1 //SPDX-License-Identifier: GPL-2.0 2 #include <linux/bpf-cgroup.h> 3 #include <linux/bpf.h> 4 #include <linux/btf.h> 5 #include <linux/bug.h> 6 #include <linux/filter.h> 7 #include <linux/mm.h> 8 #include <linux/rbtree.h> 9 #include <linux/slab.h> 10 #include <uapi/linux/btf.h> 11 12 DEFINE_PER_CPU(struct bpf_cgroup_storage*, bpf_cgroup_storage[MAX_BPF_CGROUP_STORAGE_TYPE]); 13 14 #ifdef CONFIG_CGROUP_BPF 15 16 #define LOCAL_STORAGE_CREATE_FLAG_MASK \ 17 (BPF_F_NUMA_NODE | BPF_F_ACCESS_MASK) 18 19 struct bpf_cgroup_storage_map { 20 struct bpf_map map; 21 22 spinlock_t lock; 23 struct bpf_prog_aux *aux; 24 struct rb_root root; 25 struct list_head list; 26 }; 27 28 static struct bpf_cgroup_storage_map *map_to_storage(struct bpf_map *map) 29 { 30 return container_of(map, struct bpf_cgroup_storage_map, map); 31 } 32 33 static int bpf_cgroup_storage_key_cmp( 34 const struct bpf_cgroup_storage_key *key1, 35 const struct bpf_cgroup_storage_key *key2) 36 { 37 if (key1->cgroup_inode_id < key2->cgroup_inode_id) 38 return -1; 39 else if (key1->cgroup_inode_id > key2->cgroup_inode_id) 40 return 1; 41 else if (key1->attach_type < key2->attach_type) 42 return -1; 43 else if (key1->attach_type > key2->attach_type) 44 return 1; 45 return 0; 46 } 47 48 static struct bpf_cgroup_storage *cgroup_storage_lookup( 49 struct bpf_cgroup_storage_map *map, struct bpf_cgroup_storage_key *key, 50 bool locked) 51 { 52 struct rb_root *root = &map->root; 53 struct rb_node *node; 54 55 if (!locked) 56 spin_lock_bh(&map->lock); 57 58 node = root->rb_node; 59 while (node) { 60 struct bpf_cgroup_storage *storage; 61 62 storage = container_of(node, struct bpf_cgroup_storage, node); 63 64 switch (bpf_cgroup_storage_key_cmp(key, &storage->key)) { 65 case -1: 66 node = node->rb_left; 67 break; 68 case 1: 69 node = node->rb_right; 70 break; 71 default: 72 if (!locked) 73 spin_unlock_bh(&map->lock); 74 return storage; 75 } 76 } 77 78 if (!locked) 79 spin_unlock_bh(&map->lock); 80 81 return NULL; 82 } 83 84 static int cgroup_storage_insert(struct bpf_cgroup_storage_map *map, 85 struct bpf_cgroup_storage *storage) 86 { 87 struct rb_root *root = &map->root; 88 struct rb_node **new = &(root->rb_node), *parent = NULL; 89 90 while (*new) { 91 struct bpf_cgroup_storage *this; 92 93 this = container_of(*new, struct bpf_cgroup_storage, node); 94 95 parent = *new; 96 switch (bpf_cgroup_storage_key_cmp(&storage->key, &this->key)) { 97 case -1: 98 new = &((*new)->rb_left); 99 break; 100 case 1: 101 new = &((*new)->rb_right); 102 break; 103 default: 104 return -EEXIST; 105 } 106 } 107 108 rb_link_node(&storage->node, parent, new); 109 rb_insert_color(&storage->node, root); 110 111 return 0; 112 } 113 114 static void *cgroup_storage_lookup_elem(struct bpf_map *_map, void *_key) 115 { 116 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 117 struct bpf_cgroup_storage_key *key = _key; 118 struct bpf_cgroup_storage *storage; 119 120 storage = cgroup_storage_lookup(map, key, false); 121 if (!storage) 122 return NULL; 123 124 return &READ_ONCE(storage->buf)->data[0]; 125 } 126 127 static int cgroup_storage_update_elem(struct bpf_map *map, void *_key, 128 void *value, u64 flags) 