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_LRU_HASH &&
1063                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1064                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) {
1065                                         ret = -EOPNOTSUPP;
1066                                         goto free_map_tab;
1067                                 }
1068                                 break;
1069                         case BPF_LIST_HEAD:
1070                         case BPF_RB_ROOT:
1071                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1072                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1073                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1074                                         ret = -EOPNOTSUPP;
1075                                         goto free_map_tab;
1076                                 }
1077                                 break;
1078                         default:
1079                                 /* Fail if map_type checks are missing for a field type */
1080                                 ret = -EOPNOTSUPP;
1081                                 goto free_map_tab;
1082                         }
1083                 }
1084         }
1085
1086         ret = btf_check_and_fixup_fields(btf, map->record);
1087         if (ret < 0)
1088                 goto free_map_tab;
1089
1090         if (map->ops->map_check_btf) {
1091                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1092                 if (ret < 0)
1093                         goto free_map_tab;
1094         }
1095
1096         return ret;
1097 free_map_tab:
1098         bpf_map_free_record(map);
1099         return ret;
1100 }
1101
1102 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1103 /* called via syscall */
1104 static int map_create(union bpf_attr *attr)
1105 {
1106         int numa_node = bpf_map_attr_numa_node(attr);
1107         struct btf_field_offs *foffs;
1108         struct bpf_map *map;
1109         int f_flags;
1110         int err;
1111
1112         err = CHECK_ATTR(BPF_MAP_CREATE);
1113         if (err)
1114                 return -EINVAL;
1115
1116         if (attr->btf_vmlinux_value_type_id) {
1117                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1118                     attr->btf_key_type_id || attr->btf_value_type_id)
1119                         return -EINVAL;
1120         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1121                 return -EINVAL;
1122         }
1123
1124         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1125             attr->map_extra != 0)
1126                 return -EINVAL;
1127
1128         f_flags = bpf_get_file_flag(attr->map_flags);
1129         if (f_flags < 0)
1130                 return f_flags;
1131
1132         if (numa_node != NUMA_NO_NODE &&
1133             ((unsigned int)numa_node >= nr_node_ids ||
1134              !node_online(numa_node)))
1135                 return -EINVAL;
1136
1137         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1138         map = find_and_alloc_map(attr);
1139         if (IS_ERR(map))
1140                 return PTR_ERR(map);
1141
1142         err = bpf_obj_name_cpy(map->name, attr->map_name,
1143                                sizeof(attr->map_name));
1144         if (err < 0)
1145                 goto free_map;
1146
1147         atomic64_set(&map->refcnt, 1);
1148         atomic64_set(&map->usercnt, 1);
1149         mutex_init(&map->freeze_mutex);
1150         spin_lock_init(&map->owner.lock);
1151
1152         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1153             /* Even the map's value is a kernel's struct,
1154              * the bpf_prog.o must have BTF to begin with
1155              * to figure out the corresponding kernel's
1156              * counter part.  Thus, attr->btf_fd has
1157              * to be valid also.
1158              */
1159             attr->btf_vmlinux_value_type_id) {
1160                 struct btf *btf;
1161
1162                 btf = btf_get_by_fd(attr->btf_fd);
1163                 if (IS_ERR(btf)) {
1164                         err = PTR_ERR(btf);
1165                         goto free_map;
1166                 }
1167                 if (btf_is_kernel(btf)) {
1168                         btf_put(btf);
1169                         err = -EACCES;
1170                         goto free_map;
1171                 }
1172                 map->btf = btf;
1173
1174                 if (attr->btf_value_type_id) {
1175                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1176                                             attr->btf_value_type_id);
1177                         if (err)
1178                                 goto free_map;
1179                 }
1180
1181                 map->btf_key_type_id = attr->btf_key_type_id;
1182                 map->btf_value_type_id = attr->btf_value_type_id;
1183                 map->btf_vmlinux_value_type_id =
1184                         attr->btf_vmlinux_value_type_id;
1185         }
1186
1187
1188         foffs = btf_parse_field_offs(map->record);
1189         if (IS_ERR(foffs)) {
1190                 err = PTR_ERR(foffs);
1191                 goto free_map;
1192         }
1193         map->field_offs = foffs;
1194
1195         err = security_bpf_map_alloc(map);
1196         if (err)
1197                 goto free_map_field_offs;
1198
1199         err = bpf_map_alloc_id(map);
1200         if (err)
1201                 goto free_map_sec;
1202
1203         bpf_map_save_memcg(map);
1204
1205         err = bpf_map_new_fd(map, f_flags);
1206         if (err < 0) {
1207                 /* failed to allocate fd.
1208                  * bpf_map_put_with_uref() is needed because the above
1209                  * bpf_map_alloc_id() has published the map
1210                  * to the userspace and the userspace may
1211                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1212                  */
1213                 bpf_map_put_with_uref(map);
1214                 return err;
1215         }
1216
1217         return err;
1218
1219 free_map_sec:
1220         security_bpf_map_free(map);
1221 free_map_field_offs:
1222         kfree(map->field_offs);
1223 free_map:
1224         btf_put(map->btf);
1225         map->ops->map_free(map);
1226         return err;
1227 }
1228
1229 /* if error is returned, fd is released.
1230  * On success caller should complete fd access with matching fdput()
1231  */
1232 struct bpf_map *__bpf_map_get(struct fd f)
1233 {
1234         if (!f.file)
1235                 return ERR_PTR(-EBADF);
1236         if (f.file->f_op != &bpf_map_fops) {
1237                 fdput(f);
1238                 return ERR_PTR(-EINVAL);
1239         }
1240
1241         return f.file->private_data;
1242 }
1243
1244 void bpf_map_inc(struct bpf_map *map)
1245 {
1246         atomic64_inc(&map->refcnt);
1247 }
1248 EXPORT_SYMBOL_GPL(bpf_map_inc);
1249
1250 void bpf_map_inc_with_uref(struct bpf_map *map)
1251 {
1252         atomic64_inc(&map->refcnt);
1253         atomic64_inc(&map->usercnt);
1254 }
1255 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1256
1257 struct bpf_map *bpf_map_get(u32 ufd)
1258 {
1259         struct fd f = fdget(ufd);
1260         struct bpf_map *map;
1261
1262         map = __bpf_map_get(f);
1263         if (IS_ERR(map))
1264                 return map;
1265
1266         bpf_map_inc(map);
1267         fdput(f);
1268
1269         return map;
1270 }
1271 EXPORT_SYMBOL(bpf_map_get);
1272
1273 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1274 {
1275         struct fd f = fdget(ufd);
1276         struct bpf_map *map;
1277
1278         map = __bpf_map_get(f);
1279         if (IS_ERR(map))
1280                 return map;
1281
1282         bpf_map_inc_with_uref(map);
1283         fdput(f);
1284
1285         return map;
1286 }
1287
1288 /* map_idr_lock should have been held */
1289 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1290 {
1291         int refold;
1292
1293         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1294         if (!refold)
1295                 return ERR_PTR(-ENOENT);
1296         if (uref)
1297                 atomic64_inc(&map->usercnt);
1298
1299         return map;
1300 }
1301
1302 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1303 {
1304         spin_lock_bh(&map_idr_lock);
1305         map = __bpf_map_inc_not_zero(map, false);
1306         spin_unlock_bh(&map_idr_lock);
1307
1308         return map;
1309 }
1310 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1311
1312 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1313 {
1314         return -ENOTSUPP;
1315 }
1316
1317 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1318 {
1319         if (key_size)
1320                 return vmemdup_user(ukey, key_size);
1321
1322         if (ukey)
1323                 return ERR_PTR(-EINVAL);
1324
1325         return NULL;
1326 }
1327
1328 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1329 {
1330         if (key_size)
1331                 return kvmemdup_bpfptr(ukey, key_size);
1332
1333         if (!bpfptr_is_null(ukey))
1334                 return ERR_PTR(-EINVAL);
1335
1336         return NULL;
1337 }
1338
1339 /* last field in 'union bpf_attr' used by this command */
1340 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1341
1342 static int map_lookup_elem(union bpf_attr *attr)
1343 {
1344         void __user *ukey = u64_to_user_ptr(attr->key);
1345         void __user *uvalue = u64_to_user_ptr(attr->value);
1346         int ufd = attr->map_fd;
1347         struct bpf_map *map;
1348         void *key, *value;
1349         u32 value_size;
1350         struct fd f;
1351         int err;
1352
1353         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1354                 return -EINVAL;
1355
1356         if (attr->flags & ~BPF_F_LOCK)
1357                 return -EINVAL;
1358
1359         f = fdget(ufd);
1360         map = __bpf_map_get(f);
1361         if (IS_ERR(map))
1362                 return PTR_ERR(map);
1363         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1364                 err = -EPERM;
1365                 goto err_put;
1366         }
1367
1368         if ((attr->flags & BPF_F_LOCK) &&
1369             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1370                 err = -EINVAL;
1371                 goto err_put;
1372         }
1373
1374         key = __bpf_copy_key(ukey, map->key_size);
1375         if (IS_ERR(key)) {
1376                 err = PTR_ERR(key);
1377                 goto err_put;
1378         }
1379
1380         value_size = bpf_map_value_size(map);
1381
1382         err = -ENOMEM;
1383         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1384         if (!value)
1385                 goto free_key;
1386
1387         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1388                 if (copy_from_user(value, uvalue, value_size))
1389                         err = -EFAULT;
1390                 else
1391                         err = bpf_map_copy_value(map, key, value, attr->flags);
1392                 goto free_value;
1393         }
1394
1395         err = bpf_map_copy_value(map, key, value, attr->flags);
1396         if (err)
1397                 goto free_value;
1398
1399         err = -EFAULT;
1400         if (copy_to_user(uvalue, value, value_size) != 0)
1401                 goto free_value;
1402
1403         err = 0;
1404
1405 free_value:
1406         kvfree(value);
1407 free_key:
1408         kvfree(key);
1409 err_put:
1410         fdput(f);
1411         return err;
1412 }
1413
1414
1415 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1416
1417 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1418 {
1419         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1420         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1421         int ufd = attr->map_fd;
1422         struct bpf_map *map;
1423         void *key, *value;
1424         u32 value_size;
1425         struct fd f;
1426         int err;
1427
1428         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1429                 return -EINVAL;
1430
1431         f = fdget(ufd);
1432         map = __bpf_map_get(f);
1433         if (IS_ERR(map))
1434                 return PTR_ERR(map);
1435         bpf_map_write_active_inc(map);
1436         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1437                 err = -EPERM;
1438                 goto err_put;
1439         }
1440
1441         if ((attr->flags & BPF_F_LOCK) &&
1442             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1443                 err = -EINVAL;
1444                 goto err_put;
1445         }
1446
1447         key = ___bpf_copy_key(ukey, map->key_size);
1448         if (IS_ERR(key)) {
1449                 err = PTR_ERR(key);
1450                 goto err_put;
1451         }
1452
1453         value_size = bpf_map_value_size(map);
1454         value = kvmemdup_bpfptr(uvalue, value_size);
1455         if (IS_ERR(value)) {
1456                 err = PTR_ERR(value);
1457                 goto free_key;
1458         }
1459
1460         err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1461
1462         kvfree(value);
1463 free_key:
1464         kvfree(key);
1465 err_put:
1466         bpf_map_write_active_dec(map);
1467         fdput(f);
1468         return err;
1469 }
1470
1471 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1472
1473 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1474 {
1475         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1476         int ufd = attr->map_fd;
1477         struct bpf_map *map;
1478         struct fd f;
1479         void *key;
1480         int err;
1481
1482         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1483                 return -EINVAL;
1484
1485         f = fdget(ufd);
1486         map = __bpf_map_get(f);
1487         if (IS_ERR(map))
1488                 return PTR_ERR(map);
1489         bpf_map_write_active_inc(map);
1490         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1491                 err = -EPERM;
1492                 goto err_put;
1493         }
1494
1495         key = ___bpf_copy_key(ukey, map->key_size);
1496         if (IS_ERR(key)) {
1497                 err = PTR_ERR(key);
1498                 goto err_put;
1499         }
1500
1501         if (bpf_map_is_offloaded(map)) {
1502                 err = bpf_map_offload_delete_elem(map, key);
1503                 goto out;
1504         } else if (IS_FD_PROG_ARRAY(map) ||
1505                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1506                 /* These maps require sleepable context */
1507                 err = map->ops->map_delete_elem(map, key);
1508                 goto out;
1509         }
1510
1511         bpf_disable_instrumentation();
1512         rcu_read_lock();
1513         err = map->ops->map_delete_elem(map, key);
1514         rcu_read_unlock();
1515         bpf_enable_instrumentation();
1516         maybe_wait_bpf_programs(map);
1517 out:
1518         kvfree(key);
1519 err_put:
1520         bpf_map_write_active_dec(map);
1521         fdput(f);
1522         return err;
1523 }
1524
1525 /* last field in 'union bpf_attr' used by this command */
1526 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1527
1528 static int map_get_next_key(union bpf_attr *attr)
1529 {
1530         void __user *ukey = u64_to_user_ptr(attr->key);
1531         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1532         int ufd = attr->map_fd;
1533         struct bpf_map *map;
1534         void *key, *next_key;
1535         struct fd f;
1536         int err;
1537
1538         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1539                 return -EINVAL;
1540
1541         f = fdget(ufd);
1542         map = __bpf_map_get(f);
1543         if (IS_ERR(map))
1544                 return PTR_ERR(map);
1545         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1546                 err = -EPERM;
1547                 goto err_put;
1548         }
1549
1550         if (ukey) {
1551                 key = __bpf_copy_key(ukey, map->key_size);
1552                 if (IS_ERR(key)) {
1553                         err = PTR_ERR(key);
1554                         goto err_put;
1555                 }
1556         } else {
1557                 key = NULL;
1558         }
1559
1560         err = -ENOMEM;
1561         next_key = kvmalloc(map->key_size, GFP_USER);
1562         if (!next_key)
1563                 goto free_key;
1564
1565         if (bpf_map_is_offloaded(map)) {
1566                 err = bpf_map_offload_get_next_key(map, key, next_key);
1567                 goto out;
1568         }
1569
1570         rcu_read_lock();
1571         err = map->ops->map_get_next_key(map, key, next_key);
1572         rcu_read_unlock();
1573 out:
1574         if (err)
1575                 goto free_next_key;
1576
1577         err = -EFAULT;
1578         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1579                 goto free_next_key;
1580
1581         err = 0;
1582
1583 free_next_key:
1584         kvfree(next_key);
1585 free_key:
1586         kvfree(key);
1587 err_put:
1588         fdput(f);
1589         return err;
1590 }
1591
1592 int generic_map_delete_batch(struct bpf_map *map,
1593                              const union bpf_attr *attr,
1594                              union bpf_attr __user *uattr)
1595 {
1596         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1597         u32 cp, max_count;
1598         int err = 0;
1599         void *key;
1600
1601         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1602                 return -EINVAL;
1603
1604         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1605             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1606                 return -EINVAL;
1607         }
1608
1609         max_count = attr->batch.count;
1610         if (!max_count)
1611                 return 0;
1612
1613         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1614         if (!key)
1615                 return -ENOMEM;
1616
1617         for (cp = 0; cp < max_count; cp++) {
1618                 err = -EFAULT;
1619                 if (copy_from_user(key, keys + cp * map->key_size,
1620                                    map->key_size))
1621                         break;
1622
1623                 if (bpf_map_is_offloaded(map)) {
1624                         err = bpf_map_offload_delete_elem(map, key);
1625                         break;
1626                 }
1627
1628                 bpf_disable_instrumentation();
1629                 rcu_read_lock();
1630                 err = map->ops->map_delete_elem(map, key);
1631                 rcu_read_unlock();
1632                 bpf_enable_instrumentation();
1633                 if (err)
1634                         break;
1635                 cond_resched();
1636         }
1637         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1638                 err = -EFAULT;
1639
1640         kvfree(key);
1641
1642         maybe_wait_bpf_programs(map);
1643         return err;
1644 }
1645
1646 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1647                              const union bpf_attr *attr,
1648                              union bpf_attr __user *uattr)
1649 {
1650         void __user *values = u64_to_user_ptr(attr->batch.values);
1651         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1652         u32 value_size, cp, max_count;
