diff options
| author | Martin KaFai Lau <kafai@fb.com> | 2018-05-04 14:49:51 -0700 |
|---|---|---|
| committer | Daniel Borkmann <daniel@iogearbox.net> | 2018-05-09 17:25:13 +0200 |
| commit | 78958fca7ead2f81b60a6827881c4866d1ed0c52 (patch) | |
| tree | 541a6090ce06d7bc6f708f5a70a0cd4bff187ac2 /include | |
| parent | 82e9697250977f3f87cd42e71e8daa8810e64520 (diff) | |
bpf: btf: Introduce BTF ID
This patch gives an ID to each loaded BTF. The ID is allocated by
the idr like the existing prog-id and map-id.
The bpf_put(map->btf) is moved to __bpf_map_put() so that the
userspace can stop seeing the BTF ID ASAP when the last BTF
refcnt is gone.
It also makes BTF accessible from userspace through the
1. new BPF_BTF_GET_FD_BY_ID command. It is limited to CAP_SYS_ADMIN
which is inline with the BPF_BTF_LOAD cmd and the existing
BPF_[MAP|PROG]_GET_FD_BY_ID cmd.
2. new btf_id (and btf_key_id + btf_value_id) in "struct bpf_map_info"
Once the BTF ID handler is accessible from userspace, freeing a BTF
object has to go through a rcu period. The BPF_BTF_GET_FD_BY_ID cmd
can then be done under a rcu_read_lock() instead of taking
spin_lock.
[Note: A similar rcu usage can be done to the existing
bpf_prog_get_fd_by_id() in a follow up patch]
When processing the BPF_BTF_GET_FD_BY_ID cmd,
refcount_inc_not_zero() is needed because the BTF object
could be already in the rcu dead row . btf_get() is
removed since its usage is currently limited to btf.c
alone. refcount_inc() is used directly instead.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Alexei Starovoitov <ast@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Diffstat (limited to 'include')
| -rw-r--r-- | include/linux/btf.h | 2 | ||||
| -rw-r--r-- | include/uapi/linux/bpf.h | 5 |
2 files changed, 7 insertions, 0 deletions
diff --git a/include/linux/btf.h b/include/linux/btf.h index a966dc6d61ee..e076c4697049 100644 --- a/include/linux/btf.h +++ b/include/linux/btf.h @@ -44,5 +44,7 @@ const struct btf_type *btf_type_id_size(const struct btf *btf, u32 *ret_size); void btf_type_seq_show(const struct btf *btf, u32 type_id, void *obj, struct seq_file *m); +int btf_get_fd_by_id(u32 id); +u32 btf_id(const struct btf *btf); #endif diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 93d5a4eeec2a..6106f23a9a8a 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -96,6 +96,7 @@ enum bpf_cmd { BPF_PROG_QUERY, BPF_RAW_TRACEPOINT_OPEN, BPF_BTF_LOAD, + BPF_BTF_GET_FD_BY_ID, }; enum bpf_map_type { @@ -344,6 +345,7 @@ union bpf_attr { __u32 start_id; __u32 prog_id; __u32 map_id; + __u32 btf_id; }; __u32 next_id; __u32 open_flags; @@ -2130,6 +2132,9 @@ struct bpf_map_info { __u32 ifindex; __u64 netns_dev; __u64 netns_ino; + __u32 btf_id; + __u32 btf_key_id; + __u32 btf_value_id; } __attribute__((aligned(8))); /* User bpf_sock_addr struct to access socket fields and sockaddr struct passed |