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 #include <net/netfilter/nf_bpf_link.h>
  40 #include <net/netkit.h>
  41 #include <net/tcx.h>
  42
  43 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  44                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  45                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  46 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  47 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  48 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  49                         IS_FD_HASH(map))
  50
  51 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  52
  53 DEFINE_PER_CPU(int, bpf_prog_active);
  54 static DEFINE_IDR(prog_idr);
  55 static DEFINE_SPINLOCK(prog_idr_lock);
  56 static DEFINE_IDR(map_idr);
  57 static DEFINE_SPINLOCK(map_idr_lock);
  58 static DEFINE_IDR(link_idr);
  59 static DEFINE_SPINLOCK(link_idr_lock);
  60
  61 int sysctl_unprivileged_bpf_disabled __read_mostly =
  62         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  63
  64 static const struct bpf_map_ops * const bpf_map_types[] = {
  65 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  66 #define BPF_MAP_TYPE(_id, _ops) \
  67         [_id] = &_ops,
  68 #define BPF_LINK_TYPE(_id, _name)
  69 #include <linux/bpf_types.h>
  70 #undef BPF_PROG_TYPE
  71 #undef BPF_MAP_TYPE
  72 #undef BPF_LINK_TYPE
  73 };
  74
  75 /*
  76  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  77  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  78  * we don't know about yet.
  79  *
  80  * There is a ToCToU between this function call and the following
  81  * copy_from_user() call. However, this is not a concern since this function is
  82  * meant to be a future-proofing of bits.
  83  */
  84 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  85                              size_t expected_size,
  86                              size_t actual_size)
  87 {
  88         int res;
  89
  90         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  91                 return -E2BIG;
  92
  93         if (actual_size <= expected_size)
  94                 return 0;
  95
  96         if (uaddr.is_kernel)
  97                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  98                                  actual_size - expected_size) == NULL;
  99         else
 100                 res = check_zeroed_user(uaddr.user + expected_size,
 101                                         actual_size - expected_size);
 102         if (res < 0)
 103                 return res;
 104         return res ? 0 : -E2BIG;
 105 }
 106
 107 const struct bpf_map_ops bpf_map_offload_ops = {
 108         .map_meta_equal = bpf_map_meta_equal,
 109         .map_alloc = bpf_map_offload_map_alloc,
 110         .map_free = bpf_map_offload_map_free,
 111         .map_check_btf = map_check_no_btf,
 112         .map_mem_usage = bpf_map_offload_map_mem_usage,
 113 };
 114
 115 static void bpf_map_write_active_inc(struct bpf_map *map)
 116 {
 117         atomic64_inc(&map->writecnt);
 118 }
 119
 120 static void bpf_map_write_active_dec(struct bpf_map *map)
 121 {
 122         atomic64_dec(&map->writecnt);
 123 }
 124
 125 bool bpf_map_write_active(const struct bpf_map *map)
 126 {
 127         return atomic64_read(&map->writecnt) != 0;
 128 }
 129
 130 static u32 bpf_map_value_size(const struct bpf_map *map)
 131 {
 132         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 133             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 134             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 135             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 136                 return round_up(map->value_size, 8) * num_possible_cpus();
 137         else if (IS_FD_MAP(map))
 138                 return sizeof(u32);
 139         else
 140                 return  map->value_size;
 141 }
 142
 143 static void maybe_wait_bpf_programs(struct bpf_map *map)
 144 {
 145         /* Wait for any running BPF programs to complete so that
 146          * userspace, when we return to it, knows that all programs
 147          * that could be running use the new map value.
 148          */
 149         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 150             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 151                 synchronize_rcu();
 152 }
 153
 154 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 155                                 void *key, void *value, __u64 flags)
 156 {
 157         int err;
 158
 159         /* Need to create a kthread, thus must support schedule */
 160         if (bpf_map_is_offloaded(map)) {
 161                 return bpf_map_offload_update_elem(map, key, value, flags);
 162         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 163                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 164                 return map->ops->map_update_elem(map, key, value, flags);
 165         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 166                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 167                 return sock_map_update_elem_sys(map, key, value, flags);
 168         } else if (IS_FD_PROG_ARRAY(map)) {
 169                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 170                                                     flags);
 171         }
 172
 173         bpf_disable_instrumentation();
 174         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 175             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 176                 err = bpf_percpu_hash_update(map, key, value, flags);
 177         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 178                 err = bpf_percpu_array_update(map, key, value, flags);
 179         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 180                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 181                                                        flags);
 182         } else if (IS_FD_ARRAY(map)) {
 183                 rcu_read_lock();
 184                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 185                                                    flags);
 186                 rcu_read_unlock();
 187         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 188                 rcu_read_lock();
 189                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 190                                                   flags);
 191                 rcu_read_unlock();
 192         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 193                 /* rcu_read_lock() is not needed */
 194                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 195                                                          flags);
 196         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 197                    map->map_type == BPF_MAP_TYPE_STACK ||
 198                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 199                 err = map->ops->map_push_elem(map, value, flags);
 200         } else {
 201                 rcu_read_lock();
 202                 err = map->ops->map_update_elem(map, key, value, flags);
 203                 rcu_read_unlock();
 204         }
 205         bpf_enable_instrumentation();
 206         maybe_wait_bpf_programs(map);
 207
 208         return err;
 209 }
 210
 211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 212                               __u64 flags)
 213 {
 214         void *ptr;
 215         int err;
 216
 217         if (bpf_map_is_offloaded(map))
 218                 return bpf_map_offload_lookup_elem(map, key, value);
 219
 220         bpf_disable_instrumentation();
 221         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 222             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 223                 err = bpf_percpu_hash_copy(map, key, value);
 224         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 225                 err = bpf_percpu_array_copy(map, key, value);
 226         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 227                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 228         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 229                 err = bpf_stackmap_copy(map, key, value);
 230         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 231                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 232         } else if (IS_FD_HASH(map)) {
 233                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 234         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 235                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 236         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 237                    map->map_type == BPF_MAP_TYPE_STACK ||
 238                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 239                 err = map->ops->map_peek_elem(map, value);
 240         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 241                 /* struct_ops map requires directly updating "value" */
 242                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 243         } else {
 244                 rcu_read_lock();
 245                 if (map->ops->map_lookup_elem_sys_only)
 246                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 247                 else
 248                         ptr = map->ops->map_lookup_elem(map, key);
 249                 if (IS_ERR(ptr)) {
 250                         err = PTR_ERR(ptr);
 251                 } else if (!ptr) {
 252                         err = -ENOENT;
 253                 } else {
 254                         err = 0;
 255                         if (flags & BPF_F_LOCK)
 256                                 /* lock 'ptr' and copy everything but lock */
 257                                 copy_map_value_locked(map, value, ptr, true);
 258                         else
 259                                 copy_map_value(map, value, ptr);
 260                         /* mask lock and timer, since value wasn't zero inited */
 261                         check_and_init_map_value(map, value);
 262                 }
 263                 rcu_read_unlock();
 264         }
 265
 266         bpf_enable_instrumentation();
 267         maybe_wait_bpf_programs(map);
 268
 269         return err;
 270 }
 271
 272 /* Please, do not use this function outside from the map creation path
 273  * (e.g. in map update path) without taking care of setting the active
 274  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 275  */
 276 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 277 {
 278         /* We really just want to fail instead of triggering OOM killer
 279          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 280          * which is used for lower order allocation requests.
 281          *
 282          * It has been observed that higher order allocation requests done by
 283          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 284          * to reclaim memory from the page cache, thus we set
 285          * __GFP_RETRY_MAYFAIL to avoid such situations.
 286          */
 287
 288         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 289         unsigned int flags = 0;
 290         unsigned long align = 1;
 291         void *area;
 292
 293         if (size >= SIZE_MAX)
 294                 return NULL;
 295
 296         /* kmalloc()'ed memory can't be mmap()'ed */
 297         if (mmapable) {
 298                 BUG_ON(!PAGE_ALIGNED(size));
 299                 align = SHMLBA;
 300                 flags = VM_USERMAP;
 301         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 302                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 303                                     numa_node);
 304                 if (area != NULL)
 305                         return area;
 306         }
 307
 308         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 309                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 310                         flags, numa_node, __builtin_return_address(0));
 311 }
 312
 313 void *bpf_map_area_alloc(u64 size, int numa_node)
 314 {
 315         return __bpf_map_area_alloc(size, numa_node, false);
 316 }
 317
 318 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 319 {
 320         return __bpf_map_area_alloc(size, numa_node, true);
 321 }
 322
 323 void bpf_map_area_free(void *area)
 324 {
 325         kvfree(area);
 326 }
 327
 328 static u32 bpf_map_flags_retain_permanent(u32 flags)
 329 {
 330         /* Some map creation flags are not tied to the map object but
 331          * rather to the map fd instead, so they have no meaning upon
 332          * map object inspection since multiple file descriptors with
 333          * different (access) properties can exist here. Thus, given
 334          * this has zero meaning for the map itself, lets clear these
 335          * from here.
 336          */
 337         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 338 }
 339
 340 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 341 {
 342         map->map_type = attr->map_type;
 343         map->key_size = attr->key_size;
 344         map->value_size = attr->value_size;
 345         map->max_entries = attr->max_entries;
 346         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 347         map->numa_node = bpf_map_attr_numa_node(attr);
 348         map->map_extra = attr->map_extra;
 349 }
 350
 351 static int bpf_map_alloc_id(struct bpf_map *map)
 352 {
 353         int id;
 354
 355         idr_preload(GFP_KERNEL);
 356         spin_lock_bh(&map_idr_lock);
 357         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 358         if (id > 0)
 359                 map->id = id;
 360         spin_unlock_bh(&map_idr_lock);
 361         idr_preload_end();
 362
 363         if (WARN_ON_ONCE(!id))
 364                 return -ENOSPC;
 365
 366         return id > 0 ? 0 : id;
 367 }
 368
 369 void bpf_map_free_id(struct bpf_map *map)
 370 {
 371         unsigned long flags;
 372
 373         /* Offloaded maps are removed from the IDR store when their device
 374          * disappears - even if someone holds an fd to them they are unusable,
 375          * the memory is gone, all ops will fail; they are simply waiting for
 376          * refcnt to drop to be freed.
 377          */
 378         if (!map->id)
 379                 return;
 380
 381         spin_lock_irqsave(&map_idr_lock, flags);
 382
 383         idr_remove(&map_idr, map->id);
 384         map->id = 0;
 385
 386         spin_unlock_irqrestore(&map_idr_lock, flags);
 387 }
 388
 389 #ifdef CONFIG_MEMCG_KMEM
 390 static void bpf_map_save_memcg(struct bpf_map *map)
 391 {
 392         /* Currently if a map is created by a process belonging to the root
 393          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 394          * So we have to check map->objcg for being NULL each time it's
 395          * being used.
 396          */
 397         if (memcg_bpf_enabled())
 398                 map->objcg = get_obj_cgroup_from_current();
 399 }
 400
 401 static void bpf_map_release_memcg(struct bpf_map *map)
 402 {
 403         if (map->objcg)
 404                 obj_cgroup_put(map->objcg);
 405 }
 406
 407 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 408 {
 409         if (map->objcg)
 410                 return get_mem_cgroup_from_objcg(map->objcg);
 411
 412         return root_mem_cgroup;
 413 }
 414
 415 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 416                            int node)
 417 {
 418         struct mem_cgroup *memcg, *old_memcg;
 419         void *ptr;
 420
 421         memcg = bpf_map_get_memcg(map);
 422         old_memcg = set_active_memcg(memcg);
 423         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 424         set_active_memcg(old_memcg);
 425         mem_cgroup_put(memcg);
 426
 427         return ptr;
 428 }
 429
 430 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 431 {
 432         struct mem_cgroup *memcg, *old_memcg;
 433         void *ptr;
 434
 435         memcg = bpf_map_get_memcg(map);
 436         old_memcg = set_active_memcg(memcg);
 437         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 438         set_active_memcg(old_memcg);
 439         mem_cgroup_put(memcg);
 440
 441         return ptr;
 442 }
 443
 444 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 445                        gfp_t flags)
 446 {
 447         struct mem_cgroup *memcg, *old_memcg;
 448         void *ptr;
 449
 450         memcg = bpf_map_get_memcg(map);
 451         old_memcg = set_active_memcg(memcg);
 452         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 453         set_active_memcg(old_memcg);
 454         mem_cgroup_put(memcg);
 455
 456         return ptr;
 457 }
 458
 459 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 460                                     size_t align, gfp_t flags)
 461 {
 462         struct mem_cgroup *memcg, *old_memcg;
 463         void __percpu *ptr;
 464
 465         memcg = bpf_map_get_memcg(map);
 466         old_memcg = set_active_memcg(memcg);
 467         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 468         set_active_memcg(old_memcg);
 469         mem_cgroup_put(memcg);
 470
 471         return ptr;
 472 }
 473
 474 #else
 475 static void bpf_map_save_memcg(struct bpf_map *map)
 476 {
 477 }
 478
 479 static void bpf_map_release_memcg(struct bpf_map *map)
 480 {
 481 }
 482 #endif
 483
 484 static int btf_field_cmp(const void *a, const void *b)
 485 {
 486         const struct btf_field *f1 = a, *f2 = b;
 487
 488         if (f1->offset < f2->offset)
 489                 return -1;
 490         else if (f1->offset > f2->offset)
 491                 return 1;
 492         return 0;
 493 }
 494
 495 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 496                                   u32 field_mask)
 497 {
 498         struct btf_field *field;
 499
 500         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 501                 return NULL;
 502         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 503         if (!field || !(field->type & field_mask))
 504                 return NULL;
 505         return field;
 506 }
 507
 508 void btf_record_free(struct btf_record *rec)
 509 {
 510         int i;
 511
 512         if (IS_ERR_OR_NULL(rec))
 513                 return;
 514         for (i = 0; i < rec->cnt; i++) {
 515                 switch (rec->fields[i].type) {
 516                 case BPF_KPTR_UNREF:
 517                 case BPF_KPTR_REF:
 518                 case BPF_KPTR_PERCPU:
 519                         if (rec->fields[i].kptr.module)
 520                                 module_put(rec->fields[i].kptr.module);
 521                         btf_put(rec->fields[i].kptr.btf);
 522                         break;
 523                 case BPF_LIST_HEAD:
 524                 case BPF_LIST_NODE:
 525                 case BPF_RB_ROOT:
 526                 case BPF_RB_NODE:
 527                 case BPF_SPIN_LOCK:
 528                 case BPF_TIMER:
 529                 case BPF_REFCOUNT:
 530                         /* Nothing to release */
 531                         break;
 532                 default:
 533                         WARN_ON_ONCE(1);
 534                         continue;
 535                 }
 536         }
 537         kfree(rec);
 538 }
 539
 540 void bpf_map_free_record(struct bpf_map *map)
 541 {
 542         btf_record_free(map->record);
 543         map->record = NULL;
 544 }
 545
 546 struct btf_record *btf_record_dup(const struct btf_record *rec)
 547 {
 548         const struct btf_field *fields;
 549         struct btf_record *new_rec;
 550         int ret, size, i;
 551
 552         if (IS_ERR_OR_NULL(rec))
 553                 return NULL;
 554         size = offsetof(struct btf_record, fields[rec->cnt]);
 555         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 556         if (!new_rec)
 557                 return ERR_PTR(-ENOMEM);
 558         /* Do a deep copy of the btf_record */
 559         fields = rec->fields;
 560         new_rec->cnt = 0;
 561         for (i = 0; i < rec->cnt; i++) {
 562                 switch (fields[i].type) {
 563                 case BPF_KPTR_UNREF:
 564                 case BPF_KPTR_REF:
 565                 case BPF_KPTR_PERCPU:
 566                         btf_get(fields[i].kptr.btf);
 567                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 568                                 ret = -ENXIO;
 569                                 goto free;
 570                         }
 571                         break;
 572                 case BPF_LIST_HEAD:
 573                 case BPF_LIST_NODE:
 574                 case BPF_RB_ROOT:
 575                 case BPF_RB_NODE:
 576                 case BPF_SPIN_LOCK:
 577                 case BPF_TIMER:
 578                 case BPF_REFCOUNT:
 579                         /* Nothing to acquire */
 580                         break;
 581                 default:
 582                         ret = -EFAULT;
 583                         WARN_ON_ONCE(1);
 584                         goto free;
 585                 }
 586                 new_rec->cnt++;
 587         }
 588         return new_rec;
 589 free:
 590         btf_record_free(new_rec);
 591         return ERR_PTR(ret);
 592 }
 593
 594 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 595 {
 596         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 597         int size;
 598
 599         if (!a_has_fields && !b_has_fields)
 600                 return true;
 601         if (a_has_fields != b_has_fields)
 602                 return false;
 603         if (rec_a->cnt != rec_b->cnt)
 604                 return false;
 605         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 606         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 607          * members are zeroed out. So memcmp is safe to do without worrying
 608          * about padding/unused fields.
 609          *
 610          * While spin_lock, timer, and kptr have no relation to map BTF,
 611          * list_head metadata is specific to map BTF, the btf and value_rec
 612          * members in particular. btf is the map BTF, while value_rec points to
 613          * btf_record in that map BTF.
 614          *
 615          * So while by default, we don't rely on the map BTF (which the records
 616          * were parsed from) matching for both records, which is not backwards
 617          * compatible, in case list_head is part of it, we implicitly rely on
 618          * that by way of depending on memcmp succeeding for it.
 619          */
 620         return !memcmp(rec_a, rec_b, size);
 621 }
 622
 623 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 624 {
 625         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 626                 return;
 627         bpf_timer_cancel_and_free(obj + rec->timer_off);
 628 }
 629
 630 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 631 {
 632         const struct btf_field *fields;
 633         int i;
 634
 635         if (IS_ERR_OR_NULL(rec))
 636                 return;
 637         fields = rec->fields;
 638         for (i = 0; i < rec->cnt; i++) {
 639                 struct btf_struct_meta *pointee_struct_meta;
 640                 const struct btf_field *field = &fields[i];
 641                 void *field_ptr = obj + field->offset;
 642                 void *xchgd_field;
 643
 644                 switch (fields[i].type) {
 645                 case BPF_SPIN_LOCK:
 646                         break;
 647                 case BPF_TIMER:
 648                         bpf_timer_cancel_and_free(field_ptr);
 649                         break;
 650                 case BPF_KPTR_UNREF:
 651                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 652                         break;
 653                 case BPF_KPTR_REF:
 654                 case BPF_KPTR_PERCPU:
 655                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 656                         if (!xchgd_field)
 657                                 break;
 658
 659                         if (!btf_is_kernel(field->kptr.btf)) {
 660                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 661                                                                            field->kptr.btf_id);
 662                                 migrate_disable();
 663                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 664                                                                  pointee_struct_meta->record : NULL,
 665                                                                  fields[i].type == BPF_KPTR_PERCPU);
 666                                 migrate_enable();
 667                         } else {
 668                                 field->kptr.dtor(xchgd_field);
 669                         }
 670                         break;
 671                 case BPF_LIST_HEAD:
 672                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 673                                 continue;
 674                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 675                         break;
 676                 case BPF_RB_ROOT:
 677                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 678                                 continue;
 679                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 680                         break;
 681                 case BPF_LIST_NODE:
 682                 case BPF_RB_NODE:
 683                 case BPF_REFCOUNT:
 684                         break;
 685                 default:
 686                         WARN_ON_ONCE(1);
 687                         continue;
 688                 }
 689         }
 690 }
 691
 692 /* called from workqueue */
 693 static void bpf_map_free_deferred(struct work_struct *work)
 694 {
 695         struct bpf_map *map = container_of(work, struct bpf_map, work);
 696         struct btf_record *rec = map->record;
 697
 698         security_bpf_map_free(map);
 699         bpf_map_release_memcg(map);
 700         /* implementation dependent freeing */
 701         map->ops->map_free(map);
 702         /* Delay freeing of btf_record for maps, as map_free
 703          * callback usually needs access to them. It is better to do it here
 704          * than require each callback to do the free itself manually.
 705          *
 706          * Note that the btf_record stashed in map->inner_map_meta->record was
 707          * already freed using the map_free callback for map in map case which
 708          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 709          * template bpf_map struct used during verification.
 710          */
 711         btf_record_free(rec);
 712 }
 713
 714 static void bpf_map_put_uref(struct bpf_map *map)
 715 {
 716         if (atomic64_dec_and_test(&map->usercnt)) {
 717                 if (map->ops->map_release_uref)
 718                         map->ops->map_release_uref(map);
 719         }
 720 }
 721
 722 /* decrement map refcnt and schedule it for freeing via workqueue
 723  * (underlying map implementation ops->map_free() might sleep)
 724  */
 725 void bpf_map_put(struct bpf_map *map)
 726 {
 727         if (atomic64_dec_and_test(&map->refcnt)) {
 728                 /* bpf_map_free_id() must be called first */
 729                 bpf_map_free_id(map);
 730                 btf_put(map->btf);
 731                 INIT_WORK(&map->work, bpf_map_free_deferred);
 732                 /* Avoid spawning kworkers, since they all might contend
 733                  * for the same mutex like slab_mutex.
 734                  */
 735                 queue_work(system_unbound_wq, &map->work);
 736         }
 737 }
 738 EXPORT_SYMBOL_GPL(bpf_map_put);
 739
 740 void bpf_map_put_with_uref(struct bpf_map *map)
 741 {
 742         bpf_map_put_uref(map);
 743         bpf_map_put(map);
 744 }
 745
 746 static int bpf_map_release(struct inode *inode, struct file *filp)
 747 {
 748         struct bpf_map *map = filp->private_data;
 749
 750         if (map->ops->map_release)
 751                 map->ops->map_release(map, filp);
 752
 753         bpf_map_put_with_uref(map);
 754         return 0;
 755 }
 756
 757 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 758 {
 759         fmode_t mode = f.file->f_mode;
 760
 761         /* Our file permissions may have been overridden by global
 762          * map permissions facing syscall side.
