| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
| |
Also adds badges and misc. fixes/improvements.
Closes https://github.com/canonical/lightdm/issues/293
|
|
|
|
|
|
|
|
|
|
| |
This should work now that seat_local_get_active_session returns the
proper value for non-seat0 seats.
Note: systemd-logind v245 and older erroneously report
CanMultiSession=no on non-seat0 seats even when it is supported. This
change will break those users. See
<https://github.com/systemd/systemd/pull/15337>.
|
|
|
|
|
|
| |
This makes it possible to use seat_set_supports_multi_session to
change the support status after the seat is created but before it is
started.
|
|
|
|
|
| |
The name "setup" implies early initialization, when it was actually
run just before start.
|
|
|
|
|
|
|
| |
If the seat has an existing greeter session then we should activate it
regardless of whether the seat supports user switching. (If the seat does not
support user switching then there shouldn't be an existing greeter session, I
think. Either way, it doesn't hurt to try.)
|
|
|
|
| |
This will make it easier to troubleshoot multiseat issues.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before, `seat_local_get_active_session` would always return `NULL` for
non-`seat0` seats. This broke Wayland sessions on non-`seat0` seats:
1. User logged in to a Wayland session on `seat1`.
2. LightDM properly terminated the X server to allow Wayland to
take over the seat's hardware. (See the side note below.)
3. Wayland session started.
4. The X server termination triggered a call to
`display_server_stopped_cb`.
5. `display_server_stopped_cb` called `seat_get_active_session` to
see if it needed to start a greeter to replace a terminated
session associated with the terminated display.
6. `seat_get_active_session` called
`seat_local_get_active_session`.
7. `seat_local_get_active_session` erroneously probed the active VT
even though VTs are only associated with `seat0`.
8. After finding no matching session associated with both `seat1`
and the active VT, `seat_local_get_active_session` returned
`NULL` when it should have returned the new Wayland session.
9. Due to the `NULL` response, `display_server_stopped_cb`
erroneously arrived at the conclusion that a greeter must be
started on `seat1`.
10. LightDM started a new X server and greeter on `seat1`, stomping
on the newly created Wayland session.
Side note: I don't think that terminating X to allow Wayland to take
over the seat's hardware is required because I believe X and
most?/all? Wayland compositors cooperatively share the devices via
systemd-logind or maybe libseat. When switching sessions, logind uses
a hand-off protocol to smoothly change which process is allowed to
access the devices. This makes session switching possible even
without virtual terminals. For details, see
<https://dvdhrm.wordpress.com/2013/08/25/sane-session-switching/>.
The above sequence of events is apparent in the debug log from a
minimal script (the script is not included in this commit; the
relevant lines from the log are copied below, with annotations):
#
# User logs in to seat1 with a Wayland session:
#
GREETER-X-1 AUTHENTICATION-COMPLETE USERNAME=no-password2 AUTHENTICATED=TRUE
*GREETER-X-1 START-SESSION
[+0.32s] DEBUG: Seat seat1: Creating display server of type wayland
[+0.32s] DEBUG: Seat seat1: Display server ready, running session
[+0.32s] DEBUG: Registering session with bus path /org/freedesktop/DisplayManager/Session0
[+0.32s] DEBUG: Session pid=1309577: Running command /home/rhansen/floss/lightdm/tests/src/lightdm-session test-session
