The graphics_device_info message contains the device display ID
information (device address and device display ID). Stores the data in a
hash table in vdagent.

Signed-off-by: Lukáš Hrázký <lhrazky@redhat.com>

Make sure all module files include config.h

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Remove HAVE_CONFIG_H check, it is always defined.
Remove indentation of includes, now the include is not in a condition
and also some were indented some not.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

We need to pass the original argument list, however argv is modified
by g_option_context_parse so save a copy of the array to make it
possible to restart with the same arguments.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

People can think that 32 means 32 bit while it is always a long.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Different tools (compilers or analysers) recognise missing break
in switch as possible problems.
Many recognise a "fall through" comment to tell it is wanted.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>

Modify directly the new_wbuf->buf buffer instead of writing into
a temporary buffer and then copy in the final one.
This also fixes Coverity warning:

 | uninit_use_in_call: Using uninitialized value "message_header". Field
 | "message_header.data" is uninitialized when calling "memcpy".

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Jonathon Jongsma <jjongsma@redhat.com>

Get the CRTC ID from the output struct rather than using the
output_index as an index into the array of crtcs

Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Acked-by: Lukáš Hrázký <lhrazky@redhat.com>

In the current code, output should never be negative but the check
exists after we use the variable as index. Make the check earlier.

This patch also breaks the check in two in order to provide more
accurate error log.

Found by coverity:
  | check_after_sink: You might be using variable "output" before
  | verifying that it is >= 0.

Changes in v2:
- Move overflow check before accessing the arrays (Frediano)

Changes in v3:
- Move the whole check before accessing the arrays.
- Improve commit log.

Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>

And use well defined macros for standard file descriptors.

Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>

Jonathon Jongsma authored Dec 13, 2018 and Frediano Ziglio committed Dec 15, 2018

Acked-by: Frediano Ziglio <fziglio@redhat.com>

Easier to trigger on Wayland guest by running

> xrandr --output XWAYLAND0 --rotate left

In current master, this causes the spice-vdagentd to disconnect from
the client. In 0.18 branch (latest release), mouse becomes unusable as
mentioned in the referred bug below.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1641723



Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>

The function name is XRRGetScreenSizeRange not RRGetScreenSizeRange.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

If connection fails the socket descriptor is not closed causing
a leak.

Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

The "Requisite" directive is intended to require that the virtio port
device exists before starting the vdagent socket unit. If the device
does not exist, the socket should fail to start. However, without using
"After", the socket will still start even when the device is not
present.

The man page for systemd.unit (v239+) says the following about
'Requisite=':

"Similar to Requires=. However, if the units listed here are not
started already, they will not be started and the starting of this unit
will fail immediately. Requisite= does not imply an ordering
dependency, even if both units are started in the same transaction.
Hence this setting should usually be combined with After=, to ensure
this unit is not started before the other unit."

See also https://github.com/systemd/systemd/issues/4756 for further
details (particularly the following comment:
https://github.com/systemd/systemd/issues/4756#issuecomment-267005324

)

Signed-off-by: Jonathon Jongsma <jjongsma@redhat.com>
Acked-by: Lukáš Hrázký <lhrazky@redhat.com>

Remove @type_to_string, @no_types arguments from
udscs_connect(), udscs_create_server{,_for_fd}().
udscs is used only in vdagent.c and vdagentd.c
and in both cases the args are the same
(vdagentd_messages, VDAGENTD_NO_MESSAGES).

Add debug_print_message_header().

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

All command line options now have long names
as they are required by GLib.

So the supported command line options now are:

    -h, --help

    -d, --debug
    -s, --virtio-serial-port-path
    -S, --vdagentd-socket
    -u, --uinput-device
    -f, --fake-uinput
    -x, --foreground
    -o, --one-session
    -X, --disable-session-integration

Change the types of some global variables that hold the options:
- const char * --> gchar *
- int          --> gboolean

Define DEFAULT_UINPUT_DEVICE as "/dev/uinput",
since there's already DEFAULT_VIRTIO_PORT_PATH, VDAGENTD_SOCKET.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

Victor Toso authored Sep 07, 2018 and Frediano Ziglio committed Sep 13, 2018

It is possible that g_get_user_special_dir() return NULL for
G_USER_DIRECTORY_DESKTOP or G_USER_DIRECTORY_DOWNLOAD; This is a sure
thing to happen in case xdg-user-dirs hasn't been run yet.

Keeping spice-vdagent to run on Initialization stage of
X-GNOME-Autostart-Phase could trigger a race with xdg-user-dirs that
also runs on Initialization, which is the earliest stage possible.

Moving to spice-vdagent to run WindowManager stage avoids the race
with xdg-user-dirs and still runs before any graphical application.

