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(u64 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 >= SIZE_MAX) 145 return NULL; 146 147 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { 148 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags, 149 numa_node); 150 if (area != NULL) 151 return area; 152 } 153 154 return __vmalloc_node_flags_caller(size, numa_node, 155 GFP_KERNEL | __GFP_RETRY_MAYFAIL | 156 flags, __builtin_return_address(0)); 157 } 158 159 void bpf_map_area_free(void *area) 160 { 161 kvfree(area); 162 } 163 164 static u32 bpf_map_flags_retain_permanent(u32 flags) 165 { 166 /* Some map creation flags are not tied to the map object but 167 * rather to the map fd instead, so they have no meaning upon 168 * map object inspection since multiple file descriptors with 169 * different (access) properties can exist here. Thus, given 170 * this has zero meaning for the map itself, lets clear these 171 * from here. 172 */ 173 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY); 174 } 175 176 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr) 177 { 178 map->map_type = attr->map_type; 179 map->key_size = attr->key_size; 180 map->value_size = attr->value_size; 181 map->max_entries = attr->max_entries; 182 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags); 183 map->numa_node = bpf_map_attr_numa_node(attr); 184 } 185 186 static int bpf_charge_memlock(struct user_struct *user, u32 pages) 187 { 188 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 189 190 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) { 191 atomic_long_sub(pages, &user->locked_vm); 192 return -EPERM; 193 } 194 return 0; 195 } 196 197 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages) 198 { 199 if (user) 200 atomic_long_sub(pages, &user->locked_vm); 201 } 202 203 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size) 204 { 205 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT; 206 struct user_struct *user; 207 int ret; 208 209 if (size >= U32_MAX - PAGE_SIZE) 210 return -E2BIG; 211 212 user = get_current_user(); 213 ret = bpf_charge_memlock(user, pages); 214 if (ret) { 215 free_uid(user); 216 return ret; 217 } 218 219 mem->pages = pages; 220 mem->user = user; 221 222 return 0; 223 } 224 225 void bpf_map_charge_finish(struct bpf_map_memory *mem) 226 { 227 bpf_uncharge_memlock(mem->user, mem->pages); 228 free_uid(mem->user); 229 } 230 231 void bpf_map_charge_move(struct bpf_map_memory *dst, 232 struct bpf_map_memory *src) 233 { 234 *dst = *src; 235 236 /* Make sure src will not be used for the redundant uncharging. */ 237 memset(src, 0, sizeof(struct bpf_map_memory)); 238 } 239 240 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages) 241 { 242 int ret; 243 244 ret = bpf_charge_memlock(map->memory.user, pages); 245 if (ret) 246 return ret; 247 map->memory.pages += pages; 248 return ret; 249 } 250 251 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages) 252 { 253 bpf_uncharge_memlock(map->memory.user, pages); 254 map->memory.pages -= pages; 255 } 256 257 static int bpf_map_alloc_id(struct bpf_map *map) 258 { 259 int id; 260 261 idr_preload(GFP_KERNEL); 262 spin_lock_bh(&map_idr_lock); 263 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC); 264 if (id > 0) 265 map->id = id; 266 spin_unlock_bh(&map_idr_lock); 267 idr_preload_end(); 268 269 if (WARN_ON_ONCE(!id)) 270 return -ENOSPC; 271 272 return id > 0 ? 0 : id; 273 } 274 275 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock) 276 { 277 unsigned long flags; 278 279 /* Offloaded maps are removed from the IDR store when their device 280 * disappears - even if someone holds an fd to them they are unusable, 281 * the memory is gone, all ops will fail; they are simply waiting for 282 * refcnt to drop to be freed. 283 */ 284 if (!map->id) 285 return; 286 287 if (do_idr_lock) 288 spin_lock_irqsave(&map_idr_lock, flags); 289 else 290 __acquire(&map_idr_lock); 291 292 idr_remove(&map_idr, map->id); 293 map->id = 0; 294 295 if (do_idr_lock) 296 spin_unlock_irqrestore(&map_idr_lock, flags); 297 else 298 __release(&map_idr_lock); 299 } 300 301 /* called from workqueue */ 302 static void bpf_map_free_deferred(struct work_struct *work) 303 { 304 struct bpf_map *map = container_of(work, struct bpf_map, work); 305 struct bpf_map_memory mem; 306 307 bpf_map_charge_move(&mem, &map->memory); 308 security_bpf_map_free(map); 309 /* implementation dependent freeing */ 310 map->ops->map_free(map); 311 bpf_map_charge_finish(&mem); 312 } 313 314 static void bpf_map_put_uref(struct bpf_map *map) 315 { 316 if (atomic_dec_and_test(&map->usercnt)) { 317 if (map->ops->map_release_uref) 318 map->ops->map_release_uref(map); 319 } 320 } 321 322 /* decrement map refcnt and schedule it for freeing via workqueue 323 * (unrelying map implementation ops->map_free() might sleep) 324 */ 325 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) 326 { 327 if (atomic_dec_and_test(&map->refcnt)) { 328 /* bpf_map_free_id() must be called first */ 329 bpf_map_free_id(map, do_idr_lock); 330 btf_put(map->btf); 331 INIT_WORK(&map->work, bpf_map_free_deferred); 332 schedule_work(&map->work); 333 } 334 } 335 336 void bpf_map_put(struct bpf_map *map) 337 { 338 __bpf_map_put(map, true); 339 } 340 EXPORT_SYMBOL_GPL(bpf_map_put); 341 342 void bpf_map_put_with_uref(struct bpf_map *map) 343 { 344 bpf_map_put_uref(map); 345 bpf_map_put(map); 346 } 347 348 static int bpf_map_release(struct inode *inode, struct file *filp) 349 { 350 struct bpf_map *map = filp->private_data; 351 352 if (map->ops->map_release) 353 map->ops->map_release(map, filp); 354 355 bpf_map_put_with_uref(map); 356 return 0; 357 } 358 359 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f) 360 { 361 fmode_t mode = f.file->f_mode; 362 363 /* Our file permissions may have been overridden by global 364 * map permissions facing syscall side. 365 */ 366 if (READ_ONCE(map->frozen)) 367 mode &= ~FMODE_CAN_WRITE; 368 return mode; 369 } 370 371 #ifdef CONFIG_PROC_FS 372 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) 373 { 374 const struct bpf_map *map = filp->private_data; 375 const struct bpf_array *array; 376 u32 owner_prog_type = 0; 377 u32 owner_jited = 0; 378 379 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { 380 array = container_of(map, struct bpf_array, map); 381 owner_prog_type = array->owner_prog_type; 382 owner_jited = array->owner_jited; 383 } 384 385 seq_printf(m, 386 "map_type:\t%u\n" 387 "key_size:\t%u\n" 388 "value_size:\t%u\n" 389 "max_entries:\t%u\n" 390 "map_flags:\t%#x\n" 391 "memlock:\t%llu\n" 392 "map_id:\t%u\n" 393 "frozen:\t%u\n", 394 map->map_type, 395 map->key_size, 396 map->value_size, 397 map->max_entries, 398 map->map_flags, 399 map->memory.pages * 1ULL << PAGE_SHIFT, 400 map->id, 401 READ_ONCE(map->frozen)); 402 403 if (owner_prog_type) { 404 seq_printf(m, "owner_prog_type:\t%u\n", 405 owner_prog_type); 406 seq_printf(m, "owner_jited:\t%u\n", 407 owner_jited); 408 } 409 } 410 #endif 411 412 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz, 413 loff_t *ppos) 414 { 415 /* We need this handler such that alloc_file() enables 416 * f_mode with FMODE_CAN_READ. 417 */ 418 return -EINVAL; 419 } 420 421 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf, 422 size_t siz, loff_t *ppos) 423 { 424 /* We need this handler such that alloc_file() enables 425 * f_mode with FMODE_CAN_WRITE. 426 */ 427 return -EINVAL; 428 } 429 430 const struct file_operations bpf_map_fops = { 431 #ifdef CONFIG_PROC_FS 432 .show_fdinfo = bpf_map_show_fdinfo, 433 #endif 434 .release = bpf_map_release, 435 .read = bpf_dummy_read, 436 .write = bpf_dummy_write, 437 }; 438 439 int bpf_map_new_fd(struct bpf_map *map, int flags) 440 { 441 int ret; 442 443 ret = security_bpf_map(map, OPEN_FMODE(flags)); 444 if (ret < 0) 445 return ret; 446 447 return anon_inode_getfd("bpf-map", &bpf_map_fops, map, 448 flags | O_CLOEXEC); 449 } 450 451 int bpf_get_file_flag(int flags) 452 { 453 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY)) 454 return -EINVAL; 455 if (flags & BPF_F_RDONLY) 456 return O_RDONLY; 457 if (flags & BPF_F_WRONLY) 458 return O_WRONLY; 459 return O_RDWR; 460 } 461 462 /* helper macro to check that unused fields 'union bpf_attr' are zero */ 463 #define CHECK_ATTR(CMD) \ 464 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ 465 sizeof(attr->CMD##_LAST_FIELD), 0, \ 466 sizeof(*attr) - \ 467 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ 468 sizeof(attr->CMD##_LAST_FIELD)) != NULL 469 470 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes. 471 * Return 0 on success and < 0 on error. 472 */ 473 static int bpf_obj_name_cpy(char *dst, const char *src) 474 { 475 const char *end = src + BPF_OBJ_NAME_LEN; 476 477 memset(dst, 0, BPF_OBJ_NAME_LEN); 478 /* Copy all isalnum(), '_' and '.' chars. */ 479 while (src < end && *src) { 480 if (!isalnum(*src) && 481 *src != '_' && *src != '.') 