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(VERIFY_READ, 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_type *key_type, 460 const struct btf_type *value_type) 461 { 462 return -ENOTSUPP; 463 } 464 465 static int map_check_btf(const struct bpf_map *map, const struct btf *btf, 466 u32 btf_key_id, u32 btf_value_id) 467 { 468 const struct btf_type *key_type, *value_type; 469 u32 key_size, value_size; 470 int ret = 0; 471 472 key_type = btf_type_id_size(btf, &btf_key_id, &key_size); 473 if (!key_type || key_size != map->key_size) 474 return -EINVAL; 475 476 value_type = btf_type_id_size(btf, &btf_value_id, &value_size); 477 if (!value_type || value_size != map->value_size) 478 return -EINVAL; 479 480 if (map->ops->map_check_btf) 481 ret = map->ops->map_check_btf(map, key_type, value_type); 482 483 return ret; 484 } 485 486 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id 487 /* called via syscall */ 488 static int map_create(union bpf_attr *attr) 489 { 490 int numa_node = bpf_map_attr_numa_node(attr); 491 struct bpf_map *map; 492 int f_flags; 493 int err; 494 495 err = CHECK_ATTR(BPF_MAP_CREATE); 496 if (err) 497 return -EINVAL; 498 499 f_flags = bpf_get_file_flag(attr->map_flags); 500 if (f_flags < 0) 501 return f_flags; 502 503 if (numa_node != NUMA_NO_NODE && 504 ((unsigned int)numa_node >= nr_node_ids || 505 !node_online(numa_node))) 506 return -EINVAL; 507 508 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ 509 map = find_and_alloc_map(attr); 510 if (IS_ERR(map)) 511 return PTR_ERR(map); 512 513 err = bpf_obj_name_cpy(map->name, attr->map_name); 514 if (err) 515 goto free_map_nouncharge; 516 517 atomic_set(&map->refcnt, 1); 518 atomic_set(&map->usercnt, 1); 519 520 if (attr->btf_key_type_id || attr->btf_value_type_id) { 521 struct btf *btf; 522 523 if (!attr->btf_key_type_id || !attr->btf_value_type_id) { 524 err = -EINVAL; 525 goto free_map_nouncharge; 526 } 527 528 btf = btf_get_by_fd(attr->btf_fd); 529 if (IS_ERR(btf)) { 530 err = PTR_ERR(btf); 531 goto free_map_nouncharge; 532 } 533 534 err = map_check_btf(map, btf, attr->btf_key_type_id, 535 attr->btf_value_type_id); 536 if (err) { 537 btf_put(btf); 538 goto free_map_nouncharge; 539 } 540 541 map->btf = btf; 542 map->btf_key_type_id = attr->btf_key_type_id; 543 map->btf_value_type_id = attr->btf_value_type_id; 544 } 545 546 err = security_bpf_map_alloc(map); 547 if (err) 548 goto free_map_nouncharge; 549 550 err = bpf_map_init_memlock(map); 551 if (err) 552 goto free_map_sec; 553 554 err = bpf_map_alloc_id(map); 555 if (err) 556 goto free_map; 557 558 err = bpf_map_new_fd(map, f_flags); 559 if (err < 0) { 560 /* failed to allocate fd. 561 * bpf_map_put() is needed because the above 562 * bpf_map_alloc_id() has published the map 563 * to the userspace and the userspace may 564 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. 565 */ 566 bpf_map_put(map); 567 return err; 568 } 569 570 return err; 571 572 free_map: 573 bpf_map_release_memlock(map); 574 free_map_sec: 575 security_bpf_map_free(map); 576 free_map_nouncharge: 577 btf_put(map->btf); 578 map->ops->map_free(map); 579 return err; 580 } 581 582 /* if error is returned, fd is released. 583 * On success caller should complete fd access with matching fdput() 584 */ 585 struct bpf_map *__bpf_map_get(struct fd f) 586 { 587 if (!f.file) 588 return ERR_PTR(-EBADF); 589 if (f.file->f_op != &bpf_map_fops) { 590 fdput(f); 591 return ERR_PTR(-EINVAL); 592 } 593 594 return f.file->private_data; 595 } 596 597 /* prog's and map's refcnt limit */ 598 #define BPF_MAX_REFCNT 32768 599 600 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref) 601 { 602 if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) { 603 atomic_dec(&map->refcnt); 604 return ERR_PTR(-EBUSY); 605 } 606 if (uref) 607 atomic_inc(&map->usercnt); 608 return map; 609 } 610 EXPORT_SYMBOL_GPL(bpf_map_inc); 611 612 struct bpf_map *bpf_map_get_with_uref(u32 ufd) 613 { 614 struct fd f = fdget(ufd); 615 struct bpf_map *map; 616 617 map = __bpf_map_get(f); 618 if (IS_ERR(map)) 619 return map; 620 621 map = bpf_map_inc(map, true); 622 fdput(f); 623 624 return map; 625 } 626 627 /* map_idr_lock should have been held */ 628 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, 629 bool uref) 630 { 631 int refold; 632 633 refold = atomic_fetch_add_unless(&map->refcnt, 1, 0); 634 635 if (refold >= BPF_MAX_REFCNT) { 636 __bpf_map_put(map, false); 637 return ERR_PTR(-EBUSY); 638 } 639 640 if (!refold) 641 return ERR_PTR(-ENOENT); 642 643 if (uref) 644 atomic_inc(&map->usercnt); 645 646 return map; 647 } 648 649 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) 650 { 651 return -ENOTSUPP; 652 } 653 654 /* last field in 'union bpf_attr' used by this command */ 655 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value 656 657 static int map_lookup_elem(union bpf_attr *attr) 658 { 659 void __user *ukey = u64_to_user_ptr(attr->key); 660 void __user *uvalue = u64_to_user_ptr(attr->value); 661 int ufd = attr->map_fd; 662 struct bpf_map *map; 663 void *key, *value, *ptr; 664 u32 value_size; 665 struct fd f; 666 int err; 667 668 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) 669 return -EINVAL; 670 671 f = fdget(ufd); 672 map = __bpf_map_get(f); 673 if (IS_ERR(map)) 674 return PTR_ERR(map); 675 676 if (!(f.file->f_mode & FMODE_CAN_READ)) { 677 err = -EPERM; 678 goto err_put; 679 } 680 681 key = memdup_user(ukey, map->key_size); 682 if (IS_ERR(key)) { 683 err = PTR_ERR(key); 684 goto err_put; 685 } 686 687 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 688 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 689 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) 690 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 691 else if (IS_FD_MAP(map)) 692 value_size = sizeof(u32); 693 else 694 value_size = map->value_size; 695 696 err = -ENOMEM; 697 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 698 if (!value) 699 goto free_key; 700 701 if (bpf_map_is_dev_bound(map)) { 702 err = bpf_map_offload_lookup_elem(map, key, value); 703 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 704 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 705 err = bpf_percpu_hash_copy(map, key, value); 706 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 707 err = bpf_percpu_array_copy(map, key, value); 708 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { 709 err = bpf_stackmap_copy(map, key, value); 710 } else if (IS_FD_ARRAY(map)) { 711 err = bpf_fd_array_map_lookup_elem(map, key, value); 712 } else if (IS_FD_HASH(map)) { 713 err = bpf_fd_htab_map_lookup_elem(map, key, value); 714 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 715 err = bpf_fd_reuseport_array_lookup_elem(map, key, value); 716 } else { 717 rcu_read_lock(); 718 ptr = map->ops->map_lookup_elem(map, key); 719 if (ptr) 720 memcpy(value, ptr, value_size); 721 rcu_read_unlock(); 722 err = ptr ? 