 763          */
 764         if (READ_ONCE(map->frozen))
 765                 mode &= ~FMODE_CAN_WRITE;
 766         return mode;
 767 }
 768
 769 #ifdef CONFIG_PROC_FS
 770 /* Show the memory usage of a bpf map */
 771 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 772 {
 773         return map->ops->map_mem_usage(map);
 774 }
 775
 776 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 777 {
 778         struct bpf_map *map = filp->private_data;
 779         u32 type = 0, jited = 0;
 780
 781         if (map_type_contains_progs(map)) {
 782                 spin_lock(&map->owner.lock);
 783                 type  = map->owner.type;
 784                 jited = map->owner.jited;
 785                 spin_unlock(&map->owner.lock);
 786         }
 787
 788         seq_printf(m,
 789                    "map_type:\t%u\n"
 790                    "key_size:\t%u\n"
 791                    "value_size:\t%u\n"
 792                    "max_entries:\t%u\n"
 793                    "map_flags:\t%#x\n"
 794                    "map_extra:\t%#llx\n"
 795                    "memlock:\t%llu\n"
 796                    "map_id:\t%u\n"
 797                    "frozen:\t%u\n",
 798                    map->map_type,
 799                    map->key_size,
 800                    map->value_size,
 801                    map->max_entries,
 802                    map->map_flags,
 803                    (unsigned long long)map->map_extra,
 804                    bpf_map_memory_usage(map),
 805                    map->id,
 806                    READ_ONCE(map->frozen));
 807         if (type) {
 808                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 809                 seq_printf(m, "owner_jited:\t%u\n", jited);
 810         }
 811 }
 812 #endif
 813
 814 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 815                               loff_t *ppos)
 816 {
 817         /* We need this handler such that alloc_file() enables
 818          * f_mode with FMODE_CAN_READ.
 819          */
 820         return -EINVAL;
 821 }
 822
 823 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 824                                size_t siz, loff_t *ppos)
 825 {
 826         /* We need this handler such that alloc_file() enables
 827          * f_mode with FMODE_CAN_WRITE.
 828          */
 829         return -EINVAL;
 830 }
 831
 832 /* called for any extra memory-mapped regions (except initial) */
 833 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 834 {
 835         struct bpf_map *map = vma->vm_file->private_data;
 836
 837         if (vma->vm_flags & VM_MAYWRITE)
 838                 bpf_map_write_active_inc(map);
 839 }
 840
 841 /* called for all unmapped memory region (including initial) */
 842 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 843 {
 844         struct bpf_map *map = vma->vm_file->private_data;
 845
 846         if (vma->vm_flags & VM_MAYWRITE)
 847                 bpf_map_write_active_dec(map);
 848 }
 849
 850 static const struct vm_operations_struct bpf_map_default_vmops = {
 851         .open           = bpf_map_mmap_open,
 852         .close          = bpf_map_mmap_close,
 853 };
 854
 855 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 856 {
 857         struct bpf_map *map = filp->private_data;
 858         int err;
 859
 860         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 861                 return -ENOTSUPP;
 862
 863         if (!(vma->vm_flags & VM_SHARED))
 864                 return -EINVAL;
 865
 866         mutex_lock(&map->freeze_mutex);
 867
 868         if (vma->vm_flags & VM_WRITE) {
 869                 if (map->frozen) {
 870                         err = -EPERM;
 871                         goto out;
 872                 }
 873                 /* map is meant to be read-only, so do not allow mapping as
 874                  * writable, because it's possible to leak a writable page
 875                  * reference and allows user-space to still modify it after
 876                  * freezing, while verifier will assume contents do not change
 877                  */
 878                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 879                         err = -EACCES;
 880                         goto out;
 881                 }
 882         }
 883
 884         /* set default open/close callbacks */
 885         vma->vm_ops = &bpf_map_default_vmops;
 886         vma->vm_private_data = map;
 887         vm_flags_clear(vma, VM_MAYEXEC);
 888         if (!(vma->vm_flags & VM_WRITE))
 889                 /* disallow re-mapping with PROT_WRITE */
 890                 vm_flags_clear(vma, VM_MAYWRITE);
 891
 892         err = map->ops->map_mmap(map, vma);
 893         if (err)
 894                 goto out;
 895
 896         if (vma->vm_flags & VM_MAYWRITE)
 897                 bpf_map_write_active_inc(map);
 898 out:
 899         mutex_unlock(&map->freeze_mutex);
 900         return err;
 901 }
 902
 903 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 904 {
 905         struct bpf_map *map = filp->private_data;
 906
 907         if (map->ops->map_poll)
 908                 return map->ops->map_poll(map, filp, pts);
 909
 910         return EPOLLERR;
 911 }
 912
 913 const struct file_operations bpf_map_fops = {
 914 #ifdef CONFIG_PROC_FS
 915         .show_fdinfo    = bpf_map_show_fdinfo,
 916 #endif
 917         .release        = bpf_map_release,
 918         .read           = bpf_dummy_read,
 919         .write          = bpf_dummy_write,
 920         .mmap           = bpf_map_mmap,
 921         .poll           = bpf_map_poll,
 922 };
 923
 924 int bpf_map_new_fd(struct bpf_map *map, int flags)
 925 {
 926         int ret;
 927
 928         ret = security_bpf_map(map, OPEN_FMODE(flags));
 929         if (ret < 0)
 930                 return ret;
 931
 932         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 933                                 flags | O_CLOEXEC);
 934 }
 935
 936 int bpf_get_file_flag(int flags)
 937 {
 938         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 939                 return -EINVAL;
 940         if (flags & BPF_F_RDONLY)
 941                 return O_RDONLY;
 942         if (flags & BPF_F_WRONLY)
 943                 return O_WRONLY;
 944         return O_RDWR;
 945 }
 946
 947 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 948 #define CHECK_ATTR(CMD) \
 949         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 950                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 951                    sizeof(*attr) - \
 952                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 953                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 954
 955 /* dst and src must have at least "size" number of bytes.
 956  * Return strlen on success and < 0 on error.
 957  */
 958 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 959 {
 960         const char *end = src + size;
 961         const char *orig_src = src;
 962
 963         memset(dst, 0, size);
 964         /* Copy all isalnum(), '_' and '.' chars. */
 965         while (src < end && *src) {
 966                 if (!isalnum(*src) &&
 967                     *src != '_' && *src != '.')
 968                         return -EINVAL;
 969                 *dst++ = *src++;
 970         }
 971
 972         /* No '\0' found in "size" number of bytes */
 973         if (src == end)
 974                 return -EINVAL;
 975
 976         return src - orig_src;
 977 }
 978
 979 int map_check_no_btf(const struct bpf_map *map,
 980                      const struct btf *btf,
 981                      const struct btf_type *key_type,
 982                      const struct btf_type *value_type)
 983 {
 984         return -ENOTSUPP;
 985 }
 986
 987 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 988                          u32 btf_key_id, u32 btf_value_id)
 989 {
 990         const struct btf_type *key_type, *value_type;
 991         u32 key_size, value_size;
 992         int ret = 0;
 993
 994         /* Some maps allow key to be unspecified. */
 995         if (btf_key_id) {
 996                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 997                 if (!key_type || key_size != map->key_size)
 998                         return -EINVAL;
 999         } else {
1000                 key_type = btf_type_by_id(btf, 0);
1001                 if (!map->ops->map_check_btf)
1002                         return -EINVAL;
1003         }
1004
1005         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1006         if (!value_type || value_size != map->value_size)
1007                 return -EINVAL;
1008
1009         map->record = btf_parse_fields(btf, value_type,
1010                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1011                                        BPF_RB_ROOT | BPF_REFCOUNT,
1012                                        map->value_size);
1013         if (!IS_ERR_OR_NULL(map->record)) {
1014                 int i;
1015
1016                 if (!bpf_capable()) {
1017                         ret = -EPERM;
1018                         goto free_map_tab;
1019                 }
1020                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1021                         ret = -EACCES;
1022                         goto free_map_tab;
1023                 }
1024                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1025                         switch (map->record->field_mask & (1 << i)) {
1026                         case 0:
1027                                 continue;
1028                         case BPF_SPIN_LOCK:
1029                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1030                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1031                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1032                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1033                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1034                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1035                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1036                                         ret = -EOPNOTSUPP;
1037                                         goto free_map_tab;
1038                                 }
1039                                 break;
1040                         case BPF_TIMER:
1041                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1042                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1043                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1044                                         ret = -EOPNOTSUPP;
1045                                         goto free_map_tab;
1046                                 }
1047                                 break;
1048                         case BPF_KPTR_UNREF:
1049                         case BPF_KPTR_REF:
1050                         case BPF_KPTR_PERCPU:
1051                         case BPF_REFCOUNT:
1052                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1053                                     map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1054                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1055                                     map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1056                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1057                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1058                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1059                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1060                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1061                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1062                                         ret = -EOPNOTSUPP;
1063                                         goto free_map_tab;
1064                                 }
1065                                 break;
1066                         case BPF_LIST_HEAD:
1067                         case BPF_RB_ROOT:
1068                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1069                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1070                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1071                                         ret = -EOPNOTSUPP;
1072                                         goto free_map_tab;
1073                                 }
1074                                 break;
1075                         default:
1076                                 /* Fail if map_type checks are missing for a field type */
1077                                 ret = -EOPNOTSUPP;
1078                                 goto free_map_tab;
1079                         }
1080                 }
1081         }
1082
1083         ret = btf_check_and_fixup_fields(btf, map->record);
1084         if (ret < 0)
1085                 goto free_map_tab;
1086
1087         if (map->ops->map_check_btf) {
1088                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1089                 if (ret < 0)
1090                         goto free_map_tab;
1091         }
1092
1093         return ret;
1094 free_map_tab:
1095         bpf_map_free_record(map);
1096         return ret;
1097 }
1098
1099 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1100 /* called via syscall */
1101 static int map_create(union bpf_attr *attr)
1102 {
1103         const struct bpf_map_ops *ops;
1104         int numa_node = bpf_map_attr_numa_node(attr);
1105         u32 map_type = attr->map_type;
1106         struct bpf_map *map;
1107         int f_flags;
1108         int err;
1109
1110         err = CHECK_ATTR(BPF_MAP_CREATE);
1111         if (err)
1112                 return -EINVAL;
1113
1114         if (attr->btf_vmlinux_value_type_id) {
1115                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1116                     attr->btf_key_type_id || attr->btf_value_type_id)
1117                         return -EINVAL;
1118         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1119                 return -EINVAL;
1120         }
1121
1122         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1123             attr->map_extra != 0)
1124                 return -EINVAL;
1125
1126         f_flags = bpf_get_file_flag(attr->map_flags);
1127         if (f_flags < 0)
1128                 return f_flags;
1129
1130         if (numa_node != NUMA_NO_NODE &&
1131             ((unsigned int)numa_node >= nr_node_ids ||
1132              !node_online(numa_node)))
1133                 return -EINVAL;
1134
1135         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1136         map_type = attr->map_type;
1137         if (map_type >= ARRAY_SIZE(bpf_map_types))
1138                 return -EINVAL;
1139         map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1140         ops = bpf_map_types[map_type];
1141         if (!ops)
1142                 return -EINVAL;
1143
1144         if (ops->map_alloc_check) {
1145                 err = ops->map_alloc_check(attr);
1146                 if (err)
1147                         return err;
1148         }
1149         if (attr->map_ifindex)
1150                 ops = &bpf_map_offload_ops;
1151         if (!ops->map_mem_usage)
1152                 return -EINVAL;
1153
1154         /* Intent here is for unprivileged_bpf_disabled to block BPF map
1155          * creation for unprivileged users; other actions depend
1156          * on fd availability and access to bpffs, so are dependent on
1157          * object creation success. Even with unprivileged BPF disabled,
1158          * capability checks are still carried out.
1159          */
1160         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
1161                 return -EPERM;
1162
1163         /* check privileged map type permissions */
1164         switch (map_type) {
1165         case BPF_MAP_TYPE_ARRAY:
1166         case BPF_MAP_TYPE_PERCPU_ARRAY:
1167         case BPF_MAP_TYPE_PROG_ARRAY:
1168         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1169         case BPF_MAP_TYPE_CGROUP_ARRAY:
1170         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1171         case BPF_MAP_TYPE_HASH:
1172         case BPF_MAP_TYPE_PERCPU_HASH:
1173         case BPF_MAP_TYPE_HASH_OF_MAPS:
1174         case BPF_MAP_TYPE_RINGBUF:
1175         case BPF_MAP_TYPE_USER_RINGBUF:
1176         case BPF_MAP_TYPE_CGROUP_STORAGE:
1177         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1178                 /* unprivileged */
1179                 break;
1180         case BPF_MAP_TYPE_SK_STORAGE:
1181         case BPF_MAP_TYPE_INODE_STORAGE:
1182         case BPF_MAP_TYPE_TASK_STORAGE:
1183         case BPF_MAP_TYPE_CGRP_STORAGE:
1184         case BPF_MAP_TYPE_BLOOM_FILTER:
1185         case BPF_MAP_TYPE_LPM_TRIE:
1186         case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1187         case BPF_MAP_TYPE_STACK_TRACE:
1188         case BPF_MAP_TYPE_QUEUE:
1189         case BPF_MAP_TYPE_STACK:
1190         case BPF_MAP_TYPE_LRU_HASH:
1191         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1192         case BPF_MAP_TYPE_STRUCT_OPS:
1193         case BPF_MAP_TYPE_CPUMAP:
1194                 if (!bpf_capable())
1195                         return -EPERM;
1196                 break;
1197         case BPF_MAP_TYPE_SOCKMAP:
1198         case BPF_MAP_TYPE_SOCKHASH:
1199         case BPF_MAP_TYPE_DEVMAP:
1200         case BPF_MAP_TYPE_DEVMAP_HASH:
1201         case BPF_MAP_TYPE_XSKMAP:
1202                 if (!capable(CAP_NET_ADMIN))
1203                         return -EPERM;
1204                 break;
1205         default:
1206                 WARN(1, "unsupported map type %d", map_type);
1207                 return -EPERM;
1208         }
1209
1210         map = ops->map_alloc(attr);
1211         if (IS_ERR(map))
1212                 return PTR_ERR(map);
1213         map->ops = ops;
1214         map->map_type = map_type;
1215
1216         err = bpf_obj_name_cpy(map->name, attr->map_name,
1217                                sizeof(attr->map_name));
1218         if (err < 0)
1219                 goto free_map;
1220
1221         atomic64_set(&map->refcnt, 1);
1222         atomic64_set(&map->usercnt, 1);
1223         mutex_init(&map->freeze_mutex);
1224         spin_lock_init(&map->owner.lock);
1225
1226         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1227             /* Even the map's value is a kernel's struct,
1228              * the bpf_prog.o must have BTF to begin with
1229              * to figure out the corresponding kernel's
1230              * counter part.  Thus, attr->btf_fd has
1231              * to be valid also.
1232              */
1233             attr->btf_vmlinux_value_type_id) {
1234                 struct btf *btf;
1235
1236                 btf = btf_get_by_fd(attr->btf_fd);
1237                 if (IS_ERR(btf)) {
1238                         err = PTR_ERR(btf);
1239                         goto free_map;
1240                 }
1241                 if (btf_is_kernel(btf)) {
1242                         btf_put(btf);
1243                         err = -EACCES;
1244                         goto free_map;
1245                 }
1246                 map->btf = btf;
1247
1248                 if (attr->btf_value_type_id) {
1249                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1250                                             attr->btf_value_type_id);
1251                         if (err)
1252                                 goto free_map;
1253                 }
1254
1255                 map->btf_key_type_id = attr->btf_key_type_id;
1256                 map->btf_value_type_id = attr->btf_value_type_id;
1257                 map->btf_vmlinux_value_type_id =
1258                         attr->btf_vmlinux_value_type_id;
1259         }
1260
1261         err = security_bpf_map_alloc(map);
1262         if (err)
1263                 goto free_map;
1264
1265         err = bpf_map_alloc_id(map);
1266         if (err)
1267                 goto free_map_sec;
1268
1269         bpf_map_save_memcg(map);
1270
1271         err = bpf_map_new_fd(map, f_flags);
1272         if (err < 0) {
1273                 /* failed to allocate fd.
1274                  * bpf_map_put_with_uref() is needed because the above
1275                  * bpf_map_alloc_id() has published the map
1276                  * to the userspace and the userspace may
1277                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1278                  */
1279                 bpf_map_put_with_uref(map);
1280                 return err;
1281         }
1282
1283         return err;
1284
1285 free_map_sec:
1286         security_bpf_map_free(map);
1287 free_map:
1288         btf_put(map->btf);
1289         map->ops->map_free(map);
1290         return err;
1291 }
1292
1293 /* if error is returned, fd is released.
1294  * On success caller should complete fd access with matching fdput()
1295  */
1296 struct bpf_map *__bpf_map_get(struct fd f)
1297 {
1298         if (!f.file)
1299                 return ERR_PTR(-EBADF);
1300         if (f.file->f_op != &bpf_map_fops) {
1301                 fdput(f);
1302                 return ERR_PTR(-EINVAL);
1303         }
1304
1305         return f.file->private_data;
1306 }
1307
1308 void bpf_map_inc(struct bpf_map *map)
1309 {
1310         atomic64_inc(&map->refcnt);
1311 }
1312 EXPORT_SYMBOL_GPL(bpf_map_inc);
1313
1314 void bpf_map_inc_with_uref(struct bpf_map *map)
1315 {
1316         atomic64_inc(&map->refcnt);
1317         atomic64_inc(&map->usercnt);
1318 }
1319 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1320
1321 struct bpf_map *bpf_map_get(u32 ufd)
1322 {
1323         struct fd f = fdget(ufd);
1324         struct bpf_map *map;
1325
1326         map = __bpf_map_get(f);
1327         if (IS_ERR(map))
1328                 return map;
1329
1330         bpf_map_inc(map);
1331         fdput(f);
1332
1333         return map;
1334 }
1335 EXPORT_SYMBOL(bpf_map_get);
1336
1337 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1338 {
1339         struct fd f = fdget(ufd);
1340         struct bpf_map *map;
1341
1342         map = __bpf_map_get(f);
1343         if (IS_ERR(map))
1344                 return map;
1345
1346         bpf_map_inc_with_uref(map);
1347         fdput(f);
1348
1349         return map;
1350 }
1351
1352 /* map_idr_lock should have been held or the map should have been
1353  * protected by rcu read lock.
