kernel/bpf/syscall.c

   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-cgroup.h>
   6 #include <linux/bpf_trace.h>
   7 #include <linux/bpf_lirc.h>
   8 #include <linux/bpf_verifier.h>
   9 #include <linux/bsearch.h>
  10 #include <linux/btf.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mmzone.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/fdtable.h>
  18 #include <linux/file.h>
  19 #include <linux/fs.h>
  20 #include <linux/license.h>
  21 #include <linux/filter.h>
  22 #include <linux/kernel.h>
  23 #include <linux/idr.h>
  24 #include <linux/cred.h>
  25 #include <linux/timekeeping.h>
  26 #include <linux/ctype.h>
  27 #include <linux/nospec.h>
  28 #include <linux/audit.h>
  29 #include <uapi/linux/btf.h>
  30 #include <linux/pgtable.h>
  31 #include <linux/bpf_lsm.h>
  32 #include <linux/poll.h>
  33 #include <linux/sort.h>
  34 #include <linux/bpf-netns.h>
  35 #include <linux/rcupdate_trace.h>
  36 #include <linux/memcontrol.h>
  37 #include <linux/trace_events.h>
  38 #include <net/netfilter/nf_bpf_link.h>
  39
  40 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  41                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  42                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  43 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  44 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  45 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  46                         IS_FD_HASH(map))
  47
  48 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  49
  50 DEFINE_PER_CPU(int, bpf_prog_active);
  51 static DEFINE_IDR(prog_idr);
  52 static DEFINE_SPINLOCK(prog_idr_lock);
  53 static DEFINE_IDR(map_idr);
  54 static DEFINE_SPINLOCK(map_idr_lock);
  55 static DEFINE_IDR(link_idr);
  56 static DEFINE_SPINLOCK(link_idr_lock);
  57
  58 int sysctl_unprivileged_bpf_disabled __read_mostly =
  59         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  60
  61 static const struct bpf_map_ops * const bpf_map_types[] = {
  62 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  63 #define BPF_MAP_TYPE(_id, _ops) \
  64         [_id] = &_ops,
  65 #define BPF_LINK_TYPE(_id, _name)
  66 #include <linux/bpf_types.h>
  67 #undef BPF_PROG_TYPE
  68 #undef BPF_MAP_TYPE
  69 #undef BPF_LINK_TYPE
  70 };
  71
  72 /*
  73  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  74  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  75  * we don't know about yet.
  76  *
  77  * There is a ToCToU between this function call and the following
  78  * copy_from_user() call. However, this is not a concern since this function is
  79  * meant to be a future-proofing of bits.
  80  */
  81 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  82                              size_t expected_size,
  83                              size_t actual_size)
  84 {
  85         int res;
  86
  87         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  88                 return -E2BIG;
  89
  90         if (actual_size <= expected_size)
  91                 return 0;
  92
  93         if (uaddr.is_kernel)
  94                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  95                                  actual_size - expected_size) == NULL;
  96         else
  97                 res = check_zeroed_user(uaddr.user + expected_size,
  98                                         actual_size - expected_size);
  99         if (res < 0)
 100                 return res;
 101         return res ? 0 : -E2BIG;
 102 }
 103
 104 const struct bpf_map_ops bpf_map_offload_ops = {
 105         .map_meta_equal = bpf_map_meta_equal,
 106         .map_alloc = bpf_map_offload_map_alloc,
 107         .map_free = bpf_map_offload_map_free,
 108         .map_check_btf = map_check_no_btf,
 109         .map_mem_usage = bpf_map_offload_map_mem_usage,
 110 };
 111
 112 static void bpf_map_write_active_inc(struct bpf_map *map)
 113 {
 114         atomic64_inc(&map->writecnt);
 115 }
 116
 117 static void bpf_map_write_active_dec(struct bpf_map *map)
 118 {
 119         atomic64_dec(&map->writecnt);
 120 }
 121
 122 bool bpf_map_write_active(const struct bpf_map *map)
 123 {
 124         return atomic64_read(&map->writecnt) != 0;
 125 }
 126
 127 static u32 bpf_map_value_size(const struct bpf_map *map)
 128 {
 129         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 130             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 131             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 132             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 133                 return round_up(map->value_size, 8) * num_possible_cpus();
 134         else if (IS_FD_MAP(map))
 135                 return sizeof(u32);
 136         else
 137                 return  map->value_size;
 138 }
 139
 140 static void maybe_wait_bpf_programs(struct bpf_map *map)
 141 {
 142         /* Wait for any running BPF programs to complete so that
 143          * userspace, when we return to it, knows that all programs
 144          * that could be running use the new map value.
 145          */
 146         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 147             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 148                 synchronize_rcu();
 149 }
 150
 151 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 152                                 void *key, void *value, __u64 flags)
 153 {
 154         int err;
 155
 156         /* Need to create a kthread, thus must support schedule */
 157         if (bpf_map_is_offloaded(map)) {
 158                 return bpf_map_offload_update_elem(map, key, value, flags);
 159         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 160                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 161                 return map->ops->map_update_elem(map, key, value, flags);
 162         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 163                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 164                 return sock_map_update_elem_sys(map, key, value, flags);
 165         } else if (IS_FD_PROG_ARRAY(map)) {
 166                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 167                                                     flags);
 168         }
 169
 170         bpf_disable_instrumentation();
 171         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 172             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 173                 err = bpf_percpu_hash_update(map, key, value, flags);
 174         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 175                 err = bpf_percpu_array_update(map, key, value, flags);
 176         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 177                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 178                                                        flags);
 179         } else if (IS_FD_ARRAY(map)) {
 180                 rcu_read_lock();
 181                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 182                                                    flags);
 183                 rcu_read_unlock();
 184         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 185                 rcu_read_lock();
 186                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 187                                                   flags);
 188                 rcu_read_unlock();
 189         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 190                 /* rcu_read_lock() is not needed */
 191                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 192                                                          flags);
 193         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 194                    map->map_type == BPF_MAP_TYPE_STACK ||
 195                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 196                 err = map->ops->map_push_elem(map, value, flags);
 197         } else {
 198                 rcu_read_lock();
 199                 err = map->ops->map_update_elem(map, key, value, flags);
 200                 rcu_read_unlock();
 201         }
 202         bpf_enable_instrumentation();
 203         maybe_wait_bpf_programs(map);
 204
 205         return err;
 206 }
 207
 208 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 209                               __u64 flags)
 210 {
 211         void *ptr;
 212         int err;
 213
 214         if (bpf_map_is_offloaded(map))
 215                 return bpf_map_offload_lookup_elem(map, key, value);
 216
 217         bpf_disable_instrumentation();
 218         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 219             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 220                 err = bpf_percpu_hash_copy(map, key, value);
 221         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 222                 err = bpf_percpu_array_copy(map, key, value);
 223         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 224                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 225         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 226                 err = bpf_stackmap_copy(map, key, value);
 227         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 228                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 229         } else if (IS_FD_HASH(map)) {
 230                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 231         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 232                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 233         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 234                    map->map_type == BPF_MAP_TYPE_STACK ||
 235                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 236                 err = map->ops->map_peek_elem(map, value);
 237         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 238                 /* struct_ops map requires directly updating "value" */
 239                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 240         } else {
 241                 rcu_read_lock();
 242                 if (map->ops->map_lookup_elem_sys_only)
 243                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 244                 else
 245                         ptr = map->ops->map_lookup_elem(map, key);
 246                 if (IS_ERR(ptr)) {
 247                         err = PTR_ERR(ptr);
 248                 } else if (!ptr) {
 249                         err = -ENOENT;
 250                 } else {
 251                         err = 0;
 252                         if (flags & BPF_F_LOCK)
 253                                 /* lock 'ptr' and copy everything but lock */
 254                                 copy_map_value_locked(map, value, ptr, true);
 255                         else
 256                                 copy_map_value(map, value, ptr);
 257                         /* mask lock and timer, since value wasn't zero inited */
 258                         check_and_init_map_value(map, value);
 259                 }
 260                 rcu_read_unlock();
 261         }
 262
 263         bpf_enable_instrumentation();
 264         maybe_wait_bpf_programs(map);
 265
 266         return err;
 267 }
 268
 269 /* Please, do not use this function outside from the map creation path
 270  * (e.g. in map update path) without taking care of setting the active
 271  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 272  */
 273 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 274 {
 275         /* We really just want to fail instead of triggering OOM killer
 276          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 277          * which is used for lower order allocation requests.
 278          *
 279          * It has been observed that higher order allocation requests done by
 280          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 281          * to reclaim memory from the page cache, thus we set
 282          * __GFP_RETRY_MAYFAIL to avoid such situations.
 283          */
 284
 285         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 286         unsigned int flags = 0;
 287         unsigned long align = 1;
 288         void *area;
 289
 290         if (size >= SIZE_MAX)
 291                 return NULL;
 292
 293         /* kmalloc()'ed memory can't be mmap()'ed */
 294         if (mmapable) {
 295                 BUG_ON(!PAGE_ALIGNED(size));
 296                 align = SHMLBA;
 297                 flags = VM_USERMAP;
 298         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 299                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 300                                     numa_node);
 301                 if (area != NULL)
 302                         return area;
 303         }
 304
 305         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 306                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 307                         flags, numa_node, __builtin_return_address(0));
 308 }
 309
 310 void *bpf_map_area_alloc(u64 size, int numa_node)
 311 {
 312         return __bpf_map_area_alloc(size, numa_node, false);
 313 }
 314
 315 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 316 {
 317         return __bpf_map_area_alloc(size, numa_node, true);
 318 }
 319
 320 void bpf_map_area_free(void *area)
 321 {
 322         kvfree(area);
 323 }
 324
 325 static u32 bpf_map_flags_retain_permanent(u32 flags)
 326 {
 327         /* Some map creation flags are not tied to the map object but
 328          * rather to the map fd instead, so they have no meaning upon
 329          * map object inspection since multiple file descriptors with
 330          * different (access) properties can exist here. Thus, given
 331          * this has zero meaning for the map itself, lets clear these
 332          * from here.
 333          */
 334         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 335 }
 336
 337 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 338 {
 339         map->map_type = attr->map_type;
 340         map->key_size = attr->key_size;
 341         map->value_size = attr->value_size;
 342         map->max_entries = attr->max_entries;
 343         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 344         map->numa_node = bpf_map_attr_numa_node(attr);
 345         map->map_extra = attr->map_extra;
 346 }
 347
 348 static int bpf_map_alloc_id(struct bpf_map *map)
 349 {
 350         int id;
 351
 352         idr_preload(GFP_KERNEL);
 353         spin_lock_bh(&map_idr_lock);
 354         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 355         if (id > 0)
 356                 map->id = id;
 357         spin_unlock_bh(&map_idr_lock);
 358         idr_preload_end();
 359
 360         if (WARN_ON_ONCE(!id))
 361                 return -ENOSPC;
 362
 363         return id > 0 ? 0 : id;
 364 }
 365
 366 void bpf_map_free_id(struct bpf_map *map)
 367 {
 368         unsigned long flags;
 369
 370         /* Offloaded maps are removed from the IDR store when their device
 371          * disappears - even if someone holds an fd to them they are unusable,
 372          * the memory is gone, all ops will fail; they are simply waiting for
 373          * refcnt to drop to be freed.
 374          */
 375         if (!map->id)
 376                 return;
 377
 378         spin_lock_irqsave(&map_idr_lock, flags);
 379
 380         idr_remove(&map_idr, map->id);
 381         map->id = 0;
 382
 383         spin_unlock_irqrestore(&map_idr_lock, flags);
 384 }
 385
 386 #ifdef CONFIG_MEMCG_KMEM
 387 static void bpf_map_save_memcg(struct bpf_map *map)
 388 {
 389         /* Currently if a map is created by a process belonging to the root
 390          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 391          * So we have to check map->objcg for being NULL each time it's
 392          * being used.
 393          */
 394         if (memcg_bpf_enabled())
 395                 map->objcg = get_obj_cgroup_from_current();
 396 }
 397
 398 static void bpf_map_release_memcg(struct bpf_map *map)
 399 {
 400         if (map->objcg)
 401                 obj_cgroup_put(map->objcg);
 402 }
 403
 404 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 405 {
 406         if (map->objcg)
 407                 return get_mem_cgroup_from_objcg(map->objcg);
 408
 409         return root_mem_cgroup;
 410 }
 411
 412 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 413                            int node)
 414 {
 415         struct mem_cgroup *memcg, *old_memcg;
 416         void *ptr;
 417
 418         memcg = bpf_map_get_memcg(map);
 419         old_memcg = set_active_memcg(memcg);
 420         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 421         set_active_memcg(old_memcg);
 422         mem_cgroup_put(memcg);
 423
 424         return ptr;
 425 }
 426
 427 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 428 {
 429         struct mem_cgroup *memcg, *old_memcg;
 430         void *ptr;
 431
 432         memcg = bpf_map_get_memcg(map);
 433         old_memcg = set_active_memcg(memcg);
 434         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 435         set_active_memcg(old_memcg);
 436         mem_cgroup_put(memcg);
 437
 438         return ptr;
 439 }
 440
 441 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 442                        gfp_t flags)
 443 {
 444         struct mem_cgroup *memcg, *old_memcg;
 445         void *ptr;
 446
 447         memcg = bpf_map_get_memcg(map);
 448         old_memcg = set_active_memcg(memcg);
 449         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 450         set_active_memcg(old_memcg);
 451         mem_cgroup_put(memcg);
 452
 453         return ptr;
 454 }
 455
 456 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 457                                     size_t align, gfp_t flags)
 458 {
 459         struct mem_cgroup *memcg, *old_memcg;
 460         void __percpu *ptr;
 461
 462         memcg = bpf_map_get_memcg(map);
 463         old_memcg = set_active_memcg(memcg);
 464         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 465         set_active_memcg(old_memcg);
 466         mem_cgroup_put(memcg);
 467
 468         return ptr;
 469 }
 470
 471 #else
 472 static void bpf_map_save_memcg(struct bpf_map *map)
 473 {
 474 }
 475
 476 static void bpf_map_release_memcg(struct bpf_map *map)
 477 {
 478 }
 479 #endif
 480
 481 static int btf_field_cmp(const void *a, const void *b)
 482 {
 483         const struct btf_field *f1 = a, *f2 = b;
 484
 485         if (f1->offset < f2->offset)
 486                 return -1;
 487         else if (f1->offset > f2->offset)
 488                 return 1;
 489         return 0;
 490 }
 491
 492 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 493                                   u32 field_mask)
 494 {
 495         struct btf_field *field;
 496
 497         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 498                 return NULL;
 499         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 500         if (!field || !(field->type & field_mask))
 501                 return NULL;
 502         return field;
 503 }
 504
 505 void btf_record_free(struct btf_record *rec)
 506 {
 507         int i;
 508
 509         if (IS_ERR_OR_NULL(rec))
 510                 return;
 511         for (i = 0; i < rec->cnt; i++) {
 512                 switch (rec->fields[i].type) {
 513                 case BPF_KPTR_UNREF:
 514                 case BPF_KPTR_REF:
 515                         if (rec->fields[i].kptr.module)
 516                                 module_put(rec->fields[i].kptr.module);
 517                         btf_put(rec->fields[i].kptr.btf);
 518                         break;
 519                 case BPF_LIST_HEAD:
 520                 case BPF_LIST_NODE:
 521                 case BPF_RB_ROOT:
 522                 case BPF_RB_NODE:
 523                 case BPF_SPIN_LOCK:
 524                 case BPF_TIMER:
 525                 case BPF_REFCOUNT:
 526                         /* Nothing to release */
 527                         break;
 528                 default:
 529                         WARN_ON_ONCE(1);
 530                         continue;
 531                 }
 532         }
 533         kfree(rec);
 534 }
 535
 536 void bpf_map_free_record(struct bpf_map *map)
 537 {
 538         btf_record_free(map->record);
 539         map->record = NULL;
 540 }
 541
 542 struct btf_record *btf_record_dup(const struct btf_record *rec)
 543 {
 544         const struct btf_field *fields;
 545         struct btf_record *new_rec;
 546         int ret, size, i;
 547
 548         if (IS_ERR_OR_NULL(rec))
 549                 return NULL;
 550         size = offsetof(struct btf_record, fields[rec->cnt]);
 551         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 552         if (!new_rec)
 553                 return ERR_PTR(-ENOMEM);
 554         /* Do a deep copy of the btf_record */
 555         fields = rec->fields;
 556         new_rec->cnt = 0;
 557         for (i = 0; i < rec->cnt; i++) {
 558                 switch (fields[i].type) {
 559                 case BPF_KPTR_UNREF:
 560                 case BPF_KPTR_REF:
 561                         btf_get(fields[i].kptr.btf);
 562                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 563                                 ret = -ENXIO;
 564                                 goto free;
 565                         }
 566                         break;
 567                 case BPF_LIST_HEAD:
 568                 case BPF_LIST_NODE:
 569                 case BPF_RB_ROOT:
 570                 case BPF_RB_NODE:
 571                 case BPF_SPIN_LOCK:
 572                 case BPF_TIMER:
 573                 case BPF_REFCOUNT:
 574                         /* Nothing to acquire */
 575                         break;
 576                 default:
 577                         ret = -EFAULT;
 578                         WARN_ON_ONCE(1);
 579                         goto free;
 580                 }
 581                 new_rec->cnt++;
 582         }
 583         return new_rec;
 584 free:
 585         btf_record_free(new_rec);
 586         return ERR_PTR(ret);
 587 }
 588
 589 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 590 {
 591         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 592         int size;
 593
 594         if (!a_has_fields && !b_has_fields)
 595                 return true;
 596         if (a_has_fields != b_has_fields)
 597                 return false;
 598         if (rec_a->cnt != rec_b->cnt)
 599                 return false;
 600         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 601         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 602          * members are zeroed out. So memcmp is safe to do without worrying
 603          * about padding/unused fields.
 604          *
 605          * While spin_lock, timer, and kptr have no relation to map BTF,
 606          * list_head metadata is specific to map BTF, the btf and value_rec
 607          * members in particular. btf is the map BTF, while value_rec points to
 608          * btf_record in that map BTF.
 609          *
 610          * So while by default, we don't rely on the map BTF (which the records
 611          * were parsed from) matching for both records, which is not backwards
 612          * compatible, in case list_head is part of it, we implicitly rely on
 613          * that by way of depending on memcmp succeeding for it.
 614          */
 615         return !memcmp(rec_a, rec_b, size);
 616 }
 617
 618 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 619 {
 620         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 621                 return;
 622         bpf_timer_cancel_and_free(obj + rec->timer_off);
 623 }
 624
 625 extern void __bpf_obj_drop_impl(void *p, const struct btf_record *rec);
 626
 627 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 628 {
 629         const struct btf_field *fields;
 630         int i;
 631
 632         if (IS_ERR_OR_NULL(rec))
 633                 return;
 634         fields = rec->fields;
 635         for (i = 0; i < rec->cnt; i++) {
 636                 struct btf_struct_meta *pointee_struct_meta;
 637                 const struct btf_field *field = &fields[i];
 638                 void *field_ptr = obj + field->offset;
 639                 void *xchgd_field;
 640
 641                 switch (fields[i].type) {
 642                 case BPF_SPIN_LOCK:
 643                         break;
 644                 case BPF_TIMER:
 645                         bpf_timer_cancel_and_free(field_ptr);
 646                         break;
 647                 case BPF_KPTR_UNREF:
 648                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 649                         break;
 650                 case BPF_KPTR_REF:
 651                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 652                         if (!xchgd_field)
 653                                 break;
 654
 655                         if (!btf_is_kernel(field->kptr.btf)) {
 656                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 657                                                                            field->kptr.btf_id);
 658                                 WARN_ON_ONCE(!pointee_struct_meta);
 659                                 migrate_disable();
 660                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 661                                                                  pointee_struct_meta->record :
 662                                                                  NULL);
 663                                 migrate_enable();
 664                         } else {
 665                                 field->kptr.dtor(xchgd_field);
 666                         }
 667                         break;
 668                 case BPF_LIST_HEAD:
 669                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 670                                 continue;
 671                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 672                         break;
 673                 case BPF_RB_ROOT:
 674                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 675                                 continue;
 676                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 677                         break;
 678                 case BPF_LIST_NODE:
 679                 case BPF_RB_NODE:
 680                 case BPF_REFCOUNT:
 681                         break;
 682                 default:
 683                         WARN_ON_ONCE(1);
 684                         continue;
 685                 }
 686         }
 687 }
 688
 689 /* called from workqueue */
 690 static void bpf_map_free_deferred(struct work_struct *work)
 691 {
 692         struct bpf_map *map = container_of(work, struct bpf_map, work);
 693         struct btf_record *rec = map->record;
 694
 695         security_bpf_map_free(map);
 696         bpf_map_release_memcg(map);
 697         /* implementation dependent freeing */
 698         map->ops->map_free(map);
 699         /* Delay freeing of btf_record for maps, as map_free
 700          * callback usually needs access to them. It is better to do it here
 701          * than require each callback to do the free itself manually.
