[PATCH bpf-next 0/2] bpf: Support bpf_rcu_head in hash and LRU hash maps

[Puranjay Mohan <puranjay@kernel.org>]

bpf_call_rcu() currently only takes a bpf_rcu_head in an array, because an
array element is never freed while the map is alive. Hash elements are
recycled, and an RCU callback cannot be cancelled, so a delete racing a
queued callback would hand the element back to the allocator underneath it.

This keeps the element alive until the callback has run. The map claims the
head before releasing an element; if a callback is queued or running, the
element is unlinked but not returned to the allocator, and the callback
returns it through a new map_release_elem(). Patch 1 also covers the two
places a hash recycles an element without freeing it: the per-CPU spare a
preallocated htab uses on update, and LRU eviction.

Patch 2 adds tests for the delete, replace, recycle and eviction paths.
Each was checked against a kernel with the corresponding hunk reverted.

Puranjay Mohan (2):
  bpf: Support bpf_rcu_head in hash and LRU hash maps
  selftests/bpf: Add bpf_call_rcu tests for hash and LRU hash maps

 include/linux/bpf.h                           |   5 +
 kernel/bpf/hashtab.c                          | 104 ++++++-
 kernel/bpf/helpers.c                          | 117 +++++++-
 kernel/bpf/syscall.c                          |  14 +-
 .../selftests/bpf/prog_tests/call_rcu.c       | 254 +++++++++++++++-
 tools/testing/selftests/bpf/progs/call_rcu.c  | 274 ++++++++++++++++++
 6 files changed, 740 insertions(+), 28 deletions(-)


base-commit: 4f3a5eae895b9995e93425a75235d8f1f3268caa
-- 
2.53.0-Meta