On Sun, Jun 02, 2024 at 05:30:33PM +0200, Florian Westphal wrote: > Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote: > > Yes, this is implicit case: skb is gso. This is adding an exception > > for sctp-gso. > > Is that because sctp gso carries a new / distinct sctp header > per frag? > > If so, we can't extend skb_zerocopy, as it would create > a bogus packet. > > The only "speedup" we could do in that case is that we > do not need to fix the header csum if userspace enabled > GSO support as long as we indicate csum correctness via > the NFQA_SKB_CSUMNOTREADY flag. > > But that should probably be done at a later time. I have posted a new series: https://patchwork.ozlabs.org/project/netfilter-devel/list/?series=410326 SCTP packets enter skb_gso_segment() before being sent to userspace not to break semantics. I have revisited and extended the tests to exercise GSO path from the output path.
