[PATCH net-next 0/4] net: resegment oversized TCP GSO skbs

BIG TCP is negotiated per netdevice, so BIG TCP and non-BIG TCP ports can
coexist in one path. When a BIG TCP packet goes to a port without BIG TCP
support, it loses its GSO state and the stack segments it into individual
MSS sized packets, so the port TSO is left unused.

This series cuts an oversized unencapsulated TCP GSO skb into GSO skbs
which fit the device limits instead of single segments, so the per-segment
work stays on the device TSO. On a veth -> bridge -> TAP -> guest
virtio-net path, with BIG TCP enabled on the veth endpoints and left off
in the guest, a single iperf3 TCP flow, six alternating runs per state
(-t 15 -O 5, fixed CPU affinity and port tuple):

  protocol  no BIG TCP   mixed, no reseg  mixed, resegmented
  TCP/IPv4  51.550 Gbps  15.850 Gbps      52.617 Gbps
  TCP/IPv6  52.050 Gbps  15.783 Gbps      51.933 Gbps

The middle column comes from the same kernel with the bounded path
disabled for the comparison. A BIG TCP hop which feeds a 64 KiB hop loses
69% of the throughput of a path which never enables BIG TCP, and bounded
resegmentation recovers it.

The new path is taken only when the skb is a plain TCP GSO skb which
exceeds gso_max_size or gso_max_segs, the device offloads that GSO type
and has scatter-gather and checksum offload for the protocol, and the
bound leaves room for at least two MSS segments per output skb.
Encapsulated skbs, frag-list skbs, GSO types the device cannot offload and
bounds below two segments keep the existing full segmentation path
unchanged.

The output obeys the GSO feature and limit contract the device already
advertises - gso_size stays at the MSS, gso_segs stays within the bound,
and the frame length stays below gso_max_size - so this needs no new UAPI,
no device state and no driver change, and it is applied automatically.
I considered a per-device switch and decided against it: it would spend
netlink ABI and net_device state on a decision the stack can make from
capabilities the device already advertises, and it would have to be
configured on every device created later.

Patch layout:

  [1/4] factor the device GSO limit check out of gso_features_check()
  [2/4] let the GSO engine bound the MSS segments per output skb
  [3/4] apply that bound to oversized TCP GSO skbs in the TX path
  [4/4] KUnit coverage for the bound, the device limits and the TCP path

1/4 is a preparation patch with no functional change.

Wang Zhan (4):
  net: core: factor out the GSO device limit check
  net: gso: support bounded TCP segmentation
  net: core: resegment oversized TCP GSO skbs
  net: net_test: add tests for bounded GSO segmentation

 drivers/net/tap.c          |   2 +-
 include/linux/netdevice.h  |   4 +-
 include/net/gso.h          |   6 +-
 include/net/udp.h          |   2 +-
 net/core/dev.c             | 123 ++++++++++++++++++--
 net/core/gso.c             |   5 +-
 net/core/net_test.c        | 231 +++++++++++++++++++++++++++++++++++++
 net/core/skbuff.c          |  14 ++-
 net/ipv4/tcp_offload.c     |   3 +-
 net/openvswitch/datapath.c |   2 +-
 10 files changed, 370 insertions(+), 22 deletions(-)


base-commit: 7bc21b740291c06b6437fd7ed7fab2d707b35cff
-- 
2.47.3