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