Re: [dm-devel] [PATCH v5 05/11] crypto: ahash: Remove VLA usage

From: Eric Biggers <ebiggers3@gmail.com>
To: Kees Cook <keescook@chromium.org>
Cc: Herbert Xu <herbert@gondor.apana.org.au>,
	Giovanni Cabiddu <giovanni.cabiddu@intel.com>,
	Arnd Bergmann <arnd@arndb.de>,
	"Gustavo A. R. Silva" <gustavo@embeddedor.com>,
	Mike Snitzer <snitzer@redhat.com>,
	Eric Biggers <ebiggers@google.com>,
	qat-linux@intel.com, LKML <linux-kernel@vger.kernel.org>,
	dm-devel@redhat.com, linux-crypto <linux-crypto@vger.kernel.org>,
	Lars Persson <larper@axis.com>,
	Tim Chen <tim.c.chen@linux.intel.com>,
	Alasdair Kergon <agk@redhat.com>, Rabin Vincent <rabinv@axis.com>
Subject: Re: [dm-devel] [PATCH v5 05/11] crypto: ahash: Remove VLA usage
Date: Tue, 17 Jul 2018 16:12:09 -0700	[thread overview]
Message-ID: <20180717231209.GJ75957@gmail.com> (raw)
In-Reply-To: <CAGXu5jKKD4hzeFEdvY5QdahEr4POd9ZM_B_5Z_iU-Qmx3AYSfQ@mail.gmail.com>

On Tue, Jul 17, 2018 at 01:07:50PM -0700, Kees Cook wrote:
> On Tue, Jul 17, 2018 at 9:39 AM, Eric Biggers <ebiggers3@gmail.com> wrote:
> > On Mon, Jul 16, 2018 at 09:21:44PM -0700, Kees Cook wrote:
> >> In the quest to remove all stack VLA usage from the kernel[1], this
> >> introduces max size macros for ahash, as already done for shash, and
> >> adjust the crypto user to max state size.
> >>
> >> [1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com
> >>
> >> Signed-off-by: Kees Cook <keescook@chromium.org>
> >> ---
> >>  crypto/ahash.c        | 4 ++--
> >>  crypto/algif_hash.c   | 2 +-
> >>  include/crypto/hash.h | 3 +++
> >>  3 files changed, 6 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/crypto/ahash.c b/crypto/ahash.c
> >> index a64c143165b1..6435bdbe42fd 100644
> >> --- a/crypto/ahash.c
> >> +++ b/crypto/ahash.c
> >> @@ -550,8 +550,8 @@ static int ahash_prepare_alg(struct ahash_alg *alg)
> >>  {
> >>       struct crypto_alg *base = &alg->halg.base;
> >>
> >> -     if (alg->halg.digestsize > PAGE_SIZE / 8 ||
> >> -         alg->halg.statesize > PAGE_SIZE / 8 ||
> >> +     if (alg->halg.digestsize > AHASH_MAX_DIGESTSIZE ||
> >> +         alg->halg.statesize > AHASH_MAX_STATESIZE ||
> >>           alg->halg.statesize == 0)
> >>               return -EINVAL;
> >>
> >> diff --git a/crypto/algif_hash.c b/crypto/algif_hash.c
> >> index bfcf595fd8f9..8974ee8ebead 100644
> >> --- a/crypto/algif_hash.c
> >> +++ b/crypto/algif_hash.c
> >> @@ -239,7 +239,7 @@ static int hash_accept(struct socket *sock, struct socket *newsock, int flags,
> >>       struct alg_sock *ask = alg_sk(sk);
> >>       struct hash_ctx *ctx = ask->private;
> >>       struct ahash_request *req = &ctx->req;
> >> -     char state[crypto_ahash_statesize(crypto_ahash_reqtfm(req)) ? : 1];
> >> +     char state[AHASH_MAX_STATESIZE];
> >>       struct sock *sk2;
> >>       struct alg_sock *ask2;
> >>       struct hash_ctx *ctx2;
> >> diff --git a/include/crypto/hash.h b/include/crypto/hash.h
> >> index ae14cc0e0cdb..4fcd0e2368cd 100644
> >> --- a/include/crypto/hash.h
> >> +++ b/include/crypto/hash.h
> >> @@ -64,6 +64,9 @@ struct ahash_request {
> >>       void *__ctx[] CRYPTO_MINALIGN_ATTR;
> >>  };
> >>
> >> +#define AHASH_MAX_DIGESTSIZE 512
> >> +#define AHASH_MAX_STATESIZE  512
> >> +
> >
> > Why is AHASH_MAX_DIGESTSIZE (512) so much larger than SHASH_MAX_DIGESTSIZE (64)?
> > I would have expected them to be the same.
> 
> This was a direct replacement of the PAGE_SIZE / 8 original limit.
> This this didn't trip any frame size warnings, it seemed like we
> didn't need to do the more narrow limitations done elsewhere.
> 
> I am, of course, happy to do a manual review and find a lower
> alternative if that's desired.
> 

I just don't see why ahash algorithms would need such a huge maximum digest
size.  Don't the 'ahash' algorithms all have 'shash' equivalents too?  Is there
actually any hash algorithm, either shash or ahash, in the Linux kernel that has
a digest size greater than 64 bytes (512 bits)?  Note that for a real
cryptographic hash there isn't really any need for a digest size larger than
that, since that already gives you 256-bit collision resistance; that's why
SHA-2 and SHA-3 max out at that size.

- Eric