Red Hat Bugzilla – Bug 59936
Last modified: 2007-04-18 12:40:26 EDT
Description of Problem:
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Keep logging in as different users
/dev entries that are supposed to get their permissions changed by console.perms do not get their permissions changed after a while
Ownership of selected /dev entries should be given to the first console user who logs in, /every time/, not just the first couple.
After logging in many different times as different users on various local ttys, the permissions on, say, /dev/dsp stop changing from user to user, for whoever happens to be the first user logged in at the console. The user name I use, 'nix', is the one most often used, since I'm the most frequent user of the computer in question. After logging in many times on various local ttys (X and otherwise), ownership of /dev files such as /dev/dsp remains mine even after I log out, the point at which it should revert to 600 root.root .
I know this because, when I log in remotely as root (a remote login will not trigger a change in permissions, nor a change in ownership), and do ls -l /dev/dsp, I find that the owner is still 'nix', instead of 600 root.root.
I tried removing /var/run/console.lock, followed by pam_console_apply. I also tried pam_console_apply -r. Running these worked (ownerships looked OK), but it didn't get the ownerships and permissions automagically changing again. Now, if someone wants to play, say, Tux Racer, I have to log in as root and explicitly give them permission to use /dev/dsp.
Thanks for your help.
I think I may have found both the cause and the resolution of the bug.
Cause: When I log in as any user via X, I get console ownership and various other goodies. When I kill X brutally (C-A-Backspace) instead of gracefully, Bad Things happen:
- /etc/X11/xdm/TakeConsole doesn't get executed
- /var/run/console.lock doesn't get removed
- /var/run/console now contains a file called 'nix' that contains 1 byte:'1'
Fix (as root):
Consequently, I have been able to completely and consistently reproduce the behaviour:
1. Log in remotely as root.
2. useradd -g users nix1
3. passwd nix1
4. Log into X via gdm locally as nix1
5. C-A-Backspace (kills X)
That's it, the ownership auto-assignment cycle is now broken, until (as mentioned):
[root@...]# rm /var/run/console.lock
[root@...]# rm /var/console/nix1
[root@...]# pam_console_apply -r
The crucial step seems to be
Would it be possible/wise to run these steps from someplace where the exit status of X is within the scope of the script (Checkable via $? or some such). By this I mean that you should include this kind of 'X-has-crashed-so-if-the-user-had-the-ownerships-then-fix-it-so-the-automagic-ownership-reassignment-works-again' check somewhere in the X/[xkg]dm-related scripts.
Are these steps enough ? Am I missing something due to the fact that I'm not thoroughly familiar with the workings of these ownership assignments ?
Thanks for your support.
(Using RH 7.2, pam-0.72-19, gdm-126.96.36.199-20)
AFAICT this is caused by somewhat of a problem in pam_console:
When you kill X with Ctrl-Alt-Backspace, the socket /tmp/.X11-unix/X0 (or X1, or
...) is removed.
Then, when pam_console.c:pam_sm_close_session is called (in my case from gdm's
gdm_slave_session_stop,gdm_verify_cleanup), it checks whether the 'tty' the user
was logged in to (ie. :0, :1, etc...) actually exists. In the X display case,
this translates to the existance of the mentioned /tmp/.X11-unix/X? file. This
check now clearly fails, and so the console use-count isn't decremented and
console.lock isn't removed, causing the problems mentioned in the original
To recap, the code path that causes this specific problem is:
pam_session.c:pam_close_session(), calls (through dispatch and all that)
config.y:check_console_name(), which checks the existance of /tmp/.X11-unix/X?
*** Bug 71468 has been marked as a duplicate of this bug. ***
It looks like this works ok in RHL 9 at least. I'm using gdm and I see
/tmp/X11-unix/X0 after killing X with ctrl+alt+backspace.