Re: OOPS: reproducible in vfs_follow_link 2.4.9,2.4.10-pre4

Alexander Viro (viro@math.psu.edu)
Thu, 6 Sep 2001 15:10:34 -0400 (EDT)


On Thu, 6 Sep 2001, Giacomo Catenazzi wrote:

> Hello.
>
> Since yesterdey, every time I run a 2.4.9 or 2.4.10pre-4 without the
> "devfs=nomount" I
> have two oops + /usr, /home /boot not mounted (all (also /): ext2).

Don't use devfs. One of the known bugs - devfs passes a string to
vfs_follow_link() and doesn't care to preserve it until vfs_follow_link()
is done.

E.g. rmmod during the symlink traversal will end up with

vfs_follow_link(nd, s)
[working]
[blocked on IO/allocation/whatever]
rmmod
...
kfree(s)
...
[the first process wakes up and oopses on attempt to dereference s]

There are other scenarios that end up freeing a string passed to
vfs_follow_link() (or, for that matter, being busily copied to userland by
readlink(2)). Basically, if devfs decides that symlink is gone - pray.
If it's being accessed right now you are going to end up with oops.

No idea why you've started triggering it only now - not enough
details to say.

And yes, Richard had been informed about that months ago. Sigh...

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/