1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com 3 */ 4 #include <linux/bpf.h> 5 #include <linux/bpf_trace.h> 6 #include <linux/bpf_lirc.h> 7 #include <linux/btf.h> 8 #include <linux/syscalls.h> 9 #include <linux/slab.h> 10 #include <linux/sched/signal.h> 11 #include <linux/vmalloc.h> 12 #include <linux/mmzone.h> 13 #include <linux/anon_inodes.h> 14 #include <linux/fdtable.h> 15 #include <linux/file.h> 16 #include <linux/fs.h> 17 #include <linux/license.h> 18 #include <linux/filter.h> 19 #include <linux/version.h> 20 #include <linux/kernel.h> 21 #include <linux/idr.h> 22 #include <linux/cred.h> 23 #include <linux/timekeeping.h> 24 #include <linux/ctype.h> 25 #include <linux/nospec.h> 26 #include <linux/audit.h> 27 #include <uapi/linux/btf.h> 28 29 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \ 30 (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \ 31 (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 32 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY) 33 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) 34 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \ 35 IS_FD_HASH(map)) 36 37 #define BPF_OBJ_FLAG_MASK (BPF_F_RDONLY | BPF_F_WRONLY) 38 39 DEFINE_PER_CPU(int, bpf_prog_active); 40 static DEFINE_IDR(prog_idr); 41 static DEFINE_SPINLOCK(prog_idr_lock); 42 static DEFINE_IDR(map_idr); 43 static DEFINE_SPINLOCK(map_idr_lock); 44 45 int sysctl_unprivileged_bpf_disabled __read_mostly; 46 47 static const struct bpf_map_ops * const bpf_map_types[] = { 48 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) 49 #define BPF_MAP_TYPE(_id, _ops) \ 50 [_id] = &_ops, 51 #include <linux/bpf_types.h> 52 #undef BPF_PROG_TYPE 53 #undef BPF_MAP_TYPE 54 }; 55 56 /* 57 * If we're handed a bigger struct than we know of, ensure all the unknown bits 58 * are 0 - i.e. new user-space does not rely on any kernel feature extensions 59 * we don't know about yet. 60 * 61 * There is a ToCToU between this function call and the following 62 * copy_from_user() call. However, this is not a concern since this function is 63 * meant to be a future-proofing of bits. 64 */ 65 int bpf_check_uarg_tail_zero(void __user *uaddr, 66 size_t expected_size, 67 size_t actual_size) 68 { 69 unsigned char __user *addr; 70 unsigned char __user *end; 71 unsigned char val; 72 int err; 73 74 if (unlikely(actual_size > PAGE_SIZE)) /* silly large */ 75 return -E2BIG; 76 77 if (unlikely(!access_ok(uaddr, actual_size))) 78 return -EFAULT; 79 80 if (actual_size <= expected_size) 81 return 0; 82 83 addr = uaddr + expected_size; 84 end = uaddr + actual_size; 85 86 for (; addr < end; addr++) { 87 err = get_user(val, addr); 88 if (err) 89 return err; 90 if (val) 91 return -E2BIG; 92 } 93 94 return 0; 95 } 96 97 const struct bpf_map_ops bpf_map_offload_ops = { 98 .map_alloc = bpf_map_offload_map_alloc, 99 .map_free = bpf_map_offload_map_free, 100 .map_check_btf = map_check_no_btf, 101 }; 102 103 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) 104 { 105 const struct bpf_map_ops *ops; 106 u32 type = attr->map_type; 107 struct bpf_map *map; 108 int err; 109 110 if (type >= ARRAY_SIZE(bpf_map_types)) 111 return ERR_PTR(-EINVAL); 112 type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types)); 113 ops = bpf_map_types[type]; 114 if (!ops) 115 return ERR_PTR(-EINVAL); 116 117 if (ops->map_alloc_check) { 118 err = ops->map_alloc_check(attr); 119 if (err) 120 return ERR_PTR(err); 121 } 122 if (attr->map_ifindex) 123 ops = &bpf_map_offload_ops; 124 map = ops->map_alloc(attr); 125 if (IS_ERR(map)) 126 return map; 127 map->ops = ops; 128 map->map_type = type; 129 return map; 130 } 131 132 static u32 bpf_map_value_size(struct bpf_map *map) 133 { 134 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 135 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 136 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 137 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 138 return round_up(map->value_size, 8) * num_possible_cpus(); 139 else if (IS_FD_MAP(map)) 140 return sizeof(u32); 141 else 142 return map->value_size; 143 } 144 145 static void maybe_wait_bpf_programs(struct bpf_map *map) 146 { 147 /* Wait for any running BPF programs to complete so that 148 * userspace, when we return to it, knows that all programs 149 * that could be running use the new map value. 150 */ 151 if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS || 152 map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) 153 synchronize_rcu(); 154 } 155 156 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key, 157 void *value, __u64 flags) 158 { 159 int err; 160 161 /* Need to create a kthread, thus must support schedule */ 162 if (bpf_map_is_dev_bound(map)) { 163 return bpf_map_offload_update_elem(map, key, value, flags); 164 } else if (map->map_type == BPF_MAP_TYPE_CPUMAP || 165 map->map_type == BPF_MAP_TYPE_SOCKHASH || 166 map->map_type == BPF_MAP_TYPE_SOCKMAP || 167 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { 168 return map->ops->map_update_elem(map, key, value, flags); 169 } else if (IS_FD_PROG_ARRAY(map)) { 170 return bpf_fd_array_map_update_elem(map, f.file, key, value, 171 flags); 172 } 173 174 bpf_disable_instrumentation(); 175 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 176 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 177 err = bpf_percpu_hash_update(map, key, value, flags); 178 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 179 err = bpf_percpu_array_update(map, key, value, flags); 180 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 181 err = bpf_percpu_cgroup_storage_update(map, key, value, 182 flags); 183 } else if (IS_FD_ARRAY(map)) { 184 rcu_read_lock(); 185 err = bpf_fd_array_map_update_elem(map, f.file, key, value, 186 flags); 187 rcu_read_unlock(); 188 } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) { 189 rcu_read_lock(); 190 err = bpf_fd_htab_map_update_elem(map, f.file, key, value, 191 flags); 192 rcu_read_unlock(); 193 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 194 /* rcu_read_lock() is not needed */ 195 err = bpf_fd_reuseport_array_update_elem(map, key, value, 196 flags); 197 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 198 map->map_type == BPF_MAP_TYPE_STACK) { 199 err = map->ops->map_push_elem(map, value, flags); 200 } else { 201 rcu_read_lock(); 202 err = map->ops->map_update_elem(map, key, value, flags); 203 rcu_read_unlock(); 204 } 205 bpf_enable_instrumentation(); 206 maybe_wait_bpf_programs(map); 207 208 return err; 209 } 210 211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value, 212 __u64 flags) 213 { 214 void *ptr; 215 int err; 216 217 if (bpf_map_is_dev_bound(map)) 218 return bpf_map_offload_lookup_elem(map, key, value); 219 220 bpf_disable_instrumentation(); 221 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 222 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) { 223 err = bpf_percpu_hash_copy(map, key, value); 224 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) { 225 err = bpf_percpu_array_copy(map, key, value); 226 } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) { 227 err = bpf_percpu_cgroup_storage_copy(map, key, value); 228 } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) { 229 err = bpf_stackmap_copy(map, key, value); 230 } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) { 231 err = bpf_fd_array_map_lookup_elem(map, key, value); 232 } else if (IS_FD_HASH(map)) { 233 err = bpf_fd_htab_map_lookup_elem(map, key, value); 234 } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) { 235 err = bpf_fd_reuseport_array_lookup_elem(map, key, value); 236 } else if (map->map_type == BPF_MAP_TYPE_QUEUE || 237 map->map_type == BPF_MAP_TYPE_STACK) { 238 err = map->ops->map_peek_elem(map, value); 239 } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { 240 /* struct_ops map requires directly updating "value" */ 241 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value); 242 } else { 243 rcu_read_lock(); 244 if (map->ops->map_lookup_elem_sys_only) 245 ptr = map->ops->map_lookup_elem_sys_only(map, key); 246 else 247 ptr = map->ops->map_lookup_elem(map, key); 248 if (IS_ERR(ptr)) { 249 err = PTR_ERR(ptr); 250 } else if (!ptr) { 251 err = -ENOENT; 252 } else { 253 err = 0; 254 if (flags & BPF_F_LOCK) 255 /* lock 'ptr' and copy everything but lock */ 256 copy_map_value_locked(map, value, ptr, true); 257 else 258 copy_map_value(map, value, ptr); 259 /* mask lock, since value wasn't zero inited */ 260 check_and_init_map_lock(map, value); 261 } 262 rcu_read_unlock(); 263 } 264 265 bpf_enable_instrumentation(); 266 maybe_wait_bpf_programs(map); 267 268 return err; 269 } 270 271 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable) 272 { 273 /* We really just want to fail instead of triggering OOM killer 274 * under memory pressure, therefore we set __GFP_NORETRY to kmalloc, 275 * which is used for lower order allocation requests. 276 * 277 * It has been observed that higher order allocation requests done by 278 * vmalloc with __GFP_NORETRY being set might fail due to not trying 279 * to reclaim memory from the page cache, thus we set 280 * __GFP_RETRY_MAYFAIL to avoid such situations. 281 */ 282 283 const gfp_t flags = __GFP_NOWARN | __GFP_ZERO; 284 void *area; 285 286 if (size >= SIZE_MAX) 287 return NULL; 288 289 /* kmalloc()'ed memory can't be mmap()'ed */ 290 if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { 291 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags, 292 numa_node); 293 if (area != NULL) 294 return area; 295 } 296 if (mmapable) { 297 BUG_ON(!PAGE_ALIGNED(size)); 298 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL | 299 __GFP_RETRY_MAYFAIL | flags); 300 } 301 return __vmalloc_node_flags_caller(size, numa_node, 302 GFP_KERNEL | __GFP_RETRY_MAYFAIL | 303 flags, __builtin_return_address(0)); 304 } 305 306 void *bpf_map_area_alloc(u64 size, int numa_node) 307 { 308 return __bpf_map_area_alloc(size, numa_node, false); 309 } 310 311 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node) 312 { 313 return __bpf_map_area_alloc(size, numa_node, true); 314 } 315 316 void bpf_map_area_free(void *area) 317 { 318 kvfree(area); 319 } 320 321 static u32 bpf_map_flags_retain_permanent(u32 flags) 322 { 323 /* Some map creation flags are not tied to the map object but 324 * rather to the map fd instead, so they have no meaning upon 325 * map object inspection since multiple file descriptors with 326 * different (access) properties can exist here. Thus, given 327 * this has zero meaning for the map itself, lets clear these 328 * from here. 329 */ 330 return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY); 331 } 332 333 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr) 334 { 335 map->map_type = attr->map_type; 336 map->key_size = attr->key_size; 337 map->value_size = attr->value_size; 338 map->max_entries = attr->max_entries; 339 map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags); 340 map->numa_node = bpf_map_attr_numa_node(attr); 341 } 342 343 static int bpf_charge_memlock(struct user_struct *user, u32 pages) 344 { 345 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 346 347 if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) { 348 atomic_long_sub(pages, &user->locked_vm); 349 return -EPERM; 350 } 351 return 0; 352 } 353 354 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages) 355 { 356 if (user) 357 atomic_long_sub(pages, &user->locked_vm); 358 } 359 360 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size) 361 { 362 u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT; 363 struct user_struct *user; 364 int ret; 365 366 if (size >= U32_MAX - PAGE_SIZE) 367 return -E2BIG; 368 369 user = get_current_user(); 370 ret = bpf_charge_memlock(user, pages); 371 if (ret) { 372 free_uid(user); 373 return ret; 374 } 375 376 mem->pages = pages; 377 mem->user = user; 378 379 return 0; 380 } 381 382 void bpf_map_charge_finish(struct bpf_map_memory *mem) 383 { 384 bpf_uncharge_memlock(mem->user, mem->pages); 385 free_uid(mem->user); 386 } 387 388 void bpf_map_charge_move(struct bpf_map_memory *dst, 389 struct bpf_map_memory *src) 390 { 391 *dst = *src; 392 393 /* Make sure src will not be used for the redundant uncharging. */ 394 memset(src, 0, sizeof(struct bpf_map_memory)); 395 } 396 397 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages) 398 { 399 int ret; 400 401 ret = bpf_charge_memlock(map->memory.user, pages); 402 if (ret) 403 return ret; 404 map->memory.pages += pages; 405 return ret; 406 } 407 408 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages) 409 { 410 bpf_uncharge_memlock(map->memory.user, pages); 411 map->memory.pages -= pages; 412 } 413 414 static int bpf_map_alloc_id(struct bpf_map *map) 415 { 416 int id; 417 418 idr_preload(GFP_KERNEL); 419 spin_lock_bh(&map_idr_lock); 420 id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC); 421 if (id > 0) 422 map->id = id; 423 spin_unlock_bh(&map_idr_lock); 424 idr_preload_end(); 425 426 if (WARN_ON_ONCE(!id)) 427 return -ENOSPC; 428 429 return id > 0 ? 