1653         void *key, *value;
1654         int err = 0;
1655
1656         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1657                 return -EINVAL;
1658
1659         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1660             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1661                 return -EINVAL;
1662         }
1663
1664         value_size = bpf_map_value_size(map);
1665
1666         max_count = attr->batch.count;
1667         if (!max_count)
1668                 return 0;
1669
1670         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1671         if (!key)
1672                 return -ENOMEM;
1673
1674         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1675         if (!value) {
1676                 kvfree(key);
1677                 return -ENOMEM;
1678         }
1679
1680         for (cp = 0; cp < max_count; cp++) {
1681                 err = -EFAULT;
1682                 if (copy_from_user(key, keys + cp * map->key_size,
1683                     map->key_size) ||
1684                     copy_from_user(value, values + cp * value_size, value_size))
1685                         break;
1686
1687                 err = bpf_map_update_value(map, map_file, key, value,
1688                                            attr->batch.elem_flags);
1689
1690                 if (err)
1691                         break;
1692                 cond_resched();
1693         }
1694
1695         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1696                 err = -EFAULT;
1697
1698         kvfree(value);
1699         kvfree(key);
1700         return err;
1701 }
1702
1703 #define MAP_LOOKUP_RETRIES 3
1704
1705 int generic_map_lookup_batch(struct bpf_map *map,
1706                                     const union bpf_attr *attr,
1707                                     union bpf_attr __user *uattr)
1708 {
1709         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1710         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1711         void __user *values = u64_to_user_ptr(attr->batch.values);
1712         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1713         void *buf, *buf_prevkey, *prev_key, *key, *value;
1714         int err, retry = MAP_LOOKUP_RETRIES;
1715         u32 value_size, cp, max_count;
1716
1717         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1718                 return -EINVAL;
1719
1720         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1721             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1722                 return -EINVAL;
1723
1724         value_size = bpf_map_value_size(map);
1725
1726         max_count = attr->batch.count;
1727         if (!max_count)
1728                 return 0;
1729
1730         if (put_user(0, &uattr->batch.count))
1731                 return -EFAULT;
1732
1733         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1734         if (!buf_prevkey)
1735                 return -ENOMEM;
1736
1737         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1738         if (!buf) {
1739                 kvfree(buf_prevkey);
1740                 return -ENOMEM;
1741         }
1742
1743         err = -EFAULT;
1744         prev_key = NULL;
1745         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1746                 goto free_buf;
1747         key = buf;
1748         value = key + map->key_size;
1749         if (ubatch)
1750                 prev_key = buf_prevkey;
1751
1752         for (cp = 0; cp < max_count;) {
1753                 rcu_read_lock();
1754                 err = map->ops->map_get_next_key(map, prev_key, key);
1755                 rcu_read_unlock();
1756                 if (err)
1757                         break;
1758                 err = bpf_map_copy_value(map, key, value,
1759                                          attr->batch.elem_flags);
1760
1761                 if (err == -ENOENT) {
1762                         if (retry) {
1763                                 retry--;
1764                                 continue;
1765                         }
1766                         err = -EINTR;
1767                         break;
1768                 }
1769
1770                 if (err)
1771                         goto free_buf;
1772
1773                 if (copy_to_user(keys + cp * map->key_size, key,
1774                                  map->key_size)) {
1775                         err = -EFAULT;
1776                         goto free_buf;
1777                 }
1778                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1779                         err = -EFAULT;
1780                         goto free_buf;
1781                 }
1782
1783                 if (!prev_key)
1784                         prev_key = buf_prevkey;
1785
1786                 swap(prev_key, key);
1787                 retry = MAP_LOOKUP_RETRIES;
1788                 cp++;
1789                 cond_resched();
1790         }
1791
1792         if (err == -EFAULT)
1793                 goto free_buf;
1794
1795         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1796                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1797                 err = -EFAULT;
1798
1799 free_buf:
1800         kvfree(buf_prevkey);
1801         kvfree(buf);
1802         return err;
1803 }
1804
1805 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1806
1807 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1808 {
1809         void __user *ukey = u64_to_user_ptr(attr->key);
1810         void __user *uvalue = u64_to_user_ptr(attr->value);
1811         int ufd = attr->map_fd;
1812         struct bpf_map *map;
1813         void *key, *value;
1814         u32 value_size;
1815         struct fd f;
1816         int err;
1817
1818         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1819                 return -EINVAL;
1820
1821         if (attr->flags & ~BPF_F_LOCK)
1822                 return -EINVAL;
1823
1824         f = fdget(ufd);
1825         map = __bpf_map_get(f);
1826         if (IS_ERR(map))
1827                 return PTR_ERR(map);
1828         bpf_map_write_active_inc(map);
1829         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1830             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1831                 err = -EPERM;
1832                 goto err_put;
1833         }
1834
1835         if (attr->flags &&
1836             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1837              map->map_type == BPF_MAP_TYPE_STACK)) {
1838                 err = -EINVAL;
1839                 goto err_put;
1840         }
1841
1842         if ((attr->flags & BPF_F_LOCK) &&
1843             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1844                 err = -EINVAL;
1845                 goto err_put;
1846         }
1847
1848         key = __bpf_copy_key(ukey, map->key_size);
1849         if (IS_ERR(key)) {
1850                 err = PTR_ERR(key);
1851                 goto err_put;
1852         }
1853
1854         value_size = bpf_map_value_size(map);
1855
1856         err = -ENOMEM;
1857         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1858         if (!value)
1859                 goto free_key;
1860
1861         err = -ENOTSUPP;
1862         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1863             map->map_type == BPF_MAP_TYPE_STACK) {
1864                 err = map->ops->map_pop_elem(map, value);
1865         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1866                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1867                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1868                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1869                 if (!bpf_map_is_offloaded(map)) {
1870                         bpf_disable_instrumentation();
1871                         rcu_read_lock();
1872                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1873                         rcu_read_unlock();
1874                         bpf_enable_instrumentation();
1875                 }
1876         }
1877
1878         if (err)
1879                 goto free_value;
1880
1881         if (copy_to_user(uvalue, value, value_size) != 0) {
1882                 err = -EFAULT;
1883                 goto free_value;
1884         }
1885
1886         err = 0;
1887
1888 free_value:
1889         kvfree(value);
1890 free_key:
1891         kvfree(key);
1892 err_put:
1893         bpf_map_write_active_dec(map);
1894         fdput(f);
1895         return err;
1896 }
1897
1898 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1899
1900 static int map_freeze(const union bpf_attr *attr)
1901 {
1902         int err = 0, ufd = attr->map_fd;
1903         struct bpf_map *map;
1904         struct fd f;
1905
1906         if (CHECK_ATTR(BPF_MAP_FREEZE))
1907                 return -EINVAL;
1908
1909         f = fdget(ufd);
1910         map = __bpf_map_get(f);
1911         if (IS_ERR(map))
1912                 return PTR_ERR(map);
1913
1914         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1915                 fdput(f);
1916                 return -ENOTSUPP;
1917         }
1918
1919         mutex_lock(&map->freeze_mutex);
1920         if (bpf_map_write_active(map)) {
1921                 err = -EBUSY;
1922                 goto err_put;
1923         }
1924         if (READ_ONCE(map->frozen)) {
1925                 err = -EBUSY;
1926                 goto err_put;
1927         }
1928         if (!bpf_capable()) {
1929                 err = -EPERM;
1930                 goto err_put;
1931         }
1932
1933         WRITE_ONCE(map->frozen, true);
1934 err_put:
1935         mutex_unlock(&map->freeze_mutex);
1936         fdput(f);
1937         return err;
1938 }
1939
1940 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1941 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1942         [_id] = & _name ## _prog_ops,
1943 #define BPF_MAP_TYPE(_id, _ops)
1944 #define BPF_LINK_TYPE(_id, _name)
1945 #include <linux/bpf_types.h>
1946 #undef BPF_PROG_TYPE
1947 #undef BPF_MAP_TYPE
1948 #undef BPF_LINK_TYPE
1949 };
1950
1951 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1952 {
1953         const struct bpf_prog_ops *ops;
1954
1955         if (type >= ARRAY_SIZE(bpf_prog_types))
1956                 return -EINVAL;
1957         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1958         ops = bpf_prog_types[type];
1959         if (!ops)
1960                 return -EINVAL;
1961
1962         if (!bpf_prog_is_offloaded(prog->aux))
1963                 prog->aux->ops = ops;
1964         else
1965                 prog->aux->ops = &bpf_offload_prog_ops;
1966         prog->type = type;
1967         return 0;
1968 }
1969
1970 enum bpf_audit {
1971         BPF_AUDIT_LOAD,
1972         BPF_AUDIT_UNLOAD,
1973         BPF_AUDIT_MAX,
1974 };
1975
1976 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1977         [BPF_AUDIT_LOAD]   = "LOAD",
1978         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1979 };
1980
1981 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1982 {
1983         struct audit_context *ctx = NULL;
1984         struct audit_buffer *ab;
1985
1986         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1987                 return;
1988         if (audit_enabled == AUDIT_OFF)
1989                 return;
1990         if (!in_irq() && !irqs_disabled())
1991                 ctx = audit_context();
1992         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1993         if (unlikely(!ab))
1994                 return;
1995         audit_log_format(ab, "prog-id=%u op=%s",
1996                          prog->aux->id, bpf_audit_str[op]);
1997         audit_log_end(ab);
1998 }
1999
2000 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2001 {
2002         int id;
2003
2004         idr_preload(GFP_KERNEL);
2005         spin_lock_bh(&prog_idr_lock);
2006         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2007         if (id > 0)
2008                 prog->aux->id = id;
2009         spin_unlock_bh(&prog_idr_lock);
2010         idr_preload_end();
2011
2012         /* id is in [1, INT_MAX) */
2013         if (WARN_ON_ONCE(!id))
2014                 return -ENOSPC;
2015
2016         return id > 0 ? 0 : id;
2017 }
2018
2019 void bpf_prog_free_id(struct bpf_prog *prog)
2020 {
2021         unsigned long flags;
2022
2023         /* cBPF to eBPF migrations are currently not in the idr store.
2024          * Offloaded programs are removed from the store when their device
2025          * disappears - even if someone grabs an fd to them they are unusable,
2026          * simply waiting for refcnt to drop to be freed.
2027          */
2028         if (!prog->aux->id)
2029                 return;
2030
2031         spin_lock_irqsave(&prog_idr_lock, flags);
2032         idr_remove(&prog_idr, prog->aux->id);
2033         prog->aux->id = 0;
2034         spin_unlock_irqrestore(&prog_idr_lock, flags);
2035 }
2036
2037 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2038 {
2039         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2040
2041         kvfree(aux->func_info);
2042         kfree(aux->func_info_aux);
2043         free_uid(aux->user);
2044         security_bpf_prog_free(aux);
2045         bpf_prog_free(aux->prog);
2046 }
2047
2048 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2049 {
2050         bpf_prog_kallsyms_del_all(prog);
2051         btf_put(prog->aux->btf);
2052         kvfree(prog->aux->jited_linfo);
2053         kvfree(prog->aux->linfo);
2054         kfree(prog->aux->kfunc_tab);
2055         if (prog->aux->attach_btf)
2056                 btf_put(prog->aux->attach_btf);
2057
2058         if (deferred) {
2059                 if (prog->aux->sleepable)
2060                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2061                 else
2062                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2063         } else {
2064                 __bpf_prog_put_rcu(&prog->aux->rcu);
2065         }
2066 }
2067
2068 static void bpf_prog_put_deferred(struct work_struct *work)
2069 {
2070         struct bpf_prog_aux *aux;
2071         struct bpf_prog *prog;
2072
2073         aux = container_of(work, struct bpf_prog_aux, work);
2074         prog = aux->prog;
2075         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2076         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2077         bpf_prog_free_id(prog);
2078         __bpf_prog_put_noref(prog, true);
2079 }
2080
2081 static void __bpf_prog_put(struct bpf_prog *prog)
2082 {
2083         struct bpf_prog_aux *aux = prog->aux;
2084
2085         if (atomic64_dec_and_test(&aux->refcnt)) {
2086                 if (in_irq() || irqs_disabled()) {
2087                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2088                         schedule_work(&aux->work);
2089                 } else {
2090                         bpf_prog_put_deferred(&aux->work);
2091                 }
2092         }
2093 }
2094
2095 void bpf_prog_put(struct bpf_prog *prog)
2096 {
2097         __bpf_prog_put(prog);
2098 }
2099 EXPORT_SYMBOL_GPL(bpf_prog_put);
2100
2101 static int bpf_prog_release(struct inode *inode, struct file *filp)
2102 {
2103         struct bpf_prog *prog = filp->private_data;
2104
2105         bpf_prog_put(prog);
2106         return 0;
2107 }
2108
2109 struct bpf_prog_kstats {
2110         u64 nsecs;
2111         u64 cnt;
2112         u64 misses;
2113 };
2114
2115 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2116 {
2117         struct bpf_prog_stats *stats;
2118         unsigned int flags;
2119
2120         stats = this_cpu_ptr(prog->stats);
2121         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2122         u64_stats_inc(&stats->misses);
2123         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2124 }
2125
2126 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2127                                struct bpf_prog_kstats *stats)
2128 {
2129         u64 nsecs = 0, cnt = 0, misses = 0;
2130         int cpu;
2131
2132         for_each_possible_cpu(cpu) {
2133                 const struct bpf_prog_stats *st;
2134                 unsigned int start;
2135                 u64 tnsecs, tcnt, tmisses;
2136
2137                 st = per_cpu_ptr(prog->stats, cpu);
2138                 do {
2139                         start = u64_stats_fetch_begin(&st->syncp);
2140                         tnsecs = u64_stats_read(&st->nsecs);
2141                         tcnt = u64_stats_read(&st->cnt);
2142                         tmisses = u64_stats_read(&st->misses);
2143                 } while (u64_stats_fetch_retry(&st->syncp, start));
2144                 nsecs += tnsecs;
2145                 cnt += tcnt;
2146                 misses += tmisses;
2147         }
2148         stats->nsecs = nsecs;
2149         stats->cnt = cnt;
2150         stats->misses = misses;
2151 }
2152
2153 #ifdef CONFIG_PROC_FS
2154 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2155 {
2156         const struct bpf_prog *prog = filp->private_data;
2157         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2158         struct bpf_prog_kstats stats;
2159
2160         bpf_prog_get_stats(prog, &stats);
2161         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2162         seq_printf(m,
2163                    "prog_type:\t%u\n"
2164                    "prog_jited:\t%u\n"
2165                    "prog_tag:\t%s\n"
2166                    "memlock:\t%llu\n"
2167                    "prog_id:\t%u\n"
2168                    "run_time_ns:\t%llu\n"
2169                    "run_cnt:\t%llu\n"
2170                    "recursion_misses:\t%llu\n"
2171                    "verified_insns:\t%u\n",
2172                    prog->type,
2173                    prog->jited,
2174                    prog_tag,
2175                    prog->pages * 1ULL << PAGE_SHIFT,
2176                    prog->aux->id,
2177                    stats.nsecs,
2178                    stats.cnt,
2179                    stats.misses,
2180                    prog->aux->verified_insns);
2181 }
2182 #endif
2183
2184 const struct file_operations bpf_prog_fops = {
2185 #ifdef CONFIG_PROC_FS
2186         .show_fdinfo    = bpf_prog_show_fdinfo,
2187 #endif
2188         .release        = bpf_prog_release,
2189         .read           = bpf_dummy_read,
2190         .write          = bpf_dummy_write,
2191 };
2192
2193 int bpf_prog_new_fd(struct bpf_prog *prog)
2194 {
2195         int ret;
2196
2197         ret = security_bpf_prog(prog);
2198         if (ret < 0)
2199                 return ret;
2200
2201         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2202                                 O_RDWR | O_CLOEXEC);
2203 }
2204
2205 static struct bpf_prog *____bpf_prog_get(struct fd f)
2206 {
2207         if (!f.file)
2208                 return ERR_PTR(-EBADF);
2209         if (f.file->f_op != &bpf_prog_fops) {
2210                 fdput(f);
2211                 return ERR_PTR(-EINVAL);
2212         }
2213
2214         return f.file->private_data;
2215 }
2216
2217 void bpf_prog_add(struct bpf_prog *prog, int i)
2218 {
2219         atomic64_add(i, &prog->aux->refcnt);
2220 }
2221 EXPORT_SYMBOL_GPL(bpf_prog_add);
2222
2223 void bpf_prog_sub(struct bpf_prog *prog, int i)
2224 {
2225         /* Only to be used for undoing previous bpf_prog_add() in some
2226          * error path. We still know that another entity in our call
2227          * path holds a reference to the program, thus atomic_sub() can
2228          * be safely used in such cases!
