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 goto done; 717 } 718 719 preempt_disable(); 720 this_cpu_inc(bpf_prog_active); 721 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 722 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 723 err = bpf_percpu_hash_copy(map, key, value); 724 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 725 err = bpf_percpu_array_copy(map, key, value); 726 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 727 err = bpf_percpu_cgroup_storage_copy(map, key, value); 728 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { 729 err = bpf_stackmap_copy(map, key, value); 730 } else if (IS_FD_ARRAY(map)) { 731 err = bpf_fd_array_map_lookup_elem(map, key, value); 732 } else if (IS_FD_HASH(map)) { 733 err = bpf_fd_htab_map_lookup_elem(map, key, value); 734 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 735 err = bpf_fd_reuseport_array_lookup_elem(map, key, value); 736 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 737 map->map_type == BPF_MAP_TYPE_STACK) { 738 err = map->ops->map_peek_elem(map, value); 739 } else { 740 rcu_read_lock(); 741 ptr = map->ops->map_lookup_elem(map, key); 742 if (IS_ERR(ptr)) { 743 err = PTR_ERR(ptr); 744 } else if (!ptr) { 745 err = -ENOENT; 746 } else { 747 err = 0; 748 memcpy(value, ptr, value_size); 749 } 750 rcu_read_unlock(); 751 } 752 this_cpu_dec(bpf_prog_active); 753 preempt_enable(); 754 755 done: 756 if (err) 757 goto free_value; 758 759 err = -EFAULT; 760 if (copy_to_user(uvalue, value, value_size) != 0) 761 goto free_value; 762 763 err = 0; 764 765 free_value: 766 kfree(value); 767 free_key: 768 kfree(key); 769 err_put: 770 fdput(f); 771 return err; 772 } 773 774 static void maybe_wait_bpf_programs(struct bpf_map *map) 775 { 776 /* Wait for any running BPF programs to complete so that 777 * userspace, when we return to it, knows that all programs 778 * that could be running use the new map value. 779 */ 780 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS || 781 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 782 synchronize_rcu(); 783 } 784 785 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags 786 787 static int map_update_elem(union bpf_attr *attr) 788 { 789 void __user *ukey = u64_to_user_ptr(attr->key); 790 void __user *uvalue = u64_to_user_ptr(attr->value); 791 int ufd = attr->map_fd; 792 struct bpf_map *map; 793 void *key, *value; 794 u32 value_size; 795 struct fd f; 796 int err; 797 798 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) 799 return -EINVAL; 800 801 f = fdget(ufd); 802 map = __bpf_map_get(f); 803 if (IS_ERR(map)) 804 return PTR_ERR(map); 805 806 if (!(f.file->f_mode & FMODE_CAN_WRITE)) { 807 err = -EPERM; 808 goto err_put; 809 } 810 811 key = __bpf_copy_key(ukey, map->key_size); 812 if (IS_ERR(key)) { 813 err = PTR_ERR(key); 814 goto err_put; 815 } 816 817 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 818 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 819 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 820 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 821 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 822 else 823 value_size = map->value_size; 824 825 err = -ENOMEM; 826 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 827 if (!value) 828 goto free_key; 829 830 err = -EFAULT; 831 if (copy_from_user(value, uvalue, value_size) != 0) 832 goto free_value; 833 834 /* Need to create a kthread, thus must support schedule */ 835 if (bpf_map_is_dev_bound(map)) { 836 err = bpf_map_offload_update_elem(map, key, value, attr->flags); 837 goto out; 838 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP || 839 map->map_type == BPF_MAP_TYPE_SOCKHASH || 840 map->map_type == BPF_MAP_TYPE_SOCKMAP) { 841 err = map->ops->map_update_elem(map, key, value, attr->flags); 842 goto out; 843 } 844 845 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from 846 * inside bpf map update or delete otherwise deadlocks are possible 847 */ 848 preempt_disable(); 849 __this_cpu_inc(bpf_prog_active); 850 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 851 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 852 err = bpf_percpu_hash_update(map, key, value, attr->flags); 853 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 854 err = bpf_percpu_array_update(map, key, value, attr->flags); 855 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 856 err = bpf_percpu_cgroup_storage_update(map, key, value, 857 attr->flags); 858 } else if (IS_FD_ARRAY(map)) { 859 rcu_read_lock(); 860 err = bpf_fd_array_map_update_elem(map, f.file, key, value, 861 attr->flags); 862 rcu_read_unlock(); 863 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { 864 rcu_read_lock(); 865 err = bpf_fd_htab_map_update_elem(map, f.file, key, value, 866 attr->flags); 867 rcu_read_unlock(); 868 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 869 /* rcu_read_lock() is not needed */ 870 err = bpf_fd_reuseport_array_update_elem(map, key, value, 871 attr->flags); 872 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 873 map->map_type == BPF_MAP_TYPE_STACK) { 874 err = map->ops->map_push_elem(map, value, attr->flags); 875 } else { 876 rcu_read_lock(); 877 err = map->ops->map_update_elem(map, key, value, attr->flags); 878 rcu_read_unlock(); 879 } 880 __this_cpu_dec(bpf_prog_active); 881 preempt_enable(); 882 maybe_wait_bpf_programs(map); 883 out: 884 free_value: 885 kfree(value); 886 free_key: 887 kfree(key); 888 err_put: 889 fdput(f); 890 return err; 891 } 892 893 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key 894 895 static int map_delete_elem(union bpf_attr *attr) 896 { 897 void __user *ukey = u64_to_user_ptr(attr->key); 898 int ufd = attr->map_fd; 899 struct bpf_map *map; 900 struct fd f; 901 void *key; 902 int err; 903 904 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) 905 return -EINVAL; 906 907 f = fdget(ufd); 908 map = __bpf_map_get(f); 909 if (IS_ERR(map)) 910 return PTR_ERR(map); 911 912 if (!(f.file->f_mode & FMODE_CAN_WRITE)) { 913 err = -EPERM; 914 goto err_put; 915 } 916 917 key = __bpf_copy_key(ukey, map->key_size); 918 if (IS_ERR(key)) { 919 err = PTR_ERR(key); 920 goto err_put; 921 } 922 923 if (bpf_map_is_dev_bound(map)) { 924 err = bpf_map_offload_delete_elem(map, key); 925 goto out; 926 } 927 928 preempt_disable(); 929 __this_cpu_inc(bpf_prog_active); 930 rcu_read_lock(); 931 err = map->ops->map_delete_elem(map, key); 932 rcu_read_unlock(); 933 __this_cpu_dec(bpf_prog_active); 934 preempt_enable(); 935 maybe_wait_bpf_programs(map); 936 out: 937 kfree(key); 938 err_put: 939 fdput(f); 940 return err; 941 } 942 943 /* last field in 'union bpf_attr' used by this command */ 944 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key 945 946 static int map_get_next_key(union bpf_attr *attr) 947 { 948 void __user *ukey = u64_to_user_ptr(attr->key); 949 void __user *unext_key = u64_to_user_ptr(attr->next_key); 950 int ufd = attr->map_fd; 951 struct bpf_map *map; 952 void *key, *next_key; 953 struct fd f; 954 int err; 955 956 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) 957 return -EINVAL; 958 959 f = fdget(ufd); 960 map = __bpf_map_get(f); 961 if (IS_ERR(map)) 962 return PTR_ERR(map); 963 964 if (!