summaryrefslogtreecommitdiff
path: root/daemon
Commit message (Collapse)AuthorAgeFilesLines
* session: make sure session display mode is set even if session isn't ↵Ray Strode2014-04-141-6/+6
| | | | | | | | | | | | | explicitly picked Right now we we only set the display mode when a user picks a session from the session list. That means the wrong display mode will get used if wayland is saved from a previous log in. This commit makes sure the display mode gets set as soon as the user credentials are established, since at the point log in is imminent. https://bugzilla.gnome.org/show_bug.cgi?id=728204
* session: drop remnants of WINDOWPATHRay Strode2014-04-141-6/+0
| | | | https://bugzilla.gnome.org/show_bug.cgi?id=728204
* session: don't set DISPLAY for sessions that bring their ownRay Strode2014-04-141-7/+11
| | | | | | | We only want to set DISPLAY if the user session is sharing the display with the login screen. https://bugzilla.gnome.org/show_bug.cgi?id=728204
* manager: be more accepting of callers with uids different from their sessionRay Strode2014-04-101-1/+1
| | | | | | | | | | | | | | If a user runs su in their session, that switched user will now be running in a session that doesn't belong to it. GDM won't allow a user access to the worker process associated with the session in this case. Instead, it will try to create a temporary just-in-time reauthentication channel so reauthentication can happen without having the user talking to another user's worker. Unfortunately, a logic error in the code means, the user won't access to its own just-in-time channel. This commit fixes that. https://bugzilla.gnome.org/show_bug.cgi?id=727982
* manager: don't leak session objects when client is rejectedRay Strode2014-04-101-0/+45
| | | | | | | | | | if a client creates a transient reauthentication session and then we reject the client when it tries to connect to it, then we need to clean up the associated session object. This commit does that. https://bugzilla.gnome.org/show_bug.cgi?id=727982
* session: introduce new client-rejected signalRay Strode2014-04-102-0/+25
| | | | | | | | | If a client gets rejected because it's not allowed to connect to a particular session, we really need to inform the owner of the session object so it can do any clean up it needs to do, if necessary. https://bugzilla.gnome.org/show_bug.cgi?id=727982
* worker: get PATH from parent instead of #defineRay Strode2014-04-102-12/+4
| | | | | | | | If no PATH is set, then the session worker tries to set one up, based on guess. This commit changes GDM to just use the PATH given to GDM itself, rather than guessing (and getting it wrong). https://bugzilla.gnome.org/show_bug.cgi?id=727980
* worker: don't run shell scripts for program sessionsRay Strode2014-04-071-0/+5
| | | | | | | | | | | | GDM uses a worker to manage PAM for user sessions and greeter sessions. For the latter case we shouldn't run the various Pre/Post scripts, but commit Sb48a305e71637181975fc7ec479a71079b6fbc broke that. This commit changes the code to prevent those scripts from getting run again. https://bugzilla.gnome.org/show_bug.cgi?id=727703
* gdm-session: support XDG_CURRENT_DESKTOPLars Uebernickel2014-04-031-0/+35
| | | | | | Set XDG_CURRENT_DESKTOP when DesktopNames is set in the session file. https://bugzilla.gnome.org/show_bug.cgi?id=727546
* manager: don't require session to be on a seatRay Strode2014-03-271-1/+1
| | | | vnc sessions, for instance, aren't on a seat.
