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 non-sleepable BPF programs to complete so that
 146          * userspace, when we return to it, knows that all non-sleepable
 147          * programs that could be running use the new map value. For sleepable
 148          * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
 149          * for the completions of these programs, but considering the waiting
 150          * time can be very long and userspace may think it will hang forever,
 151          * so don't handle sleepable BPF programs now.
 152          */
 153         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 154             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 155                 synchronize_rcu();
 156 }
 157
 158 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 159                                 void *key, void *value, __u64 flags)
 160 {
 161         int err;
 162
 163         /* Need to create a kthread, thus must support schedule */
 164         if (bpf_map_is_offloaded(map)) {
 165                 return bpf_map_offload_update_elem(map, key, value, flags);
 166         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 167                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 168                 return map->ops->map_update_elem(map, key, value, flags);
 169         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 170                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 171                 return sock_map_update_elem_sys(map, key, value, flags);
 172         } else if (IS_FD_PROG_ARRAY(map)) {
 173                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 174                                                     flags);
 175         }
 176
 177         bpf_disable_instrumentation();
 178         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 179             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 180                 err = bpf_percpu_hash_update(map, key, value, flags);
 181         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 182                 err = bpf_percpu_array_update(map, key, value, flags);
 183         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 184                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 185                                                        flags);
 186         } else if (IS_FD_ARRAY(map)) {
 187                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 188                                                    flags);
 189         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 190                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 191                                                   flags);
 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
 207         return err;
 208 }
 209
 210 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 211                               __u64 flags)
 212 {
 213         void *ptr;
 214         int err;
 215
 216         if (bpf_map_is_offloaded(map))
 217                 return bpf_map_offload_lookup_elem(map, key, value);
 218
 219         bpf_disable_instrumentation();
 220         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 221             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 222                 err = bpf_percpu_hash_copy(map, key, value);
 223         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 224                 err = bpf_percpu_array_copy(map, key, value);
 225         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 226                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 227         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 228                 err = bpf_stackmap_copy(map, key, value);
 229         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 230                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 231         } else if (IS_FD_HASH(map)) {
 232                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 233         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 234                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 235         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 236                    map->map_type == BPF_MAP_TYPE_STACK ||
 237                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 238                 err = map->ops->map_peek_elem(map, value);
 239         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 240                 /* struct_ops map requires directly updating "value" */
 241                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 242         } else {
 243                 rcu_read_lock();
 244                 if (map->ops->map_lookup_elem_sys_only)
 245                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 246                 else
 247                         ptr = map->ops->map_lookup_elem(map, key);
 248                 if (IS_ERR(ptr)) {
 249                         err = PTR_ERR(ptr);
 250                 } else if (!ptr) {
 251                         err = -ENOENT;
 252                 } else {
 253                         err = 0;
 254                         if (flags & BPF_F_LOCK)
 255                                 /* lock 'ptr' and copy everything but lock */
 256                                 copy_map_value_locked(map, value, ptr, true);
 257                         else
 258                                 copy_map_value(map, value, ptr);
 259                         /* mask lock and timer, since value wasn't zero inited */
 260                         check_and_init_map_value(map, value);
 261                 }
 262                 rcu_read_unlock();
 263         }
 264
 265         bpf_enable_instrumentation();
 266
 267         return err;
 268 }
 269
 270 /* Please, do not use this function outside from the map creation path
 271  * (e.g. in map update path) without taking care of setting the active
 272  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 273  */
 274 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 275 {
 276         /* We really just want to fail instead of triggering OOM killer
 277          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 278          * which is used for lower order allocation requests.
 279          *
 280          * It has been observed that higher order allocation requests done by
 281          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 282          * to reclaim memory from the page cache, thus we set
 283          * __GFP_RETRY_MAYFAIL to avoid such situations.
 284          */
 285
 286         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 287         unsigned int flags = 0;
 288         unsigned long align = 1;
 289         void *area;
 290
 291         if (size >= SIZE_MAX)
 292                 return NULL;
 293
 294         /* kmalloc()'ed memory can't be mmap()'ed */
 295         if (mmapable) {
 296                 BUG_ON(!PAGE_ALIGNED(size));
 297                 align = SHMLBA;
 298                 flags = VM_USERMAP;
 299         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 300                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 301                                     numa_node);
 302                 if (area != NULL)
 303                         return area;
 304         }
 305
 306         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 307                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 308                         flags, numa_node, __builtin_return_address(0));
 309 }
 310
 311 void *bpf_map_area_alloc(u64 size, int numa_node)
 312 {
 313         return __bpf_map_area_alloc(size, numa_node, false);
 314 }
 315
 316 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 317 {
 318         return __bpf_map_area_alloc(size, numa_node, true);
 319 }
 320
 321 void bpf_map_area_free(void *area)
 322 {
 323         kvfree(area);
 324 }
 325
 326 static u32 bpf_map_flags_retain_permanent(u32 flags)
 327 {
 328         /* Some map creation flags are not tied to the map object but
 329          * rather to the map fd instead, so they have no meaning upon
 330          * map object inspection since multiple file descriptors with
 331          * different (access) properties can exist here. Thus, given
 332          * this has zero meaning for the map itself, lets clear these
 333          * from here.
 334          */
 335         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 336 }
 337
 338 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 339 {
 340         map->map_type = attr->map_type;
 341         map->key_size = attr->key_size;
 342         map->value_size = attr->value_size;
 343         map->max_entries = attr->max_entries;
 344         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 345         map->numa_node = bpf_map_attr_numa_node(attr);
 346         map->map_extra = attr->map_extra;
 347 }
 348
 349 static int bpf_map_alloc_id(struct bpf_map *map)
 350 {
 351         int id;
 352
 353         idr_preload(GFP_KERNEL);
 354         spin_lock_bh(&map_idr_lock);
 355         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 356         if (id > 0)
 357                 map->id = id;
 358         spin_unlock_bh(&map_idr_lock);
 359         idr_preload_end();
 360
 361         if (WARN_ON_ONCE(!id))
 362                 return -ENOSPC;
 363
 364         return id > 0 ? 0 : id;
 365 }
 366
 367 void bpf_map_free_id(struct bpf_map *map)
 368 {
 369         unsigned long flags;
 370
 371         /* Offloaded maps are removed from the IDR store when their device
 372          * disappears - even if someone holds an fd to them they are unusable,
 373          * the memory is gone, all ops will fail; they are simply waiting for
 374          * refcnt to drop to be freed.
 375          */
 376         if (!map->id)
 377                 return;
 378
 379         spin_lock_irqsave(&map_idr_lock, flags);
 380
 381         idr_remove(&map_idr, map->id);
 382         map->id = 0;
 383
 384         spin_unlock_irqrestore(&map_idr_lock, flags);
 385 }
 386
 387 #ifdef CONFIG_MEMCG_KMEM
 388 static void bpf_map_save_memcg(struct bpf_map *map)
 389 {
 390         /* Currently if a map is created by a process belonging to the root
 391          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 392          * So we have to check map->objcg for being NULL each time it's
 393          * being used.
 394          */
 395         if (memcg_bpf_enabled())
 396                 map->objcg = get_obj_cgroup_from_current();
 397 }
 398
 399 static void bpf_map_release_memcg(struct bpf_map *map)
 400 {
 401         if (map->objcg)
 402                 obj_cgroup_put(map->objcg);
 403 }
 404
 405 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 406 {
 407         if (map->objcg)
 408                 return get_mem_cgroup_from_objcg(map->objcg);
 409
 410         return root_mem_cgroup;
 411 }
 412
 413 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 414                            int node)
 415 {
 416         struct mem_cgroup *memcg, *old_memcg;
 417         void *ptr;
 418
 419         memcg = bpf_map_get_memcg(map);
 420         old_memcg = set_active_memcg(memcg);
 421         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 422         set_active_memcg(old_memcg);
 423         mem_cgroup_put(memcg);
 424
 425         return ptr;
 426 }
 427
 428 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 429 {
 430         struct mem_cgroup *memcg, *old_memcg;
 431         void *ptr;
 432
 433         memcg = bpf_map_get_memcg(map);
 434         old_memcg = set_active_memcg(memcg);
 435         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 436         set_active_memcg(old_memcg);
 437         mem_cgroup_put(memcg);
 438
 439         return ptr;
 440 }
 441
 442 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 443                        gfp_t flags)
 444 {
 445         struct mem_cgroup *memcg, *old_memcg;
 446         void *ptr;
 447
 448         memcg = bpf_map_get_memcg(map);
 449         old_memcg = set_active_memcg(memcg);
 450         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 451         set_active_memcg(old_memcg);
 452         mem_cgroup_put(memcg);
 453
 454         return ptr;
 455 }
 456
 457 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 458                                     size_t align, gfp_t flags)
 459 {
 460         struct mem_cgroup *memcg, *old_memcg;
 461         void __percpu *ptr;
 462
 463         memcg = bpf_map_get_memcg(map);
 464         old_memcg = set_active_memcg(memcg);
 465         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 466         set_active_memcg(old_memcg);
 467         mem_cgroup_put(memcg);
 468
 469         return ptr;
 470 }
 471
 472 #else
 473 static void bpf_map_save_memcg(struct bpf_map *map)
 474 {
 475 }
 476
 477 static void bpf_map_release_memcg(struct bpf_map *map)
 478 {
 479 }
 480 #endif
 481
 482 static int btf_field_cmp(const void *a, const void *b)
 483 {
 484         const struct btf_field *f1 = a, *f2 = b;
 485
 486         if (f1->offset < f2->offset)
 487                 return -1;
 488         else if (f1->offset > f2->offset)
 489                 return 1;
 490         return 0;
 491 }
 492
 493 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 494                                   u32 field_mask)
 495 {
 496         struct btf_field *field;
 497
 498         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 499                 return NULL;
 500         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 501         if (!field || !(field->type & field_mask))
 502                 return NULL;
 503         return field;
 504 }
 505
 506 void btf_record_free(struct btf_record *rec)
 507 {
 508         int i;
 509
 510         if (IS_ERR_OR_NULL(rec))
 511                 return;
 512         for (i = 0; i < rec->cnt; i++) {
 513                 switch (rec->fields[i].type) {
 514                 case BPF_KPTR_UNREF:
 515                 case BPF_KPTR_REF:
 516                 case BPF_KPTR_PERCPU:
 517                         if (rec->fields[i].kptr.module)
 518                                 module_put(rec->fields[i].kptr.module);
 519                         btf_put(rec->fields[i].kptr.btf);
 520                         break;
 521                 case BPF_LIST_HEAD:
 522                 case BPF_LIST_NODE:
 523                 case BPF_RB_ROOT:
 524                 case BPF_RB_NODE:
 525                 case BPF_SPIN_LOCK:
 526                 case BPF_TIMER:
 527                 case BPF_REFCOUNT:
 528                         /* Nothing to release */
 529                         break;
 530                 default:
 531                         WARN_ON_ONCE(1);
 532                         continue;
 533                 }
 534         }
 535         kfree(rec);
 536 }
 537
 538 void bpf_map_free_record(struct bpf_map *map)
 539 {
 540         btf_record_free(map->record);
 541         map->record = NULL;
 542 }
 543
 544 struct btf_record *btf_record_dup(const struct btf_record *rec)
 545 {
 546         const struct btf_field *fields;
 547         struct btf_record *new_rec;
 548         int ret, size, i;
 549
 550         if (IS_ERR_OR_NULL(rec))
 551                 return NULL;
 552         size = offsetof(struct btf_record, fields[rec->cnt]);
 553         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 554         if (!new_rec)
 555                 return ERR_PTR(-ENOMEM);
 556         /* Do a deep copy of the btf_record */
 557         fields = rec->fields;
 558         new_rec->cnt = 0;
 559         for (i = 0; i < rec->cnt; i++) {
 560                 switch (fields[i].type) {
 561                 case BPF_KPTR_UNREF:
 562                 case BPF_KPTR_REF:
 563                 case BPF_KPTR_PERCPU:
 564                         btf_get(fields[i].kptr.btf);
 565                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 566                                 ret = -ENXIO;
 567                                 goto free;
 568                         }
 569                         break;
 570                 case BPF_LIST_HEAD:
 571                 case BPF_LIST_NODE:
 572                 case BPF_RB_ROOT:
 573                 case BPF_RB_NODE:
 574                 case BPF_SPIN_LOCK:
 575                 case BPF_TIMER:
 576                 case BPF_REFCOUNT:
 577                         /* Nothing to acquire */
 578                         break;
 579                 default:
 580                         ret = -EFAULT;
 581                         WARN_ON_ONCE(1);
 582                         goto free;
 583                 }
 584                 new_rec->cnt++;
 585         }
 586         return new_rec;
 587 free:
 588         btf_record_free(new_rec);
 589         return ERR_PTR(ret);
 590 }
 591
 592 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 593 {
 594         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 595         int size;
 596
 597         if (!a_has_fields && !b_has_fields)
 598                 return true;
 599         if (a_has_fields != b_has_fields)
 600                 return false;
 601         if (rec_a->cnt != rec_b->cnt)
 602                 return false;
 603         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 604         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 605          * members are zeroed out. So memcmp is safe to do without worrying
 606          * about padding/unused fields.
 607          *
 608          * While spin_lock, timer, and kptr have no relation to map BTF,
 609          * list_head metadata is specific to map BTF, the btf and value_rec
 610          * members in particular. btf is the map BTF, while value_rec points to
 611          * btf_record in that map BTF.
 612          *
 613          * So while by default, we don't rely on the map BTF (which the records
 614          * were parsed from) matching for both records, which is not backwards
 615          * compatible, in case list_head is part of it, we implicitly rely on
 616          * that by way of depending on memcmp succeeding for it.
 617          */
 618         return !memcmp(rec_a, rec_b, size);
 619 }
 620
 621 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 622 {
 623         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 624                 return;
 625         bpf_timer_cancel_and_free(obj + rec->timer_off);
 626 }
 627
 628 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 629 {
 630         const struct btf_field *fields;
 631         int i;
 632
 633         if (IS_ERR_OR_NULL(rec))
 634                 return;
 635         fields = rec->fields;
 636         for (i = 0; i < rec->cnt; i++) {
 637                 struct btf_struct_meta *pointee_struct_meta;
 638                 const struct btf_field *field = &fields[i];
 639                 void *field_ptr = obj + field->offset;
 640                 void *xchgd_field;
 641
 642                 switch (fields[i].type) {
 643                 case BPF_SPIN_LOCK:
 644                         break;
 645                 case BPF_TIMER:
 646                         bpf_timer_cancel_and_free(field_ptr);
 647                         break;
 648                 case BPF_KPTR_UNREF:
 649                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 650                         break;
 651                 case BPF_KPTR_REF:
 652                 case BPF_KPTR_PERCPU:
 653                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 654                         if (!xchgd_field)
 655                                 break;
 656
 657                         if (!btf_is_kernel(field->kptr.btf)) {
 658                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 659                                                                            field->kptr.btf_id);
 660                                 migrate_disable();
 661                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 662                                                                  pointee_struct_meta->record : NULL,
 663                                                                  fields[i].type == BPF_KPTR_PERCPU);
 664                                 migrate_enable();
 665                         } else {
 666                                 field->kptr.dtor(xchgd_field);
 667                         }
 668                         break;
 669                 case BPF_LIST_HEAD:
 670                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 671                                 continue;
 672                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 673                         break;
 674                 case BPF_RB_ROOT:
 675                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 676                                 continue;
 677                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 678                         break;
 679                 case BPF_LIST_NODE:
 680                 case BPF_RB_NODE:
 681                 case BPF_REFCOUNT:
 682                         break;
 683                 default:
 684                         WARN_ON_ONCE(1);
 685                         continue;
 686                 }
 687         }
 688 }
 689
 690 /* called from workqueue */
 691 static void bpf_map_free_deferred(struct work_struct *work)
 692 {
 693         struct bpf_map *map = container_of(work, struct bpf_map, work);
 694         struct btf_record *rec = map->record;
 695         struct btf *btf = map->btf;
 696
 697         security_bpf_map_free(map);
 698         bpf_map_release_memcg(map);
 699         /* implementation dependent freeing */
 700         map->ops->map_free(map);
 701         /* Delay freeing of btf_record for maps, as map_free
 702          * callback usually needs access to them. It is better to do it here
 703          * than require each callback to do the free itself manually.
 704          *
 705          * Note that the btf_record stashed in map->inner_map_meta->record was
 706          * already freed using the map_free callback for map in map case which
 707          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 708          * template bpf_map struct used during verification.
 709          */
 710         btf_record_free(rec);
 711         /* Delay freeing of btf for maps, as map_free callback may need
 712          * struct_meta info which will be freed with btf_put().
 713          */
 714         btf_put(btf);
 715 }
 716
 717 static void bpf_map_put_uref(struct bpf_map *map)
 718 {
 719         if (atomic64_dec_and_test(&map->usercnt)) {
 720                 if (map->ops->map_release_uref)
 721                         map->ops->map_release_uref(map);
 722         }
 723 }
 724
 725 static void bpf_map_free_in_work(struct bpf_map *map)
 726 {
 727         INIT_WORK(&map->work, bpf_map_free_deferred);
 728         /* Avoid spawning kworkers, since they all might contend
 729          * for the same mutex like slab_mutex.
 730          */
 731         queue_work(system_unbound_wq, &map->work);
 732 }
 733
 734 static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
 735 {
 736         bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
 737 }
 738
 739 static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
 740 {
 741         if (rcu_trace_implies_rcu_gp())
 742                 bpf_map_free_rcu_gp(rcu);
 743         else
 744                 call_rcu(rcu, bpf_map_free_rcu_gp);
 745 }
 746
 747 /* decrement map refcnt and schedule it for freeing via workqueue
 748  * (underlying map implementation ops->map_free() might sleep)
 749  */
 750 void bpf_map_put(struct bpf_map *map)
 751 {
 752         if (atomic64_dec_and_test(&map->refcnt)) {
 753                 /* bpf_map_free_id() must be called first */
 754                 bpf_map_free_id(map);
 755
 756                 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
 757                 if (READ_ONCE(map->free_after_mult_rcu_gp))
 758                         call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
 759                 else if (READ_ONCE(map->free_after_rcu_gp))
 760                         call_rcu(&map->rcu, bpf_map_free_rcu_gp);
 761                 else
 762                         bpf_map_free_in_work(map);
 763         }
 764 }
 765 EXPORT_SYMBOL_GPL(bpf_map_put);
 766
 767 void bpf_map_put_with_uref(struct bpf_map *map)
 768 {
 769         bpf_map_put_uref(map);
 770         bpf_map_put(map);
 771 }
 772
 773 static int bpf_map_release(struct inode *inode, struct file *filp)
 774 {
 775         struct bpf_map *map = filp->private_data;
 776
 777         if (map->ops->map_release)
 778                 map->ops->map_release(map, filp);
 779
 780         bpf_map_put_with_uref(map);
 781         return 0;
 782 }
 783
 784 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 785 {
 786         fmode_t mode = f.file->f_mode;
 787
 788         /* Our file permissions may have been overridden by global
 789          * map permissions facing syscall side.
 790          */
 791         if (READ_ONCE(map->frozen))
 792                 mode &= ~FMODE_CAN_WRITE;
 793         return mode;
 794 }
 795
 796 #ifdef CONFIG_PROC_FS
 797 /* Show the memory usage of a bpf map */
 798 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 799 {
 800         return map->ops->map_mem_usage(map);
 801 }
 802
 803 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 804 {
 805         struct bpf_map *map = filp->private_data;
 806         u32 type = 0, jited = 0;
 807
 808         if (map_type_contains_progs(map)) {
 809                 spin_lock(&map->owner.lock);
 810                 type  = map->owner.type;
 811                 jited = map->owner.jited;
 812                 spin_unlock(&map->owner.lock);
 813         }
 814
 815         seq_printf(m,
 816                    "map_type:\t%u\n"
 817                    "key_size:\t%u\n"
 818                    "value_size:\t%u\n"
 819                    "max_entries:\t%u\n"
 820                    "map_flags:\t%#x\n"
 821                    "map_extra:\t%#llx\n"
 822                    "memlock:\t%llu\n"
 823                    "map_id:\t%u\n"
 824                    "frozen:\t%u\n",
 825                    map->map_type,
 826                    map->key_size,
 827                    map->value_size,
 828                    map->max_entries,
 829                    map->map_flags,
 830                    (unsigned long long)map->map_extra,
 831                    bpf_map_memory_usage(map),
 832                    map->id,
 833                    READ_ONCE(map->frozen));
 834         if (type) {
 835                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 836                 seq_printf(m, "owner_jited:\t%u\n", jited);
 837         }
 838 }
 839 #endif
 840
 841 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 842                               loff_t *ppos)
 843 {
 844         /* We need this handler such that alloc_file() enables
 845          * f_mode with FMODE_CAN_READ.
 846          */
 847         return -EINVAL;
 848 }
 849
 850 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 851                                size_t siz, loff_t *ppos)
 852 {
 853         /* We need this handler such that alloc_file() enables
 854          * f_mode with FMODE_CAN_WRITE.