129 { 130 struct bpf_cgroup_storage_key *key = _key; 131 struct bpf_cgroup_storage *storage; 132 struct bpf_storage_buffer *new; 133 134 if (unlikely(flags & ~(BPF_F_LOCK | BPF_EXIST | BPF_NOEXIST))) 135 return -EINVAL; 136 137 if (unlikely(flags & BPF_NOEXIST)) 138 return -EINVAL; 139 140 if (unlikely((flags & BPF_F_LOCK) && 141 !map_value_has_spin_lock(map))) 142 return -EINVAL; 143 144 storage = cgroup_storage_lookup((struct bpf_cgroup_storage_map *)map, 145 key, false); 146 if (!storage) 147 return -ENOENT; 148 149 if (flags & BPF_F_LOCK) { 150 copy_map_value_locked(map, storage->buf->data, value, false); 151 return 0; 152 } 153 154 new = kmalloc_node(sizeof(struct bpf_storage_buffer) + 155 map->value_size, 156 __GFP_ZERO | GFP_ATOMIC | __GFP_NOWARN, 157 map->numa_node); 158 if (!new) 159 return -ENOMEM; 160 161 memcpy(&new->data[0], value, map->value_size); 162 check_and_init_map_lock(map, new->data); 163 164 new = xchg(&storage->buf, new); 165 kfree_rcu(new, rcu); 166 167 return 0; 168 } 169 170 int bpf_percpu_cgroup_storage_copy(struct bpf_map *_map, void *_key, 171 void *value) 172 { 173 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 174 struct bpf_cgroup_storage_key *key = _key; 175 struct bpf_cgroup_storage *storage; 176 int cpu, off = 0; 177 u32 size; 178 179 rcu_read_lock(); 180 storage = cgroup_storage_lookup(map, key, false); 181 if (!storage) { 182 rcu_read_unlock(); 183 return -ENOENT; 184 } 185 186 /* per_cpu areas are zero-filled and bpf programs can only 187 * access 'value_size' of them, so copying rounded areas 188 * will not leak any kernel data 189 */ 190 size = round_up(_map->value_size, 8); 191 for_each_possible_cpu(cpu) { 192 bpf_long_memcpy(value + off, 193 per_cpu_ptr(storage->percpu_buf, cpu), size); 194 off += size; 195 } 196 rcu_read_unlock(); 197 return 0; 198 } 199 200 int bpf_percpu_cgroup_storage_update(struct bpf_map *_map, void *_key, 201 void *value, u64 map_flags) 202 { 203 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 204 struct bpf_cgroup_storage_key *key = _key; 205 struct bpf_cgroup_storage *storage; 206 int cpu, off = 0; 207 u32 size; 208 209 if (map_flags != BPF_ANY && map_flags != BPF_EXIST) 210 return -EINVAL; 211 212 rcu_read_lock(); 213 storage = cgroup_storage_lookup(map, key, false); 214 if (!storage) { 215 rcu_read_unlock(); 216 return -ENOENT; 217 } 218 219 /* the user space will provide round_up(value_size, 8) bytes that 220 * will be copied into per-cpu area. bpf programs can only access 221 * value_size of it. During lookup the same extra bytes will be 222 * returned or zeros which were zero-filled by percpu_alloc, 223 * so no kernel data leaks possible 224 */ 225 size = round_up(_map->value_size, 8); 226 for_each_possible_cpu(cpu) { 227 bpf_long_memcpy(per_cpu_ptr(storage->percpu_buf, cpu), 228 value + off, size); 229 off += size; 230 } 231 rcu_read_unlock(); 232 return 0; 233 } 234 235 static int cgroup_storage_get_next_key(struct bpf_map *_map, void *_key, 236 void *_next_key) 237 { 238 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 239 struct bpf_cgroup_storage_key *key = _key; 240 struct bpf_cgroup_storage_key *next = _next_key; 241 struct