On Mon, Mar 03, 2014 at 04:48:29PM -0600, Eric Sandeen wrote:
> On 3/3/14, 4:36 PM, Dave Chinner wrote:
> > On Mon, Mar 03, 2014 at 02:41:54PM -0600, Eric Sandeen wrote:
> >> Commit "3a19fb7 libxfs: stop caching inode structures"
> >> introduced a use after free.
> >>
> >> libxfs_iput() already does the check for ip->i_itemp, and a
> >> kmem_zone_free() if it's present, and then frees the ip pointer.
> >> Re-checking ip->i_itemp after the libxfs_iput call will access
> >> the freed ip pointer, as will setting ip_>i_itemp to NULL.
> >>
> >> Simply remove the offending code to fix this up.
> >
> > which leaves the rest of the ili_done: code looking a little
> > strange.
> >
> > can you convert that now to be:
> >
> > ili_done:
> > if (iip->ili_lock_flags) {
> > iip->ili_lock_flags = 0;
> > return;
> > }
> > /* free the inode */
> > libxfs_iput(ip, 0);
> > }
>
> yeah, I actually had that first. Not sure why I didn't go with it ;)
>
> (Still looks strange to my untrained eye; "if lock flags are set, unset them
> and don't free the inode, otherwise free it")
If the lock falgs are set, it means the caller expects the
transaction commit to return the inode to it in a locked state.
Exactly the same semantics as the kernel code - the lock flags track
how the transaction releases the inode...
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
|