gdm crashes on every second LDAP login

Bug #603169 reported by Jeff Strunk on 2010-07-08
30
This bug affects 5 people
Affects Status Importance Assigned to Milestone
gdm (Ubuntu)
Medium
Unassigned

Bug Description

Binary package hint: gdm

One of our testers discovered that he has to login twice if he logs out. This happens on all of our workstations. We've seen this with nvidia, ati, intel, and virtualbox.

Here is what happens:
1. boot. result: gdm loads.
2. login. result: success.
3. logout. result: gdm loads.
4. login. result: gdm crashes and reloads.
5. login. result: success.
6. logout. result: gdm loads.
7. login. result: gdm crashes and reloads.
steps 2-4 repeat.

I found an error message in /var/log/gdm/:0-slave.log.1 after the crash. I will attach that file. The error says "could not find user...". However, I can repeatedly login on the cli and via ssh at any stage of this issue. If I log in via ssh when gdm would crash, gdm still crashes the next time I login via gdm.

This is not a duplicate of lp bug #532047. I have disabled plymouth-splash and later purged plymouth. Neither fixed the problem.

I tested with a local user. The bug does not show up when I login and logout repeatedly with the local user. The bug does happen if I try to login with an LDAP user after logging out from the local user's session.

Thank you.

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: gdm 2.30.2.is.2.30.0-0ubuntu2
ProcVersionSignature: Ubuntu 2.6.32-23.37-generic 2.6.32.15+drm33.5
Uname: Linux 2.6.32-23-generic x86_64
Architecture: amd64
Date: Thu Jul 8 08:33:36 2010
ProcEnviron:
 LANGUAGE=
 PATH=(custom, user)
 LANG=en_US.utf8
 SHELL=/bin/bash
SourcePackage: gdm

Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :

The previous common-password attachment was delivered by our configuration management system last night. This testing included removing the deprecated pam_cracklib line.

Jeff Strunk (jstrunk-math) wrote :
Jeff Strunk (jstrunk-math) wrote :
Sebastien Bacher (seb128) wrote :

Thank you for your bug report. Did that issue happens on lucid before using the stable updates? Do you tweak the greater in some way, to not display the users list for example? Do you get any crash mention the Xorg.0.log or Xorg.0.log.old or syslog?

Sebastien Bacher (seb128) wrote :

The slave log has the crash but it lacks debug symbols, could you install gdm-dbgsym and libglib2.0-0-dbgsym as indicated on http://wiki.ubuntu.com/DebuggingProgramCrash and get a new slave.log

Changed in gdm (Ubuntu):
importance: Undecided → Medium
status: New → Incomplete
Jeff Strunk (jstrunk-math) wrote :

syslog showed the same trace as the slave log, but it did not have the pam information. I have attached an X log file from /var/log/gdm. None of /var/log/Xorg* had any errors.

We have changed gdm's configuration significantly. I will revert to the default and test again. Our primary modifications are the following:
 * disable user list
 * show our logo image
 * run bsetbg with a random image from /etc/gdm/Init/Default
 * run an account request program from /etc/gdm/Init/Default

gdm-dbgsym_2.30.2.is.2.30.0-0ubuntu2_amd64.ddeb is missing from http://ddebs.ubuntu.com/pool/main/g/gdm/

# apt-get install gdm-dbgsym
Reading package lists... Done
Building dependency tree
Reading state information... Done
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
  gdm-dbgsym: Depends: gdm (= 2.30.0-0ubuntu5) but 2.30.2.is.2.30.0-0ubuntu2 is to be installed
E: Broken packages

Jeff Strunk (jstrunk-math) wrote :

It still messes up after a clean install with no modification to the gdm settings.

Sebastien Bacher (seb128) wrote :

right, it seems the debug collector had issue and didn't get this build, do you get the issue using gdm 2.30.0-0ubuntu5 or not? Could you try to sudo apt-get install gdm=2.30.0-0ubuntu5 gdm-dbgsym and see if you get the issue with it?

Jeff Strunk (jstrunk-math) wrote :

The issue went away with 2.30.0-0ubuntu5.

Sebastien Bacher (seb128) wrote :

could be similar to bug #579044

Changed in gdm (Ubuntu):
status: Incomplete → New
Jeff Strunk (jstrunk-math) wrote :

I've got some bad news. The issue is present in 2.30.0-0ubuntu5. However, the issue does not show up when the dbgsym packages are installed.

Jeff Strunk (jstrunk-math) wrote :

It looks like this bug still has a possibility of occurring even if the debug symbols are installed. I foolishly deleted the /var/log/gdm/* to try to isolate the messages after this happened. Then I couldn't reproduce it very quickly. I found the backtrace in syslog though.

Sebastien Bacher (seb128) wrote :

the stacktrace is a debug one in this log, thank you

Changed in gdm (Ubuntu):
status: New → Confirmed
Jeff Strunk (jstrunk-math) wrote :

We thought deploying with the debug symbols would be good enough. Unfortunately, on systems where multiple people log in and out all day, this bug eventually leads to no usable login screens.

I think the following is happening. Each time this bug occurs, a virtual terminal is left hanging with no login prompt, and GDM starts on a new virtual terminal. Eventually, GDM runs out of clean virtual terminals to start on, and no one can log in with GDM. At this point, restarting GDM does not reset these virtual terminals. "service gdm stop" and "service gdm restart" will always report that gdm is not running. "service gdm start" reports that it has started gdm. It seems like the only solution is to reboot.

This makes this bug a higher priority for us. We can't fully deploy lucid if people won't be able to log in.

Thank you.

I think this bug may be a duplicate of <https://bugs.launchpad.net/ubuntu/+source/libpam-mount/+bug/574329>. I witnessed GDM crashes sporadically with LDAP users in my network as well. But LDAP users---unlike local users which do not have the bug---were also using pam_mount. When I replaced pam_mount with autofs, I no longer had problems with GDM crashing. I can confirm this is a new bug in Lucid which did not exist in Karmic, as I had the same networked users setup in both.

Jeff Strunk (jstrunk-math) wrote :

We are not using pam_mount. We use autofs.

Jeff Strunk (jstrunk-math) wrote :

This bug still exists in the maverick beta. I am working on getting another debug stacktrace from my maverick test system.

I did not modify the gdm configuration in any way on maverick.

Jeff Strunk (jstrunk-math) wrote :

For anyone who needs a workaround, I discovered a legacy gdm package in karmic that still works in lucid. http://packages.ubuntu.com/karmic/gdm-2.20

It is unmaintained though, so this bug still needs to get fixed.

Ran into this bug today. Will try the workaround until this gets patched.

Adam Wolfe (awolfe-speedfc) wrote :

I've been fighting with this bug for the past few months as well.
I find it very curious that it is, predictably, every -other- login attempt.

I've tried the work-around using 2.20, but that gave me the "low graphics mode" runaround. I'll fiddle with it some more later... Or possibly switch to another DM that is LDAP capable for the time being.

Jeff Strunk (jstrunk-math) wrote :

Our new workaround has been to use SSSD instead of nscd and PADL's pam-ldap and nss-ldap.

It is important to use a newer version that what is in the Maverick repo. My PPA has a newer version. https://launchpad.net/~jstrunk-math/+archive/ppa

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers