| Message ID | 20210712125552.58705-1-m@lambda.lt (mailing list archive) |
|---|---|
| State | Accepted |
| Delegated to: | BPF |
| Headers | show |
| Series | [v2,bpf-next] libbpf: fix reuse of pinned map on older kernel | expand |
| Context | Check | Description |
|---|---|---|
| netdev/cover_letter | success | Link |
| netdev/fixes_present | success | Link |
| netdev/patch_count | success | Link |
| netdev/tree_selection | success | Clearly marked for bpf-next |
| netdev/subject_prefix | success | Link |
| netdev/cc_maintainers | warning | 6 maintainers not CCed: netdev@vger.kernel.org yhs@fb.com kpsingh@kernel.org kafai@fb.com john.fastabend@gmail.com songliubraving@fb.com |
| netdev/source_inline | success | Was 0 now: 0 |
| netdev/verify_signedoff | success | Link |
| netdev/module_param | success | Was 0 now: 0 |
| netdev/build_32bit | success | Errors and warnings before: 0 this patch: 0 |
| netdev/kdoc | success | Errors and warnings before: 0 this patch: 0 |
| netdev/verify_fixes | success | Link |
| netdev/checkpatch | success | total: 0 errors, 0 warnings, 0 checks, 69 lines checked |
| netdev/build_allmodconfig_warn | success | Errors and warnings before: 0 this patch: 0 |
| netdev/header_inline | success | Link |
Martynas Pumputis wrote: > When loading a BPF program with a pinned map, the loader checks whether > the pinned map can be reused, i.e. their properties match. To derive > such of the pinned map, the loader invokes BPF_OBJ_GET_INFO_BY_FD and > then does the comparison. > > Unfortunately, on < 4.12 kernels the BPF_OBJ_GET_INFO_BY_FD is not > available, so loading the program fails with the following error: > > libbpf: failed to get map info for map FD 5: Invalid argument > libbpf: couldn't reuse pinned map at > '/sys/fs/bpf/tc/globals/cilium_call_policy': parameter > mismatch" > libbpf: map 'cilium_call_policy': error reusing pinned map > libbpf: map 'cilium_call_policy': failed to create: > Invalid argument(-22) > libbpf: failed to load object 'bpf_overlay.o' > > To fix this, fallback to derivation of the map properties via > /proc/$PID/fdinfo/$MAP_FD if BPF_OBJ_GET_INFO_BY_FD fails with EINVAL, > which can be used as an indicator that the kernel doesn't support > the latter. > > Signed-off-by: Martynas Pumputis <m@lambda.lt> > --- LGTM. Acked-by: John Fastabend <john.fastabend@gmail.com>
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 1e04ce724240..a952acd6cd77 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -3894,6 +3894,42 @@ static int bpf_map_find_btf_info(struct bpf_object *obj, struct bpf_map *map) return 0; } +static int bpf_get_map_info_from_fdinfo(int fd, struct bpf_map_info *info) +{ + char file[PATH_MAX], buff[4096]; + FILE *fp; + __u32 val; + int err; + + snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd); + memset(info, 0, sizeof(*info)); + + fp = fopen(file, "r"); + if (!fp) { + err = -errno; + pr_warn("failed to open %s: %d. No procfs support?\n", file, + err); + return err; + } + + while (fgets(buff, sizeof(buff), fp)) { + if (sscanf(buff, "map_type:\t%u", &val) == 1) + info->type = val; + else if (sscanf(buff, "key_size:\t%u", &val) == 1) + info->key_size = val; + else if (sscanf(buff, "value_size:\t%u", &val) == 1) + info->value_size = val; + else if (sscanf(buff, "max_entries:\t%u", &val) == 1) + info->max_entries = val; + else if (sscanf(buff, "map_flags:\t%i", &val) == 1) + info->map_flags = val; + } + + fclose(fp); + + return 0; +} + int bpf_map__reuse_fd(struct bpf_map *map, int fd) { struct bpf_map_info info = {}; @@ -3902,6 +3938,8 @@ int bpf_map__reuse_fd(struct bpf_map *map, int fd) char *new_name; err = bpf_obj_get_info_by_fd(fd, &info, &len); + if (err && errno == EINVAL) + err = bpf_get_map_info_from_fdinfo(fd, &info); if (err) return libbpf_err(err); @@ -4381,12 +4419,16 @@ static bool map_is_reuse_compat(const struct bpf_map *map, int map_fd) struct bpf_map_info map_info = {}; char msg[STRERR_BUFSIZE]; __u32 map_info_len; + int err; map_info_len = sizeof(map_info); - if (bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len)) { - pr_warn("failed to get map info for map FD %d: %s\n", - map_fd, libbpf_strerror_r(errno, msg, sizeof(msg))); + err = bpf_obj_get_info_by_fd(map_fd, &map_info, &map_info_len); + if (err && errno == EINVAL) + err = bpf_get_map_info_from_fdinfo(map_fd, &map_info); + if (err) { + pr_warn("failed to get map info for map FD %d: %s\n", map_fd, + libbpf_strerror_r(errno, msg, sizeof(msg))); return false; }
When loading a BPF program with a pinned map, the loader checks whether the pinned map can be reused, i.e. their properties match. To derive such of the pinned map, the loader invokes BPF_OBJ_GET_INFO_BY_FD and then does the comparison. Unfortunately, on < 4.12 kernels the BPF_OBJ_GET_INFO_BY_FD is not available, so loading the program fails with the following error: libbpf: failed to get map info for map FD 5: Invalid argument libbpf: couldn't reuse pinned map at '/sys/fs/bpf/tc/globals/cilium_call_policy': parameter mismatch" libbpf: map 'cilium_call_policy': error reusing pinned map libbpf: map 'cilium_call_policy': failed to create: Invalid argument(-22) libbpf: failed to load object 'bpf_overlay.o' To fix this, fallback to derivation of the map properties via /proc/$PID/fdinfo/$MAP_FD if BPF_OBJ_GET_INFO_BY_FD fails with EINVAL, which can be used as an indicator that the kernel doesn't support the latter. Signed-off-by: Martynas Pumputis <m@lambda.lt> --- tools/lib/bpf/libbpf.c | 48 +++++++++++++++++++++++++++++++++++++++--- 1 file changed, 45 insertions(+), 3 deletions(-)