#
# LightDM starts shutting down X+greeter for seat1 while
# concurrently activating the new Wayland session (c1 = the
# stopping seat1 greeter session, c2 = the starting Wayland
# session).
#
[+0.33s] DEBUG: Seat seat1: Stopping greeter
[+0.33s] DEBUG: Terminating login1 session c1
[+0.33s] DEBUG: Session pid=1309572: Sending SIGTERM
[+0.33s] DEBUG: Activating login1 session c2
GREETER-X-1 TERMINATE SIGNAL=15
LOGIN1 ACTIVATE-SESSION SESSION=c2
#
# The greeter session and X exit.
#
[+0.33s] DEBUG: Session pid=1309572: Exited with return value 0
[+0.33s] DEBUG: Seat seat1: Session stopped
[+0.33s] DEBUG: Seat seat1: Stopping display server, no sessions require it
[+0.33s] DEBUG: Sending signal 15 to process 1309569
XSERVER-1 TERMINATE SIGNAL=15
[+0.33s] DEBUG: Process 1309569 exited with return value 0
[+0.33s] DEBUG: XServer 1: X server stopped
[+0.33s] DEBUG: Seat seat1: Display server stopped
#
# The start of the problem: LightDM should not restart the greeter
# until the Wayland session terminates:
#
[+0.33s] DEBUG: Seat seat1: Active display server stopped, starting greeter
[+0.33s] DEBUG: Seat seat1: Creating greeter session
[+0.33s] DEBUG: Seat seat1: Creating display server of type x
[+0.33s] DEBUG: Seat seat1: Starting local X display
[+0.33s] DEBUG: XServer 1: Launching X Server
[+0.33s] DEBUG: Launching process 1309583: /home/rhansen/floss/lightdm/tests/src/X :1 -seat seat1 -auth /var/run/lightdm/root/:1 -nolisten tcp
[+0.33s] DEBUG: XServer 1: Waiting for ready signal from X server :1
XSERVER-1 START SEAT=seat1
#
# The Wayland session has started concurrently with the
# replacement X+greeter.
#
SESSION-WAYLAND START XDG_SEAT=seat1 XDG_GREETER_DATA_DIR=/var/lib/lightdm-data/no-password2 XDG_SESSION_TYPE=wayland XDG_SESSION_DESKTOP=wayland USER=no-password2
*XSERVER-1 INDICATE-READY
XSERVER-1 INDICATE-READY
[+0.34s] DEBUG: Got signal 10 from process 1309583
[+0.34s] DEBUG: XServer 1: Got signal from X server :1
[+0.34s] DEBUG: XServer 1: Connecting to XServer :1
XSERVER-1 ACCEPT-CONNECT
[+0.34s] DEBUG: Seat seat1: Display server ready, starting session authentication
[+0.34s] DEBUG: Session pid=1309587: Started with service 'lightdm-greeter', username 'lightdm'
[+0.40s] DEBUG: Session pid=1309587: Running command /home/rhansen/floss/lightdm/tests/src/.libs/test-gobject-greeter
#
# Finally, the replacement greeter stomps on the new Wayland
# session.
#
[+0.40s] DEBUG: Locking login1 session c2
LOGIN1 LOCK-SESSION SESSION=c2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
By default, emitted status text is allowed to match a status matcher
line from anywhere in the script. Thus, a script like the
following:
#?*FOO
#?BAR
will happily accept a sequence of events like this:
BAR
*FOO
This loose ordering avoids test flakiness in the presence of
concurrent events, but can be problematic if the test wants to assert
that the BAR event is an effect of the FOO command and not
coincidental. (A concrete example: assert that a greeter was
launched *because* the user session terminated, not in parallel with
the briefly-lived user session.)
Script authors may be tempted to assert causality by introducing a
WAIT like this:
#?*WAIT
#?*FOO
#?BAR
but that would not have the desired effect because it will still
accept a BAR event before the FOO command (or even the WAIT command)
is executed.
The new FENCE command helps script authors assert causality by
ensuring that an emitted status does not match a line on the other
side of the fence. For example, the following script works as
expected and asserts a causal relationship between FOO and BAR:
#?*WAIT
#?*FENCE
#?*FOO
#?BAR
The above script only works if the WAIT is long enough to sufficiently
rule out coincidence. To see why, note that the following sequence of
events would be accepted by the above script:
*WAIT
*FENCE
BAR
*FOO
The order of the FENCE and FOO should not be switched, otherwise it
introduces a theoretical race condition: If the BAR event is caused by
the FOO command, and the BAR event happens to arrive before the FENCE
is executed, then the test will fail when it should not. (Technically
it is not possible to lose the race right now because of how
run_commands is implemented, but scripts should not rely on an
implementation detail. Besides, it doesn't eliminate the need for the
WAIT.)
This new command will be used in a future commit to add a regression
test for a Wayland multiseat bug.
|
|
|
|
|
|
|
|
|
|
|
| |
This matters when starting a Wayland session on a non-seat0 seat:
* It eliminates an erroneous attempt to switch to VT 0.
* The XDG_VTNR environment variable is no longer set.
Also add a multiseat test.
A similar change for XServerLocalPrivate is not needed because its vt field is
already initialized to -1.
|
| |
|
|
|
|
|
|
|
|
|
| |
The purpose of "cancelling" state that was recently added to the
greeter module is to notify the greeter client about session
cancellation before disconnect. This patch adds a check for
that behavior.
Signed-off-by: Paul Wolneykien <manowar@altlinux.org>
|
|
|
|
|
| |
If the hostname has changed while using a local seat, we will fail to connect and
return to the greeter. Avoid this behavior by recreating the server.
|
|
|
|
|
| |
This makes the errors much easier to find vs. downloading the zip
artifact and extracting it.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
The Ubuntu tests were disabled in commit
40c3b4d4cfd747707bae1fe3953527dcff650213 because they were failing.
