Re: [PATCH] nvmet-tcp: switch to using the crc32c library

[Sagi Grimberg]

On 27/02/2025 9:26, Hannes Reinecke wrote:
> On 2/26/25 20:01, Eric Biggers wrote:
> > On Wed, Feb 26, 2025 at 10:37:55AM +0100, Hannes Reinecke wrote:
> > > ... and it also eliminates a sporadic crash which we've seen
> > > where 'snd_hash' wasn't initialized when sending PDUs.
> > > Thanks for doing this!
> >
> > I'm not sure how that could have happened, since the ahash was 
> > allocated when
> > 'if (queue->hdr_digest || queue->data_digest)' which seemed to match the
> > conditions for when it was used.  But yeah, it's certainly nice to 
> > not have the
> > pointless allocation to worry about.
> >
> > > (Note to self: check the nvme-tls code for crc32c usage ...)
> >
> > I have patches for nvme-tls almost ready too.  Just been taking my 
> > time since
> > I've been updating all other users of "crc32" and "crc32c" in the 
> > kernel too.
> > And I need to decide what to do about skb_copy_and_hash_datagram_iter().
> If it were me I would _love_ to switch the nvme-tcp recv patch over to 
> recvmsg and kill the ->read_sock() implementation.
> ->read_sock uses a completely different codepath in tls_sw, and nvme is
> the only user of that. So there's a fair chance that we might miss any
> improvements or fixes.
>
> Plus we currently have no good way of handling TLS records from 
> ->read_sock(), which is something we might want to do in the future.
>
> So if we had an equivalent for skb_copy_and_hash_iter() for recvmsg()
> I could revisit my original patchset and work on getting ->read_sock()
> replaced.

Indeed, we want to calculate rolling crc32c as we copy the data vs. 
doing this afterwards...