Re: [btrfs] is vulnerable to a hash-DoS attack

[Chris Mason <chris.mason@fusionio.com>]

On Thu, Dec 13, 2012 at 03:07:27PM -0700, Chris Mason wrote:
> On Thu, Dec 13, 2012 at 02:34:30PM -0700, David Sterba wrote:
> > On Thu, Dec 13, 2012 at 03:52:08PM -0500, Chris Mason wrote:
> > > Thanks for taking the time to write this up.  As far as I can tell, the
> > > looping was actually fixed in an older kernel and I just misread our
> > > version string in your original email.
> > 
> > Yeah, the blogpost says 3.3.7. I did a quick test with 3.7 and was not
> > able to reproduce it.
> 
> I tried with 3.3 and every step between 3.3 and 3.7.  I'm not able to
> reproduce the problem, and I did run with Hack=True in the script
> (thanks for the flag btw, I really like that).
> 
> So, that leaves us with a few possibilities:
> 
> 1) mount -o seclabel
> 2) The small size of the device
> 3) loopback
> 
> I ran with a 1GB FS here on 3.3 and wasn't able to trigger things.  But
> Pascal, could you please help narrow the problem down?

Ok, I think what was happening on Pascal's machine is that we are
returning -EOVERFLOW and sometimes aborting the transaction (which turns
the FS readonly).

I have that fixed here.  Rename I'm reodering the rename code a little
bit to also handle the case where there are collisions on rename.

-chris