0 : -ENOENT; 723 } 724 725 if (err) 726 goto free_value; 727 728 err = -EFAULT; 729 if (copy_to_user(uvalue, value, value_size) != 0) 730 goto free_value; 731 732 err = 0; 733 734 free_value: 735 kfree(value); 736 free_key: 737 kfree(key); 738 err_put: 739 fdput(f); 740 return err; 741 } 742 743 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags 744 745 static int map_update_elem(union bpf_attr *attr) 746 { 747 void __user *ukey = u64_to_user_ptr(attr->key); 748 void __user *uvalue = u64_to_user_ptr(attr->value); 749 int ufd = attr->map_fd; 750 struct bpf_map *map; 751 void *key, *value; 752 u32 value_size; 753 struct fd f; 754 int err; 755 756 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) 757 return -EINVAL; 758 759 f = fdget(ufd); 760 map = __bpf_map_get(f); 761 if (IS_ERR(map)) 762 return PTR_ERR(map); 763 764 if (!(f.file->f_mode & FMODE_CAN_WRITE)) { 765 err = -EPERM; 766 goto err_put; 767 } 768 769 key = memdup_user(ukey, map->key_size); 770 if (IS_ERR(key)) { 771 err = PTR_ERR(key); 772 goto err_put; 773 } 774 775 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 776 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 777 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) 778 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 779 else 780 value_size = map->value_size; 781 782 err = -ENOMEM; 783 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 784 if (!value) 785 goto free_key; 786 787 err = -EFAULT; 788 if (copy_from_user(value, uvalue, value_size) != 0) 789 goto free_value; 790 791 /* Need to create a kthread, thus must support schedule */ 792 if (bpf_map_is_dev_bound(map)) { 793 err = bpf_map_offload_update_elem(map, key, value, attr->flags); 794 goto out; 795 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP || 796 map->map_type == BPF_MAP_TYPE_SOCKHASH || 797 map->map_type == BPF_MAP_TYPE_SOCKMAP) { 798 err = map->ops->map_update_elem(map, key, value, attr->flags); 799 goto out; 800 } 801 802 /* must increment bpf_prog_active to avoid kprobe+bpf triggering from 803 * inside bpf map update or delete otherwise deadlocks are possible 804 */ 805 preempt_disable(); 806 __this_cpu_inc(bpf_prog_active); 807 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 808 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 809 err = bpf_percpu_hash_update(map, key, value, attr->flags); 810 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 811 err = bpf_percpu_array_update(map, key, value, attr->flags); 812 } else if (IS_FD_ARRAY(map)) { 813 rcu_read_lock(); 814 err = bpf_fd_array_map_update_elem(map, f.file, key, value, 815 attr->flags); 816 rcu_read_unlock(); 817 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { 818 rcu_read_lock(); 819 err = bpf_fd_htab_map_update_elem(map, f.file, key, value, 820 attr->flags); 821 rcu_read_unlock(); 822 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 823 /* rcu_read_lock() is not needed */ 824 err = bpf_fd_reuseport_array_update_elem(map, key, value, 825 attr->flags); 826 } else { 827 rcu_read_lock(); 828 err = map->ops->map_update_elem(map, key, value, attr->flags); 829 rcu_read_unlock(); 830 } 831 __this_cpu_dec(bpf_prog_active); 832 preempt_enable(); 833 out: 834 free_value: 835 kfree(value); 836 free_key: 837 kfree(key); 838 err_put: 839 fdput(f); 840 return err; 841 } 842 843 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key 844 845 static int map_delete_elem(union bpf_attr *attr) 846 { 847 void __user *ukey = u64_to_user_ptr(attr->key); 848 int ufd = attr->map_fd; 849 struct bpf_map *map; 850 struct fd f; 851 void *key; 852 int err; 853 854 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) 855 return -EINVAL; 856 857 f = fdget(ufd); 858 map = __bpf_map_get(f); 859 if (IS_ERR(map)) 860 return PTR_ERR(map); 861 862 if (!(f.file->f_mode & FMODE_CAN_WRITE)) { 863 err = -EPERM; 864 goto err_put; 865 } 866 867 key = memdup_user(ukey, map->key_size); 868 if (IS_ERR(key)) { 869 err = PTR_ERR(key); 870 goto err_put; 871 } 872 873 if (bpf_map_is_dev_bound(map)) { 874 err = bpf_map_offload_delete_elem(map, key); 875 goto out; 876 } 877 878 preempt_disable(); 879 __this_cpu_inc(bpf_prog_active); 880 rcu_read_lock(); 881 err = map->ops->map_delete_elem(map, key); 882 rcu_read_unlock(); 883 __this_cpu_dec(bpf_prog_active); 884 preempt_enable(); 885 out: 886 kfree(key); 887 err_put: 888 fdput(f); 889 return err; 890 } 891 892 /* last field in 'union bpf_attr' used by this command */ 893 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key 894 895 static int map_get_next_key(union bpf_attr *attr) 896 { 897 void __user *ukey = u64_to_user_ptr(attr->key); 898 void __user *unext_key = u64_to_user_ptr(attr->next_key); 899 int ufd = attr->map_fd; 900 struct bpf_map *map; 901 void *key, *next_key; 902 struct fd f; 903 int err; 904 905 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) 906 return -EINVAL; 907 908 f = fdget(ufd); 909 map = __bpf_map_get(f); 910 if (IS_ERR(map)) 911 return PTR_ERR(map); 912 913 if (!(f.file->f_mode & FMODE_CAN_READ)) { 914 err = -EPERM; 915 goto err_put; 916 } 917 918 if (ukey) { 919 key = memdup_user(ukey, map->key_size); 920 if (IS_ERR(key)) { 921 err = PTR_ERR(key); 922 goto err_put; 923 } 924 } else { 925 key = NULL; 926 } 927 928 err = -ENOMEM; 929 next_key = kmalloc(map->key_size, GFP_USER); 930 if (!next_key) 931 goto free_key; 932 933 if (bpf_map_is_dev_bound(map)) { 934 err = bpf_map_offload_get_next_key(map, key, next_key); 935 goto out; 936 } 937 938 rcu_read_lock(); 939 err = map->ops->map_get_next_key(map, key, next_key); 940 rcu_read_unlock(); 941 out: 942 if (err) 943 goto free_next_key; 944 945 err = -EFAULT; 946 if (copy_to_user(unext_key, next_key, map->key_size) != 0) 947 goto free_next_key; 948 949 err = 0; 950 951 free_next_key: 952 kfree(next_key); 953 free_key: 954 kfree(key); 955 err_put: 956 fdput(f); 957 return err; 958 } 959 960 static const struct bpf_prog_ops * const bpf_prog_types[] = { 961 #define BPF_PROG_TYPE(_id, _name) \ 962 [_id] = & _name ## _prog_ops, 963 #define BPF_MAP_TYPE(_id, _ops) 964 #include <linux/bpf_types.h> 965 #undef BPF_PROG_TYPE 966 #undef BPF_MAP_TYPE 967 }; 968 969 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) 970 { 971 const struct bpf_prog_ops *ops; 972 973 if (type >= ARRAY_SIZE(bpf_prog_types)) 974 return -EINVAL; 975 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types)); 976 ops = bpf_prog_types[type]; 977 if (!ops) 978 return -EINVAL; 979 980 if (!bpf_prog_is_dev_bound(prog->aux)) 981 prog->aux->ops = ops; 982 else 983 prog->aux->ops = &bpf_offload_prog_ops; 984 prog->type = type; 985 return 0; 986 } 987 988 /* drop refcnt on maps used by eBPF program and free auxilary data */ 989 static void free_used_maps(struct bpf_prog_aux *aux) 990 { 991 int i; 992 993 if (aux->cgroup_storage) 994 bpf_cgroup_storage_release(aux->prog, aux->cgroup_storage); 995 996 for (i = 0; i < aux->used_map_cnt; i++) 997 bpf_map_put(aux->used_maps[i]); 998 999 kfree(aux->used_maps); 1000 } 1001 1002 int __bpf_prog_charge(struct user_struct *user, u32 pages) 1003 { 1004 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1005 unsigned long user_bufs; 1006 1007 if (user) { 1008 user_bufs = atomic_long_add_return(pages, &user->locked_vm); 1009 if (user_bufs > memlock_limit) { 1010 atomic_long_sub(pages, &user->locked_vm); 1011 return -EPERM; 1012 } 1013 } 1014 1015 return 0; 1016 } 1017 1018 void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 1019 { 1020 if (user) 1021 atomic_long_sub(pages, &user->locked_vm); 1022 } 1023 1024 static int bpf_prog_charge_memlock(struct bpf_prog *prog) 1025 { 1026 struct user_struct *user = get_current_user(); 1027 int ret; 1028 1029 ret = __bpf_prog_charge(user, prog->pages); 1030 if (ret) { 1031 free_uid(user); 1032 return ret; 1033 } 1034 1035 prog->aux->user = user; 1036 return 0; 1037 } 1038 1039 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) 1040 { 1041 struct user_struct *user = prog->aux->user; 1042 1043 __bpf_prog_uncharge(user, prog->pages); 1044 free_uid(user); 1045 } 1046 1047 static int bpf_prog_alloc_id(struct bpf_prog *prog) 1048 { 1049 int id; 1050 1051 idr_preload(GFP_KERNEL); 1052 spin_lock_bh(&prog_idr_lock); 1053 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); 1054 if (id > 0) 1055 prog->aux->id = id; 1056 spin_unlock_bh(&prog_idr_lock); 1057 idr_preload_end(); 1058 1059 /* id is in [1, INT_MAX) */ 1060 if (WARN_ON_ONCE(!id)) 1061 return -ENOSPC; 1062 1063 return id > 0 ? 0 : id; 1064 } 1065 1066 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) 1067 { 1068 /* cBPF to eBPF migrations are currently not in the idr store. 1069 * Offloaded programs are removed from the store when their device 1070 * disappears - even if someone grabs an fd to them they are unusable, 1071 * simply waiting for refcnt to drop to be freed. 1072 */ 1073 if (!prog->aux->id) 1074 return; 1075 1076 if (do_idr_lock) 1077 spin_lock_bh(&prog_idr_lock); 1078 else 1079 __acquire(&prog_idr_lock); 1080 1081 idr_remove(&prog_idr, prog->aux->id); 1082 prog->aux->id = 0; 1083 1084 if (do_idr_lock) 1085 spin_unlock_bh(&prog_idr_lock); 1086 else 1087 __release(&prog_idr_lock); 1088 } 1089 1090 static void __bpf_prog_put_rcu(struct rcu_head *rcu) 1091 { 1092 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1093 1094 free_used_maps(aux); 1095 bpf_prog_uncharge_memlock(aux->prog); 1096 security_bpf_prog_free(aux); 1097 bpf_prog_free(aux->prog); 1098 } 1099 1100 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) 1101 { 1102 if (atomic_dec_and_test(&prog->aux->refcnt)) { 1103 /* bpf_prog_free_id() must be called first */ 1104 bpf_prog_free_id(prog, do_idr_lock); 1105 bpf_prog_kallsyms_del_all(prog); 1106 1107 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1108 } 1109 } 1110 1111 void bpf_prog_put(struct bpf_prog *prog) 1112 { 1113 __bpf_prog_put(prog, true); 1114 } 1115 EXPORT_SYMBOL_GPL(bpf_prog_put); 1116 1117 static int bpf_prog_release(struct inode *inode, struct file *filp) 1118 { 1119 struct bpf_prog *prog = filp->private_data; 1120 1121 bpf_prog_put(prog); 1122 return 0; 1123 } 1124 1125 #ifdef CONFIG_PROC_FS 1126 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1127 { 1128 const struct bpf_prog *prog = filp->private_data; 1129 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1130 1131 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1132 seq_printf(m, 1133 "prog_type:\t%u\n" 1134 "prog_jited:\t%u\n" 1135 "prog_tag:\t%s\n" 1136 "memlock:\t%llu\n" 1137 "prog_id:\t%u\n", 1138 prog->type, 1139 prog->jited, 1140 prog_tag, 1141 prog->pages * 1ULL << PAGE_SHIFT, 1142 prog->aux->id); 1143 } 1144 #endif 1145 1146 const struct file_operations bpf_prog_fops = { 1147 #ifdef CONFIG_PROC_FS 1148 .show_fdinfo = bpf_prog_show_fdinfo, 1149 #endif 1150 .release = bpf_prog_release, 1151 .read = bpf_dummy_read, 1152 .write = bpf_dummy_write, 1153 }; 1154 1155 int bpf_prog_new_fd(struct bpf_prog *prog) 1156 { 1157 int ret; 1158 1159 ret = security_bpf_prog(prog); 1160 if (ret < 0) 1161 return ret; 1162 1163 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1164 O_RDWR | O_CLOEXEC); 1165 } 1166 1167 static struct bpf_prog *____bpf_prog_get(struct fd f) 1168 { 1169 if (!f.file) 1170 return ERR_PTR(-EBADF); 1171 if (f.file->f_op != &bpf_prog_fops) { 1172 fdput(f); 1173 return ERR_PTR(-EINVAL); 1174 } 1175 1176 return f.file->private_data; 1177 } 1178 1179 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i) 1180 { 1181 if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) { 1182 atomic_sub(i, &prog->aux->refcnt); 1183 return ERR_PTR(-EBUSY); 1184 } 1185 return prog; 1186 } 1187 EXPORT_SYMBOL_GPL(bpf_prog_add); 1188 1189 void bpf_prog_sub(struct bpf_prog *prog, int i) 1190 { 1191 /* Only to be used for undoing previous bpf_prog_add() in some 1192 * error path. We still know that another entity in our call 1193 * path holds a reference to the program, thus atomic_sub() can 1194 * be safely used in such cases! 