1354  */
1355 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1356 {
1357         int refold;
1358
1359         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1360         if (!refold)
1361                 return ERR_PTR(-ENOENT);
1362         if (uref)
1363                 atomic64_inc(&map->usercnt);
1364
1365         return map;
1366 }
1367
1368 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1369 {
1370         spin_lock_bh(&map_idr_lock);
1371         map = __bpf_map_inc_not_zero(map, false);
1372         spin_unlock_bh(&map_idr_lock);
1373
1374         return map;
1375 }
1376 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1377
1378 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1379 {
1380         return -ENOTSUPP;
1381 }
1382
1383 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1384 {
1385         if (key_size)
1386                 return vmemdup_user(ukey, key_size);
1387
1388         if (ukey)
1389                 return ERR_PTR(-EINVAL);
1390
1391         return NULL;
1392 }
1393
1394 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1395 {
1396         if (key_size)
1397                 return kvmemdup_bpfptr(ukey, key_size);
1398
1399         if (!bpfptr_is_null(ukey))
1400                 return ERR_PTR(-EINVAL);
1401
1402         return NULL;
1403 }
1404
1405 /* last field in 'union bpf_attr' used by this command */
1406 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1407
1408 static int map_lookup_elem(union bpf_attr *attr)
1409 {
1410         void __user *ukey = u64_to_user_ptr(attr->key);
1411         void __user *uvalue = u64_to_user_ptr(attr->value);
1412         int ufd = attr->map_fd;
1413         struct bpf_map *map;
1414         void *key, *value;
1415         u32 value_size;
1416         struct fd f;
1417         int err;
1418
1419         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1420                 return -EINVAL;
1421
1422         if (attr->flags & ~BPF_F_LOCK)
1423                 return -EINVAL;
1424
1425         f = fdget(ufd);
1426         map = __bpf_map_get(f);
1427         if (IS_ERR(map))
1428                 return PTR_ERR(map);
1429         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1430                 err = -EPERM;
1431                 goto err_put;
1432         }
1433
1434         if ((attr->flags & BPF_F_LOCK) &&
1435             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1436                 err = -EINVAL;
1437                 goto err_put;
1438         }
1439
1440         key = __bpf_copy_key(ukey, map->key_size);
1441         if (IS_ERR(key)) {
1442                 err = PTR_ERR(key);
1443                 goto err_put;
1444         }
1445
1446         value_size = bpf_map_value_size(map);
1447
1448         err = -ENOMEM;
1449         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1450         if (!value)
1451                 goto free_key;
1452
1453         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1454                 if (copy_from_user(value, uvalue, value_size))
1455                         err = -EFAULT;
1456                 else
1457                         err = bpf_map_copy_value(map, key, value, attr->flags);
1458                 goto free_value;
1459         }
1460
1461         err = bpf_map_copy_value(map, key, value, attr->flags);
1462         if (err)
1463                 goto free_value;
1464
1465         err = -EFAULT;
1466         if (copy_to_user(uvalue, value, value_size) != 0)
1467                 goto free_value;
1468
1469         err = 0;
1470
1471 free_value:
1472         kvfree(value);
1473 free_key:
1474         kvfree(key);
1475 err_put:
1476         fdput(f);
1477         return err;
1478 }
1479
1480
1481 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1482
1483 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1484 {
1485         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1486         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1487         int ufd = attr->map_fd;
1488         struct bpf_map *map;
1489         void *key, *value;
1490         u32 value_size;
1491         struct fd f;
1492         int err;
1493
1494         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1495                 return -EINVAL;
1496
1497         f = fdget(ufd);
1498         map = __bpf_map_get(f);
1499         if (IS_ERR(map))
1500                 return PTR_ERR(map);
1501         bpf_map_write_active_inc(map);
1502         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1503                 err = -EPERM;
1504                 goto err_put;
1505         }
1506
1507         if ((attr->flags & BPF_F_LOCK) &&
1508             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1509                 err = -EINVAL;
1510                 goto err_put;
1511         }
1512
1513         key = ___bpf_copy_key(ukey, map->key_size);
1514         if (IS_ERR(key)) {
1515                 err = PTR_ERR(key);
1516                 goto err_put;
1517         }
1518
1519         value_size = bpf_map_value_size(map);
1520         value = kvmemdup_bpfptr(uvalue, value_size);
1521         if (IS_ERR(value)) {
1522                 err = PTR_ERR(value);
1523                 goto free_key;
1524         }
1525
1526         err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1527
1528         kvfree(value);
1529 free_key:
1530         kvfree(key);
1531 err_put:
1532         bpf_map_write_active_dec(map);
1533         fdput(f);
1534         return err;
1535 }
1536
1537 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1538
1539 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1540 {
1541         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1542         int ufd = attr->map_fd;
1543         struct bpf_map *map;
1544         struct fd f;
1545         void *key;
1546         int err;
1547
1548         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1549                 return -EINVAL;
1550
1551         f = fdget(ufd);
1552         map = __bpf_map_get(f);
1553         if (IS_ERR(map))
1554                 return PTR_ERR(map);
1555         bpf_map_write_active_inc(map);
1556         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1557                 err = -EPERM;
1558                 goto err_put;
1559         }
1560
1561         key = ___bpf_copy_key(ukey, map->key_size);
1562         if (IS_ERR(key)) {
1563                 err = PTR_ERR(key);
1564                 goto err_put;
1565         }
1566
1567         if (bpf_map_is_offloaded(map)) {
1568                 err = bpf_map_offload_delete_elem(map, key);
1569                 goto out;
1570         } else if (IS_FD_PROG_ARRAY(map) ||
1571                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1572                 /* These maps require sleepable context */
1573                 err = map->ops->map_delete_elem(map, key);
1574                 goto out;
1575         }
1576
1577         bpf_disable_instrumentation();
1578         rcu_read_lock();
1579         err = map->ops->map_delete_elem(map, key);
1580         rcu_read_unlock();
1581         bpf_enable_instrumentation();
1582         maybe_wait_bpf_programs(map);
1583 out:
1584         kvfree(key);
1585 err_put:
1586         bpf_map_write_active_dec(map);
1587         fdput(f);
1588         return err;
1589 }
1590
1591 /* last field in 'union bpf_attr' used by this command */
1592 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1593
1594 static int map_get_next_key(union bpf_attr *attr)
1595 {
1596         void __user *ukey = u64_to_user_ptr(attr->key);
1597         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1598         int ufd = attr->map_fd;
1599         struct bpf_map *map;
1600         void *key, *next_key;
1601         struct fd f;
1602         int err;
1603
1604         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1605                 return -EINVAL;
1606
1607         f = fdget(ufd);
1608         map = __bpf_map_get(f);
1609         if (IS_ERR(map))
1610                 return PTR_ERR(map);
1611         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1612                 err = -EPERM;
1613                 goto err_put;
1614         }
1615
1616         if (ukey) {
1617                 key = __bpf_copy_key(ukey, map->key_size);
1618                 if (IS_ERR(key)) {
1619                         err = PTR_ERR(key);
1620                         goto err_put;
1621                 }
1622         } else {
1623                 key = NULL;
1624         }
1625
1626         err = -ENOMEM;
1627         next_key = kvmalloc(map->key_size, GFP_USER);
1628         if (!next_key)
1629                 goto free_key;
1630
1631         if (bpf_map_is_offloaded(map)) {
1632                 err = bpf_map_offload_get_next_key(map, key, next_key);
1633                 goto out;
1634         }
1635
1636         rcu_read_lock();
1637         err = map->ops->map_get_next_key(map, key, next_key);
1638         rcu_read_unlock();
1639 out:
1640         if (err)
1641                 goto free_next_key;
1642
1643         err = -EFAULT;
1644         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1645                 goto free_next_key;
1646
1647         err = 0;
1648
1649 free_next_key:
1650         kvfree(next_key);
1651 free_key:
1652         kvfree(key);
1653 err_put:
1654         fdput(f);
1655         return err;
1656 }
1657
1658 int generic_map_delete_batch(struct bpf_map *map,
1659                              const union bpf_attr *attr,
1660                              union bpf_attr __user *uattr)
1661 {
1662         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1663         u32 cp, max_count;
1664         int err = 0;
1665         void *key;
1666
1667         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1668                 return -EINVAL;
1669
1670         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1671             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1672                 return -EINVAL;
1673         }
1674
1675         max_count = attr->batch.count;
1676         if (!max_count)
1677                 return 0;
1678
1679         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1680         if (!key)
1681                 return -ENOMEM;
1682
1683         for (cp = 0; cp < max_count; cp++) {
1684                 err = -EFAULT;
1685                 if (copy_from_user(key, keys + cp * map->key_size,
1686                                    map->key_size))
1687                         break;
1688
1689                 if (bpf_map_is_offloaded(map)) {
1690                         err = bpf_map_offload_delete_elem(map, key);
1691                         break;
1692                 }
1693
1694                 bpf_disable_instrumentation();
1695                 rcu_read_lock();
1696                 err = map->ops->map_delete_elem(map, key);
1697                 rcu_read_unlock();
1698                 bpf_enable_instrumentation();
1699                 if (err)
1700                         break;
1701                 cond_resched();
1702         }
1703         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1704                 err = -EFAULT;
1705
1706         kvfree(key);
1707
1708         maybe_wait_bpf_programs(map);
1709         return err;
1710 }
1711
1712 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1713                              const union bpf_attr *attr,
1714                              union bpf_attr __user *uattr)
1715 {
1716         void __user *values = u64_to_user_ptr(attr->batch.values);
1717         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1718         u32 value_size, cp, max_count;
1719         void *key, *value;
1720         int err = 0;
1721
1722         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1723                 return -EINVAL;
1724
1725         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1726             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1727                 return -EINVAL;
1728         }
1729
1730         value_size = bpf_map_value_size(map);
1731
1732         max_count = attr->batch.count;
1733         if (!max_count)
1734                 return 0;
1735
1736         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1737         if (!key)
1738                 return -ENOMEM;
1739
1740         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1741         if (!value) {
1742                 kvfree(key);
1743                 return -ENOMEM;
1744         }
1745
1746         for (cp = 0; cp < max_count; cp++) {
1747                 err = -EFAULT;
1748                 if (copy_from_user(key, keys + cp * map->key_size,
1749                     map->key_size) ||
1750                     copy_from_user(value, values + cp * value_size, value_size))
1751                         break;
1752
1753                 err = bpf_map_update_value(map, map_file, key, value,
1754                                            attr->batch.elem_flags);
1755
1756                 if (err)
1757                         break;
1758                 cond_resched();
1759         }
1760
1761         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1762                 err = -EFAULT;
1763
1764         kvfree(value);
1765         kvfree(key);
1766         return err;
1767 }
1768
1769 #define MAP_LOOKUP_RETRIES 3
1770
1771 int generic_map_lookup_batch(struct bpf_map *map,
1772                                     const union bpf_attr *attr,
1773                                     union bpf_attr __user *uattr)
1774 {
1775         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1776         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1777         void __user *values = u64_to_user_ptr(attr->batch.values);
1778         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1779         void *buf, *buf_prevkey, *prev_key, *key, *value;
1780         int err, retry = MAP_LOOKUP_RETRIES;
1781         u32 value_size, cp, max_count;
1782
1783         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1784                 return -EINVAL;
1785
1786         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1787             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1788                 return -EINVAL;
1789
1790         value_size = bpf_map_value_size(map);
1791
1792         max_count = attr->batch.count;
1793         if (!max_count)
1794                 return 0;
1795
1796         if (put_user(0, &uattr->batch.count))
1797                 return -EFAULT;
1798
1799         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1800         if (!buf_prevkey)
1801                 return -ENOMEM;
1802
1803         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1804         if (!buf) {
1805                 kvfree(buf_prevkey);
1806                 return -ENOMEM;
1807         }
1808
1809         err = -EFAULT;
1810         prev_key = NULL;
1811         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1812                 goto free_buf;
1813         key = buf;
1814         value = key + map->key_size;
1815         if (ubatch)
1816                 prev_key = buf_prevkey;
1817
1818         for (cp = 0; cp < max_count;) {
1819                 rcu_read_lock();
1820                 err = map->ops->map_get_next_key(map, prev_key, key);
1821                 rcu_read_unlock();
1822                 if (err)
1823                         break;
1824                 err = bpf_map_copy_value(map, key, value,
1825                                          attr->batch.elem_flags);
1826
1827                 if (err == -ENOENT) {
1828                         if (retry) {
1829                                 retry--;
1830                                 continue;
1831                         }
1832                         err = -EINTR;
1833                         break;
1834                 }
1835
1836                 if (err)
1837                         goto free_buf;
1838
1839                 if (copy_to_user(keys + cp * map->key_size, key,
1840                                  map->key_size)) {
1841                         err = -EFAULT;
1842                         goto free_buf;
1843                 }
1844                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1845                         err = -EFAULT;
1846                         goto free_buf;
1847                 }
1848
1849                 if (!prev_key)
1850                         prev_key = buf_prevkey;
1851
1852                 swap(prev_key, key);
1853                 retry = MAP_LOOKUP_RETRIES;
1854                 cp++;
1855                 cond_resched();
1856         }
1857
1858         if (err == -EFAULT)
1859                 goto free_buf;
1860
1861         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1862                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1863                 err = -EFAULT;
1864
1865 free_buf:
1866         kvfree(buf_prevkey);
1867         kvfree(buf);
1868         return err;
1869 }
1870
1871 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1872
1873 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1874 {
1875         void __user *ukey = u64_to_user_ptr(attr->key);
1876         void __user *uvalue = u64_to_user_ptr(attr->value);
1877         int ufd = attr->map_fd;
1878         struct bpf_map *map;
1879         void *key, *value;
1880         u32 value_size;
1881         struct fd f;
1882         int err;
1883
1884         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1885                 return -EINVAL;
1886
1887         if (attr->flags & ~BPF_F_LOCK)
1888                 return -EINVAL;
1889
1890         f = fdget(ufd);
1891         map = __bpf_map_get(f);
1892         if (IS_ERR(map))
1893                 return PTR_ERR(map);
1894         bpf_map_write_active_inc(map);
1895         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1896             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1897                 err = -EPERM;
1898                 goto err_put;
1899         }
1900
1901         if (attr->flags &&
1902             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1903              map->map_type == BPF_MAP_TYPE_STACK)) {
1904                 err = -EINVAL;
1905                 goto err_put;
1906         }
1907
1908         if ((attr->flags & BPF_F_LOCK) &&
1909             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1910                 err = -EINVAL;
1911                 goto err_put;
1912         }
1913
1914         key = __bpf_copy_key(ukey, map->key_size);
1915         if (IS_ERR(key)) {
1916                 err = PTR_ERR(key);
1917                 goto err_put;
1918         }
1919
1920         value_size = bpf_map_value_size(map);
1921
1922         err = -ENOMEM;
1923         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1924         if (!value)
1925                 goto free_key;
1926
1927         err = -ENOTSUPP;
1928         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1929             map->map_type == BPF_MAP_TYPE_STACK) {
1930                 err = map->ops->map_pop_elem(map, value);
1931         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1932                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1933                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1934                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1935                 if (!bpf_map_is_offloaded(map)) {
1936                         bpf_disable_instrumentation();
1937                         rcu_read_lock();
1938                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1939                         rcu_read_unlock();
1940                         bpf_enable_instrumentation();
1941                 }
1942         }
1943
1944         if (err)
1945                 goto free_value;
1946
1947         if (copy_to_user(uvalue, value, value_size) != 0) {
1948                 err = -EFAULT;
1949                 goto free_value;
1950         }
1951
1952         err = 0;
1953
1954 free_value:
1955         kvfree(value);
1956 free_key:
1957         kvfree(key);
1958 err_put:
1959         bpf_map_write_active_dec(map);
1960         fdput(f);
1961         return err;
1962 }
1963
1964 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1965
1966 static int map_freeze(const union bpf_attr *attr)
1967 {
1968         int err = 0, ufd = attr->map_fd;
1969         struct bpf_map *map;
1970         struct fd f;
1971
1972         if (CHECK_ATTR(BPF_MAP_FREEZE))
1973                 return -EINVAL;
1974
1975         f = fdget(ufd);
1976         map = __bpf_map_get(f);
1977         if (IS_ERR(map))
1978                 return PTR_ERR(map);
1979
1980         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1981                 fdput(f);
1982                 return -ENOTSUPP;
1983         }
1984
1985         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1986                 fdput(f);
1987                 return -EPERM;
1988         }
1989
1990         mutex_lock(&map->freeze_mutex);
1991         if (bpf_map_write_active(map)) {
1992                 err = -EBUSY;
1993                 goto err_put;
1994         }
1995         if (READ_ONCE(map->frozen)) {
1996                 err = -EBUSY;
1997                 goto err_put;
1998         }
1999
2000         WRITE_ONCE(map->frozen, true);
2001 err_put:
2002         mutex_unlock(&map->freeze_mutex);
2003         fdput(f);
2004         return err;
2005 }
2006
2007 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2008 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2009         [_id] = & _name ## _prog_ops,
2010 #define BPF_MAP_TYPE(_id, _ops)
2011 #define BPF_LINK_TYPE(_id, _name)
2012 #include <linux/bpf_types.h>
2013 #undef BPF_PROG_TYPE
2014 #undef BPF_MAP_TYPE
2015 #undef BPF_LINK_TYPE
2016 };
2017
2018 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2019 {
2020         const struct bpf_prog_ops *ops;
2021
2022         if (type >= ARRAY_SIZE(bpf_prog_types))
2023                 return -EINVAL;
2024         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2025         ops = bpf_prog_types[type];
2026         if (!ops)
2027                 return -EINVAL;
2028
2029         if (!bpf_prog_is_offloaded(prog->aux))
2030                 prog->aux->ops = ops;
2031         else
2032                 prog->aux->ops = &bpf_offload_prog_ops;
2033         prog->type = type;
2034         return 0;
2035 }
2036
2037 enum bpf_audit {
2038         BPF_AUDIT_LOAD,
2039         BPF_AUDIT_UNLOAD,
2040         BPF_AUDIT_MAX,
2041 };
2042
2043 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2044         [BPF_AUDIT_LOAD]   = "LOAD",
2045         [BPF_AUDIT_UNLOAD] = "UNLOAD",
2046 };
2047
2048 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2049 {
2050         struct audit_context *ctx = NULL;
2051         struct audit_buffer *ab;
2052
2053         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2054                 return;
2055         if (audit_enabled == AUDIT_OFF)
2056                 return;
2057         if (!in_irq() && !irqs_disabled())
2058                 ctx = audit_context();
2059         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2060         if (unlikely(!ab))
2061                 return;
2062         audit_log_format(ab, "prog-id=%u op=%s",
2063                          prog->aux->id, bpf_audit_str[op]);
2064         audit_log_end(ab);
2065 }
2066
2067 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2068 {
2069         int id;
2070
2071         idr_preload(GFP_KERNEL);
2072         spin_lock_bh(&prog_idr_lock);
2073         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2074         if (id > 0)
2075                 prog->aux->id = id;
2076         spin_unlock_bh(&prog_idr_lock);
2077         idr_preload_end();
2078
2079         /* id is in [1, INT_MAX) */
2080         if (WARN_ON_ONCE(!id))
2081                 return -ENOSPC;
2082
2083         return id > 0 ? 0 : id;
2084 }
2085
2086 void bpf_prog_free_id(struct bpf_prog *prog)
2087 {
2088         unsigned long flags;
2089
2090         /* cBPF to eBPF migrations are currently not in the idr store.
2091          * Offloaded programs are removed from the store when their device
2092          * disappears - even if someone grabs an fd to them they are unusable,
2093          * simply waiting for refcnt to drop to be freed.
2094          */
2095         if (!prog->aux->id)
2096                 return;
2097
2098         spin_lock_irqsave(&prog_idr_lock, flags);
2099         idr_remove(&prog_idr, prog->aux->id);
2100         prog->aux->id = 0;
2101         spin_unlock_irqrestore(&prog_idr_lock, flags);
2102 }
2103
2104 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2105 {
2106         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2107
2108         kvfree(aux->func_info);
2109         kfree(aux->func_info_aux);
2110         free_uid(aux->user);
2111         security_bpf_prog_free(aux);
2112         bpf_prog_free(aux->prog);
2113 }
2114
2115 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2116 {
2117         bpf_prog_kallsyms_del_all(prog);
2118         btf_put(prog->aux->btf);
2119         module_put(prog->aux->mod);
2120         kvfree(prog->aux->jited_linfo);
2121         kvfree(prog->aux->linfo);
2122         kfree(prog->aux->kfunc_tab);
2123         if (prog->aux->attach_btf)
2124                 btf_put(prog->aux->attach_btf);
2125
2126         if (deferred) {
2127                 if (prog->aux->sleepable)
2128                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2129                 else
2130                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2131         } else {
2132                 __bpf_prog_put_rcu(&prog->aux->rcu);
2133         }
2134 }
2135
2136 static void bpf_prog_put_deferred(struct work_struct *work)
2137 {
2138         struct bpf_prog_aux *aux;
2139         struct bpf_prog *prog;
2140
2141         aux = container_of(work, struct bpf_prog_aux, work);
2142         prog = aux->prog;
2143         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2144         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2145         bpf_prog_free_id(prog);
2146         __bpf_prog_put_noref(prog, true);
2147 }
2148
2149 static void __bpf_prog_put(struct bpf_prog *prog)
2150 {
2151         struct bpf_prog_aux *aux = prog->aux;
2152
2153         if (atomic64_dec_and_test(&aux->refcnt)) {
2154                 if (in_irq() || irqs_disabled()) {
2155                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2156                         schedule_work(&aux->work);
2157                 } else {
2158                         bpf_prog_put_deferred(&aux->work);
2159                 }
2160         }
2161 }
2162
2163 void bpf_prog_put(struct bpf_prog *prog)
2164 {
2165         __bpf_prog_put(prog);
2166 }
2167 EXPORT_SYMBOL_GPL(bpf_prog_put);
2168
2169 static int bpf_prog_release(struct inode *inode, struct file *filp)
2170 {
2171         struct bpf_prog *prog = filp->private_data;
2172
2173         bpf_prog_put(prog);
2174         return 0;
2175 }
2176
2177 struct bpf_prog_kstats {
2178         u64 nsecs;
2179         u64 cnt;
2180         u64 misses;
2181 };
2182
2183 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2184 {
2185         struct bpf_prog_stats *stats;
2186         unsigned int flags;
2187
2188         stats = this_cpu_ptr(prog->stats);
2189         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2190         u64_stats_inc(&stats->misses);
2191         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2192 }
2193
2194 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2195                                struct bpf_prog_kstats *stats)
2196 {
2197         u64 nsecs = 0, cnt = 0, misses = 0;
2198         int cpu;
2199
2200         for_each_possible_cpu(cpu) {
2201                 const struct bpf_prog_stats *st;
2202                 unsigned int start;
2203                 u64 tnsecs, tcnt, tmisses;
2204
2205                 st = per_cpu_ptr(prog->stats, cpu);
2206                 do {
2207                         start = u64_stats_fetch_begin(&st->syncp);
2208                         tnsecs = u64_stats_read(&st->nsecs);
2209                         tcnt = u64_stats_read(&st->cnt);
2210                         tmisses = u64_stats_read(&st->misses);
2211                 } while (u64_stats_fetch_retry(&st->syncp, start));
2212                 nsecs += tnsecs;
2213                 cnt += tcnt;
2214                 misses += tmisses;
2215         }
2216         stats->nsecs = nsecs;
2217         stats->cnt = cnt;
2218         stats->misses = misses;
2219 }
2220
2221 #ifdef CONFIG_PROC_FS
2222 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2223 {
2224         const struct bpf_prog *prog = filp->private_data;
2225         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2226         struct bpf_prog_kstats stats;
2227
2228         bpf_prog_get_stats(prog, &stats);
2229         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2230         seq_printf(m,
2231                    "prog_type:\t%u\n"
2232                    "prog_jited:\t%u\n"
2233                    "prog_tag:\t%s\n"
2234                    "memlock:\t%llu\n"
2235                    "prog_id:\t%u\n"
2236                    "run_time_ns:\t%llu\n"
2237                    "run_cnt:\t%llu\n"
2238                    "recursion_misses:\t%llu\n"
2239                    "verified_insns:\t%u\n",
2240                    prog->type,
2241                    prog->jited,
2242                    prog_tag,
2243                    prog->pages * 1ULL << PAGE_SHIFT,
2244                    prog->aux->id,
2245                    stats.nsecs,
2246                    stats.cnt,
2247                    stats.misses,
2248                    prog->aux->verified_insns);
2249 }
2250 #endif
2251
2252 const struct file_operations bpf_prog_fops = {
2253 #ifdef CONFIG_PROC_FS
2254         .show_fdinfo    = bpf_prog_show_fdinfo,
2255 #endif
2256         .release        = bpf_prog_release,
2257         .read           = bpf_dummy_read,
2258         .write          = bpf_dummy_write,
2259 };
2260
2261 int bpf_prog_new_fd(struct bpf_prog *prog)
2262 {
2263         int ret;
2264
2265         ret = security_bpf_prog(prog);
2266         if (ret < 0)
2267                 return ret;
2268
2269         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2270                                 O_RDWR | O_CLOEXEC);
2271 }
2272
2273 static struct bpf_prog *____bpf_prog_get(struct fd f)
2274 {
2275         if (!f.file)
2276                 return ERR_PTR(-EBADF);
2277         if (f.file->f_op != &bpf_prog_fops) {
2278                 fdput(f);
2279                 return ERR_PTR(-EINVAL);
2280         }
2281
2282         return f.file->private_data;
2283 }
2284
2285 void bpf_prog_add(struct bpf_prog *prog, int i)
2286 {
2287         atomic64_add(i, &prog->aux->refcnt);
2288 }
2289 EXPORT_SYMBOL_GPL(bpf_prog_add);
2290
2291 void bpf_prog_sub(struct bpf_prog *prog, int i)
2292 {
2293         /* Only to be used for undoing previous bpf_prog_add() in some
2294          * error path. We still know that another entity in our call
2295          * path holds a reference to the program, thus atomic_sub() can
2296          * be safely used in such cases!