 855          */
 856         return -EINVAL;
 857 }
 858
 859 /* called for any extra memory-mapped regions (except initial) */
 860 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 861 {
 862         struct bpf_map *map = vma->vm_file->private_data;
 863
 864         if (vma->vm_flags & VM_MAYWRITE)
 865                 bpf_map_write_active_inc(map);
 866 }
 867
 868 /* called for all unmapped memory region (including initial) */
 869 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 870 {
 871         struct bpf_map *map = vma->vm_file->private_data;
 872
 873         if (vma->vm_flags & VM_MAYWRITE)
 874                 bpf_map_write_active_dec(map);
 875 }
 876
 877 static const struct vm_operations_struct bpf_map_default_vmops = {
 878         .open           = bpf_map_mmap_open,
 879         .close          = bpf_map_mmap_close,
 880 };
 881
 882 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 883 {
 884         struct bpf_map *map = filp->private_data;
 885         int err;
 886
 887         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 888                 return -ENOTSUPP;
 889
 890         if (!(vma->vm_flags & VM_SHARED))
 891                 return -EINVAL;
 892
 893         mutex_lock(&map->freeze_mutex);
 894
 895         if (vma->vm_flags & VM_WRITE) {
 896                 if (map->frozen) {
 897                         err = -EPERM;
 898                         goto out;
 899                 }
 900                 /* map is meant to be read-only, so do not allow mapping as
 901                  * writable, because it's possible to leak a writable page
 902                  * reference and allows user-space to still modify it after
 903                  * freezing, while verifier will assume contents do not change
 904                  */
 905                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 906                         err = -EACCES;
 907                         goto out;
 908                 }
 909         }
 910
 911         /* set default open/close callbacks */
 912         vma->vm_ops = &bpf_map_default_vmops;
 913         vma->vm_private_data = map;
 914         vm_flags_clear(vma, VM_MAYEXEC);
 915         if (!(vma->vm_flags & VM_WRITE))
 916                 /* disallow re-mapping with PROT_WRITE */
 917                 vm_flags_clear(vma, VM_MAYWRITE);
 918
 919         err = map->ops->map_mmap(map, vma);
 920         if (err)
 921                 goto out;
 922
 923         if (vma->vm_flags & VM_MAYWRITE)
 924                 bpf_map_write_active_inc(map);
 925 out:
 926         mutex_unlock(&map->freeze_mutex);
 927         return err;
 928 }
 929
 930 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 931 {
 932         struct bpf_map *map = filp->private_data;
 933
 934         if (map->ops->map_poll)
 935                 return map->ops->map_poll(map, filp, pts);
 936
 937         return EPOLLERR;
 938 }
 939
 940 const struct file_operations bpf_map_fops = {
 941 #ifdef CONFIG_PROC_FS
 942         .show_fdinfo    = bpf_map_show_fdinfo,
 943 #endif
 944         .release        = bpf_map_release,
 945         .read           = bpf_dummy_read,
 946         .write          = bpf_dummy_write,
 947         .mmap           = bpf_map_mmap,
 948         .poll           = bpf_map_poll,
 949 };
 950
 951 int bpf_map_new_fd(struct bpf_map *map, int flags)
 952 {
 953         int ret;
 954
 955         ret = security_bpf_map(map, OPEN_FMODE(flags));
 956         if (ret < 0)
 957                 return ret;
 958
 959         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 960                                 flags | O_CLOEXEC);
 961 }
 962
 963 int bpf_get_file_flag(int flags)
 964 {
 965         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 966                 return -EINVAL;
 967         if (flags & BPF_F_RDONLY)
 968                 return O_RDONLY;
 969         if (flags & BPF_F_WRONLY)
 970                 return O_WRONLY;
 971         return O_RDWR;
 972 }
 973
 974 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 975 #define CHECK_ATTR(CMD) \
 976         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 977                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 978                    sizeof(*attr) - \
 979                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 980                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 981
 982 /* dst and src must have at least "size" number of bytes.
 983  * Return strlen on success and < 0 on error.
 984  */
 985 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 986 {
 987         const char *end = src + size;
 988         const char *orig_src = src;
 989
 990         memset(dst, 0, size);
 991         /* Copy all isalnum(), '_' and '.' chars. */
 992         while (src < end && *src) {
 993                 if (!isalnum(*src) &&
 994                     *src != '_' && *src != '.')
 995                         return -EINVAL;
 996                 *dst++ = *src++;
 997         }
 998
 999         /* No '\0' found in "size" number of bytes */
1000         if (src == end)
1001                 return -EINVAL;
1002
1003         return src - orig_src;
1004 }
1005
1006 int map_check_no_btf(const struct bpf_map *map,
1007                      const struct btf *btf,
1008                      const struct btf_type *key_type,
1009                      const struct btf_type *value_type)
1010 {
1011         return -ENOTSUPP;
1012 }
1013
1014 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
1015                          u32 btf_key_id, u32 btf_value_id)
1016 {
1017         const struct btf_type *key_type, *value_type;
1018         u32 key_size, value_size;
1019         int ret = 0;
1020
1021         /* Some maps allow key to be unspecified. */
1022         if (btf_key_id) {
1023                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1024                 if (!key_type || key_size != map->key_size)
1025                         return -EINVAL;
1026         } else {
1027                 key_type = btf_type_by_id(btf, 0);
1028                 if (!map->ops->map_check_btf)
1029                         return -EINVAL;
1030         }
1031
1032         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1033         if (!value_type || value_size != map->value_size)
1034                 return -EINVAL;
1035
1036         map->record = btf_parse_fields(btf, value_type,
1037                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1038                                        BPF_RB_ROOT | BPF_REFCOUNT,
1039                                        map->value_size);
1040         if (!IS_ERR_OR_NULL(map->record)) {
1041                 int i;
1042
1043                 if (!bpf_capable()) {
1044                         ret = -EPERM;
1045                         goto free_map_tab;
1046                 }
1047                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1048                         ret = -EACCES;
1049                         goto free_map_tab;
1050                 }
1051                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1052                         switch (map->record->field_mask & (1 << i)) {
1053                         case 0:
1054                                 continue;
1055                         case BPF_SPIN_LOCK:
1056                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1057                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1058                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1059                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1060                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1061                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1062                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1063                                         ret = -EOPNOTSUPP;
1064                                         goto free_map_tab;
1065                                 }
1066                                 break;
1067                         case BPF_TIMER:
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                         case BPF_KPTR_UNREF:
1076                         case BPF_KPTR_REF:
1077                         case BPF_KPTR_PERCPU:
1078                         case BPF_REFCOUNT:
1079                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1080                                     map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1081                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1082                                     map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1083                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1084                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1085                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1086                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1087                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1088                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1089                                         ret = -EOPNOTSUPP;
1090                                         goto free_map_tab;
1091                                 }
1092                                 break;
1093                         case BPF_LIST_HEAD:
1094                         case BPF_RB_ROOT:
1095                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1096                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1097                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1098                                         ret = -EOPNOTSUPP;
1099                                         goto free_map_tab;
1100                                 }
1101                                 break;
1102                         default:
1103                                 /* Fail if map_type checks are missing for a field type */
1104                                 ret = -EOPNOTSUPP;
1105                                 goto free_map_tab;
1106                         }
1107                 }
1108         }
1109
1110         ret = btf_check_and_fixup_fields(btf, map->record);
1111         if (ret < 0)
1112                 goto free_map_tab;
1113
1114         if (map->ops->map_check_btf) {
1115                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1116                 if (ret < 0)
1117                         goto free_map_tab;
1118         }
1119
1120         return ret;
1121 free_map_tab:
1122         bpf_map_free_record(map);
1123         return ret;
1124 }
1125
1126 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1127 /* called via syscall */
1128 static int map_create(union bpf_attr *attr)
1129 {
1130         const struct bpf_map_ops *ops;
1131         int numa_node = bpf_map_attr_numa_node(attr);
1132         u32 map_type = attr->map_type;
1133         struct bpf_map *map;
1134         int f_flags;
1135         int err;
1136
1137         err = CHECK_ATTR(BPF_MAP_CREATE);
1138         if (err)
1139                 return -EINVAL;
1140
1141         if (attr->btf_vmlinux_value_type_id) {
1142                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1143                     attr->btf_key_type_id || attr->btf_value_type_id)
1144                         return -EINVAL;
1145         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1146                 return -EINVAL;
1147         }
1148
1149         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1150             attr->map_extra != 0)
1151                 return -EINVAL;
1152
1153         f_flags = bpf_get_file_flag(attr->map_flags);
1154         if (f_flags < 0)
1155                 return f_flags;
1156
1157         if (numa_node != NUMA_NO_NODE &&
1158             ((unsigned int)numa_node >= nr_node_ids ||
1159              !node_online(numa_node)))
1160                 return -EINVAL;
1161
1162         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1163         map_type = attr->map_type;
1164         if (map_type >= ARRAY_SIZE(bpf_map_types))
1165                 return -EINVAL;
1166         map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1167         ops = bpf_map_types[map_type];
1168         if (!ops)
1169                 return -EINVAL;
1170
1171         if (ops->map_alloc_check) {
1172                 err = ops->map_alloc_check(attr);
1173                 if (err)
1174                         return err;
1175         }
1176         if (attr->map_ifindex)
1177                 ops = &bpf_map_offload_ops;
1178         if (!ops->map_mem_usage)
1179                 return -EINVAL;
1180
1181         /* Intent here is for unprivileged_bpf_disabled to block BPF map
1182          * creation for unprivileged users; other actions depend
1183          * on fd availability and access to bpffs, so are dependent on
1184          * object creation success. Even with unprivileged BPF disabled,
1185          * capability checks are still carried out.
1186          */
1187         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
1188                 return -EPERM;
1189
1190         /* check privileged map type permissions */
1191         switch (map_type) {
1192         case BPF_MAP_TYPE_ARRAY:
1193         case BPF_MAP_TYPE_PERCPU_ARRAY:
1194         case BPF_MAP_TYPE_PROG_ARRAY:
1195         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1196         case BPF_MAP_TYPE_CGROUP_ARRAY:
1197         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1198         case BPF_MAP_TYPE_HASH:
1199         case BPF_MAP_TYPE_PERCPU_HASH:
1200         case BPF_MAP_TYPE_HASH_OF_MAPS:
1201         case BPF_MAP_TYPE_RINGBUF:
1202         case BPF_MAP_TYPE_USER_RINGBUF:
1203         case BPF_MAP_TYPE_CGROUP_STORAGE:
1204         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1205                 /* unprivileged */
1206                 break;
1207         case BPF_MAP_TYPE_SK_STORAGE:
1208         case BPF_MAP_TYPE_INODE_STORAGE:
1209         case BPF_MAP_TYPE_TASK_STORAGE:
1210         case BPF_MAP_TYPE_CGRP_STORAGE:
1211         case BPF_MAP_TYPE_BLOOM_FILTER:
1212         case BPF_MAP_TYPE_LPM_TRIE:
1213         case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1214         case BPF_MAP_TYPE_STACK_TRACE:
1215         case BPF_MAP_TYPE_QUEUE:
1216         case BPF_MAP_TYPE_STACK:
1217         case BPF_MAP_TYPE_LRU_HASH:
1218         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1219         case BPF_MAP_TYPE_STRUCT_OPS:
1220         case BPF_MAP_TYPE_CPUMAP:
1221                 if (!bpf_capable())
1222                         return -EPERM;
1223                 break;
1224         case BPF_MAP_TYPE_SOCKMAP:
1225         case BPF_MAP_TYPE_SOCKHASH:
1226         case BPF_MAP_TYPE_DEVMAP:
1227         case BPF_MAP_TYPE_DEVMAP_HASH:
1228         case BPF_MAP_TYPE_XSKMAP:
1229                 if (!capable(CAP_NET_ADMIN))
1230                         return -EPERM;
1231                 break;
1232         default:
1233                 WARN(1, "unsupported map type %d", map_type);
1234                 return -EPERM;
1235         }
1236
1237         map = ops->map_alloc(attr);
1238         if (IS_ERR(map))
1239                 return PTR_ERR(map);
1240         map->ops = ops;
1241         map->map_type = map_type;
1242
1243         err = bpf_obj_name_cpy(map->name, attr->map_name,
1244                                sizeof(attr->map_name));
1245         if (err < 0)
1246                 goto free_map;
1247
1248         atomic64_set(&map->refcnt, 1);
1249         atomic64_set(&map->usercnt, 1);
1250         mutex_init(&map->freeze_mutex);
1251         spin_lock_init(&map->owner.lock);
1252
1253         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1254             /* Even the map's value is a kernel's struct,
1255              * the bpf_prog.o must have BTF to begin with
1256              * to figure out the corresponding kernel's
1257              * counter part.  Thus, attr->btf_fd has
1258              * to be valid also.
1259              */
1260             attr->btf_vmlinux_value_type_id) {
1261                 struct btf *btf;
1262
1263                 btf = btf_get_by_fd(attr->btf_fd);
1264                 if (IS_ERR(btf)) {
1265                         err = PTR_ERR(btf);
1266                         goto free_map;
1267                 }
1268                 if (btf_is_kernel(btf)) {
1269                         btf_put(btf);
1270                         err = -EACCES;
1271                         goto free_map;
1272                 }
1273                 map->btf = btf;
1274
1275                 if (attr->btf_value_type_id) {
1276                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1277                                             attr->btf_value_type_id);
1278                         if (err)
1279                                 goto free_map;
1280                 }
1281
1282                 map->btf_key_type_id = attr->btf_key_type_id;
1283                 map->btf_value_type_id = attr->btf_value_type_id;
1284                 map->btf_vmlinux_value_type_id =
1285                         attr->btf_vmlinux_value_type_id;
1286         }
1287
1288         err = security_bpf_map_alloc(map);
1289         if (err)
1290                 goto free_map;
1291
1292         err = bpf_map_alloc_id(map);
1293         if (err)
1294                 goto free_map_sec;
1295
1296         bpf_map_save_memcg(map);
1297
1298         err = bpf_map_new_fd(map, f_flags);
1299         if (err < 0) {
1300                 /* failed to allocate fd.
1301                  * bpf_map_put_with_uref() is needed because the above
1302                  * bpf_map_alloc_id() has published the map
1303                  * to the userspace and the userspace may
1304                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1305                  */
1306                 bpf_map_put_with_uref(map);
1307                 return err;
1308         }
1309
1310         return err;
1311
1312 free_map_sec:
1313         security_bpf_map_free(map);
1314 free_map:
1315         btf_put(map->btf);
1316         map->ops->map_free(map);
1317         return err;
1318 }
1319
1320 /* if error is returned, fd is released.
1321  * On success caller should complete fd access with matching fdput()
1322  */
1323 struct bpf_map *__bpf_map_get(struct fd f)
1324 {
1325         if (!f.file)
1326                 return ERR_PTR(-EBADF);
1327         if (f.file->f_op != &bpf_map_fops) {
1328                 fdput(f);
1329                 return ERR_PTR(-EINVAL);
1330         }
1331
1332         return f.file->private_data;
1333 }
1334
1335 void bpf_map_inc(struct bpf_map *map)
1336 {
1337         atomic64_inc(&map->refcnt);
1338 }
1339 EXPORT_SYMBOL_GPL(bpf_map_inc);
1340
1341 void bpf_map_inc_with_uref(struct bpf_map *map)
1342 {
1343         atomic64_inc(&map->refcnt);
1344         atomic64_inc(&map->usercnt);
1345 }
1346 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1347
1348 struct bpf_map *bpf_map_get(u32 ufd)
1349 {
1350         struct fd f = fdget(ufd);
1351         struct bpf_map *map;
1352
1353         map = __bpf_map_get(f);
1354         if (IS_ERR(map))
1355                 return map;
1356
1357         bpf_map_inc(map);
1358         fdput(f);
1359
1360         return map;
1361 }
1362 EXPORT_SYMBOL(bpf_map_get);
1363
1364 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1365 {
1366         struct fd f = fdget(ufd);
1367         struct bpf_map *map;
1368
1369         map = __bpf_map_get(f);
1370         if (IS_ERR(map))
1371                 return map;
1372
1373         bpf_map_inc_with_uref(map);
1374         fdput(f);
1375
1376         return map;
1377 }
1378
1379 /* map_idr_lock should have been held or the map should have been
1380  * protected by rcu read lock.
1381  */
1382 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1383 {
1384         int refold;
1385
1386         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1387         if (!refold)
1388                 return ERR_PTR(-ENOENT);
1389         if (uref)
1390                 atomic64_inc(&map->usercnt);
1391
1392         return map;
1393 }
1394
1395 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1396 {
1397         spin_lock_bh(&map_idr_lock);
1398         map = __bpf_map_inc_not_zero(map, false);
1399         spin_unlock_bh(&map_idr_lock);
1400
1401         return map;
1402 }
1403 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1404
1405 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1406 {
1407         return -ENOTSUPP;
1408 }
1409
1410 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1411 {
1412         if (key_size)
1413                 return vmemdup_user(ukey, key_size);
1414
1415         if (ukey)
1416                 return ERR_PTR(-EINVAL);
1417
1418         return NULL;
1419 }
1420
1421 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1422 {
1423         if (key_size)
1424                 return kvmemdup_bpfptr(ukey, key_size);
1425
1426         if (!bpfptr_is_null(ukey))
1427                 return ERR_PTR(-EINVAL);
1428
1429         return NULL;
1430 }
1431
1432 /* last field in 'union bpf_attr' used by this command */
1433 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1434
1435 static int map_lookup_elem(union bpf_attr *attr)
1436 {
1437         void __user *ukey = u64_to_user_ptr(attr->key);
1438         void __user *uvalue = u64_to_user_ptr(attr->value);
1439         int ufd = attr->map_fd;
1440         struct bpf_map *map;
1441         void *key, *value;
1442         u32 value_size;
1443         struct fd f;
1444         int err;
1445
1446         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1447                 return -EINVAL;
1448
1449         if (attr->flags & ~BPF_F_LOCK)
1450                 return -EINVAL;
1451
1452         f = fdget(ufd);
1453         map = __bpf_map_get(f);
1454         if (IS_ERR(map))
1455                 return PTR_ERR(map);
1456         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1457                 err = -EPERM;
1458                 goto err_put;
1459         }
1460
1461         if ((attr->flags & BPF_F_LOCK) &&
1462             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1463                 err = -EINVAL;
1464                 goto err_put;
1465         }
1466
1467         key = __bpf_copy_key(ukey, map->key_size);
1468         if (IS_ERR(key)) {
1469                 err = PTR_ERR(key);
1470                 goto err_put;
1471         }
1472
1473         value_size = bpf_map_value_size(map);
1474
1475         err = -ENOMEM;
1476         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1477         if (!value)
1478                 goto free_key;
1479
1480         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1481                 if (copy_from_user(value, uvalue, value_size))
1482                         err = -EFAULT;
1483                 else
1484                         err = bpf_map_copy_value(map, key, value, attr->flags);
1485                 goto free_value;
1486         }
1487
1488         err = bpf_map_copy_value(map, key, value, attr->flags);
1489         if (err)
1490                 goto free_value;
1491
1492         err = -EFAULT;
1493         if (copy_to_user(uvalue, value, value_size) != 0)
1494                 goto free_value;
1495
1496         err = 0;
1497
1498 free_value:
1499         kvfree(value);
1500 free_key:
1501         kvfree(key);
1502 err_put:
1503         fdput(f);
1504         return err;
1505 }
1506
1507
1508 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1509
1510 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1511 {
1512         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1513         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1514         int ufd = attr->map_fd;
1515         struct bpf_map *map;
1516         void *key, *value;
1517         u32 value_size;
1518         struct fd f;
1519         int err;
1520
1521         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1522                 return -EINVAL;
1523
1524         f = fdget(ufd);
1525         map = __bpf_map_get(f);
1526         if (IS_ERR(map))
1527                 return PTR_ERR(map);
1528         bpf_map_write_active_inc(map);
1529         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1530                 err = -EPERM;
1531                 goto err_put;
1532         }
1533
1534         if ((attr->flags & BPF_F_LOCK) &&
1535             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1536                 err = -EINVAL;
1537                 goto err_put;
1538         }
1539
1540         key = ___bpf_copy_key(ukey, map->key_size);
1541         if (IS_ERR(key)) {
1542                 err = PTR_ERR(key);
1543                 goto err_put;
1544         }
1545
1546         value_size = bpf_map_value_size(map);
1547         value = kvmemdup_bpfptr(uvalue, value_size);
1548         if (IS_ERR(value)) {
1549                 err = PTR_ERR(value);
1550                 goto free_key;
1551         }
1552
1553         err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1554         if (!err)
1555                 maybe_wait_bpf_programs(map);
1556
1557         kvfree(value);
1558 free_key:
1559         kvfree(key);
1560 err_put:
1561         bpf_map_write_active_dec(map);
1562         fdput(f);
1563         return err;
1564 }
1565
1566 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1567
1568 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1569 {
1570         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1571         int ufd = attr->map_fd;
1572         struct bpf_map *map;
1573         struct fd f;
1574         void *key;
1575         int err;
1576
1577         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1578                 return -EINVAL;
1579
1580         f = fdget(ufd);
1581         map = __bpf_map_get(f);
1582         if (IS_ERR(map))
1583                 return PTR_ERR(map);
1584         bpf_map_write_active_inc(map);
1585         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1586                 err = -EPERM;
1587                 goto err_put;
1588         }
1589
1590         key = ___bpf_copy_key(ukey, map->key_size);
1591         if (IS_ERR(key)) {
1592                 err = PTR_ERR(key);
1593                 goto err_put;
1594         }
1595
1596         if (bpf_map_is_offloaded(map)) {
1597                 err = bpf_map_offload_delete_elem(map, key);
1598                 goto out;
1599         } else if (IS_FD_PROG_ARRAY(map) ||
1600                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1601                 /* These maps require sleepable context */
1602                 err = map->ops->map_delete_elem(map, key);
1603                 goto out;
1604         }
1605
1606         bpf_disable_instrumentation();
1607         rcu_read_lock();
1608         err = map->ops->map_delete_elem(map, key);
1609         rcu_read_unlock();
1610         bpf_enable_instrumentation();
1611         if (!err)
1612                 maybe_wait_bpf_programs(map);
1613 out:
1614         kvfree(key);
1615 err_put:
1616         bpf_map_write_active_dec(map);
1617         fdput(f);
1618         return err;
1619 }
