]> git.ipfire.org Git - thirdparty/iproute2.git/commitdiff
{f, m}_bpf: check map attributes when fetching as pinned
authorDaniel Borkmann <daniel@iogearbox.net>
Thu, 26 Nov 2015 14:38:43 +0000 (15:38 +0100)
committerStephen Hemminger <shemming@brocade.com>
Sun, 29 Nov 2015 19:55:16 +0000 (11:55 -0800)
Make use of the new show_fdinfo() facility and verify that when a
pinned map is being fetched that its basic attributes are the same
as the map we declared from the ELF file. I.e. when placed into the
globalns, collisions could occur. In such a case warn the user and
bail out.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
tc/tc_bpf.c

index c3adc23c0eb3eb139c6f11c5e98e23e57243ca49..b44b1237499b2c224343ebb84b49348aa0029d59 100644 (file)
@@ -205,6 +205,52 @@ void bpf_print_ops(FILE *f, struct rtattr *bpf_ops, __u16 len)
                ops[i].jf, ops[i].k);
 }
 
+static int bpf_map_selfcheck_pinned(int fd, const struct bpf_elf_map *map)
+{
+       char file[PATH_MAX], buff[4096];
+       struct bpf_elf_map tmp, zero;
+       unsigned int val;
+       FILE *fp;
+
+       snprintf(file, sizeof(file), "/proc/%d/fdinfo/%d", getpid(), fd);
+
+       fp = fopen(file, "r");
+       if (!fp) {
+               fprintf(stderr, "No procfs support?!\n");
+               return -EIO;
+       }
+
+       memset(&tmp, 0, sizeof(tmp));
+       while (fgets(buff, sizeof(buff), fp)) {
+               if (sscanf(buff, "map_type:\t%u", &val) == 1)
+                       tmp.type = val;
+               else if (sscanf(buff, "key_size:\t%u", &val) == 1)
+                       tmp.size_key = val;
+               else if (sscanf(buff, "value_size:\t%u", &val) == 1)
+                       tmp.size_value = val;
+               else if (sscanf(buff, "max_entries:\t%u", &val) == 1)
+                       tmp.max_elem = val;
+       }
+
+       fclose(fp);
+
+       if (!memcmp(&tmp, map, offsetof(struct bpf_elf_map, id))) {
+               return 0;
+       } else {
+               memset(&zero, 0, sizeof(zero));
+               /* If kernel doesn't have eBPF-related fdinfo, we cannot do much,
+                * so just accept it. We know we do have an eBPF fd and in this
+                * case, everything is 0. It is guaranteed that no such map exists
+                * since map type of 0 is unloadable BPF_MAP_TYPE_UNSPEC.
+                */
+               if (!memcmp(&tmp, &zero, offsetof(struct bpf_elf_map, id)))
+                       return 0;
+
+               fprintf(stderr, "Map specs from pinned file differ!\n");
+               return -EINVAL;
+       }
+}
+
 static int bpf_valid_mntpt(const char *mnt, unsigned long magic)
 {
        struct statfs st_fs;
@@ -816,6 +862,13 @@ static int bpf_map_attach(const char *name, const struct bpf_elf_map *map,
 
        fd = bpf_probe_pinned(name, map->pinning);
        if (fd > 0) {
+               ret = bpf_map_selfcheck_pinned(fd, map);
+               if (ret < 0) {
+                       close(fd);
+                       fprintf(stderr, "Map \'%s\' self-check failed!\n",
+                               name);
+                       return ret;
+               }
                if (verbose)
                        fprintf(stderr, "Map \'%s\' loaded as pinned!\n",
                                name);