0 : id; 430 } 431 432 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock) 433 { 434 unsigned long flags; 435 436 /* Offloaded maps are removed from the IDR store when their device 437 * disappears - even if someone holds an fd to them they are unusable, 438 * the memory is gone, all ops will fail; they are simply waiting for 439 * refcnt to drop to be freed. 440 */ 441 if (!map->id) 442 return; 443 444 if (do_idr_lock) 445 spin_lock_irqsave(&map_idr_lock, flags); 446 else 447 __acquire(&map_idr_lock); 448 449 idr_remove(&map_idr, map->id); 450 map->id = 0; 451 452 if (do_idr_lock) 453 spin_unlock_irqrestore(&map_idr_lock, flags); 454 else 455 __release(&map_idr_lock); 456 } 457 458 /* called from workqueue */ 459 static void bpf_map_free_deferred(struct work_struct *work) 460 { 461 struct bpf_map *map = container_of(work, struct bpf_map, work); 462 struct bpf_map_memory mem; 463 464 bpf_map_charge_move(&mem, &map->memory); 465 security_bpf_map_free(map); 466 /* implementation dependent freeing */ 467 map->ops->map_free(map); 468 bpf_map_charge_finish(&mem); 469 } 470 471 static void bpf_map_put_uref(struct bpf_map *map) 472 { 473 if (atomic64_dec_and_test(&map->usercnt)) { 474 if (map->ops->map_release_uref) 475 map->ops->map_release_uref(map); 476 } 477 } 478 479 /* decrement map refcnt and schedule it for freeing via workqueue 480 * (unrelying map implementation ops->map_free() might sleep) 481 */ 482 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock) 483 { 484 if (atomic64_dec_and_test(&map->refcnt)) { 485 /* bpf_map_free_id() must be called first */ 486 bpf_map_free_id(map, do_idr_lock); 487 btf_put(map->btf); 488 INIT_WORK(&map->work, bpf_map_free_deferred); 489 schedule_work(&map->work); 490 } 491 } 492 493 void bpf_map_put(struct bpf_map *map) 494 { 495 __bpf_map_put(map, true); 496 } 497 EXPORT_SYMBOL_GPL(bpf_map_put); 498 499 void bpf_map_put_with_uref(struct bpf_map *map) 500 { 501 bpf_map_put_uref(map); 502 bpf_map_put(map); 503 } 504 505 static int bpf_map_release(struct inode *inode, struct file *filp) 506 { 507 struct bpf_map *map = filp->private_data; 508 509 if (map->ops->map_release) 510 map->ops->map_release(map, filp); 511 512 bpf_map_put_with_uref(map); 513 return 0; 514 } 515 516 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f) 517 { 518 fmode_t mode = f.file->f_mode; 519 520 /* Our file permissions may have been overridden by global 521 * map permissions facing syscall side. 522 */ 523 if (READ_ONCE(map->frozen)) 524 mode &= ~FMODE_CAN_WRITE; 525 return mode; 526 } 527 528 #ifdef CONFIG_PROC_FS 529 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp) 530 { 531 const struct bpf_map *map = filp->private_data; 532 const struct bpf_array *array; 533 u32 type = 0, jited = 0; 534 535 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) { 536 array = container_of(map, struct bpf_array, map); 537 type = array->aux->type; 538 jited = array->aux->jited; 539 } 540 541 seq_printf(m, 542 "map_type:\t%u\n" 543 "key_size:\t%u\n" 544 "value_size:\t%u\n" 545 "max_entries:\t%u\n" 546 "map_flags:\t%#x\n" 547 "memlock:\t%llu\n" 548 "map_id:\t%u\n" 549 "frozen:\t%u\n", 550 map->map_type, 551 map->key_size, 552 map->value_size, 553 map->max_entries, 554 map->map_flags, 555 map->memory.pages * 1ULL << PAGE_SHIFT, 556 map->id, 557 READ_ONCE(map->frozen)); 558 if (type) { 559 seq_printf(m, "owner_prog_type:\t%u\n", type); 560 seq_printf(m, "owner_jited:\t%u\n", jited); 561 } 562 } 563 #endif 564 565 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz, 566 loff_t *ppos) 567 { 568 /* We need this handler such that alloc_file() enables 569 * f_mode with FMODE_CAN_READ. 570 */ 571 return -EINVAL; 572 } 573 574 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf, 575 size_t siz, loff_t *ppos) 576 { 577 /* We need this handler such that alloc_file() enables 578 * f_mode with FMODE_CAN_WRITE. 579 */ 580 return -EINVAL; 581 } 582 583 /* called for any extra memory-mapped regions (except initial) */ 584 static void bpf_map_mmap_open(struct vm_area_struct *vma) 585 { 586 struct bpf_map *map = vma->vm_file->private_data; 587 588 bpf_map_inc_with_uref(map); 589 590 if (vma->vm_flags & VM_WRITE) { 591 mutex_lock(&map->freeze_mutex); 592 map->writecnt++; 593 mutex_unlock(&map->freeze_mutex); 594 } 595 } 596 597 /* called for all unmapped memory region (including initial) */ 598 static void bpf_map_mmap_close(struct vm_area_struct *vma) 599 { 600 struct bpf_map *map = vma->vm_file->private_data; 601 602 if (vma->vm_flags & VM_WRITE) { 603 mutex_lock(&map->freeze_mutex); 604 map->writecnt--; 605 mutex_unlock(&map->freeze_mutex); 606 } 607 608 bpf_map_put_with_uref(map); 609 } 610 611 static const struct vm_operations_struct bpf_map_default_vmops = { 612 .open = bpf_map_mmap_open, 613 .close = bpf_map_mmap_close, 614 }; 615 616 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma) 617 { 618 struct bpf_map *map = filp->private_data; 619 int err; 620 621 if (!map->ops->map_mmap || map_value_has_spin_lock(map)) 622 return -ENOTSUPP; 623 624 if (!(vma->vm_flags & VM_SHARED)) 625 return -EINVAL; 626 627 mutex_lock(&map->freeze_mutex); 628 629 if ((vma->vm_flags & VM_WRITE) && map->frozen) { 630 err = -EPERM; 631 goto out; 632 } 633 634 /* set default open/close callbacks */ 635 vma->vm_ops = &bpf_map_default_vmops; 636 vma->vm_private_data = map; 637 638 err = map->ops->map_mmap(map, vma); 639 if (err) 640 goto out; 641 642 bpf_map_inc_with_uref(map); 643 644 if (vma->vm_flags & VM_WRITE) 645 map->writecnt++; 646 out: 647 mutex_unlock(&map->freeze_mutex); 648 return err; 649 } 650 651 const struct file_operations bpf_map_fops = { 652 #ifdef CONFIG_PROC_FS 653 .show_fdinfo = bpf_map_show_fdinfo, 654 #endif 655 .release = bpf_map_release, 656 .read = bpf_dummy_read, 657 .write = bpf_dummy_write, 658 .mmap = bpf_map_mmap, 659 }; 660 661 int bpf_map_new_fd(struct bpf_map *map, int flags) 662 { 663 int ret; 664 665 ret = security_bpf_map(map, OPEN_FMODE(flags)); 666 if (ret < 0) 667 return ret; 668 669 return anon_inode_getfd("bpf-map", &bpf_map_fops, map, 670 flags | O_CLOEXEC); 671 } 672 673 int bpf_get_file_flag(int flags) 674 { 675 if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY)) 676 return -EINVAL; 677 if (flags & BPF_F_RDONLY) 678 return O_RDONLY; 679 if (flags & BPF_F_WRONLY) 680 return O_WRONLY; 681 return O_RDWR; 682 } 683 684 /* helper macro to check that unused fields 'union bpf_attr' are zero */ 685 #define CHECK_ATTR(CMD) \ 686 memchr_inv((void *) &attr->CMD##_LAST_FIELD + \ 687 sizeof(attr->CMD##_LAST_FIELD), 0, \ 688 sizeof(*attr) - \ 689 offsetof(union bpf_attr, CMD##_LAST_FIELD) - \ 690 sizeof(attr->CMD##_LAST_FIELD)) != NULL 691 692 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes. 693 * Return 0 on success and < 0 on error. 694 */ 695 static int bpf_obj_name_cpy(char *dst, const char *src) 696 { 697 const char *end = src + BPF_OBJ_NAME_LEN; 698 699 memset(dst, 0, BPF_OBJ_NAME_LEN); 700 /* Copy all isalnum(), '_' and '.' chars. */ 701 while (src < end && *src) { 702 if (!isalnum(*src) && 703 *src != '_' && *src != '.') 704 return -EINVAL; 705 *dst++ = *src++; 706 } 707 708 /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */ 709 if (src == end) 710 return -EINVAL; 711 712 return 0; 713 } 714 715 int map_check_no_btf(const struct bpf_map *map, 716 const struct btf *btf, 717 const struct btf_type *key_type, 718 const struct btf_type *value_type) 719 { 720 return -ENOTSUPP; 721 } 722 723 static int map_check_btf(struct bpf_map *map, const struct btf *btf, 724 u32 btf_key_id, u32 btf_value_id) 725 { 726 const struct btf_type *key_type, *value_type; 727 u32 key_size, value_size; 728 int ret = 0; 729 730 /* Some maps allow key to be unspecified. */ 731 if (btf_key_id) { 732 key_type = btf_type_id_size(btf, &btf_key_id, &key_size); 733 if (!key_type || key_size != map->key_size) 734 return -EINVAL; 735 } else { 736 key_type = btf_type_by_id(btf, 0); 737 if (!map->ops->map_check_btf) 738 return -EINVAL; 739 } 740 741 value_type = btf_type_id_size(btf, &btf_value_id, &value_size); 742 if (!value_type || value_size != map->value_size) 743 return -EINVAL; 744 745 map->spin_lock_off = btf_find_spin_lock(btf, value_type); 746 747 if (map_value_has_spin_lock(map)) { 748 if (map->map_flags & BPF_F_RDONLY_PROG) 749 return -EACCES; 750 if (map->map_type != BPF_MAP_TYPE_HASH && 751 map->map_type != BPF_MAP_TYPE_ARRAY && 752 map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && 753 map->map_type != BPF_MAP_TYPE_SK_STORAGE) 754 return -ENOTSUPP; 755 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) > 756 map->value_size) { 757 WARN_ONCE(1, 758 "verifier bug spin_lock_off %d value_size %d\n", 759 map->spin_lock_off, map->value_size); 760 return -EFAULT; 761 } 762 } 763 764 if (map->ops->map_check_btf) 765 ret = map->ops->map_check_btf(map, btf, key_type, value_type); 766 767 return ret; 768 } 769 770 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id 771 /* called via syscall */ 772 static int map_create(union bpf_attr *attr) 773 { 774 int numa_node = bpf_map_attr_numa_node(attr); 775 struct bpf_map_memory mem; 776 struct bpf_map *map; 777 int f_flags; 778 int err; 779 780 err = CHECK_ATTR(BPF_MAP_CREATE); 781 if (err) 782 return -EINVAL; 783 784 if (attr->btf_vmlinux_value_type_id) { 785 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS || 786 attr->btf_key_type_id || attr->btf_value_type_id) 787 return -EINVAL; 788 } else if (attr->btf_key_type_id && !attr->btf_value_type_id) { 789 return -EINVAL; 790 } 791 792 f_flags = bpf_get_file_flag(attr->map_flags); 793 if (f_flags < 0) 794 return f_flags; 795 796 if (numa_node != NUMA_NO_NODE && 797 ((unsigned int)numa_node >= nr_node_ids || 798 !node_online(numa_node))) 799 return -EINVAL; 800 801 /* find map type and init map: hashtable vs rbtree vs bloom vs ... */ 802 map = find_and_alloc_map(attr); 803 if (IS_ERR(map)) 804 return PTR_ERR(map); 805 806 err = bpf_obj_name_cpy(map->name, attr->map_name); 807 if (err) 808 goto free_map; 809 810 atomic64_set(&map->refcnt, 1); 811 atomic64_set(&map->usercnt, 1); 812 mutex_init(&map->freeze_mutex); 813 814 map->spin_lock_off = -EINVAL; 815 if (attr->btf_key_type_id || attr->btf_value_type_id || 816 /* Even the map's value is a kernel's struct, 817 * the bpf_prog.o must have BTF to begin with 818 * to figure out the corresponding kernel's 819 * counter part. Thus, attr->btf_fd has 820 * to be valid also. 821 */ 822 attr->btf_vmlinux_value_type_id) { 823 struct btf *btf; 824 825 btf = btf_get_by_fd(attr->btf_fd); 826 if (IS_ERR(btf)) { 827 err = PTR_ERR(btf); 828 goto free_map; 829 } 830 map->btf = btf; 831 832 if (attr->btf_value_type_id) { 833 err = map_check_btf(map, btf, attr->btf_key_type_id, 834 attr->btf_value_type_id); 835 if (err) 836 goto free_map; 837 } 838 839 map->btf_key_type_id = attr->btf_key_type_id; 840 map->btf_value_type_id = attr->btf_value_type_id; 841 map->btf_vmlinux_value_type_id = 842 attr->btf_vmlinux_value_type_id; 843 } 844 845 err = security_bpf_map_alloc(map); 846 if (err) 847 goto free_map; 848 849 err = bpf_map_alloc_id(map); 850 if (err) 851 goto free_map_sec; 852 853 err = bpf_map_new_fd(map, f_flags); 854 if (err < 0) { 855 /* failed to allocate fd. 856 * bpf_map_put_with_uref() is needed because the above 857 * bpf_map_alloc_id() has published the map 858 * to the userspace and the userspace may 859 * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID. 860 */ 861 bpf_map_put_with_uref(map); 862 return err; 863 } 864 865 return err; 866 867 free_map_sec: 868 security_bpf_map_free(map); 869 free_map: 870 btf_put(map->btf); 871 bpf_map_charge_move(&mem, &map->memory); 872 map->ops->map_free(map); 873 bpf_map_charge_finish(&mem); 874 return err; 875 } 876 877 /* if error is returned, fd is released. 