1620
1621 /* last field in 'union bpf_attr' used by this command */
1622 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1623
1624 static int map_get_next_key(union bpf_attr *attr)
1625 {
1626         void __user *ukey = u64_to_user_ptr(attr->key);
1627         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1628         int ufd = attr->map_fd;
1629         struct bpf_map *map;
1630         void *key, *next_key;
1631         struct fd f;
1632         int err;
1633
1634         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1635                 return -EINVAL;
1636
1637         f = fdget(ufd);
1638         map = __bpf_map_get(f);
1639         if (IS_ERR(map))
1640                 return PTR_ERR(map);
1641         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1642                 err = -EPERM;
1643                 goto err_put;
1644         }
1645
1646         if (ukey) {
1647                 key = __bpf_copy_key(ukey, map->key_size);
1648                 if (IS_ERR(key)) {
1649                         err = PTR_ERR(key);
1650                         goto err_put;
1651                 }
1652         } else {
1653                 key = NULL;
1654         }
1655
1656         err = -ENOMEM;
1657         next_key = kvmalloc(map->key_size, GFP_USER);
1658         if (!next_key)
1659                 goto free_key;
1660
1661         if (bpf_map_is_offloaded(map)) {
1662                 err = bpf_map_offload_get_next_key(map, key, next_key);
1663                 goto out;
1664         }
1665
1666         rcu_read_lock();
1667         err = map->ops->map_get_next_key(map, key, next_key);
1668         rcu_read_unlock();
1669 out:
1670         if (err)
1671                 goto free_next_key;
1672
1673         err = -EFAULT;
1674         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1675                 goto free_next_key;
1676
1677         err = 0;
1678
1679 free_next_key:
1680         kvfree(next_key);
1681 free_key:
1682         kvfree(key);
1683 err_put:
1684         fdput(f);
1685         return err;
1686 }
1687
1688 int generic_map_delete_batch(struct bpf_map *map,
1689                              const union bpf_attr *attr,
1690                              union bpf_attr __user *uattr)
1691 {
1692         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1693         u32 cp, max_count;
1694         int err = 0;
1695         void *key;
1696
1697         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1698                 return -EINVAL;
1699
1700         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1701             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1702                 return -EINVAL;
1703         }
1704
1705         max_count = attr->batch.count;
1706         if (!max_count)
1707                 return 0;
1708
1709         if (put_user(0, &uattr->batch.count))
1710                 return -EFAULT;
1711
1712         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1713         if (!key)
1714                 return -ENOMEM;
1715
1716         for (cp = 0; cp < max_count; cp++) {
1717                 err = -EFAULT;
1718                 if (copy_from_user(key, keys + cp * map->key_size,
1719                                    map->key_size))
1720                         break;
1721
1722                 if (bpf_map_is_offloaded(map)) {
1723                         err = bpf_map_offload_delete_elem(map, key);
1724                         break;
1725                 }
1726
1727                 bpf_disable_instrumentation();
1728                 rcu_read_lock();
1729                 err = map->ops->map_delete_elem(map, key);
1730                 rcu_read_unlock();
1731                 bpf_enable_instrumentation();
1732                 if (err)
1733                         break;
1734                 cond_resched();
1735         }
1736         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1737                 err = -EFAULT;
1738
1739         kvfree(key);
1740
1741         return err;
1742 }
1743
1744 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1745                              const union bpf_attr *attr,
1746                              union bpf_attr __user *uattr)
1747 {
1748         void __user *values = u64_to_user_ptr(attr->batch.values);
1749         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1750         u32 value_size, cp, max_count;
1751         void *key, *value;
1752         int err = 0;
1753
1754         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1755                 return -EINVAL;
1756
1757         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1758             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1759                 return -EINVAL;
1760         }
1761
1762         value_size = bpf_map_value_size(map);
1763
1764         max_count = attr->batch.count;
1765         if (!max_count)
1766                 return 0;
1767
1768         if (put_user(0, &uattr->batch.count))
1769                 return -EFAULT;
1770
1771         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1772         if (!key)
1773                 return -ENOMEM;
1774
1775         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1776         if (!value) {
1777                 kvfree(key);
1778                 return -ENOMEM;
1779         }
1780
1781         for (cp = 0; cp < max_count; cp++) {
1782                 err = -EFAULT;
1783                 if (copy_from_user(key, keys + cp * map->key_size,
1784                     map->key_size) ||
1785                     copy_from_user(value, values + cp * value_size, value_size))
1786                         break;
1787
1788                 err = bpf_map_update_value(map, map_file, key, value,
1789                                            attr->batch.elem_flags);
1790
1791                 if (err)
1792                         break;
1793                 cond_resched();
1794         }
1795
1796         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1797                 err = -EFAULT;
1798
1799         kvfree(value);
1800         kvfree(key);
1801
1802         return err;
1803 }
1804
1805 #define MAP_LOOKUP_RETRIES 3
1806
1807 int generic_map_lookup_batch(struct bpf_map *map,
1808                                     const union bpf_attr *attr,
1809                                     union bpf_attr __user *uattr)
1810 {
1811         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1812         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1813         void __user *values = u64_to_user_ptr(attr->batch.values);
1814         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1815         void *buf, *buf_prevkey, *prev_key, *key, *value;
1816         int err, retry = MAP_LOOKUP_RETRIES;
1817         u32 value_size, cp, max_count;
1818
1819         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1820                 return -EINVAL;
1821
1822         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1823             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1824                 return -EINVAL;
1825
1826         value_size = bpf_map_value_size(map);
1827
1828         max_count = attr->batch.count;
1829         if (!max_count)
1830                 return 0;
1831
1832         if (put_user(0, &uattr->batch.count))
1833                 return -EFAULT;
1834
1835         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1836         if (!buf_prevkey)
1837                 return -ENOMEM;
1838
1839         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1840         if (!buf) {
1841                 kvfree(buf_prevkey);
1842                 return -ENOMEM;
1843         }
1844
1845         err = -EFAULT;
1846         prev_key = NULL;
1847         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1848                 goto free_buf;
1849         key = buf;
1850         value = key + map->key_size;
1851         if (ubatch)
1852                 prev_key = buf_prevkey;
1853
1854         for (cp = 0; cp < max_count;) {
1855                 rcu_read_lock();
1856                 err = map->ops->map_get_next_key(map, prev_key, key);
1857                 rcu_read_unlock();
1858                 if (err)
1859                         break;
1860                 err = bpf_map_copy_value(map, key, value,
1861                                          attr->batch.elem_flags);
1862
1863                 if (err == -ENOENT) {
1864                         if (retry) {
1865                                 retry--;
1866                                 continue;
1867                         }
1868                         err = -EINTR;
1869                         break;
1870                 }
1871
1872                 if (err)
1873                         goto free_buf;
1874
1875                 if (copy_to_user(keys + cp * map->key_size, key,
1876                                  map->key_size)) {
1877                         err = -EFAULT;
1878                         goto free_buf;
1879                 }
1880                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1881                         err = -EFAULT;
1882                         goto free_buf;
1883                 }
1884
1885                 if (!prev_key)
1886                         prev_key = buf_prevkey;
1887
1888                 swap(prev_key, key);
1889                 retry = MAP_LOOKUP_RETRIES;
1890                 cp++;
1891                 cond_resched();
1892         }
1893
1894         if (err == -EFAULT)
1895                 goto free_buf;
1896
1897         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1898                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1899                 err = -EFAULT;
1900
1901 free_buf:
1902         kvfree(buf_prevkey);
1903         kvfree(buf);
1904         return err;
1905 }
1906
1907 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1908
1909 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1910 {
1911         void __user *ukey = u64_to_user_ptr(attr->key);
1912         void __user *uvalue = u64_to_user_ptr(attr->value);
1913         int ufd = attr->map_fd;
1914         struct bpf_map *map;
1915         void *key, *value;
1916         u32 value_size;
1917         struct fd f;
1918         int err;
1919
1920         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1921                 return -EINVAL;
1922
1923         if (attr->flags & ~BPF_F_LOCK)
1924                 return -EINVAL;
1925
1926         f = fdget(ufd);
1927         map = __bpf_map_get(f);
1928         if (IS_ERR(map))
1929                 return PTR_ERR(map);
1930         bpf_map_write_active_inc(map);
1931         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1932             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1933                 err = -EPERM;
1934                 goto err_put;
1935         }
1936
1937         if (attr->flags &&
1938             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1939              map->map_type == BPF_MAP_TYPE_STACK)) {
1940                 err = -EINVAL;
1941                 goto err_put;
1942         }
1943
1944         if ((attr->flags & BPF_F_LOCK) &&
1945             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1946                 err = -EINVAL;
1947                 goto err_put;
1948         }
1949
1950         key = __bpf_copy_key(ukey, map->key_size);
1951         if (IS_ERR(key)) {
1952                 err = PTR_ERR(key);
1953                 goto err_put;
1954         }
1955
1956         value_size = bpf_map_value_size(map);
1957
1958         err = -ENOMEM;
1959         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1960         if (!value)
1961                 goto free_key;
1962
1963         err = -ENOTSUPP;
1964         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1965             map->map_type == BPF_MAP_TYPE_STACK) {
1966                 err = map->ops->map_pop_elem(map, value);
1967         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1968                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1969                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1970                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1971                 if (!bpf_map_is_offloaded(map)) {
1972                         bpf_disable_instrumentation();
1973                         rcu_read_lock();
1974                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1975                         rcu_read_unlock();
1976                         bpf_enable_instrumentation();
1977                 }
1978         }
1979
1980         if (err)
1981                 goto free_value;
1982
1983         if (copy_to_user(uvalue, value, value_size) != 0) {
1984                 err = -EFAULT;
1985                 goto free_value;
1986         }
1987
1988         err = 0;
1989
1990 free_value:
1991         kvfree(value);
1992 free_key:
1993         kvfree(key);
1994 err_put:
1995         bpf_map_write_active_dec(map);
1996         fdput(f);
1997         return err;
1998 }
1999
2000 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
2001
2002 static int map_freeze(const union bpf_attr *attr)
2003 {
2004         int err = 0, ufd = attr->map_fd;
2005         struct bpf_map *map;
2006         struct fd f;
2007
2008         if (CHECK_ATTR(BPF_MAP_FREEZE))
2009                 return -EINVAL;
2010
2011         f = fdget(ufd);
2012         map = __bpf_map_get(f);
2013         if (IS_ERR(map))
2014                 return PTR_ERR(map);
2015
2016         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
2017                 fdput(f);
2018                 return -ENOTSUPP;
2019         }
2020
2021         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
2022                 fdput(f);
2023                 return -EPERM;
2024         }
2025
2026         mutex_lock(&map->freeze_mutex);
2027         if (bpf_map_write_active(map)) {
2028                 err = -EBUSY;
2029                 goto err_put;
2030         }
2031         if (READ_ONCE(map->frozen)) {
2032                 err = -EBUSY;
2033                 goto err_put;
2034         }
2035
2036         WRITE_ONCE(map->frozen, true);
2037 err_put:
2038         mutex_unlock(&map->freeze_mutex);
2039         fdput(f);
2040         return err;
2041 }
2042
2043 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2044 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2045         [_id] = & _name ## _prog_ops,
2046 #define BPF_MAP_TYPE(_id, _ops)
2047 #define BPF_LINK_TYPE(_id, _name)
2048 #include <linux/bpf_types.h>
2049 #undef BPF_PROG_TYPE
2050 #undef BPF_MAP_TYPE
2051 #undef BPF_LINK_TYPE
2052 };
2053
2054 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2055 {
2056         const struct bpf_prog_ops *ops;
2057
2058         if (type >= ARRAY_SIZE(bpf_prog_types))
2059                 return -EINVAL;
2060         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2061         ops = bpf_prog_types[type];
2062         if (!ops)
2063                 return -EINVAL;
2064
2065         if (!bpf_prog_is_offloaded(prog->aux))
2066                 prog->aux->ops = ops;
2067         else
2068                 prog->aux->ops = &bpf_offload_prog_ops;
2069         prog->type = type;
2070         return 0;
2071 }
2072
2073 enum bpf_audit {
2074         BPF_AUDIT_LOAD,
2075         BPF_AUDIT_UNLOAD,
2076         BPF_AUDIT_MAX,
2077 };
2078
2079 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2080         [BPF_AUDIT_LOAD]   = "LOAD",
2081         [BPF_AUDIT_UNLOAD] = "UNLOAD",
2082 };
2083
2084 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2085 {
2086         struct audit_context *ctx = NULL;
2087         struct audit_buffer *ab;
2088
2089         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2090                 return;
2091         if (audit_enabled == AUDIT_OFF)
2092                 return;
2093         if (!in_irq() && !irqs_disabled())
2094                 ctx = audit_context();
2095         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2096         if (unlikely(!ab))
2097                 return;
2098         audit_log_format(ab, "prog-id=%u op=%s",
2099                          prog->aux->id, bpf_audit_str[op]);
2100         audit_log_end(ab);
2101 }
2102
2103 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2104 {
2105         int id;
2106
2107         idr_preload(GFP_KERNEL);
2108         spin_lock_bh(&prog_idr_lock);
2109         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2110         if (id > 0)
2111                 prog->aux->id = id;
2112         spin_unlock_bh(&prog_idr_lock);
2113         idr_preload_end();
2114
2115         /* id is in [1, INT_MAX) */
2116         if (WARN_ON_ONCE(!id))
2117                 return -ENOSPC;
2118
2119         return id > 0 ? 0 : id;
2120 }
2121
2122 void bpf_prog_free_id(struct bpf_prog *prog)
2123 {
2124         unsigned long flags;
2125
2126         /* cBPF to eBPF migrations are currently not in the idr store.
2127          * Offloaded programs are removed from the store when their device
2128          * disappears - even if someone grabs an fd to them they are unusable,
2129          * simply waiting for refcnt to drop to be freed.
2130          */
2131         if (!prog->aux->id)
2132                 return;
2133
2134         spin_lock_irqsave(&prog_idr_lock, flags);
2135         idr_remove(&prog_idr, prog->aux->id);
2136         prog->aux->id = 0;
2137         spin_unlock_irqrestore(&prog_idr_lock, flags);
2138 }
2139
2140 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2141 {
2142         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2143
2144         kvfree(aux->func_info);
2145         kfree(aux->func_info_aux);
2146         free_uid(aux->user);
2147         security_bpf_prog_free(aux);
2148         bpf_prog_free(aux->prog);
2149 }
2150
2151 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2152 {
2153         bpf_prog_kallsyms_del_all(prog);
2154         btf_put(prog->aux->btf);
2155         module_put(prog->aux->mod);
2156         kvfree(prog->aux->jited_linfo);
2157         kvfree(prog->aux->linfo);
2158         kfree(prog->aux->kfunc_tab);
2159         if (prog->aux->attach_btf)
2160                 btf_put(prog->aux->attach_btf);
2161
2162         if (deferred) {
2163                 if (prog->aux->sleepable)
2164                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2165                 else
2166                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2167         } else {
2168                 __bpf_prog_put_rcu(&prog->aux->rcu);
2169         }
2170 }
2171
2172 static void bpf_prog_put_deferred(struct work_struct *work)
2173 {
2174         struct bpf_prog_aux *aux;
2175         struct bpf_prog *prog;
2176
2177         aux = container_of(work, struct bpf_prog_aux, work);
2178         prog = aux->prog;
2179         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2180         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2181         bpf_prog_free_id(prog);
2182         __bpf_prog_put_noref(prog, true);
2183 }
2184
2185 static void __bpf_prog_put(struct bpf_prog *prog)
2186 {
2187         struct bpf_prog_aux *aux = prog->aux;
2188
2189         if (atomic64_dec_and_test(&aux->refcnt)) {
2190                 if (in_irq() || irqs_disabled()) {
2191                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2192                         schedule_work(&aux->work);
2193                 } else {
2194                         bpf_prog_put_deferred(&aux->work);
2195                 }
2196         }
2197 }
2198
2199 void bpf_prog_put(struct bpf_prog *prog)
2200 {
2201         __bpf_prog_put(prog);
2202 }
2203 EXPORT_SYMBOL_GPL(bpf_prog_put);
2204
2205 static int bpf_prog_release(struct inode *inode, struct file *filp)
2206 {
2207         struct bpf_prog *prog = filp->private_data;
2208
2209         bpf_prog_put(prog);
2210         return 0;
2211 }
2212
2213 struct bpf_prog_kstats {
2214         u64 nsecs;
2215         u64 cnt;
2216         u64 misses;
2217 };
2218
2219 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2220 {
2221         struct bpf_prog_stats *stats;
2222         unsigned int flags;
2223
2224         stats = this_cpu_ptr(prog->stats);
2225         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2226         u64_stats_inc(&stats->misses);
2227         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2228 }
2229
2230 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2231                                struct bpf_prog_kstats *stats)
2232 {
2233         u64 nsecs = 0, cnt = 0, misses = 0;
2234         int cpu;
2235
2236         for_each_possible_cpu(cpu) {
2237                 const struct bpf_prog_stats *st;
2238                 unsigned int start;
2239                 u64 tnsecs, tcnt, tmisses;
2240
2241                 st = per_cpu_ptr(prog->stats, cpu);
2242                 do {
2243                         start = u64_stats_fetch_begin(&st->syncp);
2244                         tnsecs = u64_stats_read(&st->nsecs);
2245                         tcnt = u64_stats_read(&st->cnt);
2246                         tmisses = u64_stats_read(&st->misses);
2247                 } while (u64_stats_fetch_retry(&st->syncp, start));
2248                 nsecs += tnsecs;
2249                 cnt += tcnt;
2250                 misses += tmisses;
2251         }
2252         stats->nsecs = nsecs;
2253         stats->cnt = cnt;
2254         stats->misses = misses;
2255 }
2256
2257 #ifdef CONFIG_PROC_FS
2258 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2259 {
2260         const struct bpf_prog *prog = filp->private_data;
2261         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2262         struct bpf_prog_kstats stats;
2263
2264         bpf_prog_get_stats(prog, &stats);
2265         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2266         seq_printf(m,
2267                    "prog_type:\t%u\n"
2268                    "prog_jited:\t%u\n"
2269                    "prog_tag:\t%s\n"
2270                    "memlock:\t%llu\n"
2271                    "prog_id:\t%u\n"
2272                    "run_time_ns:\t%llu\n"
2273                    "run_cnt:\t%llu\n"
2274                    "recursion_misses:\t%llu\n"
2275                    "verified_insns:\t%u\n",
2276                    prog->type,
2277                    prog->jited,
2278                    prog_tag,
2279                    prog->pages * 1ULL << PAGE_SHIFT,
2280                    prog->aux->id,
2281                    stats.nsecs,
2282                    stats.cnt,
2283                    stats.misses,
2284                    prog->aux->verified_insns);
2285 }
2286 #endif
2287
2288 const struct file_operations bpf_prog_fops = {
2289 #ifdef CONFIG_PROC_FS
2290         .show_fdinfo    = bpf_prog_show_fdinfo,
2291 #endif
2292         .release        = bpf_prog_release,
2293         .read           = bpf_dummy_read,
2294         .write          = bpf_dummy_write,
2295 };
2296
2297 int bpf_prog_new_fd(struct bpf_prog *prog)
2298 {
2299         int ret;
2300
2301         ret = security_bpf_prog(prog);
2302         if (ret < 0)
2303                 return ret;
2304
2305         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2306                                 O_RDWR | O_CLOEXEC);
2307 }
2308
2309 static struct bpf_prog *____bpf_prog_get(struct fd f)
2310 {
2311         if (!f.file)
2312                 return ERR_PTR(-EBADF);
2313         if (f.file->f_op != &bpf_prog_fops) {
2314                 fdput(f);
2315                 return ERR_PTR(-EINVAL);
2316         }
2317
2318         return f.file->private_data;
2319 }
2320
2321 void bpf_prog_add(struct bpf_prog *prog, int i)
2322 {
2323         atomic64_add(i, &prog->aux->refcnt);
2324 }
2325 EXPORT_SYMBOL_GPL(bpf_prog_add);
2326
2327 void bpf_prog_sub(struct bpf_prog *prog, int i)
2328 {
2329         /* Only to be used for undoing previous bpf_prog_add() in some
2330          * error path. We still know that another entity in our call
2331          * path holds a reference to the program, thus atomic_sub() can
2332          * be safely used in such cases!
2333          */
2334         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2335 }
2336 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2337
2338 void bpf_prog_inc(struct bpf_prog *prog)
2339 {
2340         atomic64_inc(&prog->aux->refcnt);
2341 }
2342 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2343
2344 /* prog_idr_lock should have been held */
2345 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2346 {
2347         int refold;
2348
2349         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2350
2351         if (!refold)
2352                 return ERR_PTR(-ENOENT);
2353
2354         return prog;
2355 }
2356 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2357
2358 bool bpf_prog_get_ok(struct bpf_prog *prog,
2359                             enum bpf_prog_type *attach_type, bool attach_drv)
2360 {
2361         /* not an attachment, just a refcount inc, always allow */
2362         if (!attach_type)
2363                 return true;
2364
2365         if (prog->type != *attach_type)
2366                 return false;
2367         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2368                 return false;
2369
2370         return true;
2371 }
2372
2373 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2374                                        bool attach_drv)
2375 {
2376         struct fd f = fdget(ufd);
2377         struct bpf_prog *prog;
2378
2379         prog = ____bpf_prog_get(f);
2380         if (IS_ERR(prog))
2381                 return prog;
2382         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2383                 prog = ERR_PTR(-EINVAL);
2384                 goto out;
2385         }
2386
2387         bpf_prog_inc(prog);
2388 out:
2389         fdput(f);
2390         return prog;
2391 }
2392
2393 struct bpf_prog *bpf_prog_get(u32 ufd)
2394 {
2395         return __bpf_prog_get(ufd, NULL, false);
2396 }
2397
2398 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2399                                        bool attach_drv)
2400 {
2401         return __bpf_prog_get(ufd, &type, attach_drv);
2402 }
2403 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2404
2405 /* Initially all BPF programs could be loaded w/o specifying
2406  * expected_attach_type. Later for some of them specifying expected_attach_type
2407  * at load time became required so that program could be validated properly.
2408  * Programs of types that are allowed to be loaded both w/ and w/o (for
2409  * backward compatibility) expected_attach_type, should have the default attach
2410  * type assigned to expected_attach_type for the latter case, so that it can be
2411  * validated later at attach time.
2412  *
2413  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2414  * prog type requires it but has some attach types that have to be backward
2415  * compatible.
2416  */
2417 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2418 {
2419         switch (attr->prog_type) {
2420         case BPF_PROG_TYPE_CGROUP_SOCK:
2421                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2422                  * exist so checking for non-zero is the way to go here.