482 return -EINVAL; 483 *dst++ = *src++; 484 } 485 486 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */ 487 if (src == end) 488 return -EINVAL; 489 490 return 0; 491 } 492 493 int map_check_no_btf(const struct bpf_map *map, 494 const struct btf *btf, 495 const struct btf_type *key_type, 496 const struct btf_type *value_type) 497 { 498 return -ENOTSUPP; 499 } 500 501 static int map_check_btf(struct bpf_map *map, const struct btf *btf, 502 u32 btf_key_id, u32 btf_value_id) 503 { 504 const struct btf_type *key_type, *value_type; 505 u32 key_size, value_size; 506 int ret = 0; 507 508 /* Some maps allow key to be unspecified. */ 509 if (btf_key_id) { 510 key_type = btf_type_id_size(btf, &btf_key_id, &key_size); 511 if (!key_type || key_size != map->key_size) 512 return -EINVAL; 513 } else { 514 key_type = btf_type_by_id(btf, 0); 515 if (!map->ops->map_check_btf) 516 return -EINVAL; 517 } 518 519 value_type = btf_type_id_size(btf, &btf_value_id, &value_size); 520 if (!value_type || value_size != map->value_size) 521 return -EINVAL; 522 523 map->spin_lock_off = btf_find_spin_lock(btf, value_type); 524 525 if (map_value_has_spin_lock(map)) { 526 if (map->map_flags & BPF_F_RDONLY_PROG) 527 return -EACCES; 528 if (map->map_type != BPF_MAP_TYPE_HASH && 529 map->map_type != BPF_MAP_TYPE_ARRAY && 530 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && 531 map->map_type != BPF_MAP_TYPE_SK_STORAGE) 532 return -ENOTSUPP; 533 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) > 534 map->value_size) { 535 WARN_ONCE(1, 536 "verifier bug spin_lock_off %d value_size %d\n", 537 map->spin_lock_off, map->value_size); 538 return -EFAULT; 539 } 540 } 541 542 if (map->ops->map_check_btf) 543 ret = map->ops->map_check_btf(map, btf, key_type, value_type); 544 545 return ret; 546 } 547 548 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id 549 /* called via syscall */ 550 static int map_create(union bpf_attr *attr) 551 { 552 int numa_node = bpf_map_attr_numa_node(attr); 553 struct bpf_map_memory mem; 554 struct bpf_map *map; 555 int f_flags; 556 int err; 557 558 err = CHECK_ATTR(BPF_MAP_CREATE); 559 if (err) 560 return -EINVAL; 561 562 f_flags = bpf_get_file_flag(attr->map_flags); 563 if (f_flags < 0) 564 return f_flags; 565 566 if (numa_node != NUMA_NO_NODE && 567 ((unsigned int)numa_node >= nr_node_ids || 568 !node_online(numa_node))) 569 return -EINVAL; 570 571 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ 572 map = find_and_alloc_map(attr); 573 if (IS_ERR(map)) 574 return PTR_ERR(map); 575 576 err = bpf_obj_name_cpy(map->name, attr->map_name); 577 if (err) 578 goto free_map; 579 580 atomic_set(&map->refcnt, 1); 581 atomic_set(&map->usercnt, 1); 582 583 if (attr->btf_key_type_id || attr->btf_value_type_id) { 584 struct btf *btf; 585 586 if (!attr->btf_value_type_id) { 587 err = -EINVAL; 588 goto free_map; 589 } 590 591 btf = btf_get_by_fd(attr->btf_fd); 592 if (IS_ERR(btf)) { 593 err = PTR_ERR(btf); 594 goto free_map; 595 } 596 597 err = map_check_btf(map, btf, attr->btf_key_type_id, 598 attr->btf_value_type_id); 599 if (err) { 600 btf_put(btf); 601 goto free_map; 602 } 603 604 map->btf = btf; 605 map->btf_key_type_id = attr->btf_key_type_id; 606 map->btf_value_type_id = attr->btf_value_type_id; 607 } else { 608 map->spin_lock_off = -EINVAL; 609 } 610 611 err = security_bpf_map_alloc(map); 612 if (err) 613 goto free_map; 614 615 err = bpf_map_alloc_id(map); 616 if (err) 617 goto free_map_sec; 618 619 err = bpf_map_new_fd(map, f_flags); 620 if (err < 0) { 621 /* failed to allocate fd. 622 * bpf_map_put_with_uref() is needed because the above 623 * bpf_map_alloc_id() has published the map 624 * to the userspace and the userspace may 625 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. 626 */ 627 bpf_map_put_with_uref(map); 628 return err; 629 } 630 631 return err; 632 633 free_map_sec: 634 security_bpf_map_free(map); 635 free_map: 636 btf_put(map->btf); 637 bpf_map_charge_move(&mem, &map->memory); 638 map->ops->map_free(map); 639 bpf_map_charge_finish(&mem); 640 return err; 641 } 642 643 /* if error is returned, fd is released. 644 * On success caller should complete fd access with matching fdput() 645 */ 646 struct bpf_map *__bpf_map_get(struct fd f) 647 { 648 if (!f.file) 649 return ERR_PTR(-EBADF); 650 if (f.file->f_op != &bpf_map_fops) { 651 fdput(f); 652 return ERR_PTR(-EINVAL); 653 } 654 655 return f.file->private_data; 656 } 657 658 /* prog's and map's refcnt limit */ 659 #define BPF_MAX_REFCNT 32768 660 661 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) 662 { 663 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { 664 atomic_dec(&map->refcnt); 665 return ERR_PTR(-EBUSY); 666 } 667 if (uref) 668 atomic_inc(&map->usercnt); 669 return map; 670 } 671 EXPORT_SYMBOL_GPL(bpf_map_inc); 672 673 struct bpf_map *bpf_map_get_with_uref(u32 ufd) 674 { 675 struct fd f = fdget(ufd); 676 struct bpf_map *map; 677 678 map = __bpf_map_get(f); 679 if (IS_ERR(map)) 680 return map; 681 682 map = bpf_map_inc(map, true); 683 fdput(f); 684 685 return map; 686 } 687 688 /* map_idr_lock should have been held */ 689 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, 690 bool uref) 691 { 692 int refold; 693 694 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0); 695 696 if (refold >= BPF_MAX_REFCNT) { 697 __bpf_map_put(map, false); 698 return ERR_PTR(-EBUSY); 699 } 700 701 if (!refold) 702 return ERR_PTR(-ENOENT); 703 704 if (uref) 705 atomic_inc(&map->usercnt); 706 707 return map; 708 } 709 710 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref) 711 { 712 spin_lock_bh(&map_idr_lock); 713 map = __bpf_map_inc_not_zero(map, uref); 714 spin_unlock_bh(&map_idr_lock); 715 716 return map; 717 } 718 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero); 719 720 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) 721 { 722 return -ENOTSUPP; 723 } 724 725 static void *__bpf_copy_key(void __user *ukey, u64 key_size) 726 { 727 if (key_size) 728 return memdup_user(ukey, key_size); 729 730 if (ukey) 731 return ERR_PTR(-EINVAL); 732 733 return NULL; 734 } 735 736 /* last field in 'union bpf_attr' used by this command */ 737 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags 738 739 static int map_lookup_elem(union bpf_attr *attr) 740 { 741 void __user *ukey = u64_to_user_ptr(attr->key); 742 void __user *uvalue = u64_to_user_ptr(attr->value); 743 int ufd = attr->map_fd; 744 struct bpf_map *map; 745 void *key, *value, *ptr; 746 u32 value_size; 747 struct fd f; 748 int err; 749 750 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) 751 return -EINVAL; 752 753 if (attr->flags & ~BPF_F_LOCK) 754 return -EINVAL; 755 756 f = fdget(ufd); 757 map = __bpf_map_get(f); 758 if (IS_ERR(map)) 759 return PTR_ERR(map); 760 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 761 err = -EPERM; 762 goto err_put; 763 } 764 765 if ((attr->flags & BPF_F_LOCK) && 766 !map_value_has_spin_lock(map)) { 767 err = -EINVAL; 768 goto err_put; 769 } 770 771 key = __bpf_copy_key(ukey, map->key_size); 772 if (IS_ERR(key)) { 773 err = PTR_ERR(key); 774 goto err_put; 775 } 776 777 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 778 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 779 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 780 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 781 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 782 else if (IS_FD_MAP(map)) 783 value_size = sizeof(u32); 784 else 785 value_size = map->value_size; 786 787 err = -ENOMEM; 788 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 789 if (!value) 790 goto free_key; 791 792 if (bpf_map_is_dev_bound(map)) { 793 err = bpf_map_offload_lookup_elem(map, key, value); 794 goto done; 795 } 796 797 preempt_disable(); 798 this_cpu_inc(bpf_prog_active); 799 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 800 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 801 err = bpf_percpu_hash_copy(map, key, value); 802 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 803 err = bpf_percpu_array_copy(map, key, value); 804 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 805 err = bpf_percpu_cgroup_storage_copy(map, key, value); 806 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { 807 err = bpf_stackmap_copy(map, key, value); 808 } else if (IS_FD_ARRAY(map)) { 809 err = bpf_fd_array_map_lookup_elem(map, key, value); 810 } else if (IS_FD_HASH(map)) { 811 err = bpf_fd_htab_map_lookup_elem(map, key, value); 812 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 813 err = bpf_fd_reuseport_array_lookup_elem(map, key, value); 814 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 815 map->map_type == BPF_MAP_TYPE_STACK) { 816 err = map->ops->map_peek_elem(map, value); 817 } else { 818 rcu_read_lock(); 819 if (map->ops->map_lookup_elem_sys_only) 820 ptr = map->ops->map_lookup_elem_sys_only(map, key); 821 else 822 ptr = map->ops->map_lookup_elem(map, key); 823 if (IS_ERR(ptr)) { 824 err = PTR_ERR(ptr); 825 } else if (!ptr) { 826 err = -ENOENT; 827 } else { 828 err = 0; 829 if (attr->flags & BPF_F_LOCK) 830 /* lock 'ptr' and copy everything but lock */ 831 copy_map_value_locked(map, value, ptr, true); 832 else 833 copy_map_value(map, value, ptr); 834 /* mask lock, since value wasn't zero inited */ 835 check_and_init_map_lock(map, value); 836 } 837 rcu_read_unlock(); 838 } 839 this_cpu_dec(bpf_prog_active); 840 preempt_enable(); 841 842 done: 843 if (err) 844 goto free_value; 845 846 err = -EFAULT; 847 if (copy_to_user(uvalue, value, value_size) != 0) 848 goto free_value; 849 850 err = 0; 851 852 free_value: 853 kfree(value); 854 free_key: 855 kfree(key); 856 err_put: 857 fdput(f); 858 return err; 859 } 860 861 static void maybe_wait_bpf_programs(struct bpf_map *map) 862 { 863 /* Wait for any running BPF programs to complete so that 864 * userspace, when we return to it, knows that all programs 865 * that could be running use the new map value. 