Eric Biggers <[email protected]> wrote:

> On Thu, Nov 06, 2025 at 05:44:46PM +0000, David Howells wrote:
> > The interface to this code is through the crypto_sig API as the PKCS#7 code
> > wants to use that rather than calling it directly.  As such, I've placed it
> > in crypto/ rather than lib/crypto/.  Only the verification hooks are
> > implemented; the signing hooks return an error.
> 
> As I mentioned before
> (https://lore.kernel.org/linux-crypto/20250613170456.GA1284@sol/), this
> code should go in lib/crypto/.  There seems to be a clean API in
> crypto/ml_dsa/dilithium.h already.  Just make that the library API.

Fine.  Is it ever likely to be used directly, I wonder?

There is a downside of moving stuff to lib/crypto/: dynamically loadable
algorithms now need two modules instead of one.  For initial module signing,
granted, the algorithms for that must be built in.

> If "crypto_sig" support is really needed too, then put that in
> crypto/ml-dsa.c, built on top of the library API.  It's not clear the
> crypto_sig support is very useful, though.  For one, you had to add
> ML-DSA specific logic to the calling code anyway (see "pkcs7: Allow the
> signing algo to calculate the digest itself").

The actual signature check still goes through the same code path as everything
else, so crypto_sig will remain the API.  Otherwise I have to basically
reimplement crypto_sig inside crypto/asymmetric_keys/, including the on-demand
algorithm loading.

David


Reply via email to