(f.file->f_mode & FMODE_CAN_READ)) { 965 err = -EPERM; 966 goto err_put; 967 } 968 969 if (ukey) { 970 key = __bpf_copy_key(ukey, map->key_size); 971 if (IS_ERR(key)) { 972 err = PTR_ERR(key); 973 goto err_put; 974 } 975 } else { 976 key = NULL; 977 } 978 979 err = -ENOMEM; 980 next_key = kmalloc(map->key_size, GFP_USER); 981 if (!next_key) 982 goto free_key; 983 984 if (bpf_map_is_dev_bound(map)) { 985 err = bpf_map_offload_get_next_key(map, key, next_key); 986 goto out; 987 } 988 989 rcu_read_lock(); 990 err = map->ops->map_get_next_key(map, key, next_key); 991 rcu_read_unlock(); 992 out: 993 if (err) 994 goto free_next_key; 995 996 err = -EFAULT; 997 if (copy_to_user(unext_key, next_key, map->key_size) != 0) 998 goto free_next_key; 999 1000 err = 0; 1001 1002 free_next_key: 1003 kfree(next_key); 1004 free_key: 1005 kfree(key); 1006 err_put: 1007 fdput(f); 1008 return err; 1009 } 1010 1011 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value 1012 1013 static int map_lookup_and_delete_elem(union bpf_attr *attr) 1014 { 1015 void __user *ukey = u64_to_user_ptr(attr->key); 1016 void __user *uvalue = u64_to_user_ptr(attr->value); 1017 int ufd = attr->map_fd; 1018 struct bpf_map *map; 1019 void *key, *value; 1020 u32 value_size; 1021 struct fd f; 1022 int err; 1023 1024 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM)) 1025 return -EINVAL; 1026 1027 f = fdget(ufd); 1028 map = __bpf_map_get(f); 1029 if (IS_ERR(map)) 1030 return PTR_ERR(map); 1031 1032 if (!(f.file->f_mode & FMODE_CAN_WRITE)) { 1033 err = -EPERM; 1034 goto err_put; 1035 } 1036 1037 key = __bpf_copy_key(ukey, map->key_size); 1038 if (IS_ERR(key)) { 1039 err = PTR_ERR(key); 1040 goto err_put; 1041 } 1042 1043 value_size = map->value_size; 1044 1045 err = -ENOMEM; 1046 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1047 if (!value) 1048 goto free_key; 1049 1050 if (map->map_type == BPF_MAP_TYPE_QUEUE || 1051 map->map_type == BPF_MAP_TYPE_STACK) { 1052 err = map->ops->map_pop_elem(map, value); 1053 } else { 1054 err = -ENOTSUPP; 1055 } 1056 1057 if (err) 1058 goto free_value; 1059 1060 if (copy_to_user(uvalue, value, value_size) != 0) 1061 goto free_value; 1062 1063 err = 0; 1064 1065 free_value: 1066 kfree(value); 1067 free_key: 1068 kfree(key); 1069 err_put: 1070 fdput(f); 1071 return err; 1072 } 1073 1074 static const struct bpf_prog_ops * const bpf_prog_types[] = { 1075 #define BPF_PROG_TYPE(_id, _name) \ 1076 [_id] = & _name ## _prog_ops, 1077 #define BPF_MAP_TYPE(_id, _ops) 1078 #include <linux/bpf_types.h> 1079 #undef BPF_PROG_TYPE 1080 #undef BPF_MAP_TYPE 1081 }; 1082 1083 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) 1084 { 1085 const struct bpf_prog_ops *ops; 1086 1087 if (type >= ARRAY_SIZE(bpf_prog_types)) 1088 return -EINVAL; 1089 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types)); 1090 ops = bpf_prog_types[type]; 1091 if (!ops) 1092 return -EINVAL; 1093 1094 if (!bpf_prog_is_dev_bound(prog->aux)) 1095 prog->aux->ops = ops; 1096 else 1097 prog->aux->ops = &bpf_offload_prog_ops; 1098 prog->type = type; 1099 return 0; 1100 } 1101 1102 /* drop refcnt on maps used by eBPF program and free auxilary data */ 1103 static void free_used_maps(struct bpf_prog_aux *aux) 1104 { 1105 enum bpf_cgroup_storage_type stype; 1106 int i; 1107 1108 for_each_cgroup_storage_type(stype) { 1109 if (!aux->cgroup_storage[stype]) 1110 continue; 1111 bpf_cgroup_storage_release(aux->prog, 1112 aux->cgroup_storage[stype]); 1113 } 1114 1115 for (i = 0; i < aux->used_map_cnt; i++) 1116 bpf_map_put(aux->used_maps[i]); 1117 1118 kfree(aux->used_maps); 1119 } 1120 1121 int __bpf_prog_charge(struct user_struct *user, u32 pages) 1122 { 1123 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1124 unsigned long user_bufs; 1125 1126 if (user) { 1127 user_bufs = atomic_long_add_return(pages, &user->locked_vm); 1128 if (user_bufs > memlock_limit) { 1129 atomic_long_sub(pages, &user->locked_vm); 1130 return -EPERM; 1131 } 1132 } 1133 1134 return 0; 1135 } 1136 1137 void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 1138 { 1139 if (user) 1140 atomic_long_sub(pages, &user->locked_vm); 1141 } 1142 1143 static int bpf_prog_charge_memlock(struct bpf_prog *prog) 1144 { 1145 struct user_struct *user = get_current_user(); 1146 int ret; 1147 1148 ret = __bpf_prog_charge(user, prog->pages); 1149 if (ret) { 1150 free_uid(user); 1151 return ret; 1152 } 1153 1154 prog->aux->user = user; 1155 return 0; 1156 } 1157 1158 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) 1159 { 1160 struct user_struct *user = prog->aux->user; 1161 1162 __bpf_prog_uncharge(user, prog->pages); 1163 free_uid(user); 1164 } 1165 1166 static int bpf_prog_alloc_id(struct bpf_prog *prog) 1167 { 1168 int id; 1169 1170 idr_preload(GFP_KERNEL); 1171 spin_lock_bh(&prog_idr_lock); 1172 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); 1173 if (id > 0) 1174 prog->aux->id = id; 1175 spin_unlock_bh(&prog_idr_lock); 1176 idr_preload_end(); 1177 1178 /* id is in [1, INT_MAX) */ 1179 if (WARN_ON_ONCE(!id)) 1180 return -ENOSPC; 1181 1182 return id > 0 ? 0 : id; 1183 } 1184 1185 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) 1186 { 1187 /* cBPF to eBPF migrations are currently not in the idr store. 1188 * Offloaded programs are removed from the store when their device 1189 * disappears - even if someone grabs an fd to them they are unusable, 1190 * simply waiting for refcnt to drop to be freed. 