svn commit: r361583 - head/sys/crypto/aesni

[Kornel Dulęba]

On Thu, May 28, 2020 at 8:30 PM John Baldwin <jhb at freebsd.org> wrote:

> On 5/28/20 2:13 AM, Marcin Wojtas wrote:
> > Author: mw
> > Date: Thu May 28 09:13:20 2020
> > New Revision: 361583
> > URL: https://svnweb.freebsd.org/changeset/base/361583
> >
> > Log:
> >   Change return types of hash update functions in SHA-NI
> >
> >   r359374 introduced crypto_apply function which takes as argument a
> function pointer
> >   that is expected to return an int, however aesni hash update functions
> >   return void.
> >   Because of that the function pointer passed was simply cast with
> >   its return value changed.
> >   This resulted in undefined behavior, in particular when mbuf is used,
> (ipsec)
> >   m_apply checks return value of function pointer passed to it
> >   and in our case bogusly fails after calculating hash of the first mbuf
> >   in chain.
> >   Fix it by changing signatures of sha update routines in aesni and
> >   dropping the casts.
>
> Hmm, I missed one nit in the review.  r359374 didn't introduce
> crypto_apply, it just changed some of the arguments arguments (crp
> instead of crp_buf and crp_flags).  This fix needs to be MFC'd to 12
> as well since the issue with the return type is also present there.
>
> --
> John Baldwin
>

Whoops, I should have read the diff more carefully.
Before debugging I did some bisecting and found r359374 as the culprit.
In r359374 return types of intel_sha256_update and intel_sha1_update were
changed to void which is the root cause.
SHA-NI on stable/12 should work just fine right now.
Sorry for the noise.