On Mon, 2021-05-17 at 08:57 +0200, Pavel Machek wrote: > From: Paolo Abeni <pabeni@xxxxxxxxxx> > > > > > > > > [ Upstream commit 78352f73dc5047f3f744764cc45912498c52f3c9 ] > > > > > > > > Currently the UDP protocol delivers GSO_FRAGLIST packets to > > > > the sockets without the expected segmentation. > > > > > > > > This change addresses the issue introducing and maintaining > > > > a couple of new fields to explicitly accept SKB_GSO_UDP_L4 > > > > or GSO_FRAGLIST packets. Additionally updates udp_unexpected_gso() > > > > accordingly. > > > > > > > > UDP sockets enabling UDP_GRO stil keep accept_udp_fraglist > > > > zeroed. > > > > > > What is going on here? accept_udp_fraglist variable is read-only. > > > > Thank you for checking this! > > > > The 'accept_udp_fraglist' field is implicitly initilized to zero at UDP > > socket allocation time (done by sk_alloc). > > > > So this patch effectively force segmentation of SKB_GSO_FRAGLIST > > packets via the udp_unexpected_gso() helper. > > > > We introduce the above field instead of unconditionally > > segmenting SKB_GSO_FRAGLIST, because the next patch will use it (to > > avoid unneeded segmentation for performance's sake for UDP tunnel), as > > you noted. > > Ok, but there's no follow up patch queued for 5.10...? Do we still > need it there? Patch d18931a92a0b5feddd8a39d097b90ae2867db02f is a performance improvement, I think not worthy/suitable for stable. For 5.10 78352f73dc5047f3f744764cc45912498c52f3c9 should be enough. Thanks! Paolo
