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