bpf: Consolidate spin_lock, timer management into btf_record
[ Email screwup on my end means I have to resend two weeks of email.
:/ -dan ]
Hello Kumar Kartikeya Dwivedi,
The patch db559117828d: "bpf: Consolidate spin_lock, timer management into btf_record" from Nov 4, 2022, leads to the following Smatch static checker warning:
kernel/bpf/syscall.c:1002 map_check_btf() warn: ignoring unreachable code.
kernel/bpf/syscall.c
946 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
947 u32 btf_key_id, u32 btf_value_id)
948 {
949 const struct btf_type *key_type, *value_type;
950 u32 key_size, value_size;
951 int ret = 0;
952
953 /* Some maps allow key to be unspecified. */
954 if (btf_key_id) {
955 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
956 if (!key_type || key_size != map->key_size)
957 return -EINVAL;
958 } else {
959 key_type = btf_type_by_id(btf, 0);
960 if (!map->ops->map_check_btf)
961 return -EINVAL;
962 }
963
964 value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
965 if (!value_type || value_size != map->value_size)
966 return -EINVAL;
967
968 map->record = btf_parse_fields(btf, value_type, BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR,
969 map->value_size);
970 if (!IS_ERR_OR_NULL(map->record)) {
971 int i;
972
973 if (!bpf_capable()) {
974 ret = -EPERM;
975 goto free_map_tab;
976 }
977 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
978 ret = -EACCES;
979 goto free_map_tab;
980 }
981 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
982 switch (map->record->field_mask & (1 << i)) {
983 case 0:
984 continue;
985 case BPF_SPIN_LOCK:
986 if (map->map_type != BPF_MAP_TYPE_HASH &&
987 map->map_type != BPF_MAP_TYPE_ARRAY &&
988 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
989 map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
990 map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
991 map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
992 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
993 ret = -EOPNOTSUPP;
994 goto free_map_tab;
995 }
996 break;
997 case BPF_TIMER:
998 if (map->map_type != BPF_MAP_TYPE_HASH &&
999 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1000 map->map_type != BPF_MAP_TYPE_ARRAY) {
1001 return -EOPNOTSUPP;
^^^^^^^^^^^^^^^^^^^--> 1002 goto free_map_tab;
s/return/ret =/
Surprised the coverity-bot hasn't complained about this...
1003 }
1004 break;
1005 case BPF_KPTR_UNREF:
1006 case BPF_KPTR_REF:
1007 if (map->map_type != BPF_MAP_TYPE_HASH &&
1008 map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1009 map->map_type != BPF_MAP_TYPE_ARRAY &&
1010 map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) {
1011 ret = -EOPNOTSUPP;
1012 goto free_map_tab;
1013 }
1014 break;
1015 default:
1016 /* Fail if map_type checks are missing for a field type */
1017 ret = -EOPNOTSUPP;
1018 goto free_map_tab;
1019 }
1020 }
1021 }
1022
1023 if (map->ops->map_check_btf) {
1024 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1025 if (ret < 0)
1026 goto free_map_tab;
1027 }
1028
1029 return ret;
1030 free_map_tab:
1031 bpf_map_free_record(map);
1032 return ret;
1033 }regards,
dan carpenter
Kumar Kartikeya Dwivedi
maintainerQuoted 70 lines · Dan Carpenter
[ Email screwup on my end means I have to resend two weeks of email.
:/ -dan ]Hello Kumar Kartikeya Dwivedi,
The patch
db559117828d: "bpf: Consolidatespin_lock, timer management intobtf_record" from Nov 4, 2022, leads to the following Smatch static checker warning:kernel/bpf/syscall.c:1002 map_check_btf() warn: ignoring unreachable code.
kernel/bpf/syscall.c946 static int map_check_btf(struct bpf_map *map, const struct btf *btf, 947 u32 btf_key_id, u32 btf_value_id) 948 { 949 const struct btf_type *key_type, *value_type; 950 u32 key_size, value_size; 951 int ret = 0; 952 953 /* Some maps allow key to be unspecified. */ 954 if (btf_key_id) { 955 key_type = btf_type_id_size(btf, &btf_key_id, &key_size); 956 if (!key_type || key_size != map->key_size) 957 return -EINVAL; 958 } else { 959 key_type = btf_type_by_id(btf, 0); 960 if (!map->ops->map_check_btf) 961 return -EINVAL; 962 } 963 964 value_type = btf_type_id_size(btf, &btf_value_id, &value_size); 965 if (!value_type || value_size != map->value_size) 966 return -EINVAL; 967 968 map->record = btf_parse_fields(btf, value_type, BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR, 969 map->value_size); 970 if (!IS_ERR_OR_NULL(map->record)) { 971 int i; 972 973 if (!bpf_capable()) { 974 ret = -EPERM; 975 goto free_map_tab; 976 } 977 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) { 978 ret = -EACCES; 979 goto free_map_tab; 980 } 981 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) { 982 switch (map->record->field_mask & (1 << i)) { 983 case 0: 984 continue; 985 case BPF_SPIN_LOCK: 986 if (map->map_type != BPF_MAP_TYPE_HASH && 987 map->map_type != BPF_MAP_TYPE_ARRAY && 988 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && 989 map->map_type != BPF_MAP_TYPE_SK_STORAGE && 990 map->map_type != BPF_MAP_TYPE_INODE_STORAGE && 991 map->map_type != BPF_MAP_TYPE_TASK_STORAGE && 992 map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) { 993 ret = -EOPNOTSUPP; 994 goto free_map_tab; 995 } 996 break; 997 case BPF_TIMER: 998 if (map->map_type != BPF_MAP_TYPE_HASH && 999 map->map_type != BPF_MAP_TYPE_LRU_HASH && 1000 map->map_type != BPF_MAP_TYPE_ARRAY) { 1001 return -EOPNOTSUPP; ^^^^^^^^^^^^^^^^^^^
Quoted 6 lines
--> 1002 goto
free_map_tab;s/return/ret =/
Surprised the coverity-bot hasn't complained about this...
Thanks, another bad gaffe :/. Fixed locally, will push it out with v8.