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