Tested also moving to Desktop stage, which is 4th stage (out of 5) and
worked without apparent drawbacks.

Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1623947


Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Uri Lublin <uril@redhat.com>

Move code handling VDAGENTD_FILE_XFER_STATUS messages
from agent_read_complete() to a separate function
to make the switch statement less cluttered.

Simplify the code a bit.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

Move code handling VDAGENTD_GUEST_XORG_RESOLUTION messages
from agent_read_complete() to a separate function to make
the switch statement less cluttered.

Remove unnecessary res variable.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

The functions would return -1 only if malloc() failed, otherwise 0.
Since malloc() was replaced by g_malloc(), which terminates
the program if the allocation fails, return void instead.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

The functions would return -1 only if malloc() failed, otherwise 0.
Since malloc() was replaced by g_malloc(), which terminates
the program if the allocation fails, return void instead.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

Some older parts of the code currently use
memory functions defined in stdlib.h
and usually handle allocation errors.

On the other hand, newer parts of the code
and GLib/GTK+ functions themselves commonly use
wrappers provided by GLib that terminate
the program when there isn't enough memory.

So it doesn't make much sense to check for
ENOMEM somewhere, while the program can be
terminated at any time elsewhere.

Unify the code and use GLib wrappers only.

malloc --> g_malloc, g_new
calloc --> g_malloc0, g_new0
free   --> g_free

Use g_clear_pointer where possible
for simplification.

Replace subsequent alloc and memcpy calls
with single g_memdup where possible.

Note: this commit only partially touches
src/vdagent/x11.c as the clipboard-handling
code here should be removed in the future.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

Required GLib version was bumped from 2.28 to 2.34 by
61fc548f.

Signed-off-by: Jakub Janků <jjanku@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

udscs_get_peer_cred() is currently used only in
vdagentd.c to obtain PID, so let's drop udscs_get_peer_cred()
and add udscs_get_peer_pid() instead.

Acked-by: Victor Toso <victortoso@redhat.com>

Signed-off-by: Lukáš Hrázký <lhrazky@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>

The handler is meant to ignore the X11 error, name it properly.

Signed-off-by: Lukáš Hrázký <lhrazky@redhat.com>
Acked-by: Frediano Ziglio <fziglio@redhat.com>

364b6bba introduced
session_info_session_is_locked() in session-info.h, however
dummy-session-info.c implemented session_is_locked() instead,
causing builds with --with-session-info=none to fail.

This was fixed by d66fbb9d,
but it didn't remove the incorrect session_is_locked().
Acked-by: Frediano Ziglio <fziglio@redhat.com>

Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Including some basic building tests.

Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Before this patch, the summary would have 'auto' instead of yes or no
for building with gtk.

Signed-off-by: Victor Toso <victortoso@redhat.com>
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

To have a similar output to other configure options we have.

Signed-off-by: Victor Toso <victortoso@redhat.com>
Acked-by: Christophe Fergeau <cfergeau@redhat.com>

Unlike virtual graphic cards, device-assigned graphic cards can
have more than one crtc per output.

Still we currently pick only a single crtc for each output
(but looping over them to find an active one)

Signed-off-by: Uri Lublin <uril@redhat.com>
Acked-by: Victor Toso <victortoso@redhat.com>

Acked-by: Victor Toso <victortoso@redhat.com>

Introduce new functions to handle clipboard,
add new files clipboard.[ch]

This is only a preparatory patch for
following GTK+ clipboard implementation.

Acked-by: Victor Toso <victortoso@redhat.com>

This code using Xlib was removed by commit
"x11: retrieve _NET_WM_NAME using GDK"
a4f5c7dc.
Bring it back to make GTK+ optional for now.

Acked-by: Victor Toso <victortoso@redhat.com>

Add --with-gtk configure option.
If used, favor GTK+ over Xlib.

Acked-by: Victor Toso <victortoso@redhat.com>

Since version 1.9.2, Marco sets _NET_WM_NAME to "Metacity (Marco)"
instead of "Marco", see:

http://git.mate-desktop.org/marco/commit/?id=cbf24f5d78719b930fc0cd6c0c80e00e9d62c95a



Add this new value to wms_with_icons_on_desktop[], so that
vdagent_x11_has_icons_on_desktop() returns TRUE.

Acked-by: Victor Toso <victortoso@redhat.com>

Don't cache _NET_WM_NAME, instead refetch it with each
vdagent_x11_has_icons_on_desktop() call so that correct result is returned
even if the window manager changes during the lifetime of vdagent_x11 struct.
Remove char *net_wm_name from vdagent_x11.

Remove x11 argument from vdagent_x11_has_icons_on_desktop().

Acked-by: Victor Toso <victortoso@redhat.com>