878 * On success caller should complete fd access with matching fdput() 879 */ 880 struct bpf_map *__bpf_map_get(struct fd f) 881 { 882 if (!f.file) 883 return ERR_PTR(-EBADF); 884 if (f.file->f_op != &bpf_map_fops) { 885 fdput(f); 886 return ERR_PTR(-EINVAL); 887 } 888 889 return f.file->private_data; 890 } 891 892 void bpf_map_inc(struct bpf_map *map) 893 { 894 atomic64_inc(&map->refcnt); 895 } 896 EXPORT_SYMBOL_GPL(bpf_map_inc); 897 898 void bpf_map_inc_with_uref(struct bpf_map *map) 899 { 900 atomic64_inc(&map->refcnt); 901 atomic64_inc(&map->usercnt); 902 } 903 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref); 904 905 struct bpf_map *bpf_map_get(u32 ufd) 906 { 907 struct fd f = fdget(ufd); 908 struct bpf_map *map; 909 910 map = __bpf_map_get(f); 911 if (IS_ERR(map)) 912 return map; 913 914 bpf_map_inc(map); 915 fdput(f); 916 917 return map; 918 } 919 920 struct bpf_map *bpf_map_get_with_uref(u32 ufd) 921 { 922 struct fd f = fdget(ufd); 923 struct bpf_map *map; 924 925 map = __bpf_map_get(f); 926 if (IS_ERR(map)) 927 return map; 928 929 bpf_map_inc_with_uref(map); 930 fdput(f); 931 932 return map; 933 } 934 935 /* map_idr_lock should have been held */ 936 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref) 937 { 938 int refold; 939 940 refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0); 941 if (!refold) 942 return ERR_PTR(-ENOENT); 943 if (uref) 944 atomic64_inc(&map->usercnt); 945 946 return map; 947 } 948 949 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map) 950 { 951 spin_lock_bh(&map_idr_lock); 952 map = __bpf_map_inc_not_zero(map, false); 953 spin_unlock_bh(&map_idr_lock); 954 955 return map; 956 } 957 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero); 958 959 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value) 960 { 961 return -ENOTSUPP; 962 } 963 964 static void *__bpf_copy_key(void __user *ukey, u64 key_size) 965 { 966 if (key_size) 967 return memdup_user(ukey, key_size); 968 969 if (ukey) 970 return ERR_PTR(-EINVAL); 971 972 return NULL; 973 } 974 975 /* last field in 'union bpf_attr' used by this command */ 976 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags 977 978 static int map_lookup_elem(union bpf_attr *attr) 979 { 980 void __user *ukey = u64_to_user_ptr(attr->key); 981 void __user *uvalue = u64_to_user_ptr(attr->value); 982 int ufd = attr->map_fd; 983 struct bpf_map *map; 984 void *key, *value; 985 u32 value_size; 986 struct fd f; 987 int err; 988 989 if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM)) 990 return -EINVAL; 991 992 if (attr->flags & ~BPF_F_LOCK) 993 return -EINVAL; 994 995 f = fdget(ufd); 996 map = __bpf_map_get(f); 997 if (IS_ERR(map)) 998 return PTR_ERR(map); 999 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 1000 err = -EPERM; 1001 goto err_put; 1002 } 1003 1004 if ((attr->flags & BPF_F_LOCK) && 1005 !map_value_has_spin_lock(map)) { 1006 err = -EINVAL; 1007 goto err_put; 1008 } 1009 1010 key = __bpf_copy_key(ukey, map->key_size); 1011 if (IS_ERR(key)) { 1012 err = PTR_ERR(key); 1013 goto err_put; 1014 } 1015 1016 value_size = bpf_map_value_size(map); 1017 1018 err = -ENOMEM; 1019 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1020 if (!value) 1021 goto free_key; 1022 1023 err = bpf_map_copy_value(map, key, value, attr->flags); 1024 if (err) 1025 goto free_value; 1026 1027 err = -EFAULT; 1028 if (copy_to_user(uvalue, value, value_size) != 0) 1029 goto free_value; 1030 1031 err = 0; 1032 1033 free_value: 1034 kfree(value); 1035 free_key: 1036 kfree(key); 1037 err_put: 1038 fdput(f); 1039 return err; 1040 } 1041 1042 1043 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags 1044 1045 static int map_update_elem(union bpf_attr *attr) 1046 { 1047 void __user *ukey = u64_to_user_ptr(attr->key); 1048 void __user *uvalue = u64_to_user_ptr(attr->value); 1049 int ufd = attr->map_fd; 1050 struct bpf_map *map; 1051 void *key, *value; 1052 u32 value_size; 1053 struct fd f; 1054 int err; 1055 1056 if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM)) 1057 return -EINVAL; 1058 1059 f = fdget(ufd); 1060 map = __bpf_map_get(f); 1061 if (IS_ERR(map)) 1062 return PTR_ERR(map); 1063 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1064 err = -EPERM; 1065 goto err_put; 1066 } 1067 1068 if ((attr->flags & BPF_F_LOCK) && 1069 !map_value_has_spin_lock(map)) { 1070 err = -EINVAL; 1071 goto err_put; 1072 } 1073 1074 key = __bpf_copy_key(ukey, map->key_size); 1075 if (IS_ERR(key)) { 1076 err = PTR_ERR(key); 1077 goto err_put; 1078 } 1079 1080 if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH || 1081 map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH || 1082 map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY || 1083 map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) 1084 value_size = round_up(map->value_size, 8) * num_possible_cpus(); 1085 else 1086 value_size = map->value_size; 1087 1088 err = -ENOMEM; 1089 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1090 if (!value) 1091 goto free_key; 1092 1093 err = -EFAULT; 1094 if (copy_from_user(value, uvalue, value_size) != 0) 1095 goto free_value; 1096 1097 err = bpf_map_update_value(map, f, key, value, attr->flags); 1098 1099 free_value: 1100 kfree(value); 1101 free_key: 1102 kfree(key); 1103 err_put: 1104 fdput(f); 1105 return err; 1106 } 1107 1108 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key 1109 1110 static int map_delete_elem(union bpf_attr *attr) 1111 { 1112 void __user *ukey = u64_to_user_ptr(attr->key); 1113 int ufd = attr->map_fd; 1114 struct bpf_map *map; 1115 struct fd f; 1116 void *key; 1117 int err; 1118 1119 if (CHECK_ATTR(BPF_MAP_DELETE_ELEM)) 1120 return -EINVAL; 1121 1122 f = fdget(ufd); 1123 map = __bpf_map_get(f); 1124 if (IS_ERR(map)) 1125 return PTR_ERR(map); 1126 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1127 err = -EPERM; 1128 goto err_put; 1129 } 1130 1131 key = __bpf_copy_key(ukey, map->key_size); 1132 if (IS_ERR(key)) { 1133 err = PTR_ERR(key); 1134 goto err_put; 1135 } 1136 1137 if (bpf_map_is_dev_bound(map)) { 1138 err = bpf_map_offload_delete_elem(map, key); 1139 goto out; 1140 } else if (IS_FD_PROG_ARRAY(map) || 1141 map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { 1142 /* These maps require sleepable context */ 1143 err = map->ops->map_delete_elem(map, key); 1144 goto out; 1145 } 1146 1147 bpf_disable_instrumentation(); 1148 rcu_read_lock(); 1149 err = map->ops->map_delete_elem(map, key); 1150 rcu_read_unlock(); 1151 bpf_enable_instrumentation(); 1152 maybe_wait_bpf_programs(map); 1153 out: 1154 kfree(key); 1155 err_put: 1156 fdput(f); 1157 return err; 1158 } 1159 1160 /* last field in 'union bpf_attr' used by this command */ 1161 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key 1162 1163 static int map_get_next_key(union bpf_attr *attr) 1164 { 1165 void __user *ukey = u64_to_user_ptr(attr->key); 1166 void __user *unext_key = u64_to_user_ptr(attr->next_key); 1167 int ufd = attr->map_fd; 1168 struct bpf_map *map; 1169 void *key, *next_key; 1170 struct fd f; 1171 int err; 1172 1173 if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY)) 1174 return -EINVAL; 1175 1176 f = fdget(ufd); 1177 map = __bpf_map_get(f); 1178 if (IS_ERR(map)) 1179 return PTR_ERR(map); 1180 if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 1181 err = -EPERM; 1182 goto err_put; 1183 } 1184 1185 if (ukey) { 1186 key = __bpf_copy_key(ukey, map->key_size); 1187 if (IS_ERR(key)) { 1188 err = PTR_ERR(key); 1189 goto err_put; 1190 } 1191 } else { 1192 key = NULL; 1193 } 1194 1195 err = -ENOMEM; 1196 next_key = kmalloc(map->key_size, GFP_USER); 1197 if (!next_key) 1198 goto free_key; 1199 1200 if (bpf_map_is_dev_bound(map)) { 1201 err = bpf_map_offload_get_next_key(map, key, next_key); 1202 goto out; 1203 } 1204 1205 rcu_read_lock(); 1206 err = map->ops->map_get_next_key(map, key, next_key); 1207 rcu_read_unlock(); 1208 out: 1209 if (err) 1210 goto free_next_key; 1211 1212 err = -EFAULT; 1213 if (copy_to_user(unext_key, next_key, map->key_size) != 0) 1214 goto free_next_key; 1215 1216 err = 0; 1217 1218 free_next_key: 1219 kfree(next_key); 1220 free_key: 1221 kfree(key); 1222 err_put: 1223 fdput(f); 1224 return err; 1225 } 1226 1227 int generic_map_delete_batch(struct bpf_map *map, 1228 const union bpf_attr *attr, 1229 union bpf_attr __user *uattr) 1230 { 1231 void __user *keys = u64_to_user_ptr(attr->batch.keys); 1232 u32 cp, max_count; 1233 int err = 0; 1234 void *key; 1235 1236 if (attr->batch.elem_flags & ~BPF_F_LOCK) 1237 return -EINVAL; 1238 1239 if ((attr->batch.elem_flags & BPF_F_LOCK) && 1240 !map_value_has_spin_lock(map)) { 1241 return -EINVAL; 1242 } 1243 1244 max_count = attr->batch.count; 1245 if (!max_count) 1246 return 0; 1247 1248 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN); 1249 if (!key) 1250 return -ENOMEM; 1251 1252 for (cp = 0; cp < max_count; cp++) { 1253 err = -EFAULT; 1254 if (copy_from_user(key, keys + cp * map->key_size, 1255 map->key_size)) 1256 break; 1257 1258 if (bpf_map_is_dev_bound(map)) { 1259 err = bpf_map_offload_delete_elem(map, key); 1260 break; 1261 } 1262 1263 bpf_disable_instrumentation(); 1264 rcu_read_lock(); 1265 err = map->ops->map_delete_elem(map, key); 1266 rcu_read_unlock(); 1267 bpf_enable_instrumentation(); 1268 maybe_wait_bpf_programs(map); 1269 if (err) 1270 break; 1271 } 1272 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp))) 1273 err = -EFAULT; 1274 1275 kfree(key); 1276 return err; 1277 } 1278 1279 int generic_map_update_batch(struct bpf_map *map, 1280 const union bpf_attr *attr, 1281 union bpf_attr __user *uattr) 1282 { 1283 void __user *values = u64_to_user_ptr(attr->batch.values); 1284 void __user *keys = u64_to_user_ptr(attr->batch.keys); 1285 u32 value_size, cp, max_count; 1286 int ufd = attr->map_fd; 1287 void *key, *value; 1288 struct fd f; 1289 int err = 0; 1290 1291 f = fdget(ufd); 1292 if (attr->batch.elem_flags & ~BPF_F_LOCK) 1293 return -EINVAL; 1294 1295 if ((attr->batch.elem_flags & BPF_F_LOCK) && 1296 !map_value_has_spin_lock(map)) { 1297 return -EINVAL; 1298 } 1299 1300 value_size = bpf_map_value_size(map); 1301 1302 max_count = attr->batch.count; 1303 if (!max_count) 1304 return 0; 1305 1306 key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN); 1307 if (!key) 1308 return -ENOMEM; 1309 1310 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1311 if (!value) { 1312 kfree(key); 1313 return -ENOMEM; 1314 } 1315 1316 for (cp = 0; cp < max_count; cp++) { 1317 err = -EFAULT; 1318 if (copy_from_user(key, keys + cp * map->key_size, 1319 map->key_size) || 1320 copy_from_user(value, values + cp * value_size, value_size)) 1321 break; 1322 1323 err = bpf_map_update_value(map, f, key, value, 1324 attr->batch.elem_flags); 1325 1326 if (err) 1327 break; 1328 } 1329 1330 if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp))) 1331 err = -EFAULT; 1332 1333 kfree(value); 1334 kfree(key); 1335 return err; 1336 } 1337 1338 #define MAP_LOOKUP_RETRIES 3 1339 1340 int generic_map_lookup_batch(struct bpf_map *map, 1341 const union bpf_attr *attr, 1342 union bpf_attr __user *uattr) 1343 { 1344 void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch); 1345 void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch); 1346 void __user *values = u64_to_user_ptr(attr->batch.values); 1347 void __user *keys = u64_to_user_ptr(attr->batch.keys); 1348 void *buf, *buf_prevkey, *prev_key, *key, *value; 1349 int err, retry = MAP_LOOKUP_RETRIES; 1350 u32 value_size, cp, max_count; 1351 1352 if (attr->batch.elem_flags & ~BPF_F_LOCK) 1353 return -EINVAL; 1354 1355 if ((attr->batch.elem_flags & BPF_F_LOCK) && 1356 !map_value_has_spin_lock(map)) 1357 return -EINVAL; 1358 1359 value_size = bpf_map_value_size(map); 1360 1361 max_count = attr->batch.count; 1362 if (!max_count) 1363 return 0; 1364 1365 if (put_user(0, &uattr->batch.count)) 1366 return -EFAULT; 1367 1368 buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN); 1369 if (!buf_prevkey) 1370 return -ENOMEM; 1371 1372 buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN); 1373 if (!buf) { 1374 kvfree(buf_prevkey); 1375 return -ENOMEM; 1376 } 1377 1378 err = -EFAULT; 