Re: VFS deadlock ?

[Linus Torvalds]

On Thu, Mar 21, 2013 at 2:02 PM, Dave Jones <davej@xxxxxxxxxx> wrote:
>
> here we go...
>
> WARNING: at fs/namei.c:2335 lock_rename+0x156/0x160()
> p1=irda p2=irda

Ok, good. I ssupect it's /proc or /sys, we do have that entry there.

But in fact I suspect we do want the parent name after all, because I
think we have multiple "irda" directories. There's the one in
/proc/net/ (added by net/irda/irproc.c), and there's a sysctl  CTL_DIR
"irda" directory (kernel/sysctl_binary.c). And there might even be a
few other ones in /sys too, thanks to the ldisc called "irda" etc.

I don't see where the shared inode comes from, but I suspect that
would be easier to guess if we actually see which particular case it
ends up being..

> followed by...
> =====================================
> [ BUG: bad unlock balance detected! ]

Oh, ok, that's just because the unlock path doesn't have the same
logic for unlocking identical inodes that the thing just added to the
locking path. You'd need to add a check for "same inode" and only
unlock it once.

So that was my fault in asking for a non-BUG_ON and not doing the
complete thing. See "unlock_rename()" - you'd need to change the "p1
!= p2" test there to "p1->d_inode != p2->d_inode" there to match the
logic in lock_rename()

            Linus
--
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/