1195 */ 1196 WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0); 1197 } 1198 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1199 1200 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog) 1201 { 1202 return bpf_prog_add(prog, 1); 1203 } 1204 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1205 1206 /* prog_idr_lock should have been held */ 1207 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1208 { 1209 int refold; 1210 1211 refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1212 1213 if (refold >= BPF_MAX_REFCNT) { 1214 __bpf_prog_put(prog, false); 1215 return ERR_PTR(-EBUSY); 1216 } 1217 1218 if (!refold) 1219 return ERR_PTR(-ENOENT); 1220 1221 return prog; 1222 } 1223 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1224 1225 bool bpf_prog_get_ok(struct bpf_prog *prog, 1226 enum bpf_prog_type *attach_type, bool attach_drv) 1227 { 1228 /* not an attachment, just a refcount inc, always allow */ 1229 if (!attach_type) 1230 return true; 1231 1232 if (prog->type != *attach_type) 1233 return false; 1234 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1235 return false; 1236 1237 return true; 1238 } 1239 1240 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1241 bool attach_drv) 1242 { 1243 struct fd f = fdget(ufd); 1244 struct bpf_prog *prog; 1245 1246 prog = ____bpf_prog_get(f); 1247 if (IS_ERR(prog)) 1248 return prog; 1249 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1250 prog = ERR_PTR(-EINVAL); 1251 goto out; 1252 } 1253 1254 prog = bpf_prog_inc(prog); 1255 out: 1256 fdput(f); 1257 return prog; 1258 } 1259 1260 struct bpf_prog *bpf_prog_get(u32 ufd) 1261 { 1262 return __bpf_prog_get(ufd, NULL, false); 1263 } 1264 1265 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1266 bool attach_drv) 1267 { 1268 return __bpf_prog_get(ufd, &type, attach_drv); 1269 } 1270 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1271 1272 /* Initially all BPF programs could be loaded w/o specifying 1273 * expected_attach_type. Later for some of them specifying expected_attach_type 1274 * at load time became required so that program could be validated properly. 1275 * Programs of types that are allowed to be loaded both w/ and w/o (for 1276 * backward compatibility) expected_attach_type, should have the default attach 1277 * type assigned to expected_attach_type for the latter case, so that it can be 1278 * validated later at attach time. 1279 * 1280 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1281 * prog type requires it but has some attach types that have to be backward 1282 * compatible. 1283 */ 1284 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1285 { 1286 switch (attr->prog_type) { 1287 case BPF_PROG_TYPE_CGROUP_SOCK: 1288 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1289 * exist so checking for non-zero is the way to go here. 1290 */ 1291 if (!attr->expected_attach_type) 1292 attr->expected_attach_type = 1293 BPF_CGROUP_INET_SOCK_CREATE; 1294 break; 1295 } 1296 } 1297 1298 static int 1299 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type, 1300 enum bpf_attach_type expected_attach_type) 1301 { 1302 switch (prog_type) { 1303 case BPF_PROG_TYPE_CGROUP_SOCK: 1304 switch (expected_attach_type) { 1305 case BPF_CGROUP_INET_SOCK_CREATE: 1306 case BPF_CGROUP_INET4_POST_BIND: 1307 case BPF_CGROUP_INET6_POST_BIND: 1308 return 0; 1309 default: 1310 return -EINVAL; 1311 } 1312 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1313 switch (expected_attach_type) { 1314 case BPF_CGROUP_INET4_BIND: 1315 case BPF_CGROUP_INET6_BIND: 1316 case BPF_CGROUP_INET4_CONNECT: 1317 case BPF_CGROUP_INET6_CONNECT: 1318 case BPF_CGROUP_UDP4_SENDMSG: 1319 case BPF_CGROUP_UDP6_SENDMSG: 1320 return 0; 1321 default: 1322 return -EINVAL; 1323 } 1324 default: 1325 return 0; 1326 } 1327 } 1328 1329 /* last field in 'union bpf_attr' used by this command */ 1330 #define BPF_PROG_LOAD_LAST_FIELD expected_attach_type 1331 1332 static int bpf_prog_load(union bpf_attr *attr) 1333 { 1334 enum bpf_prog_type type = attr->prog_type; 1335 struct bpf_prog *prog; 1336 int err; 1337 char license[128]; 1338 bool is_gpl; 1339 1340 if (CHECK_ATTR(BPF_PROG_LOAD)) 1341 return -EINVAL; 1342 1343 if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT) 1344 return -EINVAL; 1345 1346 /* copy eBPF program license from user space */ 1347 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 1348 sizeof(license) - 1) < 0) 1349 return -EFAULT; 1350 license[sizeof(license) - 1] = 0; 1351 1352 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 1353 is_gpl = license_is_gpl_compatible(license); 1354 1355 if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS) 1356 return -E2BIG; 1357 1358 if (type == BPF_PROG_TYPE_KPROBE && 1359 attr->kern_version != LINUX_VERSION_CODE) 1360 return -EINVAL; 1361 1362 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 1363 type != BPF_PROG_TYPE_CGROUP_SKB && 1364 !capable(CAP_SYS_ADMIN)) 1365 return -EPERM; 1366 1367 bpf_prog_load_fixup_attach_type(attr); 1368 if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type)) 1369 return -EINVAL; 1370 1371 /* plain bpf_prog allocation */ 1372 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 1373 if (!prog) 1374 return -ENOMEM; 1375 1376 prog->expected_attach_type = attr->expected_attach_type; 1377 1378 prog->aux->offload_requested = !!attr->prog_ifindex; 1379 1380 err = security_bpf_prog_alloc(prog->aux); 1381 if (err) 1382 goto free_prog_nouncharge; 1383 1384 err = bpf_prog_charge_memlock(prog); 1385 if (err) 1386 goto free_prog_sec; 1387 1388 prog->len = attr->insn_cnt; 1389 1390 err = -EFAULT; 1391 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 1392 bpf_prog_insn_size(prog)) != 0) 1393 goto free_prog; 1394 1395 prog->orig_prog = NULL; 1396 prog->jited = 0; 1397 1398 atomic_set(&prog->aux->refcnt, 1); 1399 prog->gpl_compatible = is_gpl ? 1 : 0; 1400 1401 if (bpf_prog_is_dev_bound(prog->aux)) { 1402 err = bpf_prog_offload_init(prog, attr); 1403 if (err) 1404 goto free_prog; 1405 } 1406 1407 /* find program type: socket_filter vs tracing_filter */ 1408 err = find_prog_type(type, prog); 1409 if (err < 0) 1410 goto free_prog; 1411 1412 prog->aux->load_time = ktime_get_boot_ns(); 1413 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 1414 if (err) 1415 goto free_prog; 1416 1417 /* run eBPF verifier */ 1418 err = bpf_check(&prog, attr); 1419 if (err < 0) 1420 goto free_used_maps; 1421 1422 prog = bpf_prog_select_runtime(prog, &err); 1423 if (err < 0) 1424 goto free_used_maps; 1425 1426 err = bpf_prog_alloc_id(prog); 1427 if (err) 1428 goto free_used_maps; 1429 1430 err = bpf_prog_new_fd(prog); 1431 if (err < 0) { 1432 /* failed to allocate fd. 1433 * bpf_prog_put() is needed because the above 1434 * bpf_prog_alloc_id() has published the prog 1435 * to the userspace and the userspace may 1436 * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID. 1437 */ 1438 bpf_prog_put(prog); 1439 return err; 1440 } 1441 1442 bpf_prog_kallsyms_add(prog); 1443 return err; 1444 1445 free_used_maps: 1446 bpf_prog_kallsyms_del_subprogs(prog); 1447 free_used_maps(prog->aux); 1448 free_prog: 1449 bpf_prog_uncharge_memlock(prog); 1450 free_prog_sec: 1451 