 702          *
 703          * Note that the btf_record stashed in map->inner_map_meta->record was
 704          * already freed using the map_free callback for map in map case which
 705          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 706          * template bpf_map struct used during verification.
 707          */
 708         btf_record_free(rec);
 709 }
 710
 711 static void bpf_map_put_uref(struct bpf_map *map)
 712 {
 713         if (atomic64_dec_and_test(&map->usercnt)) {
 714                 if (map->ops->map_release_uref)
 715                         map->ops->map_release_uref(map);
 716         }
 717 }
 718
 719 /* decrement map refcnt and schedule it for freeing via workqueue
 720  * (underlying map implementation ops->map_free() might sleep)
 721  */
 722 void bpf_map_put(struct bpf_map *map)
 723 {
 724         if (atomic64_dec_and_test(&map->refcnt)) {
 725                 /* bpf_map_free_id() must be called first */
 726                 bpf_map_free_id(map);
 727                 btf_put(map->btf);
 728                 INIT_WORK(&map->work, bpf_map_free_deferred);
 729                 /* Avoid spawning kworkers, since they all might contend
 730                  * for the same mutex like slab_mutex.
 731                  */
 732                 queue_work(system_unbound_wq, &map->work);
 733         }
 734 }
 735 EXPORT_SYMBOL_GPL(bpf_map_put);
 736
 737 void bpf_map_put_with_uref(struct bpf_map *map)
 738 {
 739         bpf_map_put_uref(map);
 740         bpf_map_put(map);
 741 }
 742
 743 static int bpf_map_release(struct inode *inode, struct file *filp)
 744 {
 745         struct bpf_map *map = filp->private_data;
 746
 747         if (map->ops->map_release)
 748                 map->ops->map_release(map, filp);
 749
 750         bpf_map_put_with_uref(map);
 751         return 0;
 752 }
 753
 754 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 755 {
 756         fmode_t mode = f.file->f_mode;
 757
 758         /* Our file permissions may have been overridden by global
 759          * map permissions facing syscall side.
 760          */
 761         if (READ_ONCE(map->frozen))
 762                 mode &= ~FMODE_CAN_WRITE;
 763         return mode;
 764 }
 765
 766 #ifdef CONFIG_PROC_FS
 767 /* Show the memory usage of a bpf map */
 768 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 769 {
 770         return map->ops->map_mem_usage(map);
 771 }
 772
 773 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 774 {
 775         struct bpf_map *map = filp->private_data;
 776         u32 type = 0, jited = 0;
 777
 778         if (map_type_contains_progs(map)) {
 779                 spin_lock(&map->owner.lock);
 780                 type  = map->owner.type;
 781                 jited = map->owner.jited;
 782                 spin_unlock(&map->owner.lock);
 783         }
 784
 785         seq_printf(m,
 786                    "map_type:\t%u\n"
 787                    "key_size:\t%u\n"
 788                    "value_size:\t%u\n"
 789                    "max_entries:\t%u\n"
 790                    "map_flags:\t%#x\n"
 791                    "map_extra:\t%#llx\n"
 792                    "memlock:\t%llu\n"
 793                    "map_id:\t%u\n"
 794                    "frozen:\t%u\n",
 795                    map->map_type,
 796                    map->key_size,
 797                    map->value_size,
 798                    map->max_entries,
 799                    map->map_flags,
 800                    (unsigned long long)map->map_extra,
 801                    bpf_map_memory_usage(map),
 802                    map->id,
 803                    READ_ONCE(map->frozen));
 804         if (type) {
 805                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 806                 seq_printf(m, "owner_jited:\t%u\n", jited);
 807         }
 808 }
 809 #endif
 810
 811 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 812                               loff_t *ppos)
 813 {
 814         /* We need this handler such that alloc_file() enables
 815          * f_mode with FMODE_CAN_READ.
 816          */
 817         return -EINVAL;
 818 }
 819
 820 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 821                                size_t siz, loff_t *ppos)
 822 {
 823         /* We need this handler such that alloc_file() enables
 824          * f_mode with FMODE_CAN_WRITE.
 825          */
 826         return -EINVAL;
 827 }
 828
 829 /* called for any extra memory-mapped regions (except initial) */
 830 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 831 {
 832         struct bpf_map *map = vma->vm_file->private_data;
 833
 834         if (vma->vm_flags & VM_MAYWRITE)
 835                 bpf_map_write_active_inc(map);
 836 }
 837
 838 /* called for all unmapped memory region (including initial) */
 839 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 840 {
 841         struct bpf_map *map = vma->vm_file->private_data;
 842
 843         if (vma->vm_flags & VM_MAYWRITE)
 844                 bpf_map_write_active_dec(map);
 845 }
 846
 847 static const struct vm_operations_struct bpf_map_default_vmops = {
 848         .open           = bpf_map_mmap_open,
 849         .close          = bpf_map_mmap_close,
 850 };
 851
 852 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 853 {
 854         struct bpf_map *map = filp->private_data;
 855         int err;
 856
 857         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 858                 return -ENOTSUPP;
 859
 860         if (!(vma->vm_flags & VM_SHARED))
 861                 return -EINVAL;
 862
 863         mutex_lock(&map->freeze_mutex);
 864
 865         if (vma->vm_flags & VM_WRITE) {
 866                 if (map->frozen) {
 867                         err = -EPERM;
 868                         goto out;
 869                 }
 870                 /* map is meant to be read-only, so do not allow mapping as
 871                  * writable, because it's possible to leak a writable page
 872                  * reference and allows user-space to still modify it after
 873                  * freezing, while verifier will assume contents do not change
 874                  */
 875                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 876                         err = -EACCES;
 877                         goto out;
 878                 }
 879         }
 880
 881         /* set default open/close callbacks */
 882         vma->vm_ops = &bpf_map_default_vmops;
 883         vma->vm_private_data = map;
 884         vm_flags_clear(vma, VM_MAYEXEC);
 885         if (!(vma->vm_flags & VM_WRITE))
 886                 /* disallow re-mapping with PROT_WRITE */
 887                 vm_flags_clear(vma, VM_MAYWRITE);
 888
 889         err = map->ops->map_mmap(map, vma);
 890         if (err)
 891                 goto out;
 892
 893         if (vma->vm_flags & VM_MAYWRITE)
 894                 bpf_map_write_active_inc(map);
 895 out:
 896         mutex_unlock(&map->freeze_mutex);
 897         return err;
 898 }
 899
 900 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 901 {
 902         struct bpf_map *map = filp->private_data;
 903
 904         if (map->ops->map_poll)
 905                 return map->ops->map_poll(map, filp, pts);
 906
 907         return EPOLLERR;
 908 }
 909
 910 const struct file_operations bpf_map_fops = {
 911 #ifdef CONFIG_PROC_FS
 912         .show_fdinfo    = bpf_map_show_fdinfo,
 913 #endif
 914         .release        = bpf_map_release,
 915         .read           = bpf_dummy_read,
 916         .write          = bpf_dummy_write,
 917         .mmap           = bpf_map_mmap,
 918         .poll           = bpf_map_poll,
 919 };
 920
 921 int bpf_map_new_fd(struct bpf_map *map, int flags)
 922 {
 923         int ret;
 924
 925         ret = security_bpf_map(map, OPEN_FMODE(flags));
 926         if (ret < 0)
 927                 return ret;
 928
 929         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 930                                 flags | O_CLOEXEC);
 931 }
 932
 933 int bpf_get_file_flag(int flags)
 934 {
 935         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 936                 return -EINVAL;
 937         if (flags & BPF_F_RDONLY)
 938                 return O_RDONLY;
 939         if (flags & BPF_F_WRONLY)
 940                 return O_WRONLY;
 941         return O_RDWR;
 942 }
 943
 944 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 945 #define CHECK_ATTR(CMD) \
 946         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 947                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 948                    sizeof(*attr) - \
 949                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 950                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 951
 952 /* dst and src must have at least "size" number of bytes.
 953  * Return strlen on success and < 0 on error.
 954  */
 955 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 956 {
 957         const char *end = src + size;
 958         const char *orig_src = src;
 959
 960         memset(dst, 0, size);
 961         /* Copy all isalnum(), '_' and '.' chars. */
 962         while (src < end && *src) {
 963                 if (!isalnum(*src) &&
 964                     *src != '_' && *src != '.')
 965                         return -EINVAL;
 966                 *dst++ = *src++;
 967         }
 968
 969         /* No '\0' found in "size" number of bytes */
 970         if (src == end)
 971                 return -EINVAL;
 972
 973         return src - orig_src;
 974 }
 975
 976 int map_check_no_btf(const struct bpf_map *map,
 977                      const struct btf *btf,
 978                      const struct btf_type *key_type,
 979                      const struct btf_type *value_type)
 980 {
 981         return -ENOTSUPP;
 982 }
 983
 984 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 985                          u32 btf_key_id, u32 btf_value_id)
 986 {
 987         const struct btf_type *key_type, *value_type;
 988         u32 key_size, value_size;
 989         int ret = 0;
 990
 991         /* Some maps allow key to be unspecified. */
 992         if (btf_key_id) {
 993                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 994                 if (!key_type || key_size != map->key_size)
 995                         return -EINVAL;
 996         } else {
 997                 key_type = btf_type_by_id(btf, 0);
 998                 if (!map->ops->map_check_btf)
 999                         return -EINVAL;
1000         }
1001
1002         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1003         if (!value_type || value_size != map->value_size)
1004                 return -EINVAL;
1005
1006         map->record = btf_parse_fields(btf, value_type,
1007                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1008                                        BPF_RB_ROOT | BPF_REFCOUNT,
1009                                        map->value_size);
1010         if (!IS_ERR_OR_NULL(map->record)) {
1011                 int i;
1012
1013                 if (!bpf_capable()) {
1014                         ret = -EPERM;
1015                         goto free_map_tab;
1016                 }
1017                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1018                         ret = -EACCES;
1019                         goto free_map_tab;
1020                 }
1021                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1022                         switch (map->record->field_mask & (1 << i)) {
1023                         case 0:
1024                                 continue;
1025                         case BPF_SPIN_LOCK:
1026                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1027                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1028                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1029                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1030                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1031                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1032                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1033                                         ret = -EOPNOTSUPP;
1034                                         goto free_map_tab;
1035                                 }
1036                                 break;
1037                         case BPF_TIMER:
1038                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1039                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1040                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1041                                         ret = -EOPNOTSUPP;
1042                                         goto free_map_tab;
1043                                 }
1044                                 break;
1045                         case BPF_KPTR_UNREF:
1046                         case BPF_KPTR_REF:
1047                         case BPF_REFCOUNT:
1048                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1049                                     map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1050                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1051                                     map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1052                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1053                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1054                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1055                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1056                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1057                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1058                                         ret = -EOPNOTSUPP;
1059                                         goto free_map_tab;
1060                                 }
1061                                 break;
1062                         case BPF_LIST_HEAD:
1063                         case BPF_RB_ROOT:
1064                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1065                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1066                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1067                                         ret = -EOPNOTSUPP;
1068                                         goto free_map_tab;
1069                                 }
1070                                 break;
1071                         default:
1072                                 /* Fail if map_type checks are missing for a field type */
1073                                 ret = -EOPNOTSUPP;
1074                                 goto free_map_tab;
1075                         }
1076                 }
1077         }
1078
1079         ret = btf_check_and_fixup_fields(btf, map->record);
1080         if (ret < 0)
1081                 goto free_map_tab;
1082
1083         if (map->ops->map_check_btf) {
1084                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1085                 if (ret < 0)
1086                         goto free_map_tab;
1087         }
1088
1089         return ret;
1090 free_map_tab:
1091         bpf_map_free_record(map);
1092         return ret;
1093 }
1094
1095 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1096 /* called via syscall */
1097 static int map_create(union bpf_attr *attr)
1098 {
1099         const struct bpf_map_ops *ops;
1100         int numa_node = bpf_map_attr_numa_node(attr);
1101         u32 map_type = attr->map_type;
1102         struct bpf_map *map;
1103         int f_flags;
1104         int err;
1105
1106         err = CHECK_ATTR(BPF_MAP_CREATE);
1107         if (err)
1108                 return -EINVAL;
1109
1110         if (attr->btf_vmlinux_value_type_id) {
1111                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1112                     attr->btf_key_type_id || attr->btf_value_type_id)
1113                         return -EINVAL;
1114         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1115                 return -EINVAL;
1116         }
1117
1118         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1119             attr->map_extra != 0)
1120                 return -EINVAL;
1121
1122         f_flags = bpf_get_file_flag(attr->map_flags);
1123         if (f_flags < 0)
1124                 return f_flags;
1125
1126         if (numa_node != NUMA_NO_NODE &&
1127             ((unsigned int)numa_node >= nr_node_ids ||
1128              !node_online(numa_node)))
1129                 return -EINVAL;
1130
1131         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1132         map_type = attr->map_type;
1133         if (map_type >= ARRAY_SIZE(bpf_map_types))
1134                 return -EINVAL;
1135         map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1136         ops = bpf_map_types[map_type];
1137         if (!ops)
1138                 return -EINVAL;
1139
1140         if (ops->map_alloc_check) {
1141                 err = ops->map_alloc_check(attr);
1142                 if (err)
1143                         return err;
1144         }
1145         if (attr->map_ifindex)
1146                 ops = &bpf_map_offload_ops;
1147         if (!ops->map_mem_usage)
1148                 return -EINVAL;
1149
1150         /* Intent here is for unprivileged_bpf_disabled to block BPF map
1151          * creation for unprivileged users; other actions depend
1152          * on fd availability and access to bpffs, so are dependent on
1153          * object creation success. Even with unprivileged BPF disabled,
1154          * capability checks are still carried out.
1155          */
1156         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
1157                 return -EPERM;
1158
1159         /* check privileged map type permissions */
1160         switch (map_type) {
1161         case BPF_MAP_TYPE_ARRAY:
1162         case BPF_MAP_TYPE_PERCPU_ARRAY:
1163         case BPF_MAP_TYPE_PROG_ARRAY:
1164         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1165         case BPF_MAP_TYPE_CGROUP_ARRAY:
1166         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1167         case BPF_MAP_TYPE_HASH:
1168         case BPF_MAP_TYPE_PERCPU_HASH:
1169         case BPF_MAP_TYPE_HASH_OF_MAPS:
1170         case BPF_MAP_TYPE_RINGBUF:
1171         case BPF_MAP_TYPE_USER_RINGBUF:
1172         case BPF_MAP_TYPE_CGROUP_STORAGE:
1173         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1174                 /* unprivileged */
1175                 break;
1176         case BPF_MAP_TYPE_SK_STORAGE:
1177         case BPF_MAP_TYPE_INODE_STORAGE:
1178         case BPF_MAP_TYPE_TASK_STORAGE:
1179         case BPF_MAP_TYPE_CGRP_STORAGE:
1180         case BPF_MAP_TYPE_BLOOM_FILTER:
1181         case BPF_MAP_TYPE_LPM_TRIE:
1182         case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1183         case BPF_MAP_TYPE_STACK_TRACE:
1184         case BPF_MAP_TYPE_QUEUE:
1185         case BPF_MAP_TYPE_STACK:
1186         case BPF_MAP_TYPE_LRU_HASH:
1187         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1188         case BPF_MAP_TYPE_STRUCT_OPS:
1189         case BPF_MAP_TYPE_CPUMAP:
1190                 if (!bpf_capable())
1191                         return -EPERM;
1192                 break;
1193         case BPF_MAP_TYPE_SOCKMAP:
1194         case BPF_MAP_TYPE_SOCKHASH:
1195         case BPF_MAP_TYPE_DEVMAP:
1196         case BPF_MAP_TYPE_DEVMAP_HASH:
1197         case BPF_MAP_TYPE_XSKMAP:
1198                 if (!capable(CAP_NET_ADMIN))
1199                         return -EPERM;
1200                 break;
1201         default:
1202                 WARN(1, "unsupported map type %d", map_type);
1203                 return -EPERM;
1204         }
1205
1206         map = ops->map_alloc(attr);
1207         if (IS_ERR(map))
1208                 return PTR_ERR(map);
1209         map->ops = ops;
1210         map->map_type = map_type;
1211
1212         err = bpf_obj_name_cpy(map->name, attr->map_name,
1213                                sizeof(attr->map_name));
1214         if (err < 0)
1215                 goto free_map;
1216
1217         atomic64_set(&map->refcnt, 1);
1218         atomic64_set(&map->usercnt, 1);
1219         mutex_init(&map->freeze_mutex);
1220         spin_lock_init(&map->owner.lock);
1221
1222         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1223             /* Even the map's value is a kernel's struct,
1224              * the bpf_prog.o must have BTF to begin with
1225              * to figure out the corresponding kernel's
1226              * counter part.  Thus, attr->btf_fd has
1227              * to be valid also.
1228              */
1229             attr->btf_vmlinux_value_type_id) {
1230                 struct btf *btf;
1231
1232                 btf = btf_get_by_fd(attr->btf_fd);
1233                 if (IS_ERR(btf)) {
1234                         err = PTR_ERR(btf);
1235                         goto free_map;
1236                 }
1237                 if (btf_is_kernel(btf)) {
1238                         btf_put(btf);
1239                         err = -EACCES;
1240                         goto free_map;
1241                 }
1242                 map->btf = btf;
1243
1244                 if (attr->btf_value_type_id) {
1245                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1246                                             attr->btf_value_type_id);
1247                         if (err)
1248                                 goto free_map;
1249                 }
1250
1251                 map->btf_key_type_id = attr->btf_key_type_id;
1252                 map->btf_value_type_id = attr->btf_value_type_id;
1253                 map->btf_vmlinux_value_type_id =
1254                         attr->btf_vmlinux_value_type_id;
1255         }
1256
1257         err = security_bpf_map_alloc(map);
1258         if (err)
1259                 goto free_map;
1260
1261         err = bpf_map_alloc_id(map);
1262         if (err)
1263                 goto free_map_sec;
1264
1265         bpf_map_save_memcg(map);
1266
1267         err = bpf_map_new_fd(map, f_flags);
1268         if (err < 0) {
1269                 /* failed to allocate fd.
1270                  * bpf_map_put_with_uref() is needed because the above
1271                  * bpf_map_alloc_id() has published the map
1272                  * to the userspace and the userspace may
1273                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1274                  */
1275                 bpf_map_put_with_uref(map);
1276                 return err;
1277         }
1278
1279         return err;
1280
1281 free_map_sec:
1282         security_bpf_map_free(map);
1283 free_map:
1284         btf_put(map->btf);
1285         map->ops->map_free(map);
1286         return err;
1287 }
1288
1289 /* if error is returned, fd is released.
1290  * On success caller should complete fd access with matching fdput()
1291  */
1292 struct bpf_map *__bpf_map_get(struct fd f)
1293 {
1294         if (!f.file)
1295                 return ERR_PTR(-EBADF);
1296         if (f.file->f_op != &bpf_map_fops) {
1297                 fdput(f);
1298                 return ERR_PTR(-EINVAL);
1299         }
1300
1301         return f.file->private_data;
1302 }
1303
1304 void bpf_map_inc(struct bpf_map *map)
1305 {
1306         atomic64_inc(&map->refcnt);
1307 }
1308 EXPORT_SYMBOL_GPL(bpf_map_inc);
1309
1310 void bpf_map_inc_with_uref(struct bpf_map *map)
1311 {
1312         atomic64_inc(&map->refcnt);
1313         atomic64_inc(&map->usercnt);
1314 }
1315 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1316
1317 struct bpf_map *bpf_map_get(u32 ufd)
1318 {
1319         struct fd f = fdget(ufd);
1320         struct bpf_map *map;
1321
1322         map = __bpf_map_get(f);
1323         if (IS_ERR(map))
1324                 return map;
1325
1326         bpf_map_inc(map);
1327         fdput(f);
1328
1329         return map;
1330 }
1331 EXPORT_SYMBOL(bpf_map_get);
1332
1333 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1334 {
1335         struct fd f = fdget(ufd);
1336         struct bpf_map *map;
1337
1338         map = __bpf_map_get(f);
1339         if (IS_ERR(map))
1340                 return map;
1341
1342         bpf_map_inc_with_uref(map);
1343         fdput(f);
1344
1345         return map;
1346 }
1347
1348 /* map_idr_lock should have been held or the map should have been
1349  * protected by rcu read lock.