2229          */
2230         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2231 }
2232 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2233
2234 void bpf_prog_inc(struct bpf_prog *prog)
2235 {
2236         atomic64_inc(&prog->aux->refcnt);
2237 }
2238 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2239
2240 /* prog_idr_lock should have been held */
2241 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2242 {
2243         int refold;
2244
2245         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2246
2247         if (!refold)
2248                 return ERR_PTR(-ENOENT);
2249
2250         return prog;
2251 }
2252 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2253
2254 bool bpf_prog_get_ok(struct bpf_prog *prog,
2255                             enum bpf_prog_type *attach_type, bool attach_drv)
2256 {
2257         /* not an attachment, just a refcount inc, always allow */
2258         if (!attach_type)
2259                 return true;
2260
2261         if (prog->type != *attach_type)
2262                 return false;
2263         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2264                 return false;
2265
2266         return true;
2267 }
2268
2269 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2270                                        bool attach_drv)
2271 {
2272         struct fd f = fdget(ufd);
2273         struct bpf_prog *prog;
2274
2275         prog = ____bpf_prog_get(f);
2276         if (IS_ERR(prog))
2277                 return prog;
2278         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2279                 prog = ERR_PTR(-EINVAL);
2280                 goto out;
2281         }
2282
2283         bpf_prog_inc(prog);
2284 out:
2285         fdput(f);
2286         return prog;
2287 }
2288
2289 struct bpf_prog *bpf_prog_get(u32 ufd)
2290 {
2291         return __bpf_prog_get(ufd, NULL, false);
2292 }
2293
2294 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2295                                        bool attach_drv)
2296 {
2297         return __bpf_prog_get(ufd, &type, attach_drv);
2298 }
2299 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2300
2301 /* Initially all BPF programs could be loaded w/o specifying
2302  * expected_attach_type. Later for some of them specifying expected_attach_type
2303  * at load time became required so that program could be validated properly.
2304  * Programs of types that are allowed to be loaded both w/ and w/o (for
2305  * backward compatibility) expected_attach_type, should have the default attach
2306  * type assigned to expected_attach_type for the latter case, so that it can be
2307  * validated later at attach time.
2308  *
2309  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2310  * prog type requires it but has some attach types that have to be backward
2311  * compatible.
2312  */
2313 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2314 {
2315         switch (attr->prog_type) {
2316         case BPF_PROG_TYPE_CGROUP_SOCK:
2317                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2318                  * exist so checking for non-zero is the way to go here.
2319                  */
2320                 if (!attr->expected_attach_type)
2321                         attr->expected_attach_type =
2322                                 BPF_CGROUP_INET_SOCK_CREATE;
2323                 break;
2324         case BPF_PROG_TYPE_SK_REUSEPORT:
2325                 if (!attr->expected_attach_type)
2326                         attr->expected_attach_type =
2327                                 BPF_SK_REUSEPORT_SELECT;
2328                 break;
2329         }
2330 }
2331
2332 static int
2333 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2334                            enum bpf_attach_type expected_attach_type,
2335                            struct btf *attach_btf, u32 btf_id,
2336                            struct bpf_prog *dst_prog)
2337 {
2338         if (btf_id) {
2339                 if (btf_id > BTF_MAX_TYPE)
2340                         return -EINVAL;
2341
2342                 if (!attach_btf && !dst_prog)
2343                         return -EINVAL;
2344
2345                 switch (prog_type) {
2346                 case BPF_PROG_TYPE_TRACING:
2347                 case BPF_PROG_TYPE_LSM:
2348                 case BPF_PROG_TYPE_STRUCT_OPS:
2349                 case BPF_PROG_TYPE_EXT:
2350                         break;
2351                 default:
2352                         return -EINVAL;
2353                 }
2354         }
2355
2356         if (attach_btf && (!btf_id || dst_prog))
2357                 return -EINVAL;
2358
2359         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2360             prog_type != BPF_PROG_TYPE_EXT)
2361                 return -EINVAL;
2362
2363         switch (prog_type) {
2364         case BPF_PROG_TYPE_CGROUP_SOCK:
2365                 switch (expected_attach_type) {
2366                 case BPF_CGROUP_INET_SOCK_CREATE:
2367                 case BPF_CGROUP_INET_SOCK_RELEASE:
2368                 case BPF_CGROUP_INET4_POST_BIND:
2369                 case BPF_CGROUP_INET6_POST_BIND:
2370                         return 0;
2371                 default:
2372                         return -EINVAL;
2373                 }
2374         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2375                 switch (expected_attach_type) {
2376                 case BPF_CGROUP_INET4_BIND:
2377                 case BPF_CGROUP_INET6_BIND:
2378                 case BPF_CGROUP_INET4_CONNECT:
2379                 case BPF_CGROUP_INET6_CONNECT:
2380                 case BPF_CGROUP_INET4_GETPEERNAME:
2381                 case BPF_CGROUP_INET6_GETPEERNAME:
2382                 case BPF_CGROUP_INET4_GETSOCKNAME:
2383                 case BPF_CGROUP_INET6_GETSOCKNAME:
2384                 case BPF_CGROUP_UDP4_SENDMSG:
2385                 case BPF_CGROUP_UDP6_SENDMSG:
2386                 case BPF_CGROUP_UDP4_RECVMSG:
2387                 case BPF_CGROUP_UDP6_RECVMSG:
2388                         return 0;
2389                 default:
2390                         return -EINVAL;
2391                 }
2392         case BPF_PROG_TYPE_CGROUP_SKB:
2393                 switch (expected_attach_type) {
2394                 case BPF_CGROUP_INET_INGRESS:
2395                 case BPF_CGROUP_INET_EGRESS:
2396                         return 0;
2397                 default:
2398                         return -EINVAL;
2399                 }
2400         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2401                 switch (expected_attach_type) {
2402                 case BPF_CGROUP_SETSOCKOPT:
2403                 case BPF_CGROUP_GETSOCKOPT:
2404                         return 0;
2405                 default:
2406                         return -EINVAL;
2407                 }
2408         case BPF_PROG_TYPE_SK_LOOKUP:
2409                 if (expected_attach_type == BPF_SK_LOOKUP)
2410                         return 0;
2411                 return -EINVAL;
2412         case BPF_PROG_TYPE_SK_REUSEPORT:
2413                 switch (expected_attach_type) {
2414                 case BPF_SK_REUSEPORT_SELECT:
2415                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2416                         return 0;
2417                 default:
2418                         return -EINVAL;
2419                 }
2420         case BPF_PROG_TYPE_SYSCALL:
2421         case BPF_PROG_TYPE_EXT:
2422                 if (expected_attach_type)
2423                         return -EINVAL;
2424                 fallthrough;
2425         default:
2426                 return 0;
2427         }
2428 }
2429
2430 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2431 {
2432         switch (prog_type) {
2433         case BPF_PROG_TYPE_SCHED_CLS:
2434         case BPF_PROG_TYPE_SCHED_ACT:
2435         case BPF_PROG_TYPE_XDP:
2436         case BPF_PROG_TYPE_LWT_IN:
2437         case BPF_PROG_TYPE_LWT_OUT:
2438         case BPF_PROG_TYPE_LWT_XMIT:
2439         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2440         case BPF_PROG_TYPE_SK_SKB:
2441         case BPF_PROG_TYPE_SK_MSG:
2442         case BPF_PROG_TYPE_LIRC_MODE2:
2443         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2444         case BPF_PROG_TYPE_CGROUP_DEVICE:
2445         case BPF_PROG_TYPE_CGROUP_SOCK:
2446         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2447         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2448         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2449         case BPF_PROG_TYPE_SOCK_OPS:
2450         case BPF_PROG_TYPE_EXT: /* extends any prog */
2451                 return true;
2452         case BPF_PROG_TYPE_CGROUP_SKB:
2453                 /* always unpriv */
2454         case BPF_PROG_TYPE_SK_REUSEPORT:
2455                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2456         default:
2457                 return false;
2458         }
2459 }
2460
2461 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2462 {
2463         switch (prog_type) {
2464         case BPF_PROG_TYPE_KPROBE:
2465         case BPF_PROG_TYPE_TRACEPOINT:
2466         case BPF_PROG_TYPE_PERF_EVENT:
2467         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2468         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2469         case BPF_PROG_TYPE_TRACING:
2470         case BPF_PROG_TYPE_LSM:
2471         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2472         case BPF_PROG_TYPE_EXT: /* extends any prog */
2473                 return true;
2474         default:
2475                 return false;
2476         }
2477 }
2478
2479 /* last field in 'union bpf_attr' used by this command */
2480 #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
2481
2482 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2483 {
2484         enum bpf_prog_type type = attr->prog_type;
2485         struct bpf_prog *prog, *dst_prog = NULL;
2486         struct btf *attach_btf = NULL;
2487         int err;
2488         char license[128];
2489         bool is_gpl;
2490
2491         if (CHECK_ATTR(BPF_PROG_LOAD))
2492                 return -EINVAL;
2493
2494         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2495                                  BPF_F_ANY_ALIGNMENT |
2496                                  BPF_F_TEST_STATE_FREQ |
2497                                  BPF_F_SLEEPABLE |
2498                                  BPF_F_TEST_RND_HI32 |
2499                                  BPF_F_XDP_HAS_FRAGS |
2500                                  BPF_F_XDP_DEV_BOUND_ONLY))
2501                 return -EINVAL;
2502
2503         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2504             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2505             !bpf_capable())
2506                 return -EPERM;
2507
2508         /* copy eBPF program license from user space */
2509         if (strncpy_from_bpfptr(license,
2510                                 make_bpfptr(attr->license, uattr.is_kernel),
2511                                 sizeof(license) - 1) < 0)
2512                 return -EFAULT;
2513         license[sizeof(license) - 1] = 0;
2514
2515         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2516         is_gpl = license_is_gpl_compatible(license);
2517
2518         if (attr->insn_cnt == 0 ||
2519             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2520                 return -E2BIG;
2521         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2522             type != BPF_PROG_TYPE_CGROUP_SKB &&
2523             !bpf_capable())
2524                 return -EPERM;
2525
2526         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2527                 return -EPERM;
2528         if (is_perfmon_prog_type(type) && !perfmon_capable())
2529                 return -EPERM;
2530
2531         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2532          * or btf, we need to check which one it is
2533          */
2534         if (attr->attach_prog_fd) {
2535                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2536                 if (IS_ERR(dst_prog)) {
2537                         dst_prog = NULL;
2538                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2539                         if (IS_ERR(attach_btf))
2540                                 return -EINVAL;
2541                         if (!btf_is_kernel(attach_btf)) {
2542                                 /* attaching through specifying bpf_prog's BTF
2543                                  * objects directly might be supported eventually
2544                                  */
2545                                 btf_put(attach_btf);
2546                                 return -ENOTSUPP;
2547                         }
2548                 }
2549         } else if (attr->attach_btf_id) {
2550                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2551                 attach_btf = bpf_get_btf_vmlinux();
2552                 if (IS_ERR(attach_btf))
2553                         return PTR_ERR(attach_btf);
2554                 if (!attach_btf)
2555                         return -EINVAL;
2556                 btf_get(attach_btf);
2557         }
2558
2559         bpf_prog_load_fixup_attach_type(attr);
2560         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2561                                        attach_btf, attr->attach_btf_id,
2562                                        dst_prog)) {
2563                 if (dst_prog)
2564                         bpf_prog_put(dst_prog);
2565                 if (attach_btf)
2566                         btf_put(attach_btf);
2567                 return -EINVAL;
2568         }
2569
2570         /* plain bpf_prog allocation */
2571         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2572         if (!prog) {
2573                 if (dst_prog)
2574                         bpf_prog_put(dst_prog);
2575                 if (attach_btf)
2576                         btf_put(attach_btf);
2577                 return -ENOMEM;
2578         }
2579
2580         prog->expected_attach_type = attr->expected_attach_type;
2581         prog->aux->attach_btf = attach_btf;
2582         prog->aux->attach_btf_id = attr->attach_btf_id;
2583         prog->aux->dst_prog = dst_prog;
2584         prog->aux->dev_bound = !!attr->prog_ifindex;
2585         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2586         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2587
2588         err = security_bpf_prog_alloc(prog->aux);
2589         if (err)
2590                 goto free_prog;
2591
2592         prog->aux->user = get_current_user();
2593         prog->len = attr->insn_cnt;
2594
2595         err = -EFAULT;
2596         if (copy_from_bpfptr(prog->insns,
2597                              make_bpfptr(attr->insns, uattr.is_kernel),
2598                              bpf_prog_insn_size(prog)) != 0)
2599                 goto free_prog_sec;
2600
2601         prog->orig_prog = NULL;
2602         prog->jited = 0;
2603
2604         atomic64_set(&prog->aux->refcnt, 1);
2605         prog->gpl_compatible = is_gpl ? 1 : 0;
2606
2607         if (bpf_prog_is_dev_bound(prog->aux)) {
2608                 err = bpf_prog_dev_bound_init(prog, attr);
2609                 if (err)
2610                         goto free_prog_sec;
2611         }
2612
2613         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2614             bpf_prog_is_dev_bound(dst_prog->aux)) {
2615                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2616                 if (err)
2617                         goto free_prog_sec;
2618         }
2619
2620         /* find program type: socket_filter vs tracing_filter */
2621         err = find_prog_type(type, prog);
2622         if (err < 0)
2623                 goto free_prog_sec;
2624
2625         prog->aux->load_time = ktime_get_boottime_ns();
2626         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2627                                sizeof(attr->prog_name));
2628         if (err < 0)
2629                 goto free_prog_sec;
2630
2631         /* run eBPF verifier */
2632         err = bpf_check(&prog, attr, uattr);
2633         if (err < 0)
2634                 goto free_used_maps;
2635
2636         prog = bpf_prog_select_runtime(prog, &err);
2637         if (err < 0)
2638                 goto free_used_maps;
2639
2640         err = bpf_prog_alloc_id(prog);
2641         if (err)
2642                 goto free_used_maps;
2643
2644         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2645          * effectively publicly exposed. However, retrieving via
2646          * bpf_prog_get_fd_by_id() will take another reference,
2647          * therefore it cannot be gone underneath us.
2648          *
2649          * Only for the time /after/ successful bpf_prog_new_fd()
2650          * and before returning to userspace, we might just hold
2651          * one reference and any parallel close on that fd could
2652          * rip everything out. Hence, below notifications must
2653          * happen before bpf_prog_new_fd().
2654          *
2655          * Also, any failure handling from this point onwards must
2656          * be using bpf_prog_put() given the program is exposed.
2657          */
2658         bpf_prog_kallsyms_add(prog);
2659         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2660         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2661
2662         err = bpf_prog_new_fd(prog);
2663         if (err < 0)
2664                 bpf_prog_put(prog);
2665         return err;
2666
2667 free_used_maps:
2668         /* In case we have subprogs, we need to wait for a grace
2669          * period before we can tear down JIT memory since symbols
2670          * are already exposed under kallsyms.
2671          */
2672         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2673         return err;
2674 free_prog_sec:
2675         free_uid(prog->aux->user);
2676         security_bpf_prog_free(prog->aux);
2677 free_prog:
2678         if (prog->aux->attach_btf)
2679                 btf_put(prog->aux->attach_btf);
2680         bpf_prog_free(prog);
2681         return err;
2682 }
2683
2684 #define BPF_OBJ_LAST_FIELD file_flags
2685
2686 static int bpf_obj_pin(const union bpf_attr *attr)
2687 {
2688         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2689                 return -EINVAL;
2690
2691         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2692 }
2693
2694 static int bpf_obj_get(const union bpf_attr *attr)
2695 {
2696         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2697             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2698                 return -EINVAL;
2699
2700         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2701                                 attr->file_flags);
2702 }
2703
2704 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2705                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2706 {
2707         atomic64_set(&link->refcnt, 1);
2708         link->type = type;
2709         link->id = 0;
2710         link->ops = ops;
2711         link->prog = prog;
2712 }
2713
2714 static void bpf_link_free_id(int id)
2715 {
2716         if (!id)
2717                 return;
2718
2719         spin_lock_bh(&link_idr_lock);
2720         idr_remove(&link_idr, id);
2721         spin_unlock_bh(&link_idr_lock);
2722 }
2723
2724 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2725  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2726  * anon_inode's release() call. This helper marksbpf_link as
2727  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2728  * is not decremented, it's the responsibility of a calling code that failed
2729  * to complete bpf_link initialization.