1379 prev_key = NULL; 1380 if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size)) 1381 goto free_buf; 1382 key = buf; 1383 value = key + map->key_size; 1384 if (ubatch) 1385 prev_key = buf_prevkey; 1386 1387 for (cp = 0; cp < max_count;) { 1388 rcu_read_lock(); 1389 err = map->ops->map_get_next_key(map, prev_key, key); 1390 rcu_read_unlock(); 1391 if (err) 1392 break; 1393 err = bpf_map_copy_value(map, key, value, 1394 attr->batch.elem_flags); 1395 1396 if (err == -ENOENT) { 1397 if (retry) { 1398 retry--; 1399 continue; 1400 } 1401 err = -EINTR; 1402 break; 1403 } 1404 1405 if (err) 1406 goto free_buf; 1407 1408 if (copy_to_user(keys + cp * map->key_size, key, 1409 map->key_size)) { 1410 err = -EFAULT; 1411 goto free_buf; 1412 } 1413 if (copy_to_user(values + cp * value_size, value, value_size)) { 1414 err = -EFAULT; 1415 goto free_buf; 1416 } 1417 1418 if (!prev_key) 1419 prev_key = buf_prevkey; 1420 1421 swap(prev_key, key); 1422 retry = MAP_LOOKUP_RETRIES; 1423 cp++; 1424 } 1425 1426 if (err == -EFAULT) 1427 goto free_buf; 1428 1429 if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) || 1430 (cp && copy_to_user(uobatch, prev_key, map->key_size)))) 1431 err = -EFAULT; 1432 1433 free_buf: 1434 kfree(buf_prevkey); 1435 kfree(buf); 1436 return err; 1437 } 1438 1439 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value 1440 1441 static int map_lookup_and_delete_elem(union bpf_attr *attr) 1442 { 1443 void __user *ukey = u64_to_user_ptr(attr->key); 1444 void __user *uvalue = u64_to_user_ptr(attr->value); 1445 int ufd = attr->map_fd; 1446 struct bpf_map *map; 1447 void *key, *value; 1448 u32 value_size; 1449 struct fd f; 1450 int err; 1451 1452 if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM)) 1453 return -EINVAL; 1454 1455 f = fdget(ufd); 1456 map = __bpf_map_get(f); 1457 if (IS_ERR(map)) 1458 return PTR_ERR(map); 1459 if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 1460 err = -EPERM; 1461 goto err_put; 1462 } 1463 1464 key = __bpf_copy_key(ukey, map->key_size); 1465 if (IS_ERR(key)) { 1466 err = PTR_ERR(key); 1467 goto err_put; 1468 } 1469 1470 value_size = map->value_size; 1471 1472 err = -ENOMEM; 1473 value = kmalloc(value_size, GFP_USER | __GFP_NOWARN); 1474 if (!value) 1475 goto free_key; 1476 1477 if (map->map_type == BPF_MAP_TYPE_QUEUE || 1478 map->map_type == BPF_MAP_TYPE_STACK) { 1479 err = map->ops->map_pop_elem(map, value); 1480 } else { 1481 err = -ENOTSUPP; 1482 } 1483 1484 if (err) 1485 goto free_value; 1486 1487 if (copy_to_user(uvalue, value, value_size) != 0) 1488 goto free_value; 1489 1490 err = 0; 1491 1492 free_value: 1493 kfree(value); 1494 free_key: 1495 kfree(key); 1496 err_put: 1497 fdput(f); 1498 return err; 1499 } 1500 1501 #define BPF_MAP_FREEZE_LAST_FIELD map_fd 1502 1503 static int map_freeze(const union bpf_attr *attr) 1504 { 1505 int err = 0, ufd = attr->map_fd; 1506 struct bpf_map *map; 1507 struct fd f; 1508 1509 if (CHECK_ATTR(BPF_MAP_FREEZE)) 1510 return -EINVAL; 1511 1512 f = fdget(ufd); 1513 map = __bpf_map_get(f); 1514 if (IS_ERR(map)) 1515 return PTR_ERR(map); 1516 1517 mutex_lock(&map->freeze_mutex); 1518 1519 if (map->writecnt) { 1520 err = -EBUSY; 1521 goto err_put; 1522 } 1523 if (READ_ONCE(map->frozen)) { 1524 err = -EBUSY; 1525 goto err_put; 1526 } 1527 if (!capable(CAP_SYS_ADMIN)) { 1528 err = -EPERM; 1529 goto err_put; 1530 } 1531 1532 WRITE_ONCE(map->frozen, true); 1533 err_put: 1534 mutex_unlock(&map->freeze_mutex); 1535 fdput(f); 1536 return err; 1537 } 1538 1539 static const struct bpf_prog_ops * const bpf_prog_types[] = { 1540 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ 1541 [_id] = & _name ## _prog_ops, 1542 #define BPF_MAP_TYPE(_id, _ops) 1543 #include <linux/bpf_types.h> 1544 #undef BPF_PROG_TYPE 1545 #undef BPF_MAP_TYPE 1546 }; 1547 1548 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog) 1549 { 1550 const struct bpf_prog_ops *ops; 1551 1552 if (type >= ARRAY_SIZE(bpf_prog_types)) 1553 return -EINVAL; 1554 type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types)); 1555 ops = bpf_prog_types[type]; 1556 if (!ops) 1557 return -EINVAL; 1558 1559 if (!bpf_prog_is_dev_bound(prog->aux)) 1560 prog->aux->ops = ops; 1561 else 1562 prog->aux->ops = &bpf_offload_prog_ops; 1563 prog->type = type; 1564 return 0; 1565 } 1566 1567 enum bpf_audit { 1568 BPF_AUDIT_LOAD, 1569 BPF_AUDIT_UNLOAD, 1570 BPF_AUDIT_MAX, 1571 }; 1572 1573 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = { 1574 [BPF_AUDIT_LOAD] = "LOAD", 1575 [BPF_AUDIT_UNLOAD] = "UNLOAD", 1576 }; 1577 1578 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op) 1579 { 1580 struct audit_context *ctx = NULL; 1581 struct audit_buffer *ab; 1582 1583 if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX)) 1584 return; 1585 if (audit_enabled == AUDIT_OFF) 1586 return; 1587 if (op == BPF_AUDIT_LOAD) 1588 ctx = audit_context(); 1589 ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF); 1590 if (unlikely(!ab)) 1591 return; 1592 audit_log_format(ab, "prog-id=%u op=%s", 1593 prog->aux->id, bpf_audit_str[op]); 1594 audit_log_end(ab); 1595 } 1596 1597 int __bpf_prog_charge(struct user_struct *user, u32 pages) 1598 { 1599 unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 1600 unsigned long user_bufs; 1601 1602 if (user) { 1603 user_bufs = atomic_long_add_return(pages, &user->locked_vm); 1604 if (user_bufs > memlock_limit) { 1605 atomic_long_sub(pages, &user->locked_vm); 1606 return -EPERM; 1607 } 1608 } 1609 1610 return 0; 1611 } 1612 1613 void __bpf_prog_uncharge(struct user_struct *user, u32 pages) 1614 { 1615 if (user) 1616 atomic_long_sub(pages, &user->locked_vm); 1617 } 1618 1619 static int bpf_prog_charge_memlock(struct bpf_prog *prog) 1620 { 1621 struct user_struct *user = get_current_user(); 1622 int ret; 1623 1624 ret = __bpf_prog_charge(user, prog->pages); 1625 if (ret) { 1626 free_uid(user); 1627 return ret; 1628 } 1629 1630 prog->aux->user = user; 1631 return 0; 1632 } 1633 1634 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog) 1635 { 1636 struct user_struct *user = prog->aux->user; 1637 1638 __bpf_prog_uncharge(user, prog->pages); 1639 free_uid(user); 1640 } 1641 1642 static int bpf_prog_alloc_id(struct bpf_prog *prog) 1643 { 1644 int id; 1645 1646 idr_preload(GFP_KERNEL); 1647 spin_lock_bh(&prog_idr_lock); 1648 id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC); 1649 if (id > 0) 1650 prog->aux->id = id; 1651 spin_unlock_bh(&prog_idr_lock); 1652 idr_preload_end(); 1653 1654 /* id is in [1, INT_MAX) */ 1655 if (WARN_ON_ONCE(!id)) 1656 return -ENOSPC; 1657 1658 return id > 0 ? 0 : id; 1659 } 1660 1661 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock) 1662 { 1663 /* cBPF to eBPF migrations are currently not in the idr store. 1664 * Offloaded programs are removed from the store when their device 1665 * disappears - even if someone grabs an fd to them they are unusable, 1666 * simply waiting for refcnt to drop to be freed. 1667 */ 1668 if (!prog->aux->id) 1669 return; 1670 1671 if (do_idr_lock) 1672 spin_lock_bh(&prog_idr_lock); 1673 else 1674 __acquire(&prog_idr_lock); 1675 1676 idr_remove(&prog_idr, prog->aux->id); 1677 prog->aux->id = 0; 1678 1679 if (do_idr_lock) 1680 spin_unlock_bh(&prog_idr_lock); 1681 else 1682 __release(&prog_idr_lock); 1683 } 1684 1685 static void __bpf_prog_put_rcu(struct rcu_head *rcu) 1686 { 1687 struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu); 1688 1689 kvfree(aux->func_info); 1690 kfree(aux->func_info_aux); 1691 bpf_prog_uncharge_memlock(aux->prog); 1692 security_bpf_prog_free(aux); 1693 bpf_prog_free(aux->prog); 1694 } 1695 1696 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred) 1697 { 1698 bpf_prog_kallsyms_del_all(prog); 1699 btf_put(prog->aux->btf); 1700 bpf_prog_free_linfo(prog); 1701 1702 if (deferred) 1703 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu); 1704 else 1705 __bpf_prog_put_rcu(&prog->aux->rcu); 1706 } 1707 1708 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock) 1709 { 1710 if (atomic64_dec_and_test(&prog->aux->refcnt)) { 1711 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0); 1712 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD); 1713 /* bpf_prog_free_id() must be called first */ 1714 bpf_prog_free_id(prog, do_idr_lock); 1715 __bpf_prog_put_noref(prog, true); 1716 } 1717 } 1718 1719 void bpf_prog_put(struct bpf_prog *prog) 1720 { 1721 __bpf_prog_put(prog, true); 1722 } 1723 EXPORT_SYMBOL_GPL(bpf_prog_put); 1724 1725 static int bpf_prog_release(struct inode *inode, struct file *filp) 1726 { 1727 struct bpf_prog *prog = filp->private_data; 1728 1729 bpf_prog_put(prog); 1730 return 0; 1731 } 1732 1733 static void bpf_prog_get_stats(const struct bpf_prog *prog, 1734 struct bpf_prog_stats *stats) 1735 { 1736 u64 nsecs = 0, cnt = 0; 1737 int cpu; 1738 1739 for_each_possible_cpu(cpu) { 1740 const struct bpf_prog_stats *st; 1741 unsigned int start; 1742 u64 tnsecs, tcnt; 1743 1744 st = per_cpu_ptr(prog->aux->stats, cpu); 1745 do { 1746 start = u64_stats_fetch_begin_irq(&st->syncp); 1747 tnsecs = st->nsecs; 1748 tcnt = st->cnt; 1749 } while (u64_stats_fetch_retry_irq(&st->syncp, start)); 1750 nsecs += tnsecs; 1751 cnt += tcnt; 1752 } 1753 stats->nsecs = nsecs; 1754 stats->cnt = cnt; 1755 } 1756 1757 #ifdef CONFIG_PROC_FS 1758 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp) 1759 { 1760 const struct bpf_prog *prog = filp->private_data; 1761 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 1762 struct bpf_prog_stats stats; 1763 1764 bpf_prog_get_stats(prog, &stats); 1765 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 1766 seq_printf(m, 1767 "prog_type:\t%u\n" 1768 "prog_jited:\t%u\n" 1769 "prog_tag:\t%s\n" 1770 "memlock:\t%llu\n" 1771 "prog_id:\t%u\n" 1772 "run_time_ns:\t%llu\n" 1773 "run_cnt:\t%llu\n", 1774 prog->type, 1775 prog->jited, 1776 prog_tag, 1777 prog->pages * 1ULL << PAGE_SHIFT, 1778 prog->aux->id, 1779 stats.nsecs, 1780 stats.cnt); 1781 } 1782 #endif 1783 1784 const struct file_operations bpf_prog_fops = { 1785 #ifdef CONFIG_PROC_FS 1786 .show_fdinfo = bpf_prog_show_fdinfo, 1787 #endif 1788 .release = bpf_prog_release, 1789 .read = bpf_dummy_read, 1790 .write = bpf_dummy_write, 1791 }; 1792 1793 int bpf_prog_new_fd(struct bpf_prog *prog) 1794 { 1795 int ret; 1796 1797 ret = security_bpf_prog(prog); 1798 if (ret < 0) 1799 return ret; 1800 1801 return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, 1802 O_RDWR | O_CLOEXEC); 1803 } 1804 1805 static struct bpf_prog *____bpf_prog_get(struct fd f) 1806 { 1807 if (!f.file) 1808 return ERR_PTR(-EBADF); 1809 if (f.file->f_op != &bpf_prog_fops) { 1810 fdput(f); 1811 return ERR_PTR(-EINVAL); 1812 } 1813 1814 return f.file->private_data; 1815 } 1816 1817 void bpf_prog_add(struct bpf_prog *prog, int i) 1818 { 1819 atomic64_add(i, &prog->aux->refcnt); 1820 } 1821 EXPORT_SYMBOL_GPL(bpf_prog_add); 1822 1823 void bpf_prog_sub(struct bpf_prog *prog, int i) 1824 { 1825 /* Only to be used for undoing previous bpf_prog_add() in some 1826 * error path. We still know that another entity in our call 1827 * path holds a reference to the program, thus atomic_sub() can 1828 * be safely used in such cases! 1829 */ 1830 WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0); 1831 } 1832 EXPORT_SYMBOL_GPL(bpf_prog_sub); 1833 1834 void bpf_prog_inc(struct bpf_prog *prog) 1835 { 1836 atomic64_inc(&prog->aux->refcnt); 1837 } 1838 EXPORT_SYMBOL_GPL(bpf_prog_inc); 1839 1840 /* prog_idr_lock should have been held */ 1841 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog) 1842 { 1843 int refold; 1844 1845 refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0); 1846 1847 if (!refold) 1848 return ERR_PTR(-ENOENT); 1849 1850 return prog; 1851 } 1852 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero); 1853 1854 bool bpf_prog_get_ok(struct bpf_prog *prog, 1855 enum bpf_prog_type *attach_type, bool attach_drv) 1856 { 1857 /* not an attachment, just a refcount inc, always allow */ 1858 if (!attach_type) 1859 return true; 1860 1861 if (prog->type != *attach_type) 1862 return false; 1863 if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv) 1864 return false; 1865 1866 return true; 1867 } 1868 1869 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type, 1870 bool attach_drv) 1871 { 1872 struct fd f = fdget(ufd); 1873 struct bpf_prog *prog; 1874 1875 prog = ____bpf_prog_get(f); 1876 if (IS_ERR(prog)) 1877 return prog; 1878 if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) { 1879 prog = ERR_PTR(-EINVAL); 1880 goto out; 1881 } 1882 1883 bpf_prog_inc(prog); 1884 out: 1885 fdput(f); 1886 return prog; 1887 } 1888 1889 struct bpf_prog *bpf_prog_get(u32 ufd) 1890 { 1891 return __bpf_prog_get(ufd, NULL, false); 1892 } 1893 1894 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type, 1895 bool attach_drv) 1896 { 1897 return __bpf_prog_get(ufd, &type, attach_drv); 1898 } 1899 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev); 1900 1901 /* Initially all BPF programs could be loaded w/o specifying 1902 * expected_attach_type. Later for some of them specifying expected_attach_type 1903 * at load time became required so that program could be validated properly. 