Disallow combining BPF_F_LOCK with map values that mix bpf_spin_lock and other special BTF fields, which may lead to undefined behavior. Such cases are now rejected with -EOPNOTSUPP.
Map update flag validation is centralized in bpf_map_check_op_flags() and reused across array, hash, local-storage, and task-storage update paths, with selftests added to verify the new behavior. Changes: v1 -> v2: * Reject any unknown flag bits in map_flags when updating maps (per AI review). * v1: https://lore.kernel.org/bpf/[email protected]/ Leon Hwang (2): bpf: Disallow BPF_F_LOCK with mixed special fields and centralize flag checks selftests/bpf: Add tests to verify BPF_F_LOCK restrictions include/linux/bpf.h | 7 ++ kernel/bpf/arraymap.c | 11 ++- kernel/bpf/bpf_local_storage.c | 7 -- kernel/bpf/bpf_task_storage.c | 3 - kernel/bpf/hashtab.c | 8 +-- kernel/bpf/syscall.c | 4 +- .../selftests/bpf/prog_tests/map_lock.c | 70 +++++++++++++++++++ .../selftests/bpf/progs/test_map_lock.c | 31 +++++++- 8 files changed, 117 insertions(+), 24 deletions(-) -- 2.52.0