866 */ 867 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS || 868 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 869 synchronize_rcu(); 870 } 871 872 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags 873 874 static int map_update_elem(union bpf_attr *attr) 875 { 876 void __user *ukey = u64_to_user_ptr(attr->key); 877 void __user *uvalue = u64_to_user_ptr(attr->value); 878 int ufd = attr->map_fd; 879 struct bpf_map *map; 880 void *key, *value; 881 u32 value_size; 882 struct fd f; 883 int err; 884 885 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) 886 return -EINVAL; 887 888 f = fdget(ufd); 889 map = __bpf_map_get(f); 890 if (IS_ERR(map)) 891 return PTR_ERR(map); 892 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 893 err = -EPERM; 894 goto err_put; 895 } 896 897 if ((attr->flags & BPF_F_LOCK) && 898 !map_value_has_spin_lock(map)) { 899 err = -EINVAL; 900 goto err_put; 901 } 902 903 key = __bpf_copy_key(ukey, map->key_size); 904 if (IS_ERR(key)) { 905 err = PTR_ERR(key); 906 goto err_put; 907 } 908 909 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 910 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 911 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 912 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 913 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 914 else 915 value_size = map->value_size; 916 917 err = -ENOMEM; 918 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 919 if (!value) 920 goto free_key; 921 922 err = -EFAULT; 923 if (copy_from_user(value, uvalue, value_size) != 0) 924 goto free_value; 925 926 /* Need to create a kthread, thus must support schedule */ 927 if (bpf_map_is_dev_bound(map)) { 928 err = bpf_map_offload_update_elem(map, key, value, attr->flags); 929 goto out; 930 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP || 931 map->map_type == BPF_MAP_TYPE_SOCKHASH || 932 map->map_type == BPF_MAP_TYPE_SOCKMAP) { 933 err = map->ops->map_update_elem(map, key, value, attr->flags); 934 goto out; 935 } 936 937 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from 938 * inside bpf map update or delete otherwise deadlocks are possible 939 */ 940 preempt_disable(); 941 __this_cpu_inc(bpf_prog_active); 942 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 943 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 944 err = bpf_percpu_hash_update(map, key, value, attr->flags); 945 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 946 err = bpf_percpu_array_update(map, key, value, attr->flags); 947 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 948 err = bpf_percpu_cgroup_storage_update(map, key, value, 949 attr->flags); 950 } else if (IS_FD_ARRAY(map)) { 951 rcu_read_lock(); 952 err = bpf_fd_array_map_update_elem(map, f.file, key, value, 953 attr->flags); 954 rcu_read_unlock(); 955 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { 956 rcu_read_lock(); 957 err = bpf_fd_htab_map_update_elem(map, f.file, key, value, 958 attr->flags); 959 rcu_read_unlock(); 960 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 961 /* rcu_read_lock() is not needed */ 962 err = bpf_fd_reuseport_array_update_elem(map, key, value, 963 attr->flags); 964 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 965 map->map_type == BPF_MAP_TYPE_STACK) { 966 err = map->ops->map_push_elem(map, value, attr->flags); 967 } else { 968 rcu_read_lock(); 969 err = map->ops->map_update_elem(map, key, value, attr->flags); 970 rcu_read_unlock(); 971 } 972 __this_cpu_dec(bpf_prog_active); 973 preempt_enable(); 974 maybe_wait_bpf_programs(map); 975 out: 976 free_value: 977 kfree(value); 978 free_key: 979 kfree(key); 980 err_put: 981 fdput(f); 982 return err; 983 } 984 985 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key 986 987 static int map_delete_elem(union bpf_attr *attr) 988 { 989 void __user *ukey = u64_to_user_ptr(attr->key); 990 int ufd = attr->map_fd; 991 struct bpf_map *map; 992 struct fd f; 993 void *key; 994 int err; 995 996 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) 997 return -EINVAL; 998 999 f = fdget(ufd); 1000 map = __bpf_map_get(f); 1001 if (IS_ERR(map)) 1002 return PTR_ERR(map); 1003 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1004 err = -EPERM; 1005 goto err_put; 1006 } 1007 1008 key = __bpf_copy_key(ukey, map->key_size); 1009 if (IS_ERR(key)) { 1010 err = PTR_ERR(key); 1011 goto err_put; 1012 } 1013 1014 if (bpf_map_is_dev_bound(map)) { 1015 err = bpf_map_offload_delete_elem(map, key); 1016 goto out; 1017 } 1018 1019 preempt_disable(); 1020 __this_cpu_inc(bpf_prog_active); 1021 rcu_read_lock(); 1022 err = map->ops->map_delete_elem(map, key); 1023 rcu_read_unlock(); 1024 __this_cpu_dec(bpf_prog_active); 1025 preempt_enable(); 1026 maybe_wait_bpf_programs(map); 1027 out: 1028 kfree(key); 1029 err_put: 1030 fdput(f); 1031 return err; 1032 } 1033 1034 /* last field in 'union bpf_attr' used by this command */ 1035 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key 1036 1037 static int map_get_next_key(union bpf_attr *attr) 1038 { 1039 void __user *ukey = u64_to_user_ptr(attr->key); 1040 void __user *unext_key = u64_to_user_ptr(attr->next_key); 1041 int ufd = attr->map_fd; 1042 struct bpf_map *map; 1043 void *key, *next_key; 1044 struct fd f; 1045 int err; 1046 1047 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) 1048 return -EINVAL; 1049 1050 f = fdget(ufd); 1051 map = __bpf_map_get(f); 1052 if (IS_ERR(map)) 1053 return PTR_ERR(map); 1054 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 1055 err = -EPERM; 1056 goto err_put; 1057 } 1058 1059 if (ukey) { 1060 key = __bpf_copy_key(ukey, map->key_size); 1061 if (IS_ERR(key)) { 1062 err = PTR_ERR(key); 1063 goto err_put; 1064 } 1065 } else { 1066 key = NULL; 1067 } 1068 1069 err = -ENOMEM; 1070 next_key = kmalloc(map->key_size, GFP_USER); 1071 if (!next_key) 1072 goto free_key; 1073 1074 if (bpf_map_is_dev_bound(map)) { 1075 err = bpf_map_offload_get_next_key(map, key, next_key); 1076 goto out; 1077 } 1078 1079 rcu_read_lock(); 1080 err = map->ops->map_get_next_key(map, key, next_key); 1081 rcu_read_unlock(); 1082 out: 1083 if (err) 1084 goto free_next_key; 1085 1086 err = -EFAULT; 1087 if (copy_to_user(unext_key, next_key, map->key_size) != 0) 1088 goto free_next_key; 1089 1090 err = 0; 1091 1092 free_next_key: 1093 kfree(next_key); 1094 free_key: 1095 kfree(key); 1096 err_put: 1097 fdput(f); 1098 return err; 1099 } 1100 1101 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value 1102 1103 static int map_lookup_and_delete_elem(union bpf_attr *attr) 1104 { 1105 void __user *ukey = u64_to_user_ptr(attr->key); 1106 void __user *uvalue = u64_to_user_ptr(attr->value); 1107 int ufd = attr->map_fd; 1108 struct bpf_map *map; 1109 void *key, *value; 1110 u32 value_size; 1111 struct fd f; 1112 int err; 1113 1114 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM)) 1115 return -EINVAL; 1116 1117 f = fdget(ufd); 1118 map = __bpf_map_get(f); 1119 if (IS_ERR(map)) 1120 return PTR_ERR(map); 1121 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1122 err = -EPERM; 1123 goto err_put; 1124 } 1125 1126 key = __bpf_copy_key(ukey, map->key_size); 1127 if (IS_ERR(key)) { 1128 err = PTR_ERR(key); 1129 goto err_put; 1130 } 1131 1132 value_size = map->value_size; 1133 1134 err = -ENOMEM; 1135 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1136 if (!value) 1137 goto free_key; 1138 1139 if (map->map_type == BPF_MAP_TYPE_QUEUE || 1140 map->map_type == BPF_MAP_TYPE_STACK) { 1141 err = map->ops->map_pop_elem(map, value); 1142 } else { 1143 err = -ENOTSUPP; 1144 } 1145 1146 if (err) 1147 goto free_value; 1148 1149 if (copy_to_user(uvalue, value, value_size) != 0) 1150 goto free_value; 1151 1152 err = 0; 1153 1154 free_value: 1155 kfree(value); 1156 free_key: 1157 kfree(key); 1158 err_put: 1159 fdput(f); 1160 return err; 1161 } 1162 1163 #define BPF_MAP_FREEZE_LAST_FIELD map_fd 1164 1165 static int map_freeze(const union bpf_attr *attr) 1166 { 1167 int err = 0, ufd = attr->map_fd; 1168 struct bpf_map *map; 1169 struct fd f; 1170 1171 if (CHECK_ATTR(BPF_MAP_FREEZE)) 1172 return -EINVAL; 1173 1174 f = fdget(ufd); 1175 map = __bpf_map_get(f); 1176 if (IS_ERR(map)) 1177 return PTR_ERR(map); 1178 if (READ_ONCE(map->frozen)) { 1179 err = -EBUSY; 1180 goto err_put; 1181 } 1182 if (!capable(CAP_SYS_ADMIN)) { 1183 err = -EPERM; 1184 goto err_put; 1185 } 1186 1187 WRITE_ONCE(map->frozen, true); 1188 err_put: 1189 fdput(f); 1190 return err; 1191 } 1192 1193 static const struct bpf_prog_ops * const bpf_prog_types[] = { 1194 #define BPF_PROG_TYPE(_id, _name) \ 1195 [_id] = & _name ## _prog_ops, 1196 #define BPF_MAP_TYPE(_id, _ops) 1197 #include <linux/bpf_types.h> 1198 #undef BPF_PROG_TYPE 1199 #undef BPF_MAP_TYPE 1200 }; 1201 1202 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) 1203 { 1204 const struct bpf_prog_ops *ops; 1205 1206 if (type >= ARRAY_SIZE(bpf_prog_types)) 1207 return -EINVAL; 1208 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types)); 1209 ops = bpf_prog_types[type]; 1210 if (!ops) 1211 return -EINVAL; 1212 1213 if (!bpf_prog_is_dev_bound(prog->aux)) 1214 prog->aux->ops = ops; 1215 else 1216 prog->aux->ops = &bpf_offload_prog_ops; 1217 prog->type = type; 1218 return 0; 1219 } 1220 1221 /* drop refcnt on maps used by eBPF program and