* ConsoleKit: a pair of tweaksRyan Lortie2014-03-262-2/+9
| | | | | | | | | | | | | | | | | | It looks like there were a couple of breaks when porting away from ck-connector. The first issue is that consolekit uses some hard-coded rules to decide if a given session is associated with the hard-coded 'Seat1' or not. One of these rules is the "x11-display-device" property being set. We were setting the "display-device" property instead, which is not for graphical sessions. This was causing a new seat to be dynamically created each time gdm started which was causing us to send ActivateSession requests to the wrong seat. The second issue is that ActivateSession will "fail" with "Session already active" errors. Ignore those. https://bugzilla.gnome.org/show_bug.cgi?id=727072
* manager: fix timed loginRay Strode2014-03-241-3/+2
| | | | | | | | There was some confusion over multiple competing enabled variables. This commit eliminates the confusion. https://bugzilla.gnome.org/show_bug.cgi?id=680348
* manger: rename touch_marker_file to touch_ran_once_marker_fileRay Strode2014-03-211-3/+3
| | | | | | | Rename this function to be more descriptive after some discussion with Matthias. https://bugzilla.gnome.org/show_bug.cgi?id=726813
* manager: wait until autologin is initiated before marking it soRay Strode2014-03-211-1/+2
| | | | | | | | | | | | When the slave was merged to the main daemon process, the code for doing autologin accounting was accidentally moved earlier in the autologin process. This caused the greeter to create a login screen and do autologin at the same time. This commit defers the accounting until after the operation is fully initiated. https://bugzilla.gnome.org/show_bug.cgi?id=726813
* session worker: fix one more _MODE_NEW_VT caseRyan Lortie2014-03-191-0/+2
| | | | | | | The previous patch missed one case of GDM_SESSION_DISPLAY_MODE_NEW_VT, resulting in an undefined reference to jump_to_vt(). https://bugzilla.gnome.org/show_bug.cgi?id=726380
* build-goo: drop vestigul gdm-slave.xmlRay Strode2014-03-191-2/+0
| | | | fixes distcheck
* wayland: make optional for FreeBSDRay Strode2014-03-192-0/+18
| | | | | | | FreeBSD build broke, so this commit makes wayland a configure time argument https://bugzilla.gnome.org/show_bug.cgi?id=726380
* manager: get rid of dubious weak referencesRay Strode2014-03-191-40/+28
| | | | | | | | | | The previous commit uses weak references on weak references to manage transient session lifecycles without doing proper accounting. That was a bad idea. This commit uses a hash table to track things more explicitly. https://bugzilla.gnome.org/show_bug.cgi?id=726283
* manager: support just-in-time reauthentication for non-GDM sessionsRay Strode2014-03-191-17/+188
| | | | | | | | | | | | | | | Right now, gnome-shell can't unlock screens running on an X server that isn't managed by GDM (say Xvnc or startx). This is because GDM handles the backend processing for unlocking, and it handles that backend processing from the worker associated with the session. If there is no worker associated with the session (as is the case with Xvnc and startx), then there's no process to handle reauthentication. This commit notices that case, and creates a transient worker on the fly just to perform one off authentication for unlock of non-GDM managed sessions. https://bugzilla.gnome.org/show_bug.cgi?id=726283
* manager: collect more details about bus senderRay Strode2014-03-191-19/+211
| | | | | | | | | | | | In the future we're going to need to know more details about the sender to know how to move forward (such as seat id, session id, if it's remote, etc) in order to create a transient session soley for reauthentication. To prepare for that future, this commit adds the necessary functionality to get_display_and_details_for_bus_sender. https://bugzilla.gnome.org/show_bug.cgi?id=726283
* worker: support authentication without X11 displayRay Strode2014-03-191-25/+28
| | | | | | | | | | | At the moment we unconditionally set PAM_XDISPLAY and PAM_XAUTHDATA based on values passed to the worker. In a future commit, those values are going to become stubs, so as a first step, this commit makes PAM_XDISPLAY and PAM_XAUTHDATA optional. https://bugzilla.gnome.org/show_bug.cgi?id=726283
* manager: explicitly disallow login screen from opening reauth channelRay Strode2014-03-191-3/+113
| | | | | | | | | | | | | It doesn't make sense for it to do, and right now the shell does it up front, waits for the failure, and then does the "right" thing (opens a new auth session) after. This commit makes the failure explicit, so we can subsequently make other cases where a reauth channel is requested work even if there is no session to channel to by implicitly creating a transient one just in time. That will come later. https://bugzilla.gnome.org/show_bug.cgi?id=726283
* session: Wayland sessions are allocated on a new VTJasper St. Pierre2014-03-181-0/+7
| | | | | | | ... at least for now, until we land logind integration in mutter and turn on the Xorg logind-aware codepath in gdm. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session: Wayland sessions bypass XsessionJasper St. Pierre2014-03-181-0/+4
| | | | https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session: Add a simple and naive way to detect Wayland sessionsJasper St. Pierre2014-03-181-5/+40
| | | | | | Simply look for "wayland-sessions" as a path element. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* Add wayland-sessionsJasper St. Pierre2014-03-181-0/+1
| | | | | | Add the wayland-sessions directory, as shipped by gdm. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session-worker: Implement support for the different display server modesJasper St. Pierre2014-03-184-4/+181
| | | | https://bugzilla.gnome.org/show_bug.cgi?id=726380
* manager: Don't reuse the same X server for alternate display modesJasper St. Pierre2014-03-181-1/+19