They are working now.
|
|
|
|
|
|
|
|
| |
Starting with dbus-daemon v1.15.2, unix:tmpdir no longer causes
dbus-daemon to create abstract sockets (for security reasons), so
unix:tmpdir is now equivalent to unix:dir. (Also, some systems don't
support abstract sockets.) Switch to unix:dir so that it is clear
that the tests expect a socket file to be created in the filesystem.
|
| |
|
| |
|
| |
|
|
|
|
| |
This silences some warning messages on Fedora.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
LP: #1887252
|
|
|
|
|
|
|
| |
Otherwise, no completion message is sent to the greeter leaving it in
the 'in_authentication' state
Signed-off-by: Paul Wolneykien <manowar@altlinux.org>
|
|
|
|
| |
LP: #1949970
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There is no longer support for 8 bit color depth in TigerVNC (see:
https://github.com/TigerVNC/tigervnc/commit/e86d8720ba1e79b486ca29a5c2b27fa25811e6a2);
using it causes a fatal error.
* src/x-server-xvnc.c (x_server_xvnc_init): Set default depth to 24 bit.
* tests/scripts/vnc-command.conf (command): Adjust accordingly.
* tests/scripts/vnc-guest.conf (user-session): Likewise.
* tests/scripts/vnc-login.conf (user-session): Likewise.
* tests/scripts/vnc-open-file-descriptors.conf (user-session): Likewise.
* data/lightdm.conf: Likewise.
|
|
|
|
|
|
|
|
|
|
|
|
| |
When providing the VNCServer command as 'Xvnc -SecurityTypes None',
the formatted command line used would look like:
Xvnc -SecurityTypes None :1 -auth /var/run/lightdm/root/:1
which is invalid (the display number must appear first).
* src/x-server-local.c (x_server_local_start): Format the command
placing the display number before any other provided arguments.
|
|
|
|
|
|
|
| |
Fixes #264.
* src/lightdm.c (start_display_manager): Validate if the VNCServer
command provided binary is available; fallback to 'Xvnc' otherwise.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Overwrite existing X authority files to avoid corruption
- Change logind-check-graphical default config value to true
- Block autologin if shell in nologin or false
- Fix failure if with greeters sending two login requests quickly.
- Drop Qt 4 support, it's been unsupported since 2015
- Fix lightdm_set_layout
- Improve method of determining a session is Wayland
- Don't call setenv with a NULL value - the behaviour is undefined
- Replace deprecated QAbstractItemModel::setRoleNames
- Move D-Bus conf file to $(datadir)/dbus-1/system.d
- Fix tests failing when compiled with --with-greeter-user
- Use Python 3 in tests
- Disable compiler optimizations for test programs
- Compilation fix for glibc 2.33
- Remove deprecated use of G_TYPE_INSTANCE_GET_PRIVATE, G_PARAM_PRIVATE
- Fix compile failure due to use of clearenv on FreeBSD
- Use a size_t to resolve a compile warning
- Fix DesktopManager typo in man page
* debian/lightdm.install:
- Update for new D-Bus config location
|
| |
|
| |
|
|
|
|
| |
77a7c6b7b8ca896b98ef43826641bdd520650bfb
|
|
|
|
| |
Regression introduced in 77a7c6b7b8ca896b98ef43826641bdd520650bfb
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There is a race in the greeter (greeter as in lightdm-gtk-greeter) easily triggered by double-clicking on the login button (or clicking it with a bouncing mouse button) with a password-less login.
The problem arises when the daemon has already sent SERVER_MESSAGE_END_AUTHENTICATION to the greeter,
but before it digests it, it handles more X11 events (e.g. the second click), sending the daemon a GREETER_MESSAGE_CONTINUE_AUTHENTICATION
(before the GREETER_MESSAGE_START_SESSION) which will confuse handle_continue_authentication(),
making it call session_respond_error(), which then sends a PAM_CONV_ERR to the session child,
which will in turn get confused because it's expecting the (length of the) session error file name.
As I don't really know how to avoid the race in the greeter (you would need to teach it prioritizing messages from the master over messages from X11), work around it by setting a flag in GreeterPrivate if SERVER_MESSAGE_END_AUTHENTICATION has been sent and ignore GREETER_MESSAGE_CONTINUE_AUTHENTICATION if that flag is set.
The change was verified on 1.18.3 and then adopted to HEAD.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
Without `logind-check-graphical=True` beeing set, lightdm is prone to a race condition, where lightdm has started before the graphics driver was loaded.
|
| |
|
| |
|