2423                  */
2424                 if (!attr->expected_attach_type)
2425                         attr->expected_attach_type =
2426                                 BPF_CGROUP_INET_SOCK_CREATE;
2427                 break;
2428         case BPF_PROG_TYPE_SK_REUSEPORT:
2429                 if (!attr->expected_attach_type)
2430                         attr->expected_attach_type =
2431                                 BPF_SK_REUSEPORT_SELECT;
2432                 break;
2433         }
2434 }
2435
2436 static int
2437 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2438                            enum bpf_attach_type expected_attach_type,
2439                            struct btf *attach_btf, u32 btf_id,
2440                            struct bpf_prog *dst_prog)
2441 {
2442         if (btf_id) {
2443                 if (btf_id > BTF_MAX_TYPE)
2444                         return -EINVAL;
2445
2446                 if (!attach_btf && !dst_prog)
2447                         return -EINVAL;
2448
2449                 switch (prog_type) {
2450                 case BPF_PROG_TYPE_TRACING:
2451                 case BPF_PROG_TYPE_LSM:
2452                 case BPF_PROG_TYPE_STRUCT_OPS:
2453                 case BPF_PROG_TYPE_EXT:
2454                         break;
2455                 default:
2456                         return -EINVAL;
2457                 }
2458         }
2459
2460         if (attach_btf && (!btf_id || dst_prog))
2461                 return -EINVAL;
2462
2463         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2464             prog_type != BPF_PROG_TYPE_EXT)
2465                 return -EINVAL;
2466
2467         switch (prog_type) {
2468         case BPF_PROG_TYPE_CGROUP_SOCK:
2469                 switch (expected_attach_type) {
2470                 case BPF_CGROUP_INET_SOCK_CREATE:
2471                 case BPF_CGROUP_INET_SOCK_RELEASE:
2472                 case BPF_CGROUP_INET4_POST_BIND:
2473                 case BPF_CGROUP_INET6_POST_BIND:
2474                         return 0;
2475                 default:
2476                         return -EINVAL;
2477                 }
2478         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2479                 switch (expected_attach_type) {
2480                 case BPF_CGROUP_INET4_BIND:
2481                 case BPF_CGROUP_INET6_BIND:
2482                 case BPF_CGROUP_INET4_CONNECT:
2483                 case BPF_CGROUP_INET6_CONNECT:
2484                 case BPF_CGROUP_UNIX_CONNECT:
2485                 case BPF_CGROUP_INET4_GETPEERNAME:
2486                 case BPF_CGROUP_INET6_GETPEERNAME:
2487                 case BPF_CGROUP_UNIX_GETPEERNAME:
2488                 case BPF_CGROUP_INET4_GETSOCKNAME:
2489                 case BPF_CGROUP_INET6_GETSOCKNAME:
2490                 case BPF_CGROUP_UNIX_GETSOCKNAME:
2491                 case BPF_CGROUP_UDP4_SENDMSG:
2492                 case BPF_CGROUP_UDP6_SENDMSG:
2493                 case BPF_CGROUP_UNIX_SENDMSG:
2494                 case BPF_CGROUP_UDP4_RECVMSG:
2495                 case BPF_CGROUP_UDP6_RECVMSG:
2496                 case BPF_CGROUP_UNIX_RECVMSG:
2497                         return 0;
2498                 default:
2499                         return -EINVAL;
2500                 }
2501         case BPF_PROG_TYPE_CGROUP_SKB:
2502                 switch (expected_attach_type) {
2503                 case BPF_CGROUP_INET_INGRESS:
2504                 case BPF_CGROUP_INET_EGRESS:
2505                         return 0;
2506                 default:
2507                         return -EINVAL;
2508                 }
2509         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2510                 switch (expected_attach_type) {
2511                 case BPF_CGROUP_SETSOCKOPT:
2512                 case BPF_CGROUP_GETSOCKOPT:
2513                         return 0;
2514                 default:
2515                         return -EINVAL;
2516                 }
2517         case BPF_PROG_TYPE_SK_LOOKUP:
2518                 if (expected_attach_type == BPF_SK_LOOKUP)
2519                         return 0;
2520                 return -EINVAL;
2521         case BPF_PROG_TYPE_SK_REUSEPORT:
2522                 switch (expected_attach_type) {
2523                 case BPF_SK_REUSEPORT_SELECT:
2524                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2525                         return 0;
2526                 default:
2527                         return -EINVAL;
2528                 }
2529         case BPF_PROG_TYPE_NETFILTER:
2530                 if (expected_attach_type == BPF_NETFILTER)
2531                         return 0;
2532                 return -EINVAL;
2533         case BPF_PROG_TYPE_SYSCALL:
2534         case BPF_PROG_TYPE_EXT:
2535                 if (expected_attach_type)
2536                         return -EINVAL;
2537                 fallthrough;
2538         default:
2539                 return 0;
2540         }
2541 }
2542
2543 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2544 {
2545         switch (prog_type) {
2546         case BPF_PROG_TYPE_SCHED_CLS:
2547         case BPF_PROG_TYPE_SCHED_ACT:
2548         case BPF_PROG_TYPE_XDP:
2549         case BPF_PROG_TYPE_LWT_IN:
2550         case BPF_PROG_TYPE_LWT_OUT:
2551         case BPF_PROG_TYPE_LWT_XMIT:
2552         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2553         case BPF_PROG_TYPE_SK_SKB:
2554         case BPF_PROG_TYPE_SK_MSG:
2555         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2556         case BPF_PROG_TYPE_CGROUP_DEVICE:
2557         case BPF_PROG_TYPE_CGROUP_SOCK:
2558         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2559         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2560         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2561         case BPF_PROG_TYPE_SOCK_OPS:
2562         case BPF_PROG_TYPE_EXT: /* extends any prog */
2563         case BPF_PROG_TYPE_NETFILTER:
2564                 return true;
2565         case BPF_PROG_TYPE_CGROUP_SKB:
2566                 /* always unpriv */
2567         case BPF_PROG_TYPE_SK_REUSEPORT:
2568                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2569         default:
2570                 return false;
2571         }
2572 }
2573
2574 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2575 {
2576         switch (prog_type) {
2577         case BPF_PROG_TYPE_KPROBE:
2578         case BPF_PROG_TYPE_TRACEPOINT:
2579         case BPF_PROG_TYPE_PERF_EVENT:
2580         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2581         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2582         case BPF_PROG_TYPE_TRACING:
2583         case BPF_PROG_TYPE_LSM:
2584         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2585         case BPF_PROG_TYPE_EXT: /* extends any prog */
2586                 return true;
2587         default:
2588                 return false;
2589         }
2590 }
2591
2592 /* last field in 'union bpf_attr' used by this command */
2593 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2594
2595 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2596 {
2597         enum bpf_prog_type type = attr->prog_type;
2598         struct bpf_prog *prog, *dst_prog = NULL;
2599         struct btf *attach_btf = NULL;
2600         int err;
2601         char license[128];
2602
2603         if (CHECK_ATTR(BPF_PROG_LOAD))
2604                 return -EINVAL;
2605
2606         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2607                                  BPF_F_ANY_ALIGNMENT |
2608                                  BPF_F_TEST_STATE_FREQ |
2609                                  BPF_F_SLEEPABLE |
2610                                  BPF_F_TEST_RND_HI32 |
2611                                  BPF_F_XDP_HAS_FRAGS |
2612                                  BPF_F_XDP_DEV_BOUND_ONLY |
2613                                  BPF_F_TEST_REG_INVARIANTS))
2614                 return -EINVAL;
2615
2616         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2617             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2618             !bpf_capable())
2619                 return -EPERM;
2620
2621         /* Intent here is for unprivileged_bpf_disabled to block BPF program
2622          * creation for unprivileged users; other actions depend
2623          * on fd availability and access to bpffs, so are dependent on
2624          * object creation success. Even with unprivileged BPF disabled,
2625          * capability checks are still carried out for these
2626          * and other operations.
2627          */
2628         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
2629                 return -EPERM;
2630
2631         if (attr->insn_cnt == 0 ||
2632             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2633                 return -E2BIG;
2634         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2635             type != BPF_PROG_TYPE_CGROUP_SKB &&
2636             !bpf_capable())
2637                 return -EPERM;
2638
2639         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2640                 return -EPERM;
2641         if (is_perfmon_prog_type(type) && !perfmon_capable())
2642                 return -EPERM;
2643
2644         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2645          * or btf, we need to check which one it is
2646          */
2647         if (attr->attach_prog_fd) {
2648                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2649                 if (IS_ERR(dst_prog)) {
2650                         dst_prog = NULL;
2651                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2652                         if (IS_ERR(attach_btf))
2653                                 return -EINVAL;
2654                         if (!btf_is_kernel(attach_btf)) {
2655                                 /* attaching through specifying bpf_prog's BTF
2656                                  * objects directly might be supported eventually
2657                                  */
2658                                 btf_put(attach_btf);
2659                                 return -ENOTSUPP;
2660                         }
2661                 }
2662         } else if (attr->attach_btf_id) {
2663                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2664                 attach_btf = bpf_get_btf_vmlinux();
2665                 if (IS_ERR(attach_btf))
2666                         return PTR_ERR(attach_btf);
2667                 if (!attach_btf)
2668                         return -EINVAL;
2669                 btf_get(attach_btf);
2670         }
2671
2672         bpf_prog_load_fixup_attach_type(attr);
2673         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2674                                        attach_btf, attr->attach_btf_id,
2675                                        dst_prog)) {
2676                 if (dst_prog)
2677                         bpf_prog_put(dst_prog);
2678                 if (attach_btf)
2679                         btf_put(attach_btf);
2680                 return -EINVAL;
2681         }
2682
2683         /* plain bpf_prog allocation */
2684         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2685         if (!prog) {
2686                 if (dst_prog)
2687                         bpf_prog_put(dst_prog);
2688                 if (attach_btf)
2689                         btf_put(attach_btf);
2690                 return -ENOMEM;
2691         }
2692
2693         prog->expected_attach_type = attr->expected_attach_type;
2694         prog->aux->attach_btf = attach_btf;
2695         prog->aux->attach_btf_id = attr->attach_btf_id;
2696         prog->aux->dst_prog = dst_prog;
2697         prog->aux->dev_bound = !!attr->prog_ifindex;
2698         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2699         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2700
2701         err = security_bpf_prog_alloc(prog->aux);
2702         if (err)
2703                 goto free_prog;
2704
2705         prog->aux->user = get_current_user();
2706         prog->len = attr->insn_cnt;
2707
2708         err = -EFAULT;
2709         if (copy_from_bpfptr(prog->insns,
2710                              make_bpfptr(attr->insns, uattr.is_kernel),
2711                              bpf_prog_insn_size(prog)) != 0)
2712                 goto free_prog_sec;
2713         /* copy eBPF program license from user space */
2714         if (strncpy_from_bpfptr(license,
2715                                 make_bpfptr(attr->license, uattr.is_kernel),
2716                                 sizeof(license) - 1) < 0)
2717                 goto free_prog_sec;
2718         license[sizeof(license) - 1] = 0;
2719
2720         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2721         prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2722
2723         prog->orig_prog = NULL;
2724         prog->jited = 0;
2725
2726         atomic64_set(&prog->aux->refcnt, 1);
2727
2728         if (bpf_prog_is_dev_bound(prog->aux)) {
2729                 err = bpf_prog_dev_bound_init(prog, attr);
2730                 if (err)
2731                         goto free_prog_sec;
2732         }
2733
2734         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2735             bpf_prog_is_dev_bound(dst_prog->aux)) {
2736                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2737                 if (err)
2738                         goto free_prog_sec;
2739         }
2740
2741         /*
2742          * Bookkeeping for managing the program attachment chain.
2743          *
2744          * It might be tempting to set attach_tracing_prog flag at the attachment
2745          * time, but this will not prevent from loading bunch of tracing prog
2746          * first, then attach them one to another.
2747          *
2748          * The flag attach_tracing_prog is set for the whole program lifecycle, and
2749          * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2750          * programs cannot change attachment target.
2751          */
2752         if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2753             dst_prog->type == BPF_PROG_TYPE_TRACING) {
2754                 prog->aux->attach_tracing_prog = true;
2755         }
2756
2757         /* find program type: socket_filter vs tracing_filter */
2758         err = find_prog_type(type, prog);
2759         if (err < 0)
2760                 goto free_prog_sec;
2761
2762         prog->aux->load_time = ktime_get_boottime_ns();
2763         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2764                                sizeof(attr->prog_name));
2765         if (err < 0)
2766                 goto free_prog_sec;
2767
2768         /* run eBPF verifier */
2769         err = bpf_check(&prog, attr, uattr, uattr_size);
2770         if (err < 0)
2771                 goto free_used_maps;
2772
2773         prog = bpf_prog_select_runtime(prog, &err);
2774         if (err < 0)
2775                 goto free_used_maps;
2776
2777         err = bpf_prog_alloc_id(prog);
2778         if (err)
2779                 goto free_used_maps;
2780
2781         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2782          * effectively publicly exposed. However, retrieving via
2783          * bpf_prog_get_fd_by_id() will take another reference,
2784          * therefore it cannot be gone underneath us.
2785          *
2786          * Only for the time /after/ successful bpf_prog_new_fd()
2787          * and before returning to userspace, we might just hold
2788          * one reference and any parallel close on that fd could
2789          * rip everything out. Hence, below notifications must
2790          * happen before bpf_prog_new_fd().
2791          *
2792          * Also, any failure handling from this point onwards must
2793          * be using bpf_prog_put() given the program is exposed.
2794          */
2795         bpf_prog_kallsyms_add(prog);
2796         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2797         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2798
2799         err = bpf_prog_new_fd(prog);
2800         if (err < 0)
2801                 bpf_prog_put(prog);
2802         return err;
2803
2804 free_used_maps:
2805         /* In case we have subprogs, we need to wait for a grace
2806          * period before we can tear down JIT memory since symbols
2807          * are already exposed under kallsyms.
2808          */
2809         __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2810         return err;
2811 free_prog_sec:
2812         free_uid(prog->aux->user);
2813         security_bpf_prog_free(prog->aux);
2814 free_prog:
2815         if (prog->aux->attach_btf)
2816                 btf_put(prog->aux->attach_btf);
2817         bpf_prog_free(prog);
2818         return err;
2819 }
2820
2821 #define BPF_OBJ_LAST_FIELD path_fd
2822
2823 static int bpf_obj_pin(const union bpf_attr *attr)
2824 {
2825         int path_fd;
2826
2827         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2828                 return -EINVAL;
2829
2830         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2831         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2832                 return -EINVAL;
2833
2834         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2835         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2836                                 u64_to_user_ptr(attr->pathname));
2837 }
2838
2839 static int bpf_obj_get(const union bpf_attr *attr)
2840 {
2841         int path_fd;
2842
2843         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2844             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2845                 return -EINVAL;
2846
2847         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2848         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2849                 return -EINVAL;
2850
2851         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2852         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2853                                 attr->file_flags);
2854 }
2855
2856 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2857                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2858 {
2859         atomic64_set(&link->refcnt, 1);
2860         link->type = type;
2861         link->id = 0;
2862         link->ops = ops;
2863         link->prog = prog;
2864 }
2865
2866 static void bpf_link_free_id(int id)
2867 {
2868         if (!id)
2869                 return;
2870
2871         spin_lock_bh(&link_idr_lock);
2872         idr_remove(&link_idr, id);
2873         spin_unlock_bh(&link_idr_lock);
2874 }
2875
2876 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2877  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2878  * anon_inode's release() call. This helper marks bpf_link as
2879  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2880  * is not decremented, it's the responsibility of a calling code that failed
2881  * to complete bpf_link initialization.
2882  * This helper eventually calls link's dealloc callback, but does not call
2883  * link's release callback.
2884  */
2885 void bpf_link_cleanup(struct bpf_link_primer *primer)
2886 {
2887         primer->link->prog = NULL;
2888         bpf_link_free_id(primer->id);
2889         fput(primer->file);
2890         put_unused_fd(primer->fd);
2891 }
2892
2893 void bpf_link_inc(struct bpf_link *link)
2894 {
2895         atomic64_inc(&link->refcnt);
2896 }
2897
2898 /* bpf_link_free is guaranteed to be called from process context */
2899 static void bpf_link_free(struct bpf_link *link)
2900 {
2901         bpf_link_free_id(link->id);
2902         if (link->prog) {
2903                 /* detach BPF program, clean up used resources */
2904                 link->ops->release(link);
2905                 bpf_prog_put(link->prog);
2906         }
2907         /* free bpf_link and its containing memory */
2908         link->ops->dealloc(link);
2909 }
2910
2911 static void bpf_link_put_deferred(struct work_struct *work)
2912 {
2913         struct bpf_link *link = container_of(work, struct bpf_link, work);
2914
2915         bpf_link_free(link);
2916 }
2917
2918 /* bpf_link_put might be called from atomic context. It needs to be called
2919  * from sleepable context in order to acquire sleeping locks during the process.
2920  */
2921 void bpf_link_put(struct bpf_link *link)
2922 {
2923         if (!atomic64_dec_and_test(&link->refcnt))
2924                 return;
2925
2926         INIT_WORK(&link->work, bpf_link_put_deferred);
2927         schedule_work(&link->work);
2928 }
2929 EXPORT_SYMBOL(bpf_link_put);
2930
2931 static void bpf_link_put_direct(struct bpf_link *link)
2932 {
2933         if (!atomic64_dec_and_test(&link->refcnt))
2934                 return;
2935         bpf_link_free(link);
2936 }
2937
2938 static int bpf_link_release(struct inode *inode, struct file *filp)
2939 {
2940         struct bpf_link *link = filp->private_data;
2941
2942         bpf_link_put_direct(link);
2943         return 0;
2944 }
2945
2946 #ifdef CONFIG_PROC_FS
2947 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2948 #define BPF_MAP_TYPE(_id, _ops)
2949 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2950 static const char *bpf_link_type_strs[] = {
2951         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2952 #include <linux/bpf_types.h>
2953 };
2954 #undef BPF_PROG_TYPE
2955 #undef BPF_MAP_TYPE
2956 #undef BPF_LINK_TYPE
2957
2958 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2959 {
2960         const struct bpf_link *link = filp->private_data;
2961         const struct bpf_prog *prog = link->prog;
2962         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2963
2964         seq_printf(m,
2965                    "link_type:\t%s\n"
2966                    "link_id:\t%u\n",
2967                    bpf_link_type_strs[link->type],
2968                    link->id);
2969         if (prog) {
2970                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2971                 seq_printf(m,
2972                            "prog_tag:\t%s\n"
2973                            "prog_id:\t%u\n",
2974                            prog_tag,
2975                            prog->aux->id);
2976         }
2977         if (link->ops->show_fdinfo)
2978                 link->ops->show_fdinfo(link, m);
2979 }
2980 #endif
2981
2982 static const struct file_operations bpf_link_fops = {
2983 #ifdef CONFIG_PROC_FS
2984         .show_fdinfo    = bpf_link_show_fdinfo,
2985 #endif
2986         .release        = bpf_link_release,
2987         .read           = bpf_dummy_read,
2988         .write          = bpf_dummy_write,
2989 };
2990
2991 static int bpf_link_alloc_id(struct bpf_link *link)
2992 {
2993         int id;
2994
2995         idr_preload(GFP_KERNEL);
2996         spin_lock_bh(&link_idr_lock);
2997         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2998         spin_unlock_bh(&link_idr_lock);
2999         idr_preload_end();
3000
3001         return id;
3002 }
3003
3004 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3005  * reserving unused FD and allocating ID from link_idr. This is to be paired
3006  * with bpf_link_settle() to install FD and ID and expose bpf_link to
3007  * user-space, if bpf_link is successfully attached. If not, bpf_link and
3008  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3009  * transient state is passed around in struct bpf_link_primer.
3010  * This is preferred way to create and initialize bpf_link, especially when
3011  * there are complicated and expensive operations in between creating bpf_link
3012  * itself and attaching it to BPF hook. By using bpf_link_prime() and
3013  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3014  * expensive (and potentially failing) roll back operations in a rare case
3015  * that file, FD, or ID can't be allocated.