free auxilary data */ 1222 static void free_used_maps(struct bpf_prog_aux *aux) 1223 { 1224 enum bpf_cgroup_storage_type stype; 1225 int i; 1226 1227 for_each_cgroup_storage_type(stype) { 1228 if (!aux->cgroup_storage[stype]) 1229 continue; 1230 bpf_cgroup_storage_release(aux->prog, 1231 aux->cgroup_storage[stype]); 1232 } 1233 1234 for (i = 0; i < aux->used_map_cnt; i++) 1235 bpf_map_put(aux->used_maps[i]); 1236 1237 kfree(aux->used_maps); 1238 } 1239 1240 int __bpf_prog_charge(struct user_struct *user, u32 pages) 1241 { 1242 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1243 unsigned long user_bufs; 1244 1245 if (user) { 1246 user_bufs = atomic_long_add_return(pages, &user->locked_vm); 1247 if (user_bufs > memlock_limit) { 1248 atomic_long_sub(pages, &user->locked_vm); 1249 return -EPERM; 1250 } 1251 } 1252 1253 return 0; 1254 } 1255 1256 void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 1257 { 1258 if (user) 1259 atomic_long_sub(pages, &user->locked_vm); 1260 } 1261 1262 static int bpf_prog_charge_memlock(struct bpf_prog *prog) 1263 { 1264 struct user_struct *user = get_current_user(); 1265 int ret; 1266 1267 ret = __bpf_prog_charge(user, prog->pages); 1268 if (ret) { 1269 free_uid(user); 1270 return ret; 1271 } 1272 1273 prog->aux->user = user; 1274 return 0; 1275 } 1276 1277 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) 1278 { 1279 struct user_struct *user = prog->aux->user; 1280 1281 __bpf_prog_uncharge(user, prog->pages); 1282 free_uid(user); 1283 } 1284 1285 static int bpf_prog_alloc_id(struct bpf_prog *prog) 1286 { 1287 int id; 1288 1289 idr_preload(GFP_KERNEL); 1290 spin_lock_bh(&prog_idr_lock); 1291 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); 1292 if (id > 0) 1293 prog->aux->id = id; 1294 spin_unlock_bh(&prog_idr_lock); 1295 idr_preload_end(); 1296 1297 /* id is in [1, INT_MAX) */ 1298 if (WARN_ON_ONCE(!id)) 1299 return -ENOSPC; 1300 1301 return id > 0 ? 0 : id; 1302 } 1303 1304 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) 1305 { 1306 /* cBPF to eBPF migrations are currently not in the idr store. 1307 * Offloaded programs are removed from the store when their device 1308 * disappears - even if someone grabs an fd to them they are unusable, 1309 * simply waiting for refcnt to drop to be freed. 1310 */ 1311 if (!prog->aux->id) 1312 return; 1313 1314 if (do_idr_lock) 1315 spin_lock_bh(&prog_idr_lock); 1316 else 1317 __acquire(&prog_idr_lock); 1318 1319 idr_remove(&prog_idr, prog->aux->id); 1320 prog->aux->id = 0; 1321 1322 if (do_idr_lock) 1323 spin_unlock_bh(&prog_idr_lock); 1324 else 1325 __release(&prog_idr_lock); 1326 } 1327 1328 static void __bpf_prog_put_rcu(struct rcu_head *rcu) 1329 { 1330 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1331 1332 kvfree(aux->func_info); 1333 free_used_maps(aux); 1334 bpf_prog_uncharge_memlock(aux->prog); 1335 security_bpf_prog_free(aux); 1336 bpf_prog_free(aux->prog); 1337 } 1338 1339 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred) 1340 { 1341 bpf_prog_kallsyms_del_all(prog); 1342 btf_put(prog->aux->btf); 1343 bpf_prog_free_linfo(prog); 1344 1345 if (deferred) 1346 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1347 else 1348 __bpf_prog_put_rcu(&prog->aux->rcu); 1349 } 1350 1351 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) 1352 { 1353 if (atomic_dec_and_test(&prog->aux->refcnt)) { 1354 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0); 1355 /* bpf_prog_free_id() must be called first */ 1356 bpf_prog_free_id(prog, do_idr_lock); 1357 __bpf_prog_put_noref(prog, true); 1358 } 1359 } 1360 1361 void bpf_prog_put(struct bpf_prog *prog) 1362 { 1363 __bpf_prog_put(prog, true); 1364 } 1365 EXPORT_SYMBOL_GPL(bpf_prog_put); 1366 1367 static int bpf_prog_release(struct inode *inode, struct file *filp) 1368 { 1369 struct bpf_prog *prog = filp->private_data; 1370 1371 bpf_prog_put(prog); 1372 return 0; 1373 } 1374 1375 static void bpf_prog_get_stats(const struct bpf_prog *prog, 1376 struct bpf_prog_stats *stats) 1377 { 1378 u64 nsecs = 0, cnt = 0; 1379 int cpu; 1380 1381 for_each_possible_cpu(cpu) { 1382 const struct bpf_prog_stats *st; 1383 unsigned int start; 1384 u64 tnsecs, tcnt; 1385 1386 st = per_cpu_ptr(prog->aux->stats, cpu); 1387 do { 1388 start = u64_stats_fetch_begin_irq(&st->syncp); 1389 tnsecs = st->nsecs; 1390 tcnt = st->cnt; 1391 } while (u64_stats_fetch_retry_irq(&st->syncp, start)); 1392 nsecs += tnsecs; 1393 cnt += tcnt; 1394 } 1395 stats->nsecs = nsecs; 1396 stats->cnt = cnt; 1397 } 1398 1399 #ifdef CONFIG_PROC_FS 1400 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1401 { 1402 const struct bpf_prog *prog = filp->private_data; 1403 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1404 struct bpf_prog_stats stats; 1405 1406 bpf_prog_get_stats(prog, &stats); 1407 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1408 seq_printf(m, 1409 "prog_type:\t%u\n" 1410 "prog_jited:\t%u\n" 1411 "prog_tag:\t%s\n" 1412 "memlock:\t%llu\n" 1413 "prog_id:\t%u\n" 1414 "run_time_ns:\t%llu\n" 1415 "run_cnt:\t%llu\n", 1416 prog->type, 1417 prog->jited, 1418 prog_tag, 1419 prog->pages * 1ULL << PAGE_SHIFT, 1420 prog->aux->id, 1421 stats.nsecs, 1422 stats.cnt); 1423 } 1424 #endif 1425 1426 const struct file_operations bpf_prog_fops = { 1427 #ifdef CONFIG_PROC_FS 1428 .show_fdinfo = bpf_prog_show_fdinfo, 1429 #endif 1430 .release = bpf_prog_release, 1431 .read = bpf_dummy_read, 1432 .write = bpf_dummy_write, 1433 }; 1434 1435 int bpf_prog_new_fd(struct bpf_prog *prog) 1436 { 1437 int ret; 1438 1439 ret = security_bpf_prog(prog); 1440 if (ret < 0) 1441 return ret; 1442 1443 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1444 O_RDWR | O_CLOEXEC); 1445 } 1446 1447 static struct bpf_prog *____bpf_prog_get(struct fd f) 1448 { 1449 if (!f.file) 1450 return ERR_PTR(-EBADF); 1451 if (f.file->f_op != &bpf_prog_fops) { 1452 fdput(f); 1453 return ERR_PTR(-EINVAL); 1454 } 1455 1456 return f.file->private_data; 1457 } 1458 1459 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) 1460 { 1461 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { 1462 atomic_sub(i, &prog->aux->refcnt); 1463 return ERR_PTR(-EBUSY); 1464 } 1465 return prog; 1466 } 1467 EXPORT_SYMBOL_GPL(bpf_prog_add); 1468 1469 void bpf_prog_sub(struct bpf_prog *prog, int i) 1470 { 1471 /* Only to be used for undoing previous bpf_prog_add() in some 1472 * error path. We still know that another entity in our call 1473 * path holds a reference to the program, thus atomic_sub() can 1474 * be safely used in such cases! 1475 */ 1476 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); 1477 } 1478 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1479 1480 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) 1481 { 1482 return bpf_prog_add(prog, 1); 1483 } 1484 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1485 1486 /* prog_idr_lock should have been held */ 1487 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1488 { 1489 int refold; 1490 1491 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1492 1493 if (refold >= BPF_MAX_REFCNT) { 1494 __bpf_prog_put(prog, false); 1495 return ERR_PTR(-EBUSY); 1496 } 1497 1498 if (!refold) 1499 return ERR_PTR(-ENOENT); 1500 1501 return prog; 1502 } 1503 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1504 1505 bool bpf_prog_get_ok(struct bpf_prog *prog, 1506 enum bpf_prog_type *attach_type, bool attach_drv) 1507 { 1508 /* not an attachment, just a refcount inc, always allow */ 1509 if (!attach_type) 1510 return true; 1511 1512 if (prog->type != *attach_type) 1513 return false; 1514 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1515 return false; 1516 1517 return true; 1518 } 1519 1520 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1521 bool attach_drv) 1522 { 1523 struct fd f = fdget(ufd); 1524 struct bpf_prog *prog; 1525 1526 prog = ____bpf_prog_get(f); 1527 if (IS_ERR(prog)) 1528 return prog; 1529 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1530 prog = ERR_PTR(-EINVAL); 1531 goto out; 1532 } 1533 1534 prog = bpf_prog_inc(prog); 1535 out: 1536 fdput(f); 1537 return prog; 1538 } 1539 1540 struct bpf_prog *bpf_prog_get(u32 ufd) 1541 { 1542 return __bpf_prog_get(ufd, NULL, false); 1543 } 1544 1545 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1546 bool attach_drv) 1547 { 1548 return __bpf_prog_get(ufd, &type, attach_drv); 1549 } 1550 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1551 1552 /* Initially all BPF programs could be loaded w/o specifying 1553 * expected_attach_type. Later for some of them specifying expected_attach_type 1554 * at load time became required so that program could be validated properly. 1555 * Programs of types that are allowed to be loaded both w/ and w/o (for 1556 * backward compatibility) expected_attach_type, should have the default attach 1557 * type assigned to expected_attach_type for the latter case, so that it can be 1558 * validated later at attach time. 1559 * 1560 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1561 * prog type requires it but has some attach types that have to be backward 1562 * compatible. 