Re: [PATCH resend] audit: fix mark refcounting

[Linus Torvalds <torvalds@linux-foundation.org>]

On Thu, Dec 15, 2011 at 12:40 AM, Al Viro <viro@zeniv.linux.org.uk> wrote:
>>
>> In particular, a quick grep shows that there are destroy_mark users still in:
>>
>>  - fs/notify/fanotify/fanotify_user.c
>>
>>  - fs/notify/dnotify/dnotify.c (2 of them)
>>
>>  - fs/notify/inotify/inotify_fsnotify.c
>>
>> that don't do "put_mark()" after the destroy. Why is it ok there?
>
> Um?  dnotify has fsnotify_put_mark() called in both cases...

Ok, that didn't show up in my grep, the "put_mark()" was more than
three lines away in the other case. As mentioned, I simply grepped
without looking at much context at all.

> I don't like it; it's called from ->handle_event() and parent->mark is
> exactly the inode_mark argument of that method.  It ought to be pinned
> by caller.  In other places we *do* need get/put around that destroy
> and we generally do that.

Presumably *parent* is pinned by caller, but not ->mark. So when the
parent directory is deleted, the parent data structure stays around,
but mark is cleanred, and you get the oops that was reported. See the
simple two-liner example to trigger it. I didn't test it myself, but
it looks obvious.

                             Linus