3016  */
3017 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3018 {
3019         struct file *file;
3020         int fd, id;
3021
3022         fd = get_unused_fd_flags(O_CLOEXEC);
3023         if (fd < 0)
3024                 return fd;
3025
3026
3027         id = bpf_link_alloc_id(link);
3028         if (id < 0) {
3029                 put_unused_fd(fd);
3030                 return id;
3031         }
3032
3033         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
3034         if (IS_ERR(file)) {
3035                 bpf_link_free_id(id);
3036                 put_unused_fd(fd);
3037                 return PTR_ERR(file);
3038         }
3039
3040         primer->link = link;
3041         primer->file = file;
3042         primer->fd = fd;
3043         primer->id = id;
3044         return 0;
3045 }
3046
3047 int bpf_link_settle(struct bpf_link_primer *primer)
3048 {
3049         /* make bpf_link fetchable by ID */
3050         spin_lock_bh(&link_idr_lock);
3051         primer->link->id = primer->id;
3052         spin_unlock_bh(&link_idr_lock);
3053         /* make bpf_link fetchable by FD */
3054         fd_install(primer->fd, primer->file);
3055         /* pass through installed FD */
3056         return primer->fd;
3057 }
3058
3059 int bpf_link_new_fd(struct bpf_link *link)
3060 {
3061         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
3062 }
3063
3064 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3065 {
3066         struct fd f = fdget(ufd);
3067         struct bpf_link *link;
3068
3069         if (!f.file)
3070                 return ERR_PTR(-EBADF);
3071         if (f.file->f_op != &bpf_link_fops) {
3072                 fdput(f);
3073                 return ERR_PTR(-EINVAL);
3074         }
3075
3076         link = f.file->private_data;
3077         bpf_link_inc(link);
3078         fdput(f);
3079
3080         return link;
3081 }
3082 EXPORT_SYMBOL(bpf_link_get_from_fd);
3083
3084 static void bpf_tracing_link_release(struct bpf_link *link)
3085 {
3086         struct bpf_tracing_link *tr_link =
3087                 container_of(link, struct bpf_tracing_link, link.link);
3088
3089         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3090                                                 tr_link->trampoline));
3091
3092         bpf_trampoline_put(tr_link->trampoline);
3093
3094         /* tgt_prog is NULL if target is a kernel function */
3095         if (tr_link->tgt_prog)
3096                 bpf_prog_put(tr_link->tgt_prog);
3097 }
3098
3099 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3100 {
3101         struct bpf_tracing_link *tr_link =
3102                 container_of(link, struct bpf_tracing_link, link.link);
3103
3104         kfree(tr_link);
3105 }
3106
3107 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3108                                          struct seq_file *seq)
3109 {
3110         struct bpf_tracing_link *tr_link =
3111                 container_of(link, struct bpf_tracing_link, link.link);
3112         u32 target_btf_id, target_obj_id;
3113
3114         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3115                                   &target_obj_id, &target_btf_id);
3116         seq_printf(seq,
3117                    "attach_type:\t%d\n"
3118                    "target_obj_id:\t%u\n"
3119                    "target_btf_id:\t%u\n",
3120                    tr_link->attach_type,
3121                    target_obj_id,
3122                    target_btf_id);
3123 }
3124
3125 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3126                                            struct bpf_link_info *info)
3127 {
3128         struct bpf_tracing_link *tr_link =
3129                 container_of(link, struct bpf_tracing_link, link.link);
3130
3131         info->tracing.attach_type = tr_link->attach_type;
3132         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3133                                   &info->tracing.target_obj_id,
3134                                   &info->tracing.target_btf_id);
3135
3136         return 0;
3137 }
3138
3139 static const struct bpf_link_ops bpf_tracing_link_lops = {
3140         .release = bpf_tracing_link_release,
3141         .dealloc = bpf_tracing_link_dealloc,
3142         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3143         .fill_link_info = bpf_tracing_link_fill_link_info,
3144 };
3145
3146 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3147                                    int tgt_prog_fd,
3148                                    u32 btf_id,
3149                                    u64 bpf_cookie)
3150 {
3151         struct bpf_link_primer link_primer;
3152         struct bpf_prog *tgt_prog = NULL;
3153         struct bpf_trampoline *tr = NULL;
3154         struct bpf_tracing_link *link;
3155         u64 key = 0;
3156         int err;
3157
3158         switch (prog->type) {
3159         case BPF_PROG_TYPE_TRACING:
3160                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3161                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3162                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3163                         err = -EINVAL;
3164                         goto out_put_prog;
3165                 }
3166                 break;
3167         case BPF_PROG_TYPE_EXT:
3168                 if (prog->expected_attach_type != 0) {
3169                         err = -EINVAL;
3170                         goto out_put_prog;
3171                 }
3172                 break;
3173         case BPF_PROG_TYPE_LSM:
3174                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3175                         err = -EINVAL;
3176                         goto out_put_prog;
3177                 }
3178                 break;
3179         default:
3180                 err = -EINVAL;
3181                 goto out_put_prog;
3182         }
3183
3184         if (!!tgt_prog_fd != !!btf_id) {
3185                 err = -EINVAL;
3186                 goto out_put_prog;
3187         }
3188
3189         if (tgt_prog_fd) {
3190                 /*
3191                  * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3192                  * part would be changed to implement the same for
3193                  * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3194                  * attach_tracing_prog flag is set.
3195                  */
3196                 if (prog->type != BPF_PROG_TYPE_EXT) {
3197                         err = -EINVAL;
3198                         goto out_put_prog;
3199                 }
3200
3201                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3202                 if (IS_ERR(tgt_prog)) {
3203                         err = PTR_ERR(tgt_prog);
3204                         tgt_prog = NULL;
3205                         goto out_put_prog;
3206                 }
3207
3208                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3209         }
3210
3211         link = kzalloc(sizeof(*link), GFP_USER);
3212         if (!link) {
3213                 err = -ENOMEM;
3214                 goto out_put_prog;
3215         }
3216         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3217                       &bpf_tracing_link_lops, prog);
3218         link->attach_type = prog->expected_attach_type;
3219         link->link.cookie = bpf_cookie;
3220
3221         mutex_lock(&prog->aux->dst_mutex);
3222
3223         /* There are a few possible cases here:
3224          *
3225          * - if prog->aux->dst_trampoline is set, the program was just loaded
3226          *   and not yet attached to anything, so we can use the values stored
3227          *   in prog->aux
3228          *
3229          * - if prog->aux->dst_trampoline is NULL, the program has already been
3230          *   attached to a target and its initial target was cleared (below)
3231          *
3232          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3233          *   target_btf_id using the link_create API.
3234          *
3235          * - if tgt_prog == NULL when this function was called using the old
3236          *   raw_tracepoint_open API, and we need a target from prog->aux
3237          *
3238          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3239          *   was detached and is going for re-attachment.
3240          *
3241          * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3242          *   are NULL, then program was already attached and user did not provide
3243          *   tgt_prog_fd so we have no way to find out or create trampoline
3244          */
3245         if (!prog->aux->dst_trampoline && !tgt_prog) {
3246                 /*
3247                  * Allow re-attach for TRACING and LSM programs. If it's
3248                  * currently linked, bpf_trampoline_link_prog will fail.
3249                  * EXT programs need to specify tgt_prog_fd, so they
3250                  * re-attach in separate code path.
3251                  */
3252                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3253                     prog->type != BPF_PROG_TYPE_LSM) {
3254                         err = -EINVAL;
3255                         goto out_unlock;
3256                 }
3257                 /* We can allow re-attach only if we have valid attach_btf. */
3258                 if (!prog->aux->attach_btf) {
3259                         err = -EINVAL;
3260                         goto out_unlock;
3261                 }
3262                 btf_id = prog->aux->attach_btf_id;
3263                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3264         }
3265
3266         if (!prog->aux->dst_trampoline ||
3267             (key && key != prog->aux->dst_trampoline->key)) {
3268                 /* If there is no saved target, or the specified target is
3269                  * different from the destination specified at load time, we
3270                  * need a new trampoline and a check for compatibility
3271                  */
3272                 struct bpf_attach_target_info tgt_info = {};
3273
3274                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3275                                               &tgt_info);
3276                 if (err)
3277                         goto out_unlock;
3278
3279                 if (tgt_info.tgt_mod) {
3280                         module_put(prog->aux->mod);
3281                         prog->aux->mod = tgt_info.tgt_mod;
3282                 }
3283
3284                 tr = bpf_trampoline_get(key, &tgt_info);
3285                 if (!tr) {
3286                         err = -ENOMEM;
3287                         goto out_unlock;
3288                 }
3289         } else {
3290                 /* The caller didn't specify a target, or the target was the
3291                  * same as the destination supplied during program load. This
3292                  * means we can reuse the trampoline and reference from program
3293                  * load time, and there is no need to allocate a new one. This
3294                  * can only happen once for any program, as the saved values in
3295                  * prog->aux are cleared below.
3296                  */
3297                 tr = prog->aux->dst_trampoline;
3298                 tgt_prog = prog->aux->dst_prog;
3299         }
3300
3301         err = bpf_link_prime(&link->link.link, &link_primer);
3302         if (err)
3303                 goto out_unlock;
3304
3305         err = bpf_trampoline_link_prog(&link->link, tr);
3306         if (err) {
3307                 bpf_link_cleanup(&link_primer);
3308                 link = NULL;
3309                 goto out_unlock;
3310         }
3311
3312         link->tgt_prog = tgt_prog;
3313         link->trampoline = tr;
3314
3315         /* Always clear the trampoline and target prog from prog->aux to make
3316          * sure the original attach destination is not kept alive after a
3317          * program is (re-)attached to another target.
3318          */
3319         if (prog->aux->dst_prog &&
3320             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3321                 /* got extra prog ref from syscall, or attaching to different prog */
3322                 bpf_prog_put(prog->aux->dst_prog);
3323         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3324                 /* we allocated a new trampoline, so free the old one */
3325                 bpf_trampoline_put(prog->aux->dst_trampoline);
3326
3327         prog->aux->dst_prog = NULL;
3328         prog->aux->dst_trampoline = NULL;
3329         mutex_unlock(&prog->aux->dst_mutex);
3330
3331         return bpf_link_settle(&link_primer);
3332 out_unlock:
3333         if (tr && tr != prog->aux->dst_trampoline)
3334                 bpf_trampoline_put(tr);
3335         mutex_unlock(&prog->aux->dst_mutex);
3336         kfree(link);
3337 out_put_prog:
3338         if (tgt_prog_fd && tgt_prog)
3339                 bpf_prog_put(tgt_prog);
3340         return err;
3341 }
3342
3343 struct bpf_raw_tp_link {
3344         struct bpf_link link;
3345         struct bpf_raw_event_map *btp;
3346 };
3347
3348 static void bpf_raw_tp_link_release(struct bpf_link *link)
3349 {
3350         struct bpf_raw_tp_link *raw_tp =
3351                 container_of(link, struct bpf_raw_tp_link, link);
3352
3353         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3354         bpf_put_raw_tracepoint(raw_tp->btp);
3355 }
3356
3357 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3358 {
3359         struct bpf_raw_tp_link *raw_tp =
3360                 container_of(link, struct bpf_raw_tp_link, link);
3361
3362         kfree(raw_tp);
3363 }
3364
3365 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3366                                         struct seq_file *seq)
3367 {
3368         struct bpf_raw_tp_link *raw_tp_link =
3369                 container_of(link, struct bpf_raw_tp_link, link);
3370
3371         seq_printf(seq,
3372                    "tp_name:\t%s\n",
3373                    raw_tp_link->btp->tp->name);
3374 }
3375
3376 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3377                             u32 len)
3378 {
3379         if (ulen >= len + 1) {
3380                 if (copy_to_user(ubuf, buf, len + 1))
3381                         return -EFAULT;
3382         } else {
3383                 char zero = '\0';
3384
3385                 if (copy_to_user(ubuf, buf, ulen - 1))
3386                         return -EFAULT;
3387                 if (put_user(zero, ubuf + ulen - 1))
3388                         return -EFAULT;
3389                 return -ENOSPC;
3390         }
3391
3392         return 0;
3393 }
3394
3395 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3396                                           struct bpf_link_info *info)
3397 {
3398         struct bpf_raw_tp_link *raw_tp_link =
3399                 container_of(link, struct bpf_raw_tp_link, link);
3400         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3401         const char *tp_name = raw_tp_link->btp->tp->name;
3402         u32 ulen = info->raw_tracepoint.tp_name_len;
3403         size_t tp_len = strlen(tp_name);
3404
3405         if (!ulen ^ !ubuf)
3406                 return -EINVAL;
3407
3408         info->raw_tracepoint.tp_name_len = tp_len + 1;
3409
3410         if (!ubuf)
3411                 return 0;
3412
3413         return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3414 }
3415
3416 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3417         .release = bpf_raw_tp_link_release,
3418         .dealloc = bpf_raw_tp_link_dealloc,
3419         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3420         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3421 };
3422
3423 #ifdef CONFIG_PERF_EVENTS
3424 struct bpf_perf_link {
3425         struct bpf_link link;
3426         struct file *perf_file;
3427 };
3428
3429 static void bpf_perf_link_release(struct bpf_link *link)
3430 {
3431         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3432         struct perf_event *event = perf_link->perf_file->private_data;
3433
3434         perf_event_free_bpf_prog(event);
3435         fput(perf_link->perf_file);
3436 }
3437
3438 static void bpf_perf_link_dealloc(struct bpf_link *link)
3439 {
3440         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3441
3442         kfree(perf_link);
3443 }
3444
3445 static int bpf_perf_link_fill_common(const struct perf_event *event,
3446                                      char __user *uname, u32 ulen,
3447                                      u64 *probe_offset, u64 *probe_addr,
3448                                      u32 *fd_type, unsigned long *missed)
3449 {
3450         const char *buf;
3451         u32 prog_id;
3452         size_t len;
3453         int err;
3454
3455         if (!ulen ^ !uname)
3456                 return -EINVAL;
3457
3458         err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3459                                       probe_offset, probe_addr, missed);
3460         if (err)
3461                 return err;
3462         if (!uname)
3463                 return 0;
3464         if (buf) {
3465                 len = strlen(buf);
3466                 err = bpf_copy_to_user(uname, buf, ulen, len);
3467                 if (err)
3468                         return err;
3469         } else {
3470                 char zero = '\0';
3471
3472                 if (put_user(zero, uname))
3473                         return -EFAULT;
3474         }
3475         return 0;
3476 }
3477
3478 #ifdef CONFIG_KPROBE_EVENTS
3479 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3480                                      struct bpf_link_info *info)
3481 {
3482         unsigned long missed;
3483         char __user *uname;
3484         u64 addr, offset;
3485         u32 ulen, type;
3486         int err;
3487
3488         uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3489         ulen = info->perf_event.kprobe.name_len;
3490         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3491                                         &type, &missed);
3492         if (err)
3493                 return err;
3494         if (type == BPF_FD_TYPE_KRETPROBE)
3495                 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3496         else
3497                 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3498
3499         info->perf_event.kprobe.offset = offset;
3500         info->perf_event.kprobe.missed = missed;
3501         if (!kallsyms_show_value(current_cred()))
3502                 addr = 0;
3503         info->perf_event.kprobe.addr = addr;
3504         return 0;
3505 }
3506 #endif
3507
3508 #ifdef CONFIG_UPROBE_EVENTS
3509 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3510                                      struct bpf_link_info *info)
3511 {
3512         char __user *uname;
3513         u64 addr, offset;
3514         u32 ulen, type;
3515         int err;
3516
3517         uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3518         ulen = info->perf_event.uprobe.name_len;
3519         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3520                                         &type, NULL);
3521         if (err)
3522                 return err;
3523
3524         if (type == BPF_FD_TYPE_URETPROBE)
3525                 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3526         else
3527                 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3528         info->perf_event.uprobe.offset = offset;
3529         return 0;
3530 }
3531 #endif
3532
3533 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3534                                     struct bpf_link_info *info)
3535 {
3536 #ifdef CONFIG_KPROBE_EVENTS
3537         if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3538                 return bpf_perf_link_fill_kprobe(event, info);
3539 #endif
3540 #ifdef CONFIG_UPROBE_EVENTS
3541         if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3542                 return bpf_perf_link_fill_uprobe(event, info);
3543 #endif
3544         return -EOPNOTSUPP;
3545 }
3546
3547 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3548                                          struct bpf_link_info *info)
3549 {
3550         char __user *uname;
3551         u32 ulen;
3552
3553         uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3554         ulen = info->perf_event.tracepoint.name_len;
3555         info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3556         return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3557 }
3558
3559 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3560                                          struct bpf_link_info *info)
3561 {
3562         info->perf_event.event.type = event->attr.type;
3563         info->perf_event.event.config = event->attr.config;
3564         info->perf_event.type = BPF_PERF_EVENT_EVENT;
3565         return 0;
3566 }
3567
3568 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3569                                         struct bpf_link_info *info)
3570 {
3571         struct bpf_perf_link *perf_link;
3572         const struct perf_event *event;
3573
3574         perf_link = container_of(link, struct bpf_perf_link, link);
3575         event = perf_get_event(perf_link->perf_file);
3576         if (IS_ERR(event))
3577                 return PTR_ERR(event);
3578
3579         switch (event->prog->type) {
3580         case BPF_PROG_TYPE_PERF_EVENT:
3581                 return bpf_perf_link_fill_perf_event(event, info);
3582         case BPF_PROG_TYPE_TRACEPOINT:
3583                 return bpf_perf_link_fill_tracepoint(event, info);
3584         case BPF_PROG_TYPE_KPROBE:
3585                 return bpf_perf_link_fill_probe(event, info);
3586         default:
3587                 return -EOPNOTSUPP;
3588         }
3589 }
3590
3591 static const struct bpf_link_ops bpf_perf_link_lops = {
3592         .release = bpf_perf_link_release,
3593         .dealloc = bpf_perf_link_dealloc,
3594         .fill_link_info = bpf_perf_link_fill_link_info,
3595 };
3596
3597 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3598 {
3599         struct bpf_link_primer link_primer;
3600         struct bpf_perf_link *link;
3601         struct perf_event *event;
3602         struct file *perf_file;
3603         int err;
3604
3605         if (attr->link_create.flags)
3606                 return -EINVAL;
3607
3608         perf_file = perf_event_get(attr->link_create.target_fd);
3609         if (IS_ERR(perf_file))
3610                 return PTR_ERR(perf_file);
3611
3612         link = kzalloc(sizeof(*link), GFP_USER);
3613         if (!link) {
3614                 err = -ENOMEM;
3615                 goto out_put_file;
3616         }
3617         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3618         link->perf_file = perf_file;
3619
3620         err = bpf_link_prime(&link->link, &link_primer);
3621         if (err) {
3622                 kfree(link);
3623                 goto out_put_file;
3624         }
3625
3626         event = perf_file->private_data;
3627         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3628         if (err) {
3629                 bpf_link_cleanup(&link_primer);
3630                 goto out_put_file;
3631         }
3632         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3633         bpf_prog_inc(prog);
3634
3635         return bpf_link_settle(&link_primer);
3636
3637 out_put_file:
3638         fput(perf_file);
3639         return err;
3640 }
3641 #else
3642 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3643 {
3644         return -EOPNOTSUPP;
3645 }
3646 #endif /* CONFIG_PERF_EVENTS */
3647
3648 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3649                                   const char __user *user_tp_name)
3650 {
3651         struct bpf_link_primer link_primer;
3652         struct bpf_raw_tp_link *link;
3653         struct bpf_raw_event_map *btp;
3654         const char *tp_name;
3655         char buf[128];
3656         int err;
3657
3658         switch (prog->type) {
3659         case BPF_PROG_TYPE_TRACING:
3660         case BPF_PROG_TYPE_EXT:
3661         case BPF_PROG_TYPE_LSM:
3662                 if (user_tp_name)
3663                         /* The attach point for this category of programs
3664                          * should be specified via btf_id during program load.
3665                          */
3666                         return -EINVAL;
3667                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3668                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3669                         tp_name = prog->aux->attach_func_name;
3670                         break;
3671                 }
3672                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3673         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3674         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3675                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3676                         return -EFAULT;
3677                 buf[sizeof(buf) - 1] = 0;
3678                 tp_name = buf;
3679                 break;
3680         default:
3681                 return -EINVAL;
3682         }
3683
3684         btp = bpf_get_raw_tracepoint(tp_name);
3685         if (!btp)
3686                 return -ENOENT;
3687
3688         link = kzalloc(sizeof(*link), GFP_USER);
3689         if (!link) {
3690                 err = -ENOMEM;
3691                 goto out_put_btp;
3692         }
3693         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3694                       &bpf_raw_tp_link_lops, prog);
3695         link->btp = btp;
3696
3697         err = bpf_link_prime(&link->link, &link_primer);
3698         if (err) {
3699                 kfree(link);
3700                 goto out_put_btp;
3701         }
3702
3703         err = bpf_probe_register(link->btp, prog);
3704         if (err) {
3705                 bpf_link_cleanup(&link_primer);
3706                 goto out_put_btp;
3707         }
3708
3709         return bpf_link_settle(&link_primer);
3710
3711 out_put_btp:
3712         bpf_put_raw_tracepoint(btp);
3713         return err;
3714 }
3715
3716 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3717
3718 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3719 {
3720         struct bpf_prog *prog;
3721         int fd;
3722
3723         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3724                 return -EINVAL;
3725
3726         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3727         if (IS_ERR(prog))
3728                 return PTR_ERR(prog);
3729
3730         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3731         if (fd < 0)
3732                 bpf_prog_put(prog);
3733         return fd;
3734 }
3735
3736 static enum bpf_prog_type
3737 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3738 {
3739         switch (attach_type) {
3740         case BPF_CGROUP_INET_INGRESS:
3741         case BPF_CGROUP_INET_EGRESS:
3742                 return BPF_PROG_TYPE_CGROUP_SKB;
3743         case BPF_CGROUP_INET_SOCK_CREATE:
3744         case BPF_CGROUP_INET_SOCK_RELEASE:
3745         case BPF_CGROUP_INET4_POST_BIND:
3746         case BPF_CGROUP_INET6_POST_BIND:
3747                 return BPF_PROG_TYPE_CGROUP_SOCK;
3748         case BPF_CGROUP_INET4_BIND:
3749         case BPF_CGROUP_INET6_BIND:
3750         case BPF_CGROUP_INET4_CONNECT:
3751         case BPF_CGROUP_INET6_CONNECT:
3752         case BPF_CGROUP_UNIX_CONNECT:
3753         case BPF_CGROUP_INET4_GETPEERNAME:
3754         case BPF_CGROUP_INET6_GETPEERNAME:
3755         case BPF_CGROUP_UNIX_GETPEERNAME:
3756         case BPF_CGROUP_INET4_GETSOCKNAME:
3757         case BPF_CGROUP_INET6_GETSOCKNAME:
3758         case BPF_CGROUP_UNIX_GETSOCKNAME:
3759         case BPF_CGROUP_UDP4_SENDMSG:
3760         case BPF_CGROUP_UDP6_SENDMSG:
3761         case BPF_CGROUP_UNIX_SENDMSG:
3762         case BPF_CGROUP_UDP4_RECVMSG:
3763         case BPF_CGROUP_UDP6_RECVMSG:
3764         case BPF_CGROUP_UNIX_RECVMSG:
3765                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3766         case BPF_CGROUP_SOCK_OPS:
3767                 return BPF_PROG_TYPE_SOCK_OPS;
3768         case BPF_CGROUP_DEVICE:
3769                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3770         case BPF_SK_MSG_VERDICT:
3771                 return BPF_PROG_TYPE_SK_MSG;
3772         case BPF_SK_SKB_STREAM_PARSER:
3773         case BPF_SK_SKB_STREAM_VERDICT:
3774         case BPF_SK_SKB_VERDICT:
3775                 return BPF_PROG_TYPE_SK_SKB;
3776         case BPF_LIRC_MODE2:
3777                 return BPF_PROG_TYPE_LIRC_MODE2;
3778         case BPF_FLOW_DISSECTOR:
3779                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3780         case BPF_CGROUP_SYSCTL:
3781                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3782         case BPF_CGROUP_GETSOCKOPT:
3783         case BPF_CGROUP_SETSOCKOPT:
3784                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3785         case BPF_TRACE_ITER:
3786         case BPF_TRACE_RAW_TP:
3787         case BPF_TRACE_FENTRY:
3788         case BPF_TRACE_FEXIT:
3789         case BPF_MODIFY_RETURN:
3790                 return BPF_PROG_TYPE_TRACING;
3791         case BPF_LSM_MAC:
3792                 return BPF_PROG_TYPE_LSM;
3793         case BPF_SK_LOOKUP:
3794                 return BPF_PROG_TYPE_SK_LOOKUP;
3795         case BPF_XDP:
3796                 return BPF_PROG_TYPE_XDP;
3797         case BPF_LSM_CGROUP:
3798                 return BPF_PROG_TYPE_LSM;
3799         case BPF_TCX_INGRESS:
3800         case BPF_TCX_EGRESS:
3801         case BPF_NETKIT_PRIMARY:
3802         case BPF_NETKIT_PEER:
3803                 return BPF_PROG_TYPE_SCHED_CLS;
3804         default:
3805                 return BPF_PROG_TYPE_UNSPEC;
3806         }
3807 }
3808
3809 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3810                                              enum bpf_attach_type attach_type)
3811 {
3812         enum bpf_prog_type ptype;
3813
3814         switch (prog->type) {
3815         case BPF_PROG_TYPE_CGROUP_SOCK:
3816         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3817         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3818         case BPF_PROG_TYPE_SK_LOOKUP:
3819                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3820         case BPF_PROG_TYPE_CGROUP_SKB:
3821                 if (!capable(CAP_NET_ADMIN))
3822                         /* cg-skb progs can be loaded by unpriv user.
