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 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0); 1223 /* bpf_prog_free_id() must be called first */ 1224 bpf_prog_free_id(prog, do_idr_lock); 1225 bpf_prog_kallsyms_del_all(prog); 1226 btf_put(prog->aux->btf); 1227 kvfree(prog->aux->func_info); 1228 bpf_prog_free_linfo(prog); 1229 1230 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1231 } 1232 } 1233 1234 void bpf_prog_put(struct bpf_prog *prog) 1235 { 1236 __bpf_prog_put(prog, true); 1237 } 1238 EXPORT_SYMBOL_GPL(bpf_prog_put); 1239 1240 static int bpf_prog_release(struct inode *inode, struct file *filp) 1241 { 1242 struct bpf_prog *prog = filp->private_data; 1243 1244 bpf_prog_put(prog); 1245 return 0; 1246 } 1247 1248 #ifdef CONFIG_PROC_FS 1249 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1250 { 1251 const struct bpf_prog *prog = filp->private_data; 1252 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1253 1254 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1255 seq_printf(m, 1256 "prog_type:\t%u\n" 1257 "prog_jited:\t%u\n" 1258 "prog_tag:\t%s\n" 1259 "memlock:\t%llu\n" 1260 "prog_id:\t%u\n", 1261 prog->type, 1262 prog->jited, 1263 prog_tag, 1264 prog->pages * 1ULL << PAGE_SHIFT, 1265 prog->aux->id); 1266 } 1267 #endif 1268 1269 const struct file_operations bpf_prog_fops = { 1270 #ifdef CONFIG_PROC_FS 1271 .show_fdinfo = bpf_prog_show_fdinfo, 1272 #endif 1273 .release = bpf_prog_release, 1274 .read = bpf_dummy_read, 1275 .write = bpf_dummy_write, 1276 }; 1277 1278 int bpf_prog_new_fd(struct bpf_prog *prog) 1279 { 1280 int ret; 1281 1282 ret = security_bpf_prog(prog); 1283 if (ret < 0) 1284 return ret; 1285 1286 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1287 O_RDWR | O_CLOEXEC); 1288 } 1289 1290 static struct bpf_prog *____bpf_prog_get(struct fd f) 1291 { 1292 if (!f.file) 1293 return ERR_PTR(-EBADF); 1294 if (f.file->f_op != &bpf_prog_fops) { 1295 fdput(f); 1296 return ERR_PTR(-EINVAL); 1297 } 1298 1299 return f.file->private_data; 1300 } 1301 1302 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) 1303 { 1304 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { 1305 atomic_sub(i, &prog->aux->refcnt); 1306 return ERR_PTR(-EBUSY); 1307 } 1308 return prog; 1309 } 1310 EXPORT_SYMBOL_GPL(bpf_prog_add); 1311 1312 void bpf_prog_sub(struct bpf_prog *prog, int i) 1313 { 1314 /* Only to be used for undoing previous bpf_prog_add() in some 1315 * error path. We still know that another entity in our call 1316 * path holds a reference to the program, thus atomic_sub() can 1317 * be safely used in such cases! 1318 */ 1319 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); 1320 } 1321 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1322 1323 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) 1324 { 1325 return bpf_prog_add(prog, 1); 1326 } 1327 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1328 1329 /* prog_idr_lock should have been held */ 1330 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1331 { 1332 int refold; 1333 1334 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1335 1336 if (refold >= BPF_MAX_REFCNT) { 1337 __bpf_prog_put(prog, false); 1338 return ERR_PTR(-EBUSY); 1339 } 1340 1341 if (!refold) 1342 return ERR_PTR(-ENOENT); 1343 1344 return prog; 1345 } 1346 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1347 1348 bool bpf_prog_get_ok(struct bpf_prog *prog, 1349 enum bpf_prog_type *attach_type, bool attach_drv) 1350 { 1351 /* not an attachment, just a refcount inc, always allow */ 1352 if (!attach_type) 1353 return true; 1354 1355 if (prog->type != *attach_type) 1356 return false; 1357 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1358 return false; 1359 1360 return true; 1361 } 1362 1363 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1364 bool attach_drv) 1365 { 1366 struct fd f = fdget(ufd); 1367 struct bpf_prog *prog; 1368 1369 prog = ____bpf_prog_get(f); 1370 if (IS_ERR(prog)) 1371 return prog; 1372 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1373 prog = ERR_PTR(-EINVAL); 1374 goto out; 1375 } 1376 1377 prog = bpf_prog_inc(prog); 1378 out: 1379 fdput(f); 1380 return prog; 1381 } 1382 1383 struct bpf_prog *bpf_prog_get(u32 ufd) 1384 { 1385 return __bpf_prog_get(ufd, NULL, false); 1386 } 1387 1388 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1389 bool attach_drv) 1390 { 1391 return __bpf_prog_get(ufd, &type, attach_drv); 1392 } 1393 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1394 1395 /* Initially all BPF programs could be loaded w/o specifying 1396 * expected_attach_type. Later for some of them specifying expected_attach_type 1397 * at load time became required so that program could be validated properly. 1398 * Programs of types that are allowed to be loaded both w/ and w/o (for 1399 * backward compatibility) expected_attach_type, should have the default attach 1400 * type assigned to expected_attach_type for the latter case, so that it can be 1401 * validated later at attach time. 1402 * 1403 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1404 * prog type requires it but has some attach types that have to be backward 1405 * compatible. 