1563 */ 1564 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1565 { 1566 switch (attr->prog_type) { 1567 case BPF_PROG_TYPE_CGROUP_SOCK: 1568 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1569 * exist so checking for non-zero is the way to go here. 1570 */ 1571 if (!attr->expected_attach_type) 1572 attr->expected_attach_type = 1573 BPF_CGROUP_INET_SOCK_CREATE; 1574 break; 1575 } 1576 } 1577 1578 static int 1579 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type, 1580 enum bpf_attach_type expected_attach_type) 1581 { 1582 switch (prog_type) { 1583 case BPF_PROG_TYPE_CGROUP_SOCK: 1584 switch (expected_attach_type) { 1585 case BPF_CGROUP_INET_SOCK_CREATE: 1586 case BPF_CGROUP_INET4_POST_BIND: 1587 case BPF_CGROUP_INET6_POST_BIND: 1588 return 0; 1589 default: 1590 return -EINVAL; 1591 } 1592 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1593 switch (expected_attach_type) { 1594 case BPF_CGROUP_INET4_BIND: 1595 case BPF_CGROUP_INET6_BIND: 1596 case BPF_CGROUP_INET4_CONNECT: 1597 case BPF_CGROUP_INET6_CONNECT: 1598 case BPF_CGROUP_UDP4_SENDMSG: 1599 case BPF_CGROUP_UDP6_SENDMSG: 1600 case BPF_CGROUP_UDP4_RECVMSG: 1601 case BPF_CGROUP_UDP6_RECVMSG: 1602 return 0; 1603 default: 1604 return -EINVAL; 1605 } 1606 case BPF_PROG_TYPE_CGROUP_SKB: 1607 switch (expected_attach_type) { 1608 case BPF_CGROUP_INET_INGRESS: 1609 case BPF_CGROUP_INET_EGRESS: 1610 return 0; 1611 default: 1612 return -EINVAL; 1613 } 1614 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 1615 switch (expected_attach_type) { 1616 case BPF_CGROUP_SETSOCKOPT: 1617 case BPF_CGROUP_GETSOCKOPT: 1618 return 0; 1619 default: 1620 return -EINVAL; 1621 } 1622 default: 1623 return 0; 1624 } 1625 } 1626 1627 /* last field in 'union bpf_attr' used by this command */ 1628 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt 1629 1630 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr) 1631 { 1632 enum bpf_prog_type type = attr->prog_type; 1633 struct bpf_prog *prog; 1634 int err; 1635 char license[128]; 1636 bool is_gpl; 1637 1638 if (CHECK_ATTR(BPF_PROG_LOAD)) 1639 return -EINVAL; 1640 1641 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | 1642 BPF_F_ANY_ALIGNMENT | 1643 BPF_F_TEST_STATE_FREQ | 1644 BPF_F_TEST_RND_HI32)) 1645 return -EINVAL; 1646 1647 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && 1648 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) && 1649 !capable(CAP_SYS_ADMIN)) 1650 return -EPERM; 1651 1652 /* copy eBPF program license from user space */ 1653 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 1654 sizeof(license) - 1) < 0) 1655 return -EFAULT; 1656 license[sizeof(license) - 1] = 0; 1657 1658 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1659 is_gpl = license_is_gpl_compatible(license); 1660 1661 if (attr->insn_cnt == 0 || 1662 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) 1663 return -E2BIG; 1664 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 1665 type != BPF_PROG_TYPE_CGROUP_SKB && 1666 !capable(CAP_SYS_ADMIN)) 1667 return -EPERM; 1668 1669 bpf_prog_load_fixup_attach_type(attr); 1670 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type)) 1671 return -EINVAL; 1672 1673 /* plain bpf_prog allocation */ 1674 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 1675 if (!prog) 1676 return -ENOMEM; 1677 1678 prog->expected_attach_type = attr->expected_attach_type; 1679 1680 prog->aux->offload_requested = !!attr->prog_ifindex; 1681 1682 err = security_bpf_prog_alloc(prog->aux); 1683 if (err) 1684 goto free_prog_nouncharge; 1685 1686 err = bpf_prog_charge_memlock(prog); 1687 if (err) 1688 goto free_prog_sec; 1689 1690 prog->len = attr->insn_cnt; 1691 1692 err = -EFAULT; 1693 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 1694 bpf_prog_insn_size(prog)) != 0) 1695 goto free_prog; 1696 1697 prog->orig_prog = NULL; 1698 prog->jited = 0; 1699 1700 atomic_set(&prog->aux->refcnt, 1); 1701 prog->gpl_compatible = is_gpl ? 1 : 0; 1702 1703 if (bpf_prog_is_dev_bound(prog->aux)) { 1704 err = bpf_prog_offload_init(prog, attr); 1705 if (err) 1706 goto free_prog; 1707 } 1708 1709 /* find program type: socket_filter vs tracing_filter */ 1710 err = find_prog_type(type, prog); 1711 if (err < 0) 1712 goto free_prog; 1713 1714 prog->aux->load_time = ktime_get_boottime_ns(); 1715 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 1716 if (err) 1717 goto free_prog; 1718 1719 /* run eBPF verifier */ 1720 err = bpf_check(&prog, attr, uattr); 1721 if (err < 0) 1722 goto free_used_maps; 1723 1724 prog = bpf_prog_select_runtime(prog, &err); 1725 if (err < 0) 1726 goto free_used_maps; 1727 1728 err = bpf_prog_alloc_id(prog); 1729 if (err) 1730 goto free_used_maps; 1731 1732 /* Upon success of bpf_prog_alloc_id(), the BPF prog is 1733 * effectively publicly exposed. However, retrieving via 1734 * bpf_prog_get_fd_by_id() will take another reference, 1735 * therefore it cannot be gone underneath us. 1736 * 1737 * Only for the time /after/ successful bpf_prog_new_fd() 1738 * and before returning to userspace, we might just hold 1739 * one reference and any parallel close on that fd could 1740 * rip everything out. Hence, below notifications must 1741 * happen before bpf_prog_new_fd(). 1742 * 1743 * Also, any failure handling from this point onwards must 1744 * be using bpf_prog_put() given the program is exposed. 1745 */ 1746 bpf_prog_kallsyms_add(prog); 1747 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0); 1748 1749 err = bpf_prog_new_fd(prog); 1750 if (err < 0) 1751 bpf_prog_put(prog); 1752 return err; 1753 1754 free_used_maps: 1755 /* In case we have subprogs, we need to wait for a grace 1756 * period before we can tear down JIT memory since symbols 1757 * are already exposed under kallsyms. 1758 */ 1759 __bpf_prog_put_noref(prog, prog->aux->func_cnt); 1760 return err; 1761 free_prog: 1762 bpf_prog_uncharge_memlock(prog); 1763 free_prog_sec: 1764 security_bpf_prog_free(prog->aux); 1765 free_prog_nouncharge: 1766 bpf_prog_free(prog); 1767 return err; 1768 } 1769 1770 #define BPF_OBJ_LAST_FIELD file_flags 1771 1772 static int bpf_obj_pin(const union bpf_attr *attr) 1773 { 1774 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 1775 return -EINVAL; 1776 1777 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 1778 } 1779 1780 static int bpf_obj_get(const union bpf_attr *attr) 1781 { 1782 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 1783 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 1784 return -EINVAL; 1785 1786 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 1787 attr->file_flags); 1788 } 1789 1790 struct bpf_raw_tracepoint { 1791 struct bpf_raw_event_map *btp; 1792 struct bpf_prog *prog; 1793 }; 1794 1795 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp) 1796 { 1797 struct bpf_raw_tracepoint *raw_tp = filp->private_data; 1798 1799 if (raw_tp->prog) { 1800 bpf_probe_unregister(raw_tp->btp, raw_tp->prog); 1801 bpf_prog_put(raw_tp->prog); 1802 } 1803 bpf_put_raw_tracepoint(raw_tp->btp); 1804 kfree(raw_tp); 1805 return 0; 1806 } 1807 1808 static const struct file_operations bpf_raw_tp_fops = { 1809 .release = bpf_raw_tracepoint_release, 1810 .read = bpf_dummy_read, 1811 .write = bpf_dummy_write, 1812 }; 1813 1814 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 1815 1816 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 1817 { 1818 struct bpf_raw_tracepoint *raw_tp; 1819 struct bpf_raw_event_map *btp; 1820 struct bpf_prog *prog; 1821 char tp_name[128]; 1822 int tp_fd, err; 1823 1824 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name), 1825 sizeof(tp_name) - 1) < 0) 1826 return -EFAULT; 1827 tp_name[sizeof(tp_name) - 1] = 0; 1828 1829 btp = bpf_get_raw_tracepoint(tp_name); 1830 if (!btp) 1831 return -ENOENT; 1832 1833 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER); 1834 if (!raw_tp) { 1835 err = -ENOMEM; 1836 goto out_put_btp; 1837 } 1838 raw_tp->btp = btp; 1839 1840 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd); 1841 if (IS_ERR(prog)) { 1842 err = PTR_ERR(prog); 1843 goto out_free_tp; 1844 } 1845 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT && 1846 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) { 1847 err = -EINVAL; 1848 goto out_put_prog; 1849 } 1850 1851 err = bpf_probe_register(raw_tp->btp, prog); 1852 if (err) 1853 goto out_put_prog; 1854 1855 raw_tp->prog = prog; 1856 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp, 1857 O_CLOEXEC); 1858 if (tp_fd < 0) { 1859 bpf_probe_unregister(raw_tp->btp, prog); 1860 err = tp_fd; 1861 goto out_put_prog; 1862 } 1863 return tp_fd; 1864 1865 out_put_prog: 1866 bpf_prog_put(prog); 1867 out_free_tp: 1868 kfree(raw_tp); 1869 out_put_btp: 1870 bpf_put_raw_tracepoint(btp); 1871 return err; 1872 } 1873 1874 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 1875 enum bpf_attach_type attach_type) 1876 { 1877 switch (prog->type) { 1878 case BPF_PROG_TYPE_CGROUP_SOCK: 1879 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1880 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 1881 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 1882 case BPF_PROG_TYPE_CGROUP_SKB: 1883 return prog->enforce_expected_attach_type && 1884 prog->expected_attach_type != attach_type ? 