1350  */
1351 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1352 {
1353         int refold;
1354
1355         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1356         if (!refold)
1357                 return ERR_PTR(-ENOENT);
1358         if (uref)
1359                 atomic64_inc(&map->usercnt);
1360
1361         return map;
1362 }
1363
1364 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1365 {
1366         spin_lock_bh(&map_idr_lock);
1367         map = __bpf_map_inc_not_zero(map, false);
1368         spin_unlock_bh(&map_idr_lock);
1369
1370         return map;
1371 }
1372 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1373
1374 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1375 {
1376         return -ENOTSUPP;
1377 }
1378
1379 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1380 {
1381         if (key_size)
1382                 return vmemdup_user(ukey, key_size);
1383
1384         if (ukey)
1385                 return ERR_PTR(-EINVAL);
1386
1387         return NULL;
1388 }
1389
1390 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1391 {
1392         if (key_size)
1393                 return kvmemdup_bpfptr(ukey, key_size);
1394
1395         if (!bpfptr_is_null(ukey))
1396                 return ERR_PTR(-EINVAL);
1397
1398         return NULL;
1399 }
1400
1401 /* last field in 'union bpf_attr' used by this command */
1402 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1403
1404 static int map_lookup_elem(union bpf_attr *attr)
1405 {
1406         void __user *ukey = u64_to_user_ptr(attr->key);
1407         void __user *uvalue = u64_to_user_ptr(attr->value);
1408         int ufd = attr->map_fd;
1409         struct bpf_map *map;
1410         void *key, *value;
1411         u32 value_size;
1412         struct fd f;
1413         int err;
1414
1415         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1416                 return -EINVAL;
1417
1418         if (attr->flags & ~BPF_F_LOCK)
1419                 return -EINVAL;
1420
1421         f = fdget(ufd);
1422         map = __bpf_map_get(f);
1423         if (IS_ERR(map))
1424                 return PTR_ERR(map);
1425         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1426                 err = -EPERM;
1427                 goto err_put;
1428         }
1429
1430         if ((attr->flags & BPF_F_LOCK) &&
1431             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1432                 err = -EINVAL;
1433                 goto err_put;
1434         }
1435
1436         key = __bpf_copy_key(ukey, map->key_size);
1437         if (IS_ERR(key)) {
1438                 err = PTR_ERR(key);
1439                 goto err_put;
1440         }
1441
1442         value_size = bpf_map_value_size(map);
1443
1444         err = -ENOMEM;
1445         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1446         if (!value)
1447                 goto free_key;
1448
1449         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1450                 if (copy_from_user(value, uvalue, value_size))
1451                         err = -EFAULT;
1452                 else
1453                         err = bpf_map_copy_value(map, key, value, attr->flags);
1454                 goto free_value;
1455         }
1456
1457         err = bpf_map_copy_value(map, key, value, attr->flags);
1458         if (err)
1459                 goto free_value;
1460
1461         err = -EFAULT;
1462         if (copy_to_user(uvalue, value, value_size) != 0)
1463                 goto free_value;
1464
1465         err = 0;
1466
1467 free_value:
1468         kvfree(value);
1469 free_key:
1470         kvfree(key);
1471 err_put:
1472         fdput(f);
1473         return err;
1474 }
1475
1476
1477 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1478
1479 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1480 {
1481         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1482         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1483         int ufd = attr->map_fd;
1484         struct bpf_map *map;
1485         void *key, *value;
1486         u32 value_size;
1487         struct fd f;
1488         int err;
1489
1490         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1491                 return -EINVAL;
1492
1493         f = fdget(ufd);
1494         map = __bpf_map_get(f);
1495         if (IS_ERR(map))
1496                 return PTR_ERR(map);
1497         bpf_map_write_active_inc(map);
1498         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1499                 err = -EPERM;
1500                 goto err_put;
1501         }
1502
1503         if ((attr->flags & BPF_F_LOCK) &&
1504             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1505                 err = -EINVAL;
1506                 goto err_put;
1507         }
1508
1509         key = ___bpf_copy_key(ukey, map->key_size);
1510         if (IS_ERR(key)) {
1511                 err = PTR_ERR(key);
1512                 goto err_put;
1513         }
1514
1515         value_size = bpf_map_value_size(map);
1516         value = kvmemdup_bpfptr(uvalue, value_size);
1517         if (IS_ERR(value)) {
1518                 err = PTR_ERR(value);
1519                 goto free_key;
1520         }
1521
1522         err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1523
1524         kvfree(value);
1525 free_key:
1526         kvfree(key);
1527 err_put:
1528         bpf_map_write_active_dec(map);
1529         fdput(f);
1530         return err;
1531 }
1532
1533 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1534
1535 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1536 {
1537         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1538         int ufd = attr->map_fd;
1539         struct bpf_map *map;
1540         struct fd f;
1541         void *key;
1542         int err;
1543
1544         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1545                 return -EINVAL;
1546
1547         f = fdget(ufd);
1548         map = __bpf_map_get(f);
1549         if (IS_ERR(map))
1550                 return PTR_ERR(map);
1551         bpf_map_write_active_inc(map);
1552         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1553                 err = -EPERM;
1554                 goto err_put;
1555         }
1556
1557         key = ___bpf_copy_key(ukey, map->key_size);
1558         if (IS_ERR(key)) {
1559                 err = PTR_ERR(key);
1560                 goto err_put;
1561         }
1562
1563         if (bpf_map_is_offloaded(map)) {
1564                 err = bpf_map_offload_delete_elem(map, key);
1565                 goto out;
1566         } else if (IS_FD_PROG_ARRAY(map) ||
1567                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1568                 /* These maps require sleepable context */
1569                 err = map->ops->map_delete_elem(map, key);
1570                 goto out;
1571         }
1572
1573         bpf_disable_instrumentation();
1574         rcu_read_lock();
1575         err = map->ops->map_delete_elem(map, key);
1576         rcu_read_unlock();
1577         bpf_enable_instrumentation();
1578         maybe_wait_bpf_programs(map);
1579 out:
1580         kvfree(key);
1581 err_put:
1582         bpf_map_write_active_dec(map);
1583         fdput(f);
1584         return err;
1585 }
1586
1587 /* last field in 'union bpf_attr' used by this command */
1588 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1589
1590 static int map_get_next_key(union bpf_attr *attr)
1591 {
1592         void __user *ukey = u64_to_user_ptr(attr->key);
1593         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1594         int ufd = attr->map_fd;
1595         struct bpf_map *map;
1596         void *key, *next_key;
1597         struct fd f;
1598         int err;
1599
1600         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1601                 return -EINVAL;
1602
1603         f = fdget(ufd);
1604         map = __bpf_map_get(f);
1605         if (IS_ERR(map))
1606                 return PTR_ERR(map);
1607         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1608                 err = -EPERM;
1609                 goto err_put;
1610         }
1611
1612         if (ukey) {
1613                 key = __bpf_copy_key(ukey, map->key_size);
1614                 if (IS_ERR(key)) {
1615                         err = PTR_ERR(key);
1616                         goto err_put;
1617                 }
1618         } else {
1619                 key = NULL;
1620         }
1621
1622         err = -ENOMEM;
1623         next_key = kvmalloc(map->key_size, GFP_USER);
1624         if (!next_key)
1625                 goto free_key;
1626
1627         if (bpf_map_is_offloaded(map)) {
1628                 err = bpf_map_offload_get_next_key(map, key, next_key);
1629                 goto out;
1630         }
1631
1632         rcu_read_lock();
1633         err = map->ops->map_get_next_key(map, key, next_key);
1634         rcu_read_unlock();
1635 out:
1636         if (err)
1637                 goto free_next_key;
1638
1639         err = -EFAULT;
1640         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1641                 goto free_next_key;
1642
1643         err = 0;
1644
1645 free_next_key:
1646         kvfree(next_key);
1647 free_key:
1648         kvfree(key);
1649 err_put:
1650         fdput(f);
1651         return err;
1652 }
1653
1654 int generic_map_delete_batch(struct bpf_map *map,
1655                              const union bpf_attr *attr,
1656                              union bpf_attr __user *uattr)
1657 {
1658         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1659         u32 cp, max_count;
1660         int err = 0;
1661         void *key;
1662
1663         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1664                 return -EINVAL;
1665
1666         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1667             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1668                 return -EINVAL;
1669         }
1670
1671         max_count = attr->batch.count;
1672         if (!max_count)
1673                 return 0;
1674
1675         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1676         if (!key)
1677                 return -ENOMEM;
1678
1679         for (cp = 0; cp < max_count; cp++) {
1680                 err = -EFAULT;
1681                 if (copy_from_user(key, keys + cp * map->key_size,
1682                                    map->key_size))
1683                         break;
1684
1685                 if (bpf_map_is_offloaded(map)) {
1686                         err = bpf_map_offload_delete_elem(map, key);
1687                         break;
1688                 }
1689
1690                 bpf_disable_instrumentation();
1691                 rcu_read_lock();
1692                 err = map->ops->map_delete_elem(map, key);
1693                 rcu_read_unlock();
1694                 bpf_enable_instrumentation();
1695                 if (err)
1696                         break;
1697                 cond_resched();
1698         }
1699         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1700                 err = -EFAULT;
1701
1702         kvfree(key);
1703
1704         maybe_wait_bpf_programs(map);
1705         return err;
1706 }
1707
1708 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1709                              const union bpf_attr *attr,
1710                              union bpf_attr __user *uattr)
1711 {
1712         void __user *values = u64_to_user_ptr(attr->batch.values);
1713         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1714         u32 value_size, cp, max_count;
1715         void *key, *value;
1716         int err = 0;
1717
1718         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1719                 return -EINVAL;
1720
1721         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1722             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1723                 return -EINVAL;
1724         }
1725
1726         value_size = bpf_map_value_size(map);
1727
1728         max_count = attr->batch.count;
1729         if (!max_count)
1730                 return 0;
1731
1732         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1733         if (!key)
1734                 return -ENOMEM;
1735
1736         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1737         if (!value) {
1738                 kvfree(key);
1739                 return -ENOMEM;
1740         }
1741
1742         for (cp = 0; cp < max_count; cp++) {
1743                 err = -EFAULT;
1744                 if (copy_from_user(key, keys + cp * map->key_size,
1745                     map->key_size) ||
1746                     copy_from_user(value, values + cp * value_size, value_size))
1747                         break;
1748
1749                 err = bpf_map_update_value(map, map_file, key, value,
1750                                            attr->batch.elem_flags);
1751
1752                 if (err)
1753                         break;
1754                 cond_resched();
1755         }
1756
1757         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1758                 err = -EFAULT;
1759
1760         kvfree(value);
1761         kvfree(key);
1762         return err;
1763 }
1764
1765 #define MAP_LOOKUP_RETRIES 3
1766
1767 int generic_map_lookup_batch(struct bpf_map *map,
1768                                     const union bpf_attr *attr,
1769                                     union bpf_attr __user *uattr)
1770 {
1771         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1772         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1773         void __user *values = u64_to_user_ptr(attr->batch.values);
1774         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1775         void *buf, *buf_prevkey, *prev_key, *key, *value;
1776         int err, retry = MAP_LOOKUP_RETRIES;
1777         u32 value_size, cp, max_count;
1778
1779         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1780                 return -EINVAL;
1781
1782         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1783             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1784                 return -EINVAL;
1785
1786         value_size = bpf_map_value_size(map);
1787
1788         max_count = attr->batch.count;
1789         if (!max_count)
1790                 return 0;
1791
1792         if (put_user(0, &uattr->batch.count))
1793                 return -EFAULT;
1794
1795         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1796         if (!buf_prevkey)
1797                 return -ENOMEM;
1798
1799         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1800         if (!buf) {
1801                 kvfree(buf_prevkey);
1802                 return -ENOMEM;
1803         }
1804
1805         err = -EFAULT;
1806         prev_key = NULL;
1807         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1808                 goto free_buf;
1809         key = buf;
1810         value = key + map->key_size;
1811         if (ubatch)
1812                 prev_key = buf_prevkey;
1813
1814         for (cp = 0; cp < max_count;) {
1815                 rcu_read_lock();
1816                 err = map->ops->map_get_next_key(map, prev_key, key);
1817                 rcu_read_unlock();
1818                 if (err)
1819                         break;
1820                 err = bpf_map_copy_value(map, key, value,
1821                                          attr->batch.elem_flags);
1822
1823                 if (err == -ENOENT) {
1824                         if (retry) {
1825                                 retry--;
1826                                 continue;
1827                         }
1828                         err = -EINTR;
1829                         break;
1830                 }
1831
1832                 if (err)
1833                         goto free_buf;
1834
1835                 if (copy_to_user(keys + cp * map->key_size, key,
1836                                  map->key_size)) {
1837                         err = -EFAULT;
1838                         goto free_buf;
1839                 }
1840                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1841                         err = -EFAULT;
1842                         goto free_buf;
1843                 }
1844
1845                 if (!prev_key)
1846                         prev_key = buf_prevkey;
1847
1848                 swap(prev_key, key);
1849                 retry = MAP_LOOKUP_RETRIES;
1850                 cp++;
1851                 cond_resched();
1852         }
1853
1854         if (err == -EFAULT)
1855                 goto free_buf;
1856
1857         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1858                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1859                 err = -EFAULT;
1860
1861 free_buf:
1862         kvfree(buf_prevkey);
1863         kvfree(buf);
1864         return err;
1865 }
1866
1867 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1868
1869 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1870 {
1871         void __user *ukey = u64_to_user_ptr(attr->key);
1872         void __user *uvalue = u64_to_user_ptr(attr->value);
1873         int ufd = attr->map_fd;
1874         struct bpf_map *map;
1875         void *key, *value;
1876         u32 value_size;
1877         struct fd f;
1878         int err;
1879
1880         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1881                 return -EINVAL;
1882
1883         if (attr->flags & ~BPF_F_LOCK)
1884                 return -EINVAL;
1885
1886         f = fdget(ufd);
1887         map = __bpf_map_get(f);
1888         if (IS_ERR(map))
1889                 return PTR_ERR(map);
1890         bpf_map_write_active_inc(map);
1891         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1892             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1893                 err = -EPERM;
1894                 goto err_put;
1895         }
1896
1897         if (attr->flags &&
1898             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1899              map->map_type == BPF_MAP_TYPE_STACK)) {
1900                 err = -EINVAL;
1901                 goto err_put;
1902         }
1903
1904         if ((attr->flags & BPF_F_LOCK) &&
1905             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1906                 err = -EINVAL;
1907                 goto err_put;
1908         }
1909
1910         key = __bpf_copy_key(ukey, map->key_size);
1911         if (IS_ERR(key)) {
1912                 err = PTR_ERR(key);
1913                 goto err_put;
1914         }
1915
1916         value_size = bpf_map_value_size(map);
1917
1918         err = -ENOMEM;
1919         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1920         if (!value)
1921                 goto free_key;
1922
1923         err = -ENOTSUPP;
1924         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1925             map->map_type == BPF_MAP_TYPE_STACK) {
1926                 err = map->ops->map_pop_elem(map, value);
1927         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1928                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1929                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1930                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1931                 if (!bpf_map_is_offloaded(map)) {
1932                         bpf_disable_instrumentation();
1933                         rcu_read_lock();
1934                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1935                         rcu_read_unlock();
1936                         bpf_enable_instrumentation();
1937                 }
1938         }
1939
1940         if (err)
1941                 goto free_value;
1942
1943         if (copy_to_user(uvalue, value, value_size) != 0) {
1944                 err = -EFAULT;
1945                 goto free_value;
1946         }
1947
1948         err = 0;
1949
1950 free_value:
1951         kvfree(value);
1952 free_key:
1953         kvfree(key);
1954 err_put:
1955         bpf_map_write_active_dec(map);
1956         fdput(f);
1957         return err;
1958 }
1959
1960 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1961
1962 static int map_freeze(const union bpf_attr *attr)
1963 {
1964         int err = 0, ufd = attr->map_fd;
1965         struct bpf_map *map;
1966         struct fd f;
1967
1968         if (CHECK_ATTR(BPF_MAP_FREEZE))
1969                 return -EINVAL;
1970
1971         f = fdget(ufd);
1972         map = __bpf_map_get(f);
1973         if (IS_ERR(map))
1974                 return PTR_ERR(map);
1975
1976         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1977                 fdput(f);
1978                 return -ENOTSUPP;
1979         }
1980
1981         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1982                 fdput(f);
1983                 return -EPERM;
1984         }
1985
1986         mutex_lock(&map->freeze_mutex);
1987         if (bpf_map_write_active(map)) {
1988                 err = -EBUSY;
1989                 goto err_put;
1990         }
1991         if (READ_ONCE(map->frozen)) {
1992                 err = -EBUSY;
1993                 goto err_put;
1994         }
1995
1996         WRITE_ONCE(map->frozen, true);
1997 err_put:
1998         mutex_unlock(&map->freeze_mutex);
1999         fdput(f);
2000         return err;
2001 }
2002
2003 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2004 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2005         [_id] = & _name ## _prog_ops,
2006 #define BPF_MAP_TYPE(_id, _ops)
2007 #define BPF_LINK_TYPE(_id, _name)
2008 #include <linux/bpf_types.h>
2009 #undef BPF_PROG_TYPE
2010 #undef BPF_MAP_TYPE
2011 #undef BPF_LINK_TYPE
2012 };
2013
2014 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2015 {
2016         const struct bpf_prog_ops *ops;
2017
2018         if (type >= ARRAY_SIZE(bpf_prog_types))
2019                 return -EINVAL;
2020         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2021         ops = bpf_prog_types[type];
2022         if (!ops)
2023                 return -EINVAL;
2024
2025         if (!bpf_prog_is_offloaded(prog->aux))
2026                 prog->aux->ops = ops;
2027         else
2028                 prog->aux->ops = &bpf_offload_prog_ops;
2029         prog->type = type;
2030         return 0;
2031 }
2032
2033 enum bpf_audit {
2034         BPF_AUDIT_LOAD,
2035         BPF_AUDIT_UNLOAD,
2036         BPF_AUDIT_MAX,
2037 };
2038
2039 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2040         [BPF_AUDIT_LOAD]   = "LOAD",
2041         [BPF_AUDIT_UNLOAD] = "UNLOAD",
2042 };
2043
2044 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2045 {
2046         struct audit_context *ctx = NULL;
2047         struct audit_buffer *ab;
2048
2049         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2050                 return;
2051         if (audit_enabled == AUDIT_OFF)
2052                 return;
2053         if (!in_irq() && !irqs_disabled())
2054                 ctx = audit_context();
2055         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2056         if (unlikely(!ab))
2057                 return;
2058         audit_log_format(ab, "prog-id=%u op=%s",
2059                          prog->aux->id, bpf_audit_str[op]);
2060         audit_log_end(ab);
2061 }
2062
2063 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2064 {
2065         int id;
2066
2067         idr_preload(GFP_KERNEL);
2068         spin_lock_bh(&prog_idr_lock);
2069         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2070         if (id > 0)
2071                 prog->aux->id = id;
2072         spin_unlock_bh(&prog_idr_lock);
2073         idr_preload_end();
2074
2075         /* id is in [1, INT_MAX) */
2076         if (WARN_ON_ONCE(!id))
2077                 return -ENOSPC;
2078
2079         return id > 0 ? 0 : id;
2080 }
2081
2082 void bpf_prog_free_id(struct bpf_prog *prog)
2083 {
2084         unsigned long flags;
2085
2086         /* cBPF to eBPF migrations are currently not in the idr store.
2087          * Offloaded programs are removed from the store when their device
2088          * disappears - even if someone grabs an fd to them they are unusable,
2089          * simply waiting for refcnt to drop to be freed.
