/w based plugins
- Rebase on master
v5: rebase on master
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Co-developed-by: Jin Chung Teng <jin.chung.teng@intel.com>
Co-developed-by: Hazwan Arif Mazlan <hazwan.arif.mazlan@intel.com>
Do not offset the time attempting to fix client latency.
Client should handle it by itself.
This remove entirely the delay introduced by the server.
This avoids surely possible time drifts in the client.
The server just sends it's concept of time without trying to force any
delay. Only one end should handle this delay in an attempt to
synchronize audio and video instead that doing it in both ends.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Remove:
test-display-streaming.c: In function 'get_commands':
test-display-streaming.c:218:42: error: 'calloc' sizes specified with 'sizeof' in the earlier argument and not in the later argument [-Werror=calloc-transposed-args]
218 | *commands = (Command*) calloc(sizeof(Command), *num_commands);
| ^~~~~~~
test-display-streaming.c:218:42: note: earlier argument should specify number of elements, later size of each element
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
The original check ensures we do not have an outstanding GL_DRAW. However,
in QEMU, there is no guarantee that a scanout cannot happen while the async
GL_DRAW has not returned a result yet. This introduces a race where if a
scanout is called while there is an outstanding GL_DRAW, QEMU will crash.
Signed-off-by: osy <osy@turing.llc>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
test-listen using GIO had issues running under CI for a while.
GIO is reading some desktop configuration so it's not very CI
friendly.
So instead of using GIO use OpenSSL BIO. The code does not
get much bigger or complicated.
We are already using OpenSSL so we are not adding dependencies.
This fixes CI for Fedora 39 (just released and available on docker).
This allowed to remove an old workaround for GIO in .gitlab-ci.yml
(cfr commit 89edf80821
"ci: Workaround an issue with GLib on Fedora 30")
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
If we use the x264enc encoder to encode a stream, then videoconvert
would convert the BGRx data into Y444, which is the preferred format
for x264enc. However, some decoders particularly the ones that are
h/w based cannot work with Y444 if it was the format used by the
encoder. Therefore, to address these situations, we need a way to
override the format used during the encoding stage which can be
accomplished by using the environment variable introduced in this
patch: SPICE_CONVERTER_PREFERRED_FORMAT.
For example, using NV12 as the output format for the videoconvert
element would allow us to pair a s/w based encoder (such as x264enc)
with a h/w based decoder (such as msdkh264dec) for decoding the
stream as most h/w based decoders only work with NV12 format given
its popularity.
Note that choosing an encoder format such as NV12 over Y444 would
probably result in decreased video quality although it would be
compatible with more decoders. Ideally, the client and server need
to negotiate a suitable format dynamically but the current
capabilities do not allow for such exchange.
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Based-on-patch-by: Hazwan Arif Mazlan <hazwan.arif.mazlan@intel.com>
Signed-off-by: Jin Chung Teng <jin.chung.teng@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
This fixes the following compiler error while targeting Windows platform.
../../server/smartcard.cpp:20:10: fatal error: arpa/inet.h: No such file or directory
20 | #include <arpa/inet.h>
| ^~~~~~~~~~~~~
Signed-off-by: Biswapriyo Nath <nathbappai@gmail.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
With LibreSSL SSL_OP_NO_CLIENT_RENEGOTIATION is opaque which is not
compatible with the OpenSSL 1.0.2 and earlier code path in
red-stream.cpp while SSL_OP_NO_RENEGOTIATION is not yet defined for the
newer OpenSSL code path in reds.cpp.
So with OpenSSL 1.1.0 and later if SSL_OP_NO_RENEGOTIATION is undefined
and SSL_OP_NO_CLIENT_RENEGOTIATION is defined then define the former as
the latter. This will allow the build to succeed with LibreSSL 3.7.2 and
in the future when newer LibreSSL versions add SSL_OP_NO_RENEGOTIATION
that code path will then be used automatically.
Signed-off-by: orbea <orbea@riseup.net>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Using Fedora 38 the compilation fails due to this warning:
In file included from /usr/include/gstreamer-1.0/gst/video/video.h:202,
from ../../server/gstreamer-encoder.c:27:
/usr/include/gstreamer-1.0/gst/video/video-sei.h:39:21: error: 'H265_MISP_NANOSECONDS' defined but not used [-Werror=unused-const-variable=]
39 | static const guint8 H265_MISP_NANOSECONDS[] = {
| ^~~~~~~~~~~~~~~~~~~~~
Ignore the warning for Gstreamer includes.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
The variable 'now' counts in audio sample frames, but the variable
'data' is of type uint8_t *. Multiply 'now' by the size of an audio
sample frame to get the correct source pointer.
This improves the quality of audio recordings in QEMU a little bit.
Fixes: 5d5a7bd181 ("sound: Avoid cast that could cause alignment problems")
Signed-off-by: Volker Rümelin <vr_qemu@t-online.de>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Spice uses rand() to generate the random id, but qemu (at least in the case
of qemu-system-x86) fails to initialize the RNG seed (with e.g. srand()).
The result is, that every SPICE session started (by e.g. libvirtd) has the
same client_id. Usually, this is not a problem, but running something like
a SPICE proxy, relying on the client_id to correctly route connections,
this creates problems.