1885 -EINVAL : 0; 1886 default: 1887 return 0; 1888 } 1889 } 1890 1891 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags 1892 1893 #define BPF_F_ATTACH_MASK \ 1894 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI) 1895 1896 static int bpf_prog_attach(const union bpf_attr *attr) 1897 { 1898 enum bpf_prog_type ptype; 1899 struct bpf_prog *prog; 1900 int ret; 1901 1902 if (!capable(CAP_NET_ADMIN)) 1903 return -EPERM; 1904 1905 if (CHECK_ATTR(BPF_PROG_ATTACH)) 1906 return -EINVAL; 1907 1908 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 1909 return -EINVAL; 1910 1911 switch (attr->attach_type) { 1912 case BPF_CGROUP_INET_INGRESS: 1913 case BPF_CGROUP_INET_EGRESS: 1914 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1915 break; 1916 case BPF_CGROUP_INET_SOCK_CREATE: 1917 case BPF_CGROUP_INET4_POST_BIND: 1918 case BPF_CGROUP_INET6_POST_BIND: 1919 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1920 break; 1921 case BPF_CGROUP_INET4_BIND: 1922 case BPF_CGROUP_INET6_BIND: 1923 case BPF_CGROUP_INET4_CONNECT: 1924 case BPF_CGROUP_INET6_CONNECT: 1925 case BPF_CGROUP_UDP4_SENDMSG: 1926 case BPF_CGROUP_UDP6_SENDMSG: 1927 case BPF_CGROUP_UDP4_RECVMSG: 1928 case BPF_CGROUP_UDP6_RECVMSG: 1929 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1930 break; 1931 case BPF_CGROUP_SOCK_OPS: 1932 ptype = BPF_PROG_TYPE_SOCK_OPS; 1933 break; 1934 case BPF_CGROUP_DEVICE: 1935 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1936 break; 1937 case BPF_SK_MSG_VERDICT: 1938 ptype = BPF_PROG_TYPE_SK_MSG; 1939 break; 1940 case BPF_SK_SKB_STREAM_PARSER: 1941 case BPF_SK_SKB_STREAM_VERDICT: 1942 ptype = BPF_PROG_TYPE_SK_SKB; 1943 break; 1944 case BPF_LIRC_MODE2: 1945 ptype = BPF_PROG_TYPE_LIRC_MODE2; 1946 break; 1947 case BPF_FLOW_DISSECTOR: 1948 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 1949 break; 1950 case BPF_CGROUP_SYSCTL: 1951 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 1952 break; 1953 case BPF_CGROUP_GETSOCKOPT: 1954 case BPF_CGROUP_SETSOCKOPT: 1955 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 1956 break; 1957 default: 1958 return -EINVAL; 1959 } 1960 1961 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1962 if (IS_ERR(prog)) 1963 return PTR_ERR(prog); 1964 1965 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 1966 bpf_prog_put(prog); 1967 return -EINVAL; 1968 } 1969 1970 switch (ptype) { 1971 case BPF_PROG_TYPE_SK_SKB: 1972 case BPF_PROG_TYPE_SK_MSG: 1973 ret = sock_map_get_from_fd(attr, prog); 1974 break; 1975 case BPF_PROG_TYPE_LIRC_MODE2: 1976 ret = lirc_prog_attach(attr, prog); 1977 break; 1978 case BPF_PROG_TYPE_FLOW_DISSECTOR: 1979 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 1980 break; 1981 default: 1982 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 1983 } 1984 1985 if (ret) 1986 bpf_prog_put(prog); 1987 return ret; 1988 } 1989 1990 #define BPF_PROG_DETACH_LAST_FIELD attach_type 1991 1992 static int bpf_prog_detach(const union bpf_attr *attr) 1993 { 1994 enum bpf_prog_type ptype; 1995 1996 if (!capable(CAP_NET_ADMIN)) 1997 return -EPERM; 1998 1999 if (CHECK_ATTR(BPF_PROG_DETACH)) 2000 return -EINVAL; 2001 2002 switch (attr->attach_type) { 2003 case BPF_CGROUP_INET_INGRESS: 2004 case BPF_CGROUP_INET_EGRESS: 2005 ptype = BPF_PROG_TYPE_CGROUP_SKB; 2006 break; 2007 case BPF_CGROUP_INET_SOCK_CREATE: 2008 case BPF_CGROUP_INET4_POST_BIND: 2009 case BPF_CGROUP_INET6_POST_BIND: 2010 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 2011 break; 2012 case BPF_CGROUP_INET4_BIND: 2013 case BPF_CGROUP_INET6_BIND: 2014 case BPF_CGROUP_INET4_CONNECT: 2015 case BPF_CGROUP_INET6_CONNECT: 2016 case BPF_CGROUP_UDP4_SENDMSG: 2017 case BPF_CGROUP_UDP6_SENDMSG: 2018 case BPF_CGROUP_UDP4_RECVMSG: 2019 case BPF_CGROUP_UDP6_RECVMSG: 2020 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 2021 break; 2022 case BPF_CGROUP_SOCK_OPS: 2023 ptype = BPF_PROG_TYPE_SOCK_OPS; 2024 break; 2025 case BPF_CGROUP_DEVICE: 2026 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 2027 break; 2028 case BPF_SK_MSG_VERDICT: 2029 return sock_map_get_from_fd(attr, NULL); 2030 case BPF_SK_SKB_STREAM_PARSER: 2031 case BPF_SK_SKB_STREAM_VERDICT: 2032 return sock_map_get_from_fd(attr, NULL); 2033 case BPF_LIRC_MODE2: 2034 return lirc_prog_detach(attr); 2035 case BPF_FLOW_DISSECTOR: 2036 return skb_flow_dissector_bpf_prog_detach(attr); 2037 case BPF_CGROUP_SYSCTL: 2038 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 2039 break; 2040 case BPF_CGROUP_GETSOCKOPT: 2041 case BPF_CGROUP_SETSOCKOPT: 2042 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 2043 break; 2044 default: 2045 return -EINVAL; 2046 } 2047 2048 return cgroup_bpf_prog_detach(attr, ptype); 2049 } 2050 2051 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 2052 2053 static int bpf_prog_query(const union bpf_attr *attr, 2054 union bpf_attr __user *uattr) 2055 { 2056 if (!capable(CAP_NET_ADMIN)) 2057 return -EPERM; 2058 if (CHECK_ATTR(BPF_PROG_QUERY)) 2059 return -EINVAL; 2060 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 2061 return -EINVAL; 2062 2063 switch (attr->query.attach_type) { 2064 case BPF_CGROUP_INET_INGRESS: 2065 case BPF_CGROUP_INET_EGRESS: 2066 case BPF_CGROUP_INET_SOCK_CREATE: 2067 case BPF_CGROUP_INET4_BIND: 2068 case BPF_CGROUP_INET6_BIND: 2069 case BPF_CGROUP_INET4_POST_BIND: 2070 case BPF_CGROUP_INET6_POST_BIND: 2071 case BPF_CGROUP_INET4_CONNECT: 2072 case BPF_CGROUP_INET6_CONNECT: 2073 case BPF_CGROUP_UDP4_SENDMSG: 2074 case BPF_CGROUP_UDP6_SENDMSG: 2075 case BPF_CGROUP_UDP4_RECVMSG: 2076 case BPF_CGROUP_UDP6_RECVMSG: 2077 case BPF_CGROUP_SOCK_OPS: 2078 case BPF_CGROUP_DEVICE: 2079 case BPF_CGROUP_SYSCTL: 2080 case BPF_CGROUP_GETSOCKOPT: 2081 case BPF_CGROUP_SETSOCKOPT: 2082 break; 2083 case BPF_LIRC_MODE2: 2084 return lirc_prog_query(attr, uattr); 2085 case BPF_FLOW_DISSECTOR: 2086 return skb_flow_dissector_prog_query(attr, uattr); 2087 default: 2088 return -EINVAL; 2089 } 2090 2091 return cgroup_bpf_prog_query(attr, uattr); 2092 } 2093 2094 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out 2095 2096 static int bpf_prog_test_run(const union bpf_attr *attr, 2097 union bpf_attr __user *uattr) 2098 { 2099 struct bpf_prog *prog; 2100 int ret = -ENOTSUPP; 2101 2102 if (!capable(CAP_SYS_ADMIN)) 2103 return -EPERM; 2104 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 2105 return -EINVAL; 2106 2107 if ((attr->test.ctx_size_in && !attr->test.ctx_in) || 2108 (!attr->test.ctx_size_in && attr->test.ctx_in)) 2109 return -EINVAL; 2110 2111 if ((attr->test.ctx_size_out && !attr->test.ctx_out) || 2112 (!attr->test.ctx_size_out && attr->test.ctx_out)) 2113 return -EINVAL; 2114 2115 prog = bpf_prog_get(attr->test.prog_fd); 2116 if (IS_ERR(prog)) 2117 return PTR_ERR(prog); 2118 2119 if (prog->aux->ops->test_run) 2120 ret = prog->aux->ops->test_run(prog, attr, uattr); 2121 2122 bpf_prog_put(prog); 2123 return ret; 2124 } 2125 2126 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 2127 2128 static int bpf_obj_get_next_id(const union bpf_attr *attr, 2129 union bpf_attr __user *uattr, 2130 struct idr *idr, 2131 spinlock_t *lock) 2132 { 2133 u32 next_id = attr->start_id; 2134 int err = 0; 2135 2136 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 2137 return -EINVAL; 2138 2139 if (!capable(CAP_SYS_ADMIN)) 2140 return -EPERM; 2141 2142 next_id++; 2143 spin_lock_bh(lock); 2144 if (!idr_get_next(idr, &next_id)) 2145 err = -ENOENT; 2146 spin_unlock_bh(lock); 2147 2148 if (!err) 2149 err = put_user(next_id, &uattr->next_id); 2150 2151 return err; 2152 } 2153 2154 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 2155 2156 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 2157 { 2158 struct bpf_prog *prog; 2159 u32 id = attr->prog_id; 2160 int fd; 2161 2162 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 2163 return -EINVAL; 2164 2165 if (!capable(CAP_SYS_ADMIN)) 2166 return -EPERM; 2167 2168 spin_lock_bh(&prog_idr_lock); 2169 prog = idr_find(&prog_idr, id); 2170 if (prog) 2171 prog = bpf_prog_inc_not_zero(prog); 2172 else 2173 prog = ERR_PTR(-ENOENT); 2174 spin_unlock_bh(&prog_idr_lock); 2175 2176 if (IS_ERR(prog)) 2177 return PTR_ERR(prog); 2178 2179 fd = bpf_prog_new_fd(prog); 2180 if (fd < 0) 2181 bpf_prog_put(prog); 2182 2183 return fd; 2184 } 2185 2186 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 2187 2188 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 2189 { 2190 struct bpf_map *map; 2191 u32 id = attr->map_id; 2192 int f_flags; 2193 int fd; 2194 2195 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 2196 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 2197 return -EINVAL; 2198 2199 if (!capable(CAP_SYS_ADMIN)) 2200 return -EPERM; 2201 2202 f_flags = bpf_get_file_flag(attr->open_flags); 2203 if (f_flags < 0) 2204 return f_flags; 2205 2206 spin_lock_bh(&map_idr_lock); 2207 map = idr_find(&map_idr, id); 2208 if (map) 2209 map = __bpf_map_inc_not_zero(map, true); 2210 else 2211 map = ERR_PTR(-ENOENT); 2212 spin_unlock_bh(&map_idr_lock); 2213 2214 if (IS_ERR(map)) 2215 return PTR_ERR(map); 2216 2217 fd = bpf_map_new_fd(map, f_flags); 2218 if (fd < 0) 2219 bpf_map_put_with_uref(map); 2220 2221 return fd; 2222 } 2223 2224 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 2225 unsigned long addr, u32 *off, 2226 u32 *type) 2227 { 2228 const struct bpf_map *map; 2229 int i; 2230 2231 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) { 2232 map = prog->aux->used_maps[i]; 2233 if (map == (void *)addr) { 2234 *type = BPF_PSEUDO_MAP_FD; 2235 return map; 2236 } 2237 if (!map->ops->map_direct_value_meta) 2238 continue; 2239 if (!map->ops->map_direct_value_meta(map, addr, off)) { 2240 *type = BPF_PSEUDO_MAP_VALUE; 2241 return map; 2242 } 2243 } 2244 2245 return NULL; 2246 } 2247 2248 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 2249 { 2250 const struct bpf_map *map; 2251 struct bpf_insn *insns; 2252 u32 off, type; 2253 u64 imm; 2254 int i; 2255 2256 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 2257 GFP_USER); 2258 if (!insns) 2259 return insns; 2260 2261 for (i = 0; i < prog->len; i++) { 2262 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 2263 insns[i].code = BPF_JMP | BPF_CALL; 2264 insns[i].imm = BPF_FUNC_tail_call; 2265 /* fall-through */ 2266 } 2267 if (insns[i].code == (BPF_JMP | BPF_CALL) || 2268 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 2269 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 2270 insns[i].code = BPF_JMP | BPF_CALL; 2271 if (!bpf_dump_raw_ok()) 2272 insns[i].imm = 0; 2273 continue; 2274 } 2275 2276 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 2277 continue; 2278 2279 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 2280 map = bpf_map_from_imm(prog, imm, &off, &type); 2281 if (map) { 2282 insns[i].src_reg = type; 2283 insns[i].imm = map->id; 2284 insns[i + 1].imm = off; 2285 continue; 2286 } 2287 } 2288 2289 return insns; 2290 } 2291 2292 static int set_info_rec_size(struct bpf_prog_info *info) 2293 { 2294 /* 2295 * Ensure info.