2090          */
2091         if (!prog->aux->id)
2092                 return;
2093
2094         spin_lock_irqsave(&prog_idr_lock, flags);
2095         idr_remove(&prog_idr, prog->aux->id);
2096         prog->aux->id = 0;
2097         spin_unlock_irqrestore(&prog_idr_lock, flags);
2098 }
2099
2100 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2101 {
2102         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2103
2104         kvfree(aux->func_info);
2105         kfree(aux->func_info_aux);
2106         free_uid(aux->user);
2107         security_bpf_prog_free(aux);
2108         bpf_prog_free(aux->prog);
2109 }
2110
2111 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2112 {
2113         bpf_prog_kallsyms_del_all(prog);
2114         btf_put(prog->aux->btf);
2115         module_put(prog->aux->mod);
2116         kvfree(prog->aux->jited_linfo);
2117         kvfree(prog->aux->linfo);
2118         kfree(prog->aux->kfunc_tab);
2119         if (prog->aux->attach_btf)
2120                 btf_put(prog->aux->attach_btf);
2121
2122         if (deferred) {
2123                 if (prog->aux->sleepable)
2124                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2125                 else
2126                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2127         } else {
2128                 __bpf_prog_put_rcu(&prog->aux->rcu);
2129         }
2130 }
2131
2132 static void bpf_prog_put_deferred(struct work_struct *work)
2133 {
2134         struct bpf_prog_aux *aux;
2135         struct bpf_prog *prog;
2136
2137         aux = container_of(work, struct bpf_prog_aux, work);
2138         prog = aux->prog;
2139         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2140         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2141         bpf_prog_free_id(prog);
2142         __bpf_prog_put_noref(prog, true);
2143 }
2144
2145 static void __bpf_prog_put(struct bpf_prog *prog)
2146 {
2147         struct bpf_prog_aux *aux = prog->aux;
2148
2149         if (atomic64_dec_and_test(&aux->refcnt)) {
2150                 if (in_irq() || irqs_disabled()) {
2151                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2152                         schedule_work(&aux->work);
2153                 } else {
2154                         bpf_prog_put_deferred(&aux->work);
2155                 }
2156         }
2157 }
2158
2159 void bpf_prog_put(struct bpf_prog *prog)
2160 {
2161         __bpf_prog_put(prog);
2162 }
2163 EXPORT_SYMBOL_GPL(bpf_prog_put);
2164
2165 static int bpf_prog_release(struct inode *inode, struct file *filp)
2166 {
2167         struct bpf_prog *prog = filp->private_data;
2168
2169         bpf_prog_put(prog);
2170         return 0;
2171 }
2172
2173 struct bpf_prog_kstats {
2174         u64 nsecs;
2175         u64 cnt;
2176         u64 misses;
2177 };
2178
2179 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2180 {
2181         struct bpf_prog_stats *stats;
2182         unsigned int flags;
2183
2184         stats = this_cpu_ptr(prog->stats);
2185         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2186         u64_stats_inc(&stats->misses);
2187         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2188 }
2189
2190 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2191                                struct bpf_prog_kstats *stats)
2192 {
2193         u64 nsecs = 0, cnt = 0, misses = 0;
2194         int cpu;
2195
2196         for_each_possible_cpu(cpu) {
2197                 const struct bpf_prog_stats *st;
2198                 unsigned int start;
2199                 u64 tnsecs, tcnt, tmisses;
2200
2201                 st = per_cpu_ptr(prog->stats, cpu);
2202                 do {
2203                         start = u64_stats_fetch_begin(&st->syncp);
2204                         tnsecs = u64_stats_read(&st->nsecs);
2205                         tcnt = u64_stats_read(&st->cnt);
2206                         tmisses = u64_stats_read(&st->misses);
2207                 } while (u64_stats_fetch_retry(&st->syncp, start));
2208                 nsecs += tnsecs;
2209                 cnt += tcnt;
2210                 misses += tmisses;
2211         }
2212         stats->nsecs = nsecs;
2213         stats->cnt = cnt;
2214         stats->misses = misses;
2215 }
2216
2217 #ifdef CONFIG_PROC_FS
2218 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2219 {
2220         const struct bpf_prog *prog = filp->private_data;
2221         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2222         struct bpf_prog_kstats stats;
2223
2224         bpf_prog_get_stats(prog, &stats);
2225         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2226         seq_printf(m,
2227                    "prog_type:\t%u\n"
2228                    "prog_jited:\t%u\n"
2229                    "prog_tag:\t%s\n"
2230                    "memlock:\t%llu\n"
2231                    "prog_id:\t%u\n"
2232                    "run_time_ns:\t%llu\n"
2233                    "run_cnt:\t%llu\n"
2234                    "recursion_misses:\t%llu\n"
2235                    "verified_insns:\t%u\n",
2236                    prog->type,
2237                    prog->jited,
2238                    prog_tag,
2239                    prog->pages * 1ULL << PAGE_SHIFT,
2240                    prog->aux->id,
2241                    stats.nsecs,
2242                    stats.cnt,
2243                    stats.misses,
2244                    prog->aux->verified_insns);
2245 }
2246 #endif
2247
2248 const struct file_operations bpf_prog_fops = {
2249 #ifdef CONFIG_PROC_FS
2250         .show_fdinfo    = bpf_prog_show_fdinfo,
2251 #endif
2252         .release        = bpf_prog_release,
2253         .read           = bpf_dummy_read,
2254         .write          = bpf_dummy_write,
2255 };
2256
2257 int bpf_prog_new_fd(struct bpf_prog *prog)
2258 {
2259         int ret;
2260
2261         ret = security_bpf_prog(prog);
2262         if (ret < 0)
2263                 return ret;
2264
2265         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2266                                 O_RDWR | O_CLOEXEC);
2267 }
2268
2269 static struct bpf_prog *____bpf_prog_get(struct fd f)
2270 {
2271         if (!f.file)
2272                 return ERR_PTR(-EBADF);
2273         if (f.file->f_op != &bpf_prog_fops) {
2274                 fdput(f);
2275                 return ERR_PTR(-EINVAL);
2276         }
2277
2278         return f.file->private_data;
2279 }
2280
2281 void bpf_prog_add(struct bpf_prog *prog, int i)
2282 {
2283         atomic64_add(i, &prog->aux->refcnt);
2284 }
2285 EXPORT_SYMBOL_GPL(bpf_prog_add);
2286
2287 void bpf_prog_sub(struct bpf_prog *prog, int i)
2288 {
2289         /* Only to be used for undoing previous bpf_prog_add() in some
2290          * error path. We still know that another entity in our call
2291          * path holds a reference to the program, thus atomic_sub() can
2292          * be safely used in such cases!
2293          */
2294         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2295 }
2296 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2297
2298 void bpf_prog_inc(struct bpf_prog *prog)
2299 {
2300         atomic64_inc(&prog->aux->refcnt);
2301 }
2302 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2303
2304 /* prog_idr_lock should have been held */
2305 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2306 {
2307         int refold;
2308
2309         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2310
2311         if (!refold)
2312                 return ERR_PTR(-ENOENT);
2313
2314         return prog;
2315 }
2316 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2317
2318 bool bpf_prog_get_ok(struct bpf_prog *prog,
2319                             enum bpf_prog_type *attach_type, bool attach_drv)
2320 {
2321         /* not an attachment, just a refcount inc, always allow */
2322         if (!attach_type)
2323                 return true;
2324
2325         if (prog->type != *attach_type)
2326                 return false;
2327         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2328                 return false;
2329
2330         return true;
2331 }
2332
2333 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2334                                        bool attach_drv)
2335 {
2336         struct fd f = fdget(ufd);
2337         struct bpf_prog *prog;
2338
2339         prog = ____bpf_prog_get(f);
2340         if (IS_ERR(prog))
2341                 return prog;
2342         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2343                 prog = ERR_PTR(-EINVAL);
2344                 goto out;
2345         }
2346
2347         bpf_prog_inc(prog);
2348 out:
2349         fdput(f);
2350         return prog;
2351 }
2352
2353 struct bpf_prog *bpf_prog_get(u32 ufd)
2354 {
2355         return __bpf_prog_get(ufd, NULL, false);
2356 }
2357
2358 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2359                                        bool attach_drv)
2360 {
2361         return __bpf_prog_get(ufd, &type, attach_drv);
2362 }
2363 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2364
2365 /* Initially all BPF programs could be loaded w/o specifying
2366  * expected_attach_type. Later for some of them specifying expected_attach_type
2367  * at load time became required so that program could be validated properly.
2368  * Programs of types that are allowed to be loaded both w/ and w/o (for
2369  * backward compatibility) expected_attach_type, should have the default attach
2370  * type assigned to expected_attach_type for the latter case, so that it can be
2371  * validated later at attach time.
2372  *
2373  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2374  * prog type requires it but has some attach types that have to be backward
2375  * compatible.
2376  */
2377 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2378 {
2379         switch (attr->prog_type) {
2380         case BPF_PROG_TYPE_CGROUP_SOCK:
2381                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2382                  * exist so checking for non-zero is the way to go here.
2383                  */
2384                 if (!attr->expected_attach_type)
2385                         attr->expected_attach_type =
2386                                 BPF_CGROUP_INET_SOCK_CREATE;
2387                 break;
2388         case BPF_PROG_TYPE_SK_REUSEPORT:
2389                 if (!attr->expected_attach_type)
2390                         attr->expected_attach_type =
2391                                 BPF_SK_REUSEPORT_SELECT;
2392                 break;
2393         }
2394 }
2395
2396 static int
2397 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2398                            enum bpf_attach_type expected_attach_type,
2399                            struct btf *attach_btf, u32 btf_id,
2400                            struct bpf_prog *dst_prog)
2401 {
2402         if (btf_id) {
2403                 if (btf_id > BTF_MAX_TYPE)
2404                         return -EINVAL;
2405
2406                 if (!attach_btf && !dst_prog)
2407                         return -EINVAL;
2408
2409                 switch (prog_type) {
2410                 case BPF_PROG_TYPE_TRACING:
2411                 case BPF_PROG_TYPE_LSM:
2412                 case BPF_PROG_TYPE_STRUCT_OPS:
2413                 case BPF_PROG_TYPE_EXT:
2414                         break;
2415                 default:
2416                         return -EINVAL;
2417                 }
2418         }
2419
2420         if (attach_btf && (!btf_id || dst_prog))
2421                 return -EINVAL;
2422
2423         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2424             prog_type != BPF_PROG_TYPE_EXT)
2425                 return -EINVAL;
2426
2427         switch (prog_type) {
2428         case BPF_PROG_TYPE_CGROUP_SOCK:
2429                 switch (expected_attach_type) {
2430                 case BPF_CGROUP_INET_SOCK_CREATE:
2431                 case BPF_CGROUP_INET_SOCK_RELEASE:
2432                 case BPF_CGROUP_INET4_POST_BIND:
2433                 case BPF_CGROUP_INET6_POST_BIND:
2434                         return 0;
2435                 default:
2436                         return -EINVAL;
2437                 }
2438         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2439                 switch (expected_attach_type) {
2440                 case BPF_CGROUP_INET4_BIND:
2441                 case BPF_CGROUP_INET6_BIND:
2442                 case BPF_CGROUP_INET4_CONNECT:
2443                 case BPF_CGROUP_INET6_CONNECT:
2444                 case BPF_CGROUP_INET4_GETPEERNAME:
2445                 case BPF_CGROUP_INET6_GETPEERNAME:
2446                 case BPF_CGROUP_INET4_GETSOCKNAME:
2447                 case BPF_CGROUP_INET6_GETSOCKNAME:
2448                 case BPF_CGROUP_UDP4_SENDMSG:
2449                 case BPF_CGROUP_UDP6_SENDMSG:
2450                 case BPF_CGROUP_UDP4_RECVMSG:
2451                 case BPF_CGROUP_UDP6_RECVMSG:
2452                         return 0;
2453                 default:
2454                         return -EINVAL;
2455                 }
2456         case BPF_PROG_TYPE_CGROUP_SKB:
2457                 switch (expected_attach_type) {
2458                 case BPF_CGROUP_INET_INGRESS:
2459                 case BPF_CGROUP_INET_EGRESS:
2460                         return 0;
2461                 default:
2462                         return -EINVAL;
2463                 }
2464         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2465                 switch (expected_attach_type) {
2466                 case BPF_CGROUP_SETSOCKOPT:
2467                 case BPF_CGROUP_GETSOCKOPT:
2468                         return 0;
2469                 default:
2470                         return -EINVAL;
2471                 }
2472         case BPF_PROG_TYPE_SK_LOOKUP:
2473                 if (expected_attach_type == BPF_SK_LOOKUP)
2474                         return 0;
2475                 return -EINVAL;
2476         case BPF_PROG_TYPE_SK_REUSEPORT:
2477                 switch (expected_attach_type) {
2478                 case BPF_SK_REUSEPORT_SELECT:
2479                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2480                         return 0;
2481                 default:
2482                         return -EINVAL;
2483                 }
2484         case BPF_PROG_TYPE_NETFILTER:
2485                 if (expected_attach_type == BPF_NETFILTER)
2486                         return 0;
2487                 return -EINVAL;
2488         case BPF_PROG_TYPE_SYSCALL:
2489         case BPF_PROG_TYPE_EXT:
2490                 if (expected_attach_type)
2491                         return -EINVAL;
2492                 fallthrough;
2493         default:
2494                 return 0;
2495         }
2496 }
2497
2498 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2499 {
2500         switch (prog_type) {
2501         case BPF_PROG_TYPE_SCHED_CLS:
2502         case BPF_PROG_TYPE_SCHED_ACT:
2503         case BPF_PROG_TYPE_XDP:
2504         case BPF_PROG_TYPE_LWT_IN:
2505         case BPF_PROG_TYPE_LWT_OUT:
2506         case BPF_PROG_TYPE_LWT_XMIT:
2507         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2508         case BPF_PROG_TYPE_SK_SKB:
2509         case BPF_PROG_TYPE_SK_MSG:
2510         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2511         case BPF_PROG_TYPE_CGROUP_DEVICE:
2512         case BPF_PROG_TYPE_CGROUP_SOCK:
2513         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2514         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2515         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2516         case BPF_PROG_TYPE_SOCK_OPS:
2517         case BPF_PROG_TYPE_EXT: /* extends any prog */
2518         case BPF_PROG_TYPE_NETFILTER:
2519                 return true;
2520         case BPF_PROG_TYPE_CGROUP_SKB:
2521                 /* always unpriv */
2522         case BPF_PROG_TYPE_SK_REUSEPORT:
2523                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2524         default:
2525                 return false;
2526         }
2527 }
2528
2529 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2530 {
2531         switch (prog_type) {
2532         case BPF_PROG_TYPE_KPROBE:
2533         case BPF_PROG_TYPE_TRACEPOINT:
2534         case BPF_PROG_TYPE_PERF_EVENT:
2535         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2536         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2537         case BPF_PROG_TYPE_TRACING:
2538         case BPF_PROG_TYPE_LSM:
2539         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2540         case BPF_PROG_TYPE_EXT: /* extends any prog */
2541                 return true;
2542         default:
2543                 return false;
2544         }
2545 }
2546
2547 /* last field in 'union bpf_attr' used by this command */
2548 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2549
2550 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2551 {
2552         enum bpf_prog_type type = attr->prog_type;
2553         struct bpf_prog *prog, *dst_prog = NULL;
2554         struct btf *attach_btf = NULL;
2555         int err;
2556         char license[128];
2557
2558         if (CHECK_ATTR(BPF_PROG_LOAD))
2559                 return -EINVAL;
2560
2561         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2562                                  BPF_F_ANY_ALIGNMENT |
2563                                  BPF_F_TEST_STATE_FREQ |
2564                                  BPF_F_SLEEPABLE |
2565                                  BPF_F_TEST_RND_HI32 |
2566                                  BPF_F_XDP_HAS_FRAGS |
2567                                  BPF_F_XDP_DEV_BOUND_ONLY))
2568                 return -EINVAL;
2569
2570         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2571             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2572             !bpf_capable())
2573                 return -EPERM;
2574
2575         /* Intent here is for unprivileged_bpf_disabled to block BPF program
2576          * creation for unprivileged users; other actions depend
2577          * on fd availability and access to bpffs, so are dependent on
2578          * object creation success. Even with unprivileged BPF disabled,
2579          * capability checks are still carried out for these
2580          * and other operations.
2581          */
2582         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
2583                 return -EPERM;
2584
2585         if (attr->insn_cnt == 0 ||
2586             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2587                 return -E2BIG;
2588         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2589             type != BPF_PROG_TYPE_CGROUP_SKB &&
2590             !bpf_capable())
2591                 return -EPERM;
2592
2593         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2594                 return -EPERM;
2595         if (is_perfmon_prog_type(type) && !perfmon_capable())
2596                 return -EPERM;
2597
2598         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2599          * or btf, we need to check which one it is
2600          */
2601         if (attr->attach_prog_fd) {
2602                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2603                 if (IS_ERR(dst_prog)) {
2604                         dst_prog = NULL;
2605                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2606                         if (IS_ERR(attach_btf))
2607                                 return -EINVAL;
2608                         if (!btf_is_kernel(attach_btf)) {
2609                                 /* attaching through specifying bpf_prog's BTF
2610                                  * objects directly might be supported eventually
2611                                  */
2612                                 btf_put(attach_btf);
2613                                 return -ENOTSUPP;
2614                         }
2615                 }
2616         } else if (attr->attach_btf_id) {
2617                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2618                 attach_btf = bpf_get_btf_vmlinux();
2619                 if (IS_ERR(attach_btf))
2620                         return PTR_ERR(attach_btf);
2621                 if (!attach_btf)
2622                         return -EINVAL;
2623                 btf_get(attach_btf);
2624         }
2625
2626         bpf_prog_load_fixup_attach_type(attr);
2627         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2628                                        attach_btf, attr->attach_btf_id,
2629                                        dst_prog)) {
2630                 if (dst_prog)
2631                         bpf_prog_put(dst_prog);
2632                 if (attach_btf)
2633                         btf_put(attach_btf);
2634                 return -EINVAL;
2635         }
2636
2637         /* plain bpf_prog allocation */
2638         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2639         if (!prog) {
2640                 if (dst_prog)
2641                         bpf_prog_put(dst_prog);
2642                 if (attach_btf)
2643                         btf_put(attach_btf);
2644                 return -ENOMEM;
2645         }
2646
2647         prog->expected_attach_type = attr->expected_attach_type;
2648         prog->aux->attach_btf = attach_btf;
2649         prog->aux->attach_btf_id = attr->attach_btf_id;
2650         prog->aux->dst_prog = dst_prog;
2651         prog->aux->dev_bound = !!attr->prog_ifindex;
2652         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2653         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2654
2655         err = security_bpf_prog_alloc(prog->aux);
2656         if (err)
2657                 goto free_prog;
2658
2659         prog->aux->user = get_current_user();
2660         prog->len = attr->insn_cnt;
2661
2662         err = -EFAULT;
2663         if (copy_from_bpfptr(prog->insns,
2664                              make_bpfptr(attr->insns, uattr.is_kernel),
2665                              bpf_prog_insn_size(prog)) != 0)
2666                 goto free_prog_sec;
2667         /* copy eBPF program license from user space */
2668         if (strncpy_from_bpfptr(license,
2669                                 make_bpfptr(attr->license, uattr.is_kernel),
2670                                 sizeof(license) - 1) < 0)
2671                 goto free_prog_sec;
2672         license[sizeof(license) - 1] = 0;
2673
2674         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2675         prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2676
2677         prog->orig_prog = NULL;
2678         prog->jited = 0;
2679
2680         atomic64_set(&prog->aux->refcnt, 1);
2681
2682         if (bpf_prog_is_dev_bound(prog->aux)) {
2683                 err = bpf_prog_dev_bound_init(prog, attr);
2684                 if (err)
2685                         goto free_prog_sec;
2686         }
2687
2688         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2689             bpf_prog_is_dev_bound(dst_prog->aux)) {
2690                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2691                 if (err)
2692                         goto free_prog_sec;
2693         }
2694
2695         /* find program type: socket_filter vs tracing_filter */
2696         err = find_prog_type(type, prog);
2697         if (err < 0)
2698                 goto free_prog_sec;
2699
2700         prog->aux->load_time = ktime_get_boottime_ns();
2701         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2702                                sizeof(attr->prog_name));
2703         if (err < 0)
2704                 goto free_prog_sec;
2705
2706         /* run eBPF verifier */
2707         err = bpf_check(&prog, attr, uattr, uattr_size);
2708         if (err < 0)
2709                 goto free_used_maps;
2710
2711         prog = bpf_prog_select_runtime(prog, &err);
2712         if (err < 0)
2713                 goto free_used_maps;
2714
2715         err = bpf_prog_alloc_id(prog);
2716         if (err)
2717                 goto free_used_maps;
2718
2719         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2720          * effectively publicly exposed. However, retrieving via
2721          * bpf_prog_get_fd_by_id() will take another reference,
2722          * therefore it cannot be gone underneath us.
2723          *
2724          * Only for the time /after/ successful bpf_prog_new_fd()
2725          * and before returning to userspace, we might just hold
2726          * one reference and any parallel close on that fd could
2727          * rip everything out. Hence, below notifications must
2728          * happen before bpf_prog_new_fd().
2729          *
2730          * Also, any failure handling from this point onwards must
2731          * be using bpf_prog_put() given the program is exposed.
2732          */
2733         bpf_prog_kallsyms_add(prog);
2734         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2735         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2736
2737         err = bpf_prog_new_fd(prog);
2738         if (err < 0)
2739                 bpf_prog_put(prog);
2740         return err;
2741
2742 free_used_maps:
2743         /* In case we have subprogs, we need to wait for a grace
2744          * period before we can tear down JIT memory since symbols
2745          * are already exposed under kallsyms.
2746          */
2747         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2748         return err;
2749 free_prog_sec:
2750         free_uid(prog->aux->user);
2751         security_bpf_prog_free(prog->aux);
2752 free_prog:
2753         if (prog->aux->attach_btf)
2754                 btf_put(prog->aux->attach_btf);
2755         bpf_prog_free(prog);
2756         return err;
2757 }
2758
2759 #define BPF_OBJ_LAST_FIELD path_fd
2760
2761 static int bpf_obj_pin(const union bpf_attr *attr)
2762 {
2763         int path_fd;
2764
2765         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2766                 return -EINVAL;
2767
2768         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2769         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2770                 return -EINVAL;
2771
2772         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2773         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2774                                 u64_to_user_ptr(attr->pathname));
2775 }
2776
2777 static int bpf_obj_get(const union bpf_attr *attr)
2778 {
2779         int path_fd;
2780
2781         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2782             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2783                 return -EINVAL;
2784
2785         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2786         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2787                 return -EINVAL;
2788
2789         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2790         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2791                                 attr->file_flags);
2792 }
2793
2794 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2795                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2796 {
2797         atomic64_set(&link->refcnt, 1);
2798         link->type = type;
2799         link->id = 0;
2800         link->ops = ops;
2801         link->prog = prog;
2802 }
2803
2804 static void bpf_link_free_id(int id)
2805 {
2806         if (!id)
2807                 return;
2808
2809         spin_lock_bh(&link_idr_lock);
2810         idr_remove(&link_idr, id);
2811         spin_unlock_bh(&link_idr_lock);
2812 }
2813
2814 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2815  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2816  * anon_inode's release() call. This helper marksbpf_link as
2817  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2818  * is not decremented, it's the responsibility of a calling code that failed
2819  * to complete bpf_link initialization.