1191 */ 1192 if (!prog->aux->id) 1193 return; 1194 1195 if (do_idr_lock) 1196 spin_lock_bh(&prog_idr_lock); 1197 else 1198 __acquire(&prog_idr_lock); 1199 1200 idr_remove(&prog_idr, prog->aux->id); 1201 prog->aux->id = 0; 1202 1203 if (do_idr_lock) 1204 spin_unlock_bh(&prog_idr_lock); 1205 else 1206 __release(&prog_idr_lock); 1207 } 1208 1209 static void __bpf_prog_put_rcu(struct rcu_head *rcu) 1210 { 1211 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1212 1213 free_used_maps(aux); 1214 bpf_prog_uncharge_memlock(aux->prog); 1215 security_bpf_prog_free(aux); 1216 bpf_prog_free(aux->prog); 1217 } 1218 1219 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) 1220 { 1221 if (atomic_dec_and_test(&prog->aux->refcnt)) { 1222 /* bpf_prog_free_id() must be called first */ 1223 bpf_prog_free_id(prog, do_idr_lock); 1224 bpf_prog_kallsyms_del_all(prog); 1225 btf_put(prog->aux->btf); 1226 kvfree(prog->aux->func_info); 1227 bpf_prog_free_linfo(prog); 1228 1229 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1230 } 1231 } 1232 1233 void bpf_prog_put(struct bpf_prog *prog) 1234 { 1235 __bpf_prog_put(prog, true); 1236 } 1237 EXPORT_SYMBOL_GPL(bpf_prog_put); 1238 1239 static int bpf_prog_release(struct inode *inode, struct file *filp) 1240 { 1241 struct bpf_prog *prog = filp->private_data; 1242 1243 bpf_prog_put(prog); 1244 return 0; 1245 } 1246 1247 #ifdef CONFIG_PROC_FS 1248 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1249 { 1250 const struct bpf_prog *prog = filp->private_data; 1251 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1252 1253 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1254 seq_printf(m, 1255 "prog_type:\t%u\n" 1256 "prog_jited:\t%u\n" 1257 "prog_tag:\t%s\n" 1258 "memlock:\t%llu\n" 1259 "prog_id:\t%u\n", 1260 prog->type, 1261 prog->jited, 1262 prog_tag, 1263 prog->pages * 1ULL << PAGE_SHIFT, 1264 prog->aux->id); 1265 } 1266 #endif 1267 1268 const struct file_operations bpf_prog_fops = { 1269 #ifdef CONFIG_PROC_FS 1270 .show_fdinfo = bpf_prog_show_fdinfo, 1271 #endif 1272 .release = bpf_prog_release, 1273 .read = bpf_dummy_read, 1274 .write = bpf_dummy_write, 1275 }; 1276 1277 int bpf_prog_new_fd(struct bpf_prog *prog) 1278 { 1279 int ret; 1280 1281 ret = security_bpf_prog(prog); 1282 if (ret < 0) 1283 return ret; 1284 1285 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1286 O_RDWR | O_CLOEXEC); 1287 } 1288 1289 static struct bpf_prog *____bpf_prog_get(struct fd f) 1290 { 1291 if (!f.file) 1292 return ERR_PTR(-EBADF); 1293 if (f.file->f_op != &bpf_prog_fops) { 1294 fdput(f); 1295 return ERR_PTR(-EINVAL); 1296 } 1297 1298 return f.file->private_data; 1299 } 1300 1301 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) 1302 { 1303 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { 1304 atomic_sub(i, &prog->aux->refcnt); 1305 return ERR_PTR(-EBUSY); 1306 } 1307 return prog; 1308 } 1309 EXPORT_SYMBOL_GPL(bpf_prog_add); 1310 1311 void bpf_prog_sub(struct bpf_prog *prog, int i) 1312 { 1313 /* Only to be used for undoing previous bpf_prog_add() in some 1314 * error path. We still know that another entity in our call 1315 * path holds a reference to the program, thus atomic_sub() can 1316 * be safely used in such cases! 1317 */ 1318 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); 1319 } 1320 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1321 1322 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) 1323 { 1324 return bpf_prog_add(prog, 1); 1325 } 1326 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1327 1328 /* prog_idr_lock should have been held */ 1329 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1330 { 1331 int refold; 1332 1333 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1334 1335 if (refold >= BPF_MAX_REFCNT) { 1336 __bpf_prog_put(prog, false); 1337 return ERR_PTR(-EBUSY); 1338 } 1339 1340 if (!refold) 1341 return ERR_PTR(-ENOENT); 1342 1343 return prog; 1344 } 1345 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1346 1347 bool bpf_prog_get_ok(struct bpf_prog *prog, 1348 enum bpf_prog_type *attach_type, bool attach_drv) 1349 { 1350 /* not an attachment, just a refcount inc, always allow */ 1351 if (!attach_type) 1352 return true; 1353 1354 if (prog->type != *attach_type) 1355 return false; 1356 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1357 return false; 1358 1359 return true; 1360 } 1361 1362 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1363 bool attach_drv) 1364 { 1365 struct fd f = fdget(ufd); 1366 struct bpf_prog *prog; 1367 1368 prog = ____bpf_prog_get(f); 1369 if (IS_ERR(prog)) 1370 return prog; 1371 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1372 prog = ERR_PTR(-EINVAL); 1373 goto out; 1374 } 1375 1376 prog = bpf_prog_inc(prog); 1377 out: 1378 fdput(f); 1379 return prog; 1380 } 1381 1382 struct bpf_prog *bpf_prog_get(u32 ufd) 1383 { 1384 return __bpf_prog_get(ufd, NULL, false); 1385 } 1386 1387 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1388 bool attach_drv) 1389 { 1390 return __bpf_prog_get(ufd, &type, attach_drv); 1391 } 1392 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1393 1394 /* Initially all BPF programs could be loaded w/o specifying 1395 * expected_attach_type. Later for some of them specifying expected_attach_type 1396 * at load time became required so that program could be validated properly. 1397 * Programs of types that are allowed to be loaded both w/ and w/o (for 1398 * backward compatibility) expected_attach_type, should have the default attach 1399 * type assigned to expected_attach_type for the latter case, so that it can be 1400 * validated later at attach time. 1401 * 1402 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1403 * prog type requires it but has some attach types that have to be backward 1404 * compatible. 1405 */ 1406 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1407 { 1408 switch (attr->prog_type) { 1409 case BPF_PROG_TYPE_CGROUP_SOCK: 1410 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1411 * exist so checking for non-zero is the way to go here. 1412 */ 1413 if (!attr->expected_attach_type) 1414 attr->expected_attach_type = 1415 BPF_CGROUP_INET_SOCK_CREATE; 1416 break; 1417 } 1418 } 1419 1420 static int 1421 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type, 1422 enum bpf_attach_type expected_attach_type) 1423 { 1424 switch (prog_type) { 1425 case BPF_PROG_TYPE_CGROUP_SOCK: 1426 switch (expected_attach_type) { 1427 case BPF_CGROUP_INET_SOCK_CREATE: 1428 case BPF_CGROUP_INET4_POST_BIND: 1429 case BPF_CGROUP_INET6_POST_BIND: 1430 return 0; 1431 default: 1432 return -EINVAL; 1433 } 1434 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1435 switch (expected_attach_type) { 1436 case BPF_CGROUP_INET4_BIND: 1437 case BPF_CGROUP_INET6_BIND: 1438 case BPF_CGROUP_INET4_CONNECT: 1439 case BPF_CGROUP_INET6_CONNECT: 1440 case BPF_CGROUP_UDP4_SENDMSG: 1441 case BPF_CGROUP_UDP6_SENDMSG: 1442 return 0; 1443 default: 1444 return -EINVAL; 1445 } 1446 default: 1447 return 0; 1448 } 1449 } 1450 1451 /* last field in 'union bpf_attr' used by this command */ 1452 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt 1453 1454 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr) 