3823                          * check permissions at attach time.
3824                          */
3825                         return -EPERM;
3826                 return prog->enforce_expected_attach_type &&
3827                         prog->expected_attach_type != attach_type ?
3828                         -EINVAL : 0;
3829         case BPF_PROG_TYPE_EXT:
3830                 return 0;
3831         case BPF_PROG_TYPE_NETFILTER:
3832                 if (attach_type != BPF_NETFILTER)
3833                         return -EINVAL;
3834                 return 0;
3835         case BPF_PROG_TYPE_PERF_EVENT:
3836         case BPF_PROG_TYPE_TRACEPOINT:
3837                 if (attach_type != BPF_PERF_EVENT)
3838                         return -EINVAL;
3839                 return 0;
3840         case BPF_PROG_TYPE_KPROBE:
3841                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3842                     attach_type != BPF_TRACE_KPROBE_MULTI)
3843                         return -EINVAL;
3844                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
3845                     attach_type != BPF_TRACE_UPROBE_MULTI)
3846                         return -EINVAL;
3847                 if (attach_type != BPF_PERF_EVENT &&
3848                     attach_type != BPF_TRACE_KPROBE_MULTI &&
3849                     attach_type != BPF_TRACE_UPROBE_MULTI)
3850                         return -EINVAL;
3851                 return 0;
3852         case BPF_PROG_TYPE_SCHED_CLS:
3853                 if (attach_type != BPF_TCX_INGRESS &&
3854                     attach_type != BPF_TCX_EGRESS &&
3855                     attach_type != BPF_NETKIT_PRIMARY &&
3856                     attach_type != BPF_NETKIT_PEER)
3857                         return -EINVAL;
3858                 return 0;
3859         default:
3860                 ptype = attach_type_to_prog_type(attach_type);
3861                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
3862                         return -EINVAL;
3863                 return 0;
3864         }
3865 }
3866
3867 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
3868
3869 #define BPF_F_ATTACH_MASK_BASE  \
3870         (BPF_F_ALLOW_OVERRIDE | \
3871          BPF_F_ALLOW_MULTI |    \
3872          BPF_F_REPLACE)
3873
3874 #define BPF_F_ATTACH_MASK_MPROG \
3875         (BPF_F_REPLACE |        \
3876          BPF_F_BEFORE |         \
3877          BPF_F_AFTER |          \
3878          BPF_F_ID |             \
3879          BPF_F_LINK)
3880
3881 static int bpf_prog_attach(const union bpf_attr *attr)
3882 {
3883         enum bpf_prog_type ptype;
3884         struct bpf_prog *prog;
3885         int ret;
3886
3887         if (CHECK_ATTR(BPF_PROG_ATTACH))
3888                 return -EINVAL;
3889
3890         ptype = attach_type_to_prog_type(attr->attach_type);
3891         if (ptype == BPF_PROG_TYPE_UNSPEC)
3892                 return -EINVAL;
3893         if (bpf_mprog_supported(ptype)) {
3894                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3895                         return -EINVAL;
3896         } else {
3897                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
3898                         return -EINVAL;
3899                 if (attr->relative_fd ||
3900                     attr->expected_revision)
3901                         return -EINVAL;
3902         }
3903
3904         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3905         if (IS_ERR(prog))
3906                 return PTR_ERR(prog);
3907
3908         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3909                 bpf_prog_put(prog);
3910                 return -EINVAL;
3911         }
3912
3913         switch (ptype) {
3914         case BPF_PROG_TYPE_SK_SKB:
3915         case BPF_PROG_TYPE_SK_MSG:
3916                 ret = sock_map_get_from_fd(attr, prog);
3917                 break;
3918         case BPF_PROG_TYPE_LIRC_MODE2:
3919                 ret = lirc_prog_attach(attr, prog);
3920                 break;
3921         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3922                 ret = netns_bpf_prog_attach(attr, prog);
3923                 break;
3924         case BPF_PROG_TYPE_CGROUP_DEVICE:
3925         case BPF_PROG_TYPE_CGROUP_SKB:
3926         case BPF_PROG_TYPE_CGROUP_SOCK:
3927         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3928         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3929         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3930         case BPF_PROG_TYPE_SOCK_OPS:
3931         case BPF_PROG_TYPE_LSM:
3932                 if (ptype == BPF_PROG_TYPE_LSM &&
3933                     prog->expected_attach_type != BPF_LSM_CGROUP)
3934                         ret = -EINVAL;
3935                 else
3936                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3937                 break;
3938         case BPF_PROG_TYPE_SCHED_CLS:
3939                 if (attr->attach_type == BPF_TCX_INGRESS ||
3940                     attr->attach_type == BPF_TCX_EGRESS)
3941                         ret = tcx_prog_attach(attr, prog);
3942                 else
3943                         ret = netkit_prog_attach(attr, prog);
3944                 break;
3945         default:
3946                 ret = -EINVAL;
3947         }
3948
3949         if (ret)
3950                 bpf_prog_put(prog);
3951         return ret;
3952 }
3953
3954 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
3955
3956 static int bpf_prog_detach(const union bpf_attr *attr)
3957 {
3958         struct bpf_prog *prog = NULL;
3959         enum bpf_prog_type ptype;
3960         int ret;
3961
3962         if (CHECK_ATTR(BPF_PROG_DETACH))
3963                 return -EINVAL;
3964
3965         ptype = attach_type_to_prog_type(attr->attach_type);
3966         if (bpf_mprog_supported(ptype)) {
3967                 if (ptype == BPF_PROG_TYPE_UNSPEC)
3968                         return -EINVAL;
3969                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
3970                         return -EINVAL;
3971                 if (attr->attach_bpf_fd) {
3972                         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3973                         if (IS_ERR(prog))
3974                                 return PTR_ERR(prog);
3975                 }
3976         } else if (attr->attach_flags ||
3977                    attr->relative_fd ||
3978                    attr->expected_revision) {
3979                 return -EINVAL;
3980         }
3981
3982         switch (ptype) {
3983         case BPF_PROG_TYPE_SK_MSG:
3984         case BPF_PROG_TYPE_SK_SKB:
3985                 ret = sock_map_prog_detach(attr, ptype);
3986                 break;
3987         case BPF_PROG_TYPE_LIRC_MODE2:
3988                 ret = lirc_prog_detach(attr);
3989                 break;
3990         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3991                 ret = netns_bpf_prog_detach(attr, ptype);
3992                 break;
3993         case BPF_PROG_TYPE_CGROUP_DEVICE:
3994         case BPF_PROG_TYPE_CGROUP_SKB:
3995         case BPF_PROG_TYPE_CGROUP_SOCK:
3996         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3997         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3998         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3999         case BPF_PROG_TYPE_SOCK_OPS:
4000         case BPF_PROG_TYPE_LSM:
4001                 ret = cgroup_bpf_prog_detach(attr, ptype);
4002                 break;
4003         case BPF_PROG_TYPE_SCHED_CLS:
4004                 if (attr->attach_type == BPF_TCX_INGRESS ||
4005                     attr->attach_type == BPF_TCX_EGRESS)
4006                         ret = tcx_prog_detach(attr, prog);
4007                 else
4008                         ret = netkit_prog_detach(attr, prog);
4009                 break;
4010         default:
4011                 ret = -EINVAL;
4012         }
4013
4014         if (prog)
4015                 bpf_prog_put(prog);
4016         return ret;
4017 }
4018
4019 #define BPF_PROG_QUERY_LAST_FIELD query.revision
4020
4021 static int bpf_prog_query(const union bpf_attr *attr,
4022                           union bpf_attr __user *uattr)
4023 {
4024         if (!capable(CAP_NET_ADMIN))
4025                 return -EPERM;
4026         if (CHECK_ATTR(BPF_PROG_QUERY))
4027                 return -EINVAL;
4028         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4029                 return -EINVAL;
4030
4031         switch (attr->query.attach_type) {
4032         case BPF_CGROUP_INET_INGRESS:
4033         case BPF_CGROUP_INET_EGRESS:
4034         case BPF_CGROUP_INET_SOCK_CREATE:
4035         case BPF_CGROUP_INET_SOCK_RELEASE:
4036         case BPF_CGROUP_INET4_BIND:
4037         case BPF_CGROUP_INET6_BIND:
4038         case BPF_CGROUP_INET4_POST_BIND:
4039         case BPF_CGROUP_INET6_POST_BIND:
4040         case BPF_CGROUP_INET4_CONNECT:
4041         case BPF_CGROUP_INET6_CONNECT:
4042         case BPF_CGROUP_UNIX_CONNECT:
4043         case BPF_CGROUP_INET4_GETPEERNAME:
4044         case BPF_CGROUP_INET6_GETPEERNAME:
4045         case BPF_CGROUP_UNIX_GETPEERNAME:
4046         case BPF_CGROUP_INET4_GETSOCKNAME:
4047         case BPF_CGROUP_INET6_GETSOCKNAME:
4048         case BPF_CGROUP_UNIX_GETSOCKNAME:
4049         case BPF_CGROUP_UDP4_SENDMSG:
4050         case BPF_CGROUP_UDP6_SENDMSG:
4051         case BPF_CGROUP_UNIX_SENDMSG:
4052         case BPF_CGROUP_UDP4_RECVMSG:
4053         case BPF_CGROUP_UDP6_RECVMSG:
4054         case BPF_CGROUP_UNIX_RECVMSG:
4055         case BPF_CGROUP_SOCK_OPS:
4056         case BPF_CGROUP_DEVICE:
4057         case BPF_CGROUP_SYSCTL:
4058         case BPF_CGROUP_GETSOCKOPT:
4059         case BPF_CGROUP_SETSOCKOPT:
4060         case BPF_LSM_CGROUP:
4061                 return cgroup_bpf_prog_query(attr, uattr);
4062         case BPF_LIRC_MODE2:
4063                 return lirc_prog_query(attr, uattr);
4064         case BPF_FLOW_DISSECTOR:
4065         case BPF_SK_LOOKUP:
4066                 return netns_bpf_prog_query(attr, uattr);
4067         case BPF_SK_SKB_STREAM_PARSER:
4068         case BPF_SK_SKB_STREAM_VERDICT:
4069         case BPF_SK_MSG_VERDICT:
4070         case BPF_SK_SKB_VERDICT:
4071                 return sock_map_bpf_prog_query(attr, uattr);
4072         case BPF_TCX_INGRESS:
4073         case BPF_TCX_EGRESS:
4074                 return tcx_prog_query(attr, uattr);
4075         case BPF_NETKIT_PRIMARY:
4076         case BPF_NETKIT_PEER:
4077                 return netkit_prog_query(attr, uattr);
4078         default:
4079                 return -EINVAL;
4080         }
4081 }
4082
4083 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4084
4085 static int bpf_prog_test_run(const union bpf_attr *attr,
4086                              union bpf_attr __user *uattr)
4087 {
4088         struct bpf_prog *prog;
4089         int ret = -ENOTSUPP;
4090
4091         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4092                 return -EINVAL;
4093
4094         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4095             (!attr->test.ctx_size_in && attr->test.ctx_in))
4096                 return -EINVAL;
4097
4098         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4099             (!attr->test.ctx_size_out && attr->test.ctx_out))
4100                 return -EINVAL;
4101
4102         prog = bpf_prog_get(attr->test.prog_fd);
4103         if (IS_ERR(prog))
4104                 return PTR_ERR(prog);
4105
4106         if (prog->aux->ops->test_run)
4107                 ret = prog->aux->ops->test_run(prog, attr, uattr);
4108
4109         bpf_prog_put(prog);
4110         return ret;
4111 }
4112
4113 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4114
4115 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4116                                union bpf_attr __user *uattr,
4117                                struct idr *idr,
4118                                spinlock_t *lock)
4119 {
4120         u32 next_id = attr->start_id;
4121         int err = 0;
4122
4123         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4124                 return -EINVAL;
4125
4126         if (!capable(CAP_SYS_ADMIN))
4127                 return -EPERM;
4128
4129         next_id++;
4130         spin_lock_bh(lock);
4131         if (!idr_get_next(idr, &next_id))
4132                 err = -ENOENT;
4133         spin_unlock_bh(lock);
4134
4135         if (!err)
4136                 err = put_user(next_id, &uattr->next_id);
4137
4138         return err;
4139 }
4140
4141 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4142 {
4143         struct bpf_map *map;
4144
4145         spin_lock_bh(&map_idr_lock);
4146 again:
4147         map = idr_get_next(&map_idr, id);
4148         if (map) {
4149                 map = __bpf_map_inc_not_zero(map, false);
4150                 if (IS_ERR(map)) {
4151                         (*id)++;
4152                         goto again;
4153                 }
4154         }
4155         spin_unlock_bh(&map_idr_lock);
4156
4157         return map;
4158 }
4159
4160 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4161 {
4162         struct bpf_prog *prog;
4163
4164         spin_lock_bh(&prog_idr_lock);
4165 again:
4166         prog = idr_get_next(&prog_idr, id);
4167         if (prog) {
4168                 prog = bpf_prog_inc_not_zero(prog);
4169                 if (IS_ERR(prog)) {
4170                         (*id)++;
4171                         goto again;
4172                 }
4173         }
4174         spin_unlock_bh(&prog_idr_lock);
4175
4176         return prog;
4177 }
4178
4179 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4180
4181 struct bpf_prog *bpf_prog_by_id(u32 id)
4182 {
4183         struct bpf_prog *prog;
4184
4185         if (!id)
4186                 return ERR_PTR(-ENOENT);
4187
4188         spin_lock_bh(&prog_idr_lock);
4189         prog = idr_find(&prog_idr, id);
4190         if (prog)
4191                 prog = bpf_prog_inc_not_zero(prog);
4192         else
4193                 prog = ERR_PTR(-ENOENT);
4194         spin_unlock_bh(&prog_idr_lock);
4195         return prog;
4196 }
4197
4198 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4199 {
4200         struct bpf_prog *prog;
4201         u32 id = attr->prog_id;
4202         int fd;
4203
4204         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4205                 return -EINVAL;
4206
4207         if (!capable(CAP_SYS_ADMIN))
4208                 return -EPERM;
4209
4210         prog = bpf_prog_by_id(id);
4211         if (IS_ERR(prog))
4212                 return PTR_ERR(prog);
4213
4214         fd = bpf_prog_new_fd(prog);
4215         if (fd < 0)
4216                 bpf_prog_put(prog);
4217
4218         return fd;
4219 }
4220
4221 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4222
4223 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4224 {
4225         struct bpf_map *map;
4226         u32 id = attr->map_id;
4227         int f_flags;
4228         int fd;
4229
4230         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4231             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4232                 return -EINVAL;
4233
4234         if (!capable(CAP_SYS_ADMIN))
4235                 return -EPERM;
4236
4237         f_flags = bpf_get_file_flag(attr->open_flags);
4238         if (f_flags < 0)
4239                 return f_flags;
4240
4241         spin_lock_bh(&map_idr_lock);
4242         map = idr_find(&map_idr, id);
4243         if (map)
4244                 map = __bpf_map_inc_not_zero(map, true);
4245         else
4246                 map = ERR_PTR(-ENOENT);
4247         spin_unlock_bh(&map_idr_lock);
4248
4249         if (IS_ERR(map))
4250                 return PTR_ERR(map);
4251
4252         fd = bpf_map_new_fd(map, f_flags);
4253         if (fd < 0)
4254                 bpf_map_put_with_uref(map);
4255
4256         return fd;
4257 }
4258
4259 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4260                                               unsigned long addr, u32 *off,
4261                                               u32 *type)
4262 {
4263         const struct bpf_map *map;
4264         int i;
4265
4266         mutex_lock(&prog->aux->used_maps_mutex);
4267         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4268                 map = prog->aux->used_maps[i];
4269                 if (map == (void *)addr) {
4270                         *type = BPF_PSEUDO_MAP_FD;
4271                         goto out;
4272                 }
4273                 if (!map->ops->map_direct_value_meta)
4274                         continue;
4275                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4276                         *type = BPF_PSEUDO_MAP_VALUE;
4277                         goto out;
4278                 }
4279         }
4280         map = NULL;
4281
4282 out:
4283         mutex_unlock(&prog->aux->used_maps_mutex);
4284         return map;
4285 }
4286
4287 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4288                                               const struct cred *f_cred)
4289 {
4290         const struct bpf_map *map;
4291         struct bpf_insn *insns;
4292         u32 off, type;
4293         u64 imm;
4294         u8 code;
4295         int i;
4296
4297         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4298                         GFP_USER);
4299         if (!insns)
4300                 return insns;
4301
4302         for (i = 0; i < prog->len; i++) {
4303                 code = insns[i].code;
4304
4305                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4306                         insns[i].code = BPF_JMP | BPF_CALL;
4307                         insns[i].imm = BPF_FUNC_tail_call;
4308                         /* fall-through */
4309                 }
4310                 if (code == (BPF_JMP | BPF_CALL) ||
4311                     code == (BPF_JMP | BPF_CALL_ARGS)) {
4312                         if (code == (BPF_JMP | BPF_CALL_ARGS))
4313                                 insns[i].code = BPF_JMP | BPF_CALL;
4314                         if (!bpf_dump_raw_ok(f_cred))
4315                                 insns[i].imm = 0;
4316                         continue;
4317                 }
4318                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4319                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4320                         continue;
4321                 }
4322
4323                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4324                         continue;
4325
4326                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4327                 map = bpf_map_from_imm(prog, imm, &off, &type);
4328                 if (map) {
4329                         insns[i].src_reg = type;
4330                         insns[i].imm = map->id;
4331                         insns[i + 1].imm = off;
4332                         continue;
4333                 }
4334         }
4335
4336         return insns;
4337 }
4338
4339 static int set_info_rec_size(struct bpf_prog_info *info)
4340 {
4341         /*
4342          * Ensure info.*_rec_size is the same as kernel expected size
4343          *
4344          * or
4345          *
4346          * Only allow zero *_rec_size if both _rec_size and _cnt are
4347          * zero.  In this case, the kernel will set the expected
4348          * _rec_size back to the info.