bpf_cgroup_storage *storage; 242 243 spin_lock_bh(&map->lock); 244 245 if (list_empty(&map->list)) 246 goto enoent; 247 248 if (key) { 249 storage = cgroup_storage_lookup(map, key, true); 250 if (!storage) 251 goto enoent; 252 253 storage = list_next_entry(storage, list); 254 if (!storage) 255 goto enoent; 256 } else { 257 storage = list_first_entry(&map->list, 258 struct bpf_cgroup_storage, list); 259 } 260 261 spin_unlock_bh(&map->lock); 262 next->attach_type = storage->key.attach_type; 263 next->cgroup_inode_id = storage->key.cgroup_inode_id; 264 return 0; 265 266 enoent: 267 spin_unlock_bh(&map->lock); 268 return -ENOENT; 269 } 270 271 static struct bpf_map *cgroup_storage_map_alloc(union bpf_attr *attr) 272 { 273 int numa_node = bpf_map_attr_numa_node(attr); 274 struct bpf_cgroup_storage_map *map; 275 struct bpf_map_memory mem; 276 int ret; 277 278 if (attr->key_size != sizeof(struct bpf_cgroup_storage_key)) 279 return ERR_PTR(-EINVAL); 280 281 if (attr->value_size == 0) 282 return ERR_PTR(-EINVAL); 283 284 if (attr->value_size > PAGE_SIZE) 285 return ERR_PTR(-E2BIG); 286 287 if (attr->map_flags & ~LOCAL_STORAGE_CREATE_FLAG_MASK || 288 !bpf_map_flags_access_ok(attr->map_flags)) 289 return ERR_PTR(-EINVAL); 290 291 if (attr->max_entries) 292 /* max_entries is not used and enforced to be 0 */ 293 return ERR_PTR(-EINVAL); 294 295 ret = bpf_map_charge_init(&mem, sizeof(struct bpf_cgroup_storage_map)); 296 if (ret < 0) 297 return ERR_PTR(ret); 298 299 map = kmalloc_node(sizeof(struct bpf_cgroup_storage_map), 300 __GFP_ZERO | GFP_USER, numa_node); 301 if (!map) { 302 bpf_map_charge_finish(&mem); 303 return ERR_PTR(-ENOMEM); 304 } 305 306 bpf_map_charge_move(&map->map.memory, &mem); 307 308 /* copy mandatory map attributes */ 309 bpf_map_init_from_attr(&map->map, attr); 310 311 spin_lock_init(&map->lock); 312 map->root = RB_ROOT; 313 INIT_LIST_HEAD(&map->list); 314 315 return &map->map; 316 } 317 318 static void cgroup_storage_map_free(struct bpf_map *_map) 319 { 320 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 321 322 WARN_ON(!RB_EMPTY_ROOT(&map->root)); 323 WARN_ON(!list_empty(&map->list)); 324 325 kfree(map); 326 } 327 328 static int cgroup_storage_delete_elem(struct bpf_map *map, void *key) 329 { 330 return -EINVAL; 331 } 332 333 static int cgroup_storage_check_btf(const struct bpf_map *map, 334 const struct btf *btf, 335 const struct btf_type *key_type, 336 const struct btf_type *value_type) 337 { 338 struct btf_member *m; 339 u32 offset, size; 340 341 /* Key is expected to be of struct bpf_cgroup_storage_key type, 342 * which is: 343 * struct bpf_cgroup_storage_key { 344 * __u64 cgroup_inode_id; 345 * __u32 attach_type; 346 * }; 347 */ 348 349 /* 350 * Key_type must be a structure with two fields. 351 */ 352 if (BTF_INFO_KIND(key_type->info) != BTF_KIND_STRUCT || 353 BTF_INFO_VLEN(key_type->info) != 2) 354 return -EINVAL; 355 356 /* 357 * The first field must be a 64 bit integer at 0 offset. 358 */ 359 m = (struct btf_member *)(key_type + 1); 360 size = sizeof_field(struct bpf_cgroup_storage_key, cgroup_inode_id); 361 if (!btf_member_is_reg_int(btf, key_type, m, 0, size)) 362 return -EINVAL; 363 364 /* 365 * The second field must be a 32 bit integer at 64 bit offset. 