[PATCH bpf-next v5 0/3] bpf: bidirectional VLAN support for bpf_fib_lookup()

[Avinash Duduskar]

This series adds VLAN awareness to bpf_fib_lookup() in both directions.
BPF_FIB_LOOKUP_VLAN resolves a VLAN egress to its underlying real device
plus the VLAN tag (XDP programs need this because VLAN devices have no XDP
xmit), and BPF_FIB_LOOKUP_VLAN_INPUT runs the lookup as if a tagged frame
had arrived on the matching VLAN subinterface, for iif policy routing and
VRF table selection.

The independent l3mdev/VRF flow-init fix, patch 1 in v1 and v2, was split
out and merged to bpf separately.

An unreducible VLAN egress (a QinQ egress, or a parent in another
namespace) returns BPF_FIB_LKUP_RET_VLAN_FAILURE rather than a best-effort
SUCCESS, so an XDP program cannot mistake it for a physical egress and
silently blackhole the frame at xdp_do_flush(). The code is appended after
BPF_FIB_LKUP_RET_NO_SRC_ADDR (nothing renumbered, tools/ mirror updated)
and is returned only when BPF_FIB_LOOKUP_VLAN is set, so no existing caller
can observe it. On that failure params->ifindex is left at the input; a
program that wants the VLAN device's own ifindex re-issues without the flag.

Changes v4 -> v5 (Toke's review,
https://lore.kernel.org/bpf/87y0g5ca7x.fsf@xxxxxxx/):

- Patch 1: BPF_FIB_LOOKUP_VLAN only makes sense for XDP, which cannot
  redirect to a VLAN device; a tc program can redirect to the VLAN device
  directly. So bpf_skb_fib_lookup() now rejects the flag with -EINVAL, and
  the fwd_dev out-parameter added in v4 is dropped: with the flag gone from
  the skb path there is no swap to preserve, so the deferred mtu check
  returns to the original dev_get_by_index_rcu(net, params->ifindex). The
  VLAN_FAILURE rewind moves into bpf_fib_set_fwd_params() via an input
  ifindex parameter, so each lookup ends in a plain
  "return bpf_fib_set_fwd_params(...)". The early params->ifindex =
  dev->ifindex that NO_NEIGH and NO_SRC_ADDR report stays where
  d1c362e1dd68a ("bpf: Always return target ifindex in bpf_fib_lookup") put
  it. Dropping fwd_dev also removes the i386 W=1 unused-variable warning the
  kernel test robot reported, since net is used again.

- Patch 2: no code change; add Toke's Reviewed-by.

- Patch 3: the BPF_FIB_LOOKUP_VLAN cases assert the tc helper returns
  -EINVAL and check the egress result on the XDP path, including dmac and
  (for tot_len cases) the route mtu_result; the cross-netns egress case
  runs through bpf_xdp_fib_lookup(); the obsolete skb-mtu-after-swap arm is
  dropped.

Changes v3 -> v4:

- Patch 1: return BPF_FIB_LKUP_RET_VLAN_FAILURE for an unreducible VLAN
  egress, leaving params->ifindex at the input, per Toke's v3 review.

- Patch 3: QinQ-egress and cross-namespace-egress arms expect VLAN_FAILURE;
  an escape-hatch arm re-issues without the flag; and a live-frames arm
  asserts a reducible egress is delivered and a QinQ egress is passed to
  the stack.

Taking the tag as lookup input follows the approach David Ahern suggested
in the 2021 fwmark discussion:
https://lore.kernel.org/bpf/6248c547-ad64-04d6-fcec-374893cc1ef2@xxxxxxxxx/

v4: https://lore.kernel.org/all/20260623025147.1001664-1-avinash.duduskar@xxxxxxxxx/
v3: https://lore.kernel.org/all/20260617224729.1428662-1-avinash.duduskar@xxxxxxxxx/
v2: https://lore.kernel.org/all/20260616223426.3568080-1-avinash.duduskar@xxxxxxxxx/
v1: https://lore.kernel.org/all/20260609172052.81613-1-avinash.duduskar@xxxxxxxxx/

Avinash Duduskar (3):
  bpf: Add BPF_FIB_LOOKUP_VLAN flag to bpf_fib_lookup() helper
  bpf: Add BPF_FIB_LOOKUP_VLAN_INPUT flag to bpf_fib_lookup() helper
  selftests/bpf: Add bpf_fib_lookup() VLAN flag tests

 include/uapi/linux/bpf.h                      |  50 +-
 net/core/filter.c                             |  97 ++-
 tools/include/uapi/linux/bpf.h                |  50 +-
 .../selftests/bpf/prog_tests/fib_lookup.c     | 717 +++++++++++++++++-
 .../testing/selftests/bpf/progs/fib_lookup.c  |  36 +
 5 files changed, 936 insertions(+), 14 deletions(-)


base-commit: a975094bf98ca97be9146f9d3b5681a6f9cf5ce3
-- 
2.54.0