4349          */
4350
4351         if ((info->nr_func_info || info->func_info_rec_size) &&
4352             info->func_info_rec_size != sizeof(struct bpf_func_info))
4353                 return -EINVAL;
4354
4355         if ((info->nr_line_info || info->line_info_rec_size) &&
4356             info->line_info_rec_size != sizeof(struct bpf_line_info))
4357                 return -EINVAL;
4358
4359         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4360             info->jited_line_info_rec_size != sizeof(__u64))
4361                 return -EINVAL;
4362
4363         info->func_info_rec_size = sizeof(struct bpf_func_info);
4364         info->line_info_rec_size = sizeof(struct bpf_line_info);
4365         info->jited_line_info_rec_size = sizeof(__u64);
4366
4367         return 0;
4368 }
4369
4370 static int bpf_prog_get_info_by_fd(struct file *file,
4371                                    struct bpf_prog *prog,
4372                                    const union bpf_attr *attr,
4373                                    union bpf_attr __user *uattr)
4374 {
4375         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4376         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4377         struct bpf_prog_info info;
4378         u32 info_len = attr->info.info_len;
4379         struct bpf_prog_kstats stats;
4380         char __user *uinsns;
4381         u32 ulen;
4382         int err;
4383
4384         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4385         if (err)
4386                 return err;
4387         info_len = min_t(u32, sizeof(info), info_len);
4388
4389         memset(&info, 0, sizeof(info));
4390         if (copy_from_user(&info, uinfo, info_len))
4391                 return -EFAULT;
4392
4393         info.type = prog->type;
4394         info.id = prog->aux->id;
4395         info.load_time = prog->aux->load_time;
4396         info.created_by_uid = from_kuid_munged(current_user_ns(),
4397                                                prog->aux->user->uid);
4398         info.gpl_compatible = prog->gpl_compatible;
4399
4400         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4401         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4402
4403         mutex_lock(&prog->aux->used_maps_mutex);
4404         ulen = info.nr_map_ids;
4405         info.nr_map_ids = prog->aux->used_map_cnt;
4406         ulen = min_t(u32, info.nr_map_ids, ulen);
4407         if (ulen) {
4408                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4409                 u32 i;
4410
4411                 for (i = 0; i < ulen; i++)
4412                         if (put_user(prog->aux->used_maps[i]->id,
4413                                      &user_map_ids[i])) {
4414                                 mutex_unlock(&prog->aux->used_maps_mutex);
4415                                 return -EFAULT;
4416                         }
4417         }
4418         mutex_unlock(&prog->aux->used_maps_mutex);
4419
4420         err = set_info_rec_size(&info);
4421         if (err)
4422                 return err;
4423
4424         bpf_prog_get_stats(prog, &stats);
4425         info.run_time_ns = stats.nsecs;
4426         info.run_cnt = stats.cnt;
4427         info.recursion_misses = stats.misses;
4428
4429         info.verified_insns = prog->aux->verified_insns;
4430
4431         if (!bpf_capable()) {
4432                 info.jited_prog_len = 0;
4433                 info.xlated_prog_len = 0;
4434                 info.nr_jited_ksyms = 0;
4435                 info.nr_jited_func_lens = 0;
4436                 info.nr_func_info = 0;
4437                 info.nr_line_info = 0;
4438                 info.nr_jited_line_info = 0;
4439                 goto done;
4440         }
4441
4442         ulen = info.xlated_prog_len;
4443         info.xlated_prog_len = bpf_prog_insn_size(prog);
4444         if (info.xlated_prog_len && ulen) {
4445                 struct bpf_insn *insns_sanitized;
4446                 bool fault;
4447
4448                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4449                         info.xlated_prog_insns = 0;
4450                         goto done;
4451                 }
4452                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4453                 if (!insns_sanitized)
4454                         return -ENOMEM;
4455                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4456                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4457                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4458                 kfree(insns_sanitized);
4459                 if (fault)
4460                         return -EFAULT;
4461         }
4462
4463         if (bpf_prog_is_offloaded(prog->aux)) {
4464                 err = bpf_prog_offload_info_fill(&info, prog);
4465                 if (err)
4466                         return err;
4467                 goto done;
4468         }
4469
4470         /* NOTE: the following code is supposed to be skipped for offload.
4471          * bpf_prog_offload_info_fill() is the place to fill similar fields
4472          * for offload.
4473          */
4474         ulen = info.jited_prog_len;
4475         if (prog->aux->func_cnt) {
4476                 u32 i;
4477
4478                 info.jited_prog_len = 0;
4479                 for (i = 0; i < prog->aux->func_cnt; i++)
4480                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4481         } else {
4482                 info.jited_prog_len = prog->jited_len;
4483         }
4484
4485         if (info.jited_prog_len && ulen) {
4486                 if (bpf_dump_raw_ok(file->f_cred)) {
4487                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4488                         ulen = min_t(u32, info.jited_prog_len, ulen);
4489
4490                         /* for multi-function programs, copy the JITed
4491                          * instructions for all the functions
4492                          */
4493                         if (prog->aux->func_cnt) {
4494                                 u32 len, free, i;
4495                                 u8 *img;
4496
4497                                 free = ulen;
4498                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4499                                         len = prog->aux->func[i]->jited_len;
4500                                         len = min_t(u32, len, free);
4501                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4502                                         if (copy_to_user(uinsns, img, len))
4503                                                 return -EFAULT;
4504                                         uinsns += len;
4505                                         free -= len;
4506                                         if (!free)
4507                                                 break;
4508                                 }
4509                         } else {
4510                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4511                                         return -EFAULT;
4512                         }
4513                 } else {
4514                         info.jited_prog_insns = 0;
4515                 }
4516         }
4517
4518         ulen = info.nr_jited_ksyms;
4519         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4520         if (ulen) {
4521                 if (bpf_dump_raw_ok(file->f_cred)) {
4522                         unsigned long ksym_addr;
4523                         u64 __user *user_ksyms;
4524                         u32 i;
4525
4526                         /* copy the address of the kernel symbol
4527                          * corresponding to each function
4528                          */
4529                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4530                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4531                         if (prog->aux->func_cnt) {
4532                                 for (i = 0; i < ulen; i++) {
4533                                         ksym_addr = (unsigned long)
4534                                                 prog->aux->func[i]->bpf_func;
4535                                         if (put_user((u64) ksym_addr,
4536                                                      &user_ksyms[i]))
4537                                                 return -EFAULT;
4538                                 }
4539                         } else {
4540                                 ksym_addr = (unsigned long) prog->bpf_func;
4541                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4542                                         return -EFAULT;
4543                         }
4544                 } else {
4545                         info.jited_ksyms = 0;
4546                 }
4547         }
4548
4549         ulen = info.nr_jited_func_lens;
4550         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4551         if (ulen) {
4552                 if (bpf_dump_raw_ok(file->f_cred)) {
4553                         u32 __user *user_lens;
4554                         u32 func_len, i;
4555
4556                         /* copy the JITed image lengths for each function */
4557                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4558                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4559                         if (prog->aux->func_cnt) {
4560                                 for (i = 0; i < ulen; i++) {
4561                                         func_len =
4562                                                 prog->aux->func[i]->jited_len;
4563                                         if (put_user(func_len, &user_lens[i]))
4564                                                 return -EFAULT;
4565                                 }
4566                         } else {
4567                                 func_len = prog->jited_len;
4568                                 if (put_user(func_len, &user_lens[0]))
4569                                         return -EFAULT;
4570                         }
4571                 } else {
4572                         info.jited_func_lens = 0;
4573                 }
4574         }
4575
4576         if (prog->aux->btf)
4577                 info.btf_id = btf_obj_id(prog->aux->btf);
4578         info.attach_btf_id = prog->aux->attach_btf_id;
4579         if (attach_btf)
4580                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4581
4582         ulen = info.nr_func_info;
4583         info.nr_func_info = prog->aux->func_info_cnt;
4584         if (info.nr_func_info && ulen) {
4585                 char __user *user_finfo;
4586
4587                 user_finfo = u64_to_user_ptr(info.func_info);
4588                 ulen = min_t(u32, info.nr_func_info, ulen);
4589                 if (copy_to_user(user_finfo, prog->aux->func_info,
4590                                  info.func_info_rec_size * ulen))
4591                         return -EFAULT;
4592         }
4593
4594         ulen = info.nr_line_info;
4595         info.nr_line_info = prog->aux->nr_linfo;
4596         if (info.nr_line_info && ulen) {
4597                 __u8 __user *user_linfo;
4598
4599                 user_linfo = u64_to_user_ptr(info.line_info);
4600                 ulen = min_t(u32, info.nr_line_info, ulen);
4601                 if (copy_to_user(user_linfo, prog->aux->linfo,
4602                                  info.line_info_rec_size * ulen))
4603                         return -EFAULT;
4604         }
4605
4606         ulen = info.nr_jited_line_info;
4607         if (prog->aux->jited_linfo)
4608                 info.nr_jited_line_info = prog->aux->nr_linfo;
4609         else
4610                 info.nr_jited_line_info = 0;
4611         if (info.nr_jited_line_info && ulen) {
4612                 if (bpf_dump_raw_ok(file->f_cred)) {
4613                         unsigned long line_addr;
4614                         __u64 __user *user_linfo;
4615                         u32 i;
4616
4617                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4618                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4619                         for (i = 0; i < ulen; i++) {
4620                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4621                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4622                                         return -EFAULT;
4623                         }
4624                 } else {
4625                         info.jited_line_info = 0;
4626                 }
4627         }
4628
4629         ulen = info.nr_prog_tags;
4630         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4631         if (ulen) {
4632                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4633                 u32 i;
4634
4635                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4636                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4637                 if (prog->aux->func_cnt) {
4638                         for (i = 0; i < ulen; i++) {
4639                                 if (copy_to_user(user_prog_tags[i],
4640                                                  prog->aux->func[i]->tag,
4641                                                  BPF_TAG_SIZE))
4642                                         return -EFAULT;
4643                         }
4644                 } else {
4645                         if (copy_to_user(user_prog_tags[0],
4646                                          prog->tag, BPF_TAG_SIZE))
4647                                 return -EFAULT;
4648                 }
4649         }
4650
4651 done:
4652         if (copy_to_user(uinfo, &info, info_len) ||
4653             put_user(info_len, &uattr->info.info_len))
4654                 return -EFAULT;
4655
4656         return 0;
4657 }
4658
4659 static int bpf_map_get_info_by_fd(struct file *file,
4660                                   struct bpf_map *map,
4661                                   const union bpf_attr *attr,
4662                                   union bpf_attr __user *uattr)
4663 {
4664         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4665         struct bpf_map_info info;
4666         u32 info_len = attr->info.info_len;
4667         int err;
4668
4669         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4670         if (err)
4671                 return err;
4672         info_len = min_t(u32, sizeof(info), info_len);
4673
4674         memset(&info, 0, sizeof(info));
4675         info.type = map->map_type;
4676         info.id = map->id;
4677         info.key_size = map->key_size;
4678         info.value_size = map->value_size;
4679         info.max_entries = map->max_entries;
4680         info.map_flags = map->map_flags;
4681         info.map_extra = map->map_extra;
4682         memcpy(info.name, map->name, sizeof(map->name));
4683
4684         if (map->btf) {
4685                 info.btf_id = btf_obj_id(map->btf);
4686                 info.btf_key_type_id = map->btf_key_type_id;
4687                 info.btf_value_type_id = map->btf_value_type_id;
4688         }
4689         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4690
4691         if (bpf_map_is_offloaded(map)) {
4692                 err = bpf_map_offload_info_fill(&info, map);
4693                 if (err)
4694                         return err;
4695         }
4696
4697         if (copy_to_user(uinfo, &info, info_len) ||
4698             put_user(info_len, &uattr->info.info_len))
4699                 return -EFAULT;
4700
4701         return 0;
4702 }
4703
4704 static int bpf_btf_get_info_by_fd(struct file *file,
4705                                   struct btf *btf,
4706                                   const union bpf_attr *attr,
4707                                   union bpf_attr __user *uattr)
4708 {
4709         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4710         u32 info_len = attr->info.info_len;
4711         int err;
4712
4713         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4714         if (err)
4715                 return err;
4716
4717         return btf_get_info_by_fd(btf, attr, uattr);
4718 }
4719
4720 static int bpf_link_get_info_by_fd(struct file *file,
4721                                   struct bpf_link *link,
4722                                   const union bpf_attr *attr,
4723                                   union bpf_attr __user *uattr)
4724 {
4725         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4726         struct bpf_link_info info;
4727         u32 info_len = attr->info.info_len;
4728         int err;
4729
4730         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4731         if (err)
4732                 return err;
4733         info_len = min_t(u32, sizeof(info), info_len);
4734
4735         memset(&info, 0, sizeof(info));
4736         if (copy_from_user(&info, uinfo, info_len))
4737                 return -EFAULT;
4738
4739         info.type = link->type;
4740         info.id = link->id;
4741         if (link->prog)
4742                 info.prog_id = link->prog->aux->id;
4743
4744         if (link->ops->fill_link_info) {
4745                 err = link->ops->fill_link_info(link, &info);
4746                 if (err)
4747                         return err;
4748         }
4749
4750         if (copy_to_user(uinfo, &info, info_len) ||
4751             put_user(info_len, &uattr->info.info_len))
4752                 return -EFAULT;
4753
4754         return 0;
4755 }
4756
4757
4758 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4759
4760 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4761                                   union bpf_attr __user *uattr)
4762 {
4763         int ufd = attr->info.bpf_fd;
4764         struct fd f;
4765         int err;
4766
4767         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4768                 return -EINVAL;
4769
4770         f = fdget(ufd);
4771         if (!f.file)
4772                 return -EBADFD;
4773
4774         if (f.file->f_op == &bpf_prog_fops)
4775                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4776                                               uattr);
4777         else if (f.file->f_op == &bpf_map_fops)
4778                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4779                                              uattr);
4780         else if (f.file->f_op == &btf_fops)
4781                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4782         else if (f.file->f_op == &bpf_link_fops)
4783                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4784                                               attr, uattr);
4785         else
4786                 err = -EINVAL;
4787
4788         fdput(f);
4789         return err;
4790 }
4791
4792 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4793
4794 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4795 {
4796         if (CHECK_ATTR(BPF_BTF_LOAD))
4797                 return -EINVAL;
4798
4799         if (!bpf_capable())
4800                 return -EPERM;
4801
4802         return btf_new_fd(attr, uattr, uattr_size);
4803 }
4804
4805 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4806
4807 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4808 {
4809         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4810                 return -EINVAL;
4811
4812         if (!capable(CAP_SYS_ADMIN))
4813                 return -EPERM;
4814
4815         return btf_get_fd_by_id(attr->btf_id);
4816 }
4817
4818 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4819                                     union bpf_attr __user *uattr,
4820                                     u32 prog_id, u32 fd_type,
4821                                     const char *buf, u64 probe_offset,
4822                                     u64 probe_addr)
4823 {
4824         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4825         u32 len = buf ? strlen(buf) : 0, input_len;
4826         int err = 0;
4827
4828         if (put_user(len, &uattr->task_fd_query.buf_len))
4829                 return -EFAULT;
4830         input_len = attr->task_fd_query.buf_len;
4831         if (input_len && ubuf) {
4832                 if (!len) {
4833                         /* nothing to copy, just make ubuf NULL terminated */
4834                         char zero = '\0';
4835
4836                         if (put_user(zero, ubuf))
4837                                 return -EFAULT;
4838                 } else if (input_len >= len + 1) {
4839                         /* ubuf can hold the string with NULL terminator */
4840                         if (copy_to_user(ubuf, buf, len + 1))
4841                                 return -EFAULT;
4842                 } else {
4843                         /* ubuf cannot hold the string with NULL terminator,
4844                          * do a partial copy with NULL terminator.
4845                          */
4846                         char zero = '\0';
4847
4848                         err = -ENOSPC;
4849                         if (copy_to_user(ubuf, buf, input_len - 1))
4850                                 return -EFAULT;
4851                         if (put_user(zero, ubuf + input_len - 1))
4852                                 return -EFAULT;
4853                 }
4854         }
4855
4856         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4857             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4858             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4859             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4860                 return -EFAULT;
4861
4862         return err;
4863 }
4864
4865 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4866
4867 static int bpf_task_fd_query(const union bpf_attr *attr,
4868                              union bpf_attr __user *uattr)
4869 {
4870         pid_t pid = attr->task_fd_query.pid;
4871         u32 fd = attr->task_fd_query.fd;
4872         const struct perf_event *event;
4873         struct task_struct *task;
4874         struct file *file;
4875         int err;
4876
4877         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4878                 return -EINVAL;
4879
4880         if (!capable(CAP_SYS_ADMIN))
4881                 return -EPERM;
4882
4883         if (attr->task_fd_query.flags != 0)
4884                 return -EINVAL;
4885
4886         rcu_read_lock();
4887         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4888         rcu_read_unlock();
4889         if (!task)
4890                 return -ENOENT;
4891
4892         err = 0;
4893         file = fget_task(task, fd);
4894         put_task_struct(task);
4895         if (!file)
4896                 return -EBADF;
4897
4898         if (file->f_op == &bpf_link_fops) {
4899                 struct bpf_link *link = file->private_data;
4900
4901                 if (link->ops == &bpf_raw_tp_link_lops) {
4902                         struct bpf_raw_tp_link *raw_tp =
4903                                 container_of(link, struct bpf_raw_tp_link, link);
4904                         struct bpf_raw_event_map *btp = raw_tp->btp;
4905
4906                         err = bpf_task_fd_query_copy(attr, uattr,
4907                                                      raw_tp->link.prog->aux->id,
4908                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4909                                                      btp->tp->name, 0, 0);
4910                         goto put_file;
4911                 }
4912                 goto out_not_supp;
4913         }
4914
4915         event = perf_get_event(file);
4916         if (!IS_ERR(event)) {
4917                 u64 probe_offset, probe_addr;
4918                 u32 prog_id, fd_type;
4919                 const char *buf;
4920
4921                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4922                                               &buf, &probe_offset,
4923                                               &probe_addr, NULL);
4924                 if (!err)
4925                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4926                                                      fd_type, buf,
4927                                                      probe_offset,
4928                                                      probe_addr);
4929                 goto put_file;
4930         }
4931
4932 out_not_supp:
4933         err = -ENOTSUPP;
4934 put_file:
4935         fput(file);
4936         return err;
4937 }
4938
4939 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4940
4941 #define BPF_DO_BATCH(fn, ...)                   \
4942         do {                                    \
4943                 if (!fn) {                      \
4944                         err = -ENOTSUPP;        \
4945                         goto err_put;           \
4946                 }                               \
4947                 err = fn(__VA_ARGS__);          \
4948         } while (0)
4949
4950 static int bpf_map_do_batch(const union bpf_attr *attr,
4951                             union bpf_attr __user *uattr,
4952                             int cmd)
4953 {
4954         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4955                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4956         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4957         struct bpf_map *map;
4958         int err, ufd;
4959         struct fd f;
4960
4961         if (CHECK_ATTR(BPF_MAP_BATCH))
4962                 return -EINVAL;
4963
4964         ufd = attr->batch.map_fd;
4965         f = fdget(ufd);
4966         map = __bpf_map_get(f);
4967         if (IS_ERR(map))
4968                 return PTR_ERR(map);
4969         if (has_write)
4970                 bpf_map_write_active_inc(map);
4971         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4972                 err = -EPERM;
4973                 goto err_put;
4974         }
4975         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4976                 err = -EPERM;
4977                 goto err_put;
4978         }
4979
4980         if (cmd == BPF_MAP_LOOKUP_BATCH)
4981                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4982         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4983                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4984         else if (cmd == BPF_MAP_UPDATE_BATCH)
4985                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4986         else
4987                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4988 err_put:
4989         if (has_write) {
4990                 maybe_wait_bpf_programs(map);
4991                 bpf_map_write_active_dec(map);
4992         }
4993         fdput(f);
4994         return err;
4995 }
4996
4997 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
4998 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4999 {
5000         struct bpf_prog *prog;
5001         int ret;
5002
5003         if (CHECK_ATTR(BPF_LINK_CREATE))
5004                 return -EINVAL;
5005
5006         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5007                 return bpf_struct_ops_link_create(attr);
5008
5009         prog = bpf_prog_get(attr->link_create.prog_fd);
5010         if (IS_ERR(prog))
5011                 return PTR_ERR(prog);
5012
5013         ret = bpf_prog_attach_check_attach_type(prog,
5014                                                 attr->link_create.attach_type);
5015         if (ret)
5016                 goto out;
5017
5018         switch (prog->type) {
5019         case BPF_PROG_TYPE_CGROUP_SKB:
5020         case BPF_PROG_TYPE_CGROUP_SOCK:
5021         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5022         case BPF_PROG_TYPE_SOCK_OPS:
5023         case BPF_PROG_TYPE_CGROUP_DEVICE:
5024         case BPF_PROG_TYPE_CGROUP_SYSCTL:
5025         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5026                 ret = cgroup_bpf_link_attach(attr, prog);
5027                 break;
5028         case BPF_PROG_TYPE_EXT:
5029                 ret = bpf_tracing_prog_attach(prog,
5030                                               attr->link_create.target_fd,
