GNOME.org
 

Re: [gdm-list] Your gdm changes and SunRay...

Hi,
as Brian Cameron asked me to continue the discussion on the gdm mailing list I am sending my replay about the latest MultiSeat changes in gdm here now.
As an introduction: We are currently trying to get the latest SunRay Server (SRS) 4.2 version running with Ubuntu 9.10 which in turn uses gdm 2.28. However, as SRS is heavily relying on the previous "gdmdynamic" feature the SRS software is not working out-of-the-box anymore with systems using gdm 2.28+. After searching the internet we found the submitted patches from Brian Cameron and Halton Huo for gdm and consolekit and tried to apply those to the sources used in Ubuntu 9.10. However, as the work from Brian and Halton is still in progress we found some still existing problems and thus are reporting them here now. 

Am 03.11.09 20:11, schrieb Brian Cameron:


Can you please send me a link to the git repository so that I may
eventually create own patch files while you are not able to update them
at the above locations. Thanks.


The latest code is here:

http://cgit.freedesktop.org/ConsoleKit/log/?h=multi-seat
http://cgit.freedesktop.org/ConsoleKit/tree/?h=multi-seat

I believe your problems will go away if you use the latest code in that
branch. Let me know if you have any further problems, or if this fixes
the issues you have noticed.

The patches at opensolaris.org should be updated to include the latest
fixes from the above git branch in the next day or two, by the end of
the week at the latest. Halton or I will make sure that the bugs in
bugzilla are updated with the latest patches also so that people do not
use the old patches by accident, if they don't know about git.
Unfortunately the latest patches available in the ConsoleKit git as well as in the gdm git did not fully solve our problems here. 


-- cut here --
if (kind == CK_SEAT_KIND_DYNAMIC) {
g_hash_table_steal (manager->priv->sessions, orig_ssid);
g_hash_table_remove (manager->priv->sessions, orig_ssid);
}
-- cut here --

now it doesn't crash anymore here. After that change we have been
finally able to log in to our SunRay Server. However, after a short
test we came across a second problem which I want to outline below:

>>>

Halton, have we fixed this problem in our latest patches?


FYI: It seems this bug is still present, at least in the patch file
provided at the above location. I also verified that it still crashes by
using your patches. However, after inserting the above
g_hash_table_steal() call the segmentation fault vanishes.


Do you still see this problem using the latest code in the git branch?
I believe that this is one of the issues that Halton recently fixed,
though I am not sure.
Yes, the problem is still present in the latest code in the git branch. Here I can perfectly reproduce a crash of console-kit-daemon as soon as I log into the SRS. However, by adding the above "g_hash_table_steal()" line right before the remove() finally solves that problem and console-kit-daemon does not crash anymore. 


2) With your gdm patches applied it seems that gdm itself does not
restart the corresponding X server anymore if either the X server
process is killed via (kill PID) or if the session is killed by
Ctrl-Bspc-Bspc. This seems to not only be true for the dynamic SunRay
sessions, but also for any local session (e.g. on display :0). Please
find attached the debug output where you can hopefully spot the problem
that gdm tries to start another gdm-simple-slave, however it seems to
fail and thus it ends up being aborted. Unfortunately we haven't
found a workaround or fix for this problem yet as we are not that

>>>> trained with gdm<>consolekit interactions. However, I believe that
>>>> this could be caused either by the reason that the display isn't
>>>> freed that quickly so that the new gdm slave can not start a new X
>>>> server that fast or it might be because there is still something
>>>> missing in the gdm<>consolekit interaction which causes the
>>>> X server restart to fail.
>>>

I believe that this problem is also fixed? Halton, can you confirm?


I just grabed the patches from the above directory now and recompiled
consolekit 0.3.1 and gdm 2.28.1 for Ubuntu 9.10 here. However, the net
result is the same. I am able to connect to the running SunRay service,
but as soon as I kill a session with Ctrl+Bspc+Bspc gdm never returns on
that display. The same happens with the local display :0, btw. However,
the debug output of gdm is now slightly different to the first one I
sent to you today. Now gdm is not even trying to restart
gdm-simple-slave as it seems. Perhaps you guys have an idea what might
cause that.


Do you still see this problem with the latest code in git? Again, I
think this is an issue that Halton recently fixed, though I am again not
sure.
Yes, unfortunately this problem is also still present with the latest changes from the "display-configuration" branch in the gdm git. I can perfectly verify that by simply manually killing the Xorg process. As soon as I kill it here a new Xorg will never be automatically started on display :0 like it is normally done. The same happens on the SRS displays. Only the initial session gets an X-server process but as soon as Xnewt/Xorg is killed it never returns on that display.
However, after a longer investigation my colleagues and I tracked the problem down to the newly added gdm_dynamic_display_finish() function in daemon/gdm-dynamic-display.c. We found out that there the following statements are used: 

-- cut here --
gdm_dynamic_display_finish (GdmDisplay *display)
{
  g_return_val_if_fail (GDM_IS_DISPLAY (display), FALSE);

  GDM_DISPLAY_CLASS (gdm_dynamic_display_parent_class)->finish (display);

  gdm_display_unmanage (display);

  return TRUE;
}
-- cut here --
However, this seems to cause the display to be finished but never restarted like it is done with a normal local display or with the previous unpatched gdm 2.28. Even the local display :0 seems to be treated as a "dynamic" display now. By replacing the function by the same one used in daemon/gdm-static-display.c it seems to work: 

-- cut here --
gdm_dynamic_display_finish (GdmDisplay *display)
{
  int status;

  g_return_val_if_fail (GDM_IS_DISPLAY (display), FALSE);

  /* Don't call parent's finish since we don't ever
  want to be put in the FINISHED state */

  /* restart dynamic displays */
  gdm_display_unmanage (display);

  status = gdm_display_get_status (display);
  if (status != GDM_DISPLAY_FAILED) {
     gdm_display_manage (display);
  }

  return TRUE;
}
-- cut here --
After that change the local :0 display gets a new Xorg process as soon as the old one is killed. In addition, with that change the Xserver of the Sunray Server (Xnewt) also gets respawned correctly.
Finally the above differences allowed us to get SRSS 4.2 running with gdm 2.28 under Ubuntu 9.10.
It would, however, really be nice if you guys could review my proposed changes and perhaps integrate them into your latest patchsets and/or change them in case my changes are not appropriate for the general case.
No matter what, I would like to thank Halton and Brian for getting this done in gdm 2.28+. And I would like to request that their changes should be integrated into the latest gdm trunk ASAP so that future Ubuntu versions (e.g. comming LTS 10.04) ship already with a Sunray compatible gdm version. 

best regards,
jens
--
Jens Langner                                         Ph: +49-351-2602757
Forschungszentrum Dresden-Rossendorf e.V.
Institute of Radiopharmacy - PET Center                 J Langner fzd de
Germany                                               http://www.fzd.de/
[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]