2297          */
2298         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2299 }
2300 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2301
2302 void bpf_prog_inc(struct bpf_prog *prog)
2303 {
2304         atomic64_inc(&prog->aux->refcnt);
2305 }
2306 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2307
2308 /* prog_idr_lock should have been held */
2309 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2310 {
2311         int refold;
2312
2313         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2314
2315         if (!refold)
2316                 return ERR_PTR(-ENOENT);
2317
2318         return prog;
2319 }
2320 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2321
2322 bool bpf_prog_get_ok(struct bpf_prog *prog,
2323                             enum bpf_prog_type *attach_type, bool attach_drv)
2324 {
2325         /* not an attachment, just a refcount inc, always allow */
2326         if (!attach_type)
2327                 return true;
2328
2329         if (prog->type != *attach_type)
2330                 return false;
2331         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2332                 return false;
2333
2334         return true;
2335 }
2336
2337 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2338                                        bool attach_drv)
2339 {
2340         struct fd f = fdget(ufd);
2341         struct bpf_prog *prog;
2342
2343         prog = ____bpf_prog_get(f);
2344         if (IS_ERR(prog))
2345                 return prog;
2346         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2347                 prog = ERR_PTR(-EINVAL);
2348                 goto out;
2349         }
2350
2351         bpf_prog_inc(prog);
2352 out:
2353         fdput(f);
2354         return prog;
2355 }
2356
2357 struct bpf_prog *bpf_prog_get(u32 ufd)
2358 {
2359         return __bpf_prog_get(ufd, NULL, false);
2360 }
2361
2362 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2363                                        bool attach_drv)
2364 {
2365         return __bpf_prog_get(ufd, &type, attach_drv);
2366 }
2367 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2368
2369 /* Initially all BPF programs could be loaded w/o specifying
2370  * expected_attach_type. Later for some of them specifying expected_attach_type
2371  * at load time became required so that program could be validated properly.
2372  * Programs of types that are allowed to be loaded both w/ and w/o (for
2373  * backward compatibility) expected_attach_type, should have the default attach
2374  * type assigned to expected_attach_type for the latter case, so that it can be
2375  * validated later at attach time.
2376  *
2377  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2378  * prog type requires it but has some attach types that have to be backward
2379  * compatible.
2380  */
2381 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2382 {
2383         switch (attr->prog_type) {
2384         case BPF_PROG_TYPE_CGROUP_SOCK:
2385                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2386                  * exist so checking for non-zero is the way to go here.
2387                  */
2388                 if (!attr->expected_attach_type)
2389                         attr->expected_attach_type =
2390                                 BPF_CGROUP_INET_SOCK_CREATE;
2391                 break;
2392         case BPF_PROG_TYPE_SK_REUSEPORT:
2393                 if (!attr->expected_attach_type)
2394                         attr->expected_attach_type =
2395                                 BPF_SK_REUSEPORT_SELECT;
2396                 break;
2397         }
2398 }
2399
2400 static int
2401 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2402                            enum bpf_attach_type expected_attach_type,
2403                            struct btf *attach_btf, u32 btf_id,
2404                            struct bpf_prog *dst_prog)
2405 {
2406         if (btf_id) {
2407                 if (btf_id > BTF_MAX_TYPE)
2408                         return -EINVAL;
2409
2410                 if (!attach_btf && !dst_prog)
2411                         return -EINVAL;
2412
2413                 switch (prog_type) {
2414                 case BPF_PROG_TYPE_TRACING:
2415                 case BPF_PROG_TYPE_LSM:
2416                 case BPF_PROG_TYPE_STRUCT_OPS:
2417                 case BPF_PROG_TYPE_EXT:
2418                         break;
2419                 default:
2420                         return -EINVAL;
2421                 }
2422         }
2423
2424         if (attach_btf && (!btf_id || dst_prog))
2425                 return -EINVAL;
2426
2427         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2428             prog_type != BPF_PROG_TYPE_EXT)
2429                 return -EINVAL;
2430
2431         switch (prog_type) {
2432         case BPF_PROG_TYPE_CGROUP_SOCK:
2433                 switch (expected_attach_type) {
2434                 case BPF_CGROUP_INET_SOCK_CREATE:
2435                 case BPF_CGROUP_INET_SOCK_RELEASE:
2436                 case BPF_CGROUP_INET4_POST_BIND:
2437                 case BPF_CGROUP_INET6_POST_BIND:
2438                         return 0;
2439                 default:
2440                         return -EINVAL;
2441                 }
2442         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2443                 switch (expected_attach_type) {
2444                 case BPF_CGROUP_INET4_BIND:
2445                 case BPF_CGROUP_INET6_BIND:
2446                 case BPF_CGROUP_INET4_CONNECT:
2447                 case BPF_CGROUP_INET6_CONNECT:
2448                 case BPF_CGROUP_UNIX_CONNECT:
2449                 case BPF_CGROUP_INET4_GETPEERNAME:
2450                 case BPF_CGROUP_INET6_GETPEERNAME:
2451                 case BPF_CGROUP_UNIX_GETPEERNAME:
2452                 case BPF_CGROUP_INET4_GETSOCKNAME:
2453                 case BPF_CGROUP_INET6_GETSOCKNAME:
2454                 case BPF_CGROUP_UNIX_GETSOCKNAME:
2455                 case BPF_CGROUP_UDP4_SENDMSG:
2456                 case BPF_CGROUP_UDP6_SENDMSG:
2457                 case BPF_CGROUP_UNIX_SENDMSG:
2458                 case BPF_CGROUP_UDP4_RECVMSG:
2459                 case BPF_CGROUP_UDP6_RECVMSG:
2460                 case BPF_CGROUP_UNIX_RECVMSG:
2461                         return 0;
2462                 default:
2463                         return -EINVAL;
2464                 }
2465         case BPF_PROG_TYPE_CGROUP_SKB:
2466                 switch (expected_attach_type) {
2467                 case BPF_CGROUP_INET_INGRESS:
2468                 case BPF_CGROUP_INET_EGRESS:
2469                         return 0;
2470                 default:
2471                         return -EINVAL;
2472                 }
2473         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2474                 switch (expected_attach_type) {
2475                 case BPF_CGROUP_SETSOCKOPT:
2476                 case BPF_CGROUP_GETSOCKOPT:
2477                         return 0;
2478                 default:
2479                         return -EINVAL;
2480                 }
2481         case BPF_PROG_TYPE_SK_LOOKUP:
2482                 if (expected_attach_type == BPF_SK_LOOKUP)
2483                         return 0;
2484                 return -EINVAL;
2485         case BPF_PROG_TYPE_SK_REUSEPORT:
2486                 switch (expected_attach_type) {
2487                 case BPF_SK_REUSEPORT_SELECT:
2488                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2489                         return 0;
2490                 default:
2491                         return -EINVAL;
2492                 }
2493         case BPF_PROG_TYPE_NETFILTER:
2494                 if (expected_attach_type == BPF_NETFILTER)
2495                         return 0;
2496                 return -EINVAL;
2497         case BPF_PROG_TYPE_SYSCALL:
2498         case BPF_PROG_TYPE_EXT:
2499                 if (expected_attach_type)
2500                         return -EINVAL;
2501                 fallthrough;
2502         default:
2503                 return 0;
2504         }
2505 }
2506
2507 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2508 {
2509         switch (prog_type) {
2510         case BPF_PROG_TYPE_SCHED_CLS:
2511         case BPF_PROG_TYPE_SCHED_ACT:
2512         case BPF_PROG_TYPE_XDP:
2513         case BPF_PROG_TYPE_LWT_IN:
2514         case BPF_PROG_TYPE_LWT_OUT:
2515         case BPF_PROG_TYPE_LWT_XMIT:
2516         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2517         case BPF_PROG_TYPE_SK_SKB:
2518         case BPF_PROG_TYPE_SK_MSG:
2519         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2520         case BPF_PROG_TYPE_CGROUP_DEVICE:
2521         case BPF_PROG_TYPE_CGROUP_SOCK:
2522         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2523         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2524         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2525         case BPF_PROG_TYPE_SOCK_OPS:
2526         case BPF_PROG_TYPE_EXT: /* extends any prog */
2527         case BPF_PROG_TYPE_NETFILTER:
2528                 return true;
2529         case BPF_PROG_TYPE_CGROUP_SKB:
2530                 /* always unpriv */
2531         case BPF_PROG_TYPE_SK_REUSEPORT:
2532                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2533         default:
2534                 return false;
2535         }
2536 }
2537
2538 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2539 {
2540         switch (prog_type) {
2541         case BPF_PROG_TYPE_KPROBE:
2542         case BPF_PROG_TYPE_TRACEPOINT:
2543         case BPF_PROG_TYPE_PERF_EVENT:
2544         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2545         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2546         case BPF_PROG_TYPE_TRACING:
2547         case BPF_PROG_TYPE_LSM:
2548         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2549         case BPF_PROG_TYPE_EXT: /* extends any prog */
2550                 return true;
2551         default:
2552                 return false;
2553         }
2554 }
2555
2556 /* last field in 'union bpf_attr' used by this command */
2557 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2558
2559 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2560 {
2561         enum bpf_prog_type type = attr->prog_type;
2562         struct bpf_prog *prog, *dst_prog = NULL;
2563         struct btf *attach_btf = NULL;
2564         int err;
2565         char license[128];
2566
2567         if (CHECK_ATTR(BPF_PROG_LOAD))
2568                 return -EINVAL;
2569
2570         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2571                                  BPF_F_ANY_ALIGNMENT |
2572                                  BPF_F_TEST_STATE_FREQ |
2573                                  BPF_F_SLEEPABLE |
2574                                  BPF_F_TEST_RND_HI32 |
2575                                  BPF_F_XDP_HAS_FRAGS |
2576                                  BPF_F_XDP_DEV_BOUND_ONLY))
2577                 return -EINVAL;
2578
2579         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2580             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2581             !bpf_capable())
2582                 return -EPERM;
2583
2584         /* Intent here is for unprivileged_bpf_disabled to block BPF program
2585          * creation for unprivileged users; other actions depend
2586          * on fd availability and access to bpffs, so are dependent on
2587          * object creation success. Even with unprivileged BPF disabled,
2588          * capability checks are still carried out for these
2589          * and other operations.
2590          */
2591         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
2592                 return -EPERM;
2593
2594         if (attr->insn_cnt == 0 ||
2595             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2596                 return -E2BIG;
2597         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2598             type != BPF_PROG_TYPE_CGROUP_SKB &&
2599             !bpf_capable())
2600                 return -EPERM;
2601
2602         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2603                 return -EPERM;
2604         if (is_perfmon_prog_type(type) && !perfmon_capable())
2605                 return -EPERM;
2606
2607         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2608          * or btf, we need to check which one it is
2609          */
2610         if (attr->attach_prog_fd) {
2611                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2612                 if (IS_ERR(dst_prog)) {
2613                         dst_prog = NULL;
2614                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2615                         if (IS_ERR(attach_btf))
2616                                 return -EINVAL;
2617                         if (!btf_is_kernel(attach_btf)) {
2618                                 /* attaching through specifying bpf_prog's BTF
2619                                  * objects directly might be supported eventually
2620                                  */
2621                                 btf_put(attach_btf);
2622                                 return -ENOTSUPP;
2623                         }
2624                 }
2625         } else if (attr->attach_btf_id) {
2626                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2627                 attach_btf = bpf_get_btf_vmlinux();
2628                 if (IS_ERR(attach_btf))
2629                         return PTR_ERR(attach_btf);
2630                 if (!attach_btf)
2631                         return -EINVAL;
2632                 btf_get(attach_btf);
2633         }
2634
2635         bpf_prog_load_fixup_attach_type(attr);
2636         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2637                                        attach_btf, attr->attach_btf_id,
2638                                        dst_prog)) {
2639                 if (dst_prog)
2640                         bpf_prog_put(dst_prog);
2641                 if (attach_btf)
2642                         btf_put(attach_btf);
2643                 return -EINVAL;
2644         }
2645
2646         /* plain bpf_prog allocation */
2647         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2648         if (!prog) {
2649                 if (dst_prog)
2650                         bpf_prog_put(dst_prog);
2651                 if (attach_btf)
2652                         btf_put(attach_btf);
2653                 return -ENOMEM;
2654         }
2655
2656         prog->expected_attach_type = attr->expected_attach_type;
2657         prog->aux->attach_btf = attach_btf;
2658         prog->aux->attach_btf_id = attr->attach_btf_id;
2659         prog->aux->dst_prog = dst_prog;
2660         prog->aux->dev_bound = !!attr->prog_ifindex;
2661         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2662         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2663
2664         err = security_bpf_prog_alloc(prog->aux);
2665         if (err)
2666                 goto free_prog;
2667
2668         prog->aux->user = get_current_user();
2669         prog->len = attr->insn_cnt;
2670
2671         err = -EFAULT;
2672         if (copy_from_bpfptr(prog->insns,
2673                              make_bpfptr(attr->insns, uattr.is_kernel),
2674                              bpf_prog_insn_size(prog)) != 0)
2675                 goto free_prog_sec;
2676         /* copy eBPF program license from user space */
2677         if (strncpy_from_bpfptr(license,
2678                                 make_bpfptr(attr->license, uattr.is_kernel),
2679                                 sizeof(license) - 1) < 0)
2680                 goto free_prog_sec;
2681         license[sizeof(license) - 1] = 0;
2682
2683         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2684         prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2685
2686         prog->orig_prog = NULL;
2687         prog->jited = 0;
2688
2689         atomic64_set(&prog->aux->refcnt, 1);
2690
2691         if (bpf_prog_is_dev_bound(prog->aux)) {
2692                 err = bpf_prog_dev_bound_init(prog, attr);
2693                 if (err)
2694                         goto free_prog_sec;
2695         }
2696
2697         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2698             bpf_prog_is_dev_bound(dst_prog->aux)) {
2699                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2700                 if (err)
2701                         goto free_prog_sec;
2702         }
2703
2704         /* find program type: socket_filter vs tracing_filter */
2705         err = find_prog_type(type, prog);
2706         if (err < 0)
2707                 goto free_prog_sec;
2708
2709         prog->aux->load_time = ktime_get_boottime_ns();
2710         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2711                                sizeof(attr->prog_name));
2712         if (err < 0)
2713                 goto free_prog_sec;
2714
2715         /* run eBPF verifier */
2716         err = bpf_check(&prog, attr, uattr, uattr_size);
2717         if (err < 0)
2718                 goto free_used_maps;
2719
2720         prog = bpf_prog_select_runtime(prog, &err);
2721         if (err < 0)
2722                 goto free_used_maps;
2723
2724         err = bpf_prog_alloc_id(prog);
2725         if (err)
2726                 goto free_used_maps;
2727
2728         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2729          * effectively publicly exposed. However, retrieving via
2730          * bpf_prog_get_fd_by_id() will take another reference,
2731          * therefore it cannot be gone underneath us.
2732          *
2733          * Only for the time /after/ successful bpf_prog_new_fd()
2734          * and before returning to userspace, we might just hold
2735          * one reference and any parallel close on that fd could
2736          * rip everything out. Hence, below notifications must
2737          * happen before bpf_prog_new_fd().
2738          *
2739          * Also, any failure handling from this point onwards must
2740          * be using bpf_prog_put() given the program is exposed.
2741          */
2742         bpf_prog_kallsyms_add(prog);
2743         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2744         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2745
2746         err = bpf_prog_new_fd(prog);
2747         if (err < 0)
2748                 bpf_prog_put(prog);
2749         return err;
2750
2751 free_used_maps:
2752         /* In case we have subprogs, we need to wait for a grace
2753          * period before we can tear down JIT memory since symbols
2754          * are already exposed under kallsyms.
2755          */
2756         __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2757         return err;
2758 free_prog_sec:
2759         free_uid(prog->aux->user);
2760         security_bpf_prog_free(prog->aux);
2761 free_prog:
2762         if (prog->aux->attach_btf)
2763                 btf_put(prog->aux->attach_btf);
2764         bpf_prog_free(prog);
2765         return err;
2766 }
2767
2768 #define BPF_OBJ_LAST_FIELD path_fd
2769
2770 static int bpf_obj_pin(const union bpf_attr *attr)
2771 {
2772         int path_fd;
2773
2774         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2775                 return -EINVAL;
2776
2777         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2778         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2779                 return -EINVAL;
2780
2781         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2782         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2783                                 u64_to_user_ptr(attr->pathname));
2784 }
2785
2786 static int bpf_obj_get(const union bpf_attr *attr)
2787 {
2788         int path_fd;
2789
2790         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2791             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2792                 return -EINVAL;
2793
2794         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2795         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2796                 return -EINVAL;
2797
2798         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2799         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2800                                 attr->file_flags);
2801 }
2802
2803 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2804                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2805 {
2806         atomic64_set(&link->refcnt, 1);
2807         link->type = type;
2808         link->id = 0;
2809         link->ops = ops;
2810         link->prog = prog;
2811 }
2812
2813 static void bpf_link_free_id(int id)
2814 {
2815         if (!id)
2816                 return;
2817
2818         spin_lock_bh(&link_idr_lock);
2819         idr_remove(&link_idr, id);
2820         spin_unlock_bh(&link_idr_lock);
2821 }
2822
2823 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2824  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2825  * anon_inode's release() call. This helper marks bpf_link as
2826  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2827  * is not decremented, it's the responsibility of a calling code that failed
2828  * to complete bpf_link initialization.
2829  * This helper eventually calls link's dealloc callback, but does not call
2830  * link's release callback.
2831  */
2832 void bpf_link_cleanup(struct bpf_link_primer *primer)
2833 {
2834         primer->link->prog = NULL;
2835         bpf_link_free_id(primer->id);
2836         fput(primer->file);
2837         put_unused_fd(primer->fd);
2838 }
2839
2840 void bpf_link_inc(struct bpf_link *link)
2841 {
2842         atomic64_inc(&link->refcnt);
2843 }
2844
2845 /* bpf_link_free is guaranteed to be called from process context */
2846 static void bpf_link_free(struct bpf_link *link)
2847 {
2848         bpf_link_free_id(link->id);
2849         if (link->prog) {
2850                 /* detach BPF program, clean up used resources */
2851                 link->ops->release(link);
2852                 bpf_prog_put(link->prog);
2853         }
2854         /* free bpf_link and its containing memory */
2855         link->ops->dealloc(link);
2856 }
2857
2858 static void bpf_link_put_deferred(struct work_struct *work)
2859 {
2860         struct bpf_link *link = container_of(work, struct bpf_link, work);
2861
2862         bpf_link_free(link);
2863 }
2864
2865 /* bpf_link_put might be called from atomic context. It needs to be called
2866  * from sleepable context in order to acquire sleeping locks during the process.
2867  */
2868 void bpf_link_put(struct bpf_link *link)
2869 {
2870         if (!atomic64_dec_and_test(&link->refcnt))
2871                 return;
2872
2873         INIT_WORK(&link->work, bpf_link_put_deferred);
2874         schedule_work(&link->work);
2875 }
2876 EXPORT_SYMBOL(bpf_link_put);
2877
2878 static void bpf_link_put_direct(struct bpf_link *link)
2879 {
2880         if (!atomic64_dec_and_test(&link->refcnt))
2881                 return;
2882         bpf_link_free(link);
2883 }
2884
2885 static int bpf_link_release(struct inode *inode, struct file *filp)
2886 {
2887         struct bpf_link *link = filp->private_data;
2888
2889         bpf_link_put_direct(link);
2890         return 0;
2891 }
2892
2893 #ifdef CONFIG_PROC_FS
2894 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2895 #define BPF_MAP_TYPE(_id, _ops)
2896 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2897 static const char *bpf_link_type_strs[] = {
2898         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2899 #include <linux/bpf_types.h>
2900 };
2901 #undef BPF_PROG_TYPE
2902 #undef BPF_MAP_TYPE
2903 #undef BPF_LINK_TYPE
2904
2905 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2906 {
2907         const struct bpf_link *link = filp->private_data;
2908         const struct bpf_prog *prog = link->prog;
2909         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2910
2911         seq_printf(m,
2912                    "link_type:\t%s\n"
2913                    "link_id:\t%u\n",
2914                    bpf_link_type_strs[link->type],
2915                    link->id);
2916         if (prog) {
2917                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2918                 seq_printf(m,
2919                            "prog_tag:\t%s\n"
2920                            "prog_id:\t%u\n",
2921                            prog_tag,
2922                            prog->aux->id);
2923         }
2924         if (link->ops->show_fdinfo)
2925                 link->ops->show_fdinfo(link, m);
2926 }
2927 #endif
2928
2929 static const struct file_operations bpf_link_fops = {
2930 #ifdef CONFIG_PROC_FS
2931         .show_fdinfo    = bpf_link_show_fdinfo,
2932 #endif
2933         .release        = bpf_link_release,
2934         .read           = bpf_dummy_read,
2935         .write          = bpf_dummy_write,
2936 };
2937
2938 static int bpf_link_alloc_id(struct bpf_link *link)
2939 {
2940         int id;
2941
2942         idr_preload(GFP_KERNEL);
2943         spin_lock_bh(&link_idr_lock);
2944         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2945         spin_unlock_bh(&link_idr_lock);
2946         idr_preload_end();
2947
2948         return id;
2949 }
2950
2951 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2952  * reserving unused FD and allocating ID from link_idr. This is to be paired
2953  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2954  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2955  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2956  * transient state is passed around in struct bpf_link_primer.
2957  * This is preferred way to create and initialize bpf_link, especially when
2958  * there are complicated and expensive operations in between creating bpf_link
2959  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2960  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2961  * expensive (and potentially failing) roll back operations in a rare case
2962  * that file, FD, or ID can't be allocated.