1455 { 1456 enum bpf_prog_type type = attr->prog_type; 1457 struct bpf_prog *prog; 1458 int err; 1459 char license[128]; 1460 bool is_gpl; 1461 1462 if (CHECK_ATTR(BPF_PROG_LOAD)) 1463 return -EINVAL; 1464 1465 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT)) 1466 return -EINVAL; 1467 1468 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && 1469 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) && 1470 !capable(CAP_SYS_ADMIN)) 1471 return -EPERM; 1472 1473 /* copy eBPF program license from user space */ 1474 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 1475 sizeof(license) - 1) < 0) 1476 return -EFAULT; 1477 license[sizeof(license) - 1] = 0; 1478 1479 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1480 is_gpl = license_is_gpl_compatible(license); 1481 1482 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) 1483 return -E2BIG; 1484 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 1485 type != BPF_PROG_TYPE_CGROUP_SKB && 1486 !capable(CAP_SYS_ADMIN)) 1487 return -EPERM; 1488 1489 bpf_prog_load_fixup_attach_type(attr); 1490 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type)) 1491 return -EINVAL; 1492 1493 /* plain bpf_prog allocation */ 1494 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 1495 if (!prog) 1496 return -ENOMEM; 1497 1498 prog->expected_attach_type = attr->expected_attach_type; 1499 1500 prog->aux->offload_requested = !!attr->prog_ifindex; 1501 1502 err = security_bpf_prog_alloc(prog->aux); 1503 if (err) 1504 goto free_prog_nouncharge; 1505 1506 err = bpf_prog_charge_memlock(prog); 1507 if (err) 1508 goto free_prog_sec; 1509 1510 prog->len = attr->insn_cnt; 1511 1512 err = -EFAULT; 1513 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 1514 bpf_prog_insn_size(prog)) != 0) 1515 goto free_prog; 1516 1517 prog->orig_prog = NULL; 1518 prog->jited = 0; 1519 1520 atomic_set(&prog->aux->refcnt, 1); 1521 prog->gpl_compatible = is_gpl ? 1 : 0; 1522 1523 if (bpf_prog_is_dev_bound(prog->aux)) { 1524 err = bpf_prog_offload_init(prog, attr); 1525 if (err) 1526 goto free_prog; 1527 } 1528 1529 /* find program type: socket_filter vs tracing_filter */ 1530 err = find_prog_type(type, prog); 1531 if (err < 0) 1532 goto free_prog; 1533 1534 prog->aux->load_time = ktime_get_boot_ns(); 1535 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 1536 if (err) 1537 goto free_prog; 1538 1539 /* run eBPF verifier */ 1540 err = bpf_check(&prog, attr, uattr); 1541 if (err < 0) 1542 goto free_used_maps; 1543 1544 prog = bpf_prog_select_runtime(prog, &err); 1545 if (err < 0) 1546 goto free_used_maps; 1547 1548 err = bpf_prog_alloc_id(prog); 1549 if (err) 1550 goto free_used_maps; 1551 1552 err = bpf_prog_new_fd(prog); 1553 if (err < 0) { 1554 /* failed to allocate fd. 1555 * bpf_prog_put() is needed because the above 1556 * bpf_prog_alloc_id() has published the prog 1557 * to the userspace and the userspace may 1558 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. 1559 */ 1560 bpf_prog_put(prog); 1561 return err; 1562 } 1563 1564 bpf_prog_kallsyms_add(prog); 1565 return err; 1566 1567 free_used_maps: 1568 bpf_prog_free_linfo(prog); 1569 kvfree(prog->aux->func_info); 1570 btf_put(prog->aux->btf); 1571 bpf_prog_kallsyms_del_subprogs(prog); 1572 free_used_maps(prog->aux); 1573 free_prog: 1574 bpf_prog_uncharge_memlock(prog); 1575 free_prog_sec: 1576 security_bpf_prog_free(prog->aux); 1577 free_prog_nouncharge: 1578 bpf_prog_free(prog); 1579 return err; 1580 } 1581 1582 #define BPF_OBJ_LAST_FIELD file_flags 1583 1584 static int bpf_obj_pin(const union bpf_attr *attr) 1585 { 1586 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 1587 return -EINVAL; 1588 1589 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 1590 } 1591 1592 static int bpf_obj_get(const union bpf_attr *attr) 1593 { 1594 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 1595 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 1596 return -EINVAL; 1597 1598 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 1599 attr->file_flags); 1600 } 1601 1602 struct bpf_raw_tracepoint { 1603 struct bpf_raw_event_map *btp; 1604 struct bpf_prog *prog; 1605 }; 1606 1607 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp) 1608 { 1609 struct bpf_raw_tracepoint *raw_tp = filp->private_data; 1610 1611 if (raw_tp->prog) { 1612 bpf_probe_unregister(raw_tp->btp, raw_tp->prog); 1613 bpf_prog_put(raw_tp->prog); 1614 } 1615 bpf_put_raw_tracepoint(raw_tp->btp); 1616 kfree(raw_tp); 1617 return 0; 1618 } 1619 1620 static const struct file_operations bpf_raw_tp_fops = { 1621 .release = bpf_raw_tracepoint_release, 1622 .read = bpf_dummy_read, 1623 .write = bpf_dummy_write, 1624 }; 1625 1626 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 1627 1628 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 1629 { 1630 struct bpf_raw_tracepoint *raw_tp; 1631 struct bpf_raw_event_map *btp; 1632 struct bpf_prog *prog; 1633 char tp_name[128]; 1634 int tp_fd, err; 1635 1636 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name), 1637 sizeof(tp_name) - 1) < 0) 1638 return -EFAULT; 1639 tp_name[sizeof(tp_name) - 1] = 0; 1640 1641 btp = bpf_get_raw_tracepoint(tp_name); 1642 if (!btp) 1643 return -ENOENT; 1644 1645 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER); 1646 if (!raw_tp) { 1647 err = -ENOMEM; 1648 goto out_put_btp; 1649 } 1650 raw_tp->btp = btp; 1651 1652 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd, 1653 BPF_PROG_TYPE_RAW_TRACEPOINT); 1654 if (IS_ERR(prog)) { 1655 err = PTR_ERR(prog); 1656 goto out_free_tp; 1657 } 1658 1659 err = bpf_probe_register(raw_tp->btp, prog); 1660 if (err) 1661 goto out_put_prog; 1662 1663 raw_tp->prog = prog; 1664 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp, 1665 O_CLOEXEC); 1666 if (tp_fd < 0) { 1667 bpf_probe_unregister(raw_tp->btp, prog); 1668 err = tp_fd; 1669 goto out_put_prog; 1670 } 1671 return tp_fd; 1672 1673 out_put_prog: 1674 bpf_prog_put(prog); 1675 out_free_tp: 1676 kfree(raw_tp); 1677 out_put_btp: 1678 bpf_put_raw_tracepoint(btp); 1679 return err; 1680 } 1681 1682 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 1683 enum bpf_attach_type attach_type) 1684 { 1685 switch (prog->type) { 1686 case BPF_PROG_TYPE_CGROUP_SOCK: 1687 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1688 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 1689 default: 1690 return 0; 1691 } 1692 } 1693 1694 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags 1695 1696 #define BPF_F_ATTACH_MASK \ 1697 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI) 1698 1699 static int bpf_prog_attach(const union bpf_attr *attr) 