2820  */
2821 void bpf_link_cleanup(struct bpf_link_primer *primer)
2822 {
2823         primer->link->prog = NULL;
2824         bpf_link_free_id(primer->id);
2825         fput(primer->file);
2826         put_unused_fd(primer->fd);
2827 }
2828
2829 void bpf_link_inc(struct bpf_link *link)
2830 {
2831         atomic64_inc(&link->refcnt);
2832 }
2833
2834 /* bpf_link_free is guaranteed to be called from process context */
2835 static void bpf_link_free(struct bpf_link *link)
2836 {
2837         bpf_link_free_id(link->id);
2838         if (link->prog) {
2839                 /* detach BPF program, clean up used resources */
2840                 link->ops->release(link);
2841                 bpf_prog_put(link->prog);
2842         }
2843         /* free bpf_link and its containing memory */
2844         link->ops->dealloc(link);
2845 }
2846
2847 static void bpf_link_put_deferred(struct work_struct *work)
2848 {
2849         struct bpf_link *link = container_of(work, struct bpf_link, work);
2850
2851         bpf_link_free(link);
2852 }
2853
2854 /* bpf_link_put might be called from atomic context. It needs to be called
2855  * from sleepable context in order to acquire sleeping locks during the process.
2856  */
2857 void bpf_link_put(struct bpf_link *link)
2858 {
2859         if (!atomic64_dec_and_test(&link->refcnt))
2860                 return;
2861
2862         INIT_WORK(&link->work, bpf_link_put_deferred);
2863         schedule_work(&link->work);
2864 }
2865 EXPORT_SYMBOL(bpf_link_put);
2866
2867 static void bpf_link_put_direct(struct bpf_link *link)
2868 {
2869         if (!atomic64_dec_and_test(&link->refcnt))
2870                 return;
2871         bpf_link_free(link);
2872 }
2873
2874 static int bpf_link_release(struct inode *inode, struct file *filp)
2875 {
2876         struct bpf_link *link = filp->private_data;
2877
2878         bpf_link_put_direct(link);
2879         return 0;
2880 }
2881
2882 #ifdef CONFIG_PROC_FS
2883 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2884 #define BPF_MAP_TYPE(_id, _ops)
2885 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2886 static const char *bpf_link_type_strs[] = {
2887         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2888 #include <linux/bpf_types.h>
2889 };
2890 #undef BPF_PROG_TYPE
2891 #undef BPF_MAP_TYPE
2892 #undef BPF_LINK_TYPE
2893
2894 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2895 {
2896         const struct bpf_link *link = filp->private_data;
2897         const struct bpf_prog *prog = link->prog;
2898         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2899
2900         seq_printf(m,
2901                    "link_type:\t%s\n"
2902                    "link_id:\t%u\n",
2903                    bpf_link_type_strs[link->type],
2904                    link->id);
2905         if (prog) {
2906                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2907                 seq_printf(m,
2908                            "prog_tag:\t%s\n"
2909                            "prog_id:\t%u\n",
2910                            prog_tag,
2911                            prog->aux->id);
2912         }
2913         if (link->ops->show_fdinfo)
2914                 link->ops->show_fdinfo(link, m);
2915 }
2916 #endif
2917
2918 static const struct file_operations bpf_link_fops = {
2919 #ifdef CONFIG_PROC_FS
2920         .show_fdinfo    = bpf_link_show_fdinfo,
2921 #endif
2922         .release        = bpf_link_release,
2923         .read           = bpf_dummy_read,
2924         .write          = bpf_dummy_write,
2925 };
2926
2927 static int bpf_link_alloc_id(struct bpf_link *link)
2928 {
2929         int id;
2930
2931         idr_preload(GFP_KERNEL);
2932         spin_lock_bh(&link_idr_lock);
2933         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2934         spin_unlock_bh(&link_idr_lock);
2935         idr_preload_end();
2936
2937         return id;
2938 }
2939
2940 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2941  * reserving unused FD and allocating ID from link_idr. This is to be paired
2942  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2943  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2944  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2945  * transient state is passed around in struct bpf_link_primer.
2946  * This is preferred way to create and initialize bpf_link, especially when
2947  * there are complicated and expensive operations in between creating bpf_link
2948  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2949  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2950  * expensive (and potentially failing) roll back operations in a rare case
2951  * that file, FD, or ID can't be allocated.
2952  */
2953 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2954 {
2955         struct file *file;
2956         int fd, id;
2957
2958         fd = get_unused_fd_flags(O_CLOEXEC);
2959         if (fd < 0)
2960                 return fd;
2961
2962
2963         id = bpf_link_alloc_id(link);
2964         if (id < 0) {
2965                 put_unused_fd(fd);
2966                 return id;
2967         }
2968
2969         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2970         if (IS_ERR(file)) {
2971                 bpf_link_free_id(id);
2972                 put_unused_fd(fd);
2973                 return PTR_ERR(file);
2974         }
2975
2976         primer->link = link;
2977         primer->file = file;
2978         primer->fd = fd;
2979         primer->id = id;
2980         return 0;
2981 }
2982
2983 int bpf_link_settle(struct bpf_link_primer *primer)
2984 {
2985         /* make bpf_link fetchable by ID */
2986         spin_lock_bh(&link_idr_lock);
2987         primer->link->id = primer->id;
2988         spin_unlock_bh(&link_idr_lock);
2989         /* make bpf_link fetchable by FD */
2990         fd_install(primer->fd, primer->file);
2991         /* pass through installed FD */
2992         return primer->fd;
2993 }
2994
2995 int bpf_link_new_fd(struct bpf_link *link)
2996 {
2997         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2998 }
2999
3000 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3001 {
3002         struct fd f = fdget(ufd);
3003         struct bpf_link *link;
3004
3005         if (!f.file)
3006                 return ERR_PTR(-EBADF);
3007         if (f.file->f_op != &bpf_link_fops) {
3008                 fdput(f);
3009                 return ERR_PTR(-EINVAL);
3010         }
3011
3012         link = f.file->private_data;
3013         bpf_link_inc(link);
3014         fdput(f);
3015
3016         return link;
3017 }
3018 EXPORT_SYMBOL(bpf_link_get_from_fd);
3019
3020 static void bpf_tracing_link_release(struct bpf_link *link)
3021 {
3022         struct bpf_tracing_link *tr_link =
3023                 container_of(link, struct bpf_tracing_link, link.link);
3024
3025         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3026                                                 tr_link->trampoline));
3027
3028         bpf_trampoline_put(tr_link->trampoline);
3029
3030         /* tgt_prog is NULL if target is a kernel function */
3031         if (tr_link->tgt_prog)
3032                 bpf_prog_put(tr_link->tgt_prog);
3033 }
3034
3035 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3036 {
3037         struct bpf_tracing_link *tr_link =
3038                 container_of(link, struct bpf_tracing_link, link.link);
3039
3040         kfree(tr_link);
3041 }
3042
3043 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3044                                          struct seq_file *seq)
3045 {
3046         struct bpf_tracing_link *tr_link =
3047                 container_of(link, struct bpf_tracing_link, link.link);
3048         u32 target_btf_id, target_obj_id;
3049
3050         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3051                                   &target_obj_id, &target_btf_id);
3052         seq_printf(seq,
3053                    "attach_type:\t%d\n"
3054                    "target_obj_id:\t%u\n"
3055                    "target_btf_id:\t%u\n",
3056                    tr_link->attach_type,
3057                    target_obj_id,
3058                    target_btf_id);
3059 }
3060
3061 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3062                                            struct bpf_link_info *info)
3063 {
3064         struct bpf_tracing_link *tr_link =
3065                 container_of(link, struct bpf_tracing_link, link.link);
3066
3067         info->tracing.attach_type = tr_link->attach_type;
3068         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3069                                   &info->tracing.target_obj_id,
3070                                   &info->tracing.target_btf_id);
3071
3072         return 0;
3073 }
3074
3075 static const struct bpf_link_ops bpf_tracing_link_lops = {
3076         .release = bpf_tracing_link_release,
3077         .dealloc = bpf_tracing_link_dealloc,
3078         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3079         .fill_link_info = bpf_tracing_link_fill_link_info,
3080 };
3081
3082 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3083                                    int tgt_prog_fd,
3084                                    u32 btf_id,
3085                                    u64 bpf_cookie)
3086 {
3087         struct bpf_link_primer link_primer;
3088         struct bpf_prog *tgt_prog = NULL;
3089         struct bpf_trampoline *tr = NULL;
3090         struct bpf_tracing_link *link;
3091         u64 key = 0;
3092         int err;
3093
3094         switch (prog->type) {
3095         case BPF_PROG_TYPE_TRACING:
3096                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3097                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3098                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3099                         err = -EINVAL;
3100                         goto out_put_prog;
3101                 }
3102                 break;
3103         case BPF_PROG_TYPE_EXT:
3104                 if (prog->expected_attach_type != 0) {
3105                         err = -EINVAL;
3106                         goto out_put_prog;
3107                 }
3108                 break;
3109         case BPF_PROG_TYPE_LSM:
3110                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3111                         err = -EINVAL;
3112                         goto out_put_prog;
3113                 }
3114                 break;
3115         default:
3116                 err = -EINVAL;
3117                 goto out_put_prog;
3118         }
3119
3120         if (!!tgt_prog_fd != !!btf_id) {
3121                 err = -EINVAL;
3122                 goto out_put_prog;
3123         }
3124
3125         if (tgt_prog_fd) {
3126                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3127                 if (prog->type != BPF_PROG_TYPE_EXT) {
3128                         err = -EINVAL;
3129                         goto out_put_prog;
3130                 }
3131
3132                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3133                 if (IS_ERR(tgt_prog)) {
3134                         err = PTR_ERR(tgt_prog);
3135                         tgt_prog = NULL;
3136                         goto out_put_prog;
3137                 }
3138
3139                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3140         }
3141
3142         link = kzalloc(sizeof(*link), GFP_USER);
3143         if (!link) {
3144                 err = -ENOMEM;
3145                 goto out_put_prog;
3146         }
3147         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3148                       &bpf_tracing_link_lops, prog);
3149         link->attach_type = prog->expected_attach_type;
3150         link->link.cookie = bpf_cookie;
3151
3152         mutex_lock(&prog->aux->dst_mutex);
3153
3154         /* There are a few possible cases here:
3155          *
3156          * - if prog->aux->dst_trampoline is set, the program was just loaded
3157          *   and not yet attached to anything, so we can use the values stored
3158          *   in prog->aux
3159          *
3160          * - if prog->aux->dst_trampoline is NULL, the program has already been
3161          *   attached to a target and its initial target was cleared (below)
3162          *
3163          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3164          *   target_btf_id using the link_create API.
3165          *
3166          * - if tgt_prog == NULL when this function was called using the old
3167          *   raw_tracepoint_open API, and we need a target from prog->aux
3168          *
3169          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3170          *   was detached and is going for re-attachment.
3171          */
3172         if (!prog->aux->dst_trampoline && !tgt_prog) {
3173                 /*
3174                  * Allow re-attach for TRACING and LSM programs. If it's
3175                  * currently linked, bpf_trampoline_link_prog will fail.
3176                  * EXT programs need to specify tgt_prog_fd, so they
3177                  * re-attach in separate code path.
3178                  */
3179                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3180                     prog->type != BPF_PROG_TYPE_LSM) {
3181                         err = -EINVAL;
3182                         goto out_unlock;
3183                 }
3184                 btf_id = prog->aux->attach_btf_id;
3185                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3186         }
3187
3188         if (!prog->aux->dst_trampoline ||
3189             (key && key != prog->aux->dst_trampoline->key)) {
3190                 /* If there is no saved target, or the specified target is
3191                  * different from the destination specified at load time, we
3192                  * need a new trampoline and a check for compatibility
3193                  */
3194                 struct bpf_attach_target_info tgt_info = {};
3195
3196                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3197                                               &tgt_info);
3198                 if (err)
3199                         goto out_unlock;
3200
3201                 if (tgt_info.tgt_mod) {
3202                         module_put(prog->aux->mod);
3203                         prog->aux->mod = tgt_info.tgt_mod;
3204                 }
3205
3206                 tr = bpf_trampoline_get(key, &tgt_info);
3207                 if (!tr) {
3208                         err = -ENOMEM;
3209                         goto out_unlock;
3210                 }
3211         } else {
3212                 /* The caller didn't specify a target, or the target was the
3213                  * same as the destination supplied during program load. This
3214                  * means we can reuse the trampoline and reference from program
3215                  * load time, and there is no need to allocate a new one. This
3216                  * can only happen once for any program, as the saved values in
3217                  * prog->aux are cleared below.
3218                  */
3219                 tr = prog->aux->dst_trampoline;
3220                 tgt_prog = prog->aux->dst_prog;
3221         }
3222
3223         err = bpf_link_prime(&link->link.link, &link_primer);
3224         if (err)
3225                 goto out_unlock;
3226
3227         err = bpf_trampoline_link_prog(&link->link, tr);
3228         if (err) {
3229                 bpf_link_cleanup(&link_primer);
3230                 link = NULL;
3231                 goto out_unlock;
3232         }
3233
3234         link->tgt_prog = tgt_prog;
3235         link->trampoline = tr;
3236
3237         /* Always clear the trampoline and target prog from prog->aux to make
3238          * sure the original attach destination is not kept alive after a
3239          * program is (re-)attached to another target.
3240          */
3241         if (prog->aux->dst_prog &&
3242             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3243                 /* got extra prog ref from syscall, or attaching to different prog */
3244                 bpf_prog_put(prog->aux->dst_prog);
3245         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3246                 /* we allocated a new trampoline, so free the old one */
3247                 bpf_trampoline_put(prog->aux->dst_trampoline);
3248
3249         prog->aux->dst_prog = NULL;
3250         prog->aux->dst_trampoline = NULL;
3251         mutex_unlock(&prog->aux->dst_mutex);
3252
3253         return bpf_link_settle(&link_primer);
3254 out_unlock:
3255         if (tr && tr != prog->aux->dst_trampoline)
3256                 bpf_trampoline_put(tr);
3257         mutex_unlock(&prog->aux->dst_mutex);
3258         kfree(link);
3259 out_put_prog:
3260         if (tgt_prog_fd && tgt_prog)
3261                 bpf_prog_put(tgt_prog);
3262         return err;
3263 }
3264
3265 struct bpf_raw_tp_link {
3266         struct bpf_link link;
3267         struct bpf_raw_event_map *btp;
3268 };
3269
3270 static void bpf_raw_tp_link_release(struct bpf_link *link)
3271 {
3272         struct bpf_raw_tp_link *raw_tp =
3273                 container_of(link, struct bpf_raw_tp_link, link);
3274
3275         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3276         bpf_put_raw_tracepoint(raw_tp->btp);
3277 }
3278
3279 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3280 {
3281         struct bpf_raw_tp_link *raw_tp =
3282                 container_of(link, struct bpf_raw_tp_link, link);
3283
3284         kfree(raw_tp);
3285 }
3286
3287 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3288                                         struct seq_file *seq)
3289 {
3290         struct bpf_raw_tp_link *raw_tp_link =
3291                 container_of(link, struct bpf_raw_tp_link, link);
3292
3293         seq_printf(seq,
3294                    "tp_name:\t%s\n",
3295                    raw_tp_link->btp->tp->name);
3296 }
3297
3298 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3299                                           struct bpf_link_info *info)
3300 {
3301         struct bpf_raw_tp_link *raw_tp_link =
3302                 container_of(link, struct bpf_raw_tp_link, link);
3303         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3304         const char *tp_name = raw_tp_link->btp->tp->name;
3305         u32 ulen = info->raw_tracepoint.tp_name_len;
3306         size_t tp_len = strlen(tp_name);
3307
3308         if (!ulen ^ !ubuf)
3309                 return -EINVAL;
3310
3311         info->raw_tracepoint.tp_name_len = tp_len + 1;
3312
3313         if (!ubuf)
3314                 return 0;
3315
3316         if (ulen >= tp_len + 1) {
3317                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3318                         return -EFAULT;
3319         } else {
3320                 char zero = '\0';
3321
3322                 if (copy_to_user(ubuf, tp_name, ulen - 1))
3323                         return -EFAULT;
3324                 if (put_user(zero, ubuf + ulen - 1))
3325                         return -EFAULT;
3326                 return -ENOSPC;
3327         }
3328
3329         return 0;
3330 }
3331
3332 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3333         .release = bpf_raw_tp_link_release,
3334         .dealloc = bpf_raw_tp_link_dealloc,
3335         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3336         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3337 };
3338
3339 #ifdef CONFIG_PERF_EVENTS
3340 struct bpf_perf_link {
3341         struct bpf_link link;
3342         struct file *perf_file;
3343 };
3344
3345 static void bpf_perf_link_release(struct bpf_link *link)
3346 {
3347         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3348         struct perf_event *event = perf_link->perf_file->private_data;
3349
3350         perf_event_free_bpf_prog(event);
3351         fput(perf_link->perf_file);
3352 }
3353
3354 static void bpf_perf_link_dealloc(struct bpf_link *link)
3355 {
3356         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3357
3358         kfree(perf_link);
3359 }
3360
3361 static const struct bpf_link_ops bpf_perf_link_lops = {
3362         .release = bpf_perf_link_release,
3363         .dealloc = bpf_perf_link_dealloc,
3364 };
3365
3366 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3367 {
3368         struct bpf_link_primer link_primer;
3369         struct bpf_perf_link *link;
3370         struct perf_event *event;
3371         struct file *perf_file;
3372         int err;
3373
3374         if (attr->link_create.flags)
3375                 return -EINVAL;
3376
3377         perf_file = perf_event_get(attr->link_create.target_fd);
3378         if (IS_ERR(perf_file))
3379                 return PTR_ERR(perf_file);
3380
3381         link = kzalloc(sizeof(*link), GFP_USER);
3382         if (!link) {
3383                 err = -ENOMEM;
3384                 goto out_put_file;
3385         }
3386         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3387         link->perf_file = perf_file;
3388
3389         err = bpf_link_prime(&link->link, &link_primer);
3390         if (err) {
3391                 kfree(link);
3392                 goto out_put_file;
3393         }
3394
3395         event = perf_file->private_data;
3396         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3397         if (err) {
3398                 bpf_link_cleanup(&link_primer);
3399                 goto out_put_file;
3400         }
3401         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3402         bpf_prog_inc(prog);
3403
3404         return bpf_link_settle(&link_primer);
3405
3406 out_put_file:
3407         fput(perf_file);
3408         return err;
3409 }
3410 #else
3411 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3412 {
3413         return -EOPNOTSUPP;
3414 }
3415 #endif /* CONFIG_PERF_EVENTS */
3416
3417 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3418                                   const char __user *user_tp_name)
3419 {
3420         struct bpf_link_primer link_primer;
3421         struct bpf_raw_tp_link *link;
3422         struct bpf_raw_event_map *btp;
3423         const char *tp_name;
3424         char buf[128];
3425         int err;
3426
3427         switch (prog->type) {
3428         case BPF_PROG_TYPE_TRACING:
3429         case BPF_PROG_TYPE_EXT:
3430         case BPF_PROG_TYPE_LSM:
3431                 if (user_tp_name)
3432                         /* The attach point for this category of programs
3433                          * should be specified via btf_id during program load.
3434                          */
3435                         return -EINVAL;
3436                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3437                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3438                         tp_name = prog->aux->attach_func_name;
3439                         break;
3440                 }
3441                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3442         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3443         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3444                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3445                         return -EFAULT;
3446                 buf[sizeof(buf) - 1] = 0;
3447                 tp_name = buf;
3448                 break;
3449         default:
3450                 return -EINVAL;
3451         }
3452
3453         btp = bpf_get_raw_tracepoint(tp_name);
3454         if (!btp)
3455                 return -ENOENT;
3456
3457         link = kzalloc(sizeof(*link), GFP_USER);
3458         if (!link) {
3459                 err = -ENOMEM;
3460                 goto out_put_btp;
3461         }
3462         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3463                       &bpf_raw_tp_link_lops, prog);
3464         link->btp = btp;
3465
3466         err = bpf_link_prime(&link->link, &link_primer);
3467         if (err) {
3468                 kfree(link);
3469                 goto out_put_btp;
3470         }
3471
3472         err = bpf_probe_register(link->btp, prog);
3473         if (err) {
3474                 bpf_link_cleanup(&link_primer);
3475                 goto out_put_btp;
3476         }
3477
3478         return bpf_link_settle(&link_primer);
3479
3480 out_put_btp:
3481         bpf_put_raw_tracepoint(btp);
3482         return err;
3483 }
3484
3485 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3486
3487 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3488 {
3489         struct bpf_prog *prog;
3490         int fd;
3491
3492         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3493                 return -EINVAL;
3494
3495         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3496         if (IS_ERR(prog))
3497                 return PTR_ERR(prog);
3498
3499         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3500         if (fd < 0)
3501                 bpf_prog_put(prog);
3502         return fd;
3503 }
3504
3505 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3506                                              enum bpf_attach_type attach_type)
3507 {
3508         switch (prog->type) {
3509         case BPF_PROG_TYPE_CGROUP_SOCK:
3510         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3511         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3512         case BPF_PROG_TYPE_SK_LOOKUP:
3513                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3514         case BPF_PROG_TYPE_CGROUP_SKB:
3515                 if (!capable(CAP_NET_ADMIN))
3516                         /* cg-skb progs can be loaded by unpriv user.
