Date:   Wed, 10 Apr 2019 00:02:35 -0700
From:   Eric Biggers <ebiggers@kernel.org>
To:     Daniel Axtens <dja@axtens.net>
Cc:     omosnacek@gmail.com, linux-crypto@vger.kernel.org,
        Herbert Xu <herbert@gondor.apana.org.au>,
        marcelo.cerri@canonical.com, Stephan Mueller <smueller@chronox.de>,
        leo.barbosa@canonical.com, linuxppc-dev@lists.ozlabs.org,
        nayna@linux.ibm.com, pfsmorigo@gmail.com, leitao@debian.org
Subject: Re: [PATCH] crypto: vmx - fix copy-paste error in CTR mode
Message-ID: <20190410070234.GA12406@sol.localdomain>
References: <20190315020901.16509-1-dja@axtens.net>
 <20190315022414.GA1671@sol.localdomain>
 <875zsku5mk.fsf@dja-thinkpad.axtens.net>
 <20190315043433.GC1671@sol.localdomain>
 <8736nou2x5.fsf@dja-thinkpad.axtens.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <8736nou2x5.fsf@dja-thinkpad.axtens.net>
User-Agent: Mutt/1.11.4 (2019-03-13)
Sender: linux-crypto-owner@vger.kernel.org
Precedence: bulk

Hi Daniel,

On Fri, Mar 15, 2019 at 04:23:02PM +1100, Daniel Axtens wrote:
> Eric Biggers <ebiggers@kernel.org> writes:
> 
> > Hi Daniel,
> >
> > On Fri, Mar 15, 2019 at 03:24:35PM +1100, Daniel Axtens wrote:
> >> Hi Eric,
> >> 
> >> >> The original assembly imported from OpenSSL has two copy-paste
> >> >> errors in handling CTR mode. When dealing with a 2 or 3 block tail,
> >> >> the code branches to the CBC decryption exit path, rather than to
> >> >> the CTR exit path.
> >> >
> >> > So does this need to be fixed in OpenSSL too?
> >> 
> >> Yes, I'm getting in touch with some people internally (at IBM) about
> >> doing that.
> >> 
> >> >> This leads to corruption of the IV, which leads to subsequent blocks
> >> >> being corrupted.
> >> >> 
> >> >> This can be detected with libkcapi test suite, which is available at
> >> >> https://github.com/smuellerDD/libkcapi
> >> >> 
> >> >
> >> > Is this also detected by the kernel's crypto self-tests, and if not why not?
> >> > What about with the new option CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y?
> >> 
> >> It seems the self-tests do not catch it. To catch it, there has to be a
> >> test where the blkcipher_walk creates a walk.nbytes such that
> >> [(the number of AES blocks) mod 8] is either 2 or 3. This happens with
> >> AF_ALG pretty frequently, but when I booted with self-tests it only hit
> >> 1, 4, 5, 6 and 7 - it missed 0, 2 and 3.
> >> 
> >> I don't have the EXTRA_TESTS option - I'm testing with 5.0-rc6. Is it in
> >> -next?
> >> 
> >> Regards,
> >> Daniel
> >
> > The improvements I recently made to the self-tests are intended to catch exactly
> > this sort of bug.  They were just merged for v5.1, so try the latest mainline.
> > This almost certainly would be caught by EXTRA_TESTS (and if not I'd want to
> > know), but it may be caught by the regular self-tests now too.
> 
> Well, even the patched code fails with the new self-tests, so clearly
> they're catching something! I'll investigate in more detail next week.
> 
> Regards,
> Daniel
> 
> >
> > - Eric

Are you still planning to fix the remaining bug?  I booted a ppc64le VM, and I
see the same test failure (I think) you were referring to:

alg: skcipher: p8_aes_ctr encryption test failed (wrong result) on test vector 3, cfg="uneven misaligned splits, may sleep"

- Eric