security_bpf_prog_free(prog->aux); 1452 free_prog_nouncharge: 1453 bpf_prog_free(prog); 1454 return err; 1455 } 1456 1457 #define BPF_OBJ_LAST_FIELD file_flags 1458 1459 static int bpf_obj_pin(const union bpf_attr *attr) 1460 { 1461 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 1462 return -EINVAL; 1463 1464 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 1465 } 1466 1467 static int bpf_obj_get(const union bpf_attr *attr) 1468 { 1469 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 1470 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 1471 return -EINVAL; 1472 1473 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 1474 attr->file_flags); 1475 } 1476 1477 struct bpf_raw_tracepoint { 1478 struct bpf_raw_event_map *btp; 1479 struct bpf_prog *prog; 1480 }; 1481 1482 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp) 1483 { 1484 struct bpf_raw_tracepoint *raw_tp = filp->private_data; 1485 1486 if (raw_tp->prog) { 1487 bpf_probe_unregister(raw_tp->btp, raw_tp->prog); 1488 bpf_prog_put(raw_tp->prog); 1489 } 1490 kfree(raw_tp); 1491 return 0; 1492 } 1493 1494 static const struct file_operations bpf_raw_tp_fops = { 1495 .release = bpf_raw_tracepoint_release, 1496 .read = bpf_dummy_read, 1497 .write = bpf_dummy_write, 1498 }; 1499 1500 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 1501 1502 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 1503 { 1504 struct bpf_raw_tracepoint *raw_tp; 1505 struct bpf_raw_event_map *btp; 1506 struct bpf_prog *prog; 1507 char tp_name[128]; 1508 int tp_fd, err; 1509 1510 if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name), 1511 sizeof(tp_name) - 1) < 0) 1512 return -EFAULT; 1513 tp_name[sizeof(tp_name) - 1] = 0; 1514 1515 btp = bpf_find_raw_tracepoint(tp_name); 1516 if (!btp) 1517 return -ENOENT; 1518 1519 raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER); 1520 if (!raw_tp) 1521 return -ENOMEM; 1522 raw_tp->btp = btp; 1523 1524 prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd, 1525 BPF_PROG_TYPE_RAW_TRACEPOINT); 1526 if (IS_ERR(prog)) { 1527 err = PTR_ERR(prog); 1528 goto out_free_tp; 1529 } 1530 1531 err = bpf_probe_register(raw_tp->btp, prog); 1532 if (err) 1533 goto out_put_prog; 1534 1535 raw_tp->prog = prog; 1536 tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp, 1537 O_CLOEXEC); 1538 if (tp_fd < 0) { 1539 bpf_probe_unregister(raw_tp->btp, prog); 1540 err = tp_fd; 1541 goto out_put_prog; 1542 } 1543 return tp_fd; 1544 1545 out_put_prog: 1546 bpf_prog_put(prog); 1547 out_free_tp: 1548 kfree(raw_tp); 1549 return err; 1550 } 1551 1552 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 1553 enum bpf_attach_type attach_type) 1554 { 1555 switch (prog->type) { 1556 case BPF_PROG_TYPE_CGROUP_SOCK: 1557 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1558 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 1559 default: 1560 return 0; 1561 } 1562 } 1563 1564 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags 1565 1566 #define BPF_F_ATTACH_MASK \ 1567 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI) 1568 1569 static int bpf_prog_attach(const union bpf_attr *attr) 1570 { 1571 enum bpf_prog_type ptype; 1572 struct bpf_prog *prog; 1573 int ret; 1574 1575 if (!capable(CAP_NET_ADMIN)) 1576 return -EPERM; 1577 1578 if (CHECK_ATTR(BPF_PROG_ATTACH)) 1579 return -EINVAL; 1580 1581 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 1582 return -EINVAL; 1583 1584 switch (attr->attach_type) { 1585 case BPF_CGROUP_INET_INGRESS: 1586 case BPF_CGROUP_INET_EGRESS: 1587 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1588 break; 1589 case BPF_CGROUP_INET_SOCK_CREATE: 1590 case BPF_CGROUP_INET4_POST_BIND: 1591 case BPF_CGROUP_INET6_POST_BIND: 1592 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1593 break; 1594 case BPF_CGROUP_INET4_BIND: 1595 case BPF_CGROUP_INET6_BIND: 1596 case BPF_CGROUP_INET4_CONNECT: 1597 case BPF_CGROUP_INET6_CONNECT: 1598 case BPF_CGROUP_UDP4_SENDMSG: 1599 case BPF_CGROUP_UDP6_SENDMSG: 1600 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1601 break; 1602 case BPF_CGROUP_SOCK_OPS: 1603 ptype = BPF_PROG_TYPE_SOCK_OPS; 1604 break; 1605 case BPF_CGROUP_DEVICE: 1606 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1607 break; 1608 case BPF_SK_MSG_VERDICT: 1609 ptype = BPF_PROG_TYPE_SK_MSG; 1610 break; 1611 case BPF_SK_SKB_STREAM_PARSER: 1612 case BPF_SK_SKB_STREAM_VERDICT: 1613 ptype = BPF_PROG_TYPE_SK_SKB; 1614 break; 1615 case BPF_LIRC_MODE2: 1616 ptype = BPF_PROG_TYPE_LIRC_MODE2; 1617 break; 1618 case BPF_FLOW_DISSECTOR: 1619 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 1620 break; 1621 default: 1622 return -EINVAL; 1623 } 1624 1625 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 1626 if (IS_ERR(prog)) 1627 return PTR_ERR(prog); 1628 1629 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 1630 bpf_prog_put(prog); 1631 return -EINVAL; 1632 } 1633 1634 switch (ptype) { 1635 case BPF_PROG_TYPE_SK_SKB: 1636 case BPF_PROG_TYPE_SK_MSG: 1637 ret = sockmap_get_from_fd(attr, ptype, prog); 1638 break; 1639 case BPF_PROG_TYPE_LIRC_MODE2: 1640 ret = lirc_prog_attach(attr, prog); 1641 break; 1642 case BPF_PROG_TYPE_FLOW_DISSECTOR: 1643 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 1644 break; 1645 default: 1646 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 1647 } 1648 1649 if (ret) 1650 bpf_prog_put(prog); 1651 return ret; 1652 } 1653 1654 #define BPF_PROG_DETACH_LAST_FIELD attach_type 1655 1656 static int bpf_prog_detach(const union bpf_attr *attr) 1657 { 1658 enum bpf_prog_type ptype; 1659 1660 if (!capable(CAP_NET_ADMIN)) 1661 return -EPERM; 1662 1663 if (CHECK_ATTR(BPF_PROG_DETACH)) 1664 return -EINVAL; 1665 1666 switch (attr->attach_type) { 1667 case BPF_CGROUP_INET_INGRESS: 1668 case BPF_CGROUP_INET_EGRESS: 1669 ptype = BPF_PROG_TYPE_CGROUP_SKB; 1670 break; 1671 case BPF_CGROUP_INET_SOCK_CREATE: 1672 case BPF_CGROUP_INET4_POST_BIND: 1673 case BPF_CGROUP_INET6_POST_BIND: 1674 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 1675 break; 1676 case BPF_CGROUP_INET4_BIND: 1677 case BPF_CGROUP_INET6_BIND: 1678 case BPF_CGROUP_INET4_CONNECT: 1679 case BPF_CGROUP_INET6_CONNECT: 1680 case BPF_CGROUP_UDP4_SENDMSG: 1681 case BPF_CGROUP_UDP6_SENDMSG: 1682 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 1683 break; 1684 case BPF_CGROUP_SOCK_OPS: 1685 ptype = BPF_PROG_TYPE_SOCK_OPS; 1686 break; 1687 case BPF_CGROUP_DEVICE: 1688 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 1689 break; 1690 case BPF_SK_MSG_VERDICT: 1691 