1700 { 1701 enum bpf_prog_type ptype; 1702 struct bpf_prog *prog; 1703 int ret; 1704 1705 if (!capable(CAP_NET_ADMIN)) 1706 return -EPERM; 1707 1708 if (CHECK_ATTR(BPF_PROG_ATTACH)) 1709 return -EINVAL; 1710 1711 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 1712 return -EINVAL; 1713 1714 switch (attr->attach_type) { 1715 case BPF_CGROUP_INET_INGRESS: 1716 case BPF_CGROUP_INET_EGRESS: 1717 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1718 break; 1719 case BPF_CGROUP_INET_SOCK_CREATE: 1720 case BPF_CGROUP_INET4_POST_BIND: 1721 case BPF_CGROUP_INET6_POST_BIND: 1722 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1723 break; 1724 case BPF_CGROUP_INET4_BIND: 1725 case BPF_CGROUP_INET6_BIND: 1726 case BPF_CGROUP_INET4_CONNECT: 1727 case BPF_CGROUP_INET6_CONNECT: 1728 case BPF_CGROUP_UDP4_SENDMSG: 1729 case BPF_CGROUP_UDP6_SENDMSG: 1730 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1731 break; 1732 case BPF_CGROUP_SOCK_OPS: 1733 ptype = BPF_PROG_TYPE_SOCK_OPS; 1734 break; 1735 case BPF_CGROUP_DEVICE: 1736 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1737 break; 1738 case BPF_SK_MSG_VERDICT: 1739 ptype = BPF_PROG_TYPE_SK_MSG; 1740 break; 1741 case BPF_SK_SKB_STREAM_PARSER: 1742 case BPF_SK_SKB_STREAM_VERDICT: 1743 ptype = BPF_PROG_TYPE_SK_SKB; 1744 break; 1745 case BPF_LIRC_MODE2: 1746 ptype = BPF_PROG_TYPE_LIRC_MODE2; 1747 break; 1748 case BPF_FLOW_DISSECTOR: 1749 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 1750 break; 1751 default: 1752 return -EINVAL; 1753 } 1754 1755 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1756 if (IS_ERR(prog)) 1757 return PTR_ERR(prog); 1758 1759 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 1760 bpf_prog_put(prog); 1761 return -EINVAL; 1762 } 1763 1764 switch (ptype) { 1765 case BPF_PROG_TYPE_SK_SKB: 1766 case BPF_PROG_TYPE_SK_MSG: 1767 ret = sock_map_get_from_fd(attr, prog); 1768 break; 1769 case BPF_PROG_TYPE_LIRC_MODE2: 1770 ret = lirc_prog_attach(attr, prog); 1771 break; 1772 case BPF_PROG_TYPE_FLOW_DISSECTOR: 1773 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 1774 break; 1775 default: 1776 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 1777 } 1778 1779 if (ret) 1780 bpf_prog_put(prog); 1781 return ret; 1782 } 1783 1784 #define BPF_PROG_DETACH_LAST_FIELD attach_type 1785 1786 static int bpf_prog_detach(const union bpf_attr *attr) 1787 { 1788 enum bpf_prog_type ptype; 1789 1790 if (!capable(CAP_NET_ADMIN)) 1791 return -EPERM; 1792 1793 if (CHECK_ATTR(BPF_PROG_DETACH)) 1794 return -EINVAL; 1795 1796 switch (attr->attach_type) { 1797 case BPF_CGROUP_INET_INGRESS: 1798 case BPF_CGROUP_INET_EGRESS: 1799 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1800 break; 1801 case BPF_CGROUP_INET_SOCK_CREATE: 1802 case BPF_CGROUP_INET4_POST_BIND: 1803 case BPF_CGROUP_INET6_POST_BIND: 1804 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1805 break; 1806 case BPF_CGROUP_INET4_BIND: 1807 case BPF_CGROUP_INET6_BIND: 1808 case BPF_CGROUP_INET4_CONNECT: 1809 case BPF_CGROUP_INET6_CONNECT: 1810 case BPF_CGROUP_UDP4_SENDMSG: 1811 case BPF_CGROUP_UDP6_SENDMSG: 1812 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1813 break; 1814 case BPF_CGROUP_SOCK_OPS: 1815 ptype = BPF_PROG_TYPE_SOCK_OPS; 1816 break; 1817 case BPF_CGROUP_DEVICE: 1818 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1819 break; 1820 case BPF_SK_MSG_VERDICT: 1821 return sock_map_get_from_fd(attr, NULL); 1822 case BPF_SK_SKB_STREAM_PARSER: 1823 case BPF_SK_SKB_STREAM_VERDICT: 1824 return sock_map_get_from_fd(attr, NULL); 1825 case BPF_LIRC_MODE2: 1826 return lirc_prog_detach(attr); 1827 case BPF_FLOW_DISSECTOR: 1828 return skb_flow_dissector_bpf_prog_detach(attr); 1829 default: 1830 return -EINVAL; 1831 } 1832 1833 return cgroup_bpf_prog_detach(attr, ptype); 1834 } 1835 1836 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 1837 1838 static int bpf_prog_query(const union bpf_attr *attr, 1839 union bpf_attr __user *uattr) 1840 { 1841 if (!capable(CAP_NET_ADMIN)) 1842 return -EPERM; 1843 if (CHECK_ATTR(BPF_PROG_QUERY)) 1844 return -EINVAL; 1845 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 1846 return -EINVAL; 1847 1848 switch (attr->query.attach_type) { 1849 case BPF_CGROUP_INET_INGRESS: 1850 case BPF_CGROUP_INET_EGRESS: 1851 case BPF_CGROUP_INET_SOCK_CREATE: 1852 case BPF_CGROUP_INET4_BIND: 1853 case BPF_CGROUP_INET6_BIND: 1854 case BPF_CGROUP_INET4_POST_BIND: 1855 case BPF_CGROUP_INET6_POST_BIND: 1856 case BPF_CGROUP_INET4_CONNECT: 1857 case BPF_CGROUP_INET6_CONNECT: 1858 case BPF_CGROUP_UDP4_SENDMSG: 1859 case BPF_CGROUP_UDP6_SENDMSG: 1860 case BPF_CGROUP_SOCK_OPS: 1861 case BPF_CGROUP_DEVICE: 1862 break; 1863 case BPF_LIRC_MODE2: 1864 return lirc_prog_query(attr, uattr); 1865 default: 1866 return -EINVAL; 1867 } 1868 1869 return cgroup_bpf_prog_query(attr, uattr); 1870 } 1871 1872 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration 1873 1874 static int bpf_prog_test_run(const union bpf_attr *attr, 1875 union bpf_attr __user *uattr) 1876 { 1877 struct bpf_prog *prog; 1878 int ret = -ENOTSUPP; 1879 1880 if (!capable(CAP_SYS_ADMIN)) 1881 return -EPERM; 1882 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 1883 return -EINVAL; 1884 1885 prog = bpf_prog_get(attr->test.prog_fd); 1886 if (IS_ERR(prog)) 1887 return PTR_ERR(prog); 1888 1889 if (prog->aux->ops->test_run) 1890 ret = prog->aux->ops->test_run(prog, attr, uattr); 1891 1892 bpf_prog_put(prog); 1893 return ret; 1894 } 1895 1896 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 1897 1898 static int bpf_obj_get_next_id(const union bpf_attr *attr, 1899 union bpf_attr __user *uattr, 1900 struct idr *idr, 1901 spinlock_t *lock) 1902 { 1903 u32 next_id = attr->start_id; 1904 int err = 0; 1905 1906 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 1907 return -EINVAL; 1908 1909 if (!capable(CAP_SYS_ADMIN)) 1910 return -EPERM; 1911 1912 next_id++; 1913 spin_lock_bh(lock); 1914 if (!idr_get_next(idr, &next_id)) 1915 err = -ENOENT; 1916 spin_unlock_bh(lock); 1917 1918 if (!err) 1919 err = put_user(next_id, &uattr->next_id); 1920 1921 return err; 1922 } 1923 1924 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 1925 1926 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 1927 { 1928 struct bpf_prog *prog; 1929 u32 id = attr->prog_id; 1930 int fd; 1931 1932 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 1933 return -EINVAL; 1934 1935 if (!capable(CAP_SYS_ADMIN)) 1936 return -EPERM; 1937 1938 spin_lock_bh(&prog_idr_lock); 1939 prog = idr_find(&prog_idr, id); 1940 if (prog) 1941 prog = bpf_prog_inc_not_zero(prog); 1942 else 1943 prog = ERR_PTR(-ENOENT); 1944 spin_unlock_bh(&prog_idr_lock); 1945 1946 if (IS_ERR(prog)) 1947 return PTR_ERR(prog); 1948 1949 fd = bpf_prog_new_fd(prog); 1950 if (fd < 0) 1951 bpf_prog_put(prog); 1952 1953 return fd; 1954 } 1955 1956 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 1957 1958 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 1959 { 1960 struct bpf_map *map; 1961 u32 id = attr->map_id; 1962 int f_flags; 1963 int fd; 1964 1965 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 1966 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 1967 return -EINVAL; 1968 1969 if (!capable(CAP_SYS_ADMIN)) 1970 return -EPERM; 1971 1972 f_flags = bpf_get_file_flag(attr->open_flags); 1973 if (f_flags < 0) 1974 return f_flags; 1975 1976 spin_lock_bh(&map_idr_lock); 1977 map = idr_find(&map_idr, id); 1978 if (map) 1979 map = bpf_map_inc_not_zero(map, true); 1980 else 1981 map = ERR_PTR(-ENOENT); 1982 spin_unlock_bh(&map_idr_lock); 1983 1984 if (IS_ERR(map)) 1985 return PTR_ERR(map); 1986 1987 fd = bpf_map_new_fd(map, f_flags); 1988 if (fd < 0) 1989 bpf_map_put(map); 1990 1991 return fd; 1992 } 1993 1994 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 1995 unsigned long addr) 1996 { 1997 int i; 1998 1999 for (i = 0; i < prog->aux->used_map_cnt; i++) 2000 if (prog->aux->used_maps[i] == (void *)addr) 2001 return prog->aux->used_maps[i]; 2002 return NULL; 2003 } 2004 2005 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 2006 { 2007 const struct bpf_map *map; 2008 struct bpf_insn *insns; 2009 u64 imm; 2010 int i; 2011 2012 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 2013 GFP_USER); 2014 if (!insns) 2015 return insns; 2016 2017 for (i = 0; i < prog->len; i++) { 2018 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 2019 insns[i].code = BPF_JMP | BPF_CALL; 2020 insns[i].imm = BPF_FUNC_tail_call; 2021 /* fall-through */ 2022 } 2023 if (insns[i].code == (BPF_JMP | BPF_CALL) || 2024 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 2025 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 2026 insns[i].code = BPF_JMP | BPF_CALL; 2027 if (!bpf_dump_raw_ok()) 2028 insns[i].imm = 0; 2029 continue; 2030 } 2031 2032 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 2033 continue; 2034 2035 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 2036 map = bpf_map_from_imm(prog, imm); 2037 if (map) { 2038 insns[i].src_reg = BPF_PSEUDO_MAP_FD; 2039 insns[i].imm = map->id; 2040 insns[i + 1].imm = 0; 2041 continue; 2042 } 2043 } 2044 2045 return insns; 2046 } 2047 2048 static int set_info_rec_size(struct bpf_prog_info *info) 2049 { 2050 /* 2051 * Ensure info.*_rec_size is the same as kernel expected size 2052 * 2053 * or 2054 * 2055 * Only allow zero *_rec_size if both _rec_size and _cnt are 2056 * zero. In this case, the kernel will set the expected 2057 * _rec_size back to the info. 2058 */ 2059 2060 if ((info->nr_func_info || info->func_info_rec_size) && 2061 info->func_info_rec_size != sizeof(struct bpf_func_info)) 2062 return -EINVAL; 2063 2064 if ((info->nr_line_info || info->line_info_rec_size) && 2065 info->line_info_rec_size != sizeof(struct bpf_line_info)) 2066 return -EINVAL; 2067 2068 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) && 2069 info->jited_line_info_rec_size != sizeof(__u64)) 2070 return -EINVAL; 2071 2072 info->func_info_rec_size = sizeof(struct bpf_func_info); 2073 info->line_info_rec_size = sizeof(struct bpf_line_info); 2074 info->jited_line_info_rec_size = sizeof(__u64); 2075 2076 return 0; 2077 } 2078 2079 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 2080 const union bpf_attr *attr, 2081 union bpf_attr __user *uattr) 2082 { 2083 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2084 struct bpf_prog_info info = {}; 2085 u32 info_len = attr->info.info_len; 2086 char __user *uinsns; 2087 u32 ulen; 2088 int err; 2089 2090 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2091 if (err) 2092 return err; 2093 info_len = min_t(u32, sizeof(info), info_len); 2094 2095 if (copy_from_user(&info, uinfo, info_len)) 2096 return -EFAULT; 2097 2098 info.type = prog->type; 2099 info.id = prog->aux->id; 2100 info.load_time = prog->aux->load_time; 2101 info.created_by_uid = from_kuid_munged(current_user_ns(), 2102 prog->aux->user->uid); 2103 info.gpl_compatible = prog->gpl_compatible; 2104 2105 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 2106 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 2107 2108 ulen = info.nr_map_ids; 2109 info.nr_map_ids = prog->aux->used_map_cnt; 2110 ulen = min_t(u32, info.nr_map_ids, ulen); 2111 if (ulen) { 2112 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 2113 u32 i; 2114 2115 for (i = 0; i < ulen; i++) 2116 if (put_user(prog->aux->used_maps[i]->id, 2117 &user_map_ids[i])) 2118 return -EFAULT; 2119 } 2120 2121 err = set_info_rec_size(&info); 2122 if (err) 2123 return err; 2124 2125 if (!capable(CAP_SYS_ADMIN)) { 2126 info.jited_prog_len = 0; 2127 info.xlated_prog_len = 0; 2128 info.nr_jited_ksyms = 0; 2129 info.nr_jited_func_lens = 0; 2130 info.nr_func_info = 0; 2131 info.nr_line_info = 0; 2132 info.nr_jited_line_info = 0; 2133 goto done; 2134 } 2135 2136 ulen = info.xlated_prog_len; 2137 info.xlated_prog_len = bpf_prog_insn_size(prog); 2138 if (info.xlated_prog_len && ulen) { 2139 struct bpf_insn *insns_sanitized; 2140 bool fault; 2141 2142 if (prog->blinded && !bpf_dump_raw_ok()) { 2143 info.xlated_prog_insns = 0; 2144 goto done; 2145 } 2146 insns_sanitized = bpf_insn_prepare_dump(prog); 2147 if (!insns_sanitized) 2148 return -ENOMEM; 2149 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 2150 ulen = min_t(u32, info.xlated_prog_len, ulen); 2151 fault = copy_to_user(uinsns, insns_sanitized, ulen); 2152 kfree(insns_sanitized); 2153 if (fault) 2154 return -EFAULT; 2155 } 2156 2157 if (bpf_prog_is_dev_bound(prog->aux)) { 2158 err = bpf_prog_offload_info_fill(&info, prog); 2159 if (err) 2160 return err; 2161 goto done; 2162 } 2163 2164 /* NOTE: the following code is supposed to be skipped for offload. 2165 * bpf_prog_offload_info_fill() is the place to fill similar fields 2166 * for offload. 