1406 */ 1407 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1408 { 1409 switch (attr->prog_type) { 1410 case BPF_PROG_TYPE_CGROUP_SOCK: 1411 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1412 * exist so checking for non-zero is the way to go here. 1413 */ 1414 if (!attr->expected_attach_type) 1415 attr->expected_attach_type = 1416 BPF_CGROUP_INET_SOCK_CREATE; 1417 break; 1418 } 1419 } 1420 1421 static int 1422 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type, 1423 enum bpf_attach_type expected_attach_type) 1424 { 1425 switch (prog_type) { 1426 case BPF_PROG_TYPE_CGROUP_SOCK: 1427 switch (expected_attach_type) { 1428 case BPF_CGROUP_INET_SOCK_CREATE: 1429 case BPF_CGROUP_INET4_POST_BIND: 1430 case BPF_CGROUP_INET6_POST_BIND: 1431 return 0; 1432 default: 1433 return -EINVAL; 1434 } 1435 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1436 switch (expected_attach_type) { 1437 case BPF_CGROUP_INET4_BIND: 1438 case BPF_CGROUP_INET6_BIND: 1439 case BPF_CGROUP_INET4_CONNECT: 1440 case BPF_CGROUP_INET6_CONNECT: 1441 case BPF_CGROUP_UDP4_SENDMSG: 1442 case BPF_CGROUP_UDP6_SENDMSG: 1443 return 0; 1444 default: 1445 return -EINVAL; 1446 } 1447 default: 1448 return 0; 1449 } 1450 } 1451 1452 /* last field in 'union bpf_attr' used by this command */ 1453 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt 1454 1455 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr) 1456 { 1457 enum bpf_prog_type type = attr->prog_type; 1458 struct bpf_prog *prog; 1459 int err; 1460 char license[128]; 1461 bool is_gpl; 1462 1463 if (CHECK_ATTR(BPF_PROG_LOAD)) 1464 return -EINVAL; 1465 1466 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT)) 1467 return -EINVAL; 1468 1469 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && 1470 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) && 1471 !capable(CAP_SYS_ADMIN)) 1472 return -EPERM; 1473 1474 /* copy eBPF program license from user space */ 1475 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 1476 sizeof(license) - 1) < 0) 1477 return -EFAULT; 1478 license[sizeof(license) - 1] = 0; 1479 1480 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1481 is_gpl = license_is_gpl_compatible(license); 1482 1483 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) 1484 return -E2BIG; 1485 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 1486 type != BPF_PROG_TYPE_CGROUP_SKB && 1487 !capable(CAP_SYS_ADMIN)) 1488 return -EPERM; 1489 1490 bpf_prog_load_fixup_attach_type(attr); 1491 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type)) 1492 return -EINVAL; 1493 1494 /* plain bpf_prog allocation */ 1495 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 1496 if (!prog) 1497 return -ENOMEM; 1498 1499 prog->expected_attach_type = attr->expected_attach_type; 1500 1501 prog->aux->offload_requested = !!attr->prog_ifindex; 1502 1503 err = security_bpf_prog_alloc(prog->aux); 1504 if (err) 1505 goto free_prog_nouncharge; 1506 1507 err = bpf_prog_charge_memlock(prog); 1508 if (err) 1509 goto free_prog_sec; 1510 1511 prog->len = attr->insn_cnt; 1512 1513 err = -EFAULT; 1514 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 1515 bpf_prog_insn_size(prog)) != 0) 1516 goto free_prog; 1517 1518 prog->orig_prog = NULL; 1519 prog->jited = 0; 1520 1521 atomic_set(&prog->aux->refcnt, 1); 1522 prog->gpl_compatible = is_gpl ? 1 : 0; 1523 1524 if (bpf_prog_is_dev_bound(prog->aux)) { 1525 err = bpf_prog_offload_init(prog, attr); 1526 if (err) 1527 goto free_prog; 1528 } 1529 1530 /* find program type: socket_filter vs tracing_filter */ 1531 err = find_prog_type(type, prog); 1532 if (err < 0) 1533 goto free_prog; 1534 1535 prog->aux->load_time = ktime_get_boot_ns(); 1536 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 1537 if (err) 1538 goto free_prog; 1539 1540 /* run eBPF verifier */ 1541 err = bpf_check(&prog, attr, uattr); 1542 if (err < 0) 1543 goto free_used_maps; 1544 1545 prog = bpf_prog_select_runtime(prog, &err); 1546 if (err < 0) 1547 goto free_used_maps; 1548 1549 err = bpf_prog_alloc_id(prog); 1550 if (err) 1551 goto free_used_maps; 1552 1553 err = bpf_prog_new_fd(prog); 1554 if (err < 0) { 1555 /* failed to allocate fd. 1556 * bpf_prog_put() is needed because the above 1557 * bpf_prog_alloc_id() has published the prog 1558 * to the userspace and the userspace may 1559 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. 