366 */ 367 m++; 368 offset = offsetof(struct bpf_cgroup_storage_key, attach_type); 369 size = sizeof_field(struct bpf_cgroup_storage_key, attach_type); 370 if (!btf_member_is_reg_int(btf, key_type, m, offset, size)) 371 return -EINVAL; 372 373 return 0; 374 } 375 376 static void cgroup_storage_seq_show_elem(struct bpf_map *map, void *_key, 377 struct seq_file *m) 378 { 379 enum bpf_cgroup_storage_type stype = cgroup_storage_type(map); 380 struct bpf_cgroup_storage_key *key = _key; 381 struct bpf_cgroup_storage *storage; 382 int cpu; 383 384 rcu_read_lock(); 385 storage = cgroup_storage_lookup(map_to_storage(map), key, false); 386 if (!storage) { 387 rcu_read_unlock(); 388 return; 389 } 390 391 btf_type_seq_show(map->btf, map->btf_key_type_id, key, m); 392 stype = cgroup_storage_type(map); 393 if (stype == BPF_CGROUP_STORAGE_SHARED) { 394 seq_puts(m, ": "); 395 btf_type_seq_show(map->btf, map->btf_value_type_id, 396 &READ_ONCE(storage->buf)->data[0], m); 397 seq_puts(m, "\n"); 398 } else { 399 seq_puts(m, ": {\n"); 400 for_each_possible_cpu(cpu) { 401 seq_printf(m, "\tcpu%d: ", cpu); 402 btf_type_seq_show(map->btf, map->btf_value_type_id, 403 per_cpu_ptr(storage->percpu_buf, cpu), 404 m); 405 seq_puts(m, "\n"); 406 } 407 seq_puts(m, "}\n"); 408 } 409 rcu_read_unlock(); 410 } 411 412 const struct bpf_map_ops cgroup_storage_map_ops = { 413 .map_alloc = cgroup_storage_map_alloc, 414 .map_free = cgroup_storage_map_free, 415 .map_get_next_key = cgroup_storage_get_next_key, 416 .map_lookup_elem = cgroup_storage_lookup_elem, 417 .map_update_elem = cgroup_storage_update_elem, 418 .map_delete_elem = cgroup_storage_delete_elem, 419 .map_check_btf = cgroup_storage_check_btf, 420 .map_seq_show_elem = cgroup_storage_seq_show_elem, 421 }; 422 423 int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *_map) 424 { 425 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map); 426 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 427 int ret = -EBUSY; 428 429 spin_lock_bh(&map->lock); 430 431 if (map->aux && map->aux != aux) 432 goto unlock; 433 if (aux->cgroup_storage[stype] && 434 aux->cgroup_storage[stype] != _map) 435 goto unlock; 436 437 map->aux = aux; 438 aux->cgroup_storage[stype] = _map; 439 ret = 0; 440 unlock: 441 spin_unlock_bh(&map->lock); 442 443 return ret; 444 } 445 446 void bpf_cgroup_storage_release(struct bpf_prog_aux *aux, struct bpf_map *_map) 447 { 448 enum bpf_cgroup_storage_type stype = cgroup_storage_type(_map); 449 struct bpf_cgroup_storage_map *map = map_to_storage(_map); 450 451 spin_lock_bh(&map->lock); 452 if (map->aux == aux) { 453 WARN_ON(aux->cgroup_storage[stype] != _map); 454 map->aux = NULL; 455 aux->cgroup_storage[stype] = NULL; 456 } 457 spin_unlock_bh(&map->lock); 458 } 459 460 static size_t bpf_cgroup_storage_calculate_size(struct bpf_map *map, u32 *pages) 461 { 462 size_t size; 463 464 if (cgroup_storage_type(map) == BPF_CGROUP_STORAGE_SHARED) { 465 size = sizeof(struct bpf_storage_buffer) + map->value_size; 466 *pages = round_up(sizeof(struct bpf_cgroup_storage) + size, 467 PAGE_SIZE) >> PAGE_SHIFT; 468 } else { 469 size = map->value_size; 