[gdm-list] Re: PostSession script support fades out from 2.4 to 2.6?

[Brian Cameron <Brian Cameron Sun COM>]

Jerry:

Thank you so much for doing this research.  I think that you are right that
your "ignore xioerror" patch is more deterministic and a better way to fix
the problem.

Would you mind making sure that your original patch still
applies to GDM2 CVS head and I will go ahead and apply it to gdm2.  I
think I will also apply the patch to the 2.12 branch and make a new
GDM 2.8.0.4 release with this fix so that this serious issue can finally
be resolved in a stable release.  Also there are some other nice fixes
that have gone into the 2.12 branch that would be nice to include
(the dwell gesture listener had a bug in it that caused it to never
work properly and I fixed a nasty UI problem with Timed/Automatic
login in gdmsetup for example).

I think George's comments to bug 126071 that the PAM issue is better
fixed in the PAM module asynchronously still stands and that bug 126071
isn't really a GDM issue.  Bug 126071 was closed in mid-2003 and it doesn't
seem that anybody has complained about that problem since then, so I
don't think this is an issue that is affecting many people.

Sound reasonable?  And thanks again.

Brian

> Commenting out the fix for bug 126071 leads to the following interesting behavior:
>
> The PostSession script does get called.
> You still get an xioerror, but it happens *after* the PostSession is complete.
>
> Sep 15 10:17:11 pug logger: You are in the PostSession script.
> Sep 15 10:17:11 pug gdm(pam_unix)[2576]: session closed for user jgd
> Sep 15 10:17:11 pug gdm[2576]: gdm_slave_xioerror_handler: Fatal X error - Restarting :0
>
> Test procedure
> 1) Put '/usr/bin/logger "You are in the PostSession script."' in /etc/X11/gdm/PostSession/Default
> 2) Turn on debug in gdm.conf
> 3) Install revised gdm with 126071 fix removed (and my workaround)
> 4) reboot
> 5) log in
> 6) ctl-alt-backspace
> 7) log back in and inspect the syslog (see above).
>
> I think, but I'm not certain, that the xioerror is now triggered later in slave.c, when gdm_verify_cleanup(d) is called.  So, you still
> get the xioerror signal, but it happens later.
>
> So, my "ignore xioerror" hack lets you trigger the xioerror before PostSession, and ignore it. Commenting out whack clients
> makes the xioerror happen after PostSession (provided nothing else accessed the display structure while we were getting there).
> I think mine's a little more deterministic (shorter race window), but I don't like it that much.
>
> We definitely need to find a different repair to the open filehandle problem.
>
> As for what to do w.r.t. the current source, I agree with you that not running PostSession is *much* worse than leaving an open
> filehandle that keeps an automounter from releasing a home directory.  I have no idea how to fix the interaction with pam.
>
> HTH
> J