Bug#290916: was it problem of directories being umounted unexpectedly?

[Václav Šmilauer]

Since I fixed a bug in a PreSession script supposed to mount thin
client's local media to user's home subdir, it happens MUCH less
frequently (say once in a week). I still suggest not downgrading this
bug to normal as (1) it still happens without known reason (2) gdm
should handle such conditions gracefully anyway, even if it's someone
else's fault. Detailed description follows.

On every login, we mount, as said, floppy, CDROM and flash from the thin
client (running XDMCP session @ PXES Linux) into
~/Media/{floppy,cdrom,flash}. The script also (lazily) umounts media of
users who are not logged in any more (gone etc). However, bug in the
script umounted EVERYBODY else's media on EVERY login (except the user
who was logging in), so contained files could disappear during the
session unexpectedly. This is fixed now.

Given the correlation that it happens much less since the fix, I think
that the reason for zombie processes is that some process had files on
the local media open (might be nautilus, for example) and when they
disappeared, was waiting for them (or was possibly IO blocked?)
infinitely. Gdm not being able to terminate all child processes of that
particular session got stuck and was sending SIGTERM, but without any
effect - whence the ``whack_old_slave'' message repeating in the log
until intervention.

OK, now does this make sense? Someone with better knowledge of gdm
architecture make the judgement.

Thanks to the maintainer and programmers of gdm for the (modulo bugs)
relatively great piece of software - Vaclav Smilauer

(P.S. Justification of ``renders package unusable'' was justified as
for my network gdm was barely usable at that time. The argument ``I have
it here and it works fine'' is irrelevant within BTS IMHO.)