1560 */ 1561 bpf_prog_put(prog); 1562 return err; 1563 } 1564 1565 bpf_prog_kallsyms_add(prog); 1566 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0); 1567 return err; 1568 1569 free_used_maps: 1570 bpf_prog_free_linfo(prog); 1571 kvfree(prog->aux->func_info); 1572 btf_put(prog->aux->btf); 1573 bpf_prog_kallsyms_del_subprogs(prog); 1574 free_used_maps(prog->aux); 1575 free_prog: 1576 bpf_prog_uncharge_memlock(prog); 1577 free_prog_sec: 1578 security_bpf_prog_free(prog->aux); 1579 free_prog_nouncharge: 1580 bpf_prog_free(prog); 1581 return err; 1582 } 1583 1584 #define BPF_OBJ_LAST_FIELD file_flags 1585 1586 static int bpf_obj_pin(const union bpf_attr *attr) 1587 { 1588 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 1589 return -EINVAL; 1590 1591 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 1592 } 1593 1594 static int bpf_obj_get(const union bpf_attr *attr) 1595 { 1596 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 1597 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 1598 return -EINVAL; 1599 1600 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 1601 attr->file_flags); 1602 } 1603 1604 struct bpf_raw_tracepoint { 1605 struct bpf_raw_event_map *btp; 1606 struct bpf_prog *prog; 1607 }; 1608 1609 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp) 1610 { 1611 struct bpf_raw_tracepoint *raw_tp = filp->private_data; 1612 1613 if (raw_tp->prog) { 1614 bpf_probe_unregister(raw_tp->btp, raw_tp->prog); 1615 bpf_prog_put(raw_tp->prog); 1616 } 1617 bpf_put_raw_tracepoint(raw_tp->btp); 1618 kfree(raw_tp); 1619 return 0; 1620 } 1621 1622 static const struct file_operations bpf_raw_tp_fops = { 1623 .release = bpf_raw_tracepoint_release, 1624 .read = bpf_dummy_read, 1625 .write = bpf_dummy_write, 1626 }; 1627 1628 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 1629 1630 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 1631 { 1632 struct bpf_raw_tracepoint *raw_tp; 1633 struct bpf_raw_event_map *btp; 1634 struct bpf_prog *prog; 1635 char tp_name[128]; 1636 int tp_fd, err; 1637 1638 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name), 1639 sizeof(tp_name) - 1) < 0) 1640 return -EFAULT; 1641 tp_name[sizeof(tp_name) - 1] = 0; 1642 1643 btp = bpf_get_raw_tracepoint(tp_name); 1644 if (!btp) 1645 return -ENOENT; 1646 1647 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER); 1648 if (!raw_tp) { 1649 err = -ENOMEM; 1650 goto out_put_btp; 1651 } 1652 raw_tp->btp = btp; 1653 1654 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd, 1655 BPF_PROG_TYPE_RAW_TRACEPOINT); 1656 if (IS_ERR(prog)) { 1657 err = PTR_ERR(prog); 1658 goto out_free_tp; 1659 } 1660 1661 err = bpf_probe_register(raw_tp->btp, prog); 1662 if (err) 1663 goto out_put_prog; 1664 1665 raw_tp->prog = prog; 1666 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp, 1667 O_CLOEXEC); 1668 if (tp_fd < 0) { 1669 bpf_probe_unregister(raw_tp->btp, prog); 1670 err = tp_fd; 1671 goto out_put_prog; 1672 } 1673 return tp_fd; 1674 1675 out_put_prog: 1676 bpf_prog_put(prog); 1677 out_free_tp: 1678 kfree(raw_tp); 1679 out_put_btp: 1680 bpf_put_raw_tracepoint(btp); 1681 return err; 1682 } 1683 1684 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 1685 enum bpf_attach_type attach_type) 1686 { 1687 switch (prog->type) { 1688 case BPF_PROG_TYPE_CGROUP_SOCK: 1689 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1690 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 1691 default: 1692 return 0; 1693 } 1694 } 1695 1696 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags 1697 1698 #define BPF_F_ATTACH_MASK \ 1699 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI) 1700 1701 static int bpf_prog_attach(const union bpf_attr *attr) 1702 { 1703 enum bpf_prog_type ptype; 1704 struct bpf_prog *prog; 1705 int ret; 1706 1707 if (!capable(CAP_NET_ADMIN)) 1708 return -EPERM; 1709 1710 if (CHECK_ATTR(BPF_PROG_ATTACH)) 1711 return -EINVAL; 1712 1713 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 1714 return -EINVAL; 1715 1716 switch (attr->attach_type) { 1717 case BPF_CGROUP_INET_INGRESS: 1718 case BPF_CGROUP_INET_EGRESS: 1719 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1720 break; 1721 case BPF_CGROUP_INET_SOCK_CREATE: 1722 case BPF_CGROUP_INET4_POST_BIND: 1723 case BPF_CGROUP_INET6_POST_BIND: 1724 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1725 break; 1726 case BPF_CGROUP_INET4_BIND: 1727 case BPF_CGROUP_INET6_BIND: 1728 case BPF_CGROUP_INET4_CONNECT: 1729 case BPF_CGROUP_INET6_CONNECT: 1730 case BPF_CGROUP_UDP4_SENDMSG: 1731 case BPF_CGROUP_UDP6_SENDMSG: 1732 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1733 break; 1734 case BPF_CGROUP_SOCK_OPS: 1735 ptype = BPF_PROG_TYPE_SOCK_OPS; 1736 break; 1737 case BPF_CGROUP_DEVICE: 1738 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1739 break; 1740 case BPF_SK_MSG_VERDICT: 1741 ptype = BPF_PROG_TYPE_SK_MSG; 1742 break; 1743 case BPF_SK_SKB_STREAM_PARSER: 1744 case BPF_SK_SKB_STREAM_VERDICT: 1745 ptype = BPF_PROG_TYPE_SK_SKB; 1746 break; 1747 case BPF_LIRC_MODE2: 1748 ptype = BPF_PROG_TYPE_LIRC_MODE2; 1749 break; 1750 case BPF_FLOW_DISSECTOR: 1751 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 1752 break; 1753 default: 1754 return -EINVAL; 1755 } 1756 1757 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1758 if (IS_ERR(prog)) 1759 return PTR_ERR(prog); 