1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 3 */ 4 #include <linux/bpf.h> 5 #include <linux/bpf_trace.h> 6 #include <linux/bpf_lirc.h> 7 #include <linux/btf.h> 8 #include <linux/syscalls.h> 9 #include <linux/slab.h> 10 #include <linux/sched/signal.h> 11 #include <linux/vmalloc.h> 12 #include <linux/mmzone.h> 13 #include <linux/anon_inodes.h> 14 #include <linux/fdtable.h> 15 #include <linux/file.h> 16 #include <linux/fs.h> 17 #include <linux/license.h> 18 #include <linux/filter.h> 19 #include <linux/version.h> 20 #include <linux/kernel.h> 21 #include <linux/idr.h> 22 #include <linux/cred.h> 23 #include <linux/timekeeping.h> 24 #include <linux/ctype.h> 25 #include <linux/nospec.h> 26 27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \ 28 (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \ 29 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \ 30 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) 32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map)) 33 34 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY) 35 36 DEFINE_PER_CPU(int, bpf_prog_active); 37 static DEFINE_IDR(prog_idr); 38 static DEFINE_SPINLOCK(prog_idr_lock); 39 static DEFINE_IDR(map_idr); 40 static DEFINE_SPINLOCK(map_idr_lock); 41 42 int sysctl_unprivileged_bpf_disabled __read_mostly; 43 44 static const struct bpf_map_ops * const bpf_map_types[] = { 45 #define BPF_PROG_TYPE(_id, _ops) 46 #define BPF_MAP_TYPE(_id, _ops) \ 47 [_id] = &_ops, 48 #include <linux/bpf_types.h> 49 #undef BPF_PROG_TYPE 50 #undef BPF_MAP_TYPE 51 }; 52 53 /* 54 * If we're handed a bigger struct than we know of, ensure all the unknown bits 55 * are 0 - i.e. new user-space does not rely on any kernel feature extensions 56 * we don't know about yet. 57 * 58 * There is a ToCToU between this function call and the following 59 * copy_from_user() call. However, this is not a concern since this function is 60 * meant to be a future-proofing of bits. 61 */ 62 int bpf_check_uarg_tail_zero(void __user *uaddr, 63 size_t expected_size, 64 size_t actual_size) 65 { 66 unsigned char __user *addr; 67 unsigned char __user *end; 68 unsigned char val; 69 int err; 70 71 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */ 72 return -E2BIG; 73 74 if (unlikely(!access_ok(uaddr, actual_size))) 75 return -EFAULT; 76 77 if (actual_size <= expected_size) 78 return 0; 79 80 addr = uaddr + expected_size; 81 end = uaddr + actual_size; 82 83 for (; addr < end; addr++) { 84 err = get_user(val, addr); 85 if (err) 86 return err; 87 if (val) 88 return -E2BIG; 89 } 90 91 return 0; 92 } 93 94 const struct bpf_map_ops bpf_map_offload_ops = { 95 .map_alloc = bpf_map_offload_map_alloc, 96 .map_free = bpf_map_offload_map_free, 97 .map_check_btf = map_check_no_btf, 98 }; 99 100 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) 101 { 102 const struct bpf_map_ops *ops; 103 u32 type = attr->map_type; 104 struct bpf_map *map; 105 int err; 106 107 if (type >= ARRAY_SIZE(bpf_map_types)) 108 return ERR_PTR(-EINVAL); 109 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types)); 110 ops = bpf_map_types[type]; 111 if (!ops) 112 return ERR_PTR(-EINVAL); 113 114 if (ops->map_alloc_check) { 115 err = ops->map_alloc_check(attr); 116 if (err) 117 return ERR_PTR(err); 118 } 119 if (attr->map_ifindex) 120 ops = &bpf_map_offload_ops; 121 map = ops->map_alloc(attr); 122 if (IS_ERR(map)) 123 return map; 124 map->ops = ops; 125 map->map_type = type; 126 return map; 127 } 128 129 void *bpf_map_area_alloc(size_t size, int numa_node) 130 { 131 /* We really just want to fail instead of triggering OOM killer 132 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc, 133 * which is used for lower order allocation requests. 134 * 135 * It has been observed that higher order allocation requests done by 136 * vmalloc with __GFP_NORETRY being set might fail due to not trying 137 * to reclaim memory from the page cache, thus we set 138 * __GFP_RETRY_MAYFAIL to avoid such situations. 139 */ 140 141 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO; 142 void *area; 143 144 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { 145 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags, 146 numa_node); 147 if (area != NULL) 148 return area; 149 } 150 151 return __vmalloc_node_flags_caller(size, numa_node, 152 GFP_KERNEL | __GFP_RETRY_MAYFAIL | 153 flags, __builtin_return_address(0)); 154 } 155 156 void bpf_map_area_free(void *area) 157 { 158 kvfree(area); 159 } 160 161 static u32 bpf_map_flags_retain_permanent(u32 flags) 162 { 163 /* Some map creation flags are not tied to the map object but 164 * rather to the map fd instead, so they have no meaning upon 165 * map object inspection since multiple file descriptors with 166 * different (access) properties can exist here. Thus, given 167 * this has zero meaning for the map itself, lets clear these 168 * from here. 169 */ 170 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY); 171 } 172 173 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr) 174 { 175 map->map_type = attr->map_type; 176 map->key_size = attr->key_size; 177 map->value_size = attr->value_size; 178 map->max_entries = attr->max_entries; 179 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags); 180 map->numa_node = bpf_map_attr_numa_node(attr); 181 } 182 183 static int bpf_charge_memlock(struct user_struct *user, u32 pages) 184 { 185 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 186 187 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) { 188 atomic_long_sub(pages, &user->locked_vm); 189 return -EPERM; 190 } 191 return 0; 192 } 193 194 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages) 195 { 196 if (user) 197 atomic_long_sub(pages, &user->locked_vm); 198 } 199 200 int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size) 201 { 202 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT; 203 struct user_struct *user; 204 int ret; 205 206 if (size >= U32_MAX - PAGE_SIZE) 207 return -E2BIG; 208 209 user = get_current_user(); 210 ret = bpf_charge_memlock(user, pages); 211 if (ret) { 212 free_uid(user); 213 return ret; 214 } 215 216 mem->pages = pages; 217 mem->user = user; 218 219 return 0; 220 } 221 222 void bpf_map_charge_finish(struct bpf_map_memory *mem) 223 { 224 bpf_uncharge_memlock(mem->user, mem->pages); 225 free_uid(mem->user); 226 } 227 228 void bpf_map_charge_move(struct bpf_map_memory *dst, 229 struct bpf_map_memory *src) 230 { 231 *dst = *src; 232 233 /* Make sure src will not be used for the redundant uncharging. */ 234 memset(src, 0, sizeof(struct bpf_map_memory)); 235 } 236 237 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages) 238 { 239 int ret; 240 241 ret = bpf_charge_memlock(map->memory.user, pages); 242 if (ret) 243 return ret; 244 map->memory.pages += pages; 245 return ret; 246 } 247 248 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages) 249 { 250 bpf_uncharge_memlock(map->memory.user, pages); 251 map->memory.pages -= pages; 252 } 253 254 static int bpf_map_alloc_id(struct bpf_map *map) 255 { 256 int id; 257 258 idr_preload(GFP_KERNEL); 259 spin_lock_bh(&map_idr_lock); 260 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC); 261 if (id > 0) 262 map->id = id; 263 spin_unlock_bh(&map_idr_lock); 264 idr_preload_end(); 265 266 if (WARN_ON_ONCE(!id)) 267 return -ENOSPC; 268 269 return id > 0 ? 0 : id; 270 } 271 272 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock) 273 { 274 unsigned long flags; 275 276 /* Offloaded maps are removed from the IDR store when their device 277 * disappears - even if someone holds an fd to them they are unusable, 278 * the memory is gone, all ops will fail; they are simply waiting for 279 * refcnt to drop to be freed. 280 */ 281 if (!map->id) 282 return; 283 284 if (do_idr_lock) 285 spin_lock_irqsave(&map_idr_lock, flags); 286 else 287 __acquire(&map_idr_lock); 288 289 idr_remove(&map_idr, map->id); 290 map->id = 0; 291 292 if (do_idr_lock) 293 spin_unlock_irqrestore(&map_idr_lock, flags); 294 else 295 __release(&map_idr_lock); 296 } 297 298 /* called from workqueue */ 299 static void bpf_map_free_deferred(struct work_struct *work) 300 { 301 struct bpf_map *map = container_of(work, struct bpf_map, work); 302 struct bpf_map_memory mem; 303 304 bpf_map_charge_move(&mem, &map->memory); 305 security_bpf_map_free(map); 306 /* implementation dependent freeing */ 307 map->ops->map_free(map); 308 bpf_map_charge_finish(&mem); 309 } 310 311 static void bpf_map_put_uref(struct bpf_map *map) 312 { 313 if (atomic_dec_and_test(&map->usercnt)) { 314 if (map->ops->map_release_uref) 315 map->ops->map_release_uref(map); 316 } 317 } 318 319 /* decrement map refcnt and schedule it for freeing via workqueue 320 * (unrelying map implementation ops->map_free() might sleep) 321 */ 322 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) 323 { 324 if (atomic_dec_and_test(&map->refcnt)) { 325 /* bpf_map_free_id() must be called first */ 326 bpf_map_free_id(map, do_idr_lock); 327 btf_put(map->btf); 328 INIT_WORK(&map->work, bpf_map_free_deferred); 329 schedule_work(&map->work); 330 } 331 } 332 333 void bpf_map_put(struct bpf_map *map) 334 { 335 __bpf_map_put(map, true); 336 } 337 EXPORT_SYMBOL_GPL(bpf_map_put); 338 339 void bpf_map_put_with_uref(struct bpf_map *map) 340 { 341 bpf_map_put_uref(map); 342 bpf_map_put(map); 343 } 344 345 static int bpf_map_release(struct inode *inode, struct file *filp) 346 { 347 struct bpf_map *map = filp->private_data; 348 349 if (map->ops->map_release) 350 map->ops->map_release(map, filp); 351 352 bpf_map_put_with_uref(map); 353 return 0; 354 } 355 356 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f) 357 { 358 fmode_t mode = f.file->f_mode; 359 360 /* Our file permissions may have been overridden by global 361 * map permissions facing syscall side. 362 */ 363 if (READ_ONCE(map->frozen)) 364 mode &= ~FMODE_CAN_WRITE; 365 return mode; 366 } 367 368 #ifdef CONFIG_PROC_FS 369 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) 370 { 371 const struct bpf_map *map = filp->private_data; 372 const struct bpf_array *array; 373 u32 owner_prog_type = 0; 374 u32 owner_jited = 0; 375 376 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { 377 array = container_of(map, struct bpf_array, map); 378 owner_prog_type = array->owner_prog_type; 379 owner_jited = array->owner_jited; 380 } 381 382 seq_printf(m, 383 "map_type:\t%u\n" 384 "key_size:\t%u\n" 385 "value_size:\t%u\n" 386 "max_entries:\t%u\n" 387 "map_flags:\t%#x\n" 388 "memlock:\t%llu\n" 389 "map_id:\t%u\n" 390 "frozen:\t%u\n", 391 map->map_type, 392 map->key_size, 393 map->value_size, 394 map->max_entries, 395 map->map_flags, 396 map->memory.pages * 1ULL << PAGE_SHIFT, 397 map->id, 398 READ_ONCE(map->frozen)); 399 400 if (owner_prog_type) { 401 seq_printf(m, "owner_prog_type:\t%u\n", 402 owner_prog_type); 403 seq_printf(m, "owner_jited:\t%u\n", 404 owner_jited); 405 } 406 } 407 #endif 408 409 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz, 410 loff_t *ppos) 411 { 412 /* We need this handler such that alloc_file() enables 413 * f_mode with FMODE_CAN_READ. 414 */ 415 return -EINVAL; 416 } 417 418 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf, 419 size_t siz, loff_t *ppos) 420 { 421 /* We need this handler such that alloc_file() enables 422 * f_mode with FMODE_CAN_WRITE. 