2730  */
2731 void bpf_link_cleanup(struct bpf_link_primer *primer)
2732 {
2733         primer->link->prog = NULL;
2734         bpf_link_free_id(primer->id);
2735         fput(primer->file);
2736         put_unused_fd(primer->fd);
2737 }
2738
2739 void bpf_link_inc(struct bpf_link *link)
2740 {
2741         atomic64_inc(&link->refcnt);
2742 }
2743
2744 /* bpf_link_free is guaranteed to be called from process context */
2745 static void bpf_link_free(struct bpf_link *link)
2746 {
2747         bpf_link_free_id(link->id);
2748         if (link->prog) {
2749                 /* detach BPF program, clean up used resources */
2750                 link->ops->release(link);
2751                 bpf_prog_put(link->prog);
2752         }
2753         /* free bpf_link and its containing memory */
2754         link->ops->dealloc(link);
2755 }
2756
2757 static void bpf_link_put_deferred(struct work_struct *work)
2758 {
2759         struct bpf_link *link = container_of(work, struct bpf_link, work);
2760
2761         bpf_link_free(link);
2762 }
2763
2764 /* bpf_link_put can be called from atomic context, but ensures that resources
2765  * are freed from process context
2766  */
2767 void bpf_link_put(struct bpf_link *link)
2768 {
2769         if (!atomic64_dec_and_test(&link->refcnt))
2770                 return;
2771
2772         if (in_atomic()) {
2773                 INIT_WORK(&link->work, bpf_link_put_deferred);
2774                 schedule_work(&link->work);
2775         } else {
2776                 bpf_link_free(link);
2777         }
2778 }
2779 EXPORT_SYMBOL(bpf_link_put);
2780
2781 static int bpf_link_release(struct inode *inode, struct file *filp)
2782 {
2783         struct bpf_link *link = filp->private_data;
2784
2785         bpf_link_put(link);
2786         return 0;
2787 }
2788
2789 #ifdef CONFIG_PROC_FS
2790 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2791 #define BPF_MAP_TYPE(_id, _ops)
2792 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2793 static const char *bpf_link_type_strs[] = {
2794         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2795 #include <linux/bpf_types.h>
2796 };
2797 #undef BPF_PROG_TYPE
2798 #undef BPF_MAP_TYPE
2799 #undef BPF_LINK_TYPE
2800
2801 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2802 {
2803         const struct bpf_link *link = filp->private_data;
2804         const struct bpf_prog *prog = link->prog;
2805         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2806
2807         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2808         seq_printf(m,
2809                    "link_type:\t%s\n"
2810                    "link_id:\t%u\n"
2811                    "prog_tag:\t%s\n"
2812                    "prog_id:\t%u\n",
2813                    bpf_link_type_strs[link->type],
2814                    link->id,
2815                    prog_tag,
2816                    prog->aux->id);
2817         if (link->ops->show_fdinfo)
2818                 link->ops->show_fdinfo(link, m);
2819 }
2820 #endif
2821
2822 static const struct file_operations bpf_link_fops = {
2823 #ifdef CONFIG_PROC_FS
2824         .show_fdinfo    = bpf_link_show_fdinfo,
2825 #endif
2826         .release        = bpf_link_release,
2827         .read           = bpf_dummy_read,
2828         .write          = bpf_dummy_write,
2829 };
2830
2831 static int bpf_link_alloc_id(struct bpf_link *link)
2832 {
2833         int id;
2834
2835         idr_preload(GFP_KERNEL);
2836         spin_lock_bh(&link_idr_lock);
2837         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2838         spin_unlock_bh(&link_idr_lock);
2839         idr_preload_end();
2840
2841         return id;
2842 }
2843
2844 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2845  * reserving unused FD and allocating ID from link_idr. This is to be paired
2846  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2847  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2848  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2849  * transient state is passed around in struct bpf_link_primer.
2850  * This is preferred way to create and initialize bpf_link, especially when
2851  * there are complicated and expensive operations in between creating bpf_link
2852  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2853  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2854  * expensive (and potentially failing) roll back operations in a rare case
2855  * that file, FD, or ID can't be allocated.
2856  */
2857 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2858 {
2859         struct file *file;
2860         int fd, id;
2861
2862         fd = get_unused_fd_flags(O_CLOEXEC);
2863         if (fd < 0)
2864                 return fd;
2865
2866
2867         id = bpf_link_alloc_id(link);
2868         if (id < 0) {
2869                 put_unused_fd(fd);
2870                 return id;
2871         }
2872
2873         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2874         if (IS_ERR(file)) {
2875                 bpf_link_free_id(id);
2876                 put_unused_fd(fd);
2877                 return PTR_ERR(file);
2878         }
2879
2880         primer->link = link;
2881         primer->file = file;
2882         primer->fd = fd;
2883         primer->id = id;
2884         return 0;
2885 }
2886
2887 int bpf_link_settle(struct bpf_link_primer *primer)
2888 {
2889         /* make bpf_link fetchable by ID */
2890         spin_lock_bh(&link_idr_lock);
2891         primer->link->id = primer->id;
2892         spin_unlock_bh(&link_idr_lock);
2893         /* make bpf_link fetchable by FD */
2894         fd_install(primer->fd, primer->file);
2895         /* pass through installed FD */
2896         return primer->fd;
2897 }
2898
2899 int bpf_link_new_fd(struct bpf_link *link)
2900 {
2901         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2902 }
2903
2904 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2905 {
2906         struct fd f = fdget(ufd);
2907         struct bpf_link *link;
2908
2909         if (!f.file)
2910                 return ERR_PTR(-EBADF);
2911         if (f.file->f_op != &bpf_link_fops) {
2912                 fdput(f);
2913                 return ERR_PTR(-EINVAL);
2914         }
2915
2916         link = f.file->private_data;
2917         bpf_link_inc(link);
2918         fdput(f);
2919
2920         return link;
2921 }
2922 EXPORT_SYMBOL(bpf_link_get_from_fd);
2923
2924 static void bpf_tracing_link_release(struct bpf_link *link)
2925 {
2926         struct bpf_tracing_link *tr_link =
2927                 container_of(link, struct bpf_tracing_link, link.link);
2928
2929         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2930                                                 tr_link->trampoline));
2931
2932         bpf_trampoline_put(tr_link->trampoline);
2933
2934         /* tgt_prog is NULL if target is a kernel function */
2935         if (tr_link->tgt_prog)
2936                 bpf_prog_put(tr_link->tgt_prog);
2937 }
2938
2939 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2940 {
2941         struct bpf_tracing_link *tr_link =
2942                 container_of(link, struct bpf_tracing_link, link.link);
2943
2944         kfree(tr_link);
2945 }
2946
2947 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2948                                          struct seq_file *seq)
2949 {
2950         struct bpf_tracing_link *tr_link =
2951                 container_of(link, struct bpf_tracing_link, link.link);
2952
2953         seq_printf(seq,
2954                    "attach_type:\t%d\n",
2955                    tr_link->attach_type);
2956 }
2957
2958 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2959                                            struct bpf_link_info *info)
2960 {
2961         struct bpf_tracing_link *tr_link =
2962                 container_of(link, struct bpf_tracing_link, link.link);
2963
2964         info->tracing.attach_type = tr_link->attach_type;
2965         bpf_trampoline_unpack_key(tr_link->trampoline->key,
2966                                   &info->tracing.target_obj_id,
2967                                   &info->tracing.target_btf_id);
2968
2969         return 0;
2970 }
2971
2972 static const struct bpf_link_ops bpf_tracing_link_lops = {
2973         .release = bpf_tracing_link_release,
2974         .dealloc = bpf_tracing_link_dealloc,
2975         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2976         .fill_link_info = bpf_tracing_link_fill_link_info,
2977 };
2978
2979 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2980                                    int tgt_prog_fd,
2981                                    u32 btf_id,
2982                                    u64 bpf_cookie)
2983 {
2984         struct bpf_link_primer link_primer;
2985         struct bpf_prog *tgt_prog = NULL;
2986         struct bpf_trampoline *tr = NULL;
2987         struct bpf_tracing_link *link;
2988         u64 key = 0;
2989         int err;
2990
2991         switch (prog->type) {
2992         case BPF_PROG_TYPE_TRACING:
2993                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2994                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2995                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2996                         err = -EINVAL;
2997                         goto out_put_prog;
2998                 }
2999                 break;
3000         case BPF_PROG_TYPE_EXT:
3001                 if (prog->expected_attach_type != 0) {
3002                         err = -EINVAL;
3003                         goto out_put_prog;
3004                 }
3005                 break;
3006         case BPF_PROG_TYPE_LSM:
3007                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3008                         err = -EINVAL;
3009                         goto out_put_prog;
3010                 }
3011                 break;
3012         default:
3013                 err = -EINVAL;
3014                 goto out_put_prog;
3015         }
3016
3017         if (!!tgt_prog_fd != !!btf_id) {
3018                 err = -EINVAL;
3019                 goto out_put_prog;
3020         }
3021
3022         if (tgt_prog_fd) {
3023                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3024                 if (prog->type != BPF_PROG_TYPE_EXT) {
3025                         err = -EINVAL;
3026                         goto out_put_prog;
3027                 }
3028
3029                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3030                 if (IS_ERR(tgt_prog)) {
3031                         err = PTR_ERR(tgt_prog);
3032                         tgt_prog = NULL;
3033                         goto out_put_prog;
3034                 }
3035
3036                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3037         }
3038
3039         link = kzalloc(sizeof(*link), GFP_USER);
3040         if (!link) {
3041                 err = -ENOMEM;
3042                 goto out_put_prog;
3043         }
3044         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3045                       &bpf_tracing_link_lops, prog);
3046         link->attach_type = prog->expected_attach_type;
3047         link->link.cookie = bpf_cookie;
3048
3049         mutex_lock(&prog->aux->dst_mutex);
3050
3051         /* There are a few possible cases here:
3052          *
3053          * - if prog->aux->dst_trampoline is set, the program was just loaded
3054          *   and not yet attached to anything, so we can use the values stored
3055          *   in prog->aux
3056          *
3057          * - if prog->aux->dst_trampoline is NULL, the program has already been
3058          *   attached to a target and its initial target was cleared (below)
3059          *
3060          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3061          *   target_btf_id using the link_create API.
3062          *
3063          * - if tgt_prog == NULL when this function was called using the old
3064          *   raw_tracepoint_open API, and we need a target from prog->aux
3065          *
3066          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3067          *   was detached and is going for re-attachment.
3068          */
3069         if (!prog->aux->dst_trampoline && !tgt_prog) {
3070                 /*
3071                  * Allow re-attach for TRACING and LSM programs. If it's
3072                  * currently linked, bpf_trampoline_link_prog will fail.
3073                  * EXT programs need to specify tgt_prog_fd, so they
3074                  * re-attach in separate code path.
3075                  */
3076                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3077                     prog->type != BPF_PROG_TYPE_LSM) {
3078                         err = -EINVAL;
3079                         goto out_unlock;
3080                 }
3081                 btf_id = prog->aux->attach_btf_id;
3082                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3083         }
3084
3085         if (!prog->aux->dst_trampoline ||
3086             (key && key != prog->aux->dst_trampoline->key)) {
3087                 /* If there is no saved target, or the specified target is
3088                  * different from the destination specified at load time, we
3089                  * need a new trampoline and a check for compatibility
3090                  */
3091                 struct bpf_attach_target_info tgt_info = {};
3092
3093                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3094                                               &tgt_info);
3095                 if (err)
3096                         goto out_unlock;
3097
3098                 tr = bpf_trampoline_get(key, &tgt_info);
3099                 if (!tr) {
3100                         err = -ENOMEM;
3101                         goto out_unlock;
3102                 }
3103         } else {
3104                 /* The caller didn't specify a target, or the target was the
3105                  * same as the destination supplied during program load. This
3106                  * means we can reuse the trampoline and reference from program
3107                  * load time, and there is no need to allocate a new one. This
3108                  * can only happen once for any program, as the saved values in
3109                  * prog->aux are cleared below.
3110                  */
3111                 tr = prog->aux->dst_trampoline;
3112                 tgt_prog = prog->aux->dst_prog;
3113         }
3114
3115         err = bpf_link_prime(&link->link.link, &link_primer);
3116         if (err)
3117                 goto out_unlock;
3118
3119         err = bpf_trampoline_link_prog(&link->link, tr);
3120         if (err) {
3121                 bpf_link_cleanup(&link_primer);
3122                 link = NULL;
3123                 goto out_unlock;
3124         }
3125
3126         link->tgt_prog = tgt_prog;
3127         link->trampoline = tr;
3128
3129         /* Always clear the trampoline and target prog from prog->aux to make
3130          * sure the original attach destination is not kept alive after a
3131          * program is (re-)attached to another target.
3132          */
3133         if (prog->aux->dst_prog &&
3134             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3135                 /* got extra prog ref from syscall, or attaching to different prog */
3136                 bpf_prog_put(prog->aux->dst_prog);
3137         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3138                 /* we allocated a new trampoline, so free the old one */
3139                 bpf_trampoline_put(prog->aux->dst_trampoline);
3140
3141         prog->aux->dst_prog = NULL;
3142         prog->aux->dst_trampoline = NULL;
3143         mutex_unlock(&prog->aux->dst_mutex);
3144
3145         return bpf_link_settle(&link_primer);
3146 out_unlock:
3147         if (tr && tr != prog->aux->dst_trampoline)
3148                 bpf_trampoline_put(tr);
3149         mutex_unlock(&prog->aux->dst_mutex);
3150         kfree(link);
3151 out_put_prog:
3152         if (tgt_prog_fd && tgt_prog)
3153                 bpf_prog_put(tgt_prog);
3154         return err;
3155 }
3156
3157 struct bpf_raw_tp_link {
3158         struct bpf_link link;
3159         struct bpf_raw_event_map *btp;
3160 };
3161
3162 static void bpf_raw_tp_link_release(struct bpf_link *link)
3163 {
3164         struct bpf_raw_tp_link *raw_tp =
3165                 container_of(link, struct bpf_raw_tp_link, link);
3166
3167         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3168         bpf_put_raw_tracepoint(raw_tp->btp);
3169 }
3170
3171 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3172 {
3173         struct bpf_raw_tp_link *raw_tp =
3174                 container_of(link, struct bpf_raw_tp_link, link);
3175
3176         kfree(raw_tp);
3177 }
3178
3179 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3180                                         struct seq_file *seq)
3181 {
3182         struct bpf_raw_tp_link *raw_tp_link =
3183                 container_of(link, struct bpf_raw_tp_link, link);
3184
3185         seq_printf(seq,
3186                    "tp_name:\t%s\n",
3187                    raw_tp_link->btp->tp->name);
3188 }
3189
3190 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3191                                           struct bpf_link_info *info)
3192 {
3193         struct bpf_raw_tp_link *raw_tp_link =
3194                 container_of(link, struct bpf_raw_tp_link, link);
3195         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3196         const char *tp_name = raw_tp_link->btp->tp->name;
3197         u32 ulen = info->raw_tracepoint.tp_name_len;
3198         size_t tp_len = strlen(tp_name);
3199
3200         if (!ulen ^ !ubuf)
3201                 return -EINVAL;
3202
3203         info->raw_tracepoint.tp_name_len = tp_len + 1;
3204
3205         if (!ubuf)
3206                 return 0;
3207
3208         if (ulen >= tp_len + 1) {
3209                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3210                         return -EFAULT;
3211         } else {
3212                 char zero = '\0';
3213
3214                 if (copy_to_user(ubuf, tp_name, ulen - 1))
3215                         return -EFAULT;
3216                 if (put_user(zero, ubuf + ulen - 1))
3217                         return -EFAULT;
3218                 return -ENOSPC;
3219         }
3220
3221         return 0;
3222 }
3223
3224 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3225         .release = bpf_raw_tp_link_release,
3226         .dealloc = bpf_raw_tp_link_dealloc,
3227         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3228         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3229 };
3230
3231 #ifdef CONFIG_PERF_EVENTS
3232 struct bpf_perf_link {
3233         struct bpf_link link;
3234         struct file *perf_file;
3235 };
3236
3237 static void bpf_perf_link_release(struct bpf_link *link)
3238 {
3239         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3240         struct perf_event *event = perf_link->perf_file->private_data;
3241
3242         perf_event_free_bpf_prog(event);
3243         fput(perf_link->perf_file);
3244 }
3245
3246 static void bpf_perf_link_dealloc(struct bpf_link *link)
3247 {
3248         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3249
3250         kfree(perf_link);
3251 }
3252
3253 static const struct bpf_link_ops bpf_perf_link_lops = {
3254         .release = bpf_perf_link_release,
3255         .dealloc = bpf_perf_link_dealloc,
3256 };
3257
3258 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3259 {
3260         struct bpf_link_primer link_primer;
3261         struct bpf_perf_link *link;
3262         struct perf_event *event;
3263         struct file *perf_file;
3264         int err;
3265
3266         if (attr->link_create.flags)
3267                 return -EINVAL;
3268
3269         perf_file = perf_event_get(attr->link_create.target_fd);
3270         if (IS_ERR(perf_file))
3271                 return PTR_ERR(perf_file);
3272
3273         link = kzalloc(sizeof(*link), GFP_USER);
3274         if (!link) {
3275                 err = -ENOMEM;
3276                 goto out_put_file;
3277         }
3278         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3279         link->perf_file = perf_file;
3280
3281         err = bpf_link_prime(&link->link, &link_primer);
3282         if (err) {
3283                 kfree(link);
3284                 goto out_put_file;
3285         }
3286
3287         event = perf_file->private_data;
3288         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3289         if (err) {
3290                 bpf_link_cleanup(&link_primer);
3291                 goto out_put_file;
3292         }
3293         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3294         bpf_prog_inc(prog);
3295
3296         return bpf_link_settle(&link_primer);
3297
3298 out_put_file:
3299         fput(perf_file);
3300         return err;
3301 }
3302 #else
3303 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3304 {
3305         return -EOPNOTSUPP;
3306 }
3307 #endif /* CONFIG_PERF_EVENTS */
3308
3309 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3310                                   const char __user *user_tp_name)
3311 {
3312         struct bpf_link_primer link_primer;
3313         struct bpf_raw_tp_link *link;
3314         struct bpf_raw_event_map *btp;
3315         const char *tp_name;
3316         char buf[128];
3317         int err;
3318
3319         switch (prog->type) {
3320         case BPF_PROG_TYPE_TRACING:
3321         case BPF_PROG_TYPE_EXT:
3322         case BPF_PROG_TYPE_LSM:
3323                 if (user_tp_name)
3324                         /* The attach point for this category of programs
3325                          * should be specified via btf_id during program load.