1760 1761 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 1762 bpf_prog_put(prog); 1763 return -EINVAL; 1764 } 1765 1766 switch (ptype) { 1767 case BPF_PROG_TYPE_SK_SKB: 1768 case BPF_PROG_TYPE_SK_MSG: 1769 ret = sock_map_get_from_fd(attr, prog); 1770 break; 1771 case BPF_PROG_TYPE_LIRC_MODE2: 1772 ret = lirc_prog_attach(attr, prog); 1773 break; 1774 case BPF_PROG_TYPE_FLOW_DISSECTOR: 1775 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 1776 break; 1777 default: 1778 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 1779 } 1780 1781 if (ret) 1782 bpf_prog_put(prog); 1783 return ret; 1784 } 1785 1786 #define BPF_PROG_DETACH_LAST_FIELD attach_type 1787 1788 static int bpf_prog_detach(const union bpf_attr *attr) 1789 { 1790 enum bpf_prog_type ptype; 1791 1792 if (!capable(CAP_NET_ADMIN)) 1793 return -EPERM; 1794 1795 if (CHECK_ATTR(BPF_PROG_DETACH)) 1796 return -EINVAL; 1797 1798 switch (attr->attach_type) { 1799 case BPF_CGROUP_INET_INGRESS: 1800 case BPF_CGROUP_INET_EGRESS: 1801 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1802 break; 1803 case BPF_CGROUP_INET_SOCK_CREATE: 1804 case BPF_CGROUP_INET4_POST_BIND: 1805 case BPF_CGROUP_INET6_POST_BIND: 1806 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1807 break; 1808 case BPF_CGROUP_INET4_BIND: 1809 case BPF_CGROUP_INET6_BIND: 1810 case BPF_CGROUP_INET4_CONNECT: 1811 case BPF_CGROUP_INET6_CONNECT: 1812 case BPF_CGROUP_UDP4_SENDMSG: 1813 case BPF_CGROUP_UDP6_SENDMSG: 1814 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1815 break; 1816 case BPF_CGROUP_SOCK_OPS: 1817 ptype = BPF_PROG_TYPE_SOCK_OPS; 1818 break; 1819 case BPF_CGROUP_DEVICE: 1820 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1821 break; 1822 case BPF_SK_MSG_VERDICT: 1823 return sock_map_get_from_fd(attr, NULL); 1824 case BPF_SK_SKB_STREAM_PARSER: 1825 case BPF_SK_SKB_STREAM_VERDICT: 1826 return sock_map_get_from_fd(attr, NULL); 1827 case BPF_LIRC_MODE2: 1828 return lirc_prog_detach(attr); 1829 case BPF_FLOW_DISSECTOR: 1830 return skb_flow_dissector_bpf_prog_detach(attr); 1831 default: 1832 return -EINVAL; 1833 } 1834 1835 return cgroup_bpf_prog_detach(attr, ptype); 1836 } 1837 1838 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 1839 1840 static int bpf_prog_query(const union bpf_attr *attr, 1841 union bpf_attr __user *uattr) 1842 { 1843 if (!capable(CAP_NET_ADMIN)) 1844 return -EPERM; 1845 if (CHECK_ATTR(BPF_PROG_QUERY)) 1846 return -EINVAL; 1847 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 1848 return -EINVAL; 1849 1850 switch (attr->query.attach_type) { 1851 case BPF_CGROUP_INET_INGRESS: 1852 case BPF_CGROUP_INET_EGRESS: 1853 case BPF_CGROUP_INET_SOCK_CREATE: 1854 case BPF_CGROUP_INET4_BIND: 1855 case BPF_CGROUP_INET6_BIND: 1856 case BPF_CGROUP_INET4_POST_BIND: 1857 case BPF_CGROUP_INET6_POST_BIND: 1858 case BPF_CGROUP_INET4_CONNECT: 1859 case BPF_CGROUP_INET6_CONNECT: 1860 case BPF_CGROUP_UDP4_SENDMSG: 1861 case BPF_CGROUP_UDP6_SENDMSG: 1862 case BPF_CGROUP_SOCK_OPS: 1863 case BPF_CGROUP_DEVICE: 1864 break; 1865 case BPF_LIRC_MODE2: 1866 return lirc_prog_query(attr, uattr); 1867 default: 1868 return -EINVAL; 1869 } 1870 1871 return cgroup_bpf_prog_query(attr, uattr); 1872 } 1873 1874 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration 1875 1876 static int bpf_prog_test_run(const union bpf_attr *attr, 1877 union bpf_attr __user *uattr) 1878 { 1879 struct bpf_prog *prog; 1880 int ret = -ENOTSUPP; 1881 1882 if (!capable(CAP_SYS_ADMIN)) 1883 return -EPERM; 1884 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 1885 return -EINVAL; 1886 1887 prog = bpf_prog_get(attr->test.prog_fd); 1888 if (IS_ERR(prog)) 1889 return PTR_ERR(prog); 1890 1891 if (prog->aux->ops->test_run) 1892 ret = prog->aux->ops->test_run(prog, attr, uattr); 1893 1894 bpf_prog_put(prog); 1895 return ret; 1896 } 1897 1898 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 1899 1900 static int bpf_obj_get_next_id(const union bpf_attr *attr, 1901 union bpf_attr __user *uattr, 1902 struct idr *idr, 1903 spinlock_t *lock) 1904 { 1905 u32 next_id = attr->start_id; 1906 int err = 0; 1907 1908 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 1909 return -EINVAL; 1910 1911 if (!capable(CAP_SYS_ADMIN)) 1912 return -EPERM; 1913 1914 next_id++; 1915 spin_lock_bh(lock); 1916 if (!idr_get_next(idr, &next_id)) 1917 err = -ENOENT; 1918 spin_unlock_bh(lock); 1919 1920 if (!err) 1921 err = put_user(next_id, &uattr->next_id); 1922 1923 return err; 1924 } 1925 1926 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 1927 1928 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 1929 { 1930 struct bpf_prog *prog; 1931 u32 id = attr->prog_id; 1932 int fd; 1933 1934 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 1935 return -EINVAL; 1936 1937 if (!capable(CAP_SYS_ADMIN)) 1938 return -EPERM; 1939 1940 spin_lock_bh(&prog_idr_lock); 1941 prog = idr_find(&prog_idr, id); 1942 if (prog) 1943 prog = bpf_prog_inc_not_zero(prog); 1944 else 