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_MSG, NULL); 1692 case BPF_SK_SKB_STREAM_PARSER: 1693 case BPF_SK_SKB_STREAM_VERDICT: 1694 return sockmap_get_from_fd(attr, BPF_PROG_TYPE_SK_SKB, NULL); 1695 case BPF_LIRC_MODE2: 1696 return lirc_prog_detach(attr); 1697 case BPF_FLOW_DISSECTOR: 1698 return skb_flow_dissector_bpf_prog_detach(attr); 1699 default: 1700 return -EINVAL; 1701 } 1702 1703 return cgroup_bpf_prog_detach(attr, ptype); 1704 } 1705 1706 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 1707 1708 static int bpf_prog_query(const union bpf_attr *attr, 1709 union bpf_attr __user *uattr) 1710 { 1711 if (!capable(CAP_NET_ADMIN)) 1712 return -EPERM; 1713 if (CHECK_ATTR(BPF_PROG_QUERY)) 1714 return -EINVAL; 1715 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 1716 return -EINVAL; 1717 1718 switch (attr->query.attach_type) { 1719 case BPF_CGROUP_INET_INGRESS: 1720 case BPF_CGROUP_INET_EGRESS: 1721 case BPF_CGROUP_INET_SOCK_CREATE: 1722 case BPF_CGROUP_INET4_BIND: 1723 case BPF_CGROUP_INET6_BIND: 1724 case BPF_CGROUP_INET4_POST_BIND: 1725 case BPF_CGROUP_INET6_POST_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 case BPF_CGROUP_SOCK_OPS: 1731 case BPF_CGROUP_DEVICE: 1732 break; 1733 case BPF_LIRC_MODE2: 1734 return lirc_prog_query(attr, uattr); 1735 default: 1736 return -EINVAL; 1737 } 1738 1739 return cgroup_bpf_prog_query(attr, uattr); 1740 } 1741 1742 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration 1743 1744 static int bpf_prog_test_run(const union bpf_attr *attr, 1745 union bpf_attr __user *uattr) 1746 { 1747 struct bpf_prog *prog; 1748 int ret = -ENOTSUPP; 1749 1750 if (!capable(CAP_SYS_ADMIN)) 1751 return -EPERM; 1752 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 1753 return -EINVAL; 1754 1755 prog = bpf_prog_get(attr->test.prog_fd); 1756 if (IS_ERR(prog)) 1757 return PTR_ERR(prog); 1758 1759 if (prog->aux->ops->test_run) 1760 ret = prog->aux->ops->test_run(prog, attr, uattr); 1761 1762 bpf_prog_put(prog); 1763 return ret; 1764 } 1765 1766 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 1767 1768 static int bpf_obj_get_next_id(const union bpf_attr *attr, 1769 union bpf_attr __user *uattr, 1770 struct idr *idr, 1771 spinlock_t *lock) 1772 { 1773 u32 next_id = attr->start_id; 1774 int err = 0; 1775 1776 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 1777 return -EINVAL; 1778 1779 if (!capable(CAP_SYS_ADMIN)) 1780 return -EPERM; 1781 1782 next_id++; 1783 spin_lock_bh(lock); 1784 if (!idr_get_next(idr, &next_id)) 1785 err = -ENOENT; 1786 spin_unlock_bh(lock); 1787 1788 if (!err) 1789 err = put_user(next_id, &uattr->next_id); 1790 1791 return err; 1792 } 1793 1794 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 1795 1796 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 1797 { 1798 struct bpf_prog *prog; 1799 u32 id = attr->prog_id; 1800 int fd; 1801 1802 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 1803 return -EINVAL; 1804 1805 if (!capable(CAP_SYS_ADMIN)) 1806 return -EPERM; 1807 1808 spin_lock_bh(&prog_idr_lock); 1809 prog = idr_find(&prog_idr, id); 1810 if (prog) 1811 prog = bpf_prog_inc_not_zero(prog); 1812 else 1813 prog = ERR_PTR(-ENOENT); 1814 spin_unlock_bh(&prog_idr_lock); 1815 1816 if (IS_ERR(prog)) 1817 return PTR_ERR(prog); 1818 1819 fd = bpf_prog_new_fd(prog); 1820 if (fd < 0) 1821 bpf_prog_put(prog); 1822 1823 return fd; 1824 } 1825 1826 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 1827 1828 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 1829 { 1830 struct bpf_map *map; 1831 u32 id = attr->map_id; 1832 int f_flags; 1833 int fd; 1834 1835 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 1836 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 1837 return -EINVAL; 1838 1839 if (!capable(CAP_SYS_ADMIN)) 1840 return -EPERM; 1841 1842 f_flags = bpf_get_file_flag(attr->open_flags); 1843 if (f_flags < 0) 1844 return f_flags; 1845 1846 spin_lock_bh(&map_idr_lock); 1847 map = idr_find(&map_idr, id); 1848 if (map) 1849 map = bpf_map_inc_not_zero(map, true); 1850 else 1851 map = ERR_PTR(-ENOENT); 1852 spin_unlock_bh(&map_idr_lock); 1853 1854 if (IS_ERR(map)) 1855 return PTR_ERR(map); 1856 1857 fd = bpf_map_new_fd(map, f_flags); 1858 if (fd < 0) 1859 bpf_map_put(map); 1860 1861 return fd; 1862 } 1863 1864 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 1865 unsigned long addr) 1866 { 1867 int i; 1868 1869 for (i = 0; i < prog->aux->used_map_cnt; i++) 1870 if (prog->aux->used_maps[i] == (void *)addr) 1871 return prog->aux->used_maps[i]; 1872 return NULL; 1873 } 1874 1875 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 1876 { 1877 const struct bpf_map *map; 1878 struct bpf_insn *insns; 1879 u64 imm; 1880 int i; 1881 1882 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 1883 GFP_USER); 1884 if (!insns) 1885 return insns; 1886 1887 for (i = 0; i < prog->len; i++) { 1888 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 1889 insns[i].code = BPF_JMP | BPF_CALL; 1890 insns[i].imm = BPF_FUNC_tail_call; 1891 /* fall-through */ 1892 } 1893 if (insns[i].code == (BPF_JMP | BPF_CALL) || 1894 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 1895 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 1896 insns[i].code = BPF_JMP | BPF_CALL; 1897 if (!bpf_dump_raw_ok()) 1898 insns[i].imm = 0; 1899 continue; 1900 } 1901 1902 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 1903 continue; 1904 1905 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 1906 map = bpf_map_from_imm(prog, imm); 1907 if (map) { 1908 insns[i].src_reg = BPF_PSEUDO_MAP_FD; 1909 insns[i].imm = map->id; 1910 insns[i + 1].imm = 0; 1911 continue; 1912 } 1913 1914 if (!bpf_dump_raw_ok() && 1915 imm == (unsigned long)prog->aux) { 1916 insns[i].imm = 0; 1917 insns[i + 1].imm = 0; 1918 continue; 1919 } 1920 } 1921 1922 return insns; 1923 } 1924 1925 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 1926 const union bpf_attr *attr, 1927 union bpf_attr __user *uattr) 1928 { 1929 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 1930 struct bpf_prog_info info = {}; 1931 u32 info_len = attr->info.info_len; 1932 char __user *uinsns; 1933 u32 ulen; 1934 int err; 1935 1936 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 1937 if (err) 1938 return err; 1939 info_len = min_t(u32, sizeof(info), info_len); 1940 1941 if (copy_from_user(&info, uinfo, info_len)) 1942 return -EFAULT; 1943 1944 info.type = prog->type; 1945 info.id = prog->aux->id; 1946 info.load_time = prog->aux->load_time; 1947 