3326                          */
3327                         return -EINVAL;
3328                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3329                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3330                         tp_name = prog->aux->attach_func_name;
3331                         break;
3332                 }
3333                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3334         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3335         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3336                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3337                         return -EFAULT;
3338                 buf[sizeof(buf) - 1] = 0;
3339                 tp_name = buf;
3340                 break;
3341         default:
3342                 return -EINVAL;
3343         }
3344
3345         btp = bpf_get_raw_tracepoint(tp_name);
3346         if (!btp)
3347                 return -ENOENT;
3348
3349         link = kzalloc(sizeof(*link), GFP_USER);
3350         if (!link) {
3351                 err = -ENOMEM;
3352                 goto out_put_btp;
3353         }
3354         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3355                       &bpf_raw_tp_link_lops, prog);
3356         link->btp = btp;
3357
3358         err = bpf_link_prime(&link->link, &link_primer);
3359         if (err) {
3360                 kfree(link);
3361                 goto out_put_btp;
3362         }
3363
3364         err = bpf_probe_register(link->btp, prog);
3365         if (err) {
3366                 bpf_link_cleanup(&link_primer);
3367                 goto out_put_btp;
3368         }
3369
3370         return bpf_link_settle(&link_primer);
3371
3372 out_put_btp:
3373         bpf_put_raw_tracepoint(btp);
3374         return err;
3375 }
3376
3377 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3378
3379 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3380 {
3381         struct bpf_prog *prog;
3382         int fd;
3383
3384         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3385                 return -EINVAL;
3386
3387         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3388         if (IS_ERR(prog))
3389                 return PTR_ERR(prog);
3390
3391         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3392         if (fd < 0)
3393                 bpf_prog_put(prog);
3394         return fd;
3395 }
3396
3397 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3398                                              enum bpf_attach_type attach_type)
3399 {
3400         switch (prog->type) {
3401         case BPF_PROG_TYPE_CGROUP_SOCK:
3402         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3403         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3404         case BPF_PROG_TYPE_SK_LOOKUP:
3405                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3406         case BPF_PROG_TYPE_CGROUP_SKB:
3407                 if (!capable(CAP_NET_ADMIN))
3408                         /* cg-skb progs can be loaded by unpriv user.
3409                          * check permissions at attach time.
3410                          */
3411                         return -EPERM;
3412                 return prog->enforce_expected_attach_type &&
3413                         prog->expected_attach_type != attach_type ?
3414                         -EINVAL : 0;
3415         default:
3416                 return 0;
3417         }
3418 }
3419
3420 static enum bpf_prog_type
3421 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3422 {
3423         switch (attach_type) {
3424         case BPF_CGROUP_INET_INGRESS:
3425         case BPF_CGROUP_INET_EGRESS:
3426                 return BPF_PROG_TYPE_CGROUP_SKB;
3427         case BPF_CGROUP_INET_SOCK_CREATE:
3428         case BPF_CGROUP_INET_SOCK_RELEASE:
3429         case BPF_CGROUP_INET4_POST_BIND:
3430         case BPF_CGROUP_INET6_POST_BIND:
3431                 return BPF_PROG_TYPE_CGROUP_SOCK;
3432         case BPF_CGROUP_INET4_BIND:
3433         case BPF_CGROUP_INET6_BIND:
3434         case BPF_CGROUP_INET4_CONNECT:
3435         case BPF_CGROUP_INET6_CONNECT:
3436         case BPF_CGROUP_INET4_GETPEERNAME:
3437         case BPF_CGROUP_INET6_GETPEERNAME:
3438         case BPF_CGROUP_INET4_GETSOCKNAME:
3439         case BPF_CGROUP_INET6_GETSOCKNAME:
3440         case BPF_CGROUP_UDP4_SENDMSG:
3441         case BPF_CGROUP_UDP6_SENDMSG:
3442         case BPF_CGROUP_UDP4_RECVMSG:
3443         case BPF_CGROUP_UDP6_RECVMSG:
3444                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3445         case BPF_CGROUP_SOCK_OPS:
3446                 return BPF_PROG_TYPE_SOCK_OPS;
3447         case BPF_CGROUP_DEVICE:
3448                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3449         case BPF_SK_MSG_VERDICT:
3450                 return BPF_PROG_TYPE_SK_MSG;
3451         case BPF_SK_SKB_STREAM_PARSER:
3452         case BPF_SK_SKB_STREAM_VERDICT:
3453         case BPF_SK_SKB_VERDICT:
3454                 return BPF_PROG_TYPE_SK_SKB;
3455         case BPF_LIRC_MODE2:
3456                 return BPF_PROG_TYPE_LIRC_MODE2;
3457         case BPF_FLOW_DISSECTOR:
3458                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3459         case BPF_CGROUP_SYSCTL:
3460                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3461         case BPF_CGROUP_GETSOCKOPT:
3462         case BPF_CGROUP_SETSOCKOPT:
3463                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3464         case BPF_TRACE_ITER:
3465         case BPF_TRACE_RAW_TP:
3466         case BPF_TRACE_FENTRY:
3467         case BPF_TRACE_FEXIT:
3468         case BPF_MODIFY_RETURN:
3469                 return BPF_PROG_TYPE_TRACING;
3470         case BPF_LSM_MAC:
3471                 return BPF_PROG_TYPE_LSM;
3472         case BPF_SK_LOOKUP:
3473                 return BPF_PROG_TYPE_SK_LOOKUP;
3474         case BPF_XDP:
3475                 return BPF_PROG_TYPE_XDP;
3476         case BPF_LSM_CGROUP:
3477                 return BPF_PROG_TYPE_LSM;
3478         default:
3479                 return BPF_PROG_TYPE_UNSPEC;
3480         }
3481 }
3482
3483 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3484
3485 #define BPF_F_ATTACH_MASK \
3486         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3487
3488 static int bpf_prog_attach(const union bpf_attr *attr)
3489 {
3490         enum bpf_prog_type ptype;
3491         struct bpf_prog *prog;
3492         int ret;
3493
3494         if (CHECK_ATTR(BPF_PROG_ATTACH))
3495                 return -EINVAL;
3496
3497         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3498                 return -EINVAL;
3499
3500         ptype = attach_type_to_prog_type(attr->attach_type);
3501         if (ptype == BPF_PROG_TYPE_UNSPEC)
3502                 return -EINVAL;
3503
3504         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3505         if (IS_ERR(prog))
3506                 return PTR_ERR(prog);
3507
3508         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3509                 bpf_prog_put(prog);
3510                 return -EINVAL;
3511         }
3512
3513         switch (ptype) {
3514         case BPF_PROG_TYPE_SK_SKB:
3515         case BPF_PROG_TYPE_SK_MSG:
3516                 ret = sock_map_get_from_fd(attr, prog);
3517                 break;
3518         case BPF_PROG_TYPE_LIRC_MODE2:
3519                 ret = lirc_prog_attach(attr, prog);
3520                 break;
3521         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3522                 ret = netns_bpf_prog_attach(attr, prog);
3523                 break;
3524         case BPF_PROG_TYPE_CGROUP_DEVICE:
3525         case BPF_PROG_TYPE_CGROUP_SKB:
3526         case BPF_PROG_TYPE_CGROUP_SOCK:
3527         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3528         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3529         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3530         case BPF_PROG_TYPE_SOCK_OPS:
3531         case BPF_PROG_TYPE_LSM:
3532                 if (ptype == BPF_PROG_TYPE_LSM &&
3533                     prog->expected_attach_type != BPF_LSM_CGROUP)
3534                         ret = -EINVAL;
3535                 else
3536                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3537                 break;
3538         default:
3539                 ret = -EINVAL;
3540         }
3541
3542         if (ret)
3543                 bpf_prog_put(prog);
3544         return ret;
3545 }
3546
3547 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3548
3549 static int bpf_prog_detach(const union bpf_attr *attr)
3550 {
3551         enum bpf_prog_type ptype;
3552
3553         if (CHECK_ATTR(BPF_PROG_DETACH))
3554                 return -EINVAL;
3555
3556         ptype = attach_type_to_prog_type(attr->attach_type);
3557
3558         switch (ptype) {
3559         case BPF_PROG_TYPE_SK_MSG:
3560         case BPF_PROG_TYPE_SK_SKB:
3561                 return sock_map_prog_detach(attr, ptype);
3562         case BPF_PROG_TYPE_LIRC_MODE2:
3563                 return lirc_prog_detach(attr);
3564         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3565                 return netns_bpf_prog_detach(attr, ptype);
3566         case BPF_PROG_TYPE_CGROUP_DEVICE:
3567         case BPF_PROG_TYPE_CGROUP_SKB:
3568         case BPF_PROG_TYPE_CGROUP_SOCK:
3569         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3570         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3571         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3572         case BPF_PROG_TYPE_SOCK_OPS:
3573         case BPF_PROG_TYPE_LSM:
3574                 return cgroup_bpf_prog_detach(attr, ptype);
3575         default:
3576                 return -EINVAL;
3577         }
3578 }
3579
3580 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3581
3582 static int bpf_prog_query(const union bpf_attr *attr,
3583                           union bpf_attr __user *uattr)
3584 {
3585         if (!capable(CAP_NET_ADMIN))
3586                 return -EPERM;
3587         if (CHECK_ATTR(BPF_PROG_QUERY))
3588                 return -EINVAL;
3589         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3590                 return -EINVAL;
3591
3592         switch (attr->query.attach_type) {
3593         case BPF_CGROUP_INET_INGRESS:
3594         case BPF_CGROUP_INET_EGRESS:
3595         case BPF_CGROUP_INET_SOCK_CREATE:
3596         case BPF_CGROUP_INET_SOCK_RELEASE:
3597         case BPF_CGROUP_INET4_BIND:
3598         case BPF_CGROUP_INET6_BIND:
3599         case BPF_CGROUP_INET4_POST_BIND:
3600         case BPF_CGROUP_INET6_POST_BIND:
3601         case BPF_CGROUP_INET4_CONNECT:
3602         case BPF_CGROUP_INET6_CONNECT:
3603         case BPF_CGROUP_INET4_GETPEERNAME:
3604         case BPF_CGROUP_INET6_GETPEERNAME:
3605         case BPF_CGROUP_INET4_GETSOCKNAME:
3606         case BPF_CGROUP_INET6_GETSOCKNAME:
3607         case BPF_CGROUP_UDP4_SENDMSG:
3608         case BPF_CGROUP_UDP6_SENDMSG:
3609         case BPF_CGROUP_UDP4_RECVMSG:
3610         case BPF_CGROUP_UDP6_RECVMSG:
3611         case BPF_CGROUP_SOCK_OPS:
3612         case BPF_CGROUP_DEVICE:
3613         case BPF_CGROUP_SYSCTL:
3614         case BPF_CGROUP_GETSOCKOPT:
3615         case BPF_CGROUP_SETSOCKOPT:
3616         case BPF_LSM_CGROUP:
3617                 return cgroup_bpf_prog_query(attr, uattr);
3618         case BPF_LIRC_MODE2:
3619                 return lirc_prog_query(attr, uattr);
3620         case BPF_FLOW_DISSECTOR:
3621         case BPF_SK_LOOKUP:
3622                 return netns_bpf_prog_query(attr, uattr);
3623         case BPF_SK_SKB_STREAM_PARSER:
3624         case BPF_SK_SKB_STREAM_VERDICT:
3625         case BPF_SK_MSG_VERDICT:
3626         case BPF_SK_SKB_VERDICT:
3627                 return sock_map_bpf_prog_query(attr, uattr);
3628         default:
3629                 return -EINVAL;
3630         }
3631 }
3632
3633 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3634
3635 static int bpf_prog_test_run(const union bpf_attr *attr,
3636                              union bpf_attr __user *uattr)
3637 {
3638         struct bpf_prog *prog;
3639         int ret = -ENOTSUPP;
3640
3641         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3642                 return -EINVAL;
3643
3644         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3645             (!attr->test.ctx_size_in && attr->test.ctx_in))
3646                 return -EINVAL;
3647
3648         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3649             (!attr->test.ctx_size_out && attr->test.ctx_out))
3650                 return -EINVAL;
3651
3652         prog = bpf_prog_get(attr->test.prog_fd);
3653         if (IS_ERR(prog))
3654                 return PTR_ERR(prog);
3655
3656         if (prog->aux->ops->test_run)
3657                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3658
3659         bpf_prog_put(prog);
3660         return ret;
3661 }
3662
3663 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3664
3665 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3666                                union bpf_attr __user *uattr,
3667                                struct idr *idr,
3668                                spinlock_t *lock)
3669 {
3670         u32 next_id = attr->start_id;
3671         int err = 0;
3672
3673         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3674                 return -EINVAL;
3675
3676         if (!capable(CAP_SYS_ADMIN))
3677                 return -EPERM;
3678
3679         next_id++;
3680         spin_lock_bh(lock);
3681         if (!idr_get_next(idr, &next_id))
3682                 err = -ENOENT;
3683         spin_unlock_bh(lock);
3684
3685         if (!err)
3686                 err = put_user(next_id, &uattr->next_id);
3687
3688         return err;
3689 }
3690
3691 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3692 {
3693         struct bpf_map *map;
3694
3695         spin_lock_bh(&map_idr_lock);
3696 again:
3697         map = idr_get_next(&map_idr, id);
3698         if (map) {
3699                 map = __bpf_map_inc_not_zero(map, false);
3700                 if (IS_ERR(map)) {
3701                         (*id)++;
3702                         goto again;
3703                 }
3704         }
3705         spin_unlock_bh(&map_idr_lock);
3706
3707         return map;
3708 }
3709
3710 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3711 {
3712         struct bpf_prog *prog;
3713
3714         spin_lock_bh(&prog_idr_lock);
3715 again:
3716         prog = idr_get_next(&prog_idr, id);
3717         if (prog) {
3718                 prog = bpf_prog_inc_not_zero(prog);
3719                 if (IS_ERR(prog)) {
3720                         (*id)++;
3721                         goto again;
3722                 }
3723         }
3724         spin_unlock_bh(&prog_idr_lock);
3725
3726         return prog;
3727 }
3728
3729 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3730
3731 struct bpf_prog *bpf_prog_by_id(u32 id)
3732 {
3733         struct bpf_prog *prog;
3734
3735         if (!id)
3736                 return ERR_PTR(-ENOENT);
3737
3738         spin_lock_bh(&prog_idr_lock);
3739         prog = idr_find(&prog_idr, id);
3740         if (prog)
3741                 prog = bpf_prog_inc_not_zero(prog);
3742         else
3743                 prog = ERR_PTR(-ENOENT);
3744         spin_unlock_bh(&prog_idr_lock);
3745         return prog;
3746 }
3747
3748 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3749 {
3750         struct bpf_prog *prog;
3751         u32 id = attr->prog_id;
3752         int fd;
3753
3754         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3755                 return -EINVAL;
3756
3757         if (!capable(CAP_SYS_ADMIN))
3758                 return -EPERM;
3759
3760         prog = bpf_prog_by_id(id);
3761         if (IS_ERR(prog))
3762                 return PTR_ERR(prog);
3763
3764         fd = bpf_prog_new_fd(prog);
3765         if (fd < 0)
3766                 bpf_prog_put(prog);
3767
3768         return fd;
3769 }
3770
3771 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3772
3773 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3774 {
3775         struct bpf_map *map;
3776         u32 id = attr->map_id;
3777         int f_flags;
3778         int fd;
3779
3780         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3781             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3782                 return -EINVAL;
3783
3784         if (!capable(CAP_SYS_ADMIN))
3785                 return -EPERM;
3786
3787         f_flags = bpf_get_file_flag(attr->open_flags);
3788         if (f_flags < 0)
3789                 return f_flags;
3790
3791         spin_lock_bh(&map_idr_lock);
3792         map = idr_find(&map_idr, id);
3793         if (map)
3794                 map = __bpf_map_inc_not_zero(map, true);
3795         else
3796                 map = ERR_PTR(-ENOENT);
3797         spin_unlock_bh(&map_idr_lock);
3798
3799         if (IS_ERR(map))
3800                 return PTR_ERR(map);
3801
3802         fd = bpf_map_new_fd(map, f_flags);
3803         if (fd < 0)
3804                 bpf_map_put_with_uref(map);
3805
3806         return fd;
3807 }
3808
3809 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3810                                               unsigned long addr, u32 *off,
3811                                               u32 *type)
3812 {
3813         const struct bpf_map *map;
3814         int i;
3815
3816         mutex_lock(&prog->aux->used_maps_mutex);
3817         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3818                 map = prog->aux->used_maps[i];
3819                 if (map == (void *)addr) {
3820                         *type = BPF_PSEUDO_MAP_FD;
3821                         goto out;
3822                 }
3823                 if (!map->ops->map_direct_value_meta)
3824                         continue;
3825                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3826                         *type = BPF_PSEUDO_MAP_VALUE;
3827                         goto out;
3828                 }
3829         }
3830         map = NULL;
3831
3832 out:
3833         mutex_unlock(&prog->aux->used_maps_mutex);
3834         return map;
3835 }
3836
3837 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3838                                               const struct cred *f_cred)
3839 {
3840         const struct bpf_map *map;
3841         struct bpf_insn *insns;
3842         u32 off, type;
3843         u64 imm;
3844         u8 code;
3845         int i;
3846
3847         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3848                         GFP_USER);
3849         if (!insns)
3850                 return insns;
3851
3852         for (i = 0; i < prog->len; i++) {
3853                 code = insns[i].code;
3854
3855                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3856                         insns[i].code = BPF_JMP | BPF_CALL;
3857                         insns[i].imm = BPF_FUNC_tail_call;
3858                         /* fall-through */
3859                 }
3860                 if (code == (BPF_JMP | BPF_CALL) ||
3861                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3862                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3863                                 insns[i].code = BPF_JMP | BPF_CALL;
3864                         if (!bpf_dump_raw_ok(f_cred))
3865                                 insns[i].imm = 0;
3866                         continue;
3867                 }
3868                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3869                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3870                         continue;
3871                 }
3872
3873                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3874                         continue;
3875
3876                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3877                 map = bpf_map_from_imm(prog, imm, &off, &type);
3878                 if (map) {
3879                         insns[i].src_reg = type;
3880                         insns[i].imm = map->id;
3881                         insns[i + 1].imm = off;
3882                         continue;
3883                 }
3884         }
3885
3886         return insns;
3887 }
3888
3889 static int set_info_rec_size(struct bpf_prog_info *info)
3890 {
3891         /*
3892          * Ensure info.*_rec_size is the same as kernel expected size
3893          *
3894          * or
3895          *
3896          * Only allow zero *_rec_size if both _rec_size and _cnt are
3897          * zero.  In this case, the kernel will set the expected
3898          * _rec_size back to the info.