| | | | | | | | | | | | | | | GNOME is going to need a mode of operation in GDM where the session manages its own display server. In this mode of operation, we won't morph the login screen into a user session by reusing the same X server. Instead, we reset the login screen to prepare it for a future login, and then let GNOME or the session worker activate its own session. This commit adds the prerequisite work needed to gdm-manager.c, but only as dead code, since the way to "turn it on" is stubbed out to always reuse the existing server. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session: Introduce the concept of the "session display mode"Jasper St. Pierre2014-03-182-0/+26
| | | | | | | | The session display mode describes exactly how the worker environment will set up VTs, and whether the greeter X server will be reused for the user session. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* Revert "manager: Don't reuse the same X server when we have a display server"Jasper St. Pierre2014-03-183-26/+1
| | | | | | This reverts commit 72ccb27377886b9998ad8b5b7a3b3070474f90f3. We're going to take a different approach for this.
* server: maintain active_servers list through explicit weak referencesRay Strode2014-03-181-2/+11
| | | | | | | | | It's possible for the server object to get freed before its child watch fires (if the slave is stopped early), so we can't rely on the child watch to main accounting on the active servers list. This commit changes the code over to use weak references instead.
* manager: Don't reuse the same X server when we have a display serverJasper St. Pierre2014-03-173-1/+26
| | | | | | | | | | | | | | GNOME is going to need a mode of operation in GDM where it manages its own display. In this mode of operation, we won't morph the login screen into a user session by reusing the same X server. Instead, we reset the login screen to prepare it for a future login, and then let GNOME activate its session on its own. This commit adds the prerequisite work needed to gdm-manager.c, but only as dead code, since the way to "turn it on" is stubbed out to always return FALSE. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* manager: Rename the "user session" or "session" to the "seed session"Jasper St. Pierre2014-03-171-12/+12
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | A greeter environment always has two sessions: the greeter session, and the "user session". The "user session" tracks the login PAM conversation. PAM conversations need to be created without any user, so the name "user session" is a bit of a misnomer. The "user session" is created when the greeter is created, and exists when we're sitting at the username prompt. When the username is entered, we then start a session worker. When the verification conversation is finished, we then log the user in. The same exact GdmSession follows into the user session. Traditionally, the user session is started on the same X server as the greeter, so the GdmDisplay and GdmSlave all go with it. With the new display server management, we need to make the GdmSession follow into the user session created for the login, and then start a *new* user session for the next user that logs in on the same greeter. The name "seed session" tries to capture this. The GdmSession starts out as a little seed, and with a username and password, will one day blossom into a beautiful little user session which will one day live on its own, free from the greeter that gave it life and nourished it. It's the circle of life, and it moves us all. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* manager: Make migrated a special case when starting the sessionJasper St. Pierre2014-03-171-6/+7
| | | | | | | | | | | | | | | | | | | | | | We have three cases for starting a session: 1) We migrate to an existing session. Simply tear down state and switch. 2) We're starting a new session, and it's a traditional Xorg server. Tear down the greeter, and keep the same X server and GdmDisplay, and start a full user session. 3) We're starting a new session, but it's a display server. Keep the greeter around and give it a new seed session, and start a full user session. The last two cases are really the two sides of the same coin, while the migrated case is really a fast-path special case that doesn't involve launching a user session in any case, so use a "goto out;" as a fast return rather than an if/else. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* manager: update session-id when starting user sessionRay Strode2014-03-171-0/+27
| | | | | | | | | | | In order for session unlocking to work properly, the session-id property of the display object needs to match the session running on the display. The display object's session-id property is bound to the slave's session-id property, so this commit makes sure the slave's session-id is updated to the user's session's id once the session is started. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* display: Clean up old signal handlersJasper St. Pierre2014-03-171-0/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The signal handlers here are intended to be used if the slave itself tries to stop, for instance if the server or greeter die, and will try to finish the display on its own. Letting the signal handler run again will finish the display *again* after an idle, meaning we have this path: gdm_display_finish gdm_display_unmanage on_server_stopped queue_finish ... idle ... gdm_display_finish gdm_display_unmanage on_server_stopped queue_finish ... This might have been an infinite loop, except for the fact that GDM smartly has code to detect the failure state where the X server starts up and stops sufficiently quickly, and in that case transitions the display to the FAILED state, which causes GdmDisplayStore to unref the last ref on the GdmDisplay. Since the finish is still queued, finish_idle runs, and very quickly discovers that our object is no longer a GdmDisplay, so this very quickly manifests into a: gdm_display_finish: assertion 'GDM_IS_DISPLAY (display)' failed error. Not before poking a hole and corrupting some memory trying to clear the finish_idle_id, however.