1945 prog = ERR_PTR(-ENOENT); 1946 spin_unlock_bh(&prog_idr_lock); 1947 1948 if (IS_ERR(prog)) 1949 return PTR_ERR(prog); 1950 1951 fd = bpf_prog_new_fd(prog); 1952 if (fd < 0) 1953 bpf_prog_put(prog); 1954 1955 return fd; 1956 } 1957 1958 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 1959 1960 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 1961 { 1962 struct bpf_map *map; 1963 u32 id = attr->map_id; 1964 int f_flags; 1965 int fd; 1966 1967 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 1968 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 1969 return -EINVAL; 1970 1971 if (!capable(CAP_SYS_ADMIN)) 1972 return -EPERM; 1973 1974 f_flags = bpf_get_file_flag(attr->open_flags); 1975 if (f_flags < 0) 1976 return f_flags; 1977 1978 spin_lock_bh(&map_idr_lock); 1979 map = idr_find(&map_idr, id); 1980 if (map) 1981 map = bpf_map_inc_not_zero(map, true); 1982 else 1983 map = ERR_PTR(-ENOENT); 1984 spin_unlock_bh(&map_idr_lock); 1985 1986 if (IS_ERR(map)) 1987 return PTR_ERR(map); 1988 1989 fd = bpf_map_new_fd(map, f_flags); 1990 if (fd < 0) 1991 bpf_map_put(map); 1992 1993 return fd; 1994 } 1995 1996 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 1997 unsigned long addr) 1998 { 1999 int i; 2000 2001 for (i = 0; i < prog->aux->used_map_cnt; i++) 2002 if (prog->aux->used_maps[i] == (void *)addr) 2003 return prog->aux->used_maps[i]; 2004 return NULL; 2005 } 2006 2007 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 2008 { 2009 const struct bpf_map *map; 2010 struct bpf_insn *insns; 2011 u64 imm; 2012 int i; 2013 2014 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 2015 GFP_USER); 2016 if (!insns) 2017 return insns; 2018 2019 for (i = 0; i < prog->len; i++) { 2020 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 2021 insns[i].code = BPF_JMP | BPF_CALL; 2022 insns[i].imm = BPF_FUNC_tail_call; 2023 /* fall-through */ 2024 } 2025 if (insns[i].code == (BPF_JMP | BPF_CALL) || 2026 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 2027 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 2028 insns[i].code = BPF_JMP | BPF_CALL; 2029 if (!bpf_dump_raw_ok()) 2030 insns[i].imm = 0; 2031 continue; 2032 } 2033 2034 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 2035 continue; 2036 2037 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 2038 map = bpf_map_from_imm(prog, imm); 2039 if (map) { 2040 insns[i].src_reg = BPF_PSEUDO_MAP_FD; 2041 insns[i].imm = map->id; 2042 insns[i + 1].imm = 0; 2043 continue; 2044 } 2045 } 2046 2047 return insns; 2048 } 2049 2050 static int set_info_rec_size(struct bpf_prog_info *info) 2051 { 2052 /* 2053 * Ensure info.*_rec_size is the same as kernel expected size 2054 * 2055 * or 2056 * 2057 * Only allow zero *_rec_size if both _rec_size and _cnt are 2058 * zero. In this case, the kernel will set the expected 2059 * _rec_size back to the info. 2060 */ 2061 2062 if ((info->nr_func_info || info->func_info_rec_size) && 2063 info->func_info_rec_size != sizeof(struct bpf_func_info)) 2064 return -EINVAL; 2065 2066 if ((info->nr_line_info || info->line_info_rec_size) && 2067 info->line_info_rec_size != sizeof(struct bpf_line_info)) 2068 return -EINVAL; 2069 2070 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) && 2071 info->jited_line_info_rec_size != sizeof(__u64)) 2072 return -EINVAL; 2073 2074 info->func_info_rec_size = sizeof(struct bpf_func_info); 2075 info->line_info_rec_size = sizeof(struct bpf_line_info); 2076 info->jited_line_info_rec_size = sizeof(__u64); 2077 2078 return 0; 2079 } 2080 2081 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 2082 const union bpf_attr *attr, 2083 union bpf_attr __user *uattr) 2084 { 2085 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2086 struct bpf_prog_info info = {}; 2087 u32 info_len = attr->info.info_len; 2088 char __user *uinsns; 2089 u32 ulen; 2090 int err; 2091 2092 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2093 if (err) 2094 return err; 2095 info_len = min_t(u32, sizeof(info), info_len); 2096 2097 if (copy_from_user(&info, uinfo, info_len)) 2098 return -EFAULT; 2099 2100 info.type = prog->type; 2101 info.id = prog->aux->id; 2102 info.load_time = prog->aux->load_time; 2103 info.created_by_uid = from_kuid_munged(current_user_ns(), 2104 prog->aux->user->uid); 2105 info.gpl_compatible = prog->gpl_compatible; 2106 2107 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 2108 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 2109 2110 ulen = info.nr_map_ids; 2111 info.nr_map_ids = prog->aux->used_map_cnt; 2112 ulen = min_t(u32, info.nr_map_ids, ulen); 2113 if (ulen) { 2114 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 2115 u32 i; 2116 2117 for (i = 0; i < ulen; i++) 2118 if (put_user(prog->aux->used_maps[i]->id, 2119 &user_map_ids[i])) 2120 return -EFAULT; 2121 } 2122 2123 err = set_info_rec_size(&info); 2124 if (err) 2125 return err; 2126 2127 if (!capable(CAP_SYS_ADMIN)) { 2128 info.jited_prog_len = 0; 2129 info.xlated_prog_len = 0; 2130 info.nr_jited_ksyms = 0; 2131 info.nr_jited_func_lens = 0; 2132 info.nr_func_info = 0; 2133 info.nr_line_info = 0; 2134 info.nr_jited_line_info = 0; 2135 goto done; 2136 } 2137 2138 ulen = info.xlated_prog_len; 2139 info.xlated_prog_len = bpf_prog_insn_size(prog); 2140 if (info.xlated_prog_len && ulen) { 2141 struct bpf_insn *insns_sanitized; 2142 bool fault; 2143 2144 if (prog->blinded && !bpf_dump_raw_ok()) { 2145 info.xlated_prog_insns = 0; 2146 goto done; 2147 } 2148 insns_sanitized = bpf_insn_prepare_dump(prog); 2149 if (!insns_sanitized) 2150 return -ENOMEM; 2151 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 2152 ulen = min_t(u32, info.xlated_prog_len, ulen); 2153 fault = copy_to_user(uinsns, insns_sanitized, ulen); 2154 kfree(insns_sanitized); 2155 if (fault) 2156 return -EFAULT; 2157 } 2158 2159 if (bpf_prog_is_dev_bound(prog->aux)) { 2160 err = bpf_prog_offload_info_fill(&info, prog); 2161 if (err) 2162 return err; 2163 goto done; 2164 } 2165 2166 /* NOTE: the following code is supposed to be skipped for offload. 2167 * bpf_prog_offload_info_fill() is the place to fill similar fields 2168 * for offload. 2169 */ 2170 ulen = info.jited_prog_len; 2171 if (prog->aux->func_cnt) { 2172 u32 i; 2173 2174 info.jited_prog_len = 0; 2175 for (i = 0; i < prog->aux->func_cnt; i++) 2176 info.jited_prog_len += prog->aux->func[i]->jited_len; 2177 } else { 2178 info.jited_prog_len = prog->jited_len; 2179 } 2180 2181 if (info.jited_prog_len && ulen) { 2182 if (bpf_dump_raw_ok()) { 2183 uinsns = u64_to_user_ptr(info.jited_prog_insns); 2184 ulen = min_t(u32, info.jited_prog_len, ulen); 2185 2186 /* for multi-function programs, copy the JITed 2187 * instructions for all the functions 2188 */ 2189 if (prog->aux->func_cnt) { 2190 u32 len, free, i; 2191 u8 *img; 2192 2193 free = ulen; 2194 for (i = 0; i < prog->aux->func_cnt; i++) { 2195 len = prog->aux->func[i]->jited_len; 2196 len = min_t(u32, len, free); 2197 img = (u8 *) prog->aux->func[i]->bpf_func; 2198 if (copy_to_user(uinsns, img, len)) 2199 return -EFAULT; 2200 uinsns += len; 2201 free -= len; 2202 if (!free) 2203 break; 2204 } 2205 } else { 2206 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 2207 return -EFAULT; 2208 } 2209 } else { 2210 info.jited_prog_insns = 0; 2211 } 2212 } 2213 2214 ulen = info.nr_jited_ksyms; 2215 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1; 2216 if (ulen) { 2217 if (bpf_dump_raw_ok()) { 2218 unsigned long ksym_addr; 2219 u64 __user *user_ksyms; 2220 u32 i; 2221 2222 /* copy the address of the kernel symbol 2223 * corresponding to each function 2224 */ 2225 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 2226 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 2227 if (prog->aux->func_cnt) { 2228 for (i = 0; i < ulen; i++) { 2229 ksym_addr = (unsigned long) 2230 prog->aux->func[i]->bpf_func; 2231 if (put_user((u64) ksym_addr, 2232 &user_ksyms[i])) 2233 return -EFAULT; 2234 } 2235 } else { 2236 ksym_addr = (unsigned long) prog->bpf_func; 2237 if (put_user((u64) ksym_addr, &user_ksyms[0])) 2238 return -EFAULT; 2239 } 2240 } else { 2241 info.jited_ksyms = 0; 2242 } 2243 } 2244 2245 ulen = info.nr_jited_func_lens; 2246 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1; 2247 if (ulen) { 2248 if (bpf_dump_raw_ok()) { 2249 u32 __user *user_lens; 2250 u32 func_len, i; 2251 2252 /* copy the JITed image lengths for each function */ 2253 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 2254 user_lens = u64_to_user_ptr(info.jited_func_lens); 2255 if (prog->aux->func_cnt) { 2256 for (i = 0; i < ulen; i++) { 2257 func_len = 2258 prog->aux->func[i]->jited_len; 2259 if (put_user(func_len, &user_lens[i])) 2260 return -EFAULT; 2261 } 2262 } else { 2263 func_len = prog->jited_len; 2264 if (put_user(func_len, &user_lens[0])) 2265 return -EFAULT; 2266 } 2267 } else { 2268 info.jited_func_lens = 0; 2269 } 2270 } 2271 2272 if (prog->aux->btf) 2273 info.btf_id = btf_id(prog->aux->btf); 2274 2275 ulen = info.nr_func_info; 2276 info.nr_func_info = prog->aux->func_info_cnt; 2277 if (info.nr_func_info && ulen) { 2278 char __user *user_finfo; 2279 2280 user_finfo = u64_to_user_ptr(info.func_info); 2281 ulen = min_t(u32, info.nr_func_info, ulen); 2282 if (copy_to_user(user_finfo, prog->aux->func_info, 2283 info.func_info_rec_size * ulen)) 2284 return -EFAULT; 2285 } 2286 2287 ulen = info.nr_line_info; 2288 info.nr_line_info = prog->aux->nr_linfo; 2289 if (info.nr_line_info && ulen) { 2290 __u8 __user *user_linfo; 2291 2292 user_linfo = u64_to_user_ptr(info.line_info); 2293 ulen = min_t(u32, info.nr_line_info, ulen); 2294 if (copy_to_user(user_linfo, prog->aux->linfo, 2295 info.line_info_rec_size * ulen)) 2296 return -EFAULT; 2297 } 2298 2299 ulen = info.nr_jited_line_info; 2300 if (prog->aux->jited_linfo) 2301 info.nr_jited_line_info = prog->aux->nr_linfo; 2302 else 2303 info.nr_jited_line_info = 0; 2304 if (info.nr_jited_line_info && ulen) { 2305 if (bpf_dump_raw_ok()) { 