3899          */
3900
3901         if ((info->nr_func_info || info->func_info_rec_size) &&
3902             info->func_info_rec_size != sizeof(struct bpf_func_info))
3903                 return -EINVAL;
3904
3905         if ((info->nr_line_info || info->line_info_rec_size) &&
3906             info->line_info_rec_size != sizeof(struct bpf_line_info))
3907                 return -EINVAL;
3908
3909         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3910             info->jited_line_info_rec_size != sizeof(__u64))
3911                 return -EINVAL;
3912
3913         info->func_info_rec_size = sizeof(struct bpf_func_info);
3914         info->line_info_rec_size = sizeof(struct bpf_line_info);
3915         info->jited_line_info_rec_size = sizeof(__u64);
3916
3917         return 0;
3918 }
3919
3920 static int bpf_prog_get_info_by_fd(struct file *file,
3921                                    struct bpf_prog *prog,
3922                                    const union bpf_attr *attr,
3923                                    union bpf_attr __user *uattr)
3924 {
3925         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3926         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3927         struct bpf_prog_info info;
3928         u32 info_len = attr->info.info_len;
3929         struct bpf_prog_kstats stats;
3930         char __user *uinsns;
3931         u32 ulen;
3932         int err;
3933
3934         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3935         if (err)
3936                 return err;
3937         info_len = min_t(u32, sizeof(info), info_len);
3938
3939         memset(&info, 0, sizeof(info));
3940         if (copy_from_user(&info, uinfo, info_len))
3941                 return -EFAULT;
3942
3943         info.type = prog->type;
3944         info.id = prog->aux->id;
3945         info.load_time = prog->aux->load_time;
3946         info.created_by_uid = from_kuid_munged(current_user_ns(),
3947                                                prog->aux->user->uid);
3948         info.gpl_compatible = prog->gpl_compatible;
3949
3950         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3951         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3952
3953         mutex_lock(&prog->aux->used_maps_mutex);
3954         ulen = info.nr_map_ids;
3955         info.nr_map_ids = prog->aux->used_map_cnt;
3956         ulen = min_t(u32, info.nr_map_ids, ulen);
3957         if (ulen) {
3958                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3959                 u32 i;
3960
3961                 for (i = 0; i < ulen; i++)
3962                         if (put_user(prog->aux->used_maps[i]->id,
3963                                      &user_map_ids[i])) {
3964                                 mutex_unlock(&prog->aux->used_maps_mutex);
3965                                 return -EFAULT;
3966                         }
3967         }
3968         mutex_unlock(&prog->aux->used_maps_mutex);
3969
3970         err = set_info_rec_size(&info);
3971         if (err)
3972                 return err;
3973
3974         bpf_prog_get_stats(prog, &stats);
3975         info.run_time_ns = stats.nsecs;
3976         info.run_cnt = stats.cnt;
3977         info.recursion_misses = stats.misses;
3978
3979         info.verified_insns = prog->aux->verified_insns;
3980
3981         if (!bpf_capable()) {
3982                 info.jited_prog_len = 0;
3983                 info.xlated_prog_len = 0;
3984                 info.nr_jited_ksyms = 0;
3985                 info.nr_jited_func_lens = 0;
3986                 info.nr_func_info = 0;
3987                 info.nr_line_info = 0;
3988                 info.nr_jited_line_info = 0;
3989                 goto done;
3990         }
3991
3992         ulen = info.xlated_prog_len;
3993         info.xlated_prog_len = bpf_prog_insn_size(prog);
3994         if (info.xlated_prog_len && ulen) {
3995                 struct bpf_insn *insns_sanitized;
3996                 bool fault;
3997
3998                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3999                         info.xlated_prog_insns = 0;
4000                         goto done;
4001                 }
4002                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4003                 if (!insns_sanitized)
4004                         return -ENOMEM;
4005                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4006                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4007                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4008                 kfree(insns_sanitized);
4009                 if (fault)
4010                         return -EFAULT;
4011         }
4012
4013         if (bpf_prog_is_offloaded(prog->aux)) {
4014                 err = bpf_prog_offload_info_fill(&info, prog);
4015                 if (err)
4016                         return err;
4017                 goto done;
4018         }
4019
4020         /* NOTE: the following code is supposed to be skipped for offload.
4021          * bpf_prog_offload_info_fill() is the place to fill similar fields
4022          * for offload.
4023          */
4024         ulen = info.jited_prog_len;
4025         if (prog->aux->func_cnt) {
4026                 u32 i;
4027
4028                 info.jited_prog_len = 0;
4029                 for (i = 0; i < prog->aux->func_cnt; i++)
4030                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4031         } else {
4032                 info.jited_prog_len = prog->jited_len;
4033         }
4034
4035         if (info.jited_prog_len && ulen) {
4036                 if (bpf_dump_raw_ok(file->f_cred)) {
4037                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4038                         ulen = min_t(u32, info.jited_prog_len, ulen);
4039
4040                         /* for multi-function programs, copy the JITed
4041                          * instructions for all the functions
4042                          */
4043                         if (prog->aux->func_cnt) {
4044                                 u32 len, free, i;
4045                                 u8 *img;
4046
4047                                 free = ulen;
4048                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4049                                         len = prog->aux->func[i]->jited_len;
4050                                         len = min_t(u32, len, free);
4051                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4052                                         if (copy_to_user(uinsns, img, len))
4053                                                 return -EFAULT;
4054                                         uinsns += len;
4055                                         free -= len;
4056                                         if (!free)
4057                                                 break;
4058                                 }
4059                         } else {
4060                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4061                                         return -EFAULT;
4062                         }
4063                 } else {
4064                         info.jited_prog_insns = 0;
4065                 }
4066         }
4067
4068         ulen = info.nr_jited_ksyms;
4069         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4070         if (ulen) {
4071                 if (bpf_dump_raw_ok(file->f_cred)) {
4072                         unsigned long ksym_addr;
4073                         u64 __user *user_ksyms;
4074                         u32 i;
4075
4076                         /* copy the address of the kernel symbol
4077                          * corresponding to each function
4078                          */
4079                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4080                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4081                         if (prog->aux->func_cnt) {
4082                                 for (i = 0; i < ulen; i++) {
4083                                         ksym_addr = (unsigned long)
4084                                                 prog->aux->func[i]->bpf_func;
4085                                         if (put_user((u64) ksym_addr,
4086                                                      &user_ksyms[i]))
4087                                                 return -EFAULT;
4088                                 }
4089                         } else {
4090                                 ksym_addr = (unsigned long) prog->bpf_func;
4091                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4092                                         return -EFAULT;
4093                         }
4094                 } else {
4095                         info.jited_ksyms = 0;
4096                 }
4097         }
4098
4099         ulen = info.nr_jited_func_lens;
4100         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4101         if (ulen) {
4102                 if (bpf_dump_raw_ok(file->f_cred)) {
4103                         u32 __user *user_lens;
4104                         u32 func_len, i;
4105
4106                         /* copy the JITed image lengths for each function */
4107                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4108                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4109                         if (prog->aux->func_cnt) {
4110                                 for (i = 0; i < ulen; i++) {
4111                                         func_len =
4112                                                 prog->aux->func[i]->jited_len;
4113                                         if (put_user(func_len, &user_lens[i]))
4114                                                 return -EFAULT;
4115                                 }
4116                         } else {
4117                                 func_len = prog->jited_len;
4118                                 if (put_user(func_len, &user_lens[0]))
4119                                         return -EFAULT;
4120                         }
4121                 } else {
4122                         info.jited_func_lens = 0;
4123                 }
4124         }
4125
4126         if (prog->aux->btf)
4127                 info.btf_id = btf_obj_id(prog->aux->btf);
4128         info.attach_btf_id = prog->aux->attach_btf_id;
4129         if (attach_btf)
4130                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4131
4132         ulen = info.nr_func_info;
4133         info.nr_func_info = prog->aux->func_info_cnt;
4134         if (info.nr_func_info && ulen) {
4135                 char __user *user_finfo;
4136
4137                 user_finfo = u64_to_user_ptr(info.func_info);
4138                 ulen = min_t(u32, info.nr_func_info, ulen);
4139                 if (copy_to_user(user_finfo, prog->aux->func_info,
4140                                  info.func_info_rec_size * ulen))
4141                         return -EFAULT;
4142         }
4143
4144         ulen = info.nr_line_info;
4145         info.nr_line_info = prog->aux->nr_linfo;
4146         if (info.nr_line_info && ulen) {
4147                 __u8 __user *user_linfo;
4148
4149                 user_linfo = u64_to_user_ptr(info.line_info);
4150                 ulen = min_t(u32, info.nr_line_info, ulen);
4151                 if (copy_to_user(user_linfo, prog->aux->linfo,
4152                                  info.line_info_rec_size * ulen))
4153                         return -EFAULT;
4154         }
4155
4156         ulen = info.nr_jited_line_info;
4157         if (prog->aux->jited_linfo)
4158                 info.nr_jited_line_info = prog->aux->nr_linfo;
4159         else
4160                 info.nr_jited_line_info = 0;
4161         if (info.nr_jited_line_info && ulen) {
4162                 if (bpf_dump_raw_ok(file->f_cred)) {
4163                         unsigned long line_addr;
4164                         __u64 __user *user_linfo;
4165                         u32 i;
4166
4167                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4168                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4169                         for (i = 0; i < ulen; i++) {
4170                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4171                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4172                                         return -EFAULT;
4173                         }
4174                 } else {
4175                         info.jited_line_info = 0;
4176                 }
4177         }
4178
4179         ulen = info.nr_prog_tags;
4180         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4181         if (ulen) {
4182                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4183                 u32 i;
4184
4185                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4186                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4187                 if (prog->aux->func_cnt) {
4188                         for (i = 0; i < ulen; i++) {
4189                                 if (copy_to_user(user_prog_tags[i],
4190                                                  prog->aux->func[i]->tag,
4191                                                  BPF_TAG_SIZE))
4192                                         return -EFAULT;
4193                         }
4194                 } else {
4195                         if (copy_to_user(user_prog_tags[0],
4196                                          prog->tag, BPF_TAG_SIZE))
4197                                 return -EFAULT;
4198                 }
4199         }
4200
4201 done:
4202         if (copy_to_user(uinfo, &info, info_len) ||
4203             put_user(info_len, &uattr->info.info_len))
4204                 return -EFAULT;
4205
4206         return 0;
4207 }
4208
4209 static int bpf_map_get_info_by_fd(struct file *file,
4210                                   struct bpf_map *map,
4211                                   const union bpf_attr *attr,
4212                                   union bpf_attr __user *uattr)
4213 {
4214         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4215         struct bpf_map_info info;
4216         u32 info_len = attr->info.info_len;
4217         int err;
4218
4219         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4220         if (err)
4221                 return err;
4222         info_len = min_t(u32, sizeof(info), info_len);
4223
4224         memset(&info, 0, sizeof(info));
4225         info.type = map->map_type;
4226         info.id = map->id;
4227         info.key_size = map->key_size;
4228         info.value_size = map->value_size;
4229         info.max_entries = map->max_entries;
4230         info.map_flags = map->map_flags;
4231         info.map_extra = map->map_extra;
4232         memcpy(info.name, map->name, sizeof(map->name));
4233
4234         if (map->btf) {
4235                 info.btf_id = btf_obj_id(map->btf);
4236                 info.btf_key_type_id = map->btf_key_type_id;
4237                 info.btf_value_type_id = map->btf_value_type_id;
4238         }
4239         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4240
4241         if (bpf_map_is_offloaded(map)) {
4242                 err = bpf_map_offload_info_fill(&info, map);
4243                 if (err)
4244                         return err;
4245         }
4246
4247         if (copy_to_user(uinfo, &info, info_len) ||
4248             put_user(info_len, &uattr->info.info_len))
4249                 return -EFAULT;
4250
4251         return 0;
4252 }
4253
4254 static int bpf_btf_get_info_by_fd(struct file *file,
4255                                   struct btf *btf,
4256                                   const union bpf_attr *attr,
4257                                   union bpf_attr __user *uattr)
4258 {
4259         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4260         u32 info_len = attr->info.info_len;
4261         int err;
4262
4263         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4264         if (err)
4265                 return err;
4266
4267         return btf_get_info_by_fd(btf, attr, uattr);
4268 }
4269
4270 static int bpf_link_get_info_by_fd(struct file *file,
4271                                   struct bpf_link *link,
4272                                   const union bpf_attr *attr,
4273                                   union bpf_attr __user *uattr)
4274 {
4275         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4276         struct bpf_link_info info;
4277         u32 info_len = attr->info.info_len;
4278         int err;
4279
4280         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4281         if (err)
4282                 return err;
4283         info_len = min_t(u32, sizeof(info), info_len);
4284
4285         memset(&info, 0, sizeof(info));
4286         if (copy_from_user(&info, uinfo, info_len))
4287                 return -EFAULT;
4288
4289         info.type = link->type;
4290         info.id = link->id;
4291         info.prog_id = link->prog->aux->id;
4292
4293         if (link->ops->fill_link_info) {
4294                 err = link->ops->fill_link_info(link, &info);
4295                 if (err)
4296                         return err;
4297         }
4298
4299         if (copy_to_user(uinfo, &info, info_len) ||
4300             put_user(info_len, &uattr->info.info_len))
4301                 return -EFAULT;
4302
4303         return 0;
4304 }
4305
4306
4307 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4308
4309 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4310                                   union bpf_attr __user *uattr)
4311 {
4312         int ufd = attr->info.bpf_fd;
4313         struct fd f;
4314         int err;
4315
4316         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4317                 return -EINVAL;
4318
4319         f = fdget(ufd);
4320         if (!f.file)
4321                 return -EBADFD;
4322
4323         if (f.file->f_op == &bpf_prog_fops)
4324                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4325                                               uattr);
4326         else if (f.file->f_op == &bpf_map_fops)
4327                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4328                                              uattr);
4329         else if (f.file->f_op == &btf_fops)
4330                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4331         else if (f.file->f_op == &bpf_link_fops)
4332                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4333                                               attr, uattr);
4334         else
4335                 err = -EINVAL;
4336
4337         fdput(f);
4338         return err;
4339 }
4340
4341 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4342
4343 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
4344 {
4345         if (CHECK_ATTR(BPF_BTF_LOAD))
4346                 return -EINVAL;
4347
4348         if (!bpf_capable())
4349                 return -EPERM;
4350
4351         return btf_new_fd(attr, uattr);
4352 }
4353
4354 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4355
4356 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4357 {
4358         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4359                 return -EINVAL;
4360
4361         if (!capable(CAP_SYS_ADMIN))
4362                 return -EPERM;
4363
4364         return btf_get_fd_by_id(attr->btf_id);
4365 }
4366
4367 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4368                                     union bpf_attr __user *uattr,
4369                                     u32 prog_id, u32 fd_type,
4370                                     const char *buf, u64 probe_offset,
4371                                     u64 probe_addr)
4372 {
4373         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4374         u32 len = buf ? strlen(buf) : 0, input_len;
4375         int err = 0;
4376
4377         if (put_user(len, &uattr->task_fd_query.buf_len))
4378                 return -EFAULT;
4379         input_len = attr->task_fd_query.buf_len;
4380         if (input_len && ubuf) {
4381                 if (!len) {
4382                         /* nothing to copy, just make ubuf NULL terminated */
4383                         char zero = '\0';
4384
4385                         if (put_user(zero, ubuf))
4386                                 return -EFAULT;
4387                 } else if (input_len >= len + 1) {
4388                         /* ubuf can hold the string with NULL terminator */
4389                         if (copy_to_user(ubuf, buf, len + 1))
4390                                 return -EFAULT;
4391                 } else {
4392                         /* ubuf cannot hold the string with NULL terminator,
4393                          * do a partial copy with NULL terminator.