2963  */
2964 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2965 {
2966         struct file *file;
2967         int fd, id;
2968
2969         fd = get_unused_fd_flags(O_CLOEXEC);
2970         if (fd < 0)
2971                 return fd;
2972
2973
2974         id = bpf_link_alloc_id(link);
2975         if (id < 0) {
2976                 put_unused_fd(fd);
2977                 return id;
2978         }
2979
2980         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2981         if (IS_ERR(file)) {
2982                 bpf_link_free_id(id);
2983                 put_unused_fd(fd);
2984                 return PTR_ERR(file);
2985         }
2986
2987         primer->link = link;
2988         primer->file = file;
2989         primer->fd = fd;
2990         primer->id = id;
2991         return 0;
2992 }
2993
2994 int bpf_link_settle(struct bpf_link_primer *primer)
2995 {
2996         /* make bpf_link fetchable by ID */
2997         spin_lock_bh(&link_idr_lock);
2998         primer->link->id = primer->id;
2999         spin_unlock_bh(&link_idr_lock);
3000         /* make bpf_link fetchable by FD */
3001         fd_install(primer->fd, primer->file);
3002         /* pass through installed FD */
3003         return primer->fd;
3004 }
3005
3006 int bpf_link_new_fd(struct bpf_link *link)
3007 {
3008         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3009 }
3010
3011 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3012 {
3013         struct fd f = fdget(ufd);
3014         struct bpf_link *link;
3015
3016         if (!f.file)
3017                 return ERR_PTR(-EBADF);
3018         if (f.file->f_op != &bpf_link_fops) {
3019                 fdput(f);
3020                 return ERR_PTR(-EINVAL);
3021         }
3022
3023         link = f.file->private_data;
3024         bpf_link_inc(link);
3025         fdput(f);
3026
3027         return link;
3028 }
3029 EXPORT_SYMBOL(bpf_link_get_from_fd);
3030
3031 static void bpf_tracing_link_release(struct bpf_link *link)
3032 {
3033         struct bpf_tracing_link *tr_link =
3034                 container_of(link, struct bpf_tracing_link, link.link);
3035
3036         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3037                                                 tr_link->trampoline));
3038
3039         bpf_trampoline_put(tr_link->trampoline);
3040
3041         /* tgt_prog is NULL if target is a kernel function */
3042         if (tr_link->tgt_prog)
3043                 bpf_prog_put(tr_link->tgt_prog);
3044 }
3045
3046 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3047 {
3048         struct bpf_tracing_link *tr_link =
3049                 container_of(link, struct bpf_tracing_link, link.link);
3050
3051         kfree(tr_link);
3052 }
3053
3054 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3055                                          struct seq_file *seq)
3056 {
3057         struct bpf_tracing_link *tr_link =
3058                 container_of(link, struct bpf_tracing_link, link.link);
3059         u32 target_btf_id, target_obj_id;
3060
3061         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3062                                   &target_obj_id, &target_btf_id);
3063         seq_printf(seq,
3064                    "attach_type:\t%d\n"
3065                    "target_obj_id:\t%u\n"
3066                    "target_btf_id:\t%u\n",
3067                    tr_link->attach_type,
3068                    target_obj_id,
3069                    target_btf_id);
3070 }
3071
3072 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3073                                            struct bpf_link_info *info)
3074 {
3075         struct bpf_tracing_link *tr_link =
3076                 container_of(link, struct bpf_tracing_link, link.link);
3077
3078         info->tracing.attach_type = tr_link->attach_type;
3079         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3080                                   &info->tracing.target_obj_id,
3081                                   &info->tracing.target_btf_id);
3082
3083         return 0;
3084 }
3085
3086 static const struct bpf_link_ops bpf_tracing_link_lops = {
3087         .release = bpf_tracing_link_release,
3088         .dealloc = bpf_tracing_link_dealloc,
3089         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3090         .fill_link_info = bpf_tracing_link_fill_link_info,
3091 };
3092
3093 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3094                                    int tgt_prog_fd,
3095                                    u32 btf_id,
3096                                    u64 bpf_cookie)
3097 {
3098         struct bpf_link_primer link_primer;
3099         struct bpf_prog *tgt_prog = NULL;
3100         struct bpf_trampoline *tr = NULL;
3101         struct bpf_tracing_link *link;
3102         u64 key = 0;
3103         int err;
3104
3105         switch (prog->type) {
3106         case BPF_PROG_TYPE_TRACING:
3107                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3108                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3109                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3110                         err = -EINVAL;
3111                         goto out_put_prog;
3112                 }
3113                 break;
3114         case BPF_PROG_TYPE_EXT:
3115                 if (prog->expected_attach_type != 0) {
3116                         err = -EINVAL;
3117                         goto out_put_prog;
3118                 }
3119                 break;
3120         case BPF_PROG_TYPE_LSM:
3121                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3122                         err = -EINVAL;
3123                         goto out_put_prog;
3124                 }
3125                 break;
3126         default:
3127                 err = -EINVAL;
3128                 goto out_put_prog;
3129         }
3130
3131         if (!!tgt_prog_fd != !!btf_id) {
3132                 err = -EINVAL;
3133                 goto out_put_prog;
3134         }
3135
3136         if (tgt_prog_fd) {
3137                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3138                 if (prog->type != BPF_PROG_TYPE_EXT) {
3139                         err = -EINVAL;
3140                         goto out_put_prog;
3141                 }
3142
3143                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3144                 if (IS_ERR(tgt_prog)) {
3145                         err = PTR_ERR(tgt_prog);
3146                         tgt_prog = NULL;
3147                         goto out_put_prog;
3148                 }
3149
3150                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3151         }
3152
3153         link = kzalloc(sizeof(*link), GFP_USER);
3154         if (!link) {
3155                 err = -ENOMEM;
3156                 goto out_put_prog;
3157         }
3158         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3159                       &bpf_tracing_link_lops, prog);
3160         link->attach_type = prog->expected_attach_type;
3161         link->link.cookie = bpf_cookie;
3162
3163         mutex_lock(&prog->aux->dst_mutex);
3164
3165         /* There are a few possible cases here:
3166          *
3167          * - if prog->aux->dst_trampoline is set, the program was just loaded
3168          *   and not yet attached to anything, so we can use the values stored
3169          *   in prog->aux
3170          *
3171          * - if prog->aux->dst_trampoline is NULL, the program has already been
3172          *   attached to a target and its initial target was cleared (below)
3173          *
3174          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3175          *   target_btf_id using the link_create API.
3176          *
3177          * - if tgt_prog == NULL when this function was called using the old
3178          *   raw_tracepoint_open API, and we need a target from prog->aux
3179          *
3180          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3181          *   was detached and is going for re-attachment.
3182          */
3183         if (!prog->aux->dst_trampoline && !tgt_prog) {
3184                 /*
3185                  * Allow re-attach for TRACING and LSM programs. If it's
3186                  * currently linked, bpf_trampoline_link_prog will fail.
3187                  * EXT programs need to specify tgt_prog_fd, so they
3188                  * re-attach in separate code path.
3189                  */
3190                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3191                     prog->type != BPF_PROG_TYPE_LSM) {
3192                         err = -EINVAL;
3193                         goto out_unlock;
3194                 }
3195                 btf_id = prog->aux->attach_btf_id;
3196                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3197         }
3198
3199         if (!prog->aux->dst_trampoline ||
3200             (key && key != prog->aux->dst_trampoline->key)) {
3201                 /* If there is no saved target, or the specified target is
3202                  * different from the destination specified at load time, we
3203                  * need a new trampoline and a check for compatibility
3204                  */
3205                 struct bpf_attach_target_info tgt_info = {};
3206
3207                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3208                                               &tgt_info);
3209                 if (err)
3210                         goto out_unlock;
3211
3212                 if (tgt_info.tgt_mod) {
3213                         module_put(prog->aux->mod);
3214                         prog->aux->mod = tgt_info.tgt_mod;
3215                 }
3216
3217                 tr = bpf_trampoline_get(key, &tgt_info);
3218                 if (!tr) {
3219                         err = -ENOMEM;
3220                         goto out_unlock;
3221                 }
3222         } else {
3223                 /* The caller didn't specify a target, or the target was the
3224                  * same as the destination supplied during program load. This
3225                  * means we can reuse the trampoline and reference from program
3226                  * load time, and there is no need to allocate a new one. This
3227                  * can only happen once for any program, as the saved values in
3228                  * prog->aux are cleared below.
3229                  */
3230                 tr = prog->aux->dst_trampoline;
3231                 tgt_prog = prog->aux->dst_prog;
3232         }
3233
3234         err = bpf_link_prime(&link->link.link, &link_primer);
3235         if (err)
3236                 goto out_unlock;
3237
3238         err = bpf_trampoline_link_prog(&link->link, tr);
3239         if (err) {
3240                 bpf_link_cleanup(&link_primer);
3241                 link = NULL;
3242                 goto out_unlock;
3243         }
3244
3245         link->tgt_prog = tgt_prog;
3246         link->trampoline = tr;
3247
3248         /* Always clear the trampoline and target prog from prog->aux to make
3249          * sure the original attach destination is not kept alive after a
3250          * program is (re-)attached to another target.
3251          */
3252         if (prog->aux->dst_prog &&
3253             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3254                 /* got extra prog ref from syscall, or attaching to different prog */
3255                 bpf_prog_put(prog->aux->dst_prog);
3256         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3257                 /* we allocated a new trampoline, so free the old one */
3258                 bpf_trampoline_put(prog->aux->dst_trampoline);
3259
3260         prog->aux->dst_prog = NULL;
3261         prog->aux->dst_trampoline = NULL;
3262         mutex_unlock(&prog->aux->dst_mutex);
3263
3264         return bpf_link_settle(&link_primer);
3265 out_unlock:
3266         if (tr && tr != prog->aux->dst_trampoline)
3267                 bpf_trampoline_put(tr);
3268         mutex_unlock(&prog->aux->dst_mutex);
3269         kfree(link);
3270 out_put_prog:
3271         if (tgt_prog_fd && tgt_prog)
3272                 bpf_prog_put(tgt_prog);
3273         return err;
3274 }
3275
3276 struct bpf_raw_tp_link {
3277         struct bpf_link link;
3278         struct bpf_raw_event_map *btp;
3279 };
3280
3281 static void bpf_raw_tp_link_release(struct bpf_link *link)
3282 {
3283         struct bpf_raw_tp_link *raw_tp =
3284                 container_of(link, struct bpf_raw_tp_link, link);
3285
3286         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3287         bpf_put_raw_tracepoint(raw_tp->btp);
3288 }
3289
3290 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3291 {
3292         struct bpf_raw_tp_link *raw_tp =
3293                 container_of(link, struct bpf_raw_tp_link, link);
3294
3295         kfree(raw_tp);
3296 }
3297
3298 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3299                                         struct seq_file *seq)
3300 {
3301         struct bpf_raw_tp_link *raw_tp_link =
3302                 container_of(link, struct bpf_raw_tp_link, link);
3303
3304         seq_printf(seq,
3305                    "tp_name:\t%s\n",
3306                    raw_tp_link->btp->tp->name);
3307 }
3308
3309 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3310                             u32 len)
3311 {
3312         if (ulen >= len + 1) {
3313                 if (copy_to_user(ubuf, buf, len + 1))
3314                         return -EFAULT;
3315         } else {
3316                 char zero = '\0';
3317
3318                 if (copy_to_user(ubuf, buf, ulen - 1))
3319                         return -EFAULT;
3320                 if (put_user(zero, ubuf + ulen - 1))
3321                         return -EFAULT;
3322                 return -ENOSPC;
3323         }
3324
3325         return 0;
3326 }
3327
3328 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3329                                           struct bpf_link_info *info)
3330 {
3331         struct bpf_raw_tp_link *raw_tp_link =
3332                 container_of(link, struct bpf_raw_tp_link, link);
3333         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3334         const char *tp_name = raw_tp_link->btp->tp->name;
3335         u32 ulen = info->raw_tracepoint.tp_name_len;
3336         size_t tp_len = strlen(tp_name);
3337
3338         if (!ulen ^ !ubuf)
3339                 return -EINVAL;
3340
3341         info->raw_tracepoint.tp_name_len = tp_len + 1;
3342
3343         if (!ubuf)
3344                 return 0;
3345
3346         return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3347 }
3348
3349 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3350         .release = bpf_raw_tp_link_release,
3351         .dealloc = bpf_raw_tp_link_dealloc,
3352         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3353         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3354 };
3355
3356 #ifdef CONFIG_PERF_EVENTS
3357 struct bpf_perf_link {
3358         struct bpf_link link;
3359         struct file *perf_file;
3360 };
3361
3362 static void bpf_perf_link_release(struct bpf_link *link)
3363 {
3364         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3365         struct perf_event *event = perf_link->perf_file->private_data;
3366
3367         perf_event_free_bpf_prog(event);
3368         fput(perf_link->perf_file);
3369 }
3370
3371 static void bpf_perf_link_dealloc(struct bpf_link *link)
3372 {
3373         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3374
3375         kfree(perf_link);
3376 }
3377
3378 static int bpf_perf_link_fill_common(const struct perf_event *event,
3379                                      char __user *uname, u32 ulen,
3380                                      u64 *probe_offset, u64 *probe_addr,
3381                                      u32 *fd_type, unsigned long *missed)
3382 {
3383         const char *buf;
3384         u32 prog_id;
3385         size_t len;
3386         int err;
3387
3388         if (!ulen ^ !uname)
3389                 return -EINVAL;
3390
3391         err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3392                                       probe_offset, probe_addr, missed);
3393         if (err)
3394                 return err;
3395         if (!uname)
3396                 return 0;
3397         if (buf) {
3398                 len = strlen(buf);
3399                 err = bpf_copy_to_user(uname, buf, ulen, len);
3400                 if (err)
3401                         return err;
3402         } else {
3403                 char zero = '\0';
3404
3405                 if (put_user(zero, uname))
3406                         return -EFAULT;
3407         }
3408         return 0;
3409 }
3410
3411 #ifdef CONFIG_KPROBE_EVENTS
3412 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3413                                      struct bpf_link_info *info)
3414 {
3415         unsigned long missed;
3416         char __user *uname;
3417         u64 addr, offset;
3418         u32 ulen, type;
3419         int err;
3420
3421         uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3422         ulen = info->perf_event.kprobe.name_len;
3423         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3424                                         &type, &missed);
3425         if (err)
3426                 return err;
3427         if (type == BPF_FD_TYPE_KRETPROBE)
3428                 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3429         else
3430                 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3431
3432         info->perf_event.kprobe.offset = offset;
3433         info->perf_event.kprobe.missed = missed;
3434         if (!kallsyms_show_value(current_cred()))
3435                 addr = 0;
3436         info->perf_event.kprobe.addr = addr;
3437         return 0;
3438 }
3439 #endif
3440
3441 #ifdef CONFIG_UPROBE_EVENTS
3442 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3443                                      struct bpf_link_info *info)
3444 {
3445         char __user *uname;
3446         u64 addr, offset;
3447         u32 ulen, type;
3448         int err;
3449
3450         uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3451         ulen = info->perf_event.uprobe.name_len;
3452         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3453                                         &type, NULL);
3454         if (err)
3455                 return err;
3456
3457         if (type == BPF_FD_TYPE_URETPROBE)
3458                 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3459         else
3460                 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3461         info->perf_event.uprobe.offset = offset;
3462         return 0;
3463 }
3464 #endif
3465
3466 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3467                                     struct bpf_link_info *info)
3468 {
3469 #ifdef CONFIG_KPROBE_EVENTS
3470         if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3471                 return bpf_perf_link_fill_kprobe(event, info);
3472 #endif
3473 #ifdef CONFIG_UPROBE_EVENTS
3474         if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3475                 return bpf_perf_link_fill_uprobe(event, info);
3476 #endif
3477         return -EOPNOTSUPP;
3478 }
3479
3480 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3481                                          struct bpf_link_info *info)
3482 {
3483         char __user *uname;
3484         u32 ulen;
3485
3486         uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3487         ulen = info->perf_event.tracepoint.name_len;
3488         info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3489         return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3490 }
3491
3492 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3493                                          struct bpf_link_info *info)
3494 {
3495         info->perf_event.event.type = event->attr.type;
3496         info->perf_event.event.config = event->attr.config;
3497         info->perf_event.type = BPF_PERF_EVENT_EVENT;
3498         return 0;
3499 }
3500
3501 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3502                                         struct bpf_link_info *info)
3503 {
3504         struct bpf_perf_link *perf_link;
3505         const struct perf_event *event;
3506
3507         perf_link = container_of(link, struct bpf_perf_link, link);
3508         event = perf_get_event(perf_link->perf_file);
3509         if (IS_ERR(event))
3510                 return PTR_ERR(event);
3511
3512         switch (event->prog->type) {
3513         case BPF_PROG_TYPE_PERF_EVENT:
3514                 return bpf_perf_link_fill_perf_event(event, info);
3515         case BPF_PROG_TYPE_TRACEPOINT:
3516                 return bpf_perf_link_fill_tracepoint(event, info);
3517         case BPF_PROG_TYPE_KPROBE:
3518                 return bpf_perf_link_fill_probe(event, info);
3519         default:
3520                 return -EOPNOTSUPP;
3521         }
3522 }
3523
3524 static const struct bpf_link_ops bpf_perf_link_lops = {
3525         .release = bpf_perf_link_release,
3526         .dealloc = bpf_perf_link_dealloc,
3527         .fill_link_info = bpf_perf_link_fill_link_info,
3528 };
3529
3530 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3531 {
3532         struct bpf_link_primer link_primer;
3533         struct bpf_perf_link *link;
3534         struct perf_event *event;
3535         struct file *perf_file;
3536         int err;
3537
3538         if (attr->link_create.flags)
3539                 return -EINVAL;
3540
3541         perf_file = perf_event_get(attr->link_create.target_fd);
3542         if (IS_ERR(perf_file))
3543                 return PTR_ERR(perf_file);
3544
3545         link = kzalloc(sizeof(*link), GFP_USER);
3546         if (!link) {
3547                 err = -ENOMEM;
3548                 goto out_put_file;
3549         }
3550         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3551         link->perf_file = perf_file;
3552
3553         err = bpf_link_prime(&link->link, &link_primer);
3554         if (err) {
3555                 kfree(link);
3556                 goto out_put_file;
3557         }
3558
3559         event = perf_file->private_data;
3560         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3561         if (err) {
3562                 bpf_link_cleanup(&link_primer);
3563                 goto out_put_file;
3564         }
3565         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3566         bpf_prog_inc(prog);
3567
3568         return bpf_link_settle(&link_primer);
3569
3570 out_put_file:
3571         fput(perf_file);
3572         return err;
3573 }
3574 #else
3575 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3576 {
3577         return -EOPNOTSUPP;
3578 }
3579 #endif /* CONFIG_PERF_EVENTS */
3580
3581 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3582                                   const char __user *user_tp_name)
3583 {
3584         struct bpf_link_primer link_primer;
3585         struct bpf_raw_tp_link *link;
3586         struct bpf_raw_event_map *btp;
3587         const char *tp_name;
3588         char buf[128];
3589         int err;
3590
3591         switch (prog->type) {
3592         case BPF_PROG_TYPE_TRACING:
3593         case BPF_PROG_TYPE_EXT:
3594         case BPF_PROG_TYPE_LSM:
3595                 if (user_tp_name)
3596                         /* The attach point for this category of programs
3597                          * should be specified via btf_id during program load.
3598                          */
3599                         return -EINVAL;
3600                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3601                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3602                         tp_name = prog->aux->attach_func_name;
3603                         break;
3604                 }
3605                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3606         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3607         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3608                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3609                         return -EFAULT;
3610                 buf[sizeof(buf) - 1] = 0;
3611                 tp_name = buf;
3612                 break;
3613         default:
3614                 return -EINVAL;
3615         }
3616
3617         btp = bpf_get_raw_tracepoint(tp_name);
3618         if (!btp)
3619                 return -ENOENT;
3620
3621         link = kzalloc(sizeof(*link), GFP_USER);
3622         if (!link) {
3623                 err = -ENOMEM;
3624                 goto out_put_btp;
3625         }
3626         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3627                       &bpf_raw_tp_link_lops, prog);
3628         link->btp = btp;
3629
3630         err = bpf_link_prime(&link->link, &link_primer);
3631         if (err) {
3632                 kfree(link);
3633                 goto out_put_btp;
3634         }
3635
3636         err = bpf_probe_register(link->btp, prog);
3637         if (err) {
3638                 bpf_link_cleanup(&link_primer);
3639                 goto out_put_btp;
3640         }
3641
3642         return bpf_link_settle(&link_primer);
3643
3644 out_put_btp:
3645         bpf_put_raw_tracepoint(btp);
3646         return err;
3647 }
3648
3649 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3650
3651 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3652 {
3653         struct bpf_prog *prog;
3654         int fd;
3655
3656         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3657                 return -EINVAL;
3658
3659         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3660         if (IS_ERR(prog))
3661                 return PTR_ERR(prog);
3662
3663         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3664         if (fd < 0)
3665                 bpf_prog_put(prog);
3666         return fd;
3667 }
3668
3669 static enum bpf_prog_type
3670 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3671 {
3672         switch (attach_type) {
3673         case BPF_CGROUP_INET_INGRESS:
3674         case BPF_CGROUP_INET_EGRESS:
3675                 return BPF_PROG_TYPE_CGROUP_SKB;
3676         case BPF_CGROUP_INET_SOCK_CREATE:
3677         case BPF_CGROUP_INET_SOCK_RELEASE:
3678         case BPF_CGROUP_INET4_POST_BIND:
3679         case BPF_CGROUP_INET6_POST_BIND:
3680                 return BPF_PROG_TYPE_CGROUP_SOCK;
3681         case BPF_CGROUP_INET4_BIND:
3682         case BPF_CGROUP_INET6_BIND:
3683         case BPF_CGROUP_INET4_CONNECT:
3684         case BPF_CGROUP_INET6_CONNECT:
3685         case BPF_CGROUP_UNIX_CONNECT:
3686         case BPF_CGROUP_INET4_GETPEERNAME:
3687         case BPF_CGROUP_INET6_GETPEERNAME:
3688         case BPF_CGROUP_UNIX_GETPEERNAME:
3689         case BPF_CGROUP_INET4_GETSOCKNAME:
3690         case BPF_CGROUP_INET6_GETSOCKNAME:
3691         case BPF_CGROUP_UNIX_GETSOCKNAME:
3692         case BPF_CGROUP_UDP4_SENDMSG:
3693         case BPF_CGROUP_UDP6_SENDMSG:
3694         case BPF_CGROUP_UNIX_SENDMSG:
3695         case BPF_CGROUP_UDP4_RECVMSG:
3696         case BPF_CGROUP_UDP6_RECVMSG:
3697         case BPF_CGROUP_UNIX_RECVMSG:
3698                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3699         case BPF_CGROUP_SOCK_OPS:
3700                 return BPF_PROG_TYPE_SOCK_OPS;
3701         case BPF_CGROUP_DEVICE:
3702                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3703         case BPF_SK_MSG_VERDICT:
3704                 return BPF_PROG_TYPE_SK_MSG;
3705         case BPF_SK_SKB_STREAM_PARSER:
3706         case BPF_SK_SKB_STREAM_VERDICT:
3707         case BPF_SK_SKB_VERDICT:
3708                 return BPF_PROG_TYPE_SK_SKB;
3709         case BPF_LIRC_MODE2:
3710                 return BPF_PROG_TYPE_LIRC_MODE2;
3711         case BPF_FLOW_DISSECTOR:
3712                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3713         case BPF_CGROUP_SYSCTL:
3714                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3715         case BPF_CGROUP_GETSOCKOPT:
3716         case BPF_CGROUP_SETSOCKOPT:
3717                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3718         case BPF_TRACE_ITER:
3719         case BPF_TRACE_RAW_TP:
3720         case BPF_TRACE_FENTRY:
3721         case BPF_TRACE_FEXIT:
3722         case BPF_MODIFY_RETURN:
3723                 return BPF_PROG_TYPE_TRACING;
3724         case BPF_LSM_MAC:
3725                 return BPF_PROG_TYPE_LSM;
3726         case BPF_SK_LOOKUP:
3727                 return BPF_PROG_TYPE_SK_LOOKUP;
3728         case BPF_XDP:
3729                 return BPF_PROG_TYPE_XDP;
3730         case BPF_LSM_CGROUP:
3731                 return BPF_PROG_TYPE_LSM;
3732         case BPF_TCX_INGRESS:
3733         case BPF_TCX_EGRESS:
3734         case BPF_NETKIT_PRIMARY:
3735         case BPF_NETKIT_PEER:
3736                 return BPF_PROG_TYPE_SCHED_CLS;
3737         default:
3738                 return BPF_PROG_TYPE_UNSPEC;
3739         }
3740 }
3741
3742 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3743                                              enum bpf_attach_type attach_type)
3744 {
3745         enum bpf_prog_type ptype;
3746
3747         switch (prog->type) {
3748         case BPF_PROG_TYPE_CGROUP_SOCK:
3749         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3750         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3751         case BPF_PROG_TYPE_SK_LOOKUP:
3752                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3753         case BPF_PROG_TYPE_CGROUP_SKB:
3754                 if (!capable(CAP_NET_ADMIN))
3755                         /* cg-skb progs can be loaded by unpriv user.
3756                          * check permissions at attach time.
3757                          */
3758                         return -EPERM;
3759                 return prog->enforce_expected_attach_type &&
3760                         prog->expected_attach_type != attach_type ?
3761                         -EINVAL : 0;
3762         case BPF_PROG_TYPE_EXT:
3763                 return 0;
3764         case BPF_PROG_TYPE_NETFILTER:
3765                 if (attach_type != BPF_NETFILTER)
3766                         return -EINVAL;
3767                 return 0;
3768         case BPF_PROG_TYPE_PERF_EVENT:
3769         case BPF_PROG_TYPE_TRACEPOINT:
3770                 if (attach_type != BPF_PERF_EVENT)
3771                         return -EINVAL;
3772                 return 0;
3773         case BPF_PROG_TYPE_KPROBE:
3774                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3775                     attach_type != BPF_TRACE_KPROBE_MULTI)
3776                         return -EINVAL;
3777                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
3778                     attach_type != BPF_TRACE_UPROBE_MULTI)
3779                         return -EINVAL;
3780                 if (attach_type != BPF_PERF_EVENT &&
3781                     attach_type != BPF_TRACE_KPROBE_MULTI &&
3782                     attach_type != BPF_TRACE_UPROBE_MULTI)
3783                         return -EINVAL;
3784                 return 0;
3785         case BPF_PROG_TYPE_SCHED_CLS:
3786                 if (attach_type != BPF_TCX_INGRESS &&
3787                     attach_type != BPF_TCX_EGRESS &&
3788                     attach_type != BPF_NETKIT_PRIMARY &&
3789                     attach_type != BPF_NETKIT_PEER)
3790                         return -EINVAL;
3791                 return 0;
3792         default:
3793                 ptype = attach_type_to_prog_type(attach_type);
3794                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
3795                         return -EINVAL;
3796                 return 0;
3797         }
3798 }
3799
3800 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
3801
3802 #define BPF_F_ATTACH_MASK_BASE  \
3803         (BPF_F_ALLOW_OVERRIDE | \
3804          BPF_F_ALLOW_MULTI |    \
3805          BPF_F_REPLACE)
3806
3807 #define BPF_F_ATTACH_MASK_MPROG \
3808         (BPF_F_REPLACE |        \
3809          BPF_F_BEFORE |         \
3810          BPF_F_AFTER |          \
3811          BPF_F_ID |             \
3812          BPF_F_LINK)
3813
3814 static int bpf_prog_attach(const union bpf_attr *attr)
3815 {
3816         enum bpf_prog_type ptype;
3817         struct bpf_prog *prog;
3818         int ret;
3819
3820         if (CHECK_ATTR(BPF_PROG_ATTACH))
3821                 return -EINVAL;
3822
3823         ptype = attach_type_to_prog_type(attr->attach_type);
3824         if (ptype == BPF_PROG_TYPE_UNSPEC)
3825                 return -EINVAL;
3826         if (bpf_mprog_supported(ptype)) {
3827                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3828                         return -EINVAL;
3829         } else {
3830                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
3831                         return -EINVAL;
3832                 if (attr->relative_fd ||
3833                     attr->expected_revision)
3834                         return -EINVAL;
3835         }
3836
3837         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3838         if (IS_ERR(prog))
3839                 return PTR_ERR(prog);
3840
3841         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3842                 bpf_prog_put(prog);
3843                 return -EINVAL;
3844         }
3845
3846         switch (ptype) {
3847         case BPF_PROG_TYPE_SK_SKB:
3848         case BPF_PROG_TYPE_SK_MSG:
3849                 ret = sock_map_get_from_fd(attr, prog);
3850                 break;
3851         case BPF_PROG_TYPE_LIRC_MODE2:
3852                 ret = lirc_prog_attach(attr, prog);
3853                 break;
3854         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3855                 ret = netns_bpf_prog_attach(attr, prog);
3856                 break;
3857         case BPF_PROG_TYPE_CGROUP_DEVICE:
3858         case BPF_PROG_TYPE_CGROUP_SKB:
3859         case BPF_PROG_TYPE_CGROUP_SOCK:
3860         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3861         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3862         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3863         case BPF_PROG_TYPE_SOCK_OPS:
3864         case BPF_PROG_TYPE_LSM:
3865                 if (ptype == BPF_PROG_TYPE_LSM &&
3866                     prog->expected_attach_type != BPF_LSM_CGROUP)
3867                         ret = -EINVAL;
3868                 else
3869                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3870                 break;
3871         case BPF_PROG_TYPE_SCHED_CLS:
3872                 if (attr->attach_type == BPF_TCX_INGRESS ||
3873                     attr->attach_type == BPF_TCX_EGRESS)
3874                         ret = tcx_prog_attach(attr, prog);
3875                 else
3876                         ret = netkit_prog_attach(attr, prog);
3877                 break;
3878         default:
3879                 ret = -EINVAL;
3880         }
3881
3882         if (ret)
3883                 bpf_prog_put(prog);
3884         return ret;
3885 }
3886
3887 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
3888
3889 static int bpf_prog_detach(const union bpf_attr *attr)
3890 {
3891         struct bpf_prog *prog = NULL;
3892         enum bpf_prog_type ptype;
3893         int ret;
3894
3895         if (CHECK_ATTR(BPF_PROG_DETACH))
3896                 return -EINVAL;
3897
3898         ptype = attach_type_to_prog_type(attr->attach_type);
3899         if (bpf_mprog_supported(ptype)) {
3900                 if (ptype == BPF_PROG_TYPE_UNSPEC)
3901                         return -EINVAL;
3902                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3903                         return -EINVAL;
3904                 if (attr->attach_bpf_fd) {
3905                         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3906                         if (IS_ERR(prog))
3907                                 return PTR_ERR(prog);
3908                 }
3909         } else if (attr->attach_flags ||
3910                    attr->relative_fd ||
3911                    attr->expected_revision) {
3912                 return -EINVAL;
3913         }
3914
3915         switch (ptype) {
3916         case BPF_PROG_TYPE_SK_MSG:
3917         case BPF_PROG_TYPE_SK_SKB:
3918                 ret = sock_map_prog_detach(attr, ptype);
3919                 break;
3920         case BPF_PROG_TYPE_LIRC_MODE2:
3921                 ret = lirc_prog_detach(attr);
3922                 break;
3923         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3924                 ret = netns_bpf_prog_detach(attr, ptype);
3925                 break;
3926         case BPF_PROG_TYPE_CGROUP_DEVICE:
3927         case BPF_PROG_TYPE_CGROUP_SKB:
3928         case BPF_PROG_TYPE_CGROUP_SOCK:
3929         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3930         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3931         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3932         case BPF_PROG_TYPE_SOCK_OPS:
3933         case BPF_PROG_TYPE_LSM:
3934                 ret = cgroup_bpf_prog_detach(attr, ptype);
3935                 break;
3936         case BPF_PROG_TYPE_SCHED_CLS:
3937                 if (attr->attach_type == BPF_TCX_INGRESS ||
3938                     attr->attach_type == BPF_TCX_EGRESS)
3939                         ret = tcx_prog_detach(attr, prog);
3940                 else
3941                         ret = netkit_prog_detach(attr, prog);
3942                 break;
3943         default:
3944                 ret = -EINVAL;
3945         }
3946
3947         if (prog)
3948                 bpf_prog_put(prog);
3949         return ret;
3950 }
3951
3952 #define BPF_PROG_QUERY_LAST_FIELD query.revision
3953
3954 static int bpf_prog_query(const union bpf_attr *attr,
3955                           union bpf_attr __user *uattr)
3956 {
3957         if (!capable(CAP_NET_ADMIN))
3958                 return -EPERM;
3959         if (CHECK_ATTR(BPF_PROG_QUERY))
3960                 return -EINVAL;
3961         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3962                 return -EINVAL;
3963
3964         switch (attr->query.attach_type) {
3965         case BPF_CGROUP_INET_INGRESS:
3966         case BPF_CGROUP_INET_EGRESS:
3967         case BPF_CGROUP_INET_SOCK_CREATE:
3968         case BPF_CGROUP_INET_SOCK_RELEASE:
3969         case BPF_CGROUP_INET4_BIND:
3970         case BPF_CGROUP_INET6_BIND:
3971         case BPF_CGROUP_INET4_POST_BIND:
3972         case BPF_CGROUP_INET6_POST_BIND:
3973         case BPF_CGROUP_INET4_CONNECT:
3974         case BPF_CGROUP_INET6_CONNECT:
3975         case BPF_CGROUP_UNIX_CONNECT:
3976         case BPF_CGROUP_INET4_GETPEERNAME:
3977         case BPF_CGROUP_INET6_GETPEERNAME:
3978         case BPF_CGROUP_UNIX_GETPEERNAME:
3979         case BPF_CGROUP_INET4_GETSOCKNAME:
3980         case BPF_CGROUP_INET6_GETSOCKNAME:
3981         case BPF_CGROUP_UNIX_GETSOCKNAME:
3982         case BPF_CGROUP_UDP4_SENDMSG:
3983         case BPF_CGROUP_UDP6_SENDMSG:
3984         case BPF_CGROUP_UNIX_SENDMSG:
3985         case BPF_CGROUP_UDP4_RECVMSG:
3986         case BPF_CGROUP_UDP6_RECVMSG:
3987         case BPF_CGROUP_UNIX_RECVMSG:
3988         case BPF_CGROUP_SOCK_OPS:
3989         case BPF_CGROUP_DEVICE:
3990         case BPF_CGROUP_SYSCTL:
3991         case BPF_CGROUP_GETSOCKOPT:
3992         case BPF_CGROUP_SETSOCKOPT:
3993         case BPF_LSM_CGROUP:
3994                 return cgroup_bpf_prog_query(attr, uattr);
3995         case BPF_LIRC_MODE2:
3996                 return lirc_prog_query(attr, uattr);
3997         case BPF_FLOW_DISSECTOR:
3998         case BPF_SK_LOOKUP:
3999                 return netns_bpf_prog_query(attr, uattr);
4000         case BPF_SK_SKB_STREAM_PARSER:
4001         case BPF_SK_SKB_STREAM_VERDICT:
4002         case BPF_SK_MSG_VERDICT:
4003         case BPF_SK_SKB_VERDICT:
4004                 return sock_map_bpf_prog_query(attr, uattr);
4005         case BPF_TCX_INGRESS:
4006         case BPF_TCX_EGRESS:
4007                 return tcx_prog_query(attr, uattr);
4008         case BPF_NETKIT_PRIMARY:
4009         case BPF_NETKIT_PEER:
4010                 return netkit_prog_query(attr, uattr);
4011         default:
4012                 return -EINVAL;
4013         }
4014 }
4015
4016 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4017
4018 static int bpf_prog_test_run(const union bpf_attr *attr,
4019                              union bpf_attr __user *uattr)
4020 {
4021         struct bpf_prog *prog;
4022         int ret = -ENOTSUPP;
4023
4024         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4025                 return -EINVAL;
4026
4027         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4028             (!attr->test.ctx_size_in && attr->test.ctx_in))
4029                 return -EINVAL;
4030
4031         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4032             (!attr->test.ctx_size_out && attr->test.ctx_out))
4033                 return -EINVAL;
4034
4035         prog = bpf_prog_get(attr->test.prog_fd);
4036         if (IS_ERR(prog))
4037                 return PTR_ERR(prog);
4038
4039         if (prog->aux->ops->test_run)
4040                 ret = prog->aux->ops->test_run(prog, attr, uattr);
4041
4042         bpf_prog_put(prog);
4043         return ret;
4044 }
4045
4046 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4047
4048 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4049                                union bpf_attr __user *uattr,
4050                                struct idr *idr,
4051                                spinlock_t *lock)
4052 {
4053         u32 next_id = attr->start_id;
4054         int err = 0;
4055
4056         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4057                 return -EINVAL;
4058
4059         if (!capable(CAP_SYS_ADMIN))
4060                 return -EPERM;
4061
4062         next_id++;
4063         spin_lock_bh(lock);
4064         if (!idr_get_next(idr, &next_id))
4065                 err = -ENOENT;
4066         spin_unlock_bh(lock);
4067
4068         if (!err)
4069                 err = put_user(next_id, &uattr->next_id);
4070
4071         return err;
4072 }
4073
4074 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4075 {
4076         struct bpf_map *map;
4077
4078         spin_lock_bh(&map_idr_lock);
4079 again:
4080         map = idr_get_next(&map_idr, id);
4081         if (map) {
4082                 map = __bpf_map_inc_not_zero(map, false);
4083                 if (IS_ERR(map)) {
4084                         (*id)++;
4085                         goto again;
4086                 }
4087         }
4088         spin_unlock_bh(&map_idr_lock);
4089
4090         return map;
4091 }
4092
4093 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4094 {
4095         struct bpf_prog *prog;
4096
4097         spin_lock_bh(&prog_idr_lock);
4098 again:
4099         prog = idr_get_next(&prog_idr, id);
4100         if (prog) {
4101                 prog = bpf_prog_inc_not_zero(prog);
4102                 if (IS_ERR(prog)) {
4103                         (*id)++;
4104                         goto again;
4105                 }
4106         }
4107         spin_unlock_bh(&prog_idr_lock);
4108
4109         return prog;
4110 }
4111
4112 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4113
4114 struct bpf_prog *bpf_prog_by_id(u32 id)
4115 {
4116         struct bpf_prog *prog;
4117
4118         if (!id)
4119                 return ERR_PTR(-ENOENT);
4120
4121         spin_lock_bh(&prog_idr_lock);
4122         prog = idr_find(&prog_idr, id);
4123         if (prog)
4124                 prog = bpf_prog_inc_not_zero(prog);
4125         else
4126                 prog = ERR_PTR(-ENOENT);
4127         spin_unlock_bh(&prog_idr_lock);
4128         return prog;
4129 }
4130
4131 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4132 {
4133         struct bpf_prog *prog;
4134         u32 id = attr->prog_id;
4135         int fd;
4136
4137         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4138                 return -EINVAL;
4139
4140         if (!capable(CAP_SYS_ADMIN))
4141                 return -EPERM;
4142
4143         prog = bpf_prog_by_id(id);
4144         if (IS_ERR(prog))
4145                 return PTR_ERR(prog);
4146
4147         fd = bpf_prog_new_fd(prog);
4148         if (fd < 0)
4149                 bpf_prog_put(prog);
4150
4151         return fd;
4152 }
4153
4154 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4155
4156 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4157 {
4158         struct bpf_map *map;
4159         u32 id = attr->map_id;
4160         int f_flags;
4161         int fd;
4162
4163         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4164             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4165                 return -EINVAL;
4166
4167         if (!capable(CAP_SYS_ADMIN))
4168                 return -EPERM;
4169
4170         f_flags = bpf_get_file_flag(attr->open_flags);
4171         if (f_flags < 0)
4172                 return f_flags;
4173
4174         spin_lock_bh(&map_idr_lock);
4175         map = idr_find(&map_idr, id);
4176         if (map)
4177                 map = __bpf_map_inc_not_zero(map, true);
4178         else
4179                 map = ERR_PTR(-ENOENT);
4180         spin_unlock_bh(&map_idr_lock);
4181
4182         if (IS_ERR(map))
4183                 return PTR_ERR(map);
4184
4185         fd = bpf_map_new_fd(map, f_flags);
4186         if (fd < 0)
4187                 bpf_map_put_with_uref(map);
4188
4189         return fd;
4190 }
4191
4192 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4193                                               unsigned long addr, u32 *off,
4194                                               u32 *type)
4195 {
4196         const struct bpf_map *map;
4197         int i;
4198
4199         mutex_lock(&prog->aux->used_maps_mutex);
4200         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4201                 map = prog->aux->used_maps[i];
4202                 if (map == (void *)addr) {
4203                         *type = BPF_PSEUDO_MAP_FD;
4204                         goto out;
4205                 }
4206                 if (!map->ops->map_direct_value_meta)
4207                         continue;
4208                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4209                         *type = BPF_PSEUDO_MAP_VALUE;
4210                         goto out;
4211                 }
4212         }
4213         map = NULL;
4214
4215 out:
4216         mutex_unlock(&prog->aux->used_maps_mutex);
4217         return map;
4218 }
4219
4220 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4221                                               const struct cred *f_cred)
4222 {
4223         const struct bpf_map *map;
4224         struct bpf_insn *insns;
4225         u32 off, type;
4226         u64 imm;
4227         u8 code;
4228         int i;
4229
4230         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4231                         GFP_USER);
4232         if (!insns)
4233                 return insns;
4234
4235         for (i = 0; i < prog->len; i++) {
4236                 code = insns[i].code;
4237
4238                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4239                         insns[i].code = BPF_JMP | BPF_CALL;
4240                         insns[i].imm = BPF_FUNC_tail_call;
4241                         /* fall-through */
4242                 }
4243                 if (code == (BPF_JMP | BPF_CALL) ||
4244                     code == (BPF_JMP | BPF_CALL_ARGS)) {
4245                         if (code == (BPF_JMP | BPF_CALL_ARGS))
4246                                 insns[i].code = BPF_JMP | BPF_CALL;
4247                         if (!bpf_dump_raw_ok(f_cred))
4248                                 insns[i].imm = 0;
4249                         continue;
4250                 }
4251                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4252                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4253                         continue;
4254                 }
4255
4256                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4257                         continue;
4258
4259                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4260                 map = bpf_map_from_imm(prog, imm, &off, &type);
4261                 if (map) {
4262                         insns[i].src_reg = type;
4263                         insns[i].imm = map->id;
4264                         insns[i + 1].imm = off;
4265                         continue;
4266                 }
4267         }
4268
4269         return insns;
4270 }
4271
4272 static int set_info_rec_size(struct bpf_prog_info *info)
4273 {
4274         /*
4275          * Ensure info.*_rec_size is the same as kernel expected size
4276          *
4277          * or
4278          *
4279          * Only allow zero *_rec_size if both _rec_size and _cnt are
4280          * zero.  In this case, the kernel will set the expected
4281          * _rec_size back to the info.