info.created_by_uid = from_kuid_munged(current_user_ns(), 1948 prog->aux->user->uid); 1949 info.gpl_compatible = prog->gpl_compatible; 1950 1951 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 1952 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 1953 1954 ulen = info.nr_map_ids; 1955 info.nr_map_ids = prog->aux->used_map_cnt; 1956 ulen = min_t(u32, info.nr_map_ids, ulen); 1957 if (ulen) { 1958 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 1959 u32 i; 1960 1961 for (i = 0; i < ulen; i++) 1962 if (put_user(prog->aux->used_maps[i]->id, 1963 &user_map_ids[i])) 1964 return -EFAULT; 1965 } 1966 1967 if (!capable(CAP_SYS_ADMIN)) { 1968 info.jited_prog_len = 0; 1969 info.xlated_prog_len = 0; 1970 info.nr_jited_ksyms = 0; 1971 goto done; 1972 } 1973 1974 ulen = info.xlated_prog_len; 1975 info.xlated_prog_len = bpf_prog_insn_size(prog); 1976 if (info.xlated_prog_len && ulen) { 1977 struct bpf_insn *insns_sanitized; 1978 bool fault; 1979 1980 if (prog->blinded && !bpf_dump_raw_ok()) { 1981 info.xlated_prog_insns = 0; 1982 goto done; 1983 } 1984 insns_sanitized = bpf_insn_prepare_dump(prog); 1985 if (!insns_sanitized) 1986 return -ENOMEM; 1987 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 1988 ulen = min_t(u32, info.xlated_prog_len, ulen); 1989 fault = copy_to_user(uinsns, insns_sanitized, ulen); 1990 kfree(insns_sanitized); 1991 if (fault) 1992 return -EFAULT; 1993 } 1994 1995 if (bpf_prog_is_dev_bound(prog->aux)) { 1996 err = bpf_prog_offload_info_fill(&info, prog); 1997 if (err) 1998 return err; 1999 goto done; 2000 } 2001 2002 /* NOTE: the following code is supposed to be skipped for offload. 2003 * bpf_prog_offload_info_fill() is the place to fill similar fields 2004 * for offload. 2005 */ 2006 ulen = info.jited_prog_len; 2007 if (prog->aux->func_cnt) { 2008 u32 i; 2009 2010 info.jited_prog_len = 0; 2011 for (i = 0; i < prog->aux->func_cnt; i++) 2012 info.jited_prog_len += prog->aux->func[i]->jited_len; 2013 } else { 2014 info.jited_prog_len = prog->jited_len; 2015 } 2016 2017 if (info.jited_prog_len && ulen) { 2018 if (bpf_dump_raw_ok()) { 2019 uinsns = u64_to_user_ptr(info.jited_prog_insns); 2020 ulen = min_t(u32, info.jited_prog_len, ulen); 2021 2022 /* for multi-function programs, copy the JITed 2023 * instructions for all the functions 2024 */ 2025 if (prog->aux->func_cnt) { 2026 u32 len, free, i; 2027 u8 *img; 2028 2029 free = ulen; 2030 for (i = 0; i < prog->aux->func_cnt; i++) { 2031 len = prog->aux->func[i]->jited_len; 2032 len = min_t(u32, len, free); 2033 img = (u8 *) prog->aux->func[i]->bpf_func; 2034 if (copy_to_user(uinsns, img, len)) 2035 return -EFAULT; 2036 uinsns += len; 2037 free -= len; 2038 if (!free) 2039 break; 2040 } 2041 } else { 2042 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 2043 return -EFAULT; 2044 } 2045 } else { 2046 info.jited_prog_insns = 0; 2047 } 2048 } 2049 2050 ulen = info.nr_jited_ksyms; 2051 info.nr_jited_ksyms = prog->aux->func_cnt; 2052 if (info.nr_jited_ksyms && ulen) { 2053 if (bpf_dump_raw_ok()) { 2054 u64 __user *user_ksyms; 2055 ulong ksym_addr; 2056 u32 i; 2057 2058 /* copy the address of the kernel symbol 2059 * corresponding to each function 2060 */ 2061 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 2062 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 2063 for (i = 0; i < ulen; i++) { 2064 ksym_addr = (ulong) prog->aux->func[i]->bpf_func; 2065 ksym_addr &= PAGE_MASK; 2066 if (put_user((u64) ksym_addr, &user_ksyms[i])) 2067 return -EFAULT; 2068 } 2069 } else { 2070 info.jited_ksyms = 0; 2071 } 2072 } 2073 2074 ulen = info.nr_jited_func_lens; 2075 info.nr_jited_func_lens = prog->aux->func_cnt; 2076 if (info.nr_jited_func_lens && ulen) { 2077 if (bpf_dump_raw_ok()) { 2078 u32 __user *user_lens; 2079 u32 func_len, i; 2080 2081 /* copy the JITed image lengths for each function */ 2082 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 2083 user_lens = u64_to_user_ptr(info.jited_func_lens); 2084 for (i = 0; i < ulen; i++) { 2085 func_len = prog->aux->func[i]->jited_len; 2086 if (put_user(func_len, &user_lens[i])) 2087 return -EFAULT; 2088 } 2089 } else { 2090 info.jited_func_lens = 0; 2091 } 2092 } 2093 2094 done: 2095 if (copy_to_user(uinfo, &info, info_len) || 2096 put_user(info_len, &uattr->info.info_len)) 2097 return -EFAULT; 2098 2099 return 0; 2100 } 2101 2102 static int bpf_map_get_info_by_fd(struct bpf_map *map, 2103 const union bpf_attr *attr, 2104 union bpf_attr __user *uattr) 2105 { 2106 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2107 struct bpf_map_info info = {}; 2108 u32 info_len = attr->info.info_len; 2109 int err; 2110 2111 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2112 if (err) 2113 return err; 2114 info_len = min_t(u32, sizeof(info), info_len); 2115 2116 info.type = map->map_type; 2117 info.id = map->id; 2118 info.key_size = map->key_size; 2119 info.value_size = map->value_size; 2120 info.max_entries = map->max_entries; 2121 info.map_flags = map->map_flags; 2122 memcpy(info.name, map->name, sizeof(map->name)); 2123 2124 if (map->btf) { 2125 info.btf_id = btf_id(map->btf); 2126 info.btf_key_type_id = map->btf_key_type_id; 2127 info.btf_value_type_id = map->btf_value_type_id; 2128 } 2129 2130 if (bpf_map_is_dev_bound(map)) { 2131 err = bpf_map_offload_info_fill(&info, map); 2132 if (err) 2133 return err; 2134 } 2135 2136 if (copy_to_user(uinfo, &info, info_len) || 2137 put_user(info_len, &uattr->info.info_len)) 2138 return -EFAULT; 2139 2140 return 0; 2141 } 2142 2143 static int bpf_btf_get_info_by_fd(struct btf *btf, 2144 const union bpf_attr *attr, 2145 union bpf_attr __user *uattr) 2146 { 2147 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2148 u32 info_len = attr->info.info_len; 2149 int err; 2150 2151 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 2152 if (err) 2153 return err; 2154 2155 return btf_get_info_by_fd(btf, attr, uattr); 2156 } 2157 2158 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 2159 2160 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 2161 union bpf_attr __user *uattr) 2162 { 2163 int ufd = attr->info.bpf_fd; 2164 struct fd f; 2165 int err; 2166 2167 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 2168 return -EINVAL; 2169 2170 f = fdget(ufd); 2171 if (!f.file) 2172 return -EBADFD; 2173 2174 if (f.file->f_op == &bpf_prog_fops) 2175 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 2176 uattr); 2177 else if (f.file->f_op == &bpf_map_fops) 2178 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 2179 uattr); 