2167 */ 2168 ulen = info.jited_prog_len; 2169 if (prog->aux->func_cnt) { 2170 u32 i; 2171 2172 info.jited_prog_len = 0; 2173 for (i = 0; i < prog->aux->func_cnt; i++) 2174 info.jited_prog_len += prog->aux->func[i]->jited_len; 2175 } else { 2176 info.jited_prog_len = prog->jited_len; 2177 } 2178 2179 if (info.jited_prog_len && ulen) { 2180 if (bpf_dump_raw_ok()) { 2181 uinsns = u64_to_user_ptr(info.jited_prog_insns); 2182 ulen = min_t(u32, info.jited_prog_len, ulen); 2183 2184 /* for multi-function programs, copy the JITed 2185 * instructions for all the functions 2186 */ 2187 if (prog->aux->func_cnt) { 2188 u32 len, free, i; 2189 u8 *img; 2190 2191 free = ulen; 2192 for (i = 0; i < prog->aux->func_cnt; i++) { 2193 len = prog->aux->func[i]->jited_len; 2194 len = min_t(u32, len, free); 2195 img = (u8 *) prog->aux->func[i]->bpf_func; 2196 if (copy_to_user(uinsns, img, len)) 2197 return -EFAULT; 2198 uinsns += len; 2199 free -= len; 2200 if (!free) 2201 break; 2202 } 2203 } else { 2204 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 2205 return -EFAULT; 2206 } 2207 } else { 2208 info.jited_prog_insns = 0; 2209 } 2210 } 2211 2212 ulen = info.nr_jited_ksyms; 2213 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1; 2214 if (ulen) { 2215 if (bpf_dump_raw_ok()) { 2216 unsigned long ksym_addr; 2217 u64 __user *user_ksyms; 2218 u32 i; 2219 2220 /* copy the address of the kernel symbol 2221 * corresponding to each function 2222 */ 2223 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 2224 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 2225 if (prog->aux->func_cnt) { 2226 for (i = 0; i < ulen; i++) { 2227 ksym_addr = (unsigned long) 2228 prog->aux->func[i]->bpf_func; 2229 if (put_user((u64) ksym_addr, 2230 &user_ksyms[i])) 2231 return -EFAULT; 2232 } 2233 } else { 2234 ksym_addr = (unsigned long) prog->bpf_func; 2235 if (put_user((u64) ksym_addr, &user_ksyms[0])) 2236 return -EFAULT; 2237 } 2238 } else { 2239 info.jited_ksyms = 0; 2240 } 2241 } 2242 2243 ulen = info.nr_jited_func_lens; 2244 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1; 2245 if (ulen) { 2246 if (bpf_dump_raw_ok()) { 2247 u32 __user *user_lens; 2248 u32 func_len, i; 2249 2250 /* copy the JITed image lengths for each function */ 2251 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 2252 user_lens = u64_to_user_ptr(info.jited_func_lens); 2253 if (prog->aux->func_cnt) { 2254 for (i = 0; i < ulen; i++) { 2255 func_len = 2256 prog->aux->func[i]->jited_len; 2257 if (put_user(func_len, &user_lens[i])) 2258 return -EFAULT; 2259 } 2260 } else { 2261 func_len = prog->jited_len; 2262 if (put_user(func_len, &user_lens[0])) 2263 return -EFAULT; 2264 } 2265 } else { 2266 info.jited_func_lens = 0; 2267 } 2268 } 2269 2270 if (prog->aux->btf) 2271 info.btf_id = btf_id(prog->aux->btf); 2272 2273 ulen = info.nr_func_info; 2274 info.nr_func_info = prog->aux->func_info_cnt; 2275 if (info.nr_func_info && ulen) { 2276 char __user *user_finfo; 2277 2278 user_finfo = u64_to_user_ptr(info.func_info); 2279 ulen = min_t(u32, info.nr_func_info, ulen); 2280 if (copy_to_user(user_finfo, prog->aux->func_info, 2281 info.func_info_rec_size * ulen)) 2282 return -EFAULT; 2283 } 2284 2285 ulen = info.nr_line_info; 2286 info.nr_line_info = prog->aux->nr_linfo; 2287 if (info.nr_line_info && ulen) { 2288 __u8 __user *user_linfo; 2289 2290 user_linfo = u64_to_user_ptr(info.line_info); 2291 ulen = min_t(u32, info.nr_line_info, ulen); 2292 if (copy_to_user(user_linfo, prog->aux->linfo, 2293 info.line_info_rec_size * ulen)) 2294 return -EFAULT; 2295 } 2296 2297 ulen = info.nr_jited_line_info; 2298 if (prog->aux->jited_linfo) 2299 info.nr_jited_line_info = prog->aux->nr_linfo; 2300 else 2301 info.nr_jited_line_info = 0; 2302 if (info.nr_jited_line_info && ulen) { 2303 if (bpf_dump_raw_ok()) { 2304 __u64 __user *user_linfo; 2305 u32 i; 2306 2307 user_linfo = u64_to_user_ptr(info.jited_line_info); 2308 ulen = min_t(u32, info.nr_jited_line_info, ulen); 2309 for (i = 0; i < ulen; i++) { 2310 if (put_user((__u64)(long)prog->aux->jited_linfo[i], 2311 &user_linfo[i])) 2312 return -EFAULT; 2313 } 2314 } else { 2315 info.jited_line_info = 0; 2316 } 2317 } 2318 2319 ulen = info.nr_prog_tags; 2320 info.nr_prog_tags = prog->aux->func_cnt ? : 1; 2321 if (ulen) { 2322 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE]; 2323 u32 i; 2324 2325 user_prog_tags = u64_to_user_ptr(info.prog_tags); 2326 ulen = min_t(u32, info.nr_prog_tags, ulen); 2327 if (prog->aux->func_cnt) { 2328 for (i = 0; i < ulen; i++) { 2329 if (copy_to_user(user_prog_tags[i], 2330 prog->aux->func[i]->tag, 2331 BPF_TAG_SIZE)) 2332 return -EFAULT; 2333 } 2334 } else { 2335 if (copy_to_user(user_prog_tags[0], 2336 prog->tag, BPF_TAG_SIZE)) 2337 return -EFAULT; 2338 } 2339 } 2340 2341 done: 2342 if (copy_to_user(uinfo, &info, info_len) || 2343 put_user(info_len, &uattr->info.info_len)) 2344 return -EFAULT; 2345 2346 return 0; 2347 } 2348 2349 static int bpf_map_get_info_by_fd(struct bpf_map *map, 2350 const union bpf_attr *attr, 2351 union bpf_attr __user *uattr) 2352 { 2353 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2354 struct bpf_map_info info = {}; 2355 u32 info_len = attr->info.info_len; 2356 int err; 2357 2358 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2359 if (err) 2360 return err; 2361 info_len = min_t(u32, sizeof(info), info_len); 2362 2363 info.type = map->map_type; 2364 info.id = map->id; 2365 info.key_size = map->key_size; 2366 info.value_size = map->value_size; 2367 info.max_entries = map->max_entries; 2368 info.map_flags = map->map_flags; 2369 memcpy(info.name, map->name, sizeof(map->name)); 2370 2371 if (map->btf) { 2372 info.btf_id = btf_id(map->btf); 2373 info.btf_key_type_id = map->btf_key_type_id; 2374 info.btf_value_type_id = map->btf_value_type_id; 2375 } 2376 2377 if (bpf_map_is_dev_bound(map)) { 2378 err = bpf_map_offload_info_fill(&info, map); 2379 if (err) 2380 return err; 2381 } 2382 2383 if (copy_to_user(uinfo, &info, info_len) || 2384 put_user(info_len, &uattr->info.info_len)) 2385 return -EFAULT; 2386 2387 return 0; 2388 } 2389 2390 static int bpf_btf_get_info_by_fd(struct btf *btf, 2391 const union bpf_attr *attr, 2392 union bpf_attr __user *uattr) 2393 { 2394 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2395 u32 info_len = attr->info.info_len; 2396 int err; 2397 2398 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 2399 if (err) 2400 return err; 2401 2402 return btf_get_info_by_fd(btf, attr, uattr); 2403 } 2404 2405 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 2406 2407 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 2408 union bpf_attr __user *uattr) 2409 { 2410 int ufd = attr->info.bpf_fd; 2411 struct fd f; 2412 int err; 2413 2414 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 2415 return -EINVAL; 2416 2417 f = fdget(ufd); 2418 if (!f.file) 2419 return -EBADFD; 2420 2421 if (f.file->f_op == &bpf_prog_fops) 2422 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 2423 uattr); 2424 else if (f.file->f_op == &bpf_map_fops) 2425 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 2426 uattr); 2427 