423 */ 424 return -EINVAL; 425 } 426 427 const struct file_operations bpf_map_fops = { 428 #ifdef CONFIG_PROC_FS 429 .show_fdinfo = bpf_map_show_fdinfo, 430 #endif 431 .release = bpf_map_release, 432 .read = bpf_dummy_read, 433 .write = bpf_dummy_write, 434 }; 435 436 int bpf_map_new_fd(struct bpf_map *map, int flags) 437 { 438 int ret; 439 440 ret = security_bpf_map(map, OPEN_FMODE(flags)); 441 if (ret < 0) 442 return ret; 443 444 return anon_inode_getfd("bpf-map", &bpf_map_fops, map, 445 flags | O_CLOEXEC); 446 } 447 448 int bpf_get_file_flag(int flags) 449 { 450 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY)) 451 return -EINVAL; 452 if (flags & BPF_F_RDONLY) 453 return O_RDONLY; 454 if (flags & BPF_F_WRONLY) 455 return O_WRONLY; 456 return O_RDWR; 457 } 458 459 /* helper macro to check that unused fields 'union bpf_attr' are zero */ 460 #define CHECK_ATTR(CMD) \ 461 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ 462 sizeof(attr->CMD##_LAST_FIELD), 0, \ 463 sizeof(*attr) - \ 464 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ 465 sizeof(attr->CMD##_LAST_FIELD)) != NULL 466 467 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes. 468 * Return 0 on success and < 0 on error. 469 */ 470 static int bpf_obj_name_cpy(char *dst, const char *src) 471 { 472 const char *end = src + BPF_OBJ_NAME_LEN; 473 474 memset(dst, 0, BPF_OBJ_NAME_LEN); 475 /* Copy all isalnum(), '_' and '.' chars. */ 476 while (src < end && *src) { 477 if (!isalnum(*src) && 478 *src != '_' && *src != '.') 479 return -EINVAL; 480 *dst++ = *src++; 481 } 482 483 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */ 484 if (src == end) 485 return -EINVAL; 486 487 return 0; 488 } 489 490 int map_check_no_btf(const struct bpf_map *map, 491 const struct btf *btf, 492 const struct btf_type *key_type, 493 const struct btf_type *value_type) 494 { 495 return -ENOTSUPP; 496 } 497 498 static int map_check_btf(struct bpf_map *map, const struct btf *btf, 499 u32 btf_key_id, u32 btf_value_id) 500 { 501 const struct btf_type *key_type, *value_type; 502 u32 key_size, value_size; 503 int ret = 0; 504 505 /* Some maps allow key to be unspecified. */ 506 if (btf_key_id) { 507 key_type = btf_type_id_size(btf, &btf_key_id, &key_size); 508 if (!key_type || key_size != map->key_size) 509 return -EINVAL; 510 } else { 511 key_type = btf_type_by_id(btf, 0); 512 if (!map->ops->map_check_btf) 513 return -EINVAL; 514 } 515 516 value_type = btf_type_id_size(btf, &btf_value_id, &value_size); 517 if (!value_type || value_size != map->value_size) 518 return -EINVAL; 519 520 map->spin_lock_off = btf_find_spin_lock(btf, value_type); 521 522 if (map_value_has_spin_lock(map)) { 523 if (map->map_flags & BPF_F_RDONLY_PROG) 524 return -EACCES; 525 if (map->map_type != BPF_MAP_TYPE_HASH && 526 map->map_type != BPF_MAP_TYPE_ARRAY && 527 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && 528 map->map_type != BPF_MAP_TYPE_SK_STORAGE) 529 return -ENOTSUPP; 530 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) > 531 map->value_size) { 532 WARN_ONCE(1, 533 "verifier bug spin_lock_off %d value_size %d\n", 534 map->spin_lock_off, map->value_size); 535 return -EFAULT; 536 } 537 } 538 539 if (map->ops->map_check_btf) 540 ret = map->ops->map_check_btf(map, btf, key_type, value_type); 541 542 return ret; 543 } 544 545 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id 546 /* called via syscall */ 547 static int map_create(union bpf_attr *attr) 548 { 549 int numa_node = bpf_map_attr_numa_node(attr); 550 struct bpf_map_memory mem; 551 struct bpf_map *map; 552 int f_flags; 553 int err; 554 555 err = CHECK_ATTR(BPF_MAP_CREATE); 556 if (err) 557 return -EINVAL; 558 559 f_flags = bpf_get_file_flag(attr->map_flags); 560 if (f_flags < 0) 561 return f_flags; 562 563 if (numa_node != NUMA_NO_NODE && 564 ((unsigned int)numa_node >= nr_node_ids || 565 !node_online(numa_node))) 566 return -EINVAL; 567 568 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ 569 map = find_and_alloc_map(attr); 570 if (IS_ERR(map)) 571 return PTR_ERR(map); 572 573 err = bpf_obj_name_cpy(map->name, attr->map_name); 574 if (err) 575 goto free_map; 576 577 atomic_set(&map->refcnt, 1); 578 atomic_set(&map->usercnt, 1); 579 580 if (attr->btf_key_type_id || attr->btf_value_type_id) { 581 struct btf *btf; 582 583 if (!attr->btf_value_type_id) { 584 err = -EINVAL; 585 goto free_map; 586 } 587 588 btf = btf_get_by_fd(attr->btf_fd); 589 if (IS_ERR(btf)) { 590 err = PTR_ERR(btf); 591 goto free_map; 592 } 593 594 err = map_check_btf(map, btf, attr->btf_key_type_id, 595 attr->btf_value_type_id); 596 if (err) { 597 btf_put(btf); 598 goto free_map; 599 } 600 601 map->btf = btf; 602 map->btf_key_type_id = attr->btf_key_type_id; 603 map->btf_value_type_id = attr->btf_value_type_id; 604 } else { 605 map->spin_lock_off = -EINVAL; 606 } 607 608 err = security_bpf_map_alloc(map); 609 if (err) 610 goto free_map; 611 612 err = bpf_map_alloc_id(map); 613 if (err) 614 goto free_map_sec; 615 616 err = bpf_map_new_fd(map, f_flags); 617 if (err < 0) { 618 /* failed to allocate fd. 619 * bpf_map_put_with_uref() is needed because the above 620 * bpf_map_alloc_id() has published the map 621 * to the userspace and the userspace may 622 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. 623 */ 624 bpf_map_put_with_uref(map); 625 return err; 626 } 627 628 return err; 629 630 free_map_sec: 631 security_bpf_map_free(map); 632 free_map: 633 btf_put(map->btf); 634 bpf_map_charge_move(&mem, &map->memory); 635 map->ops->map_free(map); 636 bpf_map_charge_finish(&mem); 637 return err; 638 } 639 640 /* if error is returned, fd is released. 641 * On success caller should complete fd access with matching fdput() 642 */ 643 struct bpf_map *__bpf_map_get(struct fd f) 644 { 645 if (!f.file) 646 return ERR_PTR(-EBADF); 647 if (f.file->f_op != &bpf_map_fops) { 648 fdput(f); 649 return ERR_PTR(-EINVAL); 650 } 651 652 return f.file->private_data; 653 } 654 655 /* prog's and map's refcnt limit */ 656 #define BPF_MAX_REFCNT 32768 657 658 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) 659 { 660 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { 661 atomic_dec(&map->refcnt); 662 return ERR_PTR(-EBUSY); 663 } 664 if (uref) 665 atomic_inc(&map->usercnt); 666 return map; 667 } 668 EXPORT_SYMBOL_GPL(bpf_map_inc); 669 670 struct bpf_map *bpf_map_get_with_uref(u32 ufd) 671 { 672 struct fd f = fdget(ufd); 673 struct bpf_map *map; 674 675 map = __bpf_map_get(f); 676 if (IS_ERR(map)) 677 return map; 678 679 map = bpf_map_inc(map, true); 680 fdput(f); 681 682 return map; 683 } 684 685 /* map_idr_lock should have been held */ 686 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, 687 bool uref) 688 { 689 int refold; 690 691 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0); 692 693 if (refold >= BPF_MAX_REFCNT) { 694 __bpf_map_put(map, false); 695 return ERR_PTR(-EBUSY); 696 } 697 698 if (!refold) 699 return ERR_PTR(-ENOENT); 700 701 if (uref) 702 atomic_inc(&map->usercnt); 703 704 return map; 705 } 706 707 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref) 708 { 709 spin_lock_bh(&map_idr_lock); 710 map = __bpf_map_inc_not_zero(map, uref); 711 spin_unlock_bh(&map_idr_lock); 712 713 return map; 714 } 715 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero); 716 717 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) 718 { 719 return -ENOTSUPP; 720 } 721 722 static void *__bpf_copy_key(void __user *ukey, u64 key_size) 723 { 724 if (key_size) 725 return memdup_user(ukey, key_size); 726 727 if (ukey) 728 return ERR_PTR(-EINVAL); 729 730 return NULL; 731 } 732 733 /* last field in 'union bpf_attr' used by this command */ 734 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags 735 736 static int map_lookup_elem(union bpf_attr *attr) 737 { 738 void __user *ukey = u64_to_user_ptr(attr->key); 739 void __user *uvalue = u64_to_user_ptr(attr->value); 740 int ufd = attr->map_fd; 741 struct bpf_map *map; 742 void *key, *value, *ptr; 743 u32 value_size; 744 struct fd f; 745 int err; 746 747 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) 748 return -EINVAL; 749 750 if (attr->flags & ~BPF_F_LOCK) 751 return -EINVAL; 752 753 f = fdget(ufd); 754 map = __bpf_map_get(f); 755 if (IS_ERR(map)) 756 return PTR_ERR(map); 757 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 758 err = -EPERM; 759 goto err_put; 760 } 761 762 if ((attr->flags & BPF_F_LOCK) && 763 !map_value_has_spin_lock(map)) { 764 err = -EINVAL; 765 goto err_put; 766 } 767 768 key = __bpf_copy_key(ukey, map->key_size); 769 if (IS_ERR(key)) { 770 err = PTR_ERR(key); 771 goto err_put; 772 } 773 774 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 775 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 776 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 777 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 778 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 779 else if (IS_FD_MAP(map)) 780 value_size = sizeof(u32); 781 else 782 value_size = map->value_size; 783 784 err = -ENOMEM; 785 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 786 if (!value) 787 goto free_key; 788 789 if (bpf_map_is_dev_bound(map)) { 790 err = bpf_map_offload_lookup_elem(map, key, value); 791 goto done; 792 } 793 794 preempt_disable(); 795 this_cpu_inc(bpf_prog_active); 796 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 797 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 798 err = bpf_percpu_hash_copy(map, key, value); 799 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 800 err = bpf_percpu_array_copy(map, key, value); 801 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 802 err = bpf_percpu_cgroup_storage_copy(map, key, value); 803 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { 804 err = bpf_stackmap_copy(map, key, value); 805 } else if (IS_FD_ARRAY(map)) { 806 err = bpf_fd_array_map_lookup_elem(map, key, value); 807 } else if (IS_FD_HASH(map)) { 808 err = bpf_fd_htab_map_lookup_elem(map, key, value); 809 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 810 err = bpf_fd_reuseport_array_lookup_elem(map, key, value); 811 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 812 map->map_type == BPF_MAP_TYPE_STACK) { 813 err = map->ops->map_peek_elem(map, value); 814 } else { 815 rcu_read_lock(); 816 if (map->ops->map_lookup_elem_sys_only) 817 ptr = map->ops->map_lookup_elem_sys_only(map, key); 818 else 819 ptr = map->ops->map_lookup_elem(map, key); 820 if (IS_ERR(ptr)) { 821 err = PTR_ERR(ptr); 822 } else if (!ptr) { 823 err = -ENOENT; 824 } else { 825 err = 0; 826 if (attr->flags & BPF_F_LOCK) 827 /* lock 'ptr' and copy everything but lock */ 828 copy_map_value_locked(map, value, ptr, true); 829 else 830 copy_map_value(map, value, ptr); 831 /* mask lock, since value wasn't zero inited */ 832 check_and_init_map_lock(map, value); 833 } 834 rcu_read_unlock(); 835 } 836 this_cpu_dec(bpf_prog_active); 837 preempt_enable(); 838 839 done: 840 if (err) 841 goto free_value; 842 843 err = -EFAULT; 844 if (copy_to_user(uvalue, value, value_size) != 0) 845 goto free_value; 846 847 err = 0; 848 849 free_value: 850 kfree(value); 851 free_key: 852 kfree(key); 853 err_put: 854 fdput(f); 855 return err; 856 } 857 858 static void maybe_wait_bpf_programs(struct bpf_map *map) 859 { 860 /* Wait for any running BPF programs to complete so that 861 * userspace, when we return to it, knows that all programs 862 * that could be running use the new map value. 863 */ 864 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS || 865 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 866 synchronize_rcu(); 867 } 868 869 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags 870 871 static int map_update_elem(union bpf_attr *attr) 872 { 873 void __user *ukey = u64_to_user_ptr(attr->key); 874 void __user *uvalue = u64_to_user_ptr(attr->value); 875 int ufd = attr->map_fd; 876 struct bpf_map *map; 877 void *key, *value; 878 u32 value_size; 879 struct fd f; 880 int err; 881 882 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) 883 return -EINVAL; 884 885 f = fdget(ufd); 886 map = __bpf_map_get(f); 887 if (IS_ERR(map)) 888 return PTR_ERR(map); 889 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 890 err = -EPERM; 891 goto err_put; 892 } 893 894 if ((attr->flags & BPF_F_LOCK) && 895 !map_value_has_spin_lock(map)) { 896 err = -EINVAL; 897 goto err_put; 898 } 899 900 key = __bpf_copy_key(ukey, map->key_size); 901 if (IS_ERR(key)) { 902 err = PTR_ERR(key); 903 goto err_put; 904 } 905 906 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 907 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 908 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 909 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 910 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 911 else 912 value_size = map->value_size; 913 914 err = -ENOMEM; 915 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 916 if (!value) 917 goto free_key; 918 919 err = -EFAULT; 920 if (copy_from_user(value, uvalue, value_size) != 0) 921 goto free_value; 922 923 /* Need to create a kthread, thus must support schedule */ 924 if (bpf_map_is_dev_bound(map)) { 925 err = bpf_map_offload_update_elem(map, key, value, attr->flags); 926 goto out; 927 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP || 928 map->map_type == BPF_MAP_TYPE_SOCKHASH || 929 map->map_type == BPF_MAP_TYPE_SOCKMAP) { 930 err = map->ops->map_update_elem(map, key, value, attr->flags); 931 goto out; 932 } 933 934 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from 935 * inside bpf map update or delete otherwise deadlocks are possible 936 */ 937 preempt_disable(); 938 __this_cpu_inc(bpf_prog_active); 939 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 940 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 941 err = bpf_percpu_hash_update(map, key, value, attr->flags); 942 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 943 err = bpf_percpu_array_update(map, key, value, attr->flags); 944 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 945 err = bpf_percpu_cgroup_storage_update(map, key, value, 946 attr->flags); 947 } else if (IS_FD_ARRAY(map)) { 948 rcu_read_lock(); 949 err = bpf_fd_array_map_update_elem(map, f.file, key, value, 950 attr->flags); 951 rcu_read_unlock(); 952 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { 953 rcu_read_lock(); 954 err = bpf_fd_htab_map_update_elem(map, f.file, key, value, 955 attr->flags); 956 rcu_read_unlock(); 957 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 958 /* rcu_read_lock() is not needed */ 959 err = bpf_fd_reuseport_array_update_elem(map, key, value, 960 attr->flags); 961 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 962 map->map_type == BPF_MAP_TYPE_STACK) { 963 err = map->ops->map_push_elem(map, value, attr->flags); 964 } else { 965 rcu_read_lock(); 966 err = map->ops->map_update_elem(map, key, value, attr->flags); 967 rcu_read_unlock(); 968 } 969 __this_cpu_dec(bpf_prog_active); 970 preempt_enable(); 971 maybe_wait_bpf_programs(map); 972 out: 973 free_value: 974 kfree(value); 975 free_key: 976 kfree(key); 977 err_put: 978 fdput(f); 979 return err; 980 } 981 982 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key 983 984 static int map_delete_elem(union bpf_attr *attr) 985 { 986 void __user *ukey = u64_to_user_ptr(attr->key); 987 int ufd = attr->map_fd; 988 struct bpf_map *map; 989 struct fd f; 990 void *key; 991 int err; 992 993 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) 994 return -EINVAL; 995 996 f = fdget(ufd); 997 map = __bpf_map_get(f); 998 if (IS_ERR(map)) 999 return PTR_ERR(map); 1000 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1001 err = -EPERM; 1002 goto err_put; 1003 } 1004 1005 key = __bpf_copy_key(ukey, map->key_size); 1006 if (IS_ERR(key)) { 1007 err = PTR_ERR(key); 1008 goto err_put; 1009 } 1010 1011 if (bpf_map_is_dev_bound(map)) { 1012 err = bpf_map_offload_delete_elem(map, key); 1013 goto out; 1014 } 1015 1016 preempt_disable(); 1017 __this_cpu_inc(bpf_prog_active); 1018 rcu_read_lock(); 1019 err = map->ops->map_delete_elem(map, key); 1020 rcu_read_unlock(); 1021 __this_cpu_dec(bpf_prog_active); 1022 preempt_enable(); 1023 maybe_wait_bpf_programs(map); 1024 out: 1025 kfree(key); 1026 err_put: 1027 fdput(f); 1028 return err; 1029 } 1030 1031 /* last field in 'union bpf_attr' used by this command */ 1032 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key 1033 1034 static int map_get_next_key(union bpf_attr *attr) 1035 { 1036 void __user *ukey = u64_to_user_ptr(attr->key); 1037 void __user *unext_key = u64_to_user_ptr(attr->next_key); 1038 int ufd = attr->map_fd; 1039 struct bpf_map *map; 1040 void *key, *next_key; 1041 struct fd f; 1042 int err; 1043 1044 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) 1045 return -EINVAL; 1046 1047 f = fdget(ufd); 1048 map = __bpf_map_get(f); 1049 if (IS_ERR(map)) 1050 return PTR_ERR(map); 1051 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 1052 err = -EPERM; 1053 goto err_put; 1054 } 1055 1056 if (ukey) { 1057 key = __bpf_copy_key(ukey, map->key_size); 1058 if (IS_ERR(key)) { 1059 err = PTR_ERR(key); 1060 goto err_put; 1061 } 1062 } else { 1063 key = NULL; 1064 } 1065 1066 err = -ENOMEM; 1067 next_key = kmalloc(map->key_size, GFP_USER); 1068 if (!next_key) 1069 goto free_key; 1070 1071 if (bpf_map_is_dev_bound(map)) { 1072 err = bpf_map_offload_get_next_key(map, key, next_key); 1073 goto out; 1074 } 1075 1076 rcu_read_lock(); 1077 err = map->ops->map_get_next_key(map, key, next_key); 1078 rcu_read_unlock(); 1079 out: 1080 if (err) 1081 goto free_next_key; 1082 1083 err = -EFAULT; 1084 if (copy_to_user(unext_key, next_key, map->key_size) != 0) 1085 goto free_next_key; 1086 1087 err = 0; 1088 1089 free_next_key: 1090 kfree(next_key); 1091 free_key: 1092 kfree(key); 1093 err_put: 1094 fdput(f); 1095 return err; 1096 } 1097 1098 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value 1099 1100 static int map_lookup_and_delete_elem(union bpf_attr *attr) 1101 { 1102 void __user *ukey = u64_to_user_ptr(attr->key); 1103 void __user *uvalue = u64_to_user_ptr(attr->value); 1104 int ufd = attr->map_fd; 1105 struct bpf_map *map; 1106 void *key, *value; 1107 u32 value_size; 1108 struct fd f; 1109 int err; 1110 1111 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM)) 