4394                          */
4395                         char zero = '\0';
4396
4397                         err = -ENOSPC;
4398                         if (copy_to_user(ubuf, buf, input_len - 1))
4399                                 return -EFAULT;
4400                         if (put_user(zero, ubuf + input_len - 1))
4401                                 return -EFAULT;
4402                 }
4403         }
4404
4405         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4406             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4407             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4408             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4409                 return -EFAULT;
4410
4411         return err;
4412 }
4413
4414 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4415
4416 static int bpf_task_fd_query(const union bpf_attr *attr,
4417                              union bpf_attr __user *uattr)
4418 {
4419         pid_t pid = attr->task_fd_query.pid;
4420         u32 fd = attr->task_fd_query.fd;
4421         const struct perf_event *event;
4422         struct task_struct *task;
4423         struct file *file;
4424         int err;
4425
4426         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4427                 return -EINVAL;
4428
4429         if (!capable(CAP_SYS_ADMIN))
4430                 return -EPERM;
4431
4432         if (attr->task_fd_query.flags != 0)
4433                 return -EINVAL;
4434
4435         rcu_read_lock();
4436         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4437         rcu_read_unlock();
4438         if (!task)
4439                 return -ENOENT;
4440
4441         err = 0;
4442         file = fget_task(task, fd);
4443         put_task_struct(task);
4444         if (!file)
4445                 return -EBADF;
4446
4447         if (file->f_op == &bpf_link_fops) {
4448                 struct bpf_link *link = file->private_data;
4449
4450                 if (link->ops == &bpf_raw_tp_link_lops) {
4451                         struct bpf_raw_tp_link *raw_tp =
4452                                 container_of(link, struct bpf_raw_tp_link, link);
4453                         struct bpf_raw_event_map *btp = raw_tp->btp;
4454
4455                         err = bpf_task_fd_query_copy(attr, uattr,
4456                                                      raw_tp->link.prog->aux->id,
4457                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4458                                                      btp->tp->name, 0, 0);
4459                         goto put_file;
4460                 }
4461                 goto out_not_supp;
4462         }
4463
4464         event = perf_get_event(file);
4465         if (!IS_ERR(event)) {
4466                 u64 probe_offset, probe_addr;
4467                 u32 prog_id, fd_type;
4468                 const char *buf;
4469
4470                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4471                                               &buf, &probe_offset,
4472                                               &probe_addr);
4473                 if (!err)
4474                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4475                                                      fd_type, buf,
4476                                                      probe_offset,
4477                                                      probe_addr);
4478                 goto put_file;
4479         }
4480
4481 out_not_supp:
4482         err = -ENOTSUPP;
4483 put_file:
4484         fput(file);
4485         return err;
4486 }
4487
4488 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4489
4490 #define BPF_DO_BATCH(fn, ...)                   \
4491         do {                                    \
4492                 if (!fn) {                      \
4493                         err = -ENOTSUPP;        \
4494                         goto err_put;           \
4495                 }                               \
4496                 err = fn(__VA_ARGS__);          \
4497         } while (0)
4498
4499 static int bpf_map_do_batch(const union bpf_attr *attr,
4500                             union bpf_attr __user *uattr,
4501                             int cmd)
4502 {
4503         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4504                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4505         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4506         struct bpf_map *map;
4507         int err, ufd;
4508         struct fd f;
4509
4510         if (CHECK_ATTR(BPF_MAP_BATCH))
4511                 return -EINVAL;
4512
4513         ufd = attr->batch.map_fd;
4514         f = fdget(ufd);
4515         map = __bpf_map_get(f);
4516         if (IS_ERR(map))
4517                 return PTR_ERR(map);
4518         if (has_write)
4519                 bpf_map_write_active_inc(map);
4520         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4521                 err = -EPERM;
4522                 goto err_put;
4523         }
4524         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4525                 err = -EPERM;
4526                 goto err_put;
4527         }
4528
4529         if (cmd == BPF_MAP_LOOKUP_BATCH)
4530                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4531         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4532                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4533         else if (cmd == BPF_MAP_UPDATE_BATCH)
4534                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4535         else
4536                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4537 err_put:
4538         if (has_write)
4539                 bpf_map_write_active_dec(map);
4540         fdput(f);
4541         return err;
4542 }
4543
4544 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4545 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4546 {
4547         enum bpf_prog_type ptype;
4548         struct bpf_prog *prog;
4549         int ret;
4550
4551         if (CHECK_ATTR(BPF_LINK_CREATE))
4552                 return -EINVAL;
4553
4554         prog = bpf_prog_get(attr->link_create.prog_fd);
4555         if (IS_ERR(prog))
4556                 return PTR_ERR(prog);
4557
4558         ret = bpf_prog_attach_check_attach_type(prog,
4559                                                 attr->link_create.attach_type);
4560         if (ret)
4561                 goto out;
4562
4563         switch (prog->type) {
4564         case BPF_PROG_TYPE_EXT:
4565                 break;
4566         case BPF_PROG_TYPE_PERF_EVENT:
4567         case BPF_PROG_TYPE_TRACEPOINT:
4568                 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4569                         ret = -EINVAL;
4570                         goto out;
4571                 }
4572                 break;
4573         case BPF_PROG_TYPE_KPROBE:
4574                 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4575                     attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4576                         ret = -EINVAL;
4577                         goto out;
4578                 }
4579                 break;
4580         default:
4581                 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4582                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4583                         ret = -EINVAL;
4584                         goto out;
4585                 }
4586                 break;
4587         }
4588
4589         switch (prog->type) {
4590         case BPF_PROG_TYPE_CGROUP_SKB:
4591         case BPF_PROG_TYPE_CGROUP_SOCK:
4592         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4593         case BPF_PROG_TYPE_SOCK_OPS:
4594         case BPF_PROG_TYPE_CGROUP_DEVICE:
4595         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4596         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4597                 ret = cgroup_bpf_link_attach(attr, prog);
4598                 break;
4599         case BPF_PROG_TYPE_EXT:
4600                 ret = bpf_tracing_prog_attach(prog,
4601                                               attr->link_create.target_fd,
4602                                               attr->link_create.target_btf_id,
4603                                               attr->link_create.tracing.cookie);
4604                 break;
4605         case BPF_PROG_TYPE_LSM:
4606         case BPF_PROG_TYPE_TRACING:
4607                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4608                         ret = -EINVAL;
4609                         goto out;
4610                 }
4611                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4612                         ret = bpf_raw_tp_link_attach(prog, NULL);
4613                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4614                         ret = bpf_iter_link_attach(attr, uattr, prog);
4615                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4616                         ret = cgroup_bpf_link_attach(attr, prog);
4617                 else
4618                         ret = bpf_tracing_prog_attach(prog,
4619                                                       attr->link_create.target_fd,
4620                                                       attr->link_create.target_btf_id,
4621                                                       attr->link_create.tracing.cookie);
4622                 break;
4623         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4624         case BPF_PROG_TYPE_SK_LOOKUP:
4625                 ret = netns_bpf_link_create(attr, prog);
4626                 break;
4627 #ifdef CONFIG_NET
4628         case BPF_PROG_TYPE_XDP:
4629                 ret = bpf_xdp_link_attach(attr, prog);
4630                 break;
4631 #endif
4632         case BPF_PROG_TYPE_PERF_EVENT:
4633         case BPF_PROG_TYPE_TRACEPOINT:
4634                 ret = bpf_perf_link_attach(attr, prog);
4635                 break;
4636         case BPF_PROG_TYPE_KPROBE:
4637                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4638                         ret = bpf_perf_link_attach(attr, prog);
4639                 else
4640                         ret = bpf_kprobe_multi_link_attach(attr, prog);
4641                 break;
4642         default:
4643                 ret = -EINVAL;
4644         }
4645
4646 out:
4647         if (ret < 0)
4648                 bpf_prog_put(prog);
4649         return ret;
4650 }
4651
4652 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4653
4654 static int link_update(union bpf_attr *attr)
4655 {
4656         struct bpf_prog *old_prog = NULL, *new_prog;
4657         struct bpf_link *link;
4658         u32 flags;
4659         int ret;
4660
4661         if (CHECK_ATTR(BPF_LINK_UPDATE))
4662                 return -EINVAL;
4663
4664         flags = attr->link_update.flags;
4665         if (flags & ~BPF_F_REPLACE)
4666                 return -EINVAL;
4667
4668         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4669         if (IS_ERR(link))
4670                 return PTR_ERR(link);
4671
4672         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4673         if (IS_ERR(new_prog)) {
4674                 ret = PTR_ERR(new_prog);
4675                 goto out_put_link;
4676         }
4677
4678         if (flags & BPF_F_REPLACE) {
4679                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4680                 if (IS_ERR(old_prog)) {
4681                         ret = PTR_ERR(old_prog);
4682                         old_prog = NULL;
4683                         goto out_put_progs;
4684                 }
4685         } else if (attr->link_update.old_prog_fd) {
4686                 ret = -EINVAL;
4687                 goto out_put_progs;
4688         }
4689
4690         if (link->ops->update_prog)
4691                 ret = link->ops->update_prog(link, new_prog, old_prog);
4692         else
4693                 ret = -EINVAL;
4694
4695 out_put_progs:
4696         if (old_prog)
4697                 bpf_prog_put(old_prog);
4698         if (ret)
4699                 bpf_prog_put(new_prog);
4700 out_put_link:
4701         bpf_link_put(link);
4702         return ret;
4703 }
4704
4705 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4706
4707 static int link_detach(union bpf_attr *attr)
4708 {
4709         struct bpf_link *link;
4710         int ret;
4711
4712         if (CHECK_ATTR(BPF_LINK_DETACH))
4713                 return -EINVAL;
4714
4715         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4716         if (IS_ERR(link))
4717                 return PTR_ERR(link);
4718
4719         if (link->ops->detach)
4720                 ret = link->ops->detach(link);
4721         else
4722                 ret = -EOPNOTSUPP;
4723
4724         bpf_link_put(link);
4725         return ret;
4726 }
4727
4728 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4729 {
4730         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4731 }
4732
4733 struct bpf_link *bpf_link_by_id(u32 id)
4734 {
4735         struct bpf_link *link;
4736
4737         if (!id)
4738                 return ERR_PTR(-ENOENT);
4739
4740         spin_lock_bh(&link_idr_lock);
4741         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4742         link = idr_find(&link_idr, id);
4743         if (link) {
4744                 if (link->id)
4745                         link = bpf_link_inc_not_zero(link);
4746                 else
4747                         link = ERR_PTR(-EAGAIN);
4748         } else {
4749                 link = ERR_PTR(-ENOENT);
4750         }
4751         spin_unlock_bh(&link_idr_lock);
4752         return link;
4753 }
4754
4755 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4756 {
4757         struct bpf_link *link;
4758
4759         spin_lock_bh(&link_idr_lock);
4760 again:
4761         link = idr_get_next(&link_idr, id);
4762         if (link) {
4763                 link = bpf_link_inc_not_zero(link);
4764                 if (IS_ERR(link)) {
4765                         (*id)++;
4766                         goto again;
4767                 }
4768         }
4769         spin_unlock_bh(&link_idr_lock);
4770
4771         return link;
4772 }
4773
4774 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4775
4776 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4777 {
4778         struct bpf_link *link;
4779         u32 id = attr->link_id;
4780         int fd;
4781
4782         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4783                 return -EINVAL;
4784
4785         if (!capable(CAP_SYS_ADMIN))
4786                 return -EPERM;
4787
4788         link = bpf_link_by_id(id);
4789         if (IS_ERR(link))
4790                 return PTR_ERR(link);
4791
4792         fd = bpf_link_new_fd(link);
4793         if (fd < 0)
4794                 bpf_link_put(link);
4795
4796         return fd;
4797 }
4798
4799 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4800
4801 static int bpf_stats_release(struct inode *inode, struct file *file)
4802 {
4803         mutex_lock(&bpf_stats_enabled_mutex);
4804         static_key_slow_dec(&bpf_stats_enabled_key.key);
4805         mutex_unlock(&bpf_stats_enabled_mutex);
4806         return 0;
4807 }
4808
4809 static const struct file_operations bpf_stats_fops = {
4810         .release = bpf_stats_release,
4811 };
4812
4813 static int bpf_enable_runtime_stats(void)
4814 {
4815         int fd;
4816
4817         mutex_lock(&bpf_stats_enabled_mutex);
4818
4819         /* Set a very high limit to avoid overflow */
4820         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4821                 mutex_unlock(&bpf_stats_enabled_mutex);
4822                 return -EBUSY;
4823         }
4824
4825         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4826         if (fd >= 0)
4827                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4828
4829         mutex_unlock(&bpf_stats_enabled_mutex);
4830         return fd;
4831 }
4832
4833 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4834
4835 static int bpf_enable_stats(union bpf_attr *attr)
4836 {
4837
4838         if (CHECK_ATTR(BPF_ENABLE_STATS))
4839                 return -EINVAL;
4840
4841         if (!capable(CAP_SYS_ADMIN))
4842                 return -EPERM;
4843
4844         switch (attr->enable_stats.type) {
4845         case BPF_STATS_RUN_TIME:
4846                 return bpf_enable_runtime_stats();
4847         default:
4848                 break;
4849         }
4850         return -EINVAL;
4851 }
4852
4853 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4854
4855 static int bpf_iter_create(union bpf_attr *attr)
4856 {
4857         struct bpf_link *link;
4858         int err;
4859
4860         if (CHECK_ATTR(BPF_ITER_CREATE))
4861                 return -EINVAL;
4862
4863         if (attr->iter_create.flags)
4864                 return -EINVAL;
4865
4866         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4867         if (IS_ERR(link))
4868                 return PTR_ERR(link);
4869
4870         err = bpf_iter_new_fd(link);
4871         bpf_link_put(link);
4872
4873         return err;
4874 }
4875
4876 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4877
4878 static int bpf_prog_bind_map(union bpf_attr *attr)
4879 {
4880         struct bpf_prog *prog;
4881         struct bpf_map *map;
4882         struct bpf_map **used_maps_old, **used_maps_new;
4883         int i, ret = 0;
4884
4885         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4886                 return -EINVAL;
4887
4888         if (attr->prog_bind_map.flags)
4889                 return -EINVAL;
4890
4891         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4892         if (IS_ERR(prog))
4893                 return PTR_ERR(prog);
4894
4895         map = bpf_map_get(attr->prog_bind_map.map_fd);
4896         if (IS_ERR(map)) {
4897                 ret = PTR_ERR(map);
4898                 goto out_prog_put;
4899         }
4900
4901         mutex_lock(&prog->aux->used_maps_mutex);
4902
4903         used_maps_old = prog->aux->used_maps;
4904
4905         for (i = 0; i < prog->aux->used_map_cnt; i++)
4906                 if (used_maps_old[i] == map) {
4907                         bpf_map_put(map);
4908                         goto out_unlock;
4909                 }
4910
4911         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4912                                       sizeof(used_maps_new[0]),
4913                                       GFP_KERNEL);
4914         if (!used_maps_new) {
4915                 ret = -ENOMEM;
4916                 goto out_unlock;
4917         }
4918
4919         memcpy(used_maps_new, used_maps_old,
4920                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4921         used_maps_new[prog->aux->used_map_cnt] = map;
4922
4923         prog->aux->used_map_cnt++;
4924         prog->aux->used_maps = used_maps_new;
4925
4926         kfree(used_maps_old);
4927
4928 out_unlock:
4929         mutex_unlock(&prog->aux->used_maps_mutex);
4930
4931         if (ret)
4932                 bpf_map_put(map);
4933 out_prog_put:
4934         bpf_prog_put(prog);
4935         return ret;
4936 }
4937
4938 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4939 {
4940         union bpf_attr attr;
4941         bool capable;
4942         int err;
4943
4944         capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
4945
4946         /* Intent here is for unprivileged_bpf_disabled to block key object
4947          * creation commands for unprivileged users; other actions depend
4948          * of fd availability and access to bpffs, so are dependent on
4949          * object creation success.  Capabilities are later verified for
4950          * operations such as load and map create, so even with unprivileged
4951          * BPF disabled, capability checks are still carried out for these
4952          * and other operations.