* display: Remove some unnecessary parts from disposeJasper St. Pierre2014-03-171-18/+3
| | | | | | These can't possibly be active and armed if we're disposing. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* display: Enforce a few invariantsJasper St. Pierre2014-03-171-0/+3
| | | | | | | The GdmDisplay should only be cleaned up if it's finished or failed. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* Integrate the slaves into the main daemon processJasper St. Pierre2014-03-1720-3328/+1349
| | | | | | | | | | For no particular reason, the slave has been kept as a separate process. Integrate this into the main display process by simply making the display keep a handle to a GdmSlave object. To keep the cleanup simple, we won't remove the GdmSlave subtypes yet. A future cleanup should merge the slave functionality into GdmDisplay and its subtypes. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* launch-environment: disconnect session signal handlers when destroyedRay Strode2014-03-171-28/+35
| | | | | | | | | | | | The launch environment's session can actually live longer than the environment itself. This is because the session creates internals "conversation" objects that take a reference on the session, which sometimes keeps it on life support. This commit makes sure the session signal handlers for the launch environment don't get run after the launch environment is destroyed. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session-worker: Move PostLogin / PreSession / PostSession hereJasper St. Pierre2014-03-142-84/+29
| | | | | | | | Pushing these up to the session worker dramatically simplifies the interface for the slave, which will allow us to kill the slaves as separate binaries. Ideally, in the long-term, these scripts should go away.
* server: Move Init script processing hereJasper St. Pierre2014-03-142-7/+7
| | | | | We still need to have it in the XDMCP chooser slave, since we need to initialize the X server, even if we didn't start it ourselves.
* Move gdm_slave_run_script to common codeJasper St. Pierre2014-03-144-184/+45
| | | | | | | We're going to move the code that runs Init to gdm-server.c and the code that runs PostLogin / PreSession to gdm-session-worker.c https://bugzilla.gnome.org/show_bug.cgi?id=726380
* server: Force server regeneration offJasper St. Pierre2014-03-141-1/+1
| | | | | | | | | Server regeneration causes Xorg to re-emit the SIGUSR1 signal, which *really* confuses our state machine. Since we also want to control the server lifetime fairly closely, it doesn't really make sense for it to ever trigger, so force it off. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* server: Pull out common X server argsJasper St. Pierre2014-03-141-2/+4
| | | | https://bugzilla.gnome.org/show_bug.cgi?id=726380
* session: Remove a duplicate definitionJasper St. Pierre2014-03-141-2/+1
| | | | | | And move gdm_session_cancel(); somewhere a bit related. https://bugzilla.gnome.org/show_bug.cgi?id=726380
* daemon: Remove gdm-session-private.hJasper St. Pierre2014-03-071-100/+0
| | | | This file has been flat out unused for quite a while.
* slave: Kill off some dead codeJasper St. Pierre2014-03-071-42/+0
|
* server: Kill off some dead codeJasper St. Pierre2014-03-071-41/+0
|
* xdmcp: unbreak build on OpenBSDAntoine Jacoutot2014-03-031-0/+8
| | | | | | | OpenBSD has no support for IPv4 to IPv6 mapping, so make check for the required definitions before using them. https://bugzilla.gnome.org/show_bug.cgi?id=725554
View Lorry Controller Status