1112 return -EINVAL; 1113 1114 f = fdget(ufd); 1115 map = __bpf_map_get(f); 1116 if (IS_ERR(map)) 1117 return PTR_ERR(map); 1118 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1119 err = -EPERM; 1120 goto err_put; 1121 } 1122 1123 key = __bpf_copy_key(ukey, map->key_size); 1124 if (IS_ERR(key)) { 1125 err = PTR_ERR(key); 1126 goto err_put; 1127 } 1128 1129 value_size = map->value_size; 1130 1131 err = -ENOMEM; 1132 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1133 if (!value) 1134 goto free_key; 1135 1136 if (map->map_type == BPF_MAP_TYPE_QUEUE || 1137 map->map_type == BPF_MAP_TYPE_STACK) { 1138 err = map->ops->map_pop_elem(map, value); 1139 } else { 1140 err = -ENOTSUPP; 1141 } 1142 1143 if (err) 1144 goto free_value; 1145 1146 if (copy_to_user(uvalue, value, value_size) != 0) 1147 goto free_value; 1148 1149 err = 0; 1150 1151 free_value: 1152 kfree(value); 1153 free_key: 1154 kfree(key); 1155 err_put: 1156 fdput(f); 1157 return err; 1158 } 1159 1160 #define BPF_MAP_FREEZE_LAST_FIELD map_fd 1161 1162 static int map_freeze(const union bpf_attr *attr) 1163 { 1164 int err = 0, ufd = attr->map_fd; 1165 struct bpf_map *map; 1166 struct fd f; 1167 1168 if (CHECK_ATTR(BPF_MAP_FREEZE)) 1169 return -EINVAL; 1170 1171 f = fdget(ufd); 1172 map = __bpf_map_get(f); 1173 if (IS_ERR(map)) 1174 return PTR_ERR(map); 1175 if (READ_ONCE(map->frozen)) { 1176 err = -EBUSY; 1177 goto err_put; 1178 } 1179 if (!capable(CAP_SYS_ADMIN)) { 1180 err = -EPERM; 1181 goto err_put; 1182 } 1183 1184 WRITE_ONCE(map->frozen, true); 1185 err_put: 1186 fdput(f); 1187 return err; 1188 } 1189 1190 static const struct bpf_prog_ops * const bpf_prog_types[] = { 1191 #define BPF_PROG_TYPE(_id, _name) \ 1192 [_id] = & _name ## _prog_ops, 1193 #define BPF_MAP_TYPE(_id, _ops) 1194 #include <linux/bpf_types.h> 1195 #undef BPF_PROG_TYPE 1196 #undef BPF_MAP_TYPE 1197 }; 1198 1199 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) 1200 { 1201 const struct bpf_prog_ops *ops; 1202 1203 if (type >= ARRAY_SIZE(bpf_prog_types)) 1204 return -EINVAL; 1205 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types)); 1206 ops = bpf_prog_types[type]; 1207 if (!ops) 1208 return -EINVAL; 1209 1210 if (!bpf_prog_is_dev_bound(prog->aux)) 1211 prog->aux->ops = ops; 1212 else 1213 prog->aux->ops = &bpf_offload_prog_ops; 1214 prog->type = type; 1215 return 0; 1216 } 1217 1218 /* drop refcnt on maps used by eBPF program and free auxilary data */ 1219 static void free_used_maps(struct bpf_prog_aux *aux) 1220 { 1221 enum bpf_cgroup_storage_type stype; 1222 int i; 1223 1224 for_each_cgroup_storage_type(stype) { 1225 if (!aux->cgroup_storage[stype]) 1226 continue; 1227 bpf_cgroup_storage_release(aux->prog, 1228 aux->cgroup_storage[stype]); 1229 } 1230 1231 for (i = 0; i < aux->used_map_cnt; i++) 1232 bpf_map_put(aux->used_maps[i]); 1233 1234 kfree(aux->used_maps); 1235 } 1236 1237 int __bpf_prog_charge(struct user_struct *user, u32 pages) 1238 { 1239 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1240 unsigned long user_bufs; 1241 1242 if (user) { 1243 user_bufs = atomic_long_add_return(pages, &user->locked_vm); 1244 if (user_bufs > memlock_limit) { 1245 atomic_long_sub(pages, &user->locked_vm); 1246 return -EPERM; 1247 } 1248 } 1249 1250 return 0; 1251 } 1252 1253 void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 1254 { 1255 if (user) 1256 atomic_long_sub(pages, &user->locked_vm); 1257 } 1258 1259 static int bpf_prog_charge_memlock(struct bpf_prog *prog) 1260 { 1261 struct user_struct *user = get_current_user(); 1262 int ret; 1263 1264 ret = __bpf_prog_charge(user, prog->pages); 1265 if (ret) { 1266 free_uid(user); 1267 return ret; 1268 } 1269 1270 prog->aux->user = user; 1271 return 0; 1272 } 1273 1274 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) 1275 { 1276 struct user_struct *user = prog->aux->user; 1277 1278 __bpf_prog_uncharge(user, prog->pages); 1279 free_uid(user); 1280 } 1281 1282 static int bpf_prog_alloc_id(struct bpf_prog *prog) 1283 { 1284 int id; 1285 1286 idr_preload(GFP_KERNEL); 1287 spin_lock_bh(&prog_idr_lock); 1288 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); 1289 if (id > 0) 1290 prog->aux->id = id; 1291 spin_unlock_bh(&prog_idr_lock); 1292 idr_preload_end(); 1293 1294 /* id is in [1, INT_MAX) */ 1295 if (WARN_ON_ONCE(!id)) 1296 return -ENOSPC; 1297 1298 return id > 0 ? 0 : id; 1299 } 1300 1301 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) 1302 { 1303 /* cBPF to eBPF migrations are currently not in the idr store. 1304 * Offloaded programs are removed from the store when their device 1305 * disappears - even if someone grabs an fd to them they are unusable, 1306 * simply waiting for refcnt to drop to be freed. 1307 */ 1308 if (!prog->aux->id) 1309 return; 1310 1311 if (do_idr_lock) 1312 spin_lock_bh(&prog_idr_lock); 1313 else 1314 __acquire(&prog_idr_lock); 1315 1316 idr_remove(&prog_idr, prog->aux->id); 1317 prog->aux->id = 0; 1318 1319 if (do_idr_lock) 1320 spin_unlock_bh(&prog_idr_lock); 1321 else 1322 __release(&prog_idr_lock); 1323 } 1324 1325 static void __bpf_prog_put_rcu(struct rcu_head *rcu) 1326 { 1327 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1328 1329 kvfree(aux->func_info); 1330 free_used_maps(aux); 1331 bpf_prog_uncharge_memlock(aux->prog); 1332 security_bpf_prog_free(aux); 1333 bpf_prog_free(aux->prog); 1334 } 1335 1336 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred) 1337 { 1338 bpf_prog_kallsyms_del_all(prog); 1339 btf_put(prog->aux->btf); 1340 bpf_prog_free_linfo(prog); 1341 1342 if (deferred) 1343 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1344 else 1345 __bpf_prog_put_rcu(&prog->aux->rcu); 1346 } 1347 1348 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) 1349 { 1350 if (atomic_dec_and_test(&prog->aux->refcnt)) { 1351 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0); 1352 /* bpf_prog_free_id() must be called first */ 1353 bpf_prog_free_id(prog, do_idr_lock); 1354 __bpf_prog_put_noref(prog, true); 1355 } 1356 } 1357 1358 void bpf_prog_put(struct bpf_prog *prog) 1359 { 1360 __bpf_prog_put(prog, true); 1361 } 1362 EXPORT_SYMBOL_GPL(bpf_prog_put); 1363 1364 static int bpf_prog_release(struct inode *inode, struct file *filp) 1365 { 1366 struct bpf_prog *prog = filp->private_data; 1367 1368 bpf_prog_put(prog); 1369 return 0; 1370 } 1371 1372 static void bpf_prog_get_stats(const struct bpf_prog *prog, 1373 struct bpf_prog_stats *stats) 1374 { 1375 u64 nsecs = 0, cnt = 0; 1376 int cpu; 1377 1378 for_each_possible_cpu(cpu) { 1379 const struct bpf_prog_stats *st; 1380 unsigned int start; 1381 u64 tnsecs, tcnt; 1382 1383 st = per_cpu_ptr(prog->aux->stats, cpu); 1384 do { 1385 start = u64_stats_fetch_begin_irq(&st->syncp); 1386 tnsecs = st->nsecs; 1387 tcnt = st->cnt; 1388 } while (u64_stats_fetch_retry_irq(&st->syncp, start)); 1389 nsecs += tnsecs; 1390 cnt += tcnt; 1391 } 1392 stats->nsecs = nsecs; 1393 stats->cnt = cnt; 1394 } 1395 1396 #ifdef CONFIG_PROC_FS 1397 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1398 { 1399 const struct bpf_prog *prog = filp->private_data; 1400 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1401 struct bpf_prog_stats stats; 1402 1403 bpf_prog_get_stats(prog, &stats); 1404 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1405 seq_printf(m, 1406 "prog_type:\t%u\n" 1407 "prog_jited:\t%u\n" 1408 "prog_tag:\t%s\n" 1409 "memlock:\t%llu\n" 1410 "prog_id:\t%u\n" 1411 "run_time_ns:\t%llu\n" 1412 "run_cnt:\t%llu\n", 1413 prog->type, 1414 prog->jited, 1415 prog_tag, 1416 prog->pages * 1ULL << PAGE_SHIFT, 1417 prog->aux->id, 1418 stats.nsecs, 1419 stats.cnt); 1420 } 1421 #endif 1422 1423 const struct file_operations bpf_prog_fops = { 1424 #ifdef CONFIG_PROC_FS 1425 .show_fdinfo = bpf_prog_show_fdinfo, 1426 #endif 1427 .release = bpf_prog_release, 1428 .read = bpf_dummy_read, 1429 .write = bpf_dummy_write, 1430 }; 1431 1432 int bpf_prog_new_fd(struct bpf_prog *prog) 1433 { 1434 int ret; 1435 1436 ret = security_bpf_prog(prog); 1437 if (ret < 0) 1438 return ret; 1439 1440 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1441 O_RDWR | O_CLOEXEC); 1442 } 1443 1444 static struct bpf_prog *____bpf_prog_get(struct fd f) 1445 { 1446 if (!f.file) 1447 return ERR_PTR(-EBADF); 1448 if (f.file->f_op != &bpf_prog_fops) { 1449 fdput(f); 1450 return ERR_PTR(-EINVAL); 1451 } 1452 1453 return f.file->private_data; 1454 } 1455 1456 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) 1457 { 1458 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { 1459 atomic_sub(i, &prog->aux->refcnt); 1460 return ERR_PTR(-EBUSY); 1461 } 1462 return prog; 1463 } 1464 EXPORT_SYMBOL_GPL(bpf_prog_add); 1465 1466 void bpf_prog_sub(struct bpf_prog *prog, int i) 1467 { 1468 /* Only to be used for undoing previous bpf_prog_add() in some 1469 * error path. We still know that another entity in our call 1470 * path holds a reference to the program, thus atomic_sub() can 1471 * be safely used in such cases! 