RE: flock/posix lock question

[Stuart_Hayes]

Hayes, Stuart wrote:
> Hello--
> 
> This question regards file locks and code in fs/locks.c.
> 
> I am seeing the kernel get stuck in posix_locks_deadlock() checking
> for deadlocks.  It appears that samba (smbd) is getting both an flock
> and a posix lock for the same file, which results in a circular
> dependency in the blocked_list... and posix_locks_deadlock() is
getting
> stuck in that circle.    
> The circular dependency gets into the blocked_list because deadlock
> situations aren't checked for when inserting flock requests into the
> blocked_list.  
> 
> Since flocks aren't supposed to do any deadlock checking, it seems
> like the right solution to this would be to modify
> posix_locks_deadlock() to only check for deadlock situations with
> other posix locks and lock requests, and ignore flocks.  Of course,
> samba should also probably be fixed so that it doesn't do that, too...
> but it shouldn't be able to cause a kernel hang by doing so.     
> 
> Does this sound correct?  Am I missing something?
> 
> (I am seeing this on a RHEL3 update 3 (2.4.21-20) kernel.)
> 
> Thanks!
> Stuart

Here's a patch (untested) just to illustrate what I think needs to be
done:

--- locks.c	Wed Apr 21 23:09:18 2004
+++ locks.c.new	Fri Sep 10 17:15:04 2004
@@ -685,7 +685,8 @@ next_task:
 		return 1;
 	list_for_each(tmp, &blocked_list) {
 		struct file_lock *fl = list_entry(tmp, struct file_lock,
fl_link);
-		if ((fl->fl_owner == blocked_owner)
+		if ( (fl->fl_flags & FL_POSIX)
+		    && (fl->fl_owner == blocked_owner)
 		    && (fl->fl_pid == blocked_pid)) {
 			fl = fl->fl_next;
 			blocked_owner = fl->fl_owner;


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