[RFC,net-next,0/9] xdp: allow disable GRO per packet by XDP

Message

Yan Zhai June 20, 2024, 10:19 p.m. UTC

Software GRO is currently controlled by a single switch, i.e.

  ethtool -K dev gro on|off

However, this is not always desired. When GRO is enabled, even if the
kernel cannot GRO certain traffic, it has to run through the GRO receive
handlers with no benefit.

There are also scenarios that turning off GRO is a requirement. For
example, our production environment has a scenario that a TC egress hook
may add multiple encapsulation headers to forwarded skbs for load
balancing and isolation purpose. The encapsulation is implemented via
BPF. But the problem arises then: there is no way to properly offload a
double-encapsulated packet, since skb only has network_header and
inner_network_header to track one layer of encapsulation, but not two.
On the other hand, not all the traffic through this device needs double
encapsulation. But we have to turn off GRO completely for any ingress
device as a result.

A natural approach to make this more flexible is to use XDP to control
GRO behavior. But current semantic gap between XDP buffer/frame and socket
buffer requires some new primitives.

This change set proposes a control bit gro_disabled on skbs to determine
if GRO should work on an skb or not. To manipulate this bit, we
introduce a new XDP kfunc bpf_xdp_disable_gro as well as generic helpers
xdp_frame/buff_fixup_skb_offloading.

The expected working flow is that:
* XDP program examines packets and can call bpf_xdp_disable_gro to
  disable GRO on a packet
* Device drivers need to call xdp_buff_fixup_skb_offloading (or
  equivalent version for xdp_frame), to check if skb->gro_disabled
  needs to be set.
* The kernel will elide GRO later if this bit is used.

Initially we only modified a few drivers for demonstration purpose. The
driver side changes is optional and also incremental depending on
vendors' agenda. Any suggestions are welcome!

Jesper Dangaard Brouer (1):
  mlx5: move xdp_buff scope one level up

Yan Zhai (8):
  skb: introduce gro_disabled bit
  xdp: add XDP_FLAGS_GRO_DISABLED flag
  xdp: implement bpf_xdp_disable_gro kfunc
  bnxt: apply XDP offloading fixup when building skb
  ice: apply XDP offloading fixup when building skb
  veth: apply XDP offloading fixup when building skb
  mlx5: apply XDP offloading fixup when building skb
  bpf: selftests: test disabling GRO by XDP

 drivers/net/ethernet/broadcom/bnxt/bnxt.c     |   4 +
 drivers/net/ethernet/intel/ice/ice_txrx.c     |   2 +
 drivers/net/ethernet/intel/ice/ice_xsk.c      |   6 +-
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |   6 +-
 .../ethernet/mellanox/mlx5/core/en/xsk/rx.c   |  10 +-
 .../ethernet/mellanox/mlx5/core/en/xsk/rx.h   |   6 +-
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 117 ++++++++++-------
 drivers/net/veth.c                            |   4 +
 include/linux/netdevice.h                     |   9 +-
 include/linux/skbuff.h                        |  10 ++
 include/net/xdp.h                             |  38 +++++-
 include/net/xdp_sock_drv.h                    |   2 +-
 net/Kconfig                                   |  10 ++
 net/core/gro.c                                |   2 +-
 net/core/gro_cells.c                          |   2 +-
 net/core/skbuff.c                             |   4 +
 net/core/xdp.c                                |  27 +++-
 tools/testing/selftests/bpf/config            |   1 +
 .../selftests/bpf/prog_tests/xdp_offloading.c | 122 ++++++++++++++++++
 .../selftests/bpf/progs/xdp_offloading.c      |  50 +++++++
 20 files changed, 369 insertions(+), 63 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/xdp_offloading.c
 create mode 100644 tools/testing/selftests/bpf/progs/xdp_offloading.c