4953          */
4954         if (!capable &&
4955             (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
4956                 return -EPERM;
4957
4958         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4959         if (err)
4960                 return err;
4961         size = min_t(u32, size, sizeof(attr));
4962
4963         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4964         memset(&attr, 0, sizeof(attr));
4965         if (copy_from_bpfptr(&attr, uattr, size) != 0)
4966                 return -EFAULT;
4967
4968         err = security_bpf(cmd, &attr, size);
4969         if (err < 0)
4970                 return err;
4971
4972         switch (cmd) {
4973         case BPF_MAP_CREATE:
4974                 err = map_create(&attr);
4975                 break;
4976         case BPF_MAP_LOOKUP_ELEM:
4977                 err = map_lookup_elem(&attr);
4978                 break;
4979         case BPF_MAP_UPDATE_ELEM:
4980                 err = map_update_elem(&attr, uattr);
4981                 break;
4982         case BPF_MAP_DELETE_ELEM:
4983                 err = map_delete_elem(&attr, uattr);
4984                 break;
4985         case BPF_MAP_GET_NEXT_KEY:
4986                 err = map_get_next_key(&attr);
4987                 break;
4988         case BPF_MAP_FREEZE:
4989                 err = map_freeze(&attr);
4990                 break;
4991         case BPF_PROG_LOAD:
4992                 err = bpf_prog_load(&attr, uattr);
4993                 break;
4994         case BPF_OBJ_PIN:
4995                 err = bpf_obj_pin(&attr);
4996                 break;
4997         case BPF_OBJ_GET:
4998                 err = bpf_obj_get(&attr);
4999                 break;
5000         case BPF_PROG_ATTACH:
5001                 err = bpf_prog_attach(&attr);
5002                 break;
5003         case BPF_PROG_DETACH:
5004                 err = bpf_prog_detach(&attr);
5005                 break;
5006         case BPF_PROG_QUERY:
5007                 err = bpf_prog_query(&attr, uattr.user);
5008                 break;
5009         case BPF_PROG_TEST_RUN:
5010                 err = bpf_prog_test_run(&attr, uattr.user);
5011                 break;
5012         case BPF_PROG_GET_NEXT_ID:
5013                 err = bpf_obj_get_next_id(&attr, uattr.user,
5014                                           &prog_idr, &prog_idr_lock);
5015                 break;
5016         case BPF_MAP_GET_NEXT_ID:
5017                 err = bpf_obj_get_next_id(&attr, uattr.user,
5018                                           &map_idr, &map_idr_lock);
5019                 break;
5020         case BPF_BTF_GET_NEXT_ID:
5021                 err = bpf_obj_get_next_id(&attr, uattr.user,
5022                                           &btf_idr, &btf_idr_lock);
5023                 break;
5024         case BPF_PROG_GET_FD_BY_ID:
5025                 err = bpf_prog_get_fd_by_id(&attr);
5026                 break;
5027         case BPF_MAP_GET_FD_BY_ID:
5028                 err = bpf_map_get_fd_by_id(&attr);
5029                 break;
5030         case BPF_OBJ_GET_INFO_BY_FD:
5031                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5032                 break;
5033         case BPF_RAW_TRACEPOINT_OPEN:
5034                 err = bpf_raw_tracepoint_open(&attr);
5035                 break;
5036         case BPF_BTF_LOAD:
5037                 err = bpf_btf_load(&attr, uattr);
5038                 break;
5039         case BPF_BTF_GET_FD_BY_ID:
5040                 err = bpf_btf_get_fd_by_id(&attr);
5041                 break;
5042         case BPF_TASK_FD_QUERY:
5043                 err = bpf_task_fd_query(&attr, uattr.user);
5044                 break;
5045         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5046                 err = map_lookup_and_delete_elem(&attr);
5047                 break;
5048         case BPF_MAP_LOOKUP_BATCH:
5049                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5050                 break;
5051         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5052                 err = bpf_map_do_batch(&attr, uattr.user,
5053                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5054                 break;
5055         case BPF_MAP_UPDATE_BATCH:
5056                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5057                 break;
5058         case BPF_MAP_DELETE_BATCH:
5059                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5060                 break;
5061         case BPF_LINK_CREATE:
5062                 err = link_create(&attr, uattr);
5063                 break;
5064         case BPF_LINK_UPDATE:
5065                 err = link_update(&attr);
5066                 break;
5067         case BPF_LINK_GET_FD_BY_ID:
5068                 err = bpf_link_get_fd_by_id(&attr);
5069                 break;
5070         case BPF_LINK_GET_NEXT_ID:
5071                 err = bpf_obj_get_next_id(&attr, uattr.user,
5072                                           &link_idr, &link_idr_lock);
5073                 break;
5074         case BPF_ENABLE_STATS:
5075                 err = bpf_enable_stats(&attr);
5076                 break;
5077         case BPF_ITER_CREATE:
5078                 err = bpf_iter_create(&attr);
5079                 break;
5080         case BPF_LINK_DETACH:
5081                 err = link_detach(&attr);
5082                 break;
5083         case BPF_PROG_BIND_MAP:
5084                 err = bpf_prog_bind_map(&attr);
5085                 break;
5086         default:
5087                 err = -EINVAL;
5088                 break;
5089         }
5090
5091         return err;
5092 }
5093
5094 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5095 {
5096         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5097 }
5098
5099 static bool syscall_prog_is_valid_access(int off, int size,
5100                                          enum bpf_access_type type,
5101                                          const struct bpf_prog *prog,
5102                                          struct bpf_insn_access_aux *info)
5103 {
5104         if (off < 0 || off >= U16_MAX)
5105                 return false;
5106         if (off % size != 0)
5107                 return false;
5108         return true;
5109 }
5110
5111 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5112 {
5113         switch (cmd) {
5114         case BPF_MAP_CREATE:
5115         case BPF_MAP_DELETE_ELEM:
5116         case BPF_MAP_UPDATE_ELEM:
5117         case BPF_MAP_FREEZE:
5118         case BPF_MAP_GET_FD_BY_ID:
5119         case BPF_PROG_LOAD:
5120         case BPF_BTF_LOAD:
5121         case BPF_LINK_CREATE:
5122         case BPF_RAW_TRACEPOINT_OPEN:
5123                 break;
5124         default:
5125                 return -EINVAL;
5126         }
5127         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5128 }
5129
5130
5131 /* To shut up -Wmissing-prototypes.
5132  * This function is used by the kernel light skeleton
5133  * to load bpf programs when modules are loaded or during kernel boot.
5134  * See tools/lib/bpf/skel_internal.h
5135  */
5136 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5137
5138 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5139 {
5140         struct bpf_prog * __maybe_unused prog;
5141         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5142
5143         switch (cmd) {
5144 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5145         case BPF_PROG_TEST_RUN:
5146                 if (attr->test.data_in || attr->test.data_out ||
5147                     attr->test.ctx_out || attr->test.duration ||
5148                     attr->test.repeat || attr->test.flags)
5149                         return -EINVAL;
5150
5151                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5152                 if (IS_ERR(prog))
5153                         return PTR_ERR(prog);
5154
5155                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5156                     attr->test.ctx_size_in > U16_MAX) {
5157                         bpf_prog_put(prog);
5158                         return -EINVAL;
5159                 }
5160
5161                 run_ctx.bpf_cookie = 0;
5162                 run_ctx.saved_run_ctx = NULL;
5163                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5164                         /* recursion detected */
5165                         bpf_prog_put(prog);
5166                         return -EBUSY;
5167                 }
5168                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5169                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5170                                                 &run_ctx);
5171                 bpf_prog_put(prog);
5172                 return 0;
5173 #endif
5174         default:
5175                 return ____bpf_sys_bpf(cmd, attr, size);
5176         }
5177 }
5178 EXPORT_SYMBOL(kern_sys_bpf);
5179
5180 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5181         .func           = bpf_sys_bpf,
5182         .gpl_only       = false,
5183         .ret_type       = RET_INTEGER,
5184         .arg1_type      = ARG_ANYTHING,
5185         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5186         .arg3_type      = ARG_CONST_SIZE,
5187 };
5188
5189 const struct bpf_func_proto * __weak
5190 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5191 {
5192         return bpf_base_func_proto(func_id);
5193 }
5194
5195 BPF_CALL_1(bpf_sys_close, u32, fd)
5196 {
5197         /* When bpf program calls this helper there should not be
5198          * an fdget() without matching completed fdput().
5199          * This helper is allowed in the following callchain only:
5200          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5201          */
5202         return close_fd(fd);
5203 }
5204
5205 static const struct bpf_func_proto bpf_sys_close_proto = {
5206         .func           = bpf_sys_close,
5207         .gpl_only       = false,
5208         .ret_type       = RET_INTEGER,
5209         .arg1_type      = ARG_ANYTHING,
5210 };
5211
5212 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5213 {
5214         if (flags)
5215                 return -EINVAL;
5216
5217         if (name_sz <= 1 || name[name_sz - 1])
5218                 return -EINVAL;
5219
5220         if (!bpf_dump_raw_ok(current_cred()))
5221                 return -EPERM;
5222
5223         *res = kallsyms_lookup_name(name);
5224         return *res ? 0 : -ENOENT;
5225 }
5226
5227 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5228         .func           = bpf_kallsyms_lookup_name,
5229         .gpl_only       = false,
5230         .ret_type       = RET_INTEGER,
5231         .arg1_type      = ARG_PTR_TO_MEM,
5232         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5233         .arg3_type      = ARG_ANYTHING,
5234         .arg4_type      = ARG_PTR_TO_LONG,
5235 };
5236
5237 static const struct bpf_func_proto *
5238 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5239 {
5240         switch (func_id) {
5241         case BPF_FUNC_sys_bpf:
5242                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5243         case BPF_FUNC_btf_find_by_name_kind:
5244                 return &bpf_btf_find_by_name_kind_proto;
5245         case BPF_FUNC_sys_close:
5246                 return &bpf_sys_close_proto;
5247         case BPF_FUNC_kallsyms_lookup_name:
5248                 return &bpf_kallsyms_lookup_name_proto;
5249         default:
5250                 return tracing_prog_func_proto(func_id, prog);
5251         }
5252 }
5253
5254 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5255         .get_func_proto  = syscall_prog_func_proto,
5256         .is_valid_access = syscall_prog_is_valid_access,
5257 };
5258
5259 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5260         .test_run = bpf_prog_test_run_syscall,
5261 };
5262
5263 #ifdef CONFIG_SYSCTL
5264 static int bpf_stats_handler(struct ctl_table *table, int write,
5265                              void *buffer, size_t *lenp, loff_t *ppos)
5266 {
5267         struct static_key *key = (struct static_key *)table->data;
5268         static int saved_val;
5269         int val, ret;
5270         struct ctl_table tmp = {
5271                 .data   = &val,
5272                 .maxlen = sizeof(val),
5273                 .mode   = table->mode,
5274                 .extra1 = SYSCTL_ZERO,
5275                 .extra2 = SYSCTL_ONE,
5276         };
5277
5278         if (write && !capable(CAP_SYS_ADMIN))
5279                 return -EPERM;
5280
5281         mutex_lock(&bpf_stats_enabled_mutex);
5282         val = saved_val;
5283         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5284         if (write && !ret && val != saved_val) {
5285                 if (val)
5286                         static_key_slow_inc(key);
5287                 else
5288                         static_key_slow_dec(key);
5289                 saved_val = val;
5290         }
5291         mutex_unlock(&bpf_stats_enabled_mutex);
5292         return ret;
5293 }
5294
5295 void __weak unpriv_ebpf_notify(int new_state)
5296 {
5297 }
5298
5299 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5300                               void *buffer, size_t *lenp, loff_t *ppos)
5301 {
5302         int ret, unpriv_enable = *(int *)table->data;
5303         bool locked_state = unpriv_enable == 1;
5304         struct ctl_table tmp = *table;
5305
5306         if (write && !capable(CAP_SYS_ADMIN))
5307                 return -EPERM;
5308
5309         tmp.data = &unpriv_enable;
5310         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5311         if (write && !ret) {
5312                 if (locked_state && unpriv_enable != 1)
5313                         return -EPERM;
5314                 *(int *)table->data = unpriv_enable;
5315         }
5316
5317         unpriv_ebpf_notify(unpriv_enable);
5318
5319         return ret;
5320 }
5321
5322 static struct ctl_table bpf_syscall_table[] = {
5323         {
5324                 .procname       = "unprivileged_bpf_disabled",
5325                 .data           = &sysctl_unprivileged_bpf_disabled,
5326                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5327                 .mode           = 0644,
5328                 .proc_handler   = bpf_unpriv_handler,
5329                 .extra1         = SYSCTL_ZERO,
5330                 .extra2         = SYSCTL_TWO,
5331         },
5332         {
5333                 .procname       = "bpf_stats_enabled",
5334                 .data           = &bpf_stats_enabled_key.key,
5335                 .mode           = 0644,
5336                 .proc_handler   = bpf_stats_handler,
5337         },
5338         { }
5339 };
5340
5341 static int __init bpf_syscall_sysctl_init(void)
5342 {
5343         register_sysctl_init("kernel", bpf_syscall_table);
5344         return 0;
5345 }
5346 late_initcall(bpf_syscall_sysctl_init);
5347 #endif /* CONFIG_SYSCTL */