On Thu, Sep 13, 2018 at 9:46 AM, Kees Cook [off-list ref] wrote:

On Mon, Sep 10, 2018 at 10:52 PM, Herbert Xu
[off-list ref] wrote:

quoted

On Fri, Sep 07, 2018 at 08:56:23AM +0200, Ard Biesheuvel wrote:

quoted

OK, so given that all SKCIPHER_REQUEST_ON_STACK occurrences are
updated in this series anyway, perhaps we should add
skcipher_[en|de]crypt_onstack() flavors that encapsulate the
additional check? Only question is how to enforce at compile time that
those are used instead of the ordinary ones when using a stack
allocated request. Would you mind using some macro foo here involving
__builtin_types_compatible_p() ?

Something like a completely new type which in reality is just a
wrapper around skcipher:

        struct crypto_sync_skcipher {
                struct crypto_skcipher base;
        } tfm;

        tfm = crypto_alloc_sync_skcipher(...);

        crypto_sync_skcipher_encrypt(...)
        crypto_sync_skcipher_decrypt(...)

These functions would just be trivial inline functions around their
crypto_skcipher counterparts.

This means new wrappers for the other helpers too, yes? For example:

        SKCIPHER_REQUEST_ON_STACK(nreq, ctx->null);

        skcipher_request_set_tfm(nreq, ctx->null);
        skcipher_request_set_callback(nreq, req->base.flags, NULL, NULL);
        skcipher_request_set_crypt(nreq, req->src, req->dst, nbytes, NULL);

        return crypto_skcipher_encrypt(nreq);

For the above, we'd also need:

sync_skcipher_request_set_tfm()
sync_skcipher_request_set_callback()
sync_skcipher_request_set_crypt()

Wait, I think I misunderstood you. Did you mean a new top-level thing
(tfm?) not a new request type?

That would mean at least replacing skcipher_request_set_tfm() with a
wrapper (since the tfm argument is different), but _not_
encrypt/decrypt like you mention. I could perform a type test in
SKCIPHER_REQUEST_ON_STACK().

Globally:
- add struct crypto_sync_skcipher wrapper
- add crypto_alloc_sync_skcipher() to check non-ASYNC and request size
of actual tfm
- add skcipher_request_set_sync_tfm() to attach the wrapped tfm to the request
- SKCIPHER_REQUEST_ON_STACK() would verify the tfm was a struct
crypto_sync_skcipher.

Two changes per user:
- change allocation to use new crypto_alloc_sync_skcipher() which does
the runtime checking
- change skcipher_request_set_tfm() to skcipher_request_set_sync_tfm()

This means struct skcipher_request is unchanged, along with
_set_callback, _set_crypt, _zero, and en/decrypt.

API misuse would be caught at build-time (via
SKCIPHER_REQUEST_ON_STACK type checking) and any request size problems
would be caught at allocation time.

Does this sound like what you had in mind?

-Kees

-- 
Kees Cook
Pixel Security