Fixes:
https://gitlab.com/qemu-project/qemu/-/issues/163
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Remove deprecation warnings like
In file included from ../../server/char-device.cpp:28:
../../server/safe-list.hpp:108:43: error: 'template<class _Category, class _Tp, class _Distance, class _Pointer, class _Reference> struct std::iterator' is deprecated [-Werror=deprecated-declarations]
108 | class safe_list<T>::iterator: public std::iterator<std::forward_iterator_tag, T>
| ^~~~~~~~
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
This leak causes a CI failure running test-listen test with
Valgrind.
Note that "check-valgrind" CI job will still fail but with
less issues.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
gstreamer-plugins-base 1.20 includes a new member in the
GstAppSinkCallbacks struct:
gboolean (*new_event) (GstAppSink *appsink, gpointer user_data);
So it has to be initialized in order to build test-gst.cpp
successfully.
(added in
0a657d6db5
)
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Due to reds->vm_running being initialized to TRUE (since c302e12c
"spice.h: add entries for tracking vm state") the assumption in c23cbd6f
"reds: start QXL devices if VM is running" was wrong and we can't check
on vm_running until that initialization isn't on TRUE (it is that way for
backward compatibility).
Without this revert on qemu initializing spice we will have the
display_init side of qemu not yet ready and therefore respond badly when
spice sends an event as reaction to `red_qxl_start`:
"qxl_send_events: spice-server bug: guest stopped, ignoring."
At least with qemu > v2.0 as a spice consumer is not showing issues as
`red_qxl_start` will be called just after the qemu side is ready
`qemu_spice_display_start` -> `spice_server_vm_start` ... `red_qxl_start`.
Therefore - for now to avoid the current regression - Revert c23cbd6f
"reds: start QXL devices if VM is running" until that old (2012)
initialization is updated (probably an ABI change and therefore taking
some time).
Fixes: https://gitlab.freedesktop.org/spice/spice/-/issues/64
This reverts commit c23cbd6fa8.
Using Fedora 35 the compilation fails due to this warning:
../server/gstreamer-encoder.c: In function 'create_pipeline':
../server/gstreamer-encoder.c:994:5: error: braces around scalar initializer [-Werror]
994 | GstAppSinkCallbacks appsink_cbs = {NULL, NULL, &new_sample, {NULL}};
| ^~~~~~~~~~~~~~~~~~~
../server/gstreamer-encoder.c:994:5: note: (near initialization for 'appsink_cbs.new_event')
../server/gstreamer-encoder.c:994:5: error: missing initializer for field '_gst_reserved' of 'GstAppSinkCallbacks' [-Werror=missing-field-initializers]
In file included from ../server/gstreamer-encoder.c:26:
/usr/include/gstreamer-1.0/gst/app/gstappsink.h:81:16: note: '_gst_reserved' declared here
81 | gpointer _gst_reserved[GST_PADDING - 1];
| ^~~~~~~~~~~~~
cc1: all warnings being treated as errors
Change structure initialisation to avoid the warning.
The same syntax is already used in server/tests/test-gst.cpp.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
A well behaved client implementing the SPICE protocol should check to see
if the server supports the capabilities it needs. This implementation of
the spice-server supports `SPICE_MAIN_CAP_NAME_AND_UUID` and `
SPICE_MAIN_CAP_AGENT_CONNECTED_TOKENS` so it should report this to the
client.
Acked-by: Frediano Ziglio <freddy77@gmail.com>
In OpenSSL3, the SSL_accept call now emits proper errors, which we dump
*before* emitting the expected "SSL_accept failed" error message. The
g_test_expect_message framework doesn't really allow us to discard
messages AFAICT, so instead we add a new expectation with fairly loose
criteria.
Fixes#63
Signed-off-by: Simon Chopin <simon.chopin@canonical.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Some systems use 32-bits, other 64-bits.
Some systems use signed integers, other unsigned integers.
Compute maximum constant based on time_t type.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
These conversions allowed extra conversions to for range loops in
addition to replacing macro usage with .size().
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
display_channel_surface_id_unref resets the surface so
display_channel_surface_has_canvas will return false.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Store pointers to surface object in "surfaces" and allows to
have different surfaces with same ID in memory.
The surface was keep "busy" if there was pending drawing around.
Consider the following case:
1- receive drawing command
2- queue command on DCCs
3- destroy surface
4- send draw
Previously at point 4) you would have to use a surface from
"surfaces" which was destroyed, that is we would have to maintain
the pointer (and canvas) to the surface until reference counter
was 0.
However consider this case:
1- receive drawing command
2- queue command on DCCs
3- destroy surface
4- create surface
5- send draw
What would happen in point 4) ?
We could not change the surface as it will be used by point 5).
To avoid this the code attempts to wait the commands to release the
surface. However this can be an issue, you can't force the clients
to receive pending data if network is slow.
So this patch change this allowing to create surfaces while the old
version will still be used.
This is also more clean from the reference pointer prospective,
as the reference is increased for a specific surface.
Note that now instead of checking for canvas to not be NULL a
simple check for surface pointer is enough.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Mostly left on dcc-send.cpp.
Other minor too.
The change in BitmapData seems odd but the id for cached image
was not used so the only information left was the surface.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Instead of computing the value inside the function to then
compute also in the caller return it.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>
