Re: [bug] crypto/vmx/p8_ghash memory corruption in 4.8-rc7

From: Marcelo Cerri
Date: Wed Sep 28 2016 - 08:39:20 EST

Hi Hebert,

On Wed, Sep 28, 2016 at 08:29:35PM +0800, Herbert Xu wrote:
> On Wed, Sep 28, 2016 at 03:40:51AM -0400, Jan Stancek wrote:
> >
> > Thanks for clearing up how this works in padlock-sha, but
> > we are not exactly in same situation with p8_ghash.
> >
> > p8_ghash_init_tfm() already updates descsize. Problem in original report
> > is that without custom export/import/statesize p8_ghash_alg.statesize
> > gets initialized by shash_prepare_alg() to alg->descsize:
> Right.
> > so I think we need either:
> > 1) make sure p8_ghash_alg.descsize is correct before we register shash,
> > this is what Marcelo's last patch is doing
> This approach doesn't work because there is no guarantee that
> you'll get the same fallback the next time you allocate a tfm.
> So relying on the descsize being constant can only work if all
> implementations of the fallback use the same desc struct.

The patch forces ghash-generic as the fallback. And I don't think that
is a big problem if we decide to go by this path.
> > 2) provide custom export/import/statesize for p8_ghash_alg
> This works for padlock-sha because every implementation of SHA
> uses the same state data structure from sha.h. If we can make
> all implementations of ghash agree on the exported state then
> we can use the same approach.

That would be nice because it would allow p8_ghash to keep using a
dynamic fallback, but I'm not that is viable. What do you think?

> Otherwise we can go back to allocating just ghash-generic and
> also move its data structure into an exported header file.

That would make the fix much more simple and it wouldn't require to get
the fallback descsize at runtime.

> Cheers,
> --
> Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> Home Page:
> PGP Key:


Attachment: signature.asc
Description: PGP signature