1472 */ 1473 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); 1474 } 1475 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1476 1477 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) 1478 { 1479 return bpf_prog_add(prog, 1); 1480 } 1481 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1482 1483 /* prog_idr_lock should have been held */ 1484 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1485 { 1486 int refold; 1487 1488 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1489 1490 if (refold >= BPF_MAX_REFCNT) { 1491 __bpf_prog_put(prog, false); 1492 return ERR_PTR(-EBUSY); 1493 } 1494 1495 if (!refold) 1496 return ERR_PTR(-ENOENT); 1497 1498 return prog; 1499 } 1500 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1501 1502 bool bpf_prog_get_ok(struct bpf_prog *prog, 1503 enum bpf_prog_type *attach_type, bool attach_drv) 1504 { 1505 /* not an attachment, just a refcount inc, always allow */ 1506 if (!attach_type) 1507 return true; 1508 1509 if (prog->type != *attach_type) 1510 return false; 1511 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1512 return false; 1513 1514 return true; 1515 } 1516 1517 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1518 bool attach_drv) 1519 { 1520 struct fd f = fdget(ufd); 1521 struct bpf_prog *prog; 1522 1523 prog = ____bpf_prog_get(f); 1524 if (IS_ERR(prog)) 1525 return prog; 1526 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1527 prog = ERR_PTR(-EINVAL); 1528 goto out; 1529 } 1530 1531 prog = bpf_prog_inc(prog); 1532 out: 1533 fdput(f); 1534 return prog; 1535 } 1536 1537 struct bpf_prog *bpf_prog_get(u32 ufd) 1538 { 1539 return __bpf_prog_get(ufd, NULL, false); 1540 } 1541 1542 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1543 bool attach_drv) 1544 { 1545 return __bpf_prog_get(ufd, &type, attach_drv); 1546 } 1547 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1548 1549 /* Initially all BPF programs could be loaded w/o specifying 1550 * expected_attach_type. Later for some of them specifying expected_attach_type 1551 * at load time became required so that program could be validated properly. 1552 * Programs of types that are allowed to be loaded both w/ and w/o (for 1553 * backward compatibility) expected_attach_type, should have the default attach 1554 * type assigned to expected_attach_type for the latter case, so that it can be 1555 * validated later at attach time. 1556 * 1557 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1558 * prog type requires it but has some attach types that have to be backward 1559 * compatible. 1560 */ 1561 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1562 { 1563 switch (attr->prog_type) { 1564 case BPF_PROG_TYPE_CGROUP_SOCK: 1565 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1566 * exist so checking for non-zero is the way to go here. 1567 */ 1568 if (!attr->expected_attach_type) 1569 attr->expected_attach_type = 1570 BPF_CGROUP_INET_SOCK_CREATE; 1571 break; 1572 } 1573 } 1574 1575 static int 1576 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type, 1577 enum bpf_attach_type expected_attach_type) 1578 { 1579 switch (prog_type) { 1580 case BPF_PROG_TYPE_CGROUP_SOCK: 1581 switch (expected_attach_type) { 1582 case BPF_CGROUP_INET_SOCK_CREATE: 1583 case BPF_CGROUP_INET4_POST_BIND: 1584 case BPF_CGROUP_INET6_POST_BIND: 1585 return 0; 1586 default: 1587 return -EINVAL; 1588 } 1589 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1590 switch (expected_attach_type) { 1591 case BPF_CGROUP_INET4_BIND: 1592 case BPF_CGROUP_INET6_BIND: 1593 case BPF_CGROUP_INET4_CONNECT: 1594 case BPF_CGROUP_INET6_CONNECT: 1595 case BPF_CGROUP_UDP4_SENDMSG: 1596 case BPF_CGROUP_UDP6_SENDMSG: 1597 case BPF_CGROUP_UDP4_RECVMSG: 1598 case BPF_CGROUP_UDP6_RECVMSG: 1599 return 0; 1600 default: 1601 return -EINVAL; 1602 } 1603 case BPF_PROG_TYPE_CGROUP_SKB: 1604 switch (expected_attach_type) { 1605 case BPF_CGROUP_INET_INGRESS: 1606 case BPF_CGROUP_INET_EGRESS: 1607 return 0; 1608 default: 1609 return -EINVAL; 1610 } 1611 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 1612 switch (expected_attach_type) { 1613 case BPF_CGROUP_SETSOCKOPT: 1614 case BPF_CGROUP_GETSOCKOPT: 1615 return 0; 1616 default: 1617 return -EINVAL; 1618 } 1619 default: 1620 return 0; 1621 } 1622 } 1623 1624 /* last field in 'union bpf_attr' used by this command */ 1625 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt 1626 1627 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr) 1628 { 1629 enum bpf_prog_type type = attr->prog_type; 1630 struct bpf_prog *prog; 1631 int err; 1632 char license[128]; 1633 bool is_gpl; 1634 1635 if (CHECK_ATTR(BPF_PROG_LOAD)) 1636 return -EINVAL; 1637 1638 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | 1639 BPF_F_ANY_ALIGNMENT | 1640 BPF_F_TEST_STATE_FREQ | 1641 BPF_F_TEST_RND_HI32)) 1642 return -EINVAL; 1643 1644 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && 1645 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) && 1646 !capable(CAP_SYS_ADMIN)) 1647 return -EPERM; 1648 1649 /* copy eBPF program license from user space */ 1650 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 1651 sizeof(license) - 1) < 0) 1652 return -EFAULT; 1653 license[sizeof(license) - 1] = 0; 1654 1655 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1656 is_gpl = license_is_gpl_compatible(license); 1657 1658 if (attr->insn_cnt == 0 || 1659 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) 1660 return -E2BIG; 1661 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 1662 type != BPF_PROG_TYPE_CGROUP_SKB && 1663 !capable(CAP_SYS_ADMIN)) 1664 return -EPERM; 1665 1666 bpf_prog_load_fixup_attach_type(attr); 1667 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type)) 1668 return -EINVAL; 1669 1670 /* plain bpf_prog allocation */ 1671 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 1672 if (!prog) 1673 return -ENOMEM; 1674 1675 prog->expected_attach_type = attr->expected_attach_type; 1676 1677 prog->aux->offload_requested = !!attr->prog_ifindex; 1678 1679 err = security_bpf_prog_alloc(prog->aux); 1680 if (err) 1681 goto free_prog_nouncharge; 1682 1683 err = bpf_prog_charge_memlock(prog); 1684 if (err) 1685 goto free_prog_sec; 1686 1687 prog->len = attr->insn_cnt; 1688 1689 err = -EFAULT; 1690 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 1691 bpf_prog_insn_size(prog)) != 0) 1692 goto free_prog; 1693 1694 prog->orig_prog = NULL; 1695 prog->jited = 0; 1696 1697 atomic_set(&prog->aux->refcnt, 1); 1698 prog->gpl_compatible = is_gpl ? 1 : 0; 1699 1700 if (bpf_prog_is_dev_bound(prog->aux)) { 1701 err = bpf_prog_offload_init(prog, attr); 1702 if (err) 1703 goto free_prog; 1704 } 1705 1706 /* find program type: socket_filter vs tracing_filter */ 1707 err = find_prog_type(type, prog); 1708 if (err < 0) 1709 goto free_prog; 1710 1711 prog->aux->load_time = ktime_get_boottime_ns(); 1712 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 1713 if (err) 1714 goto free_prog; 1715 1716 /* run eBPF verifier */ 1717 err = bpf_check(&prog, attr, uattr); 1718 if (err < 0) 1719 goto free_used_maps; 1720 1721 prog = bpf_prog_select_runtime(prog, &err); 1722 if (err < 0) 1723 goto free_used_maps; 1724 1725 err = bpf_prog_alloc_id(prog); 1726 if (err) 1727 goto free_used_maps; 1728 1729 /* Upon success of bpf_prog_alloc_id(), the BPF prog is 1730 * effectively publicly exposed. However, retrieving via 1731 * bpf_prog_get_fd_by_id() will take another reference, 1732 * therefore it cannot be gone underneath us. 1733 * 1734 * Only for the time /after/ successful bpf_prog_new_fd() 1735 * and before returning to userspace, we might just hold 1736 * one reference and any parallel close on that fd could 1737 * rip everything out. Hence, below notifications must 1738 * happen before bpf_prog_new_fd(). 1739 * 1740 * Also, any failure handling from this point onwards must 1741 * be using bpf_prog_put() given the program is exposed. 1742 */ 1743 bpf_prog_kallsyms_add(prog); 1744 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0); 1745 1746 err = bpf_prog_new_fd(prog); 1747 if (err < 0) 1748 bpf_prog_put(prog); 1749 return err; 1750 1751 free_used_maps: 1752 /* In case we have subprogs, we need to wait for a grace 1753 * period before we can tear down JIT memory since symbols 1754 * are already exposed under kallsyms. 1755 */ 1756 __bpf_prog_put_noref(prog, prog->aux->func_cnt); 1757 return err; 1758 free_prog: 1759 bpf_prog_uncharge_memlock(prog); 1760 free_prog_sec: 1761 security_bpf_prog_free(prog->aux); 1762 free_prog_nouncharge: 1763 bpf_prog_free(prog); 1764 return err; 1765 } 1766 1767 #define BPF_OBJ_LAST_FIELD file_flags 1768 1769 static int bpf_obj_pin(const union bpf_attr *attr) 1770 { 1771 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 1772 return -EINVAL; 1773 1774 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 1775 } 1776 1777 static int bpf_obj_get(const union bpf_attr *attr) 1778 { 1779 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 1780 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 1781 return -EINVAL; 1782 1783 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 1784 attr->file_flags); 1785 } 1786 1787 struct bpf_raw_tracepoint { 1788 struct bpf_raw_event_map *btp; 1789 struct bpf_prog *prog; 1790 }; 1791 1792 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp) 1793 { 1794 struct bpf_raw_tracepoint *raw_tp = filp->private_data; 1795 1796 if (raw_tp->prog) { 1797 bpf_probe_unregister(raw_tp->btp, raw_tp->prog); 1798 bpf_prog_put(raw_tp->prog); 1799 } 1800 bpf_put_raw_tracepoint(raw_tp->btp); 1801 kfree(raw_tp); 1802 return 0; 1803 } 1804 1805 static const struct file_operations bpf_raw_tp_fops = { 1806 .release = bpf_raw_tracepoint_release, 1807 .read = bpf_dummy_read, 1808 .write = bpf_dummy_write, 1809 }; 1810 1811 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 1812 1813 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 1814 { 1815 struct bpf_raw_tracepoint *raw_tp; 1816 struct bpf_raw_event_map *btp; 1817 struct bpf_prog *prog; 1818 char tp_name[128]; 1819 int tp_fd, err; 1820 1821 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name), 1822 sizeof(tp_name) - 1) < 0) 1823 return -EFAULT; 1824 tp_name[sizeof(tp_name) - 1] = 0; 1825 1826 btp = bpf_get_raw_tracepoint(tp_name); 1827 if (!btp) 1828 return -ENOENT; 1829 1830 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER); 1831 if (!raw_tp) { 1832 err = -ENOMEM; 1833 goto out_put_btp; 1834 } 1835 raw_tp->btp = btp; 1836 1837 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd); 1838 if (IS_ERR(prog)) { 1839 err = PTR_ERR(prog); 1840 goto out_free_tp; 1841 } 1842 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT && 1843 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) { 1844 err = -EINVAL; 1845 goto out_put_prog; 1846 } 1847 1848 err = bpf_probe_register(raw_tp->btp, prog); 1849 if (err) 1850 goto out_put_prog; 1851 1852 raw_tp->prog = prog; 1853 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp, 1854 O_CLOEXEC); 1855 if (tp_fd < 0) { 1856 bpf_probe_unregister(raw_tp->btp, prog); 1857 err = tp_fd; 1858 goto out_put_prog; 1859 } 1860 return tp_fd; 1861 1862 out_put_prog: 1863 bpf_prog_put(prog); 1864 out_free_tp: 1865 kfree(raw_tp); 1866 out_put_btp: 1867 bpf_put_raw_tracepoint(btp); 1868 return err; 1869 } 1870 1871 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 1872 enum bpf_attach_type attach_type) 1873 { 1874 switch (prog->type) { 1875 case BPF_PROG_TYPE_CGROUP_SOCK: 1876 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1877 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 1878 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 1879 case BPF_PROG_TYPE_CGROUP_SKB: 1880 return prog->enforce_expected_attach_type && 1881 prog->expected_attach_type != attach_type ? 