1904 * Programs of types that are allowed to be loaded both w/ and w/o (for 1905 * backward compatibility) expected_attach_type, should have the default attach 1906 * type assigned to expected_attach_type for the latter case, so that it can be 1907 * validated later at attach time. 1908 * 1909 * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if 1910 * prog type requires it but has some attach types that have to be backward 1911 * compatible. 1912 */ 1913 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr) 1914 { 1915 switch (attr->prog_type) { 1916 case BPF_PROG_TYPE_CGROUP_SOCK: 1917 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't 1918 * exist so checking for non-zero is the way to go here. 1919 */ 1920 if (!attr->expected_attach_type) 1921 attr->expected_attach_type = 1922 BPF_CGROUP_INET_SOCK_CREATE; 1923 break; 1924 } 1925 } 1926 1927 static int 1928 bpf_prog_load_check_attach(enum bpf_prog_type prog_type, 1929 enum bpf_attach_type expected_attach_type, 1930 u32 btf_id, u32 prog_fd) 1931 { 1932 if (btf_id) { 1933 if (btf_id > BTF_MAX_TYPE) 1934 return -EINVAL; 1935 1936 switch (prog_type) { 1937 case BPF_PROG_TYPE_TRACING: 1938 case BPF_PROG_TYPE_STRUCT_OPS: 1939 case BPF_PROG_TYPE_EXT: 1940 break; 1941 default: 1942 return -EINVAL; 1943 } 1944 } 1945 1946 if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING && 1947 prog_type != BPF_PROG_TYPE_EXT) 1948 return -EINVAL; 1949 1950 switch (prog_type) { 1951 case BPF_PROG_TYPE_CGROUP_SOCK: 1952 switch (expected_attach_type) { 1953 case BPF_CGROUP_INET_SOCK_CREATE: 1954 case BPF_CGROUP_INET4_POST_BIND: 1955 case BPF_CGROUP_INET6_POST_BIND: 1956 return 0; 1957 default: 1958 return -EINVAL; 1959 } 1960 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 1961 switch (expected_attach_type) { 1962 case BPF_CGROUP_INET4_BIND: 1963 case BPF_CGROUP_INET6_BIND: 1964 case BPF_CGROUP_INET4_CONNECT: 1965 case BPF_CGROUP_INET6_CONNECT: 1966 case BPF_CGROUP_UDP4_SENDMSG: 1967 case BPF_CGROUP_UDP6_SENDMSG: 1968 case BPF_CGROUP_UDP4_RECVMSG: 1969 case BPF_CGROUP_UDP6_RECVMSG: 1970 return 0; 1971 default: 1972 return -EINVAL; 1973 } 1974 case BPF_PROG_TYPE_CGROUP_SKB: 1975 switch (expected_attach_type) { 1976 case BPF_CGROUP_INET_INGRESS: 1977 case BPF_CGROUP_INET_EGRESS: 1978 return 0; 1979 default: 1980 return -EINVAL; 1981 } 1982 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 1983 switch (expected_attach_type) { 1984 case BPF_CGROUP_SETSOCKOPT: 1985 case BPF_CGROUP_GETSOCKOPT: 1986 return 0; 1987 default: 1988 return -EINVAL; 1989 } 1990 case BPF_PROG_TYPE_EXT: 1991 if (expected_attach_type) 1992 return -EINVAL; 1993 /* fallthrough */ 1994 default: 1995 return 0; 1996 } 1997 } 1998 1999 /* last field in 'union bpf_attr' used by this command */ 2000 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd 2001 2002 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr) 2003 { 2004 enum bpf_prog_type type = attr->prog_type; 2005 struct bpf_prog *prog; 2006 int err; 2007 char license[128]; 2008 bool is_gpl; 2009 2010 if (CHECK_ATTR(BPF_PROG_LOAD)) 2011 return -EINVAL; 2012 2013 if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | 2014 BPF_F_ANY_ALIGNMENT | 2015 BPF_F_TEST_STATE_FREQ | 2016 BPF_F_TEST_RND_HI32)) 2017 return -EINVAL; 2018 2019 if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && 2020 (attr->prog_flags & BPF_F_ANY_ALIGNMENT) && 2021 !capable(CAP_SYS_ADMIN)) 2022 return -EPERM; 2023 2024 /* copy eBPF program license from user space */ 2025 if (strncpy_from_user(license, u64_to_user_ptr(attr->license), 2026 sizeof(license) - 1) < 0) 2027 return -EFAULT; 2028 license[sizeof(license) - 1] = 0; 2029 2030 /* eBPF programs must be GPL compatible to use GPL-ed functions */ 2031 is_gpl = license_is_gpl_compatible(license); 2032 2033 if (attr->insn_cnt == 0 || 2034 attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) 2035 return -E2BIG; 2036 if (type != BPF_PROG_TYPE_SOCKET_FILTER && 2037 type != BPF_PROG_TYPE_CGROUP_SKB && 2038 !capable(CAP_SYS_ADMIN)) 2039 return -EPERM; 2040 2041 bpf_prog_load_fixup_attach_type(attr); 2042 if (bpf_prog_load_check_attach(type, attr->expected_attach_type, 2043 attr->attach_btf_id, 2044 attr->attach_prog_fd)) 2045 return -EINVAL; 2046 2047 /* plain bpf_prog allocation */ 2048 prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER); 2049 if (!prog) 2050 return -ENOMEM; 2051 2052 prog->expected_attach_type = attr->expected_attach_type; 2053 prog->aux->attach_btf_id = attr->attach_btf_id; 2054 if (attr->attach_prog_fd) { 2055 struct bpf_prog *tgt_prog; 2056 2057 tgt_prog = bpf_prog_get(attr->attach_prog_fd); 2058 if (IS_ERR(tgt_prog)) { 2059 err = PTR_ERR(tgt_prog); 2060 goto free_prog_nouncharge; 2061 } 2062 prog->aux->linked_prog = tgt_prog; 2063 } 2064 2065 prog->aux->offload_requested = !!attr->prog_ifindex; 2066 2067 err = security_bpf_prog_alloc(prog->aux); 2068 if (err) 2069 goto free_prog_nouncharge; 2070 2071 err = bpf_prog_charge_memlock(prog); 2072 if (err) 2073 goto free_prog_sec; 2074 2075 prog->len = attr->insn_cnt; 2076 2077 err = -EFAULT; 2078 if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns), 2079 bpf_prog_insn_size(prog)) != 0) 2080 goto free_prog; 2081 2082 prog->orig_prog = NULL; 2083 prog->jited = 0; 2084 2085 atomic64_set(&prog->aux->refcnt, 1); 2086 prog->gpl_compatible = is_gpl ? 1 : 0; 2087 2088 if (bpf_prog_is_dev_bound(prog->aux)) { 2089 err = bpf_prog_offload_init(prog, attr); 2090 if (err) 2091 goto free_prog; 2092 } 2093 2094 /* find program type: socket_filter vs tracing_filter */ 2095 err = find_prog_type(type, prog); 2096 if (err < 0) 2097 goto free_prog; 2098 2099 prog->aux->load_time = ktime_get_boottime_ns(); 2100 err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name); 2101 if (err) 2102 goto free_prog; 2103 2104 /* run eBPF verifier */ 2105 err = bpf_check(&prog, attr, uattr); 2106 if (err < 0) 2107 goto free_used_maps; 2108 2109 prog = bpf_prog_select_runtime(prog, &err); 2110 if (err < 0) 2111 goto free_used_maps; 2112 2113 err = bpf_prog_alloc_id(prog); 2114 if (err) 2115 goto free_used_maps; 2116 2117 /* Upon success of bpf_prog_alloc_id(), the BPF prog is 2118 * effectively publicly exposed. However, retrieving via 2119 * bpf_prog_get_fd_by_id() will take another reference, 2120 * therefore it cannot be gone underneath us. 2121 * 2122 * Only for the time /after/ successful bpf_prog_new_fd() 2123 * and before returning to userspace, we might just hold 2124 * one reference and any parallel close on that fd could 2125 * rip everything out. Hence, below notifications must 2126 * happen before bpf_prog_new_fd(). 2127 * 2128 * Also, any failure handling from this point onwards must 2129 * be using bpf_prog_put() given the program is exposed. 2130 */ 2131 bpf_prog_kallsyms_add(prog); 2132 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0); 2133 bpf_audit_prog(prog, BPF_AUDIT_LOAD); 2134 2135 err = bpf_prog_new_fd(prog); 2136 if (err < 0) 2137 bpf_prog_put(prog); 2138 return err; 2139 2140 free_used_maps: 2141 /* In case we have subprogs, we need to wait for a grace 2142 * period before we can tear down JIT memory since symbols 2143 * are already exposed under kallsyms. 2144 */ 2145 __bpf_prog_put_noref(prog, prog->aux->func_cnt); 2146 return err; 2147 free_prog: 2148 bpf_prog_uncharge_memlock(prog); 2149 free_prog_sec: 2150 security_bpf_prog_free(prog->aux); 2151 free_prog_nouncharge: 2152 bpf_prog_free(prog); 2153 return err; 2154 } 2155 2156 #define BPF_OBJ_LAST_FIELD file_flags 2157 2158 static int bpf_obj_pin(const union bpf_attr *attr) 2159 { 2160 if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0) 2161 return -EINVAL; 2162 2163 return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname)); 2164 } 2165 2166 static int bpf_obj_get(const union bpf_attr *attr) 2167 { 2168 if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 || 2169 attr->file_flags & ~BPF_OBJ_FLAG_MASK) 2170 return -EINVAL; 2171 2172 return bpf_obj_get_user(u64_to_user_ptr(attr->pathname), 2173 attr->file_flags); 2174 } 2175 2176 struct bpf_link { 2177 atomic64_t refcnt; 2178 const struct bpf_link_ops *ops; 2179 struct bpf_prog *prog; 2180 struct work_struct work; 2181 }; 2182 2183 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops, 2184 struct bpf_prog *prog) 2185 { 2186 atomic64_set(&link->refcnt, 1); 2187 link->ops = ops; 2188 link->prog = prog; 2189 } 2190 2191 void bpf_link_defunct(struct bpf_link *link) 2192 { 2193 link->prog = NULL; 2194 } 2195 2196 void bpf_link_inc(struct bpf_link *link) 2197 { 2198 atomic64_inc(&link->refcnt); 2199 } 2200 2201 /* bpf_link_free is guaranteed to be called from process context */ 2202 static void bpf_link_free(struct bpf_link *link) 2203 { 2204 if (link->prog) { 2205 /* detach BPF program, clean up used resources */ 2206 link->ops->release(link); 2207 bpf_prog_put(link->prog); 2208 } 2209 /* free bpf_link and its containing memory */ 2210 link->ops->dealloc(link); 2211 } 2212 2213 static void bpf_link_put_deferred(struct work_struct *work) 2214 { 2215 struct bpf_link *link = container_of(work, struct bpf_link, work); 2216 2217 bpf_link_free(link); 2218 } 2219 2220 /* bpf_link_put can be called from atomic context, but ensures that resources 2221 * are freed from process context 2222 */ 2223 void bpf_link_put(struct bpf_link *link) 2224 { 2225 if (!atomic64_dec_and_test(&link->refcnt)) 2226 return; 2227 2228 if (in_atomic()) { 2229 INIT_WORK(&link->work, bpf_link_put_deferred); 2230 schedule_work(&link->work); 2231 } else { 2232 bpf_link_free(link); 2233 } 2234 } 2235 2236 static int bpf_link_release(struct inode *inode, struct file *filp) 2237 { 2238 struct bpf_link *link = filp->private_data; 2239 2240 bpf_link_put(link); 2241 return 0; 2242 } 2243 2244 #ifdef CONFIG_PROC_FS 2245 static const struct bpf_link_ops bpf_raw_tp_lops; 2246 static const struct bpf_link_ops bpf_tracing_link_lops; 2247 static const struct bpf_link_ops bpf_xdp_link_lops; 2248 2249 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp) 2250 { 2251 const struct bpf_link *link = filp->private_data; 2252 const struct bpf_prog *prog = link->prog; 2253 char prog_tag[sizeof(prog->tag) * 2 + 1] = { }; 2254 const char *link_type; 2255 2256 if (link->ops == &bpf_raw_tp_lops) 2257 link_type = "raw_tracepoint"; 2258 else if (link->ops == &bpf_tracing_link_lops) 2259 link_type = "tracing"; 2260 else 2261 link_type = "unknown"; 2262 2263 bin2hex(prog_tag, prog->tag, sizeof(prog->tag)); 2264 seq_printf(m, 2265 "link_type:\t%s\n" 2266 "prog_tag:\t%s\n" 2267 "prog_id:\t%u\n", 2268 link_type, 2269 prog_tag, 2270 prog->aux->id); 2271 } 2272 #endif 2273 2274 const struct file_operations bpf_link_fops = { 2275 #ifdef CONFIG_PROC_FS 2276 .show_fdinfo = bpf_link_show_fdinfo, 2277 #endif 2278 .release = bpf_link_release, 2279 .read = bpf_dummy_read, 2280 .write = bpf_dummy_write, 2281 }; 2282 2283 int bpf_link_new_fd(struct bpf_link *link) 2284 { 2285 return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC); 2286 } 2287 2288 /* Similar to bpf_link_new_fd, create anon_inode for given bpf_link, but 2289 * instead of immediately installing fd in fdtable, just reserve it and 2290 * return. Caller then need to either install it with fd_install(fd, file) or 2291 * release with put_unused_fd(fd). 2292 * This is useful for cases when bpf_link attachment/detachment are 2293 * complicated and expensive operations and should be delayed until all the fd 2294 * reservation and anon_inode creation succeeds. 