Re: [PATCH net v2 3/4] udp: do not transition UDP fraglist to unnecessary checksum

[Antoine Tenart <atenart@kernel.org>]

Quoting Willem de Bruijn (2024-03-21 19:13:35)
> Antoine Tenart wrote:
> > Quoting Willem de Bruijn (2024-03-21 15:58:17)
> > > Antoine Tenart wrote:
> > > > 
> > > > If I sum up our discussion CHECKSUM_NONE conversion is wanted,
> > > > CHECKSUM_UNNECESSARY conversion is a no-op and CHECKSUM_PARTIAL
> > > > conversion breaks things. What about we just convert CHECKSUM_NONE to
> > > > CHECKSUM_UNNECESSARY?
> > > 
> > > CHECKSUM_NONE cannot be converted to CHECKSUM_UNNECESSARY in the
> > > receive path. Unless it is known to have been locally generated,
> > > this means that the packet has not been verified yet.
> > 
> > I'm not sure to follow, non-partial checksums are being verified by
> > skb_gro_checksum_validate_zero_check in udp4/6_gro_receive before ending
> > up in udp4/6_gro_complete. That's also probably what the original commit
> > msg refers to: "After validating the csum, we mark ip_summed as
> > CHECKSUM_UNNECESSARY for fraglist GRO packets".
> > 
> > With fraglist, the csum can then be converted to CHECKSUM_UNNECESSARY.
> 
> Oh yes, of course.
> 
> > Except for CHECKSUM_PARTIAL, as we discussed.
> 
> Because that is treated as equivalent to CHECKSUM_UNNECESSARY in the
> ingress path, and if forwarded to an egress path, csum_start and
> csum_off are set correctly. Also for all segs after segmentation.
> Okay, that sounds fine then.
> 
> There are two cases here: csum_start points to the outer header or it
> points to the inner header. I suppose that does not matter for
> correctness post segmentation.

Right. I'll send a v3 to keep the none -> unnecessary conversion and
fix build issue in patch 1.

Thanks for the review!
Antoine