Re: [PATCH v2 11/11] crypto: skcipher: Remove VLA usage for SKCIPHER_REQUEST_ON_STACK

[Herbert Xu]

On Wed, Jun 27, 2018 at 11:31:09AM -0700, Kees Cook wrote:
>
> I might be catching on... so from this list, I should only "count" the
> synchronous ones as being wrappable? The skcipher list is actually
> pretty short:
> 
> crypto/cryptd.c:        crypto_skcipher_set_reqsize(
> crypto/cryptd.c-                tfm, sizeof(struct
> cryptd_skcipher_request_ctx));

cryptd is async so you don't have to include it.
 
> These are "simple" wrappers:
> 
> crypto/lrw.c:   crypto_skcipher_set_reqsize(tfm,
> crypto_skcipher_reqsize(cipher) +
> crypto/lrw.c-                                    sizeof(struct rctx));
> 
> crypto/simd.c-  reqsize = sizeof(struct skcipher_request);
> crypto/simd.c-  reqsize += crypto_skcipher_reqsize(&cryptd_tfm->base);
> crypto/simd.c:  crypto_skcipher_set_reqsize(tfm, reqsize);

simd is async.

> crypto/xts.c:   crypto_skcipher_set_reqsize(tfm,
> crypto_skcipher_reqsize(child) +
> crypto/xts.c-                                    sizeof(struct rctx));
> 
> But what are the "legitimate" existing crypto_skcipher_reqsize() values here?
> 
> These are "complex" wrappers, with cts even adding blocksize to the mix...
> 
> crypto/ctr.c-   align = crypto_skcipher_alignmask(tfm);
> crypto/ctr.c-   align &= ~(crypto_tfm_ctx_alignment() - 1);
> crypto/ctr.c-   reqsize = align + sizeof(struct crypto_rfc3686_req_ctx) +
> crypto/ctr.c-             crypto_skcipher_reqsize(cipher);
> crypto/ctr.c:   crypto_skcipher_set_reqsize(tfm, reqsize);
> 
> crypto/cts.c-   align = crypto_skcipher_alignmask(tfm);
> crypto/cts.c-   bsize = crypto_skcipher_blocksize(cipher);
> crypto/cts.c-   reqsize = ALIGN(sizeof(struct crypto_cts_reqctx) +
> crypto/cts.c-                   crypto_skcipher_reqsize(cipher),
> crypto/cts.c-                   crypto_tfm_ctx_alignment()) +
> crypto/cts.c-             (align & ~(crypto_tfm_ctx_alignment() - 1)) + bsize;
> crypto/cts.c-
> crypto/cts.c:   crypto_skcipher_set_reqsize(tfm, reqsize);
> 
> What values might be expected here? It seems the entire blocksize
> needs to be included as well...

But otherwise yes these are the ones that count.

Cheers,
-- 
Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt