[5064] in Athena Bugs
3 ways to panic AFS
daemon@ATHENA.MIT.EDU (Jeffrey I. Schiller)
Mon Jun 4 14:53:49 1990
Date: Mon, 4 Jun 90 14:53:16 -0400
From: Jeffrey I. Schiller <jis@MIT.EDU>
To: jfc@ATHENA.MIT.EDU
Cc: bugs@ATHENA.MIT.EDU, bug-afs@ATHENA.MIT.EDU
In-Reply-To: John Carr's message of Mon, 04 Jun 90 10:44:28 EDT <9006041444.AA06322@Achates.MIT.EDU>
Date: Mon, 04 Jun 90 10:44:28 EDT
From: John Carr <jfc@ATHENA.MIT.EDU>
All these are for 7.0, but the last 2 at least have always been problems.
# cd /afs
# umount /afs
# ls
the kernel responds:
panic: afsglobalvfs
The AFS filesystem is not considered busy no matter how many file
descriptors or texts are using it. The unmount call should check this.
I agree.
# rm /usr/vice/cache/<random file name>
eventually the kernel will (1) panic because it fails to open the cache file
or (2) clobber the contents of whatever file reused the cache file's inode
number.
Foo. You shouldn't be deleting files in that directory. Heck I can
crash UNIX by doing "rm -rf /etc" (or at least make it as good as
crashed). Is this a bug, or should I prevent the deletion (even for
root) of files in /etc. Clearly not. I think that it is reasonable
that deleting files in /usr/vice/cache causes a crash, as long as only
the root can do so (which is the case). If fixing this problem is a
low energy effort (or doesn't require making the kernel bigger to add
the necessary error recovery code) then go ahead and fix it. However
my opinion is that it isn't worth much work.
# umount /usr/vice/cache (or the filesystem containing this directory)
very quickly the kernel will panic because it failed to open a cache file.
This is a similar problem to the first: there is nothing marking the
filesystem busy. It would be sufficient to hold open the file with the
cache map.
Hmmm. I have used this "feature." On my workstation I have
/usr/vice/cache on a large partition. I can gracefully shutdown my
workstation and then 'umount' that partition (and I don't do anything
to tickle AFS afterwards). I then don't need to fsck it when I
(gracefully) reboot. If it were to keep a file open, then I would lose
this "feature", at least until someone adds the graceful shutdown code
to AFS that is clearly missing. Now that would be a useful task for a
kernel wizard (though probably best for a wizard at Transarc). Perhaps
we should raise this at the user meeting.
-Jeff