3517                          * check permissions at attach time.
3518                          */
3519                         return -EPERM;
3520                 return prog->enforce_expected_attach_type &&
3521                         prog->expected_attach_type != attach_type ?
3522                         -EINVAL : 0;
3523         case BPF_PROG_TYPE_KPROBE:
3524                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3525                     attach_type != BPF_TRACE_KPROBE_MULTI)
3526                         return -EINVAL;
3527                 return 0;
3528         default:
3529                 return 0;
3530         }
3531 }
3532
3533 static enum bpf_prog_type
3534 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3535 {
3536         switch (attach_type) {
3537         case BPF_CGROUP_INET_INGRESS:
3538         case BPF_CGROUP_INET_EGRESS:
3539                 return BPF_PROG_TYPE_CGROUP_SKB;
3540         case BPF_CGROUP_INET_SOCK_CREATE:
3541         case BPF_CGROUP_INET_SOCK_RELEASE:
3542         case BPF_CGROUP_INET4_POST_BIND:
3543         case BPF_CGROUP_INET6_POST_BIND:
3544                 return BPF_PROG_TYPE_CGROUP_SOCK;
3545         case BPF_CGROUP_INET4_BIND:
3546         case BPF_CGROUP_INET6_BIND:
3547         case BPF_CGROUP_INET4_CONNECT:
3548         case BPF_CGROUP_INET6_CONNECT:
3549         case BPF_CGROUP_INET4_GETPEERNAME:
3550         case BPF_CGROUP_INET6_GETPEERNAME:
3551         case BPF_CGROUP_INET4_GETSOCKNAME:
3552         case BPF_CGROUP_INET6_GETSOCKNAME:
3553         case BPF_CGROUP_UDP4_SENDMSG:
3554         case BPF_CGROUP_UDP6_SENDMSG:
3555         case BPF_CGROUP_UDP4_RECVMSG:
3556         case BPF_CGROUP_UDP6_RECVMSG:
3557                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3558         case BPF_CGROUP_SOCK_OPS:
3559                 return BPF_PROG_TYPE_SOCK_OPS;
3560         case BPF_CGROUP_DEVICE:
3561                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3562         case BPF_SK_MSG_VERDICT:
3563                 return BPF_PROG_TYPE_SK_MSG;
3564         case BPF_SK_SKB_STREAM_PARSER:
3565         case BPF_SK_SKB_STREAM_VERDICT:
3566         case BPF_SK_SKB_VERDICT:
3567                 return BPF_PROG_TYPE_SK_SKB;
3568         case BPF_LIRC_MODE2:
3569                 return BPF_PROG_TYPE_LIRC_MODE2;
3570         case BPF_FLOW_DISSECTOR:
3571                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3572         case BPF_CGROUP_SYSCTL:
3573                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3574         case BPF_CGROUP_GETSOCKOPT:
3575         case BPF_CGROUP_SETSOCKOPT:
3576                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3577         case BPF_TRACE_ITER:
3578         case BPF_TRACE_RAW_TP:
3579         case BPF_TRACE_FENTRY:
3580         case BPF_TRACE_FEXIT:
3581         case BPF_MODIFY_RETURN:
3582                 return BPF_PROG_TYPE_TRACING;
3583         case BPF_LSM_MAC:
3584                 return BPF_PROG_TYPE_LSM;
3585         case BPF_SK_LOOKUP:
3586                 return BPF_PROG_TYPE_SK_LOOKUP;
3587         case BPF_XDP:
3588                 return BPF_PROG_TYPE_XDP;
3589         case BPF_LSM_CGROUP:
3590                 return BPF_PROG_TYPE_LSM;
3591         default:
3592                 return BPF_PROG_TYPE_UNSPEC;
3593         }
3594 }
3595
3596 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3597
3598 #define BPF_F_ATTACH_MASK \
3599         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3600
3601 static int bpf_prog_attach(const union bpf_attr *attr)
3602 {
3603         enum bpf_prog_type ptype;
3604         struct bpf_prog *prog;
3605         int ret;
3606
3607         if (CHECK_ATTR(BPF_PROG_ATTACH))
3608                 return -EINVAL;
3609
3610         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3611                 return -EINVAL;
3612
3613         ptype = attach_type_to_prog_type(attr->attach_type);
3614         if (ptype == BPF_PROG_TYPE_UNSPEC)
3615                 return -EINVAL;
3616
3617         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3618         if (IS_ERR(prog))
3619                 return PTR_ERR(prog);
3620
3621         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3622                 bpf_prog_put(prog);
3623                 return -EINVAL;
3624         }
3625
3626         switch (ptype) {
3627         case BPF_PROG_TYPE_SK_SKB:
3628         case BPF_PROG_TYPE_SK_MSG:
3629                 ret = sock_map_get_from_fd(attr, prog);
3630                 break;
3631         case BPF_PROG_TYPE_LIRC_MODE2:
3632                 ret = lirc_prog_attach(attr, prog);
3633                 break;
3634         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3635                 ret = netns_bpf_prog_attach(attr, prog);
3636                 break;
3637         case BPF_PROG_TYPE_CGROUP_DEVICE:
3638         case BPF_PROG_TYPE_CGROUP_SKB:
3639         case BPF_PROG_TYPE_CGROUP_SOCK:
3640         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3641         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3642         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3643         case BPF_PROG_TYPE_SOCK_OPS:
3644         case BPF_PROG_TYPE_LSM:
3645                 if (ptype == BPF_PROG_TYPE_LSM &&
3646                     prog->expected_attach_type != BPF_LSM_CGROUP)
3647                         ret = -EINVAL;
3648                 else
3649                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3650                 break;
3651         default:
3652                 ret = -EINVAL;
3653         }
3654
3655         if (ret)
3656                 bpf_prog_put(prog);
3657         return ret;
3658 }
3659
3660 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3661
3662 static int bpf_prog_detach(const union bpf_attr *attr)
3663 {
3664         enum bpf_prog_type ptype;
3665
3666         if (CHECK_ATTR(BPF_PROG_DETACH))
3667                 return -EINVAL;
3668
3669         ptype = attach_type_to_prog_type(attr->attach_type);
3670
3671         switch (ptype) {
3672         case BPF_PROG_TYPE_SK_MSG:
3673         case BPF_PROG_TYPE_SK_SKB:
3674                 return sock_map_prog_detach(attr, ptype);
3675         case BPF_PROG_TYPE_LIRC_MODE2:
3676                 return lirc_prog_detach(attr);
3677         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3678                 return netns_bpf_prog_detach(attr, ptype);
3679         case BPF_PROG_TYPE_CGROUP_DEVICE:
3680         case BPF_PROG_TYPE_CGROUP_SKB:
3681         case BPF_PROG_TYPE_CGROUP_SOCK:
3682         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3683         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3684         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3685         case BPF_PROG_TYPE_SOCK_OPS:
3686         case BPF_PROG_TYPE_LSM:
3687                 return cgroup_bpf_prog_detach(attr, ptype);
3688         default:
3689                 return -EINVAL;
3690         }
3691 }
3692
3693 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3694
3695 static int bpf_prog_query(const union bpf_attr *attr,
3696                           union bpf_attr __user *uattr)
3697 {
3698         if (!capable(CAP_NET_ADMIN))
3699                 return -EPERM;
3700         if (CHECK_ATTR(BPF_PROG_QUERY))
3701                 return -EINVAL;
3702         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3703                 return -EINVAL;
3704
3705         switch (attr->query.attach_type) {
3706         case BPF_CGROUP_INET_INGRESS:
3707         case BPF_CGROUP_INET_EGRESS:
3708         case BPF_CGROUP_INET_SOCK_CREATE:
3709         case BPF_CGROUP_INET_SOCK_RELEASE:
3710         case BPF_CGROUP_INET4_BIND:
3711         case BPF_CGROUP_INET6_BIND:
3712         case BPF_CGROUP_INET4_POST_BIND:
3713         case BPF_CGROUP_INET6_POST_BIND:
3714         case BPF_CGROUP_INET4_CONNECT:
3715         case BPF_CGROUP_INET6_CONNECT:
3716         case BPF_CGROUP_INET4_GETPEERNAME:
3717         case BPF_CGROUP_INET6_GETPEERNAME:
3718         case BPF_CGROUP_INET4_GETSOCKNAME:
3719         case BPF_CGROUP_INET6_GETSOCKNAME:
3720         case BPF_CGROUP_UDP4_SENDMSG:
3721         case BPF_CGROUP_UDP6_SENDMSG:
3722         case BPF_CGROUP_UDP4_RECVMSG:
3723         case BPF_CGROUP_UDP6_RECVMSG:
3724         case BPF_CGROUP_SOCK_OPS:
3725         case BPF_CGROUP_DEVICE:
3726         case BPF_CGROUP_SYSCTL:
3727         case BPF_CGROUP_GETSOCKOPT:
3728         case BPF_CGROUP_SETSOCKOPT:
3729         case BPF_LSM_CGROUP:
3730                 return cgroup_bpf_prog_query(attr, uattr);
3731         case BPF_LIRC_MODE2:
3732                 return lirc_prog_query(attr, uattr);
3733         case BPF_FLOW_DISSECTOR:
3734         case BPF_SK_LOOKUP:
3735                 return netns_bpf_prog_query(attr, uattr);
3736         case BPF_SK_SKB_STREAM_PARSER:
3737         case BPF_SK_SKB_STREAM_VERDICT:
3738         case BPF_SK_MSG_VERDICT:
3739         case BPF_SK_SKB_VERDICT:
3740                 return sock_map_bpf_prog_query(attr, uattr);
3741         default:
3742                 return -EINVAL;
3743         }
3744 }
3745
3746 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3747
3748 static int bpf_prog_test_run(const union bpf_attr *attr,
3749                              union bpf_attr __user *uattr)
3750 {
3751         struct bpf_prog *prog;
3752         int ret = -ENOTSUPP;
3753
3754         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3755                 return -EINVAL;
3756
3757         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3758             (!attr->test.ctx_size_in && attr->test.ctx_in))
3759                 return -EINVAL;
3760
3761         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3762             (!attr->test.ctx_size_out && attr->test.ctx_out))
3763                 return -EINVAL;
3764
3765         prog = bpf_prog_get(attr->test.prog_fd);
3766         if (IS_ERR(prog))
3767                 return PTR_ERR(prog);
3768
3769         if (prog->aux->ops->test_run)
3770                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3771
3772         bpf_prog_put(prog);
3773         return ret;
3774 }
3775
3776 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3777
3778 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3779                                union bpf_attr __user *uattr,
3780                                struct idr *idr,
3781                                spinlock_t *lock)
3782 {
3783         u32 next_id = attr->start_id;
3784         int err = 0;
3785
3786         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3787                 return -EINVAL;
3788
3789         if (!capable(CAP_SYS_ADMIN))
3790                 return -EPERM;
3791
3792         next_id++;
3793         spin_lock_bh(lock);
3794         if (!idr_get_next(idr, &next_id))
3795                 err = -ENOENT;
3796         spin_unlock_bh(lock);
3797
3798         if (!err)
3799                 err = put_user(next_id, &uattr->next_id);
3800
3801         return err;
3802 }
3803
3804 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3805 {
3806         struct bpf_map *map;
3807
3808         spin_lock_bh(&map_idr_lock);
3809 again:
3810         map = idr_get_next(&map_idr, id);
3811         if (map) {
3812                 map = __bpf_map_inc_not_zero(map, false);
3813                 if (IS_ERR(map)) {
3814                         (*id)++;
3815                         goto again;
3816                 }
3817         }
3818         spin_unlock_bh(&map_idr_lock);
3819
3820         return map;
3821 }
3822
3823 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3824 {
3825         struct bpf_prog *prog;
3826
3827         spin_lock_bh(&prog_idr_lock);
3828 again:
3829         prog = idr_get_next(&prog_idr, id);
3830         if (prog) {
3831                 prog = bpf_prog_inc_not_zero(prog);
3832                 if (IS_ERR(prog)) {
3833                         (*id)++;
3834                         goto again;
3835                 }
3836         }
3837         spin_unlock_bh(&prog_idr_lock);
3838
3839         return prog;
3840 }
3841
3842 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3843
3844 struct bpf_prog *bpf_prog_by_id(u32 id)
3845 {
3846         struct bpf_prog *prog;
3847
3848         if (!id)
3849                 return ERR_PTR(-ENOENT);
3850
3851         spin_lock_bh(&prog_idr_lock);
3852         prog = idr_find(&prog_idr, id);
3853         if (prog)
3854                 prog = bpf_prog_inc_not_zero(prog);
3855         else
3856                 prog = ERR_PTR(-ENOENT);
3857         spin_unlock_bh(&prog_idr_lock);
3858         return prog;
3859 }
3860
3861 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3862 {
3863         struct bpf_prog *prog;
3864         u32 id = attr->prog_id;
3865         int fd;
3866
3867         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3868                 return -EINVAL;
3869
3870         if (!capable(CAP_SYS_ADMIN))
3871                 return -EPERM;
3872
3873         prog = bpf_prog_by_id(id);
3874         if (IS_ERR(prog))
3875                 return PTR_ERR(prog);
3876
3877         fd = bpf_prog_new_fd(prog);
3878         if (fd < 0)
3879                 bpf_prog_put(prog);
3880
3881         return fd;
3882 }
3883
3884 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3885
3886 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3887 {
3888         struct bpf_map *map;
3889         u32 id = attr->map_id;
3890         int f_flags;
3891         int fd;
3892
3893         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3894             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3895                 return -EINVAL;
3896
3897         if (!capable(CAP_SYS_ADMIN))
3898                 return -EPERM;
3899
3900         f_flags = bpf_get_file_flag(attr->open_flags);
3901         if (f_flags < 0)
3902                 return f_flags;
3903
3904         spin_lock_bh(&map_idr_lock);
3905         map = idr_find(&map_idr, id);
3906         if (map)
3907                 map = __bpf_map_inc_not_zero(map, true);
3908         else
3909                 map = ERR_PTR(-ENOENT);
3910         spin_unlock_bh(&map_idr_lock);
3911
3912         if (IS_ERR(map))
3913                 return PTR_ERR(map);
3914
3915         fd = bpf_map_new_fd(map, f_flags);
3916         if (fd < 0)
3917                 bpf_map_put_with_uref(map);
3918
3919         return fd;
3920 }
3921
3922 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3923                                               unsigned long addr, u32 *off,
3924                                               u32 *type)
3925 {
3926         const struct bpf_map *map;
3927         int i;
3928
3929         mutex_lock(&prog->aux->used_maps_mutex);
3930         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3931                 map = prog->aux->used_maps[i];
3932                 if (map == (void *)addr) {
3933                         *type = BPF_PSEUDO_MAP_FD;
3934                         goto out;
3935                 }
3936                 if (!map->ops->map_direct_value_meta)
3937                         continue;
3938                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3939                         *type = BPF_PSEUDO_MAP_VALUE;
3940                         goto out;
3941                 }
3942         }
3943         map = NULL;
3944
3945 out:
3946         mutex_unlock(&prog->aux->used_maps_mutex);
3947         return map;
3948 }
3949
3950 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3951                                               const struct cred *f_cred)
3952 {
3953         const struct bpf_map *map;
3954         struct bpf_insn *insns;
3955         u32 off, type;
3956         u64 imm;
3957         u8 code;
3958         int i;
3959
3960         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3961                         GFP_USER);
3962         if (!insns)
3963                 return insns;
3964
3965         for (i = 0; i < prog->len; i++) {
3966                 code = insns[i].code;
3967
3968                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3969                         insns[i].code = BPF_JMP | BPF_CALL;
3970                         insns[i].imm = BPF_FUNC_tail_call;
3971                         /* fall-through */
3972                 }
3973                 if (code == (BPF_JMP | BPF_CALL) ||
3974                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3975                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3976                                 insns[i].code = BPF_JMP | BPF_CALL;
3977                         if (!bpf_dump_raw_ok(f_cred))
3978                                 insns[i].imm = 0;
3979                         continue;
3980                 }
3981                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3982                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3983                         continue;
3984                 }
3985
3986                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3987                         continue;
3988
3989                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3990                 map = bpf_map_from_imm(prog, imm, &off, &type);
3991                 if (map) {
3992                         insns[i].src_reg = type;
3993                         insns[i].imm = map->id;
3994                         insns[i + 1].imm = off;
3995                         continue;
3996                 }
3997         }
3998
3999         return insns;
4000 }
4001
4002 static int set_info_rec_size(struct bpf_prog_info *info)
4003 {
4004         /*
4005          * Ensure info.*_rec_size is the same as kernel expected size
4006          *
4007          * or
4008          *
4009          * Only allow zero *_rec_size if both _rec_size and _cnt are
4010          * zero.  In this case, the kernel will set the expected
4011          * _rec_size back to the info.
4012          */
4013
4014         if ((info->nr_func_info || info->func_info_rec_size) &&
4015             info->func_info_rec_size != sizeof(struct bpf_func_info))
4016                 return -EINVAL;
4017
4018         if ((info->nr_line_info || info->line_info_rec_size) &&
4019             info->line_info_rec_size != sizeof(struct bpf_line_info))
4020                 return -EINVAL;
4021
4022         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4023             info->jited_line_info_rec_size != sizeof(__u64))
4024                 return -EINVAL;
4025
4026         info->func_info_rec_size = sizeof(struct bpf_func_info);
4027         info->line_info_rec_size = sizeof(struct bpf_line_info);
4028         info->jited_line_info_rec_size = sizeof(__u64);
4029
4030         return 0;
4031 }
4032
4033 static int bpf_prog_get_info_by_fd(struct file *file,
4034                                    struct bpf_prog *prog,
4035                                    const union bpf_attr *attr,
4036                                    union bpf_attr __user *uattr)
4037 {
4038         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4039         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4040         struct bpf_prog_info info;
4041         u32 info_len = attr->info.info_len;
4042         struct bpf_prog_kstats stats;
4043         char __user *uinsns;
4044         u32 ulen;
4045         int err;
4046
4047         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4048         if (err)
4049                 return err;
4050         info_len = min_t(u32, sizeof(info), info_len);
4051
4052         memset(&info, 0, sizeof(info));
4053         if (copy_from_user(&info, uinfo, info_len))
4054                 return -EFAULT;
4055
4056         info.type = prog->type;
4057         info.id = prog->aux->id;
4058         info.load_time = prog->aux->load_time;
4059         info.created_by_uid = from_kuid_munged(current_user_ns(),
4060                                                prog->aux->user->uid);
4061         info.gpl_compatible = prog->gpl_compatible;
4062
4063         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4064         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4065
4066         mutex_lock(&prog->aux->used_maps_mutex);
4067         ulen = info.nr_map_ids;
4068         info.nr_map_ids = prog->aux->used_map_cnt;
4069         ulen = min_t(u32, info.nr_map_ids, ulen);
4070         if (ulen) {
4071                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4072                 u32 i;
4073
4074                 for (i = 0; i < ulen; i++)
4075                         if (put_user(prog->aux->used_maps[i]->id,
4076                                      &user_map_ids[i])) {
4077                                 mutex_unlock(&prog->aux->used_maps_mutex);
4078                                 return -EFAULT;
4079                         }
4080         }
4081         mutex_unlock(&prog->aux->used_maps_mutex);
4082
4083         err = set_info_rec_size(&info);
4084         if (err)
4085                 return err;
4086
4087         bpf_prog_get_stats(prog, &stats);
4088         info.run_time_ns = stats.nsecs;
4089         info.run_cnt = stats.cnt;
4090         info.recursion_misses = stats.misses;
4091
4092         info.verified_insns = prog->aux->verified_insns;
4093
4094         if (!bpf_capable()) {
4095                 info.jited_prog_len = 0;
4096                 info.xlated_prog_len = 0;
4097                 info.nr_jited_ksyms = 0;
4098                 info.nr_jited_func_lens = 0;
4099                 info.nr_func_info = 0;
4100                 info.nr_line_info = 0;
4101                 info.nr_jited_line_info = 0;
4102                 goto done;
4103         }
4104
4105         ulen = info.xlated_prog_len;
4106         info.xlated_prog_len = bpf_prog_insn_size(prog);
4107         if (info.xlated_prog_len && ulen) {
4108                 struct bpf_insn *insns_sanitized;
4109                 bool fault;
4110
4111                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4112                         info.xlated_prog_insns = 0;
4113                         goto done;
4114                 }
4115                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4116                 if (!insns_sanitized)
4117                         return -ENOMEM;
4118                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4119                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4120                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4121                 kfree(insns_sanitized);
4122                 if (fault)
4123                         return -EFAULT;
4124         }
4125
4126         if (bpf_prog_is_offloaded(prog->aux)) {
4127                 err = bpf_prog_offload_info_fill(&info, prog);
4128                 if (err)
4129                         return err;
4130                 goto done;
4131         }
4132
4133         /* NOTE: the following code is supposed to be skipped for offload.