2295 */ 2296 struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd) 2297 { 2298 struct file *file; 2299 int fd; 2300 2301 fd = get_unused_fd_flags(O_CLOEXEC); 2302 if (fd < 0) 2303 return ERR_PTR(fd); 2304 2305 file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC); 2306 if (IS_ERR(file)) { 2307 put_unused_fd(fd); 2308 return file; 2309 } 2310 2311 *reserved_fd = fd; 2312 return file; 2313 } 2314 2315 struct bpf_link *bpf_link_get_from_fd(u32 ufd) 2316 { 2317 struct fd f = fdget(ufd); 2318 struct bpf_link *link; 2319 2320 if (!f.file) 2321 return ERR_PTR(-EBADF); 2322 if (f.file->f_op != &bpf_link_fops) { 2323 fdput(f); 2324 return ERR_PTR(-EINVAL); 2325 } 2326 2327 link = f.file->private_data; 2328 bpf_link_inc(link); 2329 fdput(f); 2330 2331 return link; 2332 } 2333 2334 struct bpf_tracing_link { 2335 struct bpf_link link; 2336 }; 2337 2338 static void bpf_tracing_link_release(struct bpf_link *link) 2339 { 2340 WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog)); 2341 } 2342 2343 static void bpf_tracing_link_dealloc(struct bpf_link *link) 2344 { 2345 struct bpf_tracing_link *tr_link = 2346 container_of(link, struct bpf_tracing_link, link); 2347 2348 kfree(tr_link); 2349 } 2350 2351 static const struct bpf_link_ops bpf_tracing_link_lops = { 2352 .release = bpf_tracing_link_release, 2353 .dealloc = bpf_tracing_link_dealloc, 2354 }; 2355 2356 static int bpf_tracing_prog_attach(struct bpf_prog *prog) 2357 { 2358 struct bpf_tracing_link *link; 2359 struct file *link_file; 2360 int link_fd, err; 2361 2362 if (prog->expected_attach_type != BPF_TRACE_FENTRY && 2363 prog->expected_attach_type != BPF_TRACE_FEXIT && 2364 prog->expected_attach_type != BPF_MODIFY_RETURN && 2365 prog->type != BPF_PROG_TYPE_EXT) { 2366 err = -EINVAL; 2367 goto out_put_prog; 2368 } 2369 2370 link = kzalloc(sizeof(*link), GFP_USER); 2371 if (!link) { 2372 err = -ENOMEM; 2373 goto out_put_prog; 2374 } 2375 bpf_link_init(&link->link, &bpf_tracing_link_lops, prog); 2376 2377 link_file = bpf_link_new_file(&link->link, &link_fd); 2378 if (IS_ERR(link_file)) { 2379 kfree(link); 2380 err = PTR_ERR(link_file); 2381 goto out_put_prog; 2382 } 2383 2384 err = bpf_trampoline_link_prog(prog); 2385 if (err) { 2386 bpf_link_defunct(&link->link); 2387 fput(link_file); 2388 put_unused_fd(link_fd); 2389 goto out_put_prog; 2390 } 2391 2392 fd_install(link_fd, link_file); 2393 return link_fd; 2394 2395 out_put_prog: 2396 bpf_prog_put(prog); 2397 return err; 2398 } 2399 2400 struct bpf_raw_tp_link { 2401 struct bpf_link link; 2402 struct bpf_raw_event_map *btp; 2403 }; 2404 2405 static void bpf_raw_tp_link_release(struct bpf_link *link) 2406 { 2407 struct bpf_raw_tp_link *raw_tp = 2408 container_of(link, struct bpf_raw_tp_link, link); 2409 2410 bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog); 2411 bpf_put_raw_tracepoint(raw_tp->btp); 2412 } 2413 2414 static void bpf_raw_tp_link_dealloc(struct bpf_link *link) 2415 { 2416 struct bpf_raw_tp_link *raw_tp = 2417 container_of(link, struct bpf_raw_tp_link, link); 2418 2419 kfree(raw_tp); 2420 } 2421 2422 static const struct bpf_link_ops bpf_raw_tp_lops = { 2423 .release = bpf_raw_tp_link_release, 2424 .dealloc = bpf_raw_tp_link_dealloc, 2425 }; 2426 2427 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd 2428 2429 static int bpf_raw_tracepoint_open(const union bpf_attr *attr) 2430 { 2431 struct bpf_raw_tp_link *link; 2432 struct bpf_raw_event_map *btp; 2433 struct file *link_file; 2434 struct bpf_prog *prog; 2435 const char *tp_name; 2436 char buf[128]; 2437 int link_fd, err; 2438 2439 if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN)) 2440 return -EINVAL; 2441 2442 prog = bpf_prog_get(attr->raw_tracepoint.prog_fd); 2443 if (IS_ERR(prog)) 2444 return PTR_ERR(prog); 2445 2446 if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT && 2447 prog->type != BPF_PROG_TYPE_TRACING && 2448 prog->type != BPF_PROG_TYPE_EXT && 2449 prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) { 2450 err = -EINVAL; 2451 goto out_put_prog; 2452 } 2453 2454 if (prog->type == BPF_PROG_TYPE_TRACING || 2455 prog->type == BPF_PROG_TYPE_EXT) { 2456 if (attr->raw_tracepoint.name) { 2457 /* The attach point for this category of programs 2458 * should be specified via btf_id during program load. 2459 */ 2460 err = -EINVAL; 2461 goto out_put_prog; 2462 } 2463 if (prog->expected_attach_type == BPF_TRACE_RAW_TP) 2464 tp_name = prog->aux->attach_func_name; 2465 else 2466 return bpf_tracing_prog_attach(prog); 2467 } else { 2468 if (strncpy_from_user(buf, 2469 u64_to_user_ptr(attr->raw_tracepoint.name), 2470 sizeof(buf) - 1) < 0) { 2471 err = -EFAULT; 2472 goto out_put_prog; 2473 } 2474 buf[sizeof(buf) - 1] = 0; 2475 tp_name = buf; 2476 } 2477 2478 btp = bpf_get_raw_tracepoint(tp_name); 2479 if (!btp) { 2480 err = -ENOENT; 2481 goto out_put_prog; 2482 } 2483 2484 link = kzalloc(sizeof(*link), GFP_USER); 2485 if (!link) { 2486 err = -ENOMEM; 2487 goto out_put_btp; 2488 } 2489 bpf_link_init(&link->link, &bpf_raw_tp_lops, prog); 2490 link->btp = btp; 2491 2492 link_file = bpf_link_new_file(&link->link, &link_fd); 2493 if (IS_ERR(link_file)) { 2494 kfree(link); 2495 err = PTR_ERR(link_file); 2496 goto out_put_btp; 2497 } 2498 2499 err = bpf_probe_register(link->btp, prog); 2500 if (err) { 2501 bpf_link_defunct(&link->link); 2502 fput(link_file); 2503 put_unused_fd(link_fd); 2504 goto out_put_btp; 2505 } 2506 2507 fd_install(link_fd, link_file); 2508 return link_fd; 2509 2510 out_put_btp: 2511 bpf_put_raw_tracepoint(btp); 2512 out_put_prog: 2513 bpf_prog_put(prog); 2514 return err; 2515 } 2516 2517 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog, 2518 enum bpf_attach_type attach_type) 2519 { 2520 switch (prog->type) { 2521 case BPF_PROG_TYPE_CGROUP_SOCK: 2522 case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: 2523 case BPF_PROG_TYPE_CGROUP_SOCKOPT: 2524 return attach_type == prog->expected_attach_type ? 0 : -EINVAL; 2525 case BPF_PROG_TYPE_CGROUP_SKB: 2526 return prog->enforce_expected_attach_type && 2527 prog->expected_attach_type != attach_type ? 2528 -EINVAL : 0; 2529 default: 2530 return 0; 2531 } 2532 } 2533 2534 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd 2535 2536 #define BPF_F_ATTACH_MASK \ 2537 (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE) 2538 2539 static int bpf_prog_attach(const union bpf_attr *attr) 2540 { 2541 enum bpf_prog_type ptype; 2542 struct bpf_prog *prog; 2543 int ret; 2544 2545 if (!capable(CAP_NET_ADMIN)) 2546 return -EPERM; 2547 2548 if (CHECK_ATTR(BPF_PROG_ATTACH)) 2549 return -EINVAL; 2550 2551 if (attr->attach_flags & ~BPF_F_ATTACH_MASK) 2552 return -EINVAL; 2553 2554 switch (attr->attach_type) { 2555 case BPF_CGROUP_INET_INGRESS: 2556 case BPF_CGROUP_INET_EGRESS: 2557 ptype = BPF_PROG_TYPE_CGROUP_SKB; 2558 break; 2559 case BPF_CGROUP_INET_SOCK_CREATE: 2560 case BPF_CGROUP_INET4_POST_BIND: 2561 case BPF_CGROUP_INET6_POST_BIND: 2562 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 2563 break; 2564 case BPF_CGROUP_INET4_BIND: 2565 case BPF_CGROUP_INET6_BIND: 2566 case BPF_CGROUP_INET4_CONNECT: 2567 case BPF_CGROUP_INET6_CONNECT: 2568 case BPF_CGROUP_UDP4_SENDMSG: 2569 case BPF_CGROUP_UDP6_SENDMSG: 2570 case BPF_CGROUP_UDP4_RECVMSG: 2571 case BPF_CGROUP_UDP6_RECVMSG: 2572 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 2573 break; 2574 case BPF_CGROUP_SOCK_OPS: 2575 ptype = BPF_PROG_TYPE_SOCK_OPS; 2576 break; 2577 case BPF_CGROUP_DEVICE: 2578 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 2579 break; 2580 case BPF_SK_MSG_VERDICT: 2581 ptype = BPF_PROG_TYPE_SK_MSG; 2582 break; 2583 case BPF_SK_SKB_STREAM_PARSER: 2584 case BPF_SK_SKB_STREAM_VERDICT: 2585 ptype = BPF_PROG_TYPE_SK_SKB; 2586 break; 2587 case BPF_LIRC_MODE2: 2588 ptype = BPF_PROG_TYPE_LIRC_MODE2; 2589 break; 2590 case BPF_FLOW_DISSECTOR: 2591 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR; 2592 break; 2593 case BPF_CGROUP_SYSCTL: 2594 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 2595 break; 2596 case BPF_CGROUP_GETSOCKOPT: 2597 case BPF_CGROUP_SETSOCKOPT: 2598 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 2599 break; 2600 default: 2601 return -EINVAL; 2602 } 2603 2604 prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype); 2605 if (IS_ERR(prog)) 2606 return PTR_ERR(prog); 2607 2608 if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) { 2609 bpf_prog_put(prog); 2610 return -EINVAL; 2611 } 2612 2613 switch (ptype) { 2614 case BPF_PROG_TYPE_SK_SKB: 2615 case BPF_PROG_TYPE_SK_MSG: 2616 ret = sock_map_get_from_fd(attr, prog); 2617 break; 2618 case BPF_PROG_TYPE_LIRC_MODE2: 2619 ret = lirc_prog_attach(attr, prog); 2620 break; 2621 case BPF_PROG_TYPE_FLOW_DISSECTOR: 2622 ret = skb_flow_dissector_bpf_prog_attach(attr, prog); 2623 break; 2624 default: 2625 ret = cgroup_bpf_prog_attach(attr, ptype, prog); 2626 } 2627 2628 if (ret) 2629 bpf_prog_put(prog); 2630 return ret; 2631 } 2632 2633 #define BPF_PROG_DETACH_LAST_FIELD attach_type 2634 2635 static int bpf_prog_detach(const union bpf_attr *attr) 2636 { 2637 enum bpf_prog_type ptype; 2638 2639 if (!capable(CAP_NET_ADMIN)) 2640 return -EPERM; 2641 2642 if (CHECK_ATTR(BPF_PROG_DETACH)) 2643 return -EINVAL; 2644 2645 switch (attr->attach_type) { 2646 case BPF_CGROUP_INET_INGRESS: 2647 case BPF_CGROUP_INET_EGRESS: 2648 ptype = BPF_PROG_TYPE_CGROUP_SKB; 2649 break; 2650 case BPF_CGROUP_INET_SOCK_CREATE: 2651 case BPF_CGROUP_INET4_POST_BIND: 2652 case BPF_CGROUP_INET6_POST_BIND: 2653 ptype = BPF_PROG_TYPE_CGROUP_SOCK; 2654 break; 2655 case BPF_CGROUP_INET4_BIND: 2656 case BPF_CGROUP_INET6_BIND: 2657 case BPF_CGROUP_INET4_CONNECT: 2658 case BPF_CGROUP_INET6_CONNECT: 2659 case BPF_CGROUP_UDP4_SENDMSG: 2660 case BPF_CGROUP_UDP6_SENDMSG: 2661 case BPF_CGROUP_UDP4_RECVMSG: 2662 case BPF_CGROUP_UDP6_RECVMSG: 2663 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR; 2664 break; 2665 case BPF_CGROUP_SOCK_OPS: 2666 ptype = BPF_PROG_TYPE_SOCK_OPS; 2667 break; 2668 case BPF_CGROUP_DEVICE: 2669 ptype = BPF_PROG_TYPE_CGROUP_DEVICE; 2670 break; 2671 case BPF_SK_MSG_VERDICT: 2672 return sock_map_get_from_fd(attr, NULL); 2673 case BPF_SK_SKB_STREAM_PARSER: 2674 case BPF_SK_SKB_STREAM_VERDICT: 2675 return sock_map_get_from_fd(attr, NULL); 2676 case BPF_LIRC_MODE2: 2677 return lirc_prog_detach(attr); 2678 case BPF_FLOW_DISSECTOR: 2679 return skb_flow_dissector_bpf_prog_detach(attr); 2680 case BPF_CGROUP_SYSCTL: 2681 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL; 2682 break; 2683 case BPF_CGROUP_GETSOCKOPT: 2684 case BPF_CGROUP_SETSOCKOPT: 2685 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT; 2686 break; 2687 default: 2688 return -EINVAL; 2689 } 2690 2691 return cgroup_bpf_prog_detach(attr, ptype); 2692 } 2693 2694 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt 2695 2696 static int bpf_prog_query(const union bpf_attr *attr, 2697 union bpf_attr __user *uattr) 2698 { 2699 if (!capable(CAP_NET_ADMIN)) 2700 return -EPERM; 2701 if (CHECK_ATTR(BPF_PROG_QUERY)) 2702 return -EINVAL; 2703 if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE) 2704 return -EINVAL; 2705 2706 switch (attr->query.attach_type) { 2707 case BPF_CGROUP_INET_INGRESS: 2708 case BPF_CGROUP_INET_EGRESS: 2709 case BPF_CGROUP_INET_SOCK_CREATE: 2710 case BPF_CGROUP_INET4_BIND: 2711 case BPF_CGROUP_INET6_BIND: 2712 case BPF_CGROUP_INET4_POST_BIND: 2713 case BPF_CGROUP_INET6_POST_BIND: 2714 case BPF_CGROUP_INET4_CONNECT: 2715 case BPF_CGROUP_INET6_CONNECT: 2716 case BPF_CGROUP_UDP4_SENDMSG: 2717 case BPF_CGROUP_UDP6_SENDMSG: 2718 case BPF_CGROUP_UDP4_RECVMSG: 2719 case BPF_CGROUP_UDP6_RECVMSG: 2720 case BPF_CGROUP_SOCK_OPS: 2721 case BPF_CGROUP_DEVICE: 2722 case BPF_CGROUP_SYSCTL: 2723 case BPF_CGROUP_GETSOCKOPT: 2724 case BPF_CGROUP_SETSOCKOPT: 2725 break; 2726 case BPF_LIRC_MODE2: 2727 return lirc_prog_query(attr, uattr); 2728 case BPF_FLOW_DISSECTOR: 2729 return skb_flow_dissector_prog_query(attr, uattr); 2730 default: 2731 return -EINVAL; 2732 } 2733 2734 return cgroup_bpf_prog_query(attr, uattr); 2735 } 2736 2737 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out 2738 2739 static int bpf_prog_test_run(const union bpf_attr *attr, 2740 union bpf_attr __user *uattr) 2741 { 2742 struct bpf_prog *prog; 2743 int ret = -ENOTSUPP; 2744 2745 if (!capable(CAP_SYS_ADMIN)) 2746 return -EPERM; 2747 if (CHECK_ATTR(BPF_PROG_TEST_RUN)) 2748 return -EINVAL; 2749 2750 if ((attr->test.ctx_size_in && !attr->test.ctx_in) || 2751 (!attr->test.ctx_size_in && attr->test.ctx_in)) 2752 return -EINVAL; 2753 2754 if ((attr->test.ctx_size_out && !attr->test.ctx_out) || 2755 (!attr->test.ctx_size_out && attr->test.ctx_out)) 2756 return -EINVAL; 2757 2758 prog = bpf_prog_get(attr->test.prog_fd); 2759 if (IS_ERR(prog)) 2760 return PTR_ERR(prog); 2761 2762 if (prog->aux->ops->test_run) 2763 ret = prog->aux->ops->test_run(prog, attr, uattr); 2764 2765 bpf_prog_put(prog); 2766 return ret; 2767 } 2768 2769 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id 2770 2771 static int bpf_obj_get_next_id(const union bpf_attr *attr, 2772 union bpf_attr __user *uattr, 2773 struct idr *idr, 2774 spinlock_t *lock) 2775 { 2776 u32 next_id = attr->start_id; 2777 int err = 0; 2778 2779 if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX) 2780 return -EINVAL; 2781 2782 if (!capable(CAP_SYS_ADMIN)) 2783 return -EPERM; 2784 2785 next_id++; 2786 spin_lock_bh(lock); 2787 if (!idr_get_next(idr, &next_id)) 2788 err = -ENOENT; 2789 spin_unlock_bh(lock); 2790 2791 if (!err) 2792 err = put_user(next_id, &uattr->next_id); 2793 2794 return err; 2795 } 2796 2797 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id 2798 2799 struct bpf_prog *bpf_prog_by_id(u32 id) 2800 { 2801 struct bpf_prog *prog; 2802 2803 if (!id) 2804 return ERR_PTR(-ENOENT); 2805 2806 spin_lock_bh(&prog_idr_lock); 2807 prog = idr_find(&prog_idr, id); 2808 if (prog) 2809 prog = bpf_prog_inc_not_zero(prog); 2810 else 2811 prog = ERR_PTR(-ENOENT); 2812 spin_unlock_bh(&prog_idr_lock); 2813 return prog; 2814 } 2815 2816 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr) 2817 { 2818 struct bpf_prog *prog; 2819 u32 id = attr->prog_id; 2820 int fd; 2821 2822 if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID)) 2823 return -EINVAL; 2824 2825 if (!capable(CAP_SYS_ADMIN)) 2826 return -EPERM; 2827 2828 prog = bpf_prog_by_id(id); 2829 if (IS_ERR(prog)) 2830 return PTR_ERR(prog); 2831 2832 fd = bpf_prog_new_fd(prog); 2833 if (fd < 0) 2834 bpf_prog_put(prog); 2835 2836 return fd; 2837 } 2838 2839 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags 2840 2841 static int bpf_map_get_fd_by_id(const union bpf_attr *attr) 2842 { 2843 struct bpf_map *map; 2844 u32 id = attr->map_id; 2845 int f_flags; 2846 int fd; 2847 2848 if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) || 2849 attr->open_flags & ~BPF_OBJ_FLAG_MASK) 2850 return -EINVAL; 2851 2852 if (!capable(CAP_SYS_ADMIN)) 2853 return -EPERM; 2854 2855 f_flags = bpf_get_file_flag(attr->open_flags); 2856 if (f_flags < 0) 2857 return f_flags; 2858 2859 spin_lock_bh(&map_idr_lock); 2860 map = idr_find(&map_idr, id); 2861 if (map) 2862 map = __bpf_map_inc_not_zero(map, true); 2863 else 2864 map = ERR_PTR(-ENOENT); 2865 spin_unlock_bh(&map_idr_lock); 2866 2867 if (IS_ERR(map)) 2868 return PTR_ERR(map); 2869 2870 fd = bpf_map_new_fd(map, f_flags); 2871 if (fd < 0) 2872 bpf_map_put_with_uref(map); 2873 2874 return fd; 2875 } 2876 2877 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog, 2878 unsigned long addr, u32 *off, 2879 u32 *type) 2880 { 2881 const struct bpf_map *map; 2882 int i; 2883 2884 for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) { 2885 map = prog->aux->used_maps[i]; 2886 if (map == (void *)addr) { 2887 *type = BPF_PSEUDO_MAP_FD; 2888 return map; 2889 } 2890 if (!map->ops->map_direct_value_meta) 2891 continue; 2892 if (!map->ops->map_direct_value_meta(map, addr, off)) { 2893 *type = BPF_PSEUDO_MAP_VALUE; 2894 return map; 2895 } 2896 } 2897 2898 return NULL; 2899 } 2900 2901 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog) 2902 { 2903 const struct bpf_map *map; 2904 struct bpf_insn *insns; 2905 u32 off, type; 2906 u64 imm; 2907 int i; 2908 2909 insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog), 2910 GFP_USER); 2911 if (!insns) 2912 return insns; 2913 2914 for (i = 0; i < prog->len; i++) { 2915 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) { 2916 insns[i].code = BPF_JMP | BPF_CALL; 2917 insns[i].imm = BPF_FUNC_tail_call; 2918 /* fall-through */ 2919 } 2920 if (insns[i].code == (BPF_JMP | BPF_CALL) || 2921 insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) { 2922 if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) 2923 insns[i].code = BPF_JMP | BPF_CALL; 2924 if (!bpf_dump_raw_ok()) 2925 insns[i].imm = 0; 2926 continue; 2927 } 2928 2929 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW)) 2930 continue; 2931 2932 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm; 2933 map = bpf_map_from_imm(prog, imm, &off, &type); 2934 if (map) { 2935 insns[i].src_reg = type; 2936 insns[i].imm = map->id; 2937 insns[i + 1].imm = off; 2938 continue; 2939 } 2940 } 2941 2942 return insns; 2943 } 2944 2945 static int set_info_rec_size(struct bpf_prog_info *info) 2946 { 2947 /* 2948 * Ensure info.*_rec_size is the same as kernel expected size 2949 * 2950 * or 2951 * 2952 * Only allow zero *_rec_size if both _rec_size and _cnt are 2953 * zero. In this case, the kernel will set the expected 2954 * _rec_size back to the info. 2955 */ 2956 2957 if ((info->nr_func_info || info->func_info_rec_size) && 2958 info->func_info_rec_size != sizeof(struct bpf_func_info)) 2959 return -EINVAL; 2960 2961 if ((info->nr_line_info || info->line_info_rec_size) && 2962 info->line_info_rec_size != sizeof(struct bpf_line_info)) 2963 return -EINVAL; 2964 2965 if ((info->nr_jited_line_info || info->jited_line_info_rec_size) && 2966 info->jited_line_info_rec_size != sizeof(__u64)) 2967 return -EINVAL; 2968 2969 info->func_info_rec_size = sizeof(struct bpf_func_info); 2970 info->line_info_rec_size = sizeof(struct bpf_line_info); 2971 info->jited_line_info_rec_size = sizeof(__u64); 2972 2973 return 0; 2974 } 2975 2976 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog, 2977 const union bpf_attr *attr, 2978 union bpf_attr __user *uattr) 2979 { 2980 struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info); 2981 struct bpf_prog_info info = {}; 2982 u32 info_len = attr->info.info_len; 2983 struct bpf_prog_stats stats; 2984 char __user *uinsns; 2985 u32 ulen; 2986 int err; 2987 2988 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 2989 if (err) 2990 return err; 2991 info_len = min_t(u32, sizeof(info), info_len); 2992 2993 if (copy_from_user(&info, uinfo, info_len)) 2994 return -EFAULT; 2995 2996 info.type = prog->type; 2997 info.id = prog->aux->id; 2998 info.load_time = prog->aux->load_time; 2999 info.created_by_uid = from_kuid_munged(current_user_ns(), 3000 prog->aux->user->uid); 3001 info.gpl_compatible = prog->gpl_compatible; 3002 3003 memcpy(info.tag, prog->tag, sizeof(prog->tag)); 3004 memcpy(info.name, prog->aux->name, sizeof(prog->aux->name)); 3005 3006 ulen = info.nr_map_ids; 3007 info.nr_map_ids = prog->aux->used_map_cnt; 3008 ulen = min_t(u32, info.nr_map_ids, ulen); 3009 if (ulen) { 3010 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids); 3011 u32 i; 3012 3013 for (i = 0; i < ulen; i++) 3014 if (put_user(prog->aux->used_maps[i]->id, 3015 &user_map_ids[i])) 3016 return -EFAULT; 3017 } 3018 3019 err = set_info_rec_size(&info); 3020 if (err) 3021 return err; 3022 3023 bpf_prog_get_stats(prog, &stats); 3024 info.run_time_ns = stats.nsecs; 3025 info.run_cnt = stats.cnt; 3026 3027 if (!capable(CAP_SYS_ADMIN)) { 3028 info.jited_prog_len = 0; 3029 info.xlated_prog_len = 0; 3030 info.nr_jited_ksyms = 0; 3031 info.nr_jited_func_lens = 0; 3032 info.nr_func_info = 0; 3033 info.nr_line_info = 0; 3034 info.nr_jited_line_info = 0; 3035 goto done; 3036 } 3037 3038 ulen = info.xlated_prog_len; 3039 info.xlated_prog_len = bpf_prog_insn_size(prog); 3040 if (info.xlated_prog_len && ulen) { 3041 struct bpf_insn *insns_sanitized; 3042 bool fault; 3043 3044 if (prog->blinded && !bpf_dump_raw_ok()) { 3045 info.xlated_prog_insns = 0; 3046 goto done; 3047 } 3048 insns_sanitized = bpf_insn_prepare_dump(prog); 3049 if (!insns_sanitized) 3050 return -ENOMEM; 3051 uinsns = u64_to_user_ptr(info.xlated_prog_insns); 3052 ulen = min_t(u32, info.xlated_prog_len, ulen); 3053 fault = copy_to_user(uinsns, insns_sanitized, ulen); 3054 kfree(insns_sanitized); 3055 if (fault) 3056 return -EFAULT; 3057 } 3058 3059 if (bpf_prog_is_dev_bound(prog->aux)) { 3060 err = bpf_prog_offload_info_fill(&info, prog); 3061 if (err) 3062 return err; 3063 goto done; 3064 } 3065 3066 /* NOTE: the following code is supposed to be skipped for offload. 3067 * bpf_prog_offload_info_fill() is the place to fill similar fields 3068 * for offload. 3069 */ 3070 ulen = info.jited_prog_len; 3071 if (prog->aux->func_cnt) { 3072 u32 i; 3073 3074 info.jited_prog_len = 0; 3075 for (i = 0; i < prog->aux->func_cnt; i++) 3076 info.jited_prog_len += prog->aux->func[i]->jited_len; 3077 } else { 3078 info.jited_prog_len = prog->jited_len; 3079 } 3080 3081 if (info.jited_prog_len && ulen) { 3082 if (bpf_dump_raw_ok()) { 3083 uinsns = u64_to_user_ptr(info.jited_prog_insns); 3084 ulen = min_t(u32, info.jited_prog_len, ulen); 3085 3086 /* for multi-function programs, copy the JITed 3087 * instructions for all the functions 3088 */ 3089 if (prog->aux->func_cnt) { 3090 u32 len, free, i; 3091 u8 *img; 3092 3093 free = ulen; 3094 for (i = 0; i < prog->aux->func_cnt; i++) { 3095 len = prog->aux->func[i]->jited_len; 3096 len = min_t(u32, len, free); 3097 img = (u8 *) prog->aux->func[i]->bpf_func; 3098 if (copy_to_user(uinsns, img, len)) 3099 return -EFAULT; 3100 uinsns += len; 3101 free -= len; 3102 if (!free) 3103 break; 3104 } 3105 } else { 3106 if (copy_to_user(uinsns, prog->bpf_func, ulen)) 3107 return -EFAULT; 3108 } 3109 } else { 3110 info.jited_prog_insns = 0; 3111 } 3112 } 3113 3114 ulen = info.nr_jited_ksyms; 3115 info.nr_jited_ksyms = prog->aux->func_cnt ? : 1; 3116 if (ulen) { 3117 if (bpf_dump_raw_ok()) { 3118 unsigned long ksym_addr; 3119 u64 __user *user_ksyms; 3120 u32 i; 3121 3122 /* copy the address of the kernel symbol 3123 * corresponding to each function 3124 */ 3125 ulen = min_t(u32, info.nr_jited_ksyms, ulen); 3126 user_ksyms = u64_to_user_ptr(info.jited_ksyms); 3127 if (prog->aux->func_cnt) { 3128 for (i = 0; i < ulen; i++) { 3129 ksym_addr = (unsigned long) 3130 prog->aux->func[i]->bpf_func; 3131 if (put_user((u64) ksym_addr, 3132 &user_ksyms[i])) 3133 return -EFAULT; 3134 } 3135 } else { 3136 ksym_addr = (unsigned long) prog->bpf_func; 3137 if (put_user((u64) ksym_addr, &user_ksyms[0])) 3138 return -EFAULT; 3139 } 3140 } else { 3141 info.jited_ksyms = 0; 3142 } 3143 } 3144 3145 ulen = info.nr_jited_func_lens; 3146 info.nr_jited_func_lens = prog->aux->func_cnt ? : 1; 3147 if (ulen) { 3148 if (bpf_dump_raw_ok()) { 3149 u32 __user *user_lens; 3150 u32 func_len, i; 3151 3152 /* copy the JITed image lengths for each function */ 3153 ulen = min_t(u32, info.nr_jited_func_lens, ulen); 3154 user_lens = u64_to_user_ptr(info.jited_func_lens); 3155 if (prog->aux->func_cnt) { 3156 for (i = 0; i < ulen; i++) { 3157 func_len = 3158 prog->aux->func[i]->jited_len; 3159 if (put_user(func_len, &user_lens[i])) 3160 return -EFAULT; 3161 } 3162 } else { 3163 func_len = prog->jited_len; 3164 if (put_user(func_len, &user_lens[0])) 3165 return -EFAULT; 3166 } 3167 } else { 3168 info.jited_func_lens = 0; 3169 } 3170 } 3171 3172 if (prog->aux->btf) 3173 info.btf_id = btf_id(prog->aux->btf); 3174 3175 ulen = info.nr_func_info; 3176 info.nr_func_info = prog->aux->func_info_cnt; 3177 if (info.nr_func_info && ulen) { 3178 char __user *user_finfo; 3179 3180 user_finfo = u64_to_user_ptr(info.func_info); 3181 ulen = min_t(u32, info.nr_func_info, ulen); 3182 if (copy_to_user(user_finfo, prog->aux->func_info, 3183 info.func_info_rec_size * ulen)) 3184 return -EFAULT; 3185 } 3186 3187 ulen = info.nr_line_info; 3188 info.nr_line_info = prog->aux->nr_linfo; 3189 if (info.nr_line_info && ulen) { 3190 __u8 __user *user_linfo; 3191 3192 user_linfo = u64_to_user_ptr(info.line_info); 3193 ulen = min_t(u32, info.nr_line_info, ulen); 3194 if (copy_to_user(user_linfo, prog->aux->linfo, 3195 info.line_info_rec_size * ulen)) 3196 return -EFAULT; 3197 } 3198 3199 ulen = info.nr_jited_line_info; 3200 if (prog->aux->jited_linfo) 3201 info.nr_jited_line_info = prog->aux->nr_linfo; 3202 else 3203 info.nr_jited_line_info = 0; 3204 if (info.nr_jited_line_info && ulen) { 3205 if (bpf_dump_raw_ok()) { 3206 __u64 __user *user_linfo; 3207 u32 i; 3208 3209 user_linfo = u64_to_user_ptr(info.jited_line_info); 3210 ulen = min_t(u32, info.nr_jited_line_info, ulen); 3211 for (i = 0; i < ulen; i++) { 3212 if (put_user((__u64)(long)prog->aux->jited_linfo[i], 3213 &user_linfo[i])) 3214 return -EFAULT; 3215 } 3216 } else { 3217 info.jited_line_info = 0; 3218 } 3219 } 3220 3221 ulen = info.nr_prog_tags; 3222 info.nr_prog_tags = prog->aux->func_cnt ? : 1; 3223 if (ulen) { 3224 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE]; 3225 u32 i; 3226 3227 user_prog_tags = u64_to_user_ptr(info.prog_tags); 3228 ulen = min_t(u32, info.nr_prog_tags, ulen); 3229 if (prog->aux->func_cnt) { 3230 for (i = 0; i < ulen; i++) { 3231 if (copy_to_user(user_prog_tags[i], 3232 prog->aux->func[i]->tag, 3233 BPF_TAG_SIZE)) 3234 return -EFAULT; 3235 } 3236 } else { 3237 if (copy_to_user(user_prog_tags[0], 3238 prog->tag, BPF_TAG_SIZE)) 3239 return -EFAULT; 3240 } 3241 } 3242 3243 done: 3244 if (copy_to_user(uinfo, &info, info_len) || 3245 put_user(info_len, &uattr->info.info_len)) 3246 return -EFAULT; 3247 3248 return 0; 3249 } 3250 3251 static int bpf_map_get_info_by_fd(struct bpf_map *map, 3252 const union bpf_attr *attr, 3253 union bpf_attr __user *uattr) 3254 { 3255 struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info); 3256 struct bpf_map_info info = {}; 3257 u32 info_len = attr->info.info_len; 3258 int err; 3259 3260 err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len); 3261 if (err) 3262 return err; 3263 info_len = min_t(u32, sizeof(info), info_len); 3264 3265 info.type = map->map_type; 3266 info.id = map->id; 3267 info.key_size = map->key_size; 3268 info.value_size = map->value_size; 3269 info.max_entries = map->max_entries; 3270 info.map_flags = map->map_flags; 3271 memcpy(info.name, map->name, sizeof(map->name)); 3272 3273 if (map->btf) { 3274 info.btf_id = btf_id(map->btf); 3275 info.btf_key_type_id = map->btf_key_type_id; 3276 info.btf_value_type_id = map->btf_value_type_id; 3277 } 3278 info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id; 3279 3280 if (bpf_map_is_dev_bound(map)) { 3281 err = bpf_map_offload_info_fill(&info, map); 3282 if (err) 3283 return err; 3284 } 3285 3286 if (copy_to_user(uinfo, &info, info_len) || 3287 put_user(info_len, &uattr->info.info_len)) 3288 return -EFAULT; 3289 3290 return 0; 3291 } 3292 3293 static int bpf_btf_get_info_by_fd(struct btf *btf, 3294 const union bpf_attr *attr, 3295 union bpf_attr __user *uattr) 3296 { 3297 struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info); 3298 u32 info_len = attr->info.info_len; 3299 int err; 3300 3301 err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len); 3302 if (err) 3303 return err; 3304 3305 return btf_get_info_by_fd(btf, attr, uattr); 3306 } 3307 3308 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info 3309 3310 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr, 3311 union bpf_attr __user *uattr) 3312 { 3313 int ufd = attr->info.bpf_fd; 3314 struct fd f; 3315 int err; 3316 3317 if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD)) 3318 return -EINVAL; 3319 3320 f = fdget(ufd); 3321 if (!f.file) 3322 return -EBADFD; 3323 3324 if (f.file->f_op == &bpf_prog_fops) 3325 err = bpf_prog_get_info_by_fd(f.file->private_data, attr, 3326 uattr); 3327 else if (f.file->f_op == &bpf_map_fops) 3328 err = bpf_map_get_info_by_fd(f.file->private_data, attr, 3329 uattr); 3330 else if (f.file->f_op == &btf_fops) 3331 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr); 3332 else 3333 err = -EINVAL; 3334 3335 fdput(f); 3336 return err; 3337 } 3338 3339 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level 3340 3341 static int bpf_btf_load(const union bpf_attr *attr) 3342 { 3343 if (CHECK_ATTR(BPF_BTF_LOAD)) 3344 return -EINVAL; 3345 3346 if (!capable(CAP_SYS_ADMIN)) 3347 return -EPERM; 3348 3349 return btf_new_fd(attr); 3350 } 3351 3352 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id 3353 3354 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr) 3355 { 3356 if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID)) 3357 return -EINVAL; 3358 3359 if (!capable(CAP_SYS_ADMIN)) 3360 return -EPERM; 3361 3362 return btf_get_fd_by_id(attr->btf_id); 3363 } 3364 3365 static int bpf_task_fd_query_copy(const union bpf_attr *attr, 3366 union bpf_attr __user *uattr, 3367 u32 prog_id, u32 fd_type, 3368 const char *buf, u64 probe_offset, 3369 u64 probe_addr) 3370 { 3371 char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf); 3372 u32 len = buf ? strlen(buf) : 0, input_len; 3373 int err = 0; 3374 3375 if (put_user(len, &uattr->task_fd_query.buf_len)) 3376 return -EFAULT; 3377 input_len = attr->task_fd_query.buf_len; 3378 if (input_len && ubuf) { 3379 if (!len) { 3380 /* nothing to copy, just make ubuf NULL terminated */ 3381 char zero = '\0'; 3382 3383 if (put_user(zero, ubuf)) 3384 return -EFAULT; 3385 } else if (input_len >= len + 1) { 3386 /* ubuf can hold the string with NULL terminator */ 3387 if (copy_to_user(ubuf, buf, len + 1)) 3388 return -EFAULT; 3389 } else { 3390 /* ubuf cannot hold the string with NULL terminator, 3391 * do a partial copy with NULL terminator. 3392 */ 3393 char zero = '\0'; 3394 3395 err = -ENOSPC; 3396 if (copy_to_user(ubuf, buf, input_len - 1)) 3397 return -EFAULT; 3398 if (put_user(zero, ubuf + input_len - 1)) 3399 return -EFAULT; 3400 } 3401 } 3402 3403 if (put_user(prog_id, &uattr->task_fd_query.prog_id) || 3404 put_user(fd_type, &uattr->task_fd_query.fd_type) || 3405 put_user(probe_offset, &uattr->task_fd_query.probe_offset) || 3406 put_user(probe_addr, &uattr->task_fd_query.probe_addr)) 3407 return -EFAULT; 3408 3409 return err; 3410 } 3411 3412 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr 3413 3414 static int bpf_task_fd_query(const union bpf_attr *attr, 3415 union bpf_attr __user *uattr) 3416 { 3417 pid_t pid = attr->task_fd_query.pid; 3418 u32 fd = attr->task_fd_query.fd; 3419 const struct perf_event *event; 3420 struct files_struct *files; 3421 struct task_struct *task; 3422 struct file *file; 3423 int err; 3424 3425 if (CHECK_ATTR(BPF_TASK_FD_QUERY)) 3426 return -EINVAL; 3427 3428 if (!capable(CAP_SYS_ADMIN)) 3429 return -EPERM; 3430 3431 if (attr->task_fd_query.flags != 0) 3432 return -EINVAL; 3433 3434 task = get_pid_task(find_vpid(pid), PIDTYPE_PID); 3435 if (!task) 3436 return -ENOENT; 3437 3438 files = get_files_struct(task); 3439 put_task_struct(task); 3440 if (!files) 3441 return -ENOENT; 3442 3443 err = 0; 3444 spin_lock(&files->file_lock); 3445 file = fcheck_files(files, fd); 3446 if (!file) 3447 err = -EBADF; 3448 else 3449 get_file(file); 3450 spin_unlock(&files->file_lock); 3451 put_files_struct(files); 3452 3453 if (err) 3454 goto out; 3455 3456 if (file->f_op == &bpf_link_fops) { 3457 struct bpf_link *link = file->private_data; 3458 3459 if (link->ops == &bpf_raw_tp_lops) { 3460 struct bpf_raw_tp_link *raw_tp = 3461 container_of(link, struct bpf_raw_tp_link, link); 3462 struct bpf_raw_event_map *btp = raw_tp->btp; 3463 3464 err = bpf_task_fd_query_copy(attr, uattr, 3465 raw_tp->link.prog->aux->id, 3466 BPF_FD_TYPE_RAW_TRACEPOINT, 3467 btp->tp->name, 0, 0); 3468 goto put_file; 3469 } 3470 goto out_not_supp; 3471 } 3472 3473 event = perf_get_event(file); 3474 if (!IS_ERR(event)) { 3475 u64 probe_offset, probe_addr; 3476 u32 prog_id, fd_type; 3477 const char *buf; 3478 3479 err = bpf_get_perf_event_info(event, &prog_id, &fd_type, 3480 &buf, &probe_offset, 3481 &probe_addr); 3482 if (!err) 3483 err = bpf_task_fd_query_copy(attr, uattr, prog_id, 3484 fd_type, buf, 3485 probe_offset, 3486 probe_addr); 3487 goto put_file; 3488 } 3489 3490 out_not_supp: 3491 err = -ENOTSUPP; 3492 put_file: 3493 fput(file); 3494 out: 3495 return err; 3496 } 3497 3498 #define BPF_MAP_BATCH_LAST_FIELD batch.flags 3499 3500 #define BPF_DO_BATCH(fn) \ 3501 do { \ 3502 if (!fn) { \ 3503 err = -ENOTSUPP; \ 3504 goto err_put; \ 3505 } \ 3506 err = fn(map, attr, uattr); \ 3507 } while (0) 3508 3509 static int bpf_map_do_batch(const union bpf_attr *attr, 3510 union bpf_attr __user *uattr, 3511 int cmd) 3512 { 3513 struct bpf_map *map; 3514 int err, ufd; 3515 struct fd f; 3516 3517 if (CHECK_ATTR(BPF_MAP_BATCH)) 3518 return -EINVAL; 3519 3520 ufd = attr->batch.map_fd; 3521 f = fdget(ufd); 3522 map = __bpf_map_get(f); 3523 if (IS_ERR(map)) 3524 return PTR_ERR(map); 3525 3526 if ((cmd == BPF_MAP_LOOKUP_BATCH || 3527 cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) && 3528 !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) { 3529 err = -EPERM; 3530 goto err_put; 3531 } 3532 3533 if (cmd != BPF_MAP_LOOKUP_BATCH && 3534 !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) { 3535 err = -EPERM; 3536 goto err_put; 3537 } 3538 3539 if (cmd == BPF_MAP_LOOKUP_BATCH) 3540 BPF_DO_BATCH(map->ops->map_lookup_batch); 3541 else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) 3542 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch); 3543 else if (cmd == BPF_MAP_UPDATE_BATCH) 3544 BPF_DO_BATCH(map->ops->map_update_batch); 3545 else 3546 BPF_DO_BATCH(map->ops->map_delete_batch); 3547 3548 err_put: 3549 fdput(f); 3550 return err; 3551 } 3552 3553 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size) 3554 { 3555 union bpf_attr attr = {}; 3556 int err; 3557 3558 if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN)) 3559 return -EPERM; 3560 3561 err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size); 3562 if (err) 3563 return err; 3564 size = min_t(u32, size, sizeof(attr)); 3565 3566 /* copy attributes from user space, may be less than sizeof(bpf_attr) */ 3567 if (copy_from_user(&attr, uattr, size) != 0) 3568 return -EFAULT; 3569 3570 err = security_bpf(cmd, &attr, size); 3571 if (err < 0) 3572 return err; 3573 3574 switch (cmd) { 3575 case BPF_MAP_CREATE: 3576 err = map_create(&attr); 3577 break; 3578 case BPF_MAP_LOOKUP_ELEM: 3579 err = map_lookup_elem(&attr); 3580 break; 3581 case BPF_MAP_UPDATE_ELEM: 3582 err = map_update_elem(&attr); 3583 break; 3584 case BPF_MAP_DELETE_ELEM: 3585 err = map_delete_elem(&attr); 3586 break; 3587 case BPF_MAP_GET_NEXT_KEY: 3588 err = map_get_next_key(&attr); 3589 break; 3590 case BPF_MAP_FREEZE: 3591 err = map_freeze(&attr); 3592 break; 3593 case BPF_PROG_LOAD: 3594 err = bpf_prog_load(&attr, uattr); 3595 break; 3596 case BPF_OBJ_PIN: 3597 err = bpf_obj_pin(&attr); 3598 break; 3599 case BPF_OBJ_GET: 3600 err = bpf_obj_get(&attr); 3601 break; 3602 case BPF_PROG_ATTACH: 3603 err = bpf_prog_attach(&attr); 3604 break; 3605 case BPF_PROG_DETACH: 3606 err = bpf_prog_detach(&attr); 3607 break; 3608 case BPF_PROG_QUERY: 3609 err = bpf_prog_query(&attr, uattr); 3610 break; 3611 case BPF_PROG_TEST_RUN: 3612 err = bpf_prog_test_run(&attr, uattr); 3613 break; 3614 case BPF_PROG_GET_NEXT_ID: 3615 err = bpf_obj_get_next_id(&attr, uattr, 3616 &prog_idr, &prog_idr_lock); 3617 break; 3618 case BPF_MAP_GET_NEXT_ID: 3619 err = bpf_obj_get_next_id(&attr, uattr, 3620 &map_idr, &map_idr_lock); 3621 break; 3622 case BPF_BTF_GET_NEXT_ID: 3623 err = bpf_obj_get_next_id(&attr, uattr, 3624 &btf_idr, &btf_idr_lock); 3625 break; 3626 case BPF_PROG_GET_FD_BY_ID: 3627 err = bpf_prog_get_fd_by_id(&attr); 3628 break; 3629 case BPF_MAP_GET_FD_BY_ID: 3630 err = bpf_map_get_fd_by_id(&attr); 3631 break; 3632 case BPF_OBJ_GET_INFO_BY_FD: 3633 err = bpf_obj_get_info_by_fd(&attr, uattr); 3634 break; 3635 case BPF_RAW_TRACEPOINT_OPEN: 3636 err = bpf_raw_tracepoint_open(&attr); 3637 break; 3638 case BPF_BTF_LOAD: 3639 err = bpf_btf_load(&attr); 3640 break; 3641 case BPF_BTF_GET_FD_BY_ID: 3642 err = bpf_btf_get_fd_by_id(&attr); 3643 break; 3644 case BPF_TASK_FD_QUERY: 3645 err = bpf_task_fd_query(&attr, uattr); 3646 break; 3647 case BPF_MAP_LOOKUP_AND_DELETE_ELEM: 3648 err = map_lookup_and_delete_elem(&attr); 3649 break; 3650 case BPF_MAP_LOOKUP_BATCH: 3651 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH); 3652 break; 3653 case BPF_MAP_LOOKUP_AND_DELETE_BATCH: 3654 err = bpf_map_do_batch(&attr, uattr, 3655 BPF_MAP_LOOKUP_AND_DELETE_BATCH); 3656 break; 3657 case BPF_MAP_UPDATE_BATCH: 3658 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH); 3659 break; 3660 case BPF_MAP_DELETE_BATCH: 3661 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH); 3662 break; 3663 default: 3664 err = -EINVAL; 3665 break; 3666 } 3667 3668 return err; 3669 } 3670