2180 else if (f.file->f_op == &btf_fops) 2181 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 2182 else 2183 err = -EINVAL; 2184 2185 fdput(f); 2186 return err; 2187 } 2188 2189 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 2190 2191 static int bpf_btf_load(const union bpf_attr *attr) 2192 { 2193 if (CHECK_ATTR(BPF_BTF_LOAD)) 2194 return -EINVAL; 2195 2196 if (!capable(CAP_SYS_ADMIN)) 2197 return -EPERM; 2198 2199 return btf_new_fd(attr); 2200 } 2201 2202 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 2203 2204 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 2205 { 2206 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 2207 return -EINVAL; 2208 2209 if (!capable(CAP_SYS_ADMIN)) 2210 return -EPERM; 2211 2212 return btf_get_fd_by_id(attr->btf_id); 2213 } 2214 2215 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 2216 union bpf_attr __user *uattr, 2217 u32 prog_id, u32 fd_type, 2218 const char *buf, u64 probe_offset, 2219 u64 probe_addr) 2220 { 2221 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 2222 u32 len = buf ? strlen(buf) : 0, input_len; 2223 int err = 0; 2224 2225 if (put_user(len, &uattr->task_fd_query.buf_len)) 2226 return -EFAULT; 2227 input_len = attr->task_fd_query.buf_len; 2228 if (input_len && ubuf) { 2229 if (!len) { 2230 /* nothing to copy, just make ubuf NULL terminated */ 2231 char zero = '\0'; 2232 2233 if (put_user(zero, ubuf)) 2234 return -EFAULT; 2235 } else if (input_len >= len + 1) { 2236 /* ubuf can hold the string with NULL terminator */ 2237 if (copy_to_user(ubuf, buf, len + 1)) 2238 return -EFAULT; 2239 } else { 2240 /* ubuf cannot hold the string with NULL terminator, 2241 * do a partial copy with NULL terminator. 2242 */ 2243 char zero = '\0'; 2244 2245 err = -ENOSPC; 2246 if (copy_to_user(ubuf, buf, input_len - 1)) 2247 return -EFAULT; 2248 if (put_user(zero, ubuf + input_len - 1)) 2249 return -EFAULT; 2250 } 2251 } 2252 2253 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 2254 put_user(fd_type, &uattr->task_fd_query.fd_type) || 2255 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 2256 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 2257 return -EFAULT; 2258 2259 return err; 2260 } 2261 2262 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 2263 2264 static int bpf_task_fd_query(const union bpf_attr *attr, 2265 union bpf_attr __user *uattr) 2266 { 2267 pid_t pid = attr->task_fd_query.pid; 2268 u32 fd = attr->task_fd_query.fd; 2269 const struct perf_event *event; 2270 struct files_struct *files; 2271 struct task_struct *task; 2272 struct file *file; 2273 int err; 2274 2275 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 2276 return -EINVAL; 2277 2278 if (!capable(CAP_SYS_ADMIN)) 2279 return -EPERM; 2280 2281 if (attr->task_fd_query.flags != 0) 2282 return -EINVAL; 2283 2284 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 2285 if (!task) 2286 return -ENOENT; 2287 2288 files = get_files_struct(task); 2289 put_task_struct(task); 2290 if (!files) 2291 return -ENOENT; 2292 2293 err = 0; 2294 spin_lock(&files->file_lock); 2295 file = fcheck_files(files, fd); 2296 if (!file) 2297 err = -EBADF; 2298 else 2299 get_file(file); 2300 spin_unlock(&files->file_lock); 2301 put_files_struct(files); 2302 2303 if (err) 2304 goto out; 2305 2306 if (file->f_op == &bpf_raw_tp_fops) { 2307 struct bpf_raw_tracepoint *raw_tp = file->private_data; 2308 struct bpf_raw_event_map *btp = raw_tp->btp; 2309 2310 err = bpf_task_fd_query_copy(attr, uattr, 2311 raw_tp->prog->aux->id, 2312 BPF_FD_TYPE_RAW_TRACEPOINT, 2313 btp->tp->name, 0, 0); 2314 goto put_file; 2315 } 2316 2317 event = perf_get_event(file); 2318 if (!IS_ERR(event)) { 2319 u64 probe_offset, probe_addr; 2320 u32 prog_id, fd_type; 2321 const char *buf; 2322 2323 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 2324 &buf, &probe_offset, 2325 &probe_addr); 2326 if (!err) 2327 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 2328 fd_type, buf, 2329 probe_offset, 2330 probe_addr); 2331 goto put_file; 2332 } 2333 2334 err = -ENOTSUPP; 2335 put_file: 2336 fput(file); 2337 out: 2338 return err; 2339 } 2340 2341 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 2342 { 2343 union bpf_attr attr = {}; 2344 int err; 2345 2346 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 2347 return -EPERM; 2348 2349 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 2350 if (err) 2351 return err; 2352 size = min_t(u32, size, sizeof(attr)); 2353 2354 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 2355 if (copy_from_user(&attr, uattr, size) != 0) 2356 return -EFAULT; 2357 2358 err = security_bpf(cmd, &attr, size); 2359 if (err < 0) 2360 return err; 2361 2362 switch (cmd) { 2363 case BPF_MAP_CREATE: 2364 err = map_create(&attr); 2365 break; 2366 case BPF_MAP_LOOKUP_ELEM: 2367 err = map_lookup_elem(&attr); 2368 break; 2369 case BPF_MAP_UPDATE_ELEM: 2370 err = map_update_elem(&attr); 2371 break; 2372 case BPF_MAP_DELETE_ELEM: 2373 err = map_delete_elem(&attr); 2374 break; 2375 case BPF_MAP_GET_NEXT_KEY: 2376 err = map_get_next_key(&attr); 2377 break; 2378 case BPF_PROG_LOAD: 2379 err = bpf_prog_load(&attr); 2380 break; 2381 case BPF_OBJ_PIN: 2382 err = bpf_obj_pin(&attr); 2383 break; 2384 case BPF_OBJ_GET: 2385 err = bpf_obj_get(&attr); 2386 break; 2387 case BPF_PROG_ATTACH: 2388 err = bpf_prog_attach(&attr); 2389 break; 2390 case BPF_PROG_DETACH: 2391 err = bpf_prog_detach(&attr); 2392 break; 2393 case BPF_PROG_QUERY: 2394 err = bpf_prog_query(&attr, uattr); 2395 break; 2396 case BPF_PROG_TEST_RUN: 2397 err = bpf_prog_test_run(&attr, uattr); 2398 break; 2399 case BPF_PROG_GET_NEXT_ID: 2400 err = bpf_obj_get_next_id(&attr, uattr, 2401 &prog_idr, &prog_idr_lock); 2402 break; 2403 case BPF_MAP_GET_NEXT_ID: 2404 err = bpf_obj_get_next_id(&attr, uattr, 2405 &map_idr, &map_idr_lock); 2406 break; 2407 case BPF_PROG_GET_FD_BY_ID: 2408 err = bpf_prog_get_fd_by_id(&attr); 2409 break; 2410 case BPF_MAP_GET_FD_BY_ID: 2411 err = bpf_map_get_fd_by_id(&attr); 2412 break; 2413 case BPF_OBJ_GET_INFO_BY_FD: 2414 err = bpf_obj_get_info_by_fd(&attr, uattr); 2415 break; 2416 case BPF_RAW_TRACEPOINT_OPEN: 2417 err = bpf_raw_tracepoint_open(&attr); 2418 break; 2419 case BPF_BTF_LOAD: 2420 err = bpf_btf_load(&attr); 2421 break; 2422 case BPF_BTF_GET_FD_BY_ID: 2423 err = bpf_btf_get_fd_by_id(&attr); 2424 break; 2425 case BPF_TASK_FD_QUERY: 2426 err = bpf_task_fd_query(&attr, uattr); 2427 break; 2428 default: 2429 err = -EINVAL; 2430 break; 2431 } 2432 2433 return err; 2434 } 2435