5031                                               attr->link_create.target_btf_id,
5032                                               attr->link_create.tracing.cookie);
5033                 break;
5034         case BPF_PROG_TYPE_LSM:
5035         case BPF_PROG_TYPE_TRACING:
5036                 if (attr->link_create.attach_type != prog->expected_attach_type) {
5037                         ret = -EINVAL;
5038                         goto out;
5039                 }
5040                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5041                         ret = bpf_raw_tp_link_attach(prog, NULL);
5042                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5043                         ret = bpf_iter_link_attach(attr, uattr, prog);
5044                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5045                         ret = cgroup_bpf_link_attach(attr, prog);
5046                 else
5047                         ret = bpf_tracing_prog_attach(prog,
5048                                                       attr->link_create.target_fd,
5049                                                       attr->link_create.target_btf_id,
5050                                                       attr->link_create.tracing.cookie);
5051                 break;
5052         case BPF_PROG_TYPE_FLOW_DISSECTOR:
5053         case BPF_PROG_TYPE_SK_LOOKUP:
5054                 ret = netns_bpf_link_create(attr, prog);
5055                 break;
5056 #ifdef CONFIG_NET
5057         case BPF_PROG_TYPE_XDP:
5058                 ret = bpf_xdp_link_attach(attr, prog);
5059                 break;
5060         case BPF_PROG_TYPE_SCHED_CLS:
5061                 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5062                     attr->link_create.attach_type == BPF_TCX_EGRESS)
5063                         ret = tcx_link_attach(attr, prog);
5064                 else
5065                         ret = netkit_link_attach(attr, prog);
5066                 break;
5067         case BPF_PROG_TYPE_NETFILTER:
5068                 ret = bpf_nf_link_attach(attr, prog);
5069                 break;
5070 #endif
5071         case BPF_PROG_TYPE_PERF_EVENT:
5072         case BPF_PROG_TYPE_TRACEPOINT:
5073                 ret = bpf_perf_link_attach(attr, prog);
5074                 break;
5075         case BPF_PROG_TYPE_KPROBE:
5076                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5077                         ret = bpf_perf_link_attach(attr, prog);
5078                 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI)
5079                         ret = bpf_kprobe_multi_link_attach(attr, prog);
5080                 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5081                         ret = bpf_uprobe_multi_link_attach(attr, prog);
5082                 break;
5083         default:
5084                 ret = -EINVAL;
5085         }
5086
5087 out:
5088         if (ret < 0)
5089                 bpf_prog_put(prog);
5090         return ret;
5091 }
5092
5093 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5094 {
5095         struct bpf_map *new_map, *old_map = NULL;
5096         int ret;
5097
5098         new_map = bpf_map_get(attr->link_update.new_map_fd);
5099         if (IS_ERR(new_map))
5100                 return PTR_ERR(new_map);
5101
5102         if (attr->link_update.flags & BPF_F_REPLACE) {
5103                 old_map = bpf_map_get(attr->link_update.old_map_fd);
5104                 if (IS_ERR(old_map)) {
5105                         ret = PTR_ERR(old_map);
5106                         goto out_put;
5107                 }
5108         } else if (attr->link_update.old_map_fd) {
5109                 ret = -EINVAL;
5110                 goto out_put;
5111         }
5112
5113         ret = link->ops->update_map(link, new_map, old_map);
5114
5115         if (old_map)
5116                 bpf_map_put(old_map);
5117 out_put:
5118         bpf_map_put(new_map);
5119         return ret;
5120 }
5121
5122 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5123
5124 static int link_update(union bpf_attr *attr)
5125 {
5126         struct bpf_prog *old_prog = NULL, *new_prog;
5127         struct bpf_link *link;
5128         u32 flags;
5129         int ret;
5130
5131         if (CHECK_ATTR(BPF_LINK_UPDATE))
5132                 return -EINVAL;
5133
5134         flags = attr->link_update.flags;
5135         if (flags & ~BPF_F_REPLACE)
5136                 return -EINVAL;
5137
5138         link = bpf_link_get_from_fd(attr->link_update.link_fd);
5139         if (IS_ERR(link))
5140                 return PTR_ERR(link);
5141
5142         if (link->ops->update_map) {
5143                 ret = link_update_map(link, attr);
5144                 goto out_put_link;
5145         }
5146
5147         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5148         if (IS_ERR(new_prog)) {
5149                 ret = PTR_ERR(new_prog);
5150                 goto out_put_link;
5151         }
5152
5153         if (flags & BPF_F_REPLACE) {
5154                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5155                 if (IS_ERR(old_prog)) {
5156                         ret = PTR_ERR(old_prog);
5157                         old_prog = NULL;
5158                         goto out_put_progs;
5159                 }
5160         } else if (attr->link_update.old_prog_fd) {
5161                 ret = -EINVAL;
5162                 goto out_put_progs;
5163         }
5164
5165         if (link->ops->update_prog)
5166                 ret = link->ops->update_prog(link, new_prog, old_prog);
5167         else
5168                 ret = -EINVAL;
5169
5170 out_put_progs:
5171         if (old_prog)
5172                 bpf_prog_put(old_prog);
5173         if (ret)
5174                 bpf_prog_put(new_prog);
5175 out_put_link:
5176         bpf_link_put_direct(link);
5177         return ret;
5178 }
5179
5180 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5181
5182 static int link_detach(union bpf_attr *attr)
5183 {
5184         struct bpf_link *link;
5185         int ret;
5186
5187         if (CHECK_ATTR(BPF_LINK_DETACH))
5188                 return -EINVAL;
5189
5190         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5191         if (IS_ERR(link))
5192                 return PTR_ERR(link);
5193
5194         if (link->ops->detach)
5195                 ret = link->ops->detach(link);
5196         else
5197                 ret = -EOPNOTSUPP;
5198
5199         bpf_link_put_direct(link);
5200         return ret;
5201 }
5202
5203 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5204 {
5205         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5206 }
5207
5208 struct bpf_link *bpf_link_by_id(u32 id)
5209 {
5210         struct bpf_link *link;
5211
5212         if (!id)
5213                 return ERR_PTR(-ENOENT);
5214
5215         spin_lock_bh(&link_idr_lock);
5216         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5217         link = idr_find(&link_idr, id);
5218         if (link) {
5219                 if (link->id)
5220                         link = bpf_link_inc_not_zero(link);
5221                 else
5222                         link = ERR_PTR(-EAGAIN);
5223         } else {
5224                 link = ERR_PTR(-ENOENT);
5225         }
5226         spin_unlock_bh(&link_idr_lock);
5227         return link;
5228 }
5229
5230 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5231 {
5232         struct bpf_link *link;
5233
5234         spin_lock_bh(&link_idr_lock);
5235 again:
5236         link = idr_get_next(&link_idr, id);
5237         if (link) {
5238                 link = bpf_link_inc_not_zero(link);
5239                 if (IS_ERR(link)) {
5240                         (*id)++;
5241                         goto again;
5242                 }
5243         }
5244         spin_unlock_bh(&link_idr_lock);
5245
5246         return link;
5247 }
5248
5249 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5250
5251 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5252 {
5253         struct bpf_link *link;
5254         u32 id = attr->link_id;
5255         int fd;
5256
5257         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5258                 return -EINVAL;
5259
5260         if (!capable(CAP_SYS_ADMIN))
5261                 return -EPERM;
5262
5263         link = bpf_link_by_id(id);
5264         if (IS_ERR(link))
5265                 return PTR_ERR(link);
5266
5267         fd = bpf_link_new_fd(link);
5268         if (fd < 0)
5269                 bpf_link_put_direct(link);
5270
5271         return fd;
5272 }
5273
5274 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5275
5276 static int bpf_stats_release(struct inode *inode, struct file *file)
5277 {
5278         mutex_lock(&bpf_stats_enabled_mutex);
5279         static_key_slow_dec(&bpf_stats_enabled_key.key);
5280         mutex_unlock(&bpf_stats_enabled_mutex);
5281         return 0;
5282 }
5283
5284 static const struct file_operations bpf_stats_fops = {
5285         .release = bpf_stats_release,
5286 };
5287
5288 static int bpf_enable_runtime_stats(void)
5289 {
5290         int fd;
5291
5292         mutex_lock(&bpf_stats_enabled_mutex);
5293
5294         /* Set a very high limit to avoid overflow */
5295         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5296                 mutex_unlock(&bpf_stats_enabled_mutex);
5297                 return -EBUSY;
5298         }
5299
5300         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5301         if (fd >= 0)
5302                 static_key_slow_inc(&bpf_stats_enabled_key.key);
5303
5304         mutex_unlock(&bpf_stats_enabled_mutex);
5305         return fd;
5306 }
5307
5308 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5309
5310 static int bpf_enable_stats(union bpf_attr *attr)
5311 {
5312
5313         if (CHECK_ATTR(BPF_ENABLE_STATS))
5314                 return -EINVAL;
5315
5316         if (!capable(CAP_SYS_ADMIN))
5317                 return -EPERM;
5318
5319         switch (attr->enable_stats.type) {
5320         case BPF_STATS_RUN_TIME:
5321                 return bpf_enable_runtime_stats();
5322         default:
5323                 break;
5324         }
5325         return -EINVAL;
5326 }
5327
5328 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5329
5330 static int bpf_iter_create(union bpf_attr *attr)
5331 {
5332         struct bpf_link *link;
5333         int err;
5334
5335         if (CHECK_ATTR(BPF_ITER_CREATE))
5336                 return -EINVAL;
5337
5338         if (attr->iter_create.flags)
5339                 return -EINVAL;
5340
5341         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5342         if (IS_ERR(link))
5343                 return PTR_ERR(link);
5344
5345         err = bpf_iter_new_fd(link);
5346         bpf_link_put_direct(link);
5347
5348         return err;
5349 }
5350
5351 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5352
5353 static int bpf_prog_bind_map(union bpf_attr *attr)
5354 {
5355         struct bpf_prog *prog;
5356         struct bpf_map *map;
5357         struct bpf_map **used_maps_old, **used_maps_new;
5358         int i, ret = 0;
5359
5360         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5361                 return -EINVAL;
5362
5363         if (attr->prog_bind_map.flags)
5364                 return -EINVAL;
5365
5366         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5367         if (IS_ERR(prog))
5368                 return PTR_ERR(prog);
5369
5370         map = bpf_map_get(attr->prog_bind_map.map_fd);
5371         if (IS_ERR(map)) {
5372                 ret = PTR_ERR(map);
5373                 goto out_prog_put;
5374         }
5375
5376         mutex_lock(&prog->aux->used_maps_mutex);
5377
5378         used_maps_old = prog->aux->used_maps;
5379
5380         for (i = 0; i < prog->aux->used_map_cnt; i++)
5381                 if (used_maps_old[i] == map) {
5382                         bpf_map_put(map);
5383                         goto out_unlock;
5384                 }
5385
5386         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5387                                       sizeof(used_maps_new[0]),
5388                                       GFP_KERNEL);
5389         if (!used_maps_new) {
5390                 ret = -ENOMEM;
5391                 goto out_unlock;
5392         }
5393
5394         /* The bpf program will not access the bpf map, but for the sake of
5395          * simplicity, increase sleepable_refcnt for sleepable program as well.
5396          */
5397         if (prog->aux->sleepable)
5398                 atomic64_inc(&map->sleepable_refcnt);
5399         memcpy(used_maps_new, used_maps_old,
5400                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5401         used_maps_new[prog->aux->used_map_cnt] = map;
5402
5403         prog->aux->used_map_cnt++;
5404         prog->aux->used_maps = used_maps_new;
5405
5406         kfree(used_maps_old);
5407
5408 out_unlock:
5409         mutex_unlock(&prog->aux->used_maps_mutex);
5410
5411         if (ret)
5412                 bpf_map_put(map);
5413 out_prog_put:
5414         bpf_prog_put(prog);
5415         return ret;
5416 }
5417
5418 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5419 {
5420         union bpf_attr attr;
5421         int err;
5422
5423         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5424         if (err)
5425                 return err;
5426         size = min_t(u32, size, sizeof(attr));
5427
5428         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5429         memset(&attr, 0, sizeof(attr));
5430         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5431                 return -EFAULT;
5432
5433         err = security_bpf(cmd, &attr, size);
5434         if (err < 0)
5435                 return err;
5436
5437         switch (cmd) {
5438         case BPF_MAP_CREATE:
5439                 err = map_create(&attr);
5440                 break;
5441         case BPF_MAP_LOOKUP_ELEM:
5442                 err = map_lookup_elem(&attr);
5443                 break;
5444         case BPF_MAP_UPDATE_ELEM:
5445                 err = map_update_elem(&attr, uattr);
5446                 break;
5447         case BPF_MAP_DELETE_ELEM:
5448                 err = map_delete_elem(&attr, uattr);
5449                 break;
5450         case BPF_MAP_GET_NEXT_KEY:
5451                 err = map_get_next_key(&attr);
5452                 break;
5453         case BPF_MAP_FREEZE:
5454                 err = map_freeze(&attr);
5455                 break;
5456         case BPF_PROG_LOAD:
5457                 err = bpf_prog_load(&attr, uattr, size);
5458                 break;
5459         case BPF_OBJ_PIN:
5460                 err = bpf_obj_pin(&attr);
5461                 break;
5462         case BPF_OBJ_GET:
5463                 err = bpf_obj_get(&attr);
5464                 break;
5465         case BPF_PROG_ATTACH:
5466                 err = bpf_prog_attach(&attr);
5467                 break;
5468         case BPF_PROG_DETACH:
5469                 err = bpf_prog_detach(&attr);
5470                 break;
5471         case BPF_PROG_QUERY:
5472                 err = bpf_prog_query(&attr, uattr.user);
5473                 break;
5474         case BPF_PROG_TEST_RUN:
5475                 err = bpf_prog_test_run(&attr, uattr.user);
5476                 break;
5477         case BPF_PROG_GET_NEXT_ID:
5478                 err = bpf_obj_get_next_id(&attr, uattr.user,
5479                                           &prog_idr, &prog_idr_lock);
5480                 break;
5481         case BPF_MAP_GET_NEXT_ID:
5482                 err = bpf_obj_get_next_id(&attr, uattr.user,
5483                                           &map_idr, &map_idr_lock);
5484                 break;
5485         case BPF_BTF_GET_NEXT_ID:
5486                 err = bpf_obj_get_next_id(&attr, uattr.user,
5487                                           &btf_idr, &btf_idr_lock);
5488                 break;
5489         case BPF_PROG_GET_FD_BY_ID:
5490                 err = bpf_prog_get_fd_by_id(&attr);
5491                 break;
5492         case BPF_MAP_GET_FD_BY_ID:
5493                 err = bpf_map_get_fd_by_id(&attr);
5494                 break;
5495         case BPF_OBJ_GET_INFO_BY_FD:
5496                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5497                 break;
5498         case BPF_RAW_TRACEPOINT_OPEN:
5499                 err = bpf_raw_tracepoint_open(&attr);
5500                 break;
5501         case BPF_BTF_LOAD:
5502                 err = bpf_btf_load(&attr, uattr, size);
5503                 break;
5504         case BPF_BTF_GET_FD_BY_ID:
5505                 err = bpf_btf_get_fd_by_id(&attr);
5506                 break;
5507         case BPF_TASK_FD_QUERY:
5508                 err = bpf_task_fd_query(&attr, uattr.user);
5509                 break;
5510         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5511                 err = map_lookup_and_delete_elem(&attr);
5512                 break;
5513         case BPF_MAP_LOOKUP_BATCH:
5514                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5515                 break;
5516         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5517                 err = bpf_map_do_batch(&attr, uattr.user,
5518                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5519                 break;
5520         case BPF_MAP_UPDATE_BATCH:
5521                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5522                 break;
5523         case BPF_MAP_DELETE_BATCH:
5524                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5525                 break;
5526         case BPF_LINK_CREATE:
5527                 err = link_create(&attr, uattr);
5528                 break;
5529         case BPF_LINK_UPDATE:
5530                 err = link_update(&attr);
5531                 break;
5532         case BPF_LINK_GET_FD_BY_ID:
5533                 err = bpf_link_get_fd_by_id(&attr);
5534                 break;
5535         case BPF_LINK_GET_NEXT_ID:
5536                 err = bpf_obj_get_next_id(&attr, uattr.user,
5537                                           &link_idr, &link_idr_lock);
5538                 break;
5539         case BPF_ENABLE_STATS:
5540                 err = bpf_enable_stats(&attr);
5541                 break;
5542         case BPF_ITER_CREATE:
5543                 err = bpf_iter_create(&attr);
5544                 break;
5545         case BPF_LINK_DETACH:
5546                 err = link_detach(&attr);
5547                 break;
5548         case BPF_PROG_BIND_MAP:
5549                 err = bpf_prog_bind_map(&attr);
5550                 break;
5551         default:
5552                 err = -EINVAL;
5553                 break;
5554         }
5555
5556         return err;
5557 }
5558
5559 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5560 {
5561         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5562 }
5563
5564 static bool syscall_prog_is_valid_access(int off, int size,
5565                                          enum bpf_access_type type,
5566                                          const struct bpf_prog *prog,
5567                                          struct bpf_insn_access_aux *info)
5568 {
5569         if (off < 0 || off >= U16_MAX)
5570                 return false;
5571         if (off % size != 0)
5572                 return false;
5573         return true;
5574 }
5575
5576 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5577 {
5578         switch (cmd) {
5579         case BPF_MAP_CREATE:
5580         case BPF_MAP_DELETE_ELEM:
5581         case BPF_MAP_UPDATE_ELEM:
5582         case BPF_MAP_FREEZE:
5583         case BPF_MAP_GET_FD_BY_ID:
5584         case BPF_PROG_LOAD:
5585         case BPF_BTF_LOAD:
5586         case BPF_LINK_CREATE:
5587         case BPF_RAW_TRACEPOINT_OPEN:
5588                 break;
5589         default:
5590                 return -EINVAL;
5591         }
5592         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5593 }
5594
5595
5596 /* To shut up -Wmissing-prototypes.
5597  * This function is used by the kernel light skeleton
5598  * to load bpf programs when modules are loaded or during kernel boot.
5599  * See tools/lib/bpf/skel_internal.h
5600  */
5601 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5602
5603 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5604 {
5605         struct bpf_prog * __maybe_unused prog;
5606         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5607
5608         switch (cmd) {
5609 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5610         case BPF_PROG_TEST_RUN:
5611                 if (attr->test.data_in || attr->test.data_out ||
5612                     attr->test.ctx_out || attr->test.duration ||
5613                     attr->test.repeat || attr->test.flags)
5614                         return -EINVAL;
5615
5616                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5617                 if (IS_ERR(prog))
5618                         return PTR_ERR(prog);
5619
5620                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5621                     attr->test.ctx_size_in > U16_MAX) {
5622                         bpf_prog_put(prog);
5623                         return -EINVAL;
5624                 }
5625
5626                 run_ctx.bpf_cookie = 0;
5627                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5628                         /* recursion detected */
5629                         __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5630                         bpf_prog_put(prog);
5631                         return -EBUSY;
5632                 }
5633                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5634                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5635                                                 &run_ctx);
5636                 bpf_prog_put(prog);
5637                 return 0;
5638 #endif
5639         default:
5640                 return ____bpf_sys_bpf(cmd, attr, size);
5641         }
5642 }
5643 EXPORT_SYMBOL(kern_sys_bpf);
5644
5645 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5646         .func           = bpf_sys_bpf,
5647         .gpl_only       = false,
5648         .ret_type       = RET_INTEGER,
5649         .arg1_type      = ARG_ANYTHING,
5650         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5651         .arg3_type      = ARG_CONST_SIZE,
5652 };
5653
5654 const struct bpf_func_proto * __weak
5655 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5656 {
5657         return bpf_base_func_proto(func_id);
5658 }
5659
5660 BPF_CALL_1(bpf_sys_close, u32, fd)
5661 {
5662         /* When bpf program calls this helper there should not be
5663          * an fdget() without matching completed fdput().
5664          * This helper is allowed in the following callchain only:
5665          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5666          */
5667         return close_fd(fd);
5668 }
5669
5670 static const struct bpf_func_proto bpf_sys_close_proto = {
5671         .func           = bpf_sys_close,
5672         .gpl_only       = false,
5673         .ret_type       = RET_INTEGER,
5674         .arg1_type      = ARG_ANYTHING,
5675 };
5676
5677 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5678 {
5679         if (flags)
5680                 return -EINVAL;
5681
5682         if (name_sz <= 1 || name[name_sz - 1])
5683                 return -EINVAL;
5684
5685         if (!bpf_dump_raw_ok(current_cred()))
5686                 return -EPERM;
5687
5688         *res = kallsyms_lookup_name(name);
5689         return *res ? 0 : -ENOENT;
5690 }
5691
5692 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5693         .func           = bpf_kallsyms_lookup_name,
5694         .gpl_only       = false,
5695         .ret_type       = RET_INTEGER,
5696         .arg1_type      = ARG_PTR_TO_MEM,
5697         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5698         .arg3_type      = ARG_ANYTHING,
5699         .arg4_type      = ARG_PTR_TO_LONG,
5700 };
5701
5702 static const struct bpf_func_proto *
5703 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5704 {
5705         switch (func_id) {
5706         case BPF_FUNC_sys_bpf:
5707                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5708         case BPF_FUNC_btf_find_by_name_kind:
5709                 return &bpf_btf_find_by_name_kind_proto;
5710         case BPF_FUNC_sys_close:
5711                 return &bpf_sys_close_proto;
5712         case BPF_FUNC_kallsyms_lookup_name:
5713                 return &bpf_kallsyms_lookup_name_proto;
5714         default:
5715                 return tracing_prog_func_proto(func_id, prog);
5716         }
5717 }
5718
5719 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5720         .get_func_proto  = syscall_prog_func_proto,
5721         .is_valid_access = syscall_prog_is_valid_access,
5722 };
5723
5724 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5725         .test_run = bpf_prog_test_run_syscall,
5726 };
5727
5728 #ifdef CONFIG_SYSCTL
5729 static int bpf_stats_handler(struct ctl_table *table, int write,
5730                              void *buffer, size_t *lenp, loff_t *ppos)
5731 {
5732         struct static_key *key = (struct static_key *)table->data;
5733         static int saved_val;
5734         int val, ret;
5735         struct ctl_table tmp = {
5736                 .data   = &val,
5737                 .maxlen = sizeof(val),
5738                 .mode   = table->mode,
5739                 .extra1 = SYSCTL_ZERO,
5740                 .extra2 = SYSCTL_ONE,
5741         };
5742
5743         if (write && !capable(CAP_SYS_ADMIN))
5744                 return -EPERM;
5745
5746         mutex_lock(&bpf_stats_enabled_mutex);
5747         val = saved_val;
5748         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5749         if (write && !ret && val != saved_val) {
5750                 if (val)
5751                         static_key_slow_inc(key);
5752                 else
5753                         static_key_slow_dec(key);
5754                 saved_val = val;
5755         }
5756         mutex_unlock(&bpf_stats_enabled_mutex);
5757         return ret;
5758 }
5759
5760 void __weak unpriv_ebpf_notify(int new_state)
5761 {
5762 }
5763
5764 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5765                               void *buffer, size_t *lenp, loff_t *ppos)
5766 {
5767         int ret, unpriv_enable = *(int *)table->data;
5768         bool locked_state = unpriv_enable == 1;
5769         struct ctl_table tmp = *table;
5770
5771         if (write && !capable(CAP_SYS_ADMIN))
5772                 return -EPERM;
5773
5774         tmp.data = &unpriv_enable;
5775         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5776         if (write && !ret) {
5777                 if (locked_state && unpriv_enable != 1)
5778                         return -EPERM;
5779                 *(int *)table->data = unpriv_enable;
5780         }
5781
5782         if (write)
5783                 unpriv_ebpf_notify(unpriv_enable);
5784
5785         return ret;
5786 }
5787
5788 static struct ctl_table bpf_syscall_table[] = {
5789         {
5790                 .procname       = "unprivileged_bpf_disabled",
5791                 .data           = &sysctl_unprivileged_bpf_disabled,
5792                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5793                 .mode           = 0644,
5794                 .proc_handler   = bpf_unpriv_handler,
5795                 .extra1         = SYSCTL_ZERO,
5796                 .extra2         = SYSCTL_TWO,
5797         },
5798         {
5799                 .procname       = "bpf_stats_enabled",
5800                 .data           = &bpf_stats_enabled_key.key,
5801                 .mode           = 0644,
5802                 .proc_handler   = bpf_stats_handler,
5803         },
5804         { }
5805 };
5806
5807 static int __init bpf_syscall_sysctl_init(void)
5808 {
5809         register_sysctl_init("kernel", bpf_syscall_table);
5810         return 0;
5811 }
5812 late_initcall(bpf_syscall_sysctl_init);
5813 #endif /* CONFIG_SYSCTL */