[bpf-next,v3,0/6] bpf: Support atomic update for htab of maps

Message

Hou Tao April 1, 2025, 6:22 a.m. UTC

From: Hou Tao <[e-mail]>

Hi,

The motivation for the patch set comes from the question raised by Cody
Haas [1]. When trying to concurrently lookup and update an existing
element in a htab of maps, the lookup procedure may return -ENOENT
unexpectedly. The first revision of the patch set tried to resolve the
problem by making the insertion of the new element and the deletion of
the old element being atomic from the perspective of the lookup process.
While the solution would benefit all hash maps, it does not fully
resolved the problem due to the immediate reuse issue. Therefore, in v2
of the patch set, it only fixes the problem for fd htab.

Please see individual patches for details. Comments are always welcome.

v3:
 * rebase on bpf_next/for-next
 * add Acked-by tags

v2: https://lore.kernel.org/bpf/[e-mail]/
  * only support atomic update for fd htab

v1: https://lore.kernel.org/bpf/[e-mail]

[1]: https://lore.kernel.org/xdp-newbies/[e-mail]/


Hou Tao (6):
  bpf: Factor out htab_elem_value helper()
  bpf: Rename __htab_percpu_map_update_elem to
    htab_map_update_elem_in_place
  bpf: Support atomic update for htab of maps
  bpf: Add is_fd_htab() helper
  bpf: Don't allocate per-cpu extra_elems for fd htab
  selftests/bpf: Add test case for atomic update of fd htab

 kernel/bpf/hashtab.c                          | 148 +++++++-------
 .../selftests/bpf/prog_tests/fd_htab_lookup.c | 192 ++++++++++++++++++
 .../selftests/bpf/progs/fd_htab_lookup.c      |  25 +++
 3 files changed, 289 insertions(+), 76 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/fd_htab_lookup.c
 create mode 100644 tools/testing/selftests/bpf/progs/fd_htab_lookup.c