2306 __u64 __user *user_linfo; 2307 u32 i; 2308 2309 user_linfo = u64_to_user_ptr(info.jited_line_info); 2310 ulen = min_t(u32, info.nr_jited_line_info, ulen); 2311 for (i = 0; i < ulen; i++) { 2312 if (put_user((__u64)(long)prog->aux->jited_linfo[i], 2313 &user_linfo[i])) 2314 return -EFAULT; 2315 } 2316 } else { 2317 info.jited_line_info = 0; 2318 } 2319 } 2320 2321 ulen = info.nr_prog_tags; 2322 info.nr_prog_tags = prog->aux->func_cnt ? : 1; 2323 if (ulen) { 2324 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE]; 2325 u32 i; 2326 2327 user_prog_tags = u64_to_user_ptr(info.prog_tags); 2328 ulen = min_t(u32, info.nr_prog_tags, ulen); 2329 if (prog->aux->func_cnt) { 2330 for (i = 0; i < ulen; i++) { 2331 if (copy_to_user(user_prog_tags[i], 2332 prog->aux->func[i]->tag, 2333 BPF_TAG_SIZE)) 2334 return -EFAULT; 2335 } 2336 } else { 2337 if (copy_to_user(user_prog_tags[0], 2338 prog->tag, BPF_TAG_SIZE)) 2339 return -EFAULT; 2340 } 2341 } 2342 2343 done: 2344 if (copy_to_user(uinfo, &info, info_len) || 2345 put_user(info_len, &uattr->info.info_len)) 2346 return -EFAULT; 2347 2348 return 0; 2349 } 2350 2351 static int bpf_map_get_info_by_fd(struct bpf_map *map, 2352 const union bpf_attr *attr, 2353 union bpf_attr __user *uattr) 2354 { 2355 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2356 struct bpf_map_info info = {}; 2357 u32 info_len = attr->info.info_len; 2358 int err; 2359 2360 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2361 if (err) 2362 return err; 2363 info_len = min_t(u32, sizeof(info), info_len); 2364 2365 info.type = map->map_type; 2366 info.id = map->id; 2367 info.key_size = map->key_size; 2368 info.value_size = map->value_size; 2369 info.max_entries = map->max_entries; 2370 info.map_flags = map->map_flags; 2371 memcpy(info.name, map->name, sizeof(map->name)); 2372 2373 if (map->btf) { 2374 info.btf_id = btf_id(map->btf); 2375 info.btf_key_type_id = map->btf_key_type_id; 2376 info.btf_value_type_id = map->btf_value_type_id; 2377 } 2378 2379 if (bpf_map_is_dev_bound(map)) { 2380 err = bpf_map_offload_info_fill(&info, map); 2381 if (err) 2382 return err; 2383 } 2384 2385 if (copy_to_user(uinfo, &info, info_len) || 2386 put_user(info_len, &uattr->info.info_len)) 2387 return -EFAULT; 2388 2389 return 0; 2390 } 2391 2392 static int bpf_btf_get_info_by_fd(struct btf *btf, 2393 const union bpf_attr *attr, 2394 union bpf_attr __user *uattr) 2395 { 2396 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2397 u32 info_len = attr->info.info_len; 2398 int err; 2399 2400 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 2401 if (err) 2402 return err; 2403 2404 return btf_get_info_by_fd(btf, attr, uattr); 2405 } 2406 2407 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 2408 2409 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 2410 union bpf_attr __user *uattr) 2411 { 2412 int ufd = attr->info.bpf_fd; 2413 struct fd f; 2414 int err; 2415 2416 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 2417 return -EINVAL; 2418 2419 f = fdget(ufd); 2420 if (!f.file) 2421 return -EBADFD; 2422 2423 if (f.file->f_op == &bpf_prog_fops) 2424 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 2425 uattr); 2426 else if (f.file->f_op == &bpf_map_fops) 2427 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 2428 uattr); 2429 else if (f.file->f_op == &btf_fops) 2430 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 2431 else 2432 err = -EINVAL; 2433 2434 fdput(f); 2435 return err; 2436 } 2437 2438 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 2439 2440 static int bpf_btf_load(const union bpf_attr *attr) 2441 { 2442 if (CHECK_ATTR(BPF_BTF_LOAD)) 2443 return -EINVAL; 2444 2445 if (!capable(CAP_SYS_ADMIN)) 2446 return -EPERM; 2447 2448 return btf_new_fd(attr); 2449 } 2450 2451 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 2452 2453 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 2454 { 2455 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 2456 return -EINVAL; 2457 2458 if (!capable(CAP_SYS_ADMIN)) 2459 return -EPERM; 2460 2461 return btf_get_fd_by_id(attr->btf_id); 2462 } 2463 2464 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 2465 union bpf_attr __user *uattr, 2466 u32 prog_id, u32 fd_type, 2467 const char *buf, u64 probe_offset, 2468 u64 probe_addr) 2469 { 2470 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 2471 u32 len = buf ? strlen(buf) : 0, input_len; 2472 int err = 0; 2473 2474 if (put_user(len, &uattr->task_fd_query.buf_len)) 2475 return -EFAULT; 2476 input_len = attr->task_fd_query.buf_len; 2477 if (input_len && ubuf) { 2478 if (!len) { 2479 /* nothing to copy, just make ubuf NULL terminated */ 2480 char zero = '\0'; 2481 2482 if (put_user(zero, ubuf)) 2483 return -EFAULT; 2484 } else if (input_len >= len + 1) { 2485 /* ubuf can hold the string with NULL terminator */ 2486 if (copy_to_user(ubuf, buf, len + 1)) 2487 return -EFAULT; 2488 } else { 2489 /* ubuf cannot hold the string with NULL terminator, 2490 * do a partial copy with NULL terminator. 