470 *pages = round_up(round_up(size, 8) * num_possible_cpus(), 471 PAGE_SIZE) >> PAGE_SHIFT; 472 } 473 474 return size; 475 } 476 477 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog, 478 enum bpf_cgroup_storage_type stype) 479 { 480 struct bpf_cgroup_storage *storage; 481 struct bpf_map *map; 482 gfp_t flags; 483 size_t size; 484 u32 pages; 485 486 map = prog->aux->cgroup_storage[stype]; 487 if (!map) 488 return NULL; 489 490 size = bpf_cgroup_storage_calculate_size(map, &pages); 491 492 if (bpf_map_charge_memlock(map, pages)) 493 return ERR_PTR(-EPERM); 494 495 storage = kmalloc_node(sizeof(struct bpf_cgroup_storage), 496 __GFP_ZERO | GFP_USER, map->numa_node); 497 if (!storage) 498 goto enomem; 499 500 flags = __GFP_ZERO | GFP_USER; 501 502 if (stype == BPF_CGROUP_STORAGE_SHARED) { 503 storage->buf = kmalloc_node(size, flags, map->numa_node); 504 if (!storage->buf) 505 goto enomem; 506 check_and_init_map_lock(map, storage->buf->data); 507 } else { 508 storage->percpu_buf = __alloc_percpu_gfp(size, 8, flags); 509 if (!storage->percpu_buf) 510 goto enomem; 511 } 512 513 storage->map = (struct bpf_cgroup_storage_map *)map; 514 515 return storage; 516 517 enomem: 518 bpf_map_uncharge_memlock(map, pages); 519 kfree(storage); 520 return ERR_PTR(-ENOMEM); 521 } 522 523 static void free_shared_cgroup_storage_rcu(struct rcu_head *rcu) 524 { 525 struct bpf_cgroup_storage *storage = 526 container_of(rcu, struct bpf_cgroup_storage, rcu); 527 528 kfree(storage->buf); 529 kfree(storage); 530 } 531 532 static void free_percpu_cgroup_storage_rcu(struct rcu_head *rcu) 533 { 534 struct bpf_cgroup_storage *storage = 535 container_of(rcu, struct bpf_cgroup_storage, rcu); 536 537 free_percpu(storage->percpu_buf); 538 kfree(storage); 539 } 540 541 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage) 542 { 543 enum bpf_cgroup_storage_type stype; 544 struct bpf_map *map; 545 u32 pages; 546 547 if (!storage) 548 return; 549 550 map = &storage->map->map; 551 552 bpf_cgroup_storage_calculate_size(map, &pages); 553 bpf_map_uncharge_memlock(map, pages); 554 555 stype = cgroup_storage_type(map); 556 if (stype == BPF_CGROUP_STORAGE_SHARED) 557 call_rcu(&storage->rcu, free_shared_cgroup_storage_rcu); 558 else 559 call_rcu(&storage->rcu, free_percpu_cgroup_storage_rcu); 560 } 561 562 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage, 563 struct cgroup *cgroup, 564 enum bpf_attach_type type) 565 { 566 struct bpf_cgroup_storage_map *map; 567 568 if (!storage) 569 return; 570 571 storage->key.attach_type = type; 572 storage->key.cgroup_inode_id = cgroup_id(cgroup); 573 574 map = storage->map; 575 576 spin_lock_bh(&map->lock); 577 WARN_ON(cgroup_storage_insert(map, storage)); 578 list_add(&storage->list, &map->list); 579 spin_unlock_bh(&map->lock); 580 } 581 582 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage) 583 { 584 struct bpf_cgroup_storage_map *map; 585 struct rb_root *root; 586 587 if (!storage) 588 return; 589 590 map = storage->map; 591 592 spin_lock_bh(&map->lock); 593 root = &map->root; 594 rb_erase(&storage->node, root); 595 596 list_del(&storage->list); 597 spin_unlock_bh(&map->lock); 598 } 599 600 #endif 601