*_rec_size is the same as kernel expected size 2296 * 2297 * or 2298 * 2299 * Only allow zero *_rec_size if both _rec_size and _cnt are 2300 * zero. In this case, the kernel will set the expected 2301 * _rec_size back to the info. 2302 */ 2303 2304 if ((info->nr_func_info || info->func_info_rec_size) && 2305 info->func_info_rec_size != sizeof(struct bpf_func_info)) 2306 return -EINVAL; 2307 2308 if ((info->nr_line_info || info->line_info_rec_size) && 2309 info->line_info_rec_size != sizeof(struct bpf_line_info)) 2310 return -EINVAL; 2311 2312 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) && 2313 info->jited_line_info_rec_size != sizeof(__u64)) 2314 return -EINVAL; 2315 2316 info->func_info_rec_size = sizeof(struct bpf_func_info); 2317 info->line_info_rec_size = sizeof(struct bpf_line_info); 2318 info->jited_line_info_rec_size = sizeof(__u64); 2319 2320 return 0; 2321 } 2322 2323 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 2324 const union bpf_attr *attr, 2325 union bpf_attr __user *uattr) 2326 { 2327 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2328 struct bpf_prog_info info = {}; 2329 u32 info_len = attr->info.info_len; 2330 struct bpf_prog_stats stats; 2331 char __user *uinsns; 2332 u32 ulen; 2333 int err; 2334 2335 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2336 if (err) 2337 return err; 2338 info_len = min_t(u32, sizeof(info), info_len); 2339 2340 if (copy_from_user(&info, uinfo, info_len)) 2341 return -EFAULT; 2342 2343 info.type = prog->type; 2344 info.id = prog->aux->id; 2345 info.load_time = prog->aux->load_time; 2346 info.created_by_uid = from_kuid_munged(current_user_ns(), 2347 prog->aux->user->uid); 2348 info.gpl_compatible = prog->gpl_compatible; 2349 2350 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 2351 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 2352 2353 ulen = info.nr_map_ids; 2354 info.nr_map_ids = prog->aux->used_map_cnt; 2355 ulen = min_t(u32, info.nr_map_ids, ulen); 2356 if (ulen) { 2357 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 2358 u32 i; 2359 2360 for (i = 0; i < ulen; i++) 2361 if (put_user(prog->aux->used_maps[i]->id, 2362 &user_map_ids[i])) 2363 return -EFAULT; 2364 } 2365 2366 err = set_info_rec_size(&info); 2367 if (err) 2368 return err; 2369 2370 bpf_prog_get_stats(prog, &stats); 2371 info.run_time_ns = stats.nsecs; 2372 info.run_cnt = stats.cnt; 2373 2374 if (!capable(CAP_SYS_ADMIN)) { 2375 info.jited_prog_len = 0; 2376 info.xlated_prog_len = 0; 2377 info.nr_jited_ksyms = 0; 2378 info.nr_jited_func_lens = 0; 2379 info.nr_func_info = 0; 2380 info.nr_line_info = 0; 2381 info.nr_jited_line_info = 0; 2382 goto done; 2383 } 2384 2385 ulen = info.xlated_prog_len; 2386 info.xlated_prog_len = bpf_prog_insn_size(prog); 2387 if (info.xlated_prog_len && ulen) { 2388 struct bpf_insn *insns_sanitized; 2389 bool fault; 2390 2391 if (prog->blinded && !bpf_dump_raw_ok()) { 2392 info.xlated_prog_insns = 0; 2393 goto done; 2394 } 2395 insns_sanitized = bpf_insn_prepare_dump(prog); 2396 if (!insns_sanitized) 2397 return -ENOMEM; 2398 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 2399 ulen = min_t(u32, info.xlated_prog_len, ulen); 2400 fault = copy_to_user(uinsns, insns_sanitized, ulen); 2401 kfree(insns_sanitized); 2402 if (fault) 2403 return -EFAULT; 2404 } 2405 2406 if (bpf_prog_is_dev_bound(prog->aux)) { 2407 err = bpf_prog_offload_info_fill(&info, prog); 2408 if (err) 2409 return err; 2410 goto done; 2411 } 2412 2413 /* NOTE: the following code is supposed to be skipped for offload. 2414 * bpf_prog_offload_info_fill() is the place to fill similar fields 2415 * for offload. 2416 */ 2417 ulen = info.jited_prog_len; 2418 if (prog->aux->func_cnt) { 2419 u32 i; 2420 2421 info.jited_prog_len = 0; 2422 for (i = 0; i < prog->aux->func_cnt; i++) 2423 info.jited_prog_len += prog->aux->func[i]->jited_len; 2424 } else { 2425 info.jited_prog_len = prog->jited_len; 2426 } 2427 2428 if (info.jited_prog_len && ulen) { 2429 if (bpf_dump_raw_ok()) { 2430 uinsns = u64_to_user_ptr(info.jited_prog_insns); 2431 ulen = min_t(u32, info.jited_prog_len, ulen); 2432 2433 /* for multi-function programs, copy the JITed 2434 * instructions for all the functions 2435 */ 2436 if (prog->aux->func_cnt) { 2437 u32 len, free, i; 2438 u8 *img; 2439 2440 free = ulen; 2441 for (i = 0; i < prog->aux->func_cnt; i++) { 2442 len = prog->aux->func[i]->jited_len; 2443 len = min_t(u32, len, free); 2444 img = (u8 *) prog->aux->func[i]->bpf_func; 2445 if (copy_to_user(uinsns, img, len)) 2446 return -EFAULT; 2447 uinsns += len; 2448 free -= len; 2449 if (!free) 2450 break; 2451 } 2452 } else { 2453 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 2454 return -EFAULT; 2455 } 2456 } else { 2457 info.jited_prog_insns = 0; 2458 } 2459 } 2460 2461 ulen = info.nr_jited_ksyms; 2462 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1; 2463 if (ulen) { 2464 if (bpf_dump_raw_ok()) { 2465 unsigned long ksym_addr; 2466 u64 __user *user_ksyms; 2467 u32 i; 2468 2469 /* copy the address of the kernel symbol 2470 * corresponding to each function 2471 */ 2472 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 2473 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 2474 if (prog->aux->func_cnt) { 2475 for (i = 0; i < ulen; i++) { 2476 ksym_addr = (unsigned long) 2477 prog->aux->func[i]->bpf_func; 2478 if (put_user((u64) ksym_addr, 2479 &user_ksyms[i])) 2480 return -EFAULT; 2481 } 2482 } else { 2483 ksym_addr = (unsigned long) prog->bpf_func; 2484 if (put_user((u64) ksym_addr, &user_ksyms[0])) 2485 return -EFAULT; 2486 } 2487 } else { 2488 info.jited_ksyms = 0; 2489 } 2490 } 2491 2492 ulen = info.nr_jited_func_lens; 2493 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1; 2494 if (ulen) { 2495 if (bpf_dump_raw_ok()) { 2496 u32 __user *user_lens; 2497 u32 func_len, i; 2498 2499 /* copy the JITed image lengths for each function */ 2500 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 2501 user_lens = u64_to_user_ptr(info.jited_func_lens); 2502 if (prog->aux->func_cnt) { 2503 for (i = 0; i < ulen; i++) { 2504 func_len = 2505 prog->aux->func[i]->jited_len; 2506 if (put_user(func_len, &user_lens[i])) 2507 return -EFAULT; 2508 } 2509 } else { 2510 func_len = prog->jited_len; 2511 if (put_user(func_len, &user_lens[0])) 2512 return -EFAULT; 2513 } 2514 } else { 2515 info.jited_func_lens = 0; 2516 } 2517 } 2518 2519 if (prog->aux->btf) 2520 info.btf_id = btf_id(prog->aux->btf); 2521 2522 ulen = info.nr_func_info; 2523 info.nr_func_info = prog->aux->func_info_cnt; 2524 if (info.nr_func_info && ulen) { 2525 char __user *user_finfo; 2526 2527 user_finfo = u64_to_user_ptr(info.func_info); 2528 ulen = min_t(u32, info.nr_func_info, ulen); 2529 if (copy_to_user(user_finfo, prog->aux->func_info, 2530 info.func_info_rec_size * ulen)) 2531 return -EFAULT; 2532 } 2533 2534 ulen = info.nr_line_info; 2535 info.nr_line_info = prog->aux->nr_linfo; 2536 if (info.nr_line_info && ulen) { 2537 __u8 __user *user_linfo; 2538 2539 user_linfo = u64_to_user_ptr(info.line_info); 2540 ulen = min_t(u32, info.nr_line_info, ulen); 2541 if (copy_to_user(user_linfo, prog->aux->linfo, 2542 info.line_info_rec_size * ulen)) 2543 return -EFAULT; 2544 } 2545 2546 ulen = info.nr_jited_line_info; 2547 if (prog->aux->jited_linfo) 2548 info.nr_jited_line_info = prog->aux->nr_linfo; 2549 else 2550 info.nr_jited_line_info = 0; 2551 if (info.nr_jited_line_info && ulen) { 2552 if (bpf_dump_raw_ok()) { 2553 __u64 __user *user_linfo; 2554 u32 i; 2555 2556 user_linfo = u64_to_user_ptr(info.jited_line_info); 2557 ulen = min_t(u32, info.nr_jited_line_info, ulen); 2558 for (i = 0; i < ulen; i++) { 2559 if (put_user((__u64)(long)prog->aux->jited_linfo[i], 2560 &user_linfo[i])) 2561 return -EFAULT; 2562 } 2563 } else { 2564 info.jited_line_info = 0; 2565 } 2566 } 2567 2568 ulen = info.nr_prog_tags; 2569 info.nr_prog_tags = prog->aux->func_cnt ? : 1; 2570 if (ulen) { 2571 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE]; 2572 u32 i; 2573 2574 user_prog_tags = u64_to_user_ptr(info.prog_tags); 2575 ulen = min_t(u32, info.nr_prog_tags, ulen); 2576 if (prog->aux->func_cnt) { 2577 for (i = 0; i < ulen; i++) { 2578 if (copy_to_user(user_prog_tags[i], 