else if (f.file->f_op == &btf_fops) 2428 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 2429 else 2430 err = -EINVAL; 2431 2432 fdput(f); 2433 return err; 2434 } 2435 2436 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 2437 2438 static int bpf_btf_load(const union bpf_attr *attr) 2439 { 2440 if (CHECK_ATTR(BPF_BTF_LOAD)) 2441 return -EINVAL; 2442 2443 if (!capable(CAP_SYS_ADMIN)) 2444 return -EPERM; 2445 2446 return btf_new_fd(attr); 2447 } 2448 2449 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 2450 2451 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 2452 { 2453 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 2454 return -EINVAL; 2455 2456 if (!capable(CAP_SYS_ADMIN)) 2457 return -EPERM; 2458 2459 return btf_get_fd_by_id(attr->btf_id); 2460 } 2461 2462 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 2463 union bpf_attr __user *uattr, 2464 u32 prog_id, u32 fd_type, 2465 const char *buf, u64 probe_offset, 2466 u64 probe_addr) 2467 { 2468 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 2469 u32 len = buf ? strlen(buf) : 0, input_len; 2470 int err = 0; 2471 2472 if (put_user(len, &uattr->task_fd_query.buf_len)) 2473 return -EFAULT; 2474 input_len = attr->task_fd_query.buf_len; 2475 if (input_len && ubuf) { 2476 if (!len) { 2477 /* nothing to copy, just make ubuf NULL terminated */ 2478 char zero = '\0'; 2479 2480 if (put_user(zero, ubuf)) 2481 return -EFAULT; 2482 } else if (input_len >= len + 1) { 2483 /* ubuf can hold the string with NULL terminator */ 2484 if (copy_to_user(ubuf, buf, len + 1)) 2485 return -EFAULT; 2486 } else { 2487 /* ubuf cannot hold the string with NULL terminator, 2488 * do a partial copy with NULL terminator. 2489 */ 2490 char zero = '\0'; 2491 2492 err = -ENOSPC; 2493 if (copy_to_user(ubuf, buf, input_len - 1)) 2494 return -EFAULT; 2495 if (put_user(zero, ubuf + input_len - 1)) 2496 return -EFAULT; 2497 } 2498 } 2499 2500 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 2501 put_user(fd_type, &uattr->task_fd_query.fd_type) || 2502 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 2503 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 2504 return -EFAULT; 2505 2506 return err; 2507 } 2508 2509 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 2510 2511 static int bpf_task_fd_query(const union bpf_attr *attr, 2512 union bpf_attr __user *uattr) 2513 { 2514 pid_t pid = attr->task_fd_query.pid; 2515 u32 fd = attr->task_fd_query.fd; 2516 const struct perf_event *event; 2517 struct files_struct *files; 2518 struct task_struct *task; 2519 struct file *file; 2520 int err; 2521 2522 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 2523 return -EINVAL; 2524 2525 if (!capable(CAP_SYS_ADMIN)) 2526 return -EPERM; 2527 2528 if (attr->task_fd_query.flags != 0) 2529 return -EINVAL; 2530 2531 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 2532 if (!task) 2533 return -ENOENT; 2534 2535 files = get_files_struct(task); 2536 put_task_struct(task); 2537 if (!files) 2538 return -ENOENT; 2539 2540 err = 0; 2541 spin_lock(&files->file_lock); 2542 file = fcheck_files(files, fd); 2543 if (!file) 2544 err = -EBADF; 2545 else 2546 get_file(file); 2547 spin_unlock(&files->file_lock); 2548 put_files_struct(files); 2549 2550 if (err) 2551 goto out; 2552 2553 if (file->f_op == &bpf_raw_tp_fops) { 2554 struct bpf_raw_tracepoint *raw_tp = file->private_data; 2555 struct bpf_raw_event_map *btp = raw_tp->btp; 2556 2557 err = bpf_task_fd_query_copy(attr, uattr, 2558 raw_tp->prog->aux->id, 2559 BPF_FD_TYPE_RAW_TRACEPOINT, 2560 btp->tp->name, 0, 0); 2561 goto put_file; 2562 } 2563 2564 event = perf_get_event(file); 2565 if (!IS_ERR(event)) { 2566 u64 probe_offset, probe_addr; 2567 u32 prog_id, fd_type; 2568 const char *buf; 2569 2570 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 2571 &buf, &probe_offset, 2572 &probe_addr); 2573 if (!err) 2574 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 2575 fd_type, buf, 2576 probe_offset, 2577 probe_addr); 2578 goto put_file; 2579 } 2580 2581 err = -ENOTSUPP; 2582 put_file: 2583 fput(file); 2584 out: 2585 return err; 2586 } 2587 2588 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 2589 { 2590 union bpf_attr attr = {}; 2591 int err; 2592 2593 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 2594 return -EPERM; 2595 2596 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 2597 if (err) 2598 return err; 2599 size = min_t(u32, size, sizeof(attr)); 2600 2601 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 2602 if (copy_from_user(&attr, uattr, size) != 0) 2603 return -EFAULT; 2604 2605 err = security_bpf(cmd, &attr, size); 2606 if (err < 0) 2607 return err; 2608 2609 switch (cmd) { 2610 case BPF_MAP_CREATE: 2611 err = map_create(&attr); 2612 break; 2613 case BPF_MAP_LOOKUP_ELEM: 2614 err = map_lookup_elem(&attr); 2615 break; 2616 case BPF_MAP_UPDATE_ELEM: 2617 err = map_update_elem(&attr); 2618 break; 2619 case BPF_MAP_DELETE_ELEM: 2620 err = map_delete_elem(&attr); 2621 break; 2622 case BPF_MAP_GET_NEXT_KEY: 2623 err = map_get_next_key(&attr); 2624 break; 2625 case BPF_PROG_LOAD: 2626 err = bpf_prog_load(&attr, uattr); 2627 break; 2628 case BPF_OBJ_PIN: 2629 err = bpf_obj_pin(&attr); 2630 break; 2631 case BPF_OBJ_GET: 2632 err = bpf_obj_get(&attr); 2633 break; 2634 case BPF_PROG_ATTACH: 2635 err = bpf_prog_attach(&attr); 2636 break; 2637 case BPF_PROG_DETACH: 2638 err = bpf_prog_detach(&attr); 2639 break; 2640 case BPF_PROG_QUERY: 2641 err = bpf_prog_query(&attr, uattr); 2642 break; 2643 case BPF_PROG_TEST_RUN: 2644 err = bpf_prog_test_run(&attr, uattr); 2645 break; 2646 case BPF_PROG_GET_NEXT_ID: 2647 err = bpf_obj_get_next_id(&attr, uattr, 2648 &prog_idr, &prog_idr_lock); 2649 break; 2650 case BPF_MAP_GET_NEXT_ID: 2651 err = bpf_obj_get_next_id(&attr, uattr, 2652 &map_idr, &map_idr_lock); 2653 break; 2654 case BPF_PROG_GET_FD_BY_ID: 2655 err = bpf_prog_get_fd_by_id(&attr); 2656 break; 2657 case BPF_MAP_GET_FD_BY_ID: 2658 err = bpf_map_get_fd_by_id(&attr); 2659 break; 2660 case BPF_OBJ_GET_INFO_BY_FD: 2661 err = bpf_obj_get_info_by_fd(&attr, uattr); 2662 break; 2663 case BPF_RAW_TRACEPOINT_OPEN: 2664 err = bpf_raw_tracepoint_open(&attr); 2665 break; 2666 case BPF_BTF_LOAD: 2667 err = bpf_btf_load(&attr); 2668 break; 2669 case BPF_BTF_GET_FD_BY_ID: 2670 err = bpf_btf_get_fd_by_id(&attr); 2671 break; 2672 case BPF_TASK_FD_QUERY: 2673 err = bpf_task_fd_query(&attr, uattr); 2674 break; 2675 case BPF_MAP_LOOKUP_AND_DELETE_ELEM: 2676 err = map_lookup_and_delete_elem(&attr); 2677 break; 2678 default: 2679 err = -EINVAL; 2680 break; 2681 } 2682 2683 return err; 2684 } 2685