4134          * bpf_prog_offload_info_fill() is the place to fill similar fields
4135          * for offload.
4136          */
4137         ulen = info.jited_prog_len;
4138         if (prog->aux->func_cnt) {
4139                 u32 i;
4140
4141                 info.jited_prog_len = 0;
4142                 for (i = 0; i < prog->aux->func_cnt; i++)
4143                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4144         } else {
4145                 info.jited_prog_len = prog->jited_len;
4146         }
4147
4148         if (info.jited_prog_len && ulen) {
4149                 if (bpf_dump_raw_ok(file->f_cred)) {
4150                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4151                         ulen = min_t(u32, info.jited_prog_len, ulen);
4152
4153                         /* for multi-function programs, copy the JITed
4154                          * instructions for all the functions
4155                          */
4156                         if (prog->aux->func_cnt) {
4157                                 u32 len, free, i;
4158                                 u8 *img;
4159
4160                                 free = ulen;
4161                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4162                                         len = prog->aux->func[i]->jited_len;
4163                                         len = min_t(u32, len, free);
4164                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4165                                         if (copy_to_user(uinsns, img, len))
4166                                                 return -EFAULT;
4167                                         uinsns += len;
4168                                         free -= len;
4169                                         if (!free)
4170                                                 break;
4171                                 }
4172                         } else {
4173                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4174                                         return -EFAULT;
4175                         }
4176                 } else {
4177                         info.jited_prog_insns = 0;
4178                 }
4179         }
4180
4181         ulen = info.nr_jited_ksyms;
4182         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4183         if (ulen) {
4184                 if (bpf_dump_raw_ok(file->f_cred)) {
4185                         unsigned long ksym_addr;
4186                         u64 __user *user_ksyms;
4187                         u32 i;
4188
4189                         /* copy the address of the kernel symbol
4190                          * corresponding to each function
4191                          */
4192                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4193                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4194                         if (prog->aux->func_cnt) {
4195                                 for (i = 0; i < ulen; i++) {
4196                                         ksym_addr = (unsigned long)
4197                                                 prog->aux->func[i]->bpf_func;
4198                                         if (put_user((u64) ksym_addr,
4199                                                      &user_ksyms[i]))
4200                                                 return -EFAULT;
4201                                 }
4202                         } else {
4203                                 ksym_addr = (unsigned long) prog->bpf_func;
4204                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4205                                         return -EFAULT;
4206                         }
4207                 } else {
4208                         info.jited_ksyms = 0;
4209                 }
4210         }
4211
4212         ulen = info.nr_jited_func_lens;
4213         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4214         if (ulen) {
4215                 if (bpf_dump_raw_ok(file->f_cred)) {
4216                         u32 __user *user_lens;
4217                         u32 func_len, i;
4218
4219                         /* copy the JITed image lengths for each function */
4220                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4221                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4222                         if (prog->aux->func_cnt) {
4223                                 for (i = 0; i < ulen; i++) {
4224                                         func_len =
4225                                                 prog->aux->func[i]->jited_len;
4226                                         if (put_user(func_len, &user_lens[i]))
4227                                                 return -EFAULT;
4228                                 }
4229                         } else {
4230                                 func_len = prog->jited_len;
4231                                 if (put_user(func_len, &user_lens[0]))
4232                                         return -EFAULT;
4233                         }
4234                 } else {
4235                         info.jited_func_lens = 0;
4236                 }
4237         }
4238
4239         if (prog->aux->btf)
4240                 info.btf_id = btf_obj_id(prog->aux->btf);
4241         info.attach_btf_id = prog->aux->attach_btf_id;
4242         if (attach_btf)
4243                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4244
4245         ulen = info.nr_func_info;
4246         info.nr_func_info = prog->aux->func_info_cnt;
4247         if (info.nr_func_info && ulen) {
4248                 char __user *user_finfo;
4249
4250                 user_finfo = u64_to_user_ptr(info.func_info);
4251                 ulen = min_t(u32, info.nr_func_info, ulen);
4252                 if (copy_to_user(user_finfo, prog->aux->func_info,
4253                                  info.func_info_rec_size * ulen))
4254                         return -EFAULT;
4255         }
4256
4257         ulen = info.nr_line_info;
4258         info.nr_line_info = prog->aux->nr_linfo;
4259         if (info.nr_line_info && ulen) {
4260                 __u8 __user *user_linfo;
4261
4262                 user_linfo = u64_to_user_ptr(info.line_info);
4263                 ulen = min_t(u32, info.nr_line_info, ulen);
4264                 if (copy_to_user(user_linfo, prog->aux->linfo,
4265                                  info.line_info_rec_size * ulen))
4266                         return -EFAULT;
4267         }
4268
4269         ulen = info.nr_jited_line_info;
4270         if (prog->aux->jited_linfo)
4271                 info.nr_jited_line_info = prog->aux->nr_linfo;
4272         else
4273                 info.nr_jited_line_info = 0;
4274         if (info.nr_jited_line_info && ulen) {
4275                 if (bpf_dump_raw_ok(file->f_cred)) {
4276                         unsigned long line_addr;
4277                         __u64 __user *user_linfo;
4278                         u32 i;
4279
4280                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4281                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4282                         for (i = 0; i < ulen; i++) {
4283                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4284                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4285                                         return -EFAULT;
4286                         }
4287                 } else {
4288                         info.jited_line_info = 0;
4289                 }
4290         }
4291
4292         ulen = info.nr_prog_tags;
4293         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4294         if (ulen) {
4295                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4296                 u32 i;
4297
4298                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4299                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4300                 if (prog->aux->func_cnt) {
4301                         for (i = 0; i < ulen; i++) {
4302                                 if (copy_to_user(user_prog_tags[i],
4303                                                  prog->aux->func[i]->tag,
4304                                                  BPF_TAG_SIZE))
4305                                         return -EFAULT;
4306                         }
4307                 } else {
4308                         if (copy_to_user(user_prog_tags[0],
4309                                          prog->tag, BPF_TAG_SIZE))
4310                                 return -EFAULT;
4311                 }
4312         }
4313
4314 done:
4315         if (copy_to_user(uinfo, &info, info_len) ||
4316             put_user(info_len, &uattr->info.info_len))
4317                 return -EFAULT;
4318
4319         return 0;
4320 }
4321
4322 static int bpf_map_get_info_by_fd(struct file *file,
4323                                   struct bpf_map *map,
4324                                   const union bpf_attr *attr,
4325                                   union bpf_attr __user *uattr)
4326 {
4327         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4328         struct bpf_map_info info;
4329         u32 info_len = attr->info.info_len;
4330         int err;
4331
4332         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4333         if (err)
4334                 return err;
4335         info_len = min_t(u32, sizeof(info), info_len);
4336
4337         memset(&info, 0, sizeof(info));
4338         info.type = map->map_type;
4339         info.id = map->id;
4340         info.key_size = map->key_size;
4341         info.value_size = map->value_size;
4342         info.max_entries = map->max_entries;
4343         info.map_flags = map->map_flags;
4344         info.map_extra = map->map_extra;
4345         memcpy(info.name, map->name, sizeof(map->name));
4346
4347         if (map->btf) {
4348                 info.btf_id = btf_obj_id(map->btf);
4349                 info.btf_key_type_id = map->btf_key_type_id;
4350                 info.btf_value_type_id = map->btf_value_type_id;
4351         }
4352         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4353
4354         if (bpf_map_is_offloaded(map)) {
4355                 err = bpf_map_offload_info_fill(&info, map);
4356                 if (err)
4357                         return err;
4358         }
4359
4360         if (copy_to_user(uinfo, &info, info_len) ||
4361             put_user(info_len, &uattr->info.info_len))
4362                 return -EFAULT;
4363
4364         return 0;
4365 }
4366
4367 static int bpf_btf_get_info_by_fd(struct file *file,
4368                                   struct btf *btf,
4369                                   const union bpf_attr *attr,
4370                                   union bpf_attr __user *uattr)
4371 {
4372         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4373         u32 info_len = attr->info.info_len;
4374         int err;
4375
4376         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4377         if (err)
4378                 return err;
4379
4380         return btf_get_info_by_fd(btf, attr, uattr);
4381 }
4382
4383 static int bpf_link_get_info_by_fd(struct file *file,
4384                                   struct bpf_link *link,
4385                                   const union bpf_attr *attr,
4386                                   union bpf_attr __user *uattr)
4387 {
4388         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4389         struct bpf_link_info info;
4390         u32 info_len = attr->info.info_len;
4391         int err;
4392
4393         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4394         if (err)
4395                 return err;
4396         info_len = min_t(u32, sizeof(info), info_len);
4397
4398         memset(&info, 0, sizeof(info));
4399         if (copy_from_user(&info, uinfo, info_len))
4400                 return -EFAULT;
4401
4402         info.type = link->type;
4403         info.id = link->id;
4404         if (link->prog)
4405                 info.prog_id = link->prog->aux->id;
4406
4407         if (link->ops->fill_link_info) {
4408                 err = link->ops->fill_link_info(link, &info);
4409                 if (err)
4410                         return err;
4411         }
4412
4413         if (copy_to_user(uinfo, &info, info_len) ||
4414             put_user(info_len, &uattr->info.info_len))
4415                 return -EFAULT;
4416
4417         return 0;
4418 }
4419
4420
4421 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4422
4423 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4424                                   union bpf_attr __user *uattr)
4425 {
4426         int ufd = attr->info.bpf_fd;
4427         struct fd f;
4428         int err;
4429
4430         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4431                 return -EINVAL;
4432
4433         f = fdget(ufd);
4434         if (!f.file)
4435                 return -EBADFD;
4436
4437         if (f.file->f_op == &bpf_prog_fops)
4438                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4439                                               uattr);
4440         else if (f.file->f_op == &bpf_map_fops)
4441                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4442                                              uattr);
4443         else if (f.file->f_op == &btf_fops)
4444                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4445         else if (f.file->f_op == &bpf_link_fops)
4446                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4447                                               attr, uattr);
4448         else
4449                 err = -EINVAL;
4450
4451         fdput(f);
4452         return err;
4453 }
4454
4455 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4456
4457 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4458 {
4459         if (CHECK_ATTR(BPF_BTF_LOAD))
4460                 return -EINVAL;
4461
4462         if (!bpf_capable())
4463                 return -EPERM;
4464
4465         return btf_new_fd(attr, uattr, uattr_size);
4466 }
4467
4468 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4469
4470 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4471 {
4472         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4473                 return -EINVAL;
4474
4475         if (!capable(CAP_SYS_ADMIN))
4476                 return -EPERM;
4477
4478         return btf_get_fd_by_id(attr->btf_id);
4479 }
4480
4481 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4482                                     union bpf_attr __user *uattr,
4483                                     u32 prog_id, u32 fd_type,
4484                                     const char *buf, u64 probe_offset,
4485                                     u64 probe_addr)
4486 {
4487         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4488         u32 len = buf ? strlen(buf) : 0, input_len;
4489         int err = 0;
4490
4491         if (put_user(len, &uattr->task_fd_query.buf_len))
4492                 return -EFAULT;
4493         input_len = attr->task_fd_query.buf_len;
4494         if (input_len && ubuf) {
4495                 if (!len) {
4496                         /* nothing to copy, just make ubuf NULL terminated */
4497                         char zero = '\0';
4498
4499                         if (put_user(zero, ubuf))
4500                                 return -EFAULT;
4501                 } else if (input_len >= len + 1) {
4502                         /* ubuf can hold the string with NULL terminator */
4503                         if (copy_to_user(ubuf, buf, len + 1))
4504                                 return -EFAULT;
4505                 } else {
4506                         /* ubuf cannot hold the string with NULL terminator,
4507                          * do a partial copy with NULL terminator.
4508                          */
4509                         char zero = '\0';
4510
4511                         err = -ENOSPC;
4512                         if (copy_to_user(ubuf, buf, input_len - 1))
4513                                 return -EFAULT;
4514                         if (put_user(zero, ubuf + input_len - 1))
4515                                 return -EFAULT;
4516                 }
4517         }
4518
4519         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4520             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4521             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4522             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4523                 return -EFAULT;
4524
4525         return err;
4526 }
4527
4528 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4529
4530 static int bpf_task_fd_query(const union bpf_attr *attr,
4531                              union bpf_attr __user *uattr)
4532 {
4533         pid_t pid = attr->task_fd_query.pid;
4534         u32 fd = attr->task_fd_query.fd;
4535         const struct perf_event *event;
4536         struct task_struct *task;
4537         struct file *file;
4538         int err;
4539
4540         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4541                 return -EINVAL;
4542
4543         if (!capable(CAP_SYS_ADMIN))
4544                 return -EPERM;
4545
4546         if (attr->task_fd_query.flags != 0)
4547                 return -EINVAL;
4548
4549         rcu_read_lock();
4550         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4551         rcu_read_unlock();
4552         if (!task)
4553                 return -ENOENT;
4554
4555         err = 0;
4556         file = fget_task(task, fd);
4557         put_task_struct(task);
4558         if (!file)
4559                 return -EBADF;
4560
4561         if (file->f_op == &bpf_link_fops) {
4562                 struct bpf_link *link = file->private_data;
4563
4564                 if (link->ops == &bpf_raw_tp_link_lops) {
4565                         struct bpf_raw_tp_link *raw_tp =
4566                                 container_of(link, struct bpf_raw_tp_link, link);
4567                         struct bpf_raw_event_map *btp = raw_tp->btp;
4568
4569                         err = bpf_task_fd_query_copy(attr, uattr,
4570                                                      raw_tp->link.prog->aux->id,
4571                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4572                                                      btp->tp->name, 0, 0);
4573                         goto put_file;
4574                 }
4575                 goto out_not_supp;
4576         }
4577
4578         event = perf_get_event(file);
4579         if (!IS_ERR(event)) {
4580                 u64 probe_offset, probe_addr;
4581                 u32 prog_id, fd_type;
4582                 const char *buf;
4583
4584                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4585                                               &buf, &probe_offset,
4586                                               &probe_addr);
4587                 if (!err)
4588                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4589                                                      fd_type, buf,
4590                                                      probe_offset,
4591                                                      probe_addr);
4592                 goto put_file;
4593         }
4594
4595 out_not_supp:
4596         err = -ENOTSUPP;
4597 put_file:
4598         fput(file);
4599         return err;
4600 }
4601
4602 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4603
4604 #define BPF_DO_BATCH(fn, ...)                   \
4605         do {                                    \
4606                 if (!fn) {                      \
4607                         err = -ENOTSUPP;        \
4608                         goto err_put;           \
4609                 }                               \
4610                 err = fn(__VA_ARGS__);          \
4611         } while (0)
4612
4613 static int bpf_map_do_batch(const union bpf_attr *attr,
4614                             union bpf_attr __user *uattr,
4615                             int cmd)
4616 {
4617         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4618                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4619         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4620         struct bpf_map *map;
4621         int err, ufd;
4622         struct fd f;
4623
4624         if (CHECK_ATTR(BPF_MAP_BATCH))
4625                 return -EINVAL;
4626
4627         ufd = attr->batch.map_fd;
4628         f = fdget(ufd);
4629         map = __bpf_map_get(f);
4630         if (IS_ERR(map))
4631                 return PTR_ERR(map);
4632         if (has_write)
4633                 bpf_map_write_active_inc(map);
4634         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4635                 err = -EPERM;
4636                 goto err_put;
4637         }
4638         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4639                 err = -EPERM;
4640                 goto err_put;
4641         }
4642
4643         if (cmd == BPF_MAP_LOOKUP_BATCH)
4644                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4645         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4646                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4647         else if (cmd == BPF_MAP_UPDATE_BATCH)
4648                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4649         else
4650                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4651 err_put:
4652         if (has_write)
4653                 bpf_map_write_active_dec(map);
4654         fdput(f);
4655         return err;
4656 }
4657
4658 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4659 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4660 {
4661         enum bpf_prog_type ptype;
4662         struct bpf_prog *prog;
4663         int ret;
4664
4665         if (CHECK_ATTR(BPF_LINK_CREATE))
4666                 return -EINVAL;
4667
4668         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
4669                 return bpf_struct_ops_link_create(attr);
4670
4671         prog = bpf_prog_get(attr->link_create.prog_fd);
4672         if (IS_ERR(prog))
4673                 return PTR_ERR(prog);
4674
4675         ret = bpf_prog_attach_check_attach_type(prog,
4676                                                 attr->link_create.attach_type);
4677         if (ret)
4678                 goto out;
4679
4680         switch (prog->type) {
4681         case BPF_PROG_TYPE_EXT:
4682                 break;
4683         case BPF_PROG_TYPE_NETFILTER:
4684                 if (attr->link_create.attach_type != BPF_NETFILTER) {
4685                         ret = -EINVAL;
4686                         goto out;
4687                 }
4688                 break;
4689         case BPF_PROG_TYPE_PERF_EVENT:
4690         case BPF_PROG_TYPE_TRACEPOINT:
4691                 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4692                         ret = -EINVAL;
4693                         goto out;
4694                 }
4695                 break;
4696         case BPF_PROG_TYPE_KPROBE:
4697                 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4698                     attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4699                         ret = -EINVAL;
4700                         goto out;
4701                 }
4702                 break;
4703         default:
4704                 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4705                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4706                         ret = -EINVAL;
4707                         goto out;
4708                 }
4709                 break;
4710         }
4711
4712         switch (prog->type) {
4713         case BPF_PROG_TYPE_CGROUP_SKB:
4714         case BPF_PROG_TYPE_CGROUP_SOCK:
4715         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4716         case BPF_PROG_TYPE_SOCK_OPS:
4717         case BPF_PROG_TYPE_CGROUP_DEVICE:
4718         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4719         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4720                 ret = cgroup_bpf_link_attach(attr, prog);
4721                 break;
4722         case BPF_PROG_TYPE_EXT:
4723                 ret = bpf_tracing_prog_attach(prog,
4724                                               attr->link_create.target_fd,
4725                                               attr->link_create.target_btf_id,
4726                                               attr->link_create.tracing.cookie);
4727                 break;
4728         case BPF_PROG_TYPE_LSM:
4729         case BPF_PROG_TYPE_TRACING:
4730                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4731                         ret = -EINVAL;
4732                         goto out;
4733                 }
4734                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4735                         ret = bpf_raw_tp_link_attach(prog, NULL);
4736                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4737                         ret = bpf_iter_link_attach(attr, uattr, prog);
4738                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4739                         ret = cgroup_bpf_link_attach(attr, prog);
4740                 else
4741                         ret = bpf_tracing_prog_attach(prog,
4742                                                       attr->link_create.target_fd,
4743                                                       attr->link_create.target_btf_id,
4744                                                       attr->link_create.tracing.cookie);
4745                 break;
4746         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4747         case BPF_PROG_TYPE_SK_LOOKUP:
4748                 ret = netns_bpf_link_create(attr, prog);