4282          */
4283
4284         if ((info->nr_func_info || info->func_info_rec_size) &&
4285             info->func_info_rec_size != sizeof(struct bpf_func_info))
4286                 return -EINVAL;
4287
4288         if ((info->nr_line_info || info->line_info_rec_size) &&
4289             info->line_info_rec_size != sizeof(struct bpf_line_info))
4290                 return -EINVAL;
4291
4292         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4293             info->jited_line_info_rec_size != sizeof(__u64))
4294                 return -EINVAL;
4295
4296         info->func_info_rec_size = sizeof(struct bpf_func_info);
4297         info->line_info_rec_size = sizeof(struct bpf_line_info);
4298         info->jited_line_info_rec_size = sizeof(__u64);
4299
4300         return 0;
4301 }
4302
4303 static int bpf_prog_get_info_by_fd(struct file *file,
4304                                    struct bpf_prog *prog,
4305                                    const union bpf_attr *attr,
4306                                    union bpf_attr __user *uattr)
4307 {
4308         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4309         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4310         struct bpf_prog_info info;
4311         u32 info_len = attr->info.info_len;
4312         struct bpf_prog_kstats stats;
4313         char __user *uinsns;
4314         u32 ulen;
4315         int err;
4316
4317         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4318         if (err)
4319                 return err;
4320         info_len = min_t(u32, sizeof(info), info_len);
4321
4322         memset(&info, 0, sizeof(info));
4323         if (copy_from_user(&info, uinfo, info_len))
4324                 return -EFAULT;
4325
4326         info.type = prog->type;
4327         info.id = prog->aux->id;
4328         info.load_time = prog->aux->load_time;
4329         info.created_by_uid = from_kuid_munged(current_user_ns(),
4330                                                prog->aux->user->uid);
4331         info.gpl_compatible = prog->gpl_compatible;
4332
4333         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4334         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4335
4336         mutex_lock(&prog->aux->used_maps_mutex);
4337         ulen = info.nr_map_ids;
4338         info.nr_map_ids = prog->aux->used_map_cnt;
4339         ulen = min_t(u32, info.nr_map_ids, ulen);
4340         if (ulen) {
4341                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4342                 u32 i;
4343
4344                 for (i = 0; i < ulen; i++)
4345                         if (put_user(prog->aux->used_maps[i]->id,
4346                                      &user_map_ids[i])) {
4347                                 mutex_unlock(&prog->aux->used_maps_mutex);
4348                                 return -EFAULT;
4349                         }
4350         }
4351         mutex_unlock(&prog->aux->used_maps_mutex);
4352
4353         err = set_info_rec_size(&info);
4354         if (err)
4355                 return err;
4356
4357         bpf_prog_get_stats(prog, &stats);
4358         info.run_time_ns = stats.nsecs;
4359         info.run_cnt = stats.cnt;
4360         info.recursion_misses = stats.misses;
4361
4362         info.verified_insns = prog->aux->verified_insns;
4363
4364         if (!bpf_capable()) {
4365                 info.jited_prog_len = 0;
4366                 info.xlated_prog_len = 0;
4367                 info.nr_jited_ksyms = 0;
4368                 info.nr_jited_func_lens = 0;
4369                 info.nr_func_info = 0;
4370                 info.nr_line_info = 0;
4371                 info.nr_jited_line_info = 0;
4372                 goto done;
4373         }
4374
4375         ulen = info.xlated_prog_len;
4376         info.xlated_prog_len = bpf_prog_insn_size(prog);
4377         if (info.xlated_prog_len && ulen) {
4378                 struct bpf_insn *insns_sanitized;
4379                 bool fault;
4380
4381                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4382                         info.xlated_prog_insns = 0;
4383                         goto done;
4384                 }
4385                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4386                 if (!insns_sanitized)
4387                         return -ENOMEM;
4388                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4389                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4390                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4391                 kfree(insns_sanitized);
4392                 if (fault)
4393                         return -EFAULT;
4394         }
4395
4396         if (bpf_prog_is_offloaded(prog->aux)) {
4397                 err = bpf_prog_offload_info_fill(&info, prog);
4398                 if (err)
4399                         return err;
4400                 goto done;
4401         }
4402
4403         /* NOTE: the following code is supposed to be skipped for offload.
4404          * bpf_prog_offload_info_fill() is the place to fill similar fields
4405          * for offload.
4406          */
4407         ulen = info.jited_prog_len;
4408         if (prog->aux->func_cnt) {
4409                 u32 i;
4410
4411                 info.jited_prog_len = 0;
4412                 for (i = 0; i < prog->aux->func_cnt; i++)
4413                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4414         } else {
4415                 info.jited_prog_len = prog->jited_len;
4416         }
4417
4418         if (info.jited_prog_len && ulen) {
4419                 if (bpf_dump_raw_ok(file->f_cred)) {
4420                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4421                         ulen = min_t(u32, info.jited_prog_len, ulen);
4422
4423                         /* for multi-function programs, copy the JITed
4424                          * instructions for all the functions
4425                          */
4426                         if (prog->aux->func_cnt) {
4427                                 u32 len, free, i;
4428                                 u8 *img;
4429
4430                                 free = ulen;
4431                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4432                                         len = prog->aux->func[i]->jited_len;
4433                                         len = min_t(u32, len, free);
4434                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4435                                         if (copy_to_user(uinsns, img, len))
4436                                                 return -EFAULT;
4437                                         uinsns += len;
4438                                         free -= len;
4439                                         if (!free)
4440                                                 break;
4441                                 }
4442                         } else {
4443                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4444                                         return -EFAULT;
4445                         }
4446                 } else {
4447                         info.jited_prog_insns = 0;
4448                 }
4449         }
4450
4451         ulen = info.nr_jited_ksyms;
4452         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4453         if (ulen) {
4454                 if (bpf_dump_raw_ok(file->f_cred)) {
4455                         unsigned long ksym_addr;
4456                         u64 __user *user_ksyms;
4457                         u32 i;
4458
4459                         /* copy the address of the kernel symbol
4460                          * corresponding to each function
4461                          */
4462                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4463                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4464                         if (prog->aux->func_cnt) {
4465                                 for (i = 0; i < ulen; i++) {
4466                                         ksym_addr = (unsigned long)
4467                                                 prog->aux->func[i]->bpf_func;
4468                                         if (put_user((u64) ksym_addr,
4469                                                      &user_ksyms[i]))
4470                                                 return -EFAULT;
4471                                 }
4472                         } else {
4473                                 ksym_addr = (unsigned long) prog->bpf_func;
4474                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4475                                         return -EFAULT;
4476                         }
4477                 } else {
4478                         info.jited_ksyms = 0;
4479                 }
4480         }
4481
4482         ulen = info.nr_jited_func_lens;
4483         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4484         if (ulen) {
4485                 if (bpf_dump_raw_ok(file->f_cred)) {
4486                         u32 __user *user_lens;
4487                         u32 func_len, i;
4488
4489                         /* copy the JITed image lengths for each function */
4490                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4491                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4492                         if (prog->aux->func_cnt) {
4493                                 for (i = 0; i < ulen; i++) {
4494                                         func_len =
4495                                                 prog->aux->func[i]->jited_len;
4496                                         if (put_user(func_len, &user_lens[i]))
4497                                                 return -EFAULT;
4498                                 }
4499                         } else {
4500                                 func_len = prog->jited_len;
4501                                 if (put_user(func_len, &user_lens[0]))
4502                                         return -EFAULT;
4503                         }
4504                 } else {
4505                         info.jited_func_lens = 0;
4506                 }
4507         }
4508
4509         if (prog->aux->btf)
4510                 info.btf_id = btf_obj_id(prog->aux->btf);
4511         info.attach_btf_id = prog->aux->attach_btf_id;
4512         if (attach_btf)
4513                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4514
4515         ulen = info.nr_func_info;
4516         info.nr_func_info = prog->aux->func_info_cnt;
4517         if (info.nr_func_info && ulen) {
4518                 char __user *user_finfo;
4519
4520                 user_finfo = u64_to_user_ptr(info.func_info);
4521                 ulen = min_t(u32, info.nr_func_info, ulen);
4522                 if (copy_to_user(user_finfo, prog->aux->func_info,
4523                                  info.func_info_rec_size * ulen))
4524                         return -EFAULT;
4525         }
4526
4527         ulen = info.nr_line_info;
4528         info.nr_line_info = prog->aux->nr_linfo;
4529         if (info.nr_line_info && ulen) {
4530                 __u8 __user *user_linfo;
4531
4532                 user_linfo = u64_to_user_ptr(info.line_info);
4533                 ulen = min_t(u32, info.nr_line_info, ulen);
4534                 if (copy_to_user(user_linfo, prog->aux->linfo,
4535                                  info.line_info_rec_size * ulen))
4536                         return -EFAULT;
4537         }
4538
4539         ulen = info.nr_jited_line_info;
4540         if (prog->aux->jited_linfo)
4541                 info.nr_jited_line_info = prog->aux->nr_linfo;
4542         else
4543                 info.nr_jited_line_info = 0;
4544         if (info.nr_jited_line_info && ulen) {
4545                 if (bpf_dump_raw_ok(file->f_cred)) {
4546                         unsigned long line_addr;
4547                         __u64 __user *user_linfo;
4548                         u32 i;
4549
4550                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4551                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4552                         for (i = 0; i < ulen; i++) {
4553                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4554                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4555                                         return -EFAULT;
4556                         }
4557                 } else {
4558                         info.jited_line_info = 0;
4559                 }
4560         }
4561
4562         ulen = info.nr_prog_tags;
4563         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4564         if (ulen) {
4565                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4566                 u32 i;
4567
4568                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4569                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4570                 if (prog->aux->func_cnt) {
4571                         for (i = 0; i < ulen; i++) {
4572                                 if (copy_to_user(user_prog_tags[i],
4573                                                  prog->aux->func[i]->tag,
4574                                                  BPF_TAG_SIZE))
4575                                         return -EFAULT;
4576                         }
4577                 } else {
4578                         if (copy_to_user(user_prog_tags[0],
4579                                          prog->tag, BPF_TAG_SIZE))
4580                                 return -EFAULT;
4581                 }
4582         }
4583
4584 done:
4585         if (copy_to_user(uinfo, &info, info_len) ||
4586             put_user(info_len, &uattr->info.info_len))
4587                 return -EFAULT;
4588
4589         return 0;
4590 }
4591
4592 static int bpf_map_get_info_by_fd(struct file *file,
4593                                   struct bpf_map *map,
4594                                   const union bpf_attr *attr,
4595                                   union bpf_attr __user *uattr)
4596 {
4597         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4598         struct bpf_map_info info;
4599         u32 info_len = attr->info.info_len;
4600         int err;
4601
4602         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4603         if (err)
4604                 return err;
4605         info_len = min_t(u32, sizeof(info), info_len);
4606
4607         memset(&info, 0, sizeof(info));
4608         info.type = map->map_type;
4609         info.id = map->id;
4610         info.key_size = map->key_size;
4611         info.value_size = map->value_size;
4612         info.max_entries = map->max_entries;
4613         info.map_flags = map->map_flags;
4614         info.map_extra = map->map_extra;
4615         memcpy(info.name, map->name, sizeof(map->name));
4616
4617         if (map->btf) {
4618                 info.btf_id = btf_obj_id(map->btf);
4619                 info.btf_key_type_id = map->btf_key_type_id;
4620                 info.btf_value_type_id = map->btf_value_type_id;
4621         }
4622         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4623
4624         if (bpf_map_is_offloaded(map)) {
4625                 err = bpf_map_offload_info_fill(&info, map);
4626                 if (err)
4627                         return err;
4628         }
4629
4630         if (copy_to_user(uinfo, &info, info_len) ||
4631             put_user(info_len, &uattr->info.info_len))
4632                 return -EFAULT;
4633
4634         return 0;
4635 }
4636
4637 static int bpf_btf_get_info_by_fd(struct file *file,
4638                                   struct btf *btf,
4639                                   const union bpf_attr *attr,
4640                                   union bpf_attr __user *uattr)
4641 {
4642         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4643         u32 info_len = attr->info.info_len;
4644         int err;
4645
4646         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4647         if (err)
4648                 return err;
4649
4650         return btf_get_info_by_fd(btf, attr, uattr);
4651 }
4652
4653 static int bpf_link_get_info_by_fd(struct file *file,
4654                                   struct bpf_link *link,
4655                                   const union bpf_attr *attr,
4656                                   union bpf_attr __user *uattr)
4657 {
4658         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4659         struct bpf_link_info info;
4660         u32 info_len = attr->info.info_len;
4661         int err;
4662
4663         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4664         if (err)
4665                 return err;
4666         info_len = min_t(u32, sizeof(info), info_len);
4667
4668         memset(&info, 0, sizeof(info));
4669         if (copy_from_user(&info, uinfo, info_len))
4670                 return -EFAULT;
4671
4672         info.type = link->type;
4673         info.id = link->id;
4674         if (link->prog)
4675                 info.prog_id = link->prog->aux->id;
4676
4677         if (link->ops->fill_link_info) {
4678                 err = link->ops->fill_link_info(link, &info);
4679                 if (err)
4680                         return err;
4681         }
4682
4683         if (copy_to_user(uinfo, &info, info_len) ||
4684             put_user(info_len, &uattr->info.info_len))
4685                 return -EFAULT;
4686
4687         return 0;
4688 }
4689
4690
4691 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4692
4693 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4694                                   union bpf_attr __user *uattr)
4695 {
4696         int ufd = attr->info.bpf_fd;
4697         struct fd f;
4698         int err;
4699
4700         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4701                 return -EINVAL;
4702
4703         f = fdget(ufd);
4704         if (!f.file)
4705                 return -EBADFD;
4706
4707         if (f.file->f_op == &bpf_prog_fops)
4708                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4709                                               uattr);
4710         else if (f.file->f_op == &bpf_map_fops)
4711                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4712                                              uattr);
4713         else if (f.file->f_op == &btf_fops)
4714                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4715         else if (f.file->f_op == &bpf_link_fops)
4716                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4717                                               attr, uattr);
4718         else
4719                 err = -EINVAL;
4720
4721         fdput(f);
4722         return err;
4723 }
4724
4725 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4726
4727 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4728 {
4729         if (CHECK_ATTR(BPF_BTF_LOAD))
4730                 return -EINVAL;
4731
4732         if (!bpf_capable())
4733                 return -EPERM;
4734
4735         return btf_new_fd(attr, uattr, uattr_size);
4736 }
4737
4738 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4739
4740 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4741 {
4742         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4743                 return -EINVAL;
4744
4745         if (!capable(CAP_SYS_ADMIN))
4746                 return -EPERM;
4747
4748         return btf_get_fd_by_id(attr->btf_id);
4749 }
4750
4751 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4752                                     union bpf_attr __user *uattr,
4753                                     u32 prog_id, u32 fd_type,
4754                                     const char *buf, u64 probe_offset,
4755                                     u64 probe_addr)
4756 {
4757         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4758         u32 len = buf ? strlen(buf) : 0, input_len;
4759         int err = 0;
4760
4761         if (put_user(len, &uattr->task_fd_query.buf_len))
4762                 return -EFAULT;
4763         input_len = attr->task_fd_query.buf_len;
4764         if (input_len && ubuf) {
4765                 if (!len) {
4766                         /* nothing to copy, just make ubuf NULL terminated */
4767                         char zero = '\0';
4768
4769                         if (put_user(zero, ubuf))
4770                                 return -EFAULT;
4771                 } else if (input_len >= len + 1) {
4772                         /* ubuf can hold the string with NULL terminator */
4773                         if (copy_to_user(ubuf, buf, len + 1))
4774                                 return -EFAULT;
4775                 } else {
4776                         /* ubuf cannot hold the string with NULL terminator,
4777                          * do a partial copy with NULL terminator.
4778                          */
4779                         char zero = '\0';
4780
4781                         err = -ENOSPC;
4782                         if (copy_to_user(ubuf, buf, input_len - 1))
4783                                 return -EFAULT;
4784                         if (put_user(zero, ubuf + input_len - 1))
4785                                 return -EFAULT;
4786                 }
4787         }
4788
4789         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4790             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4791             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4792             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4793                 return -EFAULT;
4794
4795         return err;
4796 }
4797
4798 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4799
4800 static int bpf_task_fd_query(const union bpf_attr *attr,
4801                              union bpf_attr __user *uattr)
4802 {
4803         pid_t pid = attr->task_fd_query.pid;
4804         u32 fd = attr->task_fd_query.fd;
4805         const struct perf_event *event;
4806         struct task_struct *task;
4807         struct file *file;
4808         int err;
4809
4810         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4811                 return -EINVAL;
4812
4813         if (!capable(CAP_SYS_ADMIN))
4814                 return -EPERM;
4815
4816         if (attr->task_fd_query.flags != 0)
4817                 return -EINVAL;
4818
4819         rcu_read_lock();
4820         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4821         rcu_read_unlock();
4822         if (!task)
4823                 return -ENOENT;
4824
4825         err = 0;
4826         file = fget_task(task, fd);
4827         put_task_struct(task);
4828         if (!file)
4829                 return -EBADF;
4830
4831         if (file->f_op == &bpf_link_fops) {
4832                 struct bpf_link *link = file->private_data;
4833
4834                 if (link->ops == &bpf_raw_tp_link_lops) {
4835                         struct bpf_raw_tp_link *raw_tp =
4836                                 container_of(link, struct bpf_raw_tp_link, link);
4837                         struct bpf_raw_event_map *btp = raw_tp->btp;
4838
4839                         err = bpf_task_fd_query_copy(attr, uattr,
4840                                                      raw_tp->link.prog->aux->id,
4841                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4842                                                      btp->tp->name, 0, 0);
4843                         goto put_file;
4844                 }
4845                 goto out_not_supp;
4846         }
4847
4848         event = perf_get_event(file);
4849         if (!IS_ERR(event)) {
4850                 u64 probe_offset, probe_addr;
4851                 u32 prog_id, fd_type;
4852                 const char *buf;
4853
4854                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4855                                               &buf, &probe_offset,
4856                                               &probe_addr, NULL);
4857                 if (!err)
4858                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4859                                                      fd_type, buf,
4860                                                      probe_offset,
4861                                                      probe_addr);
4862                 goto put_file;
4863         }
4864
4865 out_not_supp:
4866         err = -ENOTSUPP;
4867 put_file:
4868         fput(file);
4869         return err;
4870 }
4871
4872 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4873
4874 #define BPF_DO_BATCH(fn, ...)                   \
4875         do {                                    \
4876                 if (!fn) {                      \
4877                         err = -ENOTSUPP;        \
4878                         goto err_put;           \
4879                 }                               \
4880                 err = fn(__VA_ARGS__);          \
4881         } while (0)
4882
4883 static int bpf_map_do_batch(const union bpf_attr *attr,
4884                             union bpf_attr __user *uattr,
4885                             int cmd)
4886 {
4887         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4888                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4889         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4890         struct bpf_map *map;
4891         int err, ufd;
4892         struct fd f;
4893
4894         if (CHECK_ATTR(BPF_MAP_BATCH))
4895                 return -EINVAL;
4896
4897         ufd = attr->batch.map_fd;
4898         f = fdget(ufd);
4899         map = __bpf_map_get(f);
4900         if (IS_ERR(map))
4901                 return PTR_ERR(map);
4902         if (has_write)
4903                 bpf_map_write_active_inc(map);
4904         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4905                 err = -EPERM;
4906                 goto err_put;
4907         }
4908         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4909                 err = -EPERM;
4910                 goto err_put;
4911         }
4912
4913         if (cmd == BPF_MAP_LOOKUP_BATCH)
4914                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4915         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4916                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4917         else if (cmd == BPF_MAP_UPDATE_BATCH)
4918                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4919         else
4920                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4921 err_put:
4922         if (has_write)
4923                 bpf_map_write_active_dec(map);
4924         fdput(f);
4925         return err;
4926 }
4927
4928 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
4929 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4930 {
4931         struct bpf_prog *prog;
4932         int ret;
4933
4934         if (CHECK_ATTR(BPF_LINK_CREATE))
4935                 return -EINVAL;
4936
4937         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
4938                 return bpf_struct_ops_link_create(attr);
4939
4940         prog = bpf_prog_get(attr->link_create.prog_fd);
4941         if (IS_ERR(prog))
4942                 return PTR_ERR(prog);
4943
4944         ret = bpf_prog_attach_check_attach_type(prog,
4945                                                 attr->link_create.attach_type);
4946         if (ret)
4947                 goto out;
4948
4949         switch (prog->type) {
4950         case BPF_PROG_TYPE_CGROUP_SKB:
4951         case BPF_PROG_TYPE_CGROUP_SOCK:
4952         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4953         case BPF_PROG_TYPE_SOCK_OPS:
4954         case BPF_PROG_TYPE_CGROUP_DEVICE:
4955         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4956         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4957                 ret = cgroup_bpf_link_attach(attr, prog);
4958                 break;
4959         case BPF_PROG_TYPE_EXT:
4960                 ret = bpf_tracing_prog_attach(prog,
4961                                               attr->link_create.target_fd,
4962                                               attr->link_create.target_btf_id,
4963                                               attr->link_create.tracing.cookie);
4964                 break;
4965         case BPF_PROG_TYPE_LSM:
4966         case BPF_PROG_TYPE_TRACING:
4967                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4968                         ret = -EINVAL;
4969                         goto out;
4970                 }
4971                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4972                         ret = bpf_raw_tp_link_attach(prog, NULL);
4973                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4974                         ret = bpf_iter_link_attach(attr, uattr, prog);
4975                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4976                         ret = cgroup_bpf_link_attach(attr, prog);
4977                 else
4978                         ret = bpf_tracing_prog_attach(prog,
4979                                                       attr->link_create.target_fd,
4980                                                       attr->link_create.target_btf_id,
4981                                                       attr->link_create.tracing.cookie);
4982                 break;
4983         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4984         case BPF_PROG_TYPE_SK_LOOKUP:
4985                 ret = netns_bpf_link_create(attr, prog);
4986                 break;
4987 #ifdef CONFIG_NET
4988         case BPF_PROG_TYPE_XDP:
4989                 ret = bpf_xdp_link_attach(attr, prog);
4990                 break;
4991         case BPF_PROG_TYPE_SCHED_CLS:
4992                 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
4993                     attr->link_create.attach_type == BPF_TCX_EGRESS)
4994                         ret = tcx_link_attach(attr, prog);
4995                 else
4996                         ret = netkit_link_attach(attr, prog);
4997                 break;
4998         case BPF_PROG_TYPE_NETFILTER:
4999                 ret = bpf_nf_link_attach(attr, prog);
5000                 break;
5001 #endif
5002         case BPF_PROG_TYPE_PERF_EVENT:
5003         case BPF_PROG_TYPE_TRACEPOINT:
5004                 ret = bpf_perf_link_attach(attr, prog);
5005                 break;
5006         case BPF_PROG_TYPE_KPROBE:
5007                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5008                         ret = bpf_perf_link_attach(attr, prog);
5009                 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5010                         ret = bpf_kprobe_multi_link_attach(attr, prog);
5011                 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5012                         ret = bpf_uprobe_multi_link_attach(attr, prog);
5013                 break;
5014         default:
5015                 ret = -EINVAL;
5016         }
5017
5018 out:
5019         if (ret < 0)
5020                 bpf_prog_put(prog);
5021         return ret;
5022 }
5023
5024 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5025 {
5026         struct bpf_map *new_map, *old_map = NULL;
5027         int ret;
5028
5029         new_map = bpf_map_get(attr->link_update.new_map_fd);
5030         if (IS_ERR(new_map))
5031                 return PTR_ERR(new_map);
5032
5033         if (attr->link_update.flags & BPF_F_REPLACE) {
5034                 old_map = bpf_map_get(attr->link_update.old_map_fd);
5035                 if (IS_ERR(old_map)) {
5036                         ret = PTR_ERR(old_map);
5037                         goto out_put;
5038                 }
5039         } else if (attr->link_update.old_map_fd) {
5040                 ret = -EINVAL;
5041                 goto out_put;
5042         }
5043
5044         ret = link->ops->update_map(link, new_map, old_map);
5045
5046         if (old_map)
5047                 bpf_map_put(old_map);
5048 out_put:
5049         bpf_map_put(new_map);
5050         return ret;
5051 }
5052
5053 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5054
5055 static int link_update(union bpf_attr *attr)
5056 {
5057         struct bpf_prog *old_prog = NULL, *new_prog;
5058         struct bpf_link *link;
5059         u32 flags;
5060         int ret;
5061
5062         if (CHECK_ATTR(BPF_LINK_UPDATE))
5063                 return -EINVAL;
5064
5065         flags = attr->link_update.flags;
5066         if (flags & ~BPF_F_REPLACE)
5067                 return -EINVAL;
5068
5069         link = bpf_link_get_from_fd(attr->link_update.link_fd);
5070         if (IS_ERR(link))
5071                 return PTR_ERR(link);
5072
5073         if (link->ops->update_map) {
5074                 ret = link_update_map(link, attr);
5075                 goto out_put_link;
5076         }
5077
5078         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5079         if (IS_ERR(new_prog)) {
5080                 ret = PTR_ERR(new_prog);
5081                 goto out_put_link;
5082         }
5083
5084         if (flags & BPF_F_REPLACE) {
5085                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5086                 if (IS_ERR(old_prog)) {
5087                         ret = PTR_ERR(old_prog);
5088                         old_prog = NULL;
5089                         goto out_put_progs;
5090                 }
5091         } else if (attr->link_update.old_prog_fd) {
5092                 ret = -EINVAL;
5093                 goto out_put_progs;
5094         }
5095
5096         if (link->ops->update_prog)
5097                 ret = link->ops->update_prog(link, new_prog, old_prog);
5098         else
5099                 ret = -EINVAL;
5100
5101 out_put_progs:
5102         if (old_prog)
5103                 bpf_prog_put(old_prog);
5104         if (ret)
5105                 bpf_prog_put(new_prog);
5106 out_put_link:
5107         bpf_link_put_direct(link);
5108         return ret;
5109 }
5110
5111 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5112
5113 static int link_detach(union bpf_attr *attr)
5114 {
5115         struct bpf_link *link;
5116         int ret;
5117
5118         if (CHECK_ATTR(BPF_LINK_DETACH))
5119                 return -EINVAL;
5120
5121         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5122         if (IS_ERR(link))
5123                 return PTR_ERR(link);
5124
5125         if (link->ops->detach)
5126                 ret = link->ops->detach(link);
5127         else
5128                 ret = -EOPNOTSUPP;
5129
5130         bpf_link_put_direct(link);
5131         return ret;
5132 }
5133
5134 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5135 {
5136         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5137 }
5138
5139 struct bpf_link *bpf_link_by_id(u32 id)
5140 {
5141         struct bpf_link *link;
5142
5143         if (!id)
5144                 return ERR_PTR(-ENOENT);
5145
5146         spin_lock_bh(&link_idr_lock);
5147         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5148         link = idr_find(&link_idr, id);
5149         if (link) {
5150                 if (link->id)
5151                         link = bpf_link_inc_not_zero(link);
5152                 else
5153                         link = ERR_PTR(-EAGAIN);
5154         } else {
5155                 link = ERR_PTR(-ENOENT);
5156         }
5157         spin_unlock_bh(&link_idr_lock);
5158         return link;
5159 }
5160
5161 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5162 {
5163         struct bpf_link *link;
5164
5165         spin_lock_bh(&link_idr_lock);
5166 again:
5167         link = idr_get_next(&link_idr, id);
5168         if (link) {
5169                 link = bpf_link_inc_not_zero(link);
5170                 if (IS_ERR(link)) {
5171                         (*id)++;
5172                         goto again;
5173                 }
5174         }
5175         spin_unlock_bh(&link_idr_lock);
5176
5177         return link;
5178 }
5179
5180 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5181
5182 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5183 {
5184         struct bpf_link *link;
5185         u32 id = attr->link_id;
5186         int fd;
5187
5188         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5189                 return -EINVAL;
5190
5191         if (!capable(CAP_SYS_ADMIN))
5192                 return -EPERM;
5193
5194         link = bpf_link_by_id(id);
5195         if (IS_ERR(link))
5196                 return PTR_ERR(link);
5197
5198         fd = bpf_link_new_fd(link);
5199         if (fd < 0)
5200                 bpf_link_put_direct(link);
5201
5202         return fd;
5203 }
5204
5205 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5206
5207 static int bpf_stats_release(struct inode *inode, struct file *file)
5208 {
5209         mutex_lock(&bpf_stats_enabled_mutex);
5210         static_key_slow_dec(&bpf_stats_enabled_key.key);
5211         mutex_unlock(&bpf_stats_enabled_mutex);
5212         return 0;
5213 }
5214
5215 static const struct file_operations bpf_stats_fops = {
5216         .release = bpf_stats_release,
5217 };
5218
5219 static int bpf_enable_runtime_stats(void)
5220 {
5221         int fd;
5222
5223         mutex_lock(&bpf_stats_enabled_mutex);
5224
5225         /* Set a very high limit to avoid overflow */
5226         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5227                 mutex_unlock(&bpf_stats_enabled_mutex);
5228                 return -EBUSY;
5229         }
5230
5231         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5232         if (fd >= 0)
5233                 static_key_slow_inc(&bpf_stats_enabled_key.key);
5234
5235         mutex_unlock(&bpf_stats_enabled_mutex);
5236         return fd;
5237 }
5238
5239 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5240
5241 static int bpf_enable_stats(union bpf_attr *attr)
5242 {
5243
5244         if (CHECK_ATTR(BPF_ENABLE_STATS))
5245                 return -EINVAL;
5246
5247         if (!capable(CAP_SYS_ADMIN))
5248                 return -EPERM;
5249
5250         switch (attr->enable_stats.type) {
5251         case BPF_STATS_RUN_TIME:
5252                 return bpf_enable_runtime_stats();
5253         default:
5254                 break;
5255         }
5256         return -EINVAL;
5257 }
5258
5259 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5260
5261 static int bpf_iter_create(union bpf_attr *attr)
5262 {
5263         struct bpf_link *link;
5264         int err;
5265
5266         if (CHECK_ATTR(BPF_ITER_CREATE))
5267                 return -EINVAL;
5268
5269         if (attr->iter_create.flags)
5270                 return -EINVAL;
5271
5272         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5273         if (IS_ERR(link))
5274                 return PTR_ERR(link);
5275
5276         err = bpf_iter_new_fd(link);
5277         bpf_link_put_direct(link);
5278
5279         return err;
5280 }
5281
5282 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5283
5284 static int bpf_prog_bind_map(union bpf_attr *attr)
5285 {
5286         struct bpf_prog *prog;
5287         struct bpf_map *map;
5288         struct bpf_map **used_maps_old, **used_maps_new;
5289         int i, ret = 0;
5290
5291         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5292                 return -EINVAL;
5293
5294         if (attr->prog_bind_map.flags)
5295                 return -EINVAL;
5296
5297         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5298         if (IS_ERR(prog))
5299                 return PTR_ERR(prog);
5300
5301         map = bpf_map_get(attr->prog_bind_map.map_fd);
5302         if (IS_ERR(map)) {
5303                 ret = PTR_ERR(map);
5304                 goto out_prog_put;
5305         }
5306
5307         mutex_lock(&prog->aux->used_maps_mutex);
5308
5309         used_maps_old = prog->aux->used_maps;
5310
5311         for (i = 0; i < prog->aux->used_map_cnt; i++)
5312                 if (used_maps_old[i] == map) {
5313                         bpf_map_put(map);
5314                         goto out_unlock;
5315                 }
5316
5317         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5318                                       sizeof(used_maps_new[0]),
5319                                       GFP_KERNEL);
5320         if (!used_maps_new) {
5321                 ret = -ENOMEM;
5322                 goto out_unlock;
5323         }
5324
5325         memcpy(used_maps_new, used_maps_old,
5326                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5327         used_maps_new[prog->aux->used_map_cnt] = map;
5328
5329         prog->aux->used_map_cnt++;
5330         prog->aux->used_maps = used_maps_new;
5331
5332         kfree(used_maps_old);
5333
5334 out_unlock:
5335         mutex_unlock(&prog->aux->used_maps_mutex);
5336
5337         if (ret)
5338                 bpf_map_put(map);
5339 out_prog_put:
5340         bpf_prog_put(prog);
5341         return ret;
5342 }
5343
5344 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5345 {
5346         union bpf_attr attr;
5347         int err;
5348
5349         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5350         if (err)
5351                 return err;
5352         size = min_t(u32, size, sizeof(attr));
5353
5354         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5355         memset(&attr, 0, sizeof(attr));
5356         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5357                 return -EFAULT;
5358
5359         err = security_bpf(cmd, &attr, size);
5360         if (err < 0)
5361                 return err;
5362
5363         switch (cmd) {
5364         case BPF_MAP_CREATE:
5365                 err = map_create(&attr);
5366                 break;
5367         case BPF_MAP_LOOKUP_ELEM:
5368                 err = map_lookup_elem(&attr);
5369                 break;
5370         case BPF_MAP_UPDATE_ELEM:
5371                 err = map_update_elem(&attr, uattr);
5372                 break;
5373         case BPF_MAP_DELETE_ELEM:
5374                 err = map_delete_elem(&attr, uattr);
5375                 break;
5376         case BPF_MAP_GET_NEXT_KEY:
5377                 err = map_get_next_key(&attr);
5378                 break;
5379         case BPF_MAP_FREEZE:
5380                 err = map_freeze(&attr);
5381                 break;
5382         case BPF_PROG_LOAD:
5383                 err = bpf_prog_load(&attr, uattr, size);
5384                 break;
5385         case BPF_OBJ_PIN:
5386                 err = bpf_obj_pin(&attr);
5387                 break;
5388         case BPF_OBJ_GET:
5389                 err = bpf_obj_get(&attr);
5390                 break;
5391         case BPF_PROG_ATTACH:
5392                 err = bpf_prog_attach(&attr);
5393                 break;
5394         case BPF_PROG_DETACH:
5395                 err = bpf_prog_detach(&attr);
5396                 break;
5397         case BPF_PROG_QUERY:
5398                 err = bpf_prog_query(&attr, uattr.user);
5399                 break;
5400         case BPF_PROG_TEST_RUN:
5401                 err = bpf_prog_test_run(&attr, uattr.user);
5402                 break;
5403         case BPF_PROG_GET_NEXT_ID:
5404                 err = bpf_obj_get_next_id(&attr, uattr.user,
5405                                           &prog_idr, &prog_idr_lock);
5406                 break;
5407         case BPF_MAP_GET_NEXT_ID:
5408                 err = bpf_obj_get_next_id(&attr, uattr.user,
5409                                           &map_idr, &map_idr_lock);
5410                 break;
5411         case BPF_BTF_GET_NEXT_ID:
5412                 err = bpf_obj_get_next_id(&attr, uattr.user,
5413                                           &btf_idr, &btf_idr_lock);
5414                 break;
5415         case BPF_PROG_GET_FD_BY_ID:
5416                 err = bpf_prog_get_fd_by_id(&attr);
5417                 break;
5418         case BPF_MAP_GET_FD_BY_ID:
5419                 err = bpf_map_get_fd_by_id(&attr);
5420                 break;
5421         case BPF_OBJ_GET_INFO_BY_FD:
5422                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5423                 break;
5424         case BPF_RAW_TRACEPOINT_OPEN:
5425                 err = bpf_raw_tracepoint_open(&attr);
5426                 break;
5427         case BPF_BTF_LOAD:
5428                 err = bpf_btf_load(&attr, uattr, size);
5429                 break;
5430         case BPF_BTF_GET_FD_BY_ID:
5431                 err = bpf_btf_get_fd_by_id(&attr);
5432                 break;
5433         case BPF_TASK_FD_QUERY:
5434                 err = bpf_task_fd_query(&attr, uattr.user);
5435                 break;
5436         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5437                 err = map_lookup_and_delete_elem(&attr);
5438                 break;
5439         case BPF_MAP_LOOKUP_BATCH:
5440                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5441                 break;
5442         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5443                 err = bpf_map_do_batch(&attr, uattr.user,
5444                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5445                 break;
5446         case BPF_MAP_UPDATE_BATCH:
5447                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5448                 break;
5449         case BPF_MAP_DELETE_BATCH:
5450                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5451                 break;
5452         case BPF_LINK_CREATE:
5453                 err = link_create(&attr, uattr);
5454                 break;
5455         case BPF_LINK_UPDATE:
5456                 err = link_update(&attr);
5457                 break;
5458         case BPF_LINK_GET_FD_BY_ID:
5459                 err = bpf_link_get_fd_by_id(&attr);
5460                 break;
5461         case BPF_LINK_GET_NEXT_ID:
5462                 err = bpf_obj_get_next_id(&attr, uattr.user,
5463                                           &link_idr, &link_idr_lock);
5464                 break;
5465         case BPF_ENABLE_STATS:
5466                 err = bpf_enable_stats(&attr);
5467                 break;
5468         case BPF_ITER_CREATE:
5469                 err = bpf_iter_create(&attr);
5470                 break;
5471         case BPF_LINK_DETACH:
5472                 err = link_detach(&attr);
5473                 break;
5474         case BPF_PROG_BIND_MAP:
5475                 err = bpf_prog_bind_map(&attr);
5476                 break;
5477         default:
5478                 err = -EINVAL;
5479                 break;
5480         }
5481
5482         return err;
5483 }
5484
5485 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5486 {
5487         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5488 }
5489
5490 static bool syscall_prog_is_valid_access(int off, int size,
5491                                          enum bpf_access_type type,
5492                                          const struct bpf_prog *prog,
5493                                          struct bpf_insn_access_aux *info)
5494 {
5495         if (off < 0 || off >= U16_MAX)
5496                 return false;
5497         if (off % size != 0)
5498                 return false;
5499         return true;
5500 }
5501
5502 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5503 {
5504         switch (cmd) {
5505         case BPF_MAP_CREATE:
5506         case BPF_MAP_DELETE_ELEM:
5507         case BPF_MAP_UPDATE_ELEM:
5508         case BPF_MAP_FREEZE:
5509         case BPF_MAP_GET_FD_BY_ID:
5510         case BPF_PROG_LOAD:
5511         case BPF_BTF_LOAD:
5512         case BPF_LINK_CREATE:
5513         case BPF_RAW_TRACEPOINT_OPEN:
5514                 break;
5515         default:
5516                 return -EINVAL;
5517         }
5518         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5519 }
5520
5521
5522 /* To shut up -Wmissing-prototypes.
5523  * This function is used by the kernel light skeleton
5524  * to load bpf programs when modules are loaded or during kernel boot.
5525  * See tools/lib/bpf/skel_internal.h
5526  */
5527 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5528
5529 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5530 {
5531         struct bpf_prog * __maybe_unused prog;
5532         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5533
5534         switch (cmd) {
5535 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5536         case BPF_PROG_TEST_RUN:
5537                 if (attr->test.data_in || attr->test.data_out ||
5538                     attr->test.ctx_out || attr->test.duration ||
5539                     attr->test.repeat || attr->test.flags)
5540                         return -EINVAL;
5541
5542                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5543                 if (IS_ERR(prog))
5544                         return PTR_ERR(prog);
5545
5546                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5547                     attr->test.ctx_size_in > U16_MAX) {
5548                         bpf_prog_put(prog);
5549                         return -EINVAL;
5550                 }
5551
5552                 run_ctx.bpf_cookie = 0;
5553                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5554                         /* recursion detected */
5555                         __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5556                         bpf_prog_put(prog);
5557                         return -EBUSY;
5558                 }
5559                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5560                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5561                                                 &run_ctx);
5562                 bpf_prog_put(prog);
5563                 return 0;
5564 #endif
5565         default:
5566                 return ____bpf_sys_bpf(cmd, attr, size);
5567         }
5568 }
5569 EXPORT_SYMBOL(kern_sys_bpf);
5570
5571 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5572         .func           = bpf_sys_bpf,
5573         .gpl_only       = false,
5574         .ret_type       = RET_INTEGER,
5575         .arg1_type      = ARG_ANYTHING,
5576         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5577         .arg3_type      = ARG_CONST_SIZE,
5578 };
5579
5580 const struct bpf_func_proto * __weak
5581 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5582 {
5583         return bpf_base_func_proto(func_id);
5584 }
5585
5586 BPF_CALL_1(bpf_sys_close, u32, fd)
5587 {
5588         /* When bpf program calls this helper there should not be
5589          * an fdget() without matching completed fdput().
5590          * This helper is allowed in the following callchain only:
5591          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5592          */
5593         return close_fd(fd);
5594 }
5595
5596 static const struct bpf_func_proto bpf_sys_close_proto = {
5597         .func           = bpf_sys_close,
5598         .gpl_only       = false,
5599         .ret_type       = RET_INTEGER,
5600         .arg1_type      = ARG_ANYTHING,
5601 };
5602
5603 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5604 {
5605         if (flags)
5606                 return -EINVAL;
5607
5608         if (name_sz <= 1 || name[name_sz - 1])
5609                 return -EINVAL;
5610
5611         if (!bpf_dump_raw_ok(current_cred()))
5612                 return -EPERM;
5613
5614         *res = kallsyms_lookup_name(name);
5615         return *res ? 0 : -ENOENT;
5616 }
5617
5618 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5619         .func           = bpf_kallsyms_lookup_name,
5620         .gpl_only       = false,
5621         .ret_type       = RET_INTEGER,
5622         .arg1_type      = ARG_PTR_TO_MEM,
5623         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5624         .arg3_type      = ARG_ANYTHING,
5625         .arg4_type      = ARG_PTR_TO_LONG,
5626 };
5627
5628 static const struct bpf_func_proto *
5629 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5630 {
5631         switch (func_id) {
5632         case BPF_FUNC_sys_bpf:
5633                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5634         case BPF_FUNC_btf_find_by_name_kind:
5635                 return &bpf_btf_find_by_name_kind_proto;
5636         case BPF_FUNC_sys_close:
5637                 return &bpf_sys_close_proto;
5638         case BPF_FUNC_kallsyms_lookup_name:
5639                 return &bpf_kallsyms_lookup_name_proto;
5640         default:
5641                 return tracing_prog_func_proto(func_id, prog);
5642         }
5643 }
5644
5645 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5646         .get_func_proto  = syscall_prog_func_proto,
5647         .is_valid_access = syscall_prog_is_valid_access,
5648 };
5649
5650 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5651         .test_run = bpf_prog_test_run_syscall,
5652 };
5653
5654 #ifdef CONFIG_SYSCTL
5655 static int bpf_stats_handler(struct ctl_table *table, int write,
5656                              void *buffer, size_t *lenp, loff_t *ppos)
5657 {
5658         struct static_key *key = (struct static_key *)table->data;
5659         static int saved_val;
5660         int val, ret;
5661         struct ctl_table tmp = {
5662                 .data   = &val,
5663                 .maxlen = sizeof(val),
5664                 .mode   = table->mode,
5665                 .extra1 = SYSCTL_ZERO,
5666                 .extra2 = SYSCTL_ONE,
5667         };
5668
5669         if (write && !capable(CAP_SYS_ADMIN))
5670                 return -EPERM;
5671
5672         mutex_lock(&bpf_stats_enabled_mutex);
5673         val = saved_val;
5674         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5675         if (write && !ret && val != saved_val) {
5676                 if (val)
5677                         static_key_slow_inc(key);
5678                 else
5679                         static_key_slow_dec(key);
5680                 saved_val = val;
5681         }
5682         mutex_unlock(&bpf_stats_enabled_mutex);
5683         return ret;
5684 }
5685
5686 void __weak unpriv_ebpf_notify(int new_state)
5687 {
5688 }
5689
5690 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5691                               void *buffer, size_t *lenp, loff_t *ppos)
5692 {
5693         int ret, unpriv_enable = *(int *)table->data;
5694         bool locked_state = unpriv_enable == 1;
5695         struct ctl_table tmp = *table;
5696
5697         if (write && !capable(CAP_SYS_ADMIN))
5698                 return -EPERM;
5699
5700         tmp.data = &unpriv_enable;
5701         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5702         if (write && !ret) {
5703                 if (locked_state && unpriv_enable != 1)
5704                         return -EPERM;
5705                 *(int *)table->data = unpriv_enable;
5706         }
5707
5708         if (write)
5709                 unpriv_ebpf_notify(unpriv_enable);
5710
5711         return ret;
5712 }
5713
5714 static struct ctl_table bpf_syscall_table[] = {
5715         {
5716                 .procname       = "unprivileged_bpf_disabled",
5717                 .data           = &sysctl_unprivileged_bpf_disabled,
5718                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5719                 .mode           = 0644,
5720                 .proc_handler   = bpf_unpriv_handler,
5721                 .extra1         = SYSCTL_ZERO,
5722                 .extra2         = SYSCTL_TWO,
5723         },
5724         {
5725                 .procname       = "bpf_stats_enabled",
5726                 .data           = &bpf_stats_enabled_key.key,
5727                 .mode           = 0644,
5728                 .proc_handler   = bpf_stats_handler,
5729         },
5730         { }
5731 };
5732
5733 static int __init bpf_syscall_sysctl_init(void)
5734 {
5735         register_sysctl_init("kernel", bpf_syscall_table);
5736         return 0;
5737 }
5738 late_initcall(bpf_syscall_sysctl_init);
5739 #endif /* CONFIG_SYSCTL */