1882 -EINVAL : 0; 1883 default: 1884 return 0; 1885 } 1886 } 1887 1888 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags 1889 1890 #define BPF_F_ATTACH_MASK \ 1891 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI) 1892 1893 static int bpf_prog_attach(const union bpf_attr *attr) 1894 { 1895 enum bpf_prog_type ptype; 1896 struct bpf_prog *prog; 1897 int ret; 1898 1899 if (!capable(CAP_NET_ADMIN)) 1900 return -EPERM; 1901 1902 if (CHECK_ATTR(BPF_PROG_ATTACH)) 1903 return -EINVAL; 1904 1905 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 1906 return -EINVAL; 1907 1908 switch (attr->attach_type) { 1909 case BPF_CGROUP_INET_INGRESS: 1910 case BPF_CGROUP_INET_EGRESS: 1911 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1912 break; 1913 case BPF_CGROUP_INET_SOCK_CREATE: 1914 case BPF_CGROUP_INET4_POST_BIND: 1915 case BPF_CGROUP_INET6_POST_BIND: 1916 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1917 break; 1918 case BPF_CGROUP_INET4_BIND: 1919 case BPF_CGROUP_INET6_BIND: 1920 case BPF_CGROUP_INET4_CONNECT: 1921 case BPF_CGROUP_INET6_CONNECT: 1922 case BPF_CGROUP_UDP4_SENDMSG: 1923 case BPF_CGROUP_UDP6_SENDMSG: 1924 case BPF_CGROUP_UDP4_RECVMSG: 1925 case BPF_CGROUP_UDP6_RECVMSG: 1926 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1927 break; 1928 case BPF_CGROUP_SOCK_OPS: 1929 ptype = BPF_PROG_TYPE_SOCK_OPS; 1930 break; 1931 case BPF_CGROUP_DEVICE: 1932 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1933 break; 1934 case BPF_SK_MSG_VERDICT: 1935 ptype = BPF_PROG_TYPE_SK_MSG; 1936 break; 1937 case BPF_SK_SKB_STREAM_PARSER: 1938 case BPF_SK_SKB_STREAM_VERDICT: 1939 ptype = BPF_PROG_TYPE_SK_SKB; 1940 break; 1941 case BPF_LIRC_MODE2: 1942 ptype = BPF_PROG_TYPE_LIRC_MODE2; 1943 break; 1944 case BPF_FLOW_DISSECTOR: 1945 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 1946 break; 1947 case BPF_CGROUP_SYSCTL: 1948 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 1949 break; 1950 case BPF_CGROUP_GETSOCKOPT: 1951 case BPF_CGROUP_SETSOCKOPT: 1952 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 1953 break; 1954 default: 1955 return -EINVAL; 1956 } 1957 1958 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1959 if (IS_ERR(prog)) 1960 return PTR_ERR(prog); 1961 1962 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 1963 bpf_prog_put(prog); 1964 return -EINVAL; 1965 } 1966 1967 switch (ptype) { 1968 case BPF_PROG_TYPE_SK_SKB: 1969 case BPF_PROG_TYPE_SK_MSG: 1970 ret = sock_map_get_from_fd(attr, prog); 1971 break; 1972 case BPF_PROG_TYPE_LIRC_MODE2: 1973 ret = lirc_prog_attach(attr, prog); 1974 break; 1975 case BPF_PROG_TYPE_FLOW_DISSECTOR: 1976 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 1977 break; 1978 default: 1979 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 1980 } 1981 1982 if (ret) 1983 bpf_prog_put(prog); 1984 return ret; 1985 } 1986 1987 #define BPF_PROG_DETACH_LAST_FIELD attach_type 1988 1989 static int bpf_prog_detach(const union bpf_attr *attr) 1990 { 1991 enum bpf_prog_type ptype; 1992 1993 if (!capable(CAP_NET_ADMIN)) 1994 return -EPERM; 1995 1996 if (CHECK_ATTR(BPF_PROG_DETACH)) 1997 return -EINVAL; 1998 1999 switch (attr->attach_type) { 2000 case BPF_CGROUP_INET_INGRESS: 2001 case BPF_CGROUP_INET_EGRESS: 2002 ptype = BPF_PROG_TYPE_CGROUP_SKB; 2003 break; 2004 case BPF_CGROUP_INET_SOCK_CREATE: 2005 case BPF_CGROUP_INET4_POST_BIND: 2006 case BPF_CGROUP_INET6_POST_BIND: 2007 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 2008 break; 2009 case BPF_CGROUP_INET4_BIND: 2010 case BPF_CGROUP_INET6_BIND: 2011 case BPF_CGROUP_INET4_CONNECT: 2012 case BPF_CGROUP_INET6_CONNECT: 2013 case BPF_CGROUP_UDP4_SENDMSG: 2014 case BPF_CGROUP_UDP6_SENDMSG: 2015 case BPF_CGROUP_UDP4_RECVMSG: 2016 case BPF_CGROUP_UDP6_RECVMSG: 2017 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 2018 break; 2019 case BPF_CGROUP_SOCK_OPS: 2020 ptype = BPF_PROG_TYPE_SOCK_OPS; 2021 break; 2022 case BPF_CGROUP_DEVICE: 2023 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 2024 break; 2025 case BPF_SK_MSG_VERDICT: 2026 return sock_map_get_from_fd(attr, NULL); 2027 case BPF_SK_SKB_STREAM_PARSER: 2028 case BPF_SK_SKB_STREAM_VERDICT: 2029 return sock_map_get_from_fd(attr, NULL); 2030 case BPF_LIRC_MODE2: 2031 return lirc_prog_detach(attr); 2032 case BPF_FLOW_DISSECTOR: 2033 return skb_flow_dissector_bpf_prog_detach(attr); 2034 case BPF_CGROUP_SYSCTL: 2035 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 2036 break; 2037 case BPF_CGROUP_GETSOCKOPT: 2038 case BPF_CGROUP_SETSOCKOPT: 2039 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 2040 break; 2041 default: 2042 return -EINVAL; 2043 } 2044 2045 return cgroup_bpf_prog_detach(attr, ptype); 2046 } 2047 2048 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 2049 2050 static int bpf_prog_query(const union bpf_attr *attr, 2051 union bpf_attr __user *uattr) 2052 { 2053 if (!capable(CAP_NET_ADMIN)) 2054 return -EPERM; 2055 if (CHECK_ATTR(BPF_PROG_QUERY)) 2056 return -EINVAL; 2057 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 2058 return -EINVAL; 2059 2060 switch (attr->query.attach_type) { 2061 case BPF_CGROUP_INET_INGRESS: 2062 case BPF_CGROUP_INET_EGRESS: 2063 case BPF_CGROUP_INET_SOCK_CREATE: 2064 case BPF_CGROUP_INET4_BIND: 2065 case BPF_CGROUP_INET6_BIND: 2066 case BPF_CGROUP_INET4_POST_BIND: 2067 case BPF_CGROUP_INET6_POST_BIND: 2068 case BPF_CGROUP_INET4_CONNECT: 2069 case BPF_CGROUP_INET6_CONNECT: 2070 case BPF_CGROUP_UDP4_SENDMSG: 2071 case BPF_CGROUP_UDP6_SENDMSG: 2072 case BPF_CGROUP_UDP4_RECVMSG: 2073 case BPF_CGROUP_UDP6_RECVMSG: 2074 case BPF_CGROUP_SOCK_OPS: 2075 case BPF_CGROUP_DEVICE: 2076 case BPF_CGROUP_SYSCTL: 2077 case BPF_CGROUP_GETSOCKOPT: 2078 case BPF_CGROUP_SETSOCKOPT: 2079 break; 2080 case BPF_LIRC_MODE2: 2081 return lirc_prog_query(attr, uattr); 2082 case BPF_FLOW_DISSECTOR: 2083 return skb_flow_dissector_prog_query(attr, uattr); 2084 default: 2085 return -EINVAL; 2086 } 2087 2088 return cgroup_bpf_prog_query(attr, uattr); 2089 } 2090 2091 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out 2092 2093 static int bpf_prog_test_run(const union bpf_attr *attr, 2094 union bpf_attr __user *uattr) 2095 { 2096 struct bpf_prog *prog; 2097 int ret = -ENOTSUPP; 2098 2099 if (!capable(CAP_SYS_ADMIN)) 2100 return -EPERM; 2101 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 2102 return -EINVAL; 2103 2104 if ((attr->test.ctx_size_in && !attr->test.ctx_in) || 2105 (!attr->test.ctx_size_in && attr->test.ctx_in)) 2106 return -EINVAL; 2107 2108 if ((attr->test.ctx_size_out && !attr->test.ctx_out) || 2109 (!attr->test.ctx_size_out && attr->test.ctx_out)) 2110 return -EINVAL; 2111 2112 prog = bpf_prog_get(attr->test.prog_fd); 2113 if (IS_ERR(prog)) 2114 return PTR_ERR(prog); 2115 2116 if (prog->aux->ops->test_run) 2117 ret = prog->aux->ops->test_run(prog, attr, uattr); 2118 2119 bpf_prog_put(prog); 2120 return ret; 2121 } 2122 2123 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 2124 2125 static int bpf_obj_get_next_id(const union bpf_attr *attr, 2126 union bpf_attr __user *uattr, 2127 struct idr *idr, 2128 spinlock_t *lock) 2129 { 2130 u32 next_id = attr->start_id; 2131 int err = 0; 2132 2133 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 2134 return -EINVAL; 2135 2136 if (!capable(CAP_SYS_ADMIN)) 2137 return -EPERM; 2138 2139 next_id++; 2140 spin_lock_bh(lock); 2141 if (!idr_get_next(idr, &next_id)) 2142 err = -ENOENT; 2143 spin_unlock_bh(lock); 2144 2145 if (!err) 2146 err = put_user(next_id, &uattr->next_id); 2147 2148 return err; 2149 } 2150 2151 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 2152 2153 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 2154 { 2155 struct bpf_prog *prog; 2156 u32 id = attr->prog_id; 2157 int fd; 2158 2159 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 2160 return -EINVAL; 2161 2162 if (!capable(CAP_SYS_ADMIN)) 2163 return -EPERM; 2164 2165 spin_lock_bh(&prog_idr_lock); 2166 prog = idr_find(&prog_idr, id); 2167 if (prog) 2168 prog = bpf_prog_inc_not_zero(prog); 2169 else 2170 prog = ERR_PTR(-ENOENT); 2171 spin_unlock_bh(&prog_idr_lock); 2172 2173 if (IS_ERR(prog)) 2174 return PTR_ERR(prog); 2175 2176 fd = bpf_prog_new_fd(prog); 2177 if (fd < 0) 2178 bpf_prog_put(prog); 2179 2180 return fd; 2181 } 2182 2183 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 2184 2185 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 2186 { 2187 struct bpf_map *map; 2188 u32 id = attr->map_id; 2189 int f_flags; 2190 int fd; 2191 2192 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 2193 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 2194 return -EINVAL; 2195 2196 if (!capable(CAP_SYS_ADMIN)) 2197 return -EPERM; 2198 2199 f_flags = bpf_get_file_flag(attr->open_flags); 2200 if (f_flags < 0) 2201 return f_flags; 2202 2203 spin_lock_bh(&map_idr_lock); 2204 map = idr_find(&map_idr, id); 2205 if (map) 2206 map = __bpf_map_inc_not_zero(map, true); 2207 else 2208 map = ERR_PTR(-ENOENT); 2209 spin_unlock_bh(&map_idr_lock); 2210 2211 if (IS_ERR(map)) 2212 return PTR_ERR(map); 2213 2214 fd = bpf_map_new_fd(map, f_flags); 2215 if (fd < 0) 2216 bpf_map_put_with_uref(map); 2217 2218 return fd; 2219 } 2220 2221 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 2222 unsigned long addr, u32 *off, 2223 u32 *type) 2224 { 2225 const struct bpf_map *map; 2226 int i; 2227 2228 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) { 2229 map = prog->aux->used_maps[i]; 2230 if (map == (void *)addr) { 2231 *type = BPF_PSEUDO_MAP_FD; 2232 return map; 2233 } 2234 if (!map->ops->map_direct_value_meta) 2235 continue; 2236 if (!map->ops->map_direct_value_meta(map, addr, off)) { 2237 *type = BPF_PSEUDO_MAP_VALUE; 2238 return map; 2239 } 2240 } 2241 2242 return NULL; 2243 } 2244 2245 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 2246 { 2247 const struct bpf_map *map; 2248 struct bpf_insn *insns; 2249 u32 off, type; 2250 u64 imm; 2251 int i; 2252 2253 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 2254 GFP_USER); 2255 if (!insns) 2256 return insns; 2257 2258 for (i = 0; i < prog->len; i++) { 2259 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 2260 insns[i].code = BPF_JMP | BPF_CALL; 2261 insns[i].imm = BPF_FUNC_tail_call; 2262 /* fall-through */ 2263 } 2264 if (insns[i].code == (BPF_JMP | BPF_CALL) || 2265 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 2266 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 2267 insns[i].code = BPF_JMP | BPF_CALL; 2268 if (!bpf_dump_raw_ok()) 2269 insns[i].imm = 0; 2270 continue; 2271 } 2272 2273 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 2274 continue; 2275 2276 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 2277 map = bpf_map_from_imm(prog, imm, &off, &type); 2278 if (map) { 2279 insns[i].src_reg = type; 2280 insns[i].imm = map->id; 2281 insns[i + 1].imm = off; 2282 continue; 2283 } 2284 } 2285 2286 return insns; 2287 } 2288 2289 static int set_info_rec_size(struct bpf_prog_info *info) 2290 { 2291 /* 2292 * Ensure info.*_rec_size is the same as kernel expected size 2293 * 2294 * or 2295 * 2296 * Only allow zero *_rec_size if both _rec_size and _cnt are 2297 * zero. In this case, the kernel will set the expected 2298 * _rec_size back to the info. 2299 */ 2300 2301 if ((info->nr_func_info || info->func_info_rec_size) && 2302 info->func_info_rec_size != sizeof(struct bpf_func_info)) 2303 return -EINVAL; 2304 2305 if ((info->nr_line_info || info->line_info_rec_size) && 2306 info->line_info_rec_size != sizeof(struct bpf_line_info)) 2307 return -EINVAL; 2308 2309 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) && 2310 info->jited_line_info_rec_size != sizeof(__u64)) 2311 return -EINVAL; 2312 2313 info->func_info_rec_size = sizeof(struct bpf_func_info); 2314 info->line_info_rec_size = sizeof(struct bpf_line_info); 2315 info->jited_line_info_rec_size = sizeof(__u64); 2316 2317 return 0; 2318 } 2319 2320 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 2321 const union bpf_attr *attr, 2322 union bpf_attr __user *uattr) 2323 { 2324 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2325 struct bpf_prog_info info = {}; 2326 u32 info_len = attr->info.info_len; 2327 struct bpf_prog_stats stats; 2328 char __user *uinsns; 2329 u32 ulen; 2330 int err; 2331 2332 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2333 if (err) 2334 return err; 2335 info_len = min_t(u32, sizeof(info), info_len); 2336 2337 if (copy_from_user(&info, uinfo, info_len)) 2338 return -EFAULT; 2339 2340 info.type = prog->type; 2341 info.id = prog->aux->id; 2342 info.load_time = prog->aux->load_time; 2343 info.created_by_uid = from_kuid_munged(current_user_ns(), 2344 prog->aux->user->uid); 2345 info.gpl_compatible = prog->gpl_compatible; 2346 2347 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 2348 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 2349 2350 ulen = info.nr_map_ids; 2351 info.nr_map_ids = prog->aux->used_map_cnt; 2352 ulen = min_t(u32, info.nr_map_ids, ulen); 2353 if (ulen) { 2354 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 2355 u32 i; 2356 2357 for (i = 0; i < ulen; i++) 2358 if (put_user(prog->aux->used_maps[i]->id, 2359 &user_map_ids[i])) 2360 return -EFAULT; 2361 } 2362 2363 err = set_info_rec_size(&info); 2364 if (err) 2365 return err; 2366 2367 bpf_prog_get_stats(prog, &stats); 2368 info.run_time_ns = stats.nsecs; 2369 info.run_cnt = stats.cnt; 2370 2371 if (!capable(CAP_SYS_ADMIN)) { 2372 info.jited_prog_len = 0; 2373 info.xlated_prog_len = 0; 2374 info.nr_jited_ksyms = 0; 2375 info.nr_jited_func_lens = 0; 2376 info.nr_func_info = 0; 2377 info.nr_line_info = 0; 2378 info.nr_jited_line_info = 0; 2379 goto done; 2380 } 2381 2382 ulen = info.xlated_prog_len; 2383 info.xlated_prog_len = bpf_prog_insn_size(prog); 2384 if (info.xlated_prog_len && ulen) { 2385 struct bpf_insn *insns_sanitized; 2386 bool fault; 2387 2388 if (prog->blinded && !bpf_dump_raw_ok()) { 2389 info.xlated_prog_insns = 0; 2390 goto done; 2391 } 2392 insns_sanitized = bpf_insn_prepare_dump(prog); 2393 if (!insns_sanitized) 2394 return -ENOMEM; 2395 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 2396 ulen = min_t(u32, info.xlated_prog_len, ulen); 2397 fault = copy_to_user(uinsns, insns_sanitized, ulen); 2398 kfree(insns_sanitized); 2399 if (fault) 2400 return -EFAULT; 2401 } 2402 2403 if (bpf_prog_is_dev_bound(prog->aux)) { 2404 err = bpf_prog_offload_info_fill(&info, prog); 2405 if (err) 2406 return err; 2407 goto done; 2408 } 2409 2410 /* NOTE: the following code is supposed to be skipped for offload. 2411 * bpf_prog_offload_info_fill() is the place to fill similar fields 2412 * for offload. 2413 */ 2414 ulen = info.jited_prog_len; 2415 if (prog->aux->func_cnt) { 2416 u32 i; 2417 2418 info.jited_prog_len = 0; 2419 for (i = 0; i < prog->aux->func_cnt; i++) 2420 info.jited_prog_len += prog->aux->func[i]->jited_len; 2421 } else { 2422 info.jited_prog_len = prog->jited_len; 2423 } 2424 2425 if (info.jited_prog_len && ulen) { 2426 if (bpf_dump_raw_ok()) { 2427 uinsns = u64_to_user_ptr(info.jited_prog_insns); 2428 ulen = min_t(u32, info.jited_prog_len, ulen); 2429 2430 /* for multi-function programs, copy the JITed 2431 * instructions for all the functions 2432 */ 2433 if (prog->aux->func_cnt) { 2434 u32 len, free, i; 2435 u8 *img; 2436 2437 free = ulen; 2438 for (i = 0; i < prog->aux->func_cnt; i++) { 2439 len = prog->aux->func[i]->jited_len; 2440 len = min_t(u32, len, free); 2441 img = (u8 *) prog->aux->func[i]->bpf_func; 2442 if (copy_to_user(uinsns, img, len)) 2443 return -EFAULT; 2444 uinsns += len; 2445 free -= len; 2446 if (!free) 2447 break; 2448 } 2449 } else { 2450 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 2451 return -EFAULT; 2452 } 2453 } else { 2454 info.jited_prog_insns = 0; 2455 } 2456 } 2457 2458 ulen = info.nr_jited_ksyms; 2459 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1; 2460 if (ulen) { 2461 if (bpf_dump_raw_ok()) { 2462 unsigned long ksym_addr; 2463 u64 __user *user_ksyms; 2464 u32 i; 2465 2466 /* copy the address of the kernel symbol 2467 * corresponding to each function 2468 */ 2469 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 2470 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 2471 if (prog->aux->func_cnt) { 2472 for (i = 0; i < ulen; i++) { 2473 ksym_addr = (unsigned long) 2474 prog->aux->func[i]->bpf_func; 2475 if (put_user((u64) ksym_addr, 2476 &user_ksyms[i])) 2477 return -EFAULT; 2478 } 2479 } else { 2480 ksym_addr = (unsigned long) prog->bpf_func; 2481 if (put_user((u64) ksym_addr, &user_ksyms[0])) 2482 return -EFAULT; 2483 } 2484 } else { 2485 info.jited_ksyms = 0; 2486 } 2487 } 2488 2489 ulen = info.nr_jited_func_lens; 2490 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1; 2491 if (ulen) { 2492 if (bpf_dump_raw_ok()) { 2493 u32 __user *user_lens; 2494 u32 func_len, i; 2495 2496 /* copy the JITed image lengths for each function */ 2497 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 2498 user_lens = u64_to_user_ptr(info.jited_func_lens); 2499 if (prog->aux->func_cnt) { 2500 for (i = 0; i < ulen; i++) { 2501 func_len = 2502 prog->aux->func[i]->jited_len; 2503 if (put_user(func_len, &user_lens[i])) 2504 return -EFAULT; 2505 } 2506 } else { 2507 func_len = prog->jited_len; 2508 if (put_user(func_len, &user_lens[0])) 2509 return -EFAULT; 2510 } 2511 } else { 2512 info.jited_func_lens = 0; 2513 } 2514 } 2515 2516 if (prog->aux->btf) 2517 info.btf_id = btf_id(prog->aux->btf); 2518 2519 ulen = info.nr_func_info; 2520 info.nr_func_info = prog->aux->func_info_cnt; 2521 if (info.nr_func_info && ulen) { 2522 char __user *user_finfo; 2523 2524 user_finfo = u64_to_user_ptr(info.func_info); 2525 ulen = min_t(u32, info.nr_func_info, ulen); 2526 if (copy_to_user(user_finfo, prog->aux->func_info, 2527 info.func_info_rec_size * ulen)) 2528 return -EFAULT; 2529 } 2530 2531 ulen = info.nr_line_info; 2532 info.nr_line_info = prog->aux->nr_linfo; 2533 if (info.nr_line_info && ulen) { 2534 __u8 __user *user_linfo; 2535 2536 user_linfo = u64_to_user_ptr(info.line_info); 2537 ulen = min_t(u32, info.nr_line_info, ulen); 2538 if (copy_to_user(user_linfo, prog->aux->linfo, 2539 info.line_info_rec_size * ulen)) 2540 return -EFAULT; 2541 } 2542 2543 ulen = info.nr_jited_line_info; 2544 if (prog->aux->jited_linfo) 2545 info.nr_jited_line_info = prog->aux->nr_linfo; 2546 else 2547 info.nr_jited_line_info = 0; 2548 if (info.nr_jited_line_info && ulen) { 2549 if (bpf_dump_raw_ok()) { 2550 __u64 __user *user_linfo; 2551 u32 i; 2552 2553 user_linfo = u64_to_user_ptr(info.jited_line_info); 2554 ulen = min_t(u32, info.nr_jited_line_info, ulen); 2555 for (i = 0; i < ulen; i++) { 2556 if (put_user((__u64)(long)prog->aux->jited_linfo[i], 2557 &user_linfo[i])) 2558 return -EFAULT; 2559 } 2560 } else { 2561 info.jited_line_info = 0; 2562 } 2563 } 2564 2565 ulen = info.nr_prog_tags; 2566 info.nr_prog_tags = prog->aux->func_cnt ? : 1; 2567 if (ulen) { 2568 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE]; 2569 u32 i; 2570 2571 user_prog_tags = u64_to_user_ptr(info.prog_tags); 2572 ulen = min_t(u32, info.nr_prog_tags, ulen); 2573 if (prog->aux->func_cnt) { 2574 for (i = 0; i < ulen; i++) { 2575 if (copy_to_user(user_prog_tags[i], 2576 prog->aux->func[i]->tag, 2577 BPF_TAG_SIZE)) 2578 return -EFAULT; 2579 } 2580 } else { 2581 if (copy_to_user(user_prog_tags[0], 2582 prog->tag, BPF_TAG_SIZE)) 2583 return -EFAULT; 2584 } 2585 } 2586 2587 done: 2588 if (copy_to_user(uinfo, &info, info_len) || 2589 put_user(info_len, &uattr->info.info_len)) 2590 return -EFAULT; 2591 2592 return 0; 2593 } 2594 2595 static int bpf_map_get_info_by_fd(struct bpf_map *map, 2596 const union bpf_attr *attr, 2597 union bpf_attr __user *uattr) 2598 { 2599 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2600 struct bpf_map_info info = {}; 2601 u32 info_len = attr->info.info_len; 2602 int err; 2603 2604 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2605 if (err) 2606 return err; 2607 info_len = min_t(u32, sizeof(info), info_len); 2608 2609 info.type = map->map_type; 2610 info.id = map->id; 2611 info.key_size = map->key_size; 2612 info.value_size = map->value_size; 2613 info.max_entries = map->max_entries; 2614 info.map_flags = map->map_flags; 2615 memcpy(info.name, map->name, sizeof(map->name)); 2616 2617 if (map->btf) { 2618 info.btf_id = btf_id(map->btf); 2619 info.btf_key_type_id = map->btf_key_type_id; 2620 info.btf_value_type_id = map->btf_value_type_id; 2621 } 2622 2623 if (bpf_map_is_dev_bound(map)) { 2624 err = bpf_map_offload_info_fill(&info, map); 2625 if (err) 2626 return err; 2627 } 2628 2629 if (copy_to_user(uinfo, &info, info_len) || 2630 put_user(info_len, &uattr->info.info_len)) 2631 return -EFAULT; 2632 2633 return 0; 2634 } 2635 2636 static int bpf_btf_get_info_by_fd(struct btf *btf, 2637 const union bpf_attr *attr, 2638 union bpf_attr __user *uattr) 2639 { 2640 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2641 u32 info_len = attr->info.info_len; 2642 int err; 2643 2644 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 2645 if (err) 2646 return err; 2647 2648 return btf_get_info_by_fd(btf, attr, uattr); 2649 } 2650 2651 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 2652 2653 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 2654 union bpf_attr __user *uattr) 2655 { 2656 int ufd = attr->info.bpf_fd; 2657 struct fd f; 2658 int err; 2659 2660 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 2661 return -EINVAL; 2662 2663 f = fdget(ufd); 2664 if (!f.file) 2665 return -EBADFD; 2666 2667 if (f.file->f_op == &bpf_prog_fops) 2668 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 2669 uattr); 2670 else if (f.file->f_op == &bpf_map_fops) 2671 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 2672 uattr); 2673 else if (f.file->f_op == &btf_fops) 2674 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 2675 else 2676 err = -EINVAL; 2677 2678 fdput(f); 2679 return err; 2680 } 2681 2682 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 2683 2684 static int bpf_btf_load(const union bpf_attr *attr) 2685 { 2686 if (CHECK_ATTR(BPF_BTF_LOAD)) 2687 return -EINVAL; 2688 2689 if (!capable(CAP_SYS_ADMIN)) 2690 return -EPERM; 2691 2692 return btf_new_fd(attr); 2693 } 2694 2695 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 2696 2697 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 2698 { 2699 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 2700 return -EINVAL; 2701 2702 if (!capable(CAP_SYS_ADMIN)) 2703 return -EPERM; 2704 2705 return btf_get_fd_by_id(attr->btf_id); 2706 } 2707 2708 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 2709 union bpf_attr __user *uattr, 2710 u32 prog_id, u32 fd_type, 2711 const char *buf, u64 probe_offset, 2712 u64 probe_addr) 2713 { 2714 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 2715 u32 len = buf ? strlen(buf) : 0, input_len; 2716 int err = 0; 2717 2718 if (put_user(len, &uattr->task_fd_query.buf_len)) 2719 return -EFAULT; 2720 input_len = attr->task_fd_query.buf_len; 2721 if (input_len && ubuf) { 2722 if (!len) { 2723 /* nothing to copy, just make ubuf NULL terminated */ 2724 char zero = '\0'; 2725 2726 if (put_user(zero, ubuf)) 2727 return -EFAULT; 2728 } else if (input_len >= len + 1) { 2729 /* ubuf can hold the string with NULL terminator */ 2730 if (copy_to_user(ubuf, buf, len + 1)) 2731 return -EFAULT; 2732 } else { 2733 /* ubuf cannot hold the string with NULL terminator, 2734 * do a partial copy with NULL terminator. 2735 */ 2736 char zero = '\0'; 2737 2738 err = -ENOSPC; 2739 if (copy_to_user(ubuf, buf, input_len - 1)) 2740 return -EFAULT; 2741 if (put_user(zero, ubuf + input_len - 1)) 2742 return -EFAULT; 2743 } 2744 } 2745 2746 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 2747 put_user(fd_type, &uattr->task_fd_query.fd_type) || 2748 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 2749 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 2750 return -EFAULT; 2751 2752 return err; 2753 } 2754 2755 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 2756 2757 static int bpf_task_fd_query(const union bpf_attr *attr, 2758 union bpf_attr __user *uattr) 2759 { 2760 pid_t pid = attr->task_fd_query.pid; 2761 u32 fd = attr->task_fd_query.fd; 2762 const struct perf_event *event; 2763 struct files_struct *files; 2764 struct task_struct *task; 2765 struct file *file; 2766 int err; 2767 2768 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 2769 return -EINVAL; 2770 2771 if (!capable(CAP_SYS_ADMIN)) 2772 return -EPERM; 2773 2774 if (attr->task_fd_query.flags != 0) 2775 return -EINVAL; 2776 2777 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 2778 if (!task) 2779 return -ENOENT; 2780 2781 files = get_files_struct(task); 2782 put_task_struct(task); 2783 if (!files) 2784 return -ENOENT; 2785 2786 err = 0; 2787 spin_lock(&files->file_lock); 2788 file = fcheck_files(files, fd); 2789 if (!file) 2790 err = -EBADF; 2791 else 2792 get_file(file); 2793 spin_unlock(&files->file_lock); 2794 put_files_struct(files); 2795 2796 if (err) 2797 goto out; 2798 2799 if (file->f_op == &bpf_raw_tp_fops) { 2800 struct bpf_raw_tracepoint *raw_tp = file->private_data; 2801 struct bpf_raw_event_map *btp = raw_tp->btp; 2802 2803 err = bpf_task_fd_query_copy(attr, uattr, 2804 raw_tp->prog->aux->id, 2805 BPF_FD_TYPE_RAW_TRACEPOINT, 2806 btp->tp->name, 0, 0); 2807 goto put_file; 2808 } 2809 2810 event = perf_get_event(file); 2811 if (!IS_ERR(event)) { 2812 u64 probe_offset, probe_addr; 2813 u32 prog_id, fd_type; 2814 const char *buf; 2815 2816 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 2817 &buf, &probe_offset, 2818 &probe_addr); 2819 if (!err) 2820 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 2821 fd_type, buf, 2822 probe_offset, 2823 probe_addr); 2824 goto put_file; 2825 } 2826 2827 err = -ENOTSUPP; 2828 put_file: 2829 fput(file); 2830 out: 2831 return err; 2832 } 2833 2834 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 2835 { 2836 union bpf_attr attr = {}; 2837 int err; 2838 2839 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 2840 return -EPERM; 2841 2842 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 2843 if (err) 2844 return err; 2845 size = min_t(u32, size, sizeof(attr)); 2846 2847 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 2848 if (copy_from_user(&attr, uattr, size) != 0) 2849 return -EFAULT; 2850 2851 err = security_bpf(cmd, &attr, size); 2852 if (err < 0) 2853 return err; 2854 2855 switch (cmd) { 2856 case BPF_MAP_CREATE: 2857 err = map_create(&attr); 2858 break; 2859 case BPF_MAP_LOOKUP_ELEM: 2860 err = map_lookup_elem(&attr); 2861 break; 2862 case BPF_MAP_UPDATE_ELEM: 2863 err = map_update_elem(&attr); 2864 break; 2865 case BPF_MAP_DELETE_ELEM: 2866 err = map_delete_elem(&attr); 2867 break; 2868 case BPF_MAP_GET_NEXT_KEY: 2869 err = map_get_next_key(&attr); 2870 break; 2871 case BPF_MAP_FREEZE: 2872 err = map_freeze(&attr); 2873 break; 2874 case BPF_PROG_LOAD: 2875 err = bpf_prog_load(&attr, uattr); 2876 break; 2877 case BPF_OBJ_PIN: 2878 err = bpf_obj_pin(&attr); 2879 break; 2880 case BPF_OBJ_GET: 2881 err = bpf_obj_get(&attr); 2882 break; 2883 case BPF_PROG_ATTACH: 2884 err = bpf_prog_attach(&attr); 2885 break; 2886 case BPF_PROG_DETACH: 2887 err = bpf_prog_detach(&attr); 2888 break; 2889 case BPF_PROG_QUERY: 2890 err = bpf_prog_query(&attr, uattr); 2891 break; 2892 case BPF_PROG_TEST_RUN: 2893 err = bpf_prog_test_run(&attr, uattr); 2894 break; 2895 case BPF_PROG_GET_NEXT_ID: 2896 err = bpf_obj_get_next_id(&attr, uattr, 2897 &prog_idr, &prog_idr_lock); 2898 break; 2899 case BPF_MAP_GET_NEXT_ID: 2900 err = bpf_obj_get_next_id(&attr, uattr, 2901 &map_idr, &map_idr_lock); 2902 break; 2903 case BPF_BTF_GET_NEXT_ID: 2904 err = bpf_obj_get_next_id(&attr, uattr, 2905 &btf_idr, &btf_idr_lock); 2906 break; 2907 case BPF_PROG_GET_FD_BY_ID: 2908 err = bpf_prog_get_fd_by_id(&attr); 2909 break; 2910 case BPF_MAP_GET_FD_BY_ID: 2911 err = bpf_map_get_fd_by_id(&attr); 2912 break; 2913 case BPF_OBJ_GET_INFO_BY_FD: 2914 err = bpf_obj_get_info_by_fd(&attr, uattr); 2915 break; 2916 case BPF_RAW_TRACEPOINT_OPEN: 2917 err = bpf_raw_tracepoint_open(&attr); 2918 break; 2919 case BPF_BTF_LOAD: 2920 err = bpf_btf_load(&attr); 2921 break; 2922 case BPF_BTF_GET_FD_BY_ID: 2923 err = bpf_btf_get_fd_by_id(&attr); 2924 break; 2925 case BPF_TASK_FD_QUERY: 2926 err = bpf_task_fd_query(&attr, uattr); 2927 break; 2928 case BPF_MAP_LOOKUP_AND_DELETE_ELEM: 2929 err = map_lookup_and_delete_elem(&attr); 2930 break; 2931 default: 2932 err = -EINVAL; 2933 break; 2934 } 2935 2936 return err; 2937 } 2938