2491 */ 2492 char zero = '\0'; 2493 2494 err = -ENOSPC; 2495 if (copy_to_user(ubuf, buf, input_len - 1)) 2496 return -EFAULT; 2497 if (put_user(zero, ubuf + input_len - 1)) 2498 return -EFAULT; 2499 } 2500 } 2501 2502 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 2503 put_user(fd_type, &uattr->task_fd_query.fd_type) || 2504 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 2505 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 2506 return -EFAULT; 2507 2508 return err; 2509 } 2510 2511 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 2512 2513 static int bpf_task_fd_query(const union bpf_attr *attr, 2514 union bpf_attr __user *uattr) 2515 { 2516 pid_t pid = attr->task_fd_query.pid; 2517 u32 fd = attr->task_fd_query.fd; 2518 const struct perf_event *event; 2519 struct files_struct *files; 2520 struct task_struct *task; 2521 struct file *file; 2522 int err; 2523 2524 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 2525 return -EINVAL; 2526 2527 if (!capable(CAP_SYS_ADMIN)) 2528 return -EPERM; 2529 2530 if (attr->task_fd_query.flags != 0) 2531 return -EINVAL; 2532 2533 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 2534 if (!task) 2535 return -ENOENT; 2536 2537 files = get_files_struct(task); 2538 put_task_struct(task); 2539 if (!files) 2540 return -ENOENT; 2541 2542 err = 0; 2543 spin_lock(&files->file_lock); 2544 file = fcheck_files(files, fd); 2545 if (!file) 2546 err = -EBADF; 2547 else 2548 get_file(file); 2549 spin_unlock(&files->file_lock); 2550 put_files_struct(files); 2551 2552 if (err) 2553 goto out; 2554 2555 if (file->f_op == &bpf_raw_tp_fops) { 2556 struct bpf_raw_tracepoint *raw_tp = file->private_data; 2557 struct bpf_raw_event_map *btp = raw_tp->btp; 2558 2559 err = bpf_task_fd_query_copy(attr, uattr, 2560 raw_tp->prog->aux->id, 2561 BPF_FD_TYPE_RAW_TRACEPOINT, 2562 btp->tp->name, 0, 0); 2563 goto put_file; 2564 } 2565 2566 event = perf_get_event(file); 2567 if (!IS_ERR(event)) { 2568 u64 probe_offset, probe_addr; 2569 u32 prog_id, fd_type; 2570 const char *buf; 2571 2572 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 2573 &buf, &probe_offset, 2574 &probe_addr); 2575 if (!err) 2576 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 2577 fd_type, buf, 2578 probe_offset, 2579 probe_addr); 2580 goto put_file; 2581 } 2582 2583 err = -ENOTSUPP; 2584 put_file: 2585 fput(file); 2586 out: 2587 return err; 2588 } 2589 2590 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 2591 { 2592 union bpf_attr attr = {}; 2593 int err; 2594 2595 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 2596 return -EPERM; 2597 2598 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 2599 if (err) 2600 return err; 2601 size = min_t(u32, size, sizeof(attr)); 2602 2603 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 2604 if (copy_from_user(&attr, uattr, size) != 0) 2605 return -EFAULT; 2606 2607 err = security_bpf(cmd, &attr, size); 2608 if (err < 0) 2609 return err; 2610 2611 switch (cmd) { 2612 case BPF_MAP_CREATE: 2613 err = map_create(&attr); 2614 break; 2615 case BPF_MAP_LOOKUP_ELEM: 2616 err = map_lookup_elem(&attr); 2617 break; 2618 case BPF_MAP_UPDATE_ELEM: 2619 err = map_update_elem(&attr); 2620 break; 2621 case BPF_MAP_DELETE_ELEM: 2622 err = map_delete_elem(&attr); 2623 break; 2624 case BPF_MAP_GET_NEXT_KEY: 2625 err = map_get_next_key(&attr); 2626 break; 2627 case BPF_PROG_LOAD: 2628 err = bpf_prog_load(&attr, uattr); 2629 break; 2630 case BPF_OBJ_PIN: 2631 err = bpf_obj_pin(&attr); 2632 break; 2633 case BPF_OBJ_GET: 2634 err = bpf_obj_get(&attr); 2635 break; 2636 case BPF_PROG_ATTACH: 2637 err = bpf_prog_attach(&attr); 2638 break; 2639 case BPF_PROG_DETACH: 2640 err = bpf_prog_detach(&attr); 2641 break; 2642 case BPF_PROG_QUERY: 2643 err = bpf_prog_query(&attr, uattr); 2644 break; 2645 case BPF_PROG_TEST_RUN: 2646 err = bpf_prog_test_run(&attr, uattr); 2647 break; 2648 case BPF_PROG_GET_NEXT_ID: 2649 err = bpf_obj_get_next_id(&attr, uattr, 2650 &prog_idr, &prog_idr_lock); 2651 break; 2652 case BPF_MAP_GET_NEXT_ID: 2653 err = bpf_obj_get_next_id(&attr, uattr, 2654 &map_idr, &map_idr_lock); 2655 break; 2656 case BPF_PROG_GET_FD_BY_ID: 2657 err = bpf_prog_get_fd_by_id(&attr); 2658 break; 2659 case BPF_MAP_GET_FD_BY_ID: 2660 err = bpf_map_get_fd_by_id(&attr); 2661 break; 2662 case BPF_OBJ_GET_INFO_BY_FD: 2663 err = bpf_obj_get_info_by_fd(&attr, uattr); 2664 break; 2665 case BPF_RAW_TRACEPOINT_OPEN: 2666 err = bpf_raw_tracepoint_open(&attr); 2667 break; 2668 case BPF_BTF_LOAD: 2669 err = bpf_btf_load(&attr); 2670 break; 2671 case BPF_BTF_GET_FD_BY_ID: 2672 err = bpf_btf_get_fd_by_id(&attr); 2673 break; 2674 case BPF_TASK_FD_QUERY: 2675 err = bpf_task_fd_query(&attr, uattr); 2676 break; 2677 case BPF_MAP_LOOKUP_AND_DELETE_ELEM: 2678 err = map_lookup_and_delete_elem(&attr); 2679 break; 2680 default: 2681 err = -EINVAL; 2682 break; 2683 } 2684 2685 return err; 2686 } 2687