2579 prog->aux->func[i]->tag, 2580 BPF_TAG_SIZE)) 2581 return -EFAULT; 2582 } 2583 } else { 2584 if (copy_to_user(user_prog_tags[0], 2585 prog->tag, BPF_TAG_SIZE)) 2586 return -EFAULT; 2587 } 2588 } 2589 2590 done: 2591 if (copy_to_user(uinfo, &info, info_len) || 2592 put_user(info_len, &uattr->info.info_len)) 2593 return -EFAULT; 2594 2595 return 0; 2596 } 2597 2598 static int bpf_map_get_info_by_fd(struct bpf_map *map, 2599 const union bpf_attr *attr, 2600 union bpf_attr __user *uattr) 2601 { 2602 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2603 struct bpf_map_info info = {}; 2604 u32 info_len = attr->info.info_len; 2605 int err; 2606 2607 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2608 if (err) 2609 return err; 2610 info_len = min_t(u32, sizeof(info), info_len); 2611 2612 info.type = map->map_type; 2613 info.id = map->id; 2614 info.key_size = map->key_size; 2615 info.value_size = map->value_size; 2616 info.max_entries = map->max_entries; 2617 info.map_flags = map->map_flags; 2618 memcpy(info.name, map->name, sizeof(map->name)); 2619 2620 if (map->btf) { 2621 info.btf_id = btf_id(map->btf); 2622 info.btf_key_type_id = map->btf_key_type_id; 2623 info.btf_value_type_id = map->btf_value_type_id; 2624 } 2625 2626 if (bpf_map_is_dev_bound(map)) { 2627 err = bpf_map_offload_info_fill(&info, map); 2628 if (err) 2629 return err; 2630 } 2631 2632 if (copy_to_user(uinfo, &info, info_len) || 2633 put_user(info_len, &uattr->info.info_len)) 2634 return -EFAULT; 2635 2636 return 0; 2637 } 2638 2639 static int bpf_btf_get_info_by_fd(struct btf *btf, 2640 const union bpf_attr *attr, 2641 union bpf_attr __user *uattr) 2642 { 2643 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2644 u32 info_len = attr->info.info_len; 2645 int err; 2646 2647 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 2648 if (err) 2649 return err; 2650 2651 return btf_get_info_by_fd(btf, attr, uattr); 2652 } 2653 2654 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 2655 2656 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 2657 union bpf_attr __user *uattr) 2658 { 2659 int ufd = attr->info.bpf_fd; 2660 struct fd f; 2661 int err; 2662 2663 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 2664 return -EINVAL; 2665 2666 f = fdget(ufd); 2667 if (!f.file) 2668 return -EBADFD; 2669 2670 if (f.file->f_op == &bpf_prog_fops) 2671 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 2672 uattr); 2673 else if (f.file->f_op == &bpf_map_fops) 2674 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 2675 uattr); 2676 else if (f.file->f_op == &btf_fops) 2677 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 2678 else 2679 err = -EINVAL; 2680 2681 fdput(f); 2682 return err; 2683 } 2684 2685 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 2686 2687 static int bpf_btf_load(const union bpf_attr *attr) 2688 { 2689 if (CHECK_ATTR(BPF_BTF_LOAD)) 2690 return -EINVAL; 2691 2692 if (!capable(CAP_SYS_ADMIN)) 2693 return -EPERM; 2694 2695 return btf_new_fd(attr); 2696 } 2697 2698 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 2699 2700 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 2701 { 2702 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 2703 return -EINVAL; 2704 2705 if (!capable(CAP_SYS_ADMIN)) 2706 return -EPERM; 2707 2708 return btf_get_fd_by_id(attr->btf_id); 2709 } 2710 2711 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 2712 union bpf_attr __user *uattr, 2713 u32 prog_id, u32 fd_type, 2714 const char *buf, u64 probe_offset, 2715 u64 probe_addr) 2716 { 2717 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 2718 u32 len = buf ? strlen(buf) : 0, input_len; 2719 int err = 0; 2720 2721 if (put_user(len, &uattr->task_fd_query.buf_len)) 2722 return -EFAULT; 2723 input_len = attr->task_fd_query.buf_len; 2724 if (input_len && ubuf) { 2725 if (!len) { 2726 /* nothing to copy, just make ubuf NULL terminated */ 2727 char zero = '\0'; 2728 2729 if (put_user(zero, ubuf)) 2730 return -EFAULT; 2731 } else if (input_len >= len + 1) { 2732 /* ubuf can hold the string with NULL terminator */ 2733 if (copy_to_user(ubuf, buf, len + 1)) 2734 return -EFAULT; 2735 } else { 2736 /* ubuf cannot hold the string with NULL terminator, 2737 * do a partial copy with NULL terminator. 2738 */ 2739 char zero = '\0'; 2740 2741 err = -ENOSPC; 2742 if (copy_to_user(ubuf, buf, input_len - 1)) 2743 return -EFAULT; 2744 if (put_user(zero, ubuf + input_len - 1)) 2745 return -EFAULT; 2746 } 2747 } 2748 2749 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 2750 put_user(fd_type, &uattr->task_fd_query.fd_type) || 2751 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 2752 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 2753 return -EFAULT; 2754 2755 return err; 2756 } 2757 2758 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 2759 2760 static int bpf_task_fd_query(const union bpf_attr *attr, 2761 union bpf_attr __user *uattr) 2762 { 2763 pid_t pid = attr->task_fd_query.pid; 2764 u32 fd = attr->task_fd_query.fd; 2765 const struct perf_event *event; 2766 struct files_struct *files; 2767 struct task_struct *task; 2768 struct file *file; 2769 int err; 2770 2771 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 2772 return -EINVAL; 2773 2774 if (!capable(CAP_SYS_ADMIN)) 2775 return -EPERM; 2776 2777 if (attr->task_fd_query.flags != 0) 2778 return -EINVAL; 2779 2780 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 2781 if (!task) 2782 return -ENOENT; 2783 2784 files = get_files_struct(task); 2785 put_task_struct(task); 2786 if (!files) 2787 return -ENOENT; 2788 2789 err = 0; 2790 spin_lock(&files->file_lock); 2791 file = fcheck_files(files, fd); 2792 if (!file) 2793 err = -EBADF; 2794 else 2795 get_file(file); 2796 spin_unlock(&files->file_lock); 2797 put_files_struct(files); 2798 2799 if (err) 2800 goto out; 2801 2802 if (file->f_op == &bpf_raw_tp_fops) { 2803 struct bpf_raw_tracepoint *raw_tp = file->private_data; 2804 struct bpf_raw_event_map *btp = raw_tp->btp; 2805 2806 err = bpf_task_fd_query_copy(attr, uattr, 2807 raw_tp->prog->aux->id, 2808 BPF_FD_TYPE_RAW_TRACEPOINT, 2809 btp->tp->name, 0, 0); 2810 goto put_file; 2811 } 2812 2813 event = perf_get_event(file); 2814 if (!IS_ERR(event)) { 2815 u64 probe_offset, probe_addr; 2816 u32 prog_id, fd_type; 2817 const char *buf; 2818 2819 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 2820 &buf, &probe_offset, 2821 &probe_addr); 2822 if (!err) 2823 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 2824 fd_type, buf, 2825 probe_offset, 2826 probe_addr); 2827 goto put_file; 2828 } 2829 2830 err = -ENOTSUPP; 2831 put_file: 2832 fput(file); 2833 out: 2834 return err; 2835 } 2836 2837 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 2838 { 2839 union bpf_attr attr = {}; 2840 int err; 2841 2842 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 2843 return -EPERM; 2844 2845 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 2846 if (err) 2847 return err; 2848 size = min_t(u32, size, sizeof(attr)); 2849 2850 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 2851 if (copy_from_user(&attr, uattr, size) != 0) 2852 return -EFAULT; 2853 2854 err = security_bpf(cmd, &attr, size); 2855 if (err < 0) 2856 return err; 2857 2858 switch (cmd) { 2859 case BPF_MAP_CREATE: 2860 err = map_create(&attr); 2861 break; 2862 case BPF_MAP_LOOKUP_ELEM: 2863 err = map_lookup_elem(&attr); 2864 break; 2865 case BPF_MAP_UPDATE_ELEM: 2866 err = map_update_elem(&attr); 2867 break; 2868 case BPF_MAP_DELETE_ELEM: 2869 err = map_delete_elem(&attr); 2870 break; 2871 case BPF_MAP_GET_NEXT_KEY: 2872 err = map_get_next_key(&attr); 2873 break; 2874 case BPF_MAP_FREEZE: 2875 err = map_freeze(&attr); 2876 break; 2877 case BPF_PROG_LOAD: 2878 err = bpf_prog_load(&attr, uattr); 2879 break; 2880 case BPF_OBJ_PIN: 2881 err = bpf_obj_pin(&attr); 2882 break; 2883 case BPF_OBJ_GET: 2884 err = bpf_obj_get(&attr); 2885 break; 2886 case BPF_PROG_ATTACH: 2887 err = bpf_prog_attach(&attr); 2888 break; 2889 case BPF_PROG_DETACH: 2890 err = bpf_prog_detach(&attr); 2891 break; 2892 case BPF_PROG_QUERY: 2893 err = bpf_prog_query(&attr, uattr); 2894 break; 2895 case BPF_PROG_TEST_RUN: 2896 err = bpf_prog_test_run(&attr, uattr); 2897 break; 2898 case BPF_PROG_GET_NEXT_ID: 2899 err = bpf_obj_get_next_id(&attr, uattr, 2900 &prog_idr, &prog_idr_lock); 2901 break; 2902 case BPF_MAP_GET_NEXT_ID: 2903 err = bpf_obj_get_next_id(&attr, uattr, 2904 &map_idr, &map_idr_lock); 2905 break; 2906 case BPF_BTF_GET_NEXT_ID: 2907 err = bpf_obj_get_next_id(&attr, uattr, 2908 &btf_idr, &btf_idr_lock); 2909 break; 2910 case BPF_PROG_GET_FD_BY_ID: 2911 err = bpf_prog_get_fd_by_id(&attr); 2912 break; 2913 case BPF_MAP_GET_FD_BY_ID: 2914 err = bpf_map_get_fd_by_id(&attr); 2915 break; 2916 case BPF_OBJ_GET_INFO_BY_FD: 2917 err = bpf_obj_get_info_by_fd(&attr, uattr); 2918 break; 2919 case BPF_RAW_TRACEPOINT_OPEN: 2920 err = bpf_raw_tracepoint_open(&attr); 2921 break; 2922 case BPF_BTF_LOAD: 2923 err = bpf_btf_load(&attr); 2924 break; 2925 case BPF_BTF_GET_FD_BY_ID: 2926 err = bpf_btf_get_fd_by_id(&attr); 2927 break; 2928 case BPF_TASK_FD_QUERY: 2929 err = bpf_task_fd_query(&attr, uattr); 2930 break; 2931 case BPF_MAP_LOOKUP_AND_DELETE_ELEM: 2932 err = map_lookup_and_delete_elem(&attr); 2933 break; 2934 default: 2935 err = -EINVAL; 2936 break; 2937 } 2938 2939 return err; 2940 } 2941