4749                 break;
4750 #ifdef CONFIG_NET
4751         case BPF_PROG_TYPE_XDP:
4752                 ret = bpf_xdp_link_attach(attr, prog);
4753                 break;
4754         case BPF_PROG_TYPE_NETFILTER:
4755                 ret = bpf_nf_link_attach(attr, prog);
4756                 break;
4757 #endif
4758         case BPF_PROG_TYPE_PERF_EVENT:
4759         case BPF_PROG_TYPE_TRACEPOINT:
4760                 ret = bpf_perf_link_attach(attr, prog);
4761                 break;
4762         case BPF_PROG_TYPE_KPROBE:
4763                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4764                         ret = bpf_perf_link_attach(attr, prog);
4765                 else
4766                         ret = bpf_kprobe_multi_link_attach(attr, prog);
4767                 break;
4768         default:
4769                 ret = -EINVAL;
4770         }
4771
4772 out:
4773         if (ret < 0)
4774                 bpf_prog_put(prog);
4775         return ret;
4776 }
4777
4778 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
4779 {
4780         struct bpf_map *new_map, *old_map = NULL;
4781         int ret;
4782
4783         new_map = bpf_map_get(attr->link_update.new_map_fd);
4784         if (IS_ERR(new_map))
4785                 return PTR_ERR(new_map);
4786
4787         if (attr->link_update.flags & BPF_F_REPLACE) {
4788                 old_map = bpf_map_get(attr->link_update.old_map_fd);
4789                 if (IS_ERR(old_map)) {
4790                         ret = PTR_ERR(old_map);
4791                         goto out_put;
4792                 }
4793         } else if (attr->link_update.old_map_fd) {
4794                 ret = -EINVAL;
4795                 goto out_put;
4796         }
4797
4798         ret = link->ops->update_map(link, new_map, old_map);
4799
4800         if (old_map)
4801                 bpf_map_put(old_map);
4802 out_put:
4803         bpf_map_put(new_map);
4804         return ret;
4805 }
4806
4807 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4808
4809 static int link_update(union bpf_attr *attr)
4810 {
4811         struct bpf_prog *old_prog = NULL, *new_prog;
4812         struct bpf_link *link;
4813         u32 flags;
4814         int ret;
4815
4816         if (CHECK_ATTR(BPF_LINK_UPDATE))
4817                 return -EINVAL;
4818
4819         flags = attr->link_update.flags;
4820         if (flags & ~BPF_F_REPLACE)
4821                 return -EINVAL;
4822
4823         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4824         if (IS_ERR(link))
4825                 return PTR_ERR(link);
4826
4827         if (link->ops->update_map) {
4828                 ret = link_update_map(link, attr);
4829                 goto out_put_link;
4830         }
4831
4832         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4833         if (IS_ERR(new_prog)) {
4834                 ret = PTR_ERR(new_prog);
4835                 goto out_put_link;
4836         }
4837
4838         if (flags & BPF_F_REPLACE) {
4839                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4840                 if (IS_ERR(old_prog)) {
4841                         ret = PTR_ERR(old_prog);
4842                         old_prog = NULL;
4843                         goto out_put_progs;
4844                 }
4845         } else if (attr->link_update.old_prog_fd) {
4846                 ret = -EINVAL;
4847                 goto out_put_progs;
4848         }
4849
4850         if (link->ops->update_prog)
4851                 ret = link->ops->update_prog(link, new_prog, old_prog);
4852         else
4853                 ret = -EINVAL;
4854
4855 out_put_progs:
4856         if (old_prog)
4857                 bpf_prog_put(old_prog);
4858         if (ret)
4859                 bpf_prog_put(new_prog);
4860 out_put_link:
4861         bpf_link_put_direct(link);
4862         return ret;
4863 }
4864
4865 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4866
4867 static int link_detach(union bpf_attr *attr)
4868 {
4869         struct bpf_link *link;
4870         int ret;
4871
4872         if (CHECK_ATTR(BPF_LINK_DETACH))
4873                 return -EINVAL;
4874
4875         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4876         if (IS_ERR(link))
4877                 return PTR_ERR(link);
4878
4879         if (link->ops->detach)
4880                 ret = link->ops->detach(link);
4881         else
4882                 ret = -EOPNOTSUPP;
4883
4884         bpf_link_put_direct(link);
4885         return ret;
4886 }
4887
4888 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4889 {
4890         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4891 }
4892
4893 struct bpf_link *bpf_link_by_id(u32 id)
4894 {
4895         struct bpf_link *link;
4896
4897         if (!id)
4898                 return ERR_PTR(-ENOENT);
4899
4900         spin_lock_bh(&link_idr_lock);
4901         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4902         link = idr_find(&link_idr, id);
4903         if (link) {
4904                 if (link->id)
4905                         link = bpf_link_inc_not_zero(link);
4906                 else
4907                         link = ERR_PTR(-EAGAIN);
4908         } else {
4909                 link = ERR_PTR(-ENOENT);
4910         }
4911         spin_unlock_bh(&link_idr_lock);
4912         return link;
4913 }
4914
4915 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4916 {
4917         struct bpf_link *link;
4918
4919         spin_lock_bh(&link_idr_lock);
4920 again:
4921         link = idr_get_next(&link_idr, id);
4922         if (link) {
4923                 link = bpf_link_inc_not_zero(link);
4924                 if (IS_ERR(link)) {
4925                         (*id)++;
4926                         goto again;
4927                 }
4928         }
4929         spin_unlock_bh(&link_idr_lock);
4930
4931         return link;
4932 }
4933
4934 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4935
4936 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4937 {
4938         struct bpf_link *link;
4939         u32 id = attr->link_id;
4940         int fd;
4941
4942         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4943                 return -EINVAL;
4944
4945         if (!capable(CAP_SYS_ADMIN))
4946                 return -EPERM;
4947
4948         link = bpf_link_by_id(id);
4949         if (IS_ERR(link))
4950                 return PTR_ERR(link);
4951
4952         fd = bpf_link_new_fd(link);
4953         if (fd < 0)
4954                 bpf_link_put_direct(link);
4955
4956         return fd;
4957 }
4958
4959 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4960
4961 static int bpf_stats_release(struct inode *inode, struct file *file)
4962 {
4963         mutex_lock(&bpf_stats_enabled_mutex);
4964         static_key_slow_dec(&bpf_stats_enabled_key.key);
4965         mutex_unlock(&bpf_stats_enabled_mutex);
4966         return 0;
4967 }
4968
4969 static const struct file_operations bpf_stats_fops = {
4970         .release = bpf_stats_release,
4971 };
4972
4973 static int bpf_enable_runtime_stats(void)
4974 {
4975         int fd;
4976
4977         mutex_lock(&bpf_stats_enabled_mutex);
4978
4979         /* Set a very high limit to avoid overflow */
4980         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4981                 mutex_unlock(&bpf_stats_enabled_mutex);
4982                 return -EBUSY;
4983         }
4984
4985         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4986         if (fd >= 0)
4987                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4988
4989         mutex_unlock(&bpf_stats_enabled_mutex);
4990         return fd;
4991 }
4992
4993 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4994
4995 static int bpf_enable_stats(union bpf_attr *attr)
4996 {
4997
4998         if (CHECK_ATTR(BPF_ENABLE_STATS))
4999                 return -EINVAL;
5000
5001         if (!capable(CAP_SYS_ADMIN))
5002                 return -EPERM;
5003
5004         switch (attr->enable_stats.type) {
5005         case BPF_STATS_RUN_TIME:
5006                 return bpf_enable_runtime_stats();
5007         default:
5008                 break;
5009         }
5010         return -EINVAL;
5011 }
5012
5013 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5014
5015 static int bpf_iter_create(union bpf_attr *attr)
5016 {
5017         struct bpf_link *link;
5018         int err;
5019
5020         if (CHECK_ATTR(BPF_ITER_CREATE))
5021                 return -EINVAL;
5022
5023         if (attr->iter_create.flags)
5024                 return -EINVAL;
5025
5026         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5027         if (IS_ERR(link))
5028                 return PTR_ERR(link);
5029
5030         err = bpf_iter_new_fd(link);
5031         bpf_link_put_direct(link);
5032
5033         return err;
5034 }
5035
5036 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5037
5038 static int bpf_prog_bind_map(union bpf_attr *attr)
5039 {
5040         struct bpf_prog *prog;
5041         struct bpf_map *map;
5042         struct bpf_map **used_maps_old, **used_maps_new;
5043         int i, ret = 0;
5044
5045         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5046                 return -EINVAL;
5047
5048         if (attr->prog_bind_map.flags)
5049                 return -EINVAL;
5050
5051         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5052         if (IS_ERR(prog))
5053                 return PTR_ERR(prog);
5054
5055         map = bpf_map_get(attr->prog_bind_map.map_fd);
5056         if (IS_ERR(map)) {
5057                 ret = PTR_ERR(map);
5058                 goto out_prog_put;
5059         }
5060
5061         mutex_lock(&prog->aux->used_maps_mutex);
5062
5063         used_maps_old = prog->aux->used_maps;
5064
5065         for (i = 0; i < prog->aux->used_map_cnt; i++)
5066                 if (used_maps_old[i] == map) {
5067                         bpf_map_put(map);
5068                         goto out_unlock;
5069                 }
5070
5071         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5072                                       sizeof(used_maps_new[0]),
5073                                       GFP_KERNEL);
5074         if (!used_maps_new) {
5075                 ret = -ENOMEM;
5076                 goto out_unlock;
5077         }
5078
5079         memcpy(used_maps_new, used_maps_old,
5080                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5081         used_maps_new[prog->aux->used_map_cnt] = map;
5082
5083         prog->aux->used_map_cnt++;
5084         prog->aux->used_maps = used_maps_new;
5085
5086         kfree(used_maps_old);
5087
5088 out_unlock:
5089         mutex_unlock(&prog->aux->used_maps_mutex);
5090
5091         if (ret)
5092                 bpf_map_put(map);
5093 out_prog_put:
5094         bpf_prog_put(prog);
5095         return ret;
5096 }
5097
5098 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5099 {
5100         union bpf_attr attr;
5101         int err;
5102
5103         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5104         if (err)
5105                 return err;
5106         size = min_t(u32, size, sizeof(attr));
5107
5108         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5109         memset(&attr, 0, sizeof(attr));
5110         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5111                 return -EFAULT;
5112
5113         err = security_bpf(cmd, &attr, size);
5114         if (err < 0)
5115                 return err;
5116
5117         switch (cmd) {
5118         case BPF_MAP_CREATE:
5119                 err = map_create(&attr);
5120                 break;
5121         case BPF_MAP_LOOKUP_ELEM:
5122                 err = map_lookup_elem(&attr);
5123                 break;
5124         case BPF_MAP_UPDATE_ELEM:
5125                 err = map_update_elem(&attr, uattr);
5126                 break;
5127         case BPF_MAP_DELETE_ELEM:
5128                 err = map_delete_elem(&attr, uattr);
5129                 break;
5130         case BPF_MAP_GET_NEXT_KEY:
5131                 err = map_get_next_key(&attr);
5132                 break;
5133         case BPF_MAP_FREEZE:
5134                 err = map_freeze(&attr);
5135                 break;
5136         case BPF_PROG_LOAD:
5137                 err = bpf_prog_load(&attr, uattr, size);
5138                 break;
5139         case BPF_OBJ_PIN:
5140                 err = bpf_obj_pin(&attr);
5141                 break;
5142         case BPF_OBJ_GET:
5143                 err = bpf_obj_get(&attr);
5144                 break;
5145         case BPF_PROG_ATTACH:
5146                 err = bpf_prog_attach(&attr);
5147                 break;
5148         case BPF_PROG_DETACH:
5149                 err = bpf_prog_detach(&attr);
5150                 break;
5151         case BPF_PROG_QUERY:
5152                 err = bpf_prog_query(&attr, uattr.user);
5153                 break;
5154         case BPF_PROG_TEST_RUN:
5155                 err = bpf_prog_test_run(&attr, uattr.user);
5156                 break;
5157         case BPF_PROG_GET_NEXT_ID:
5158                 err = bpf_obj_get_next_id(&attr, uattr.user,
5159                                           &prog_idr, &prog_idr_lock);
5160                 break;
5161         case BPF_MAP_GET_NEXT_ID:
5162                 err = bpf_obj_get_next_id(&attr, uattr.user,
5163                                           &map_idr, &map_idr_lock);
5164                 break;
5165         case BPF_BTF_GET_NEXT_ID:
5166                 err = bpf_obj_get_next_id(&attr, uattr.user,
5167                                           &btf_idr, &btf_idr_lock);
5168                 break;
5169         case BPF_PROG_GET_FD_BY_ID:
5170                 err = bpf_prog_get_fd_by_id(&attr);
5171                 break;
5172         case BPF_MAP_GET_FD_BY_ID:
5173                 err = bpf_map_get_fd_by_id(&attr);
5174                 break;
5175         case BPF_OBJ_GET_INFO_BY_FD:
5176                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5177                 break;
5178         case BPF_RAW_TRACEPOINT_OPEN:
5179                 err = bpf_raw_tracepoint_open(&attr);
5180                 break;
5181         case BPF_BTF_LOAD:
5182                 err = bpf_btf_load(&attr, uattr, size);
5183                 break;
5184         case BPF_BTF_GET_FD_BY_ID:
5185                 err = bpf_btf_get_fd_by_id(&attr);
5186                 break;
5187         case BPF_TASK_FD_QUERY:
5188                 err = bpf_task_fd_query(&attr, uattr.user);
5189                 break;
5190         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5191                 err = map_lookup_and_delete_elem(&attr);
5192                 break;
5193         case BPF_MAP_LOOKUP_BATCH:
5194                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5195                 break;
5196         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5197                 err = bpf_map_do_batch(&attr, uattr.user,
5198                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5199                 break;
5200         case BPF_MAP_UPDATE_BATCH:
5201                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5202                 break;
5203         case BPF_MAP_DELETE_BATCH:
5204                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5205                 break;
5206         case BPF_LINK_CREATE:
5207                 err = link_create(&attr, uattr);
5208                 break;
5209         case BPF_LINK_UPDATE:
5210                 err = link_update(&attr);
5211                 break;
5212         case BPF_LINK_GET_FD_BY_ID:
5213                 err = bpf_link_get_fd_by_id(&attr);
5214                 break;
5215         case BPF_LINK_GET_NEXT_ID:
5216                 err = bpf_obj_get_next_id(&attr, uattr.user,
5217                                           &link_idr, &link_idr_lock);
5218                 break;
5219         case BPF_ENABLE_STATS:
5220                 err = bpf_enable_stats(&attr);
5221                 break;
5222         case BPF_ITER_CREATE:
5223                 err = bpf_iter_create(&attr);
5224                 break;
5225         case BPF_LINK_DETACH:
5226                 err = link_detach(&attr);
5227                 break;
5228         case BPF_PROG_BIND_MAP:
5229                 err = bpf_prog_bind_map(&attr);
5230                 break;
5231         default:
5232                 err = -EINVAL;
5233                 break;
5234         }
5235
5236         return err;
5237 }
5238
5239 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5240 {
5241         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5242 }
5243
5244 static bool syscall_prog_is_valid_access(int off, int size,
5245                                          enum bpf_access_type type,
5246                                          const struct bpf_prog *prog,
5247                                          struct bpf_insn_access_aux *info)
5248 {
5249         if (off < 0 || off >= U16_MAX)
5250                 return false;
5251         if (off % size != 0)
5252                 return false;
5253         return true;
5254 }
5255
5256 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5257 {
5258         switch (cmd) {
5259         case BPF_MAP_CREATE:
5260         case BPF_MAP_DELETE_ELEM:
5261         case BPF_MAP_UPDATE_ELEM:
5262         case BPF_MAP_FREEZE:
5263         case BPF_MAP_GET_FD_BY_ID:
5264         case BPF_PROG_LOAD:
5265         case BPF_BTF_LOAD:
5266         case BPF_LINK_CREATE:
5267         case BPF_RAW_TRACEPOINT_OPEN:
5268                 break;
5269         default:
5270                 return -EINVAL;
5271         }
5272         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5273 }
5274
5275
5276 /* To shut up -Wmissing-prototypes.
5277  * This function is used by the kernel light skeleton
5278  * to load bpf programs when modules are loaded or during kernel boot.
5279  * See tools/lib/bpf/skel_internal.h
5280  */
5281 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5282
5283 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5284 {
5285         struct bpf_prog * __maybe_unused prog;
5286         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5287
5288         switch (cmd) {
5289 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5290         case BPF_PROG_TEST_RUN:
5291                 if (attr->test.data_in || attr->test.data_out ||
5292                     attr->test.ctx_out || attr->test.duration ||
5293                     attr->test.repeat || attr->test.flags)
5294                         return -EINVAL;
5295
5296                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5297                 if (IS_ERR(prog))
5298                         return PTR_ERR(prog);
5299
5300                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5301                     attr->test.ctx_size_in > U16_MAX) {
5302                         bpf_prog_put(prog);
5303                         return -EINVAL;
5304                 }
5305
5306                 run_ctx.bpf_cookie = 0;
5307                 run_ctx.saved_run_ctx = NULL;
5308                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5309                         /* recursion detected */
5310                         bpf_prog_put(prog);
5311                         return -EBUSY;
5312                 }
5313                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5314                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5315                                                 &run_ctx);
5316                 bpf_prog_put(prog);
5317                 return 0;
5318 #endif
5319         default:
5320                 return ____bpf_sys_bpf(cmd, attr, size);
5321         }
5322 }
5323 EXPORT_SYMBOL(kern_sys_bpf);
5324
5325 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5326         .func           = bpf_sys_bpf,
5327         .gpl_only       = false,
5328         .ret_type       = RET_INTEGER,
5329         .arg1_type      = ARG_ANYTHING,
5330         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5331         .arg3_type      = ARG_CONST_SIZE,
5332 };
5333
5334 const struct bpf_func_proto * __weak
5335 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5336 {
5337         return bpf_base_func_proto(func_id);
5338 }
5339
5340 BPF_CALL_1(bpf_sys_close, u32, fd)
5341 {
5342         /* When bpf program calls this helper there should not be
5343          * an fdget() without matching completed fdput().
5344          * This helper is allowed in the following callchain only:
5345          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5346          */
5347         return close_fd(fd);
5348 }
5349
5350 static const struct bpf_func_proto bpf_sys_close_proto = {
5351         .func           = bpf_sys_close,
5352         .gpl_only       = false,
5353         .ret_type       = RET_INTEGER,
5354         .arg1_type      = ARG_ANYTHING,
5355 };
5356
5357 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5358 {
5359         if (flags)
5360                 return -EINVAL;
5361
5362         if (name_sz <= 1 || name[name_sz - 1])
5363                 return -EINVAL;
5364
5365         if (!bpf_dump_raw_ok(current_cred()))
5366                 return -EPERM;
5367
5368         *res = kallsyms_lookup_name(name);
5369         return *res ? 0 : -ENOENT;
5370 }
5371
5372 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5373         .func           = bpf_kallsyms_lookup_name,
5374         .gpl_only       = false,
5375         .ret_type       = RET_INTEGER,
5376         .arg1_type      = ARG_PTR_TO_MEM,
5377         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5378         .arg3_type      = ARG_ANYTHING,
5379         .arg4_type      = ARG_PTR_TO_LONG,
5380 };
5381
5382 static const struct bpf_func_proto *
5383 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5384 {
5385         switch (func_id) {
5386         case BPF_FUNC_sys_bpf:
5387                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5388         case BPF_FUNC_btf_find_by_name_kind:
5389                 return &bpf_btf_find_by_name_kind_proto;
5390         case BPF_FUNC_sys_close:
5391                 return &bpf_sys_close_proto;
5392         case BPF_FUNC_kallsyms_lookup_name:
5393                 return &bpf_kallsyms_lookup_name_proto;
5394         default:
5395                 return tracing_prog_func_proto(func_id, prog);
5396         }
5397 }
5398
5399 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5400         .get_func_proto  = syscall_prog_func_proto,
5401         .is_valid_access = syscall_prog_is_valid_access,
5402 };
5403
5404 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5405         .test_run = bpf_prog_test_run_syscall,
5406 };
5407
5408 #ifdef CONFIG_SYSCTL
5409 static int bpf_stats_handler(struct ctl_table *table, int write,
5410                              void *buffer, size_t *lenp, loff_t *ppos)
5411 {
5412         struct static_key *key = (struct static_key *)table->data;
5413         static int saved_val;
5414         int val, ret;
5415         struct ctl_table tmp = {
5416                 .data   = &val,
5417                 .maxlen = sizeof(val),
5418                 .mode   = table->mode,
5419                 .extra1 = SYSCTL_ZERO,
5420                 .extra2 = SYSCTL_ONE,
5421         };
5422
5423         if (write && !capable(CAP_SYS_ADMIN))
5424                 return -EPERM;
5425
5426         mutex_lock(&bpf_stats_enabled_mutex);
5427         val = saved_val;
5428         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5429         if (write && !ret && val != saved_val) {
5430                 if (val)
5431                         static_key_slow_inc(key);
5432                 else
5433                         static_key_slow_dec(key);
5434                 saved_val = val;
5435         }
5436         mutex_unlock(&bpf_stats_enabled_mutex);
5437         return ret;
5438 }
5439
5440 void __weak unpriv_ebpf_notify(int new_state)
5441 {
5442 }
5443
5444 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5445                               void *buffer, size_t *lenp, loff_t *ppos)
5446 {
5447         int ret, unpriv_enable = *(int *)table->data;
5448         bool locked_state = unpriv_enable == 1;
5449         struct ctl_table tmp = *table;
5450
5451         if (write && !capable(CAP_SYS_ADMIN))
5452                 return -EPERM;
5453
5454         tmp.data = &unpriv_enable;
5455         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5456         if (write && !ret) {
5457                 if (locked_state && unpriv_enable != 1)
5458                         return -EPERM;
5459                 *(int *)table->data = unpriv_enable;
5460         }
5461
5462         if (write)
5463                 unpriv_ebpf_notify(unpriv_enable);
5464
5465         return ret;
5466 }
5467
5468 static struct ctl_table bpf_syscall_table[] = {
5469         {
5470                 .procname       = "unprivileged_bpf_disabled",
5471                 .data           = &sysctl_unprivileged_bpf_disabled,
5472                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5473                 .mode           = 0644,
5474                 .proc_handler   = bpf_unpriv_handler,
5475                 .extra1         = SYSCTL_ZERO,
5476                 .extra2         = SYSCTL_TWO,
5477         },
5478         {
5479                 .procname       = "bpf_stats_enabled",
5480                 .data           = &bpf_stats_enabled_key.key,
5481                 .mode           = 0644,
5482                 .proc_handler   = bpf_stats_handler,
5483         },
5484         { }
5485 };
5486
5487 static int __init bpf_syscall_sysctl_init(void)
5488 {
5489         register_sysctl_init("kernel", bpf_syscall_table);
5490         return 0;
5491 }
5492 late_initcall(bpf_syscall_sysctl_init);
5493 #endif /* CONFIG_SYSCTL */