kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf-cgroup.h>
   6 #include <linux/bpf_trace.h>
   7 #include <linux/bpf_lirc.h>
   8 #include <linux/bpf_verifier.h>
   9 #include <linux/bsearch.h>
  10 #include <linux/btf.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mmzone.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/fdtable.h>
  18 #include <linux/file.h>
  19 #include <linux/fs.h>
  20 #include <linux/license.h>
  21 #include <linux/filter.h>
  22 #include <linux/kernel.h>
  23 #include <linux/idr.h>
  24 #include <linux/cred.h>
  25 #include <linux/timekeeping.h>
  26 #include <linux/ctype.h>
  27 #include <linux/nospec.h>
  28 #include <linux/audit.h>
  29 #include <uapi/linux/btf.h>
  30 #include <linux/pgtable.h>
  31 #include <linux/bpf_lsm.h>
  32 #include <linux/poll.h>
  33 #include <linux/sort.h>
  34 #include <linux/bpf-netns.h>
  35 #include <linux/rcupdate_trace.h>
  36 #include <linux/memcontrol.h>
  37 #include <linux/trace_events.h>
  38 #include <net/netfilter/nf_bpf_link.h>
  39
  40 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  41                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  42                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  43 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  44 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  45 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  46                         IS_FD_HASH(map))
  47
  48 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  49
  50 DEFINE_PER_CPU(int, bpf_prog_active);
  51 static DEFINE_IDR(prog_idr);
  52 static DEFINE_SPINLOCK(prog_idr_lock);
  53 static DEFINE_IDR(map_idr);
  54 static DEFINE_SPINLOCK(map_idr_lock);
  55 static DEFINE_IDR(link_idr);
  56 static DEFINE_SPINLOCK(link_idr_lock);
  57
  58 int sysctl_unprivileged_bpf_disabled __read_mostly =
  59         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  60
  61 static const struct bpf_map_ops * const bpf_map_types[] = {
  62 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  63 #define BPF_MAP_TYPE(_id, _ops) \
  64         [_id] = &_ops,
  65 #define BPF_LINK_TYPE(_id, _name)
  66 #include <linux/bpf_types.h>
  67 #undef BPF_PROG_TYPE
  68 #undef BPF_MAP_TYPE
  69 #undef BPF_LINK_TYPE
  70 };
  71
  72 /*
  73  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  74  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  75  * we don't know about yet.
  76  *
  77  * There is a ToCToU between this function call and the following
  78  * copy_from_user() call. However, this is not a concern since this function is
  79  * meant to be a future-proofing of bits.
  80  */
  81 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  82                              size_t expected_size,
  83                              size_t actual_size)
  84 {
  85         int res;
  86
  87         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  88                 return -E2BIG;
  89
  90         if (actual_size <= expected_size)
  91                 return 0;
  92
  93         if (uaddr.is_kernel)
  94                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  95                                  actual_size - expected_size) == NULL;
  96         else
  97                 res = check_zeroed_user(uaddr.user + expected_size,
  98                                         actual_size - expected_size);
  99         if (res < 0)
 100                 return res;
 101         return res ? 0 : -E2BIG;
 102 }
 103
 104 const struct bpf_map_ops bpf_map_offload_ops = {
 105         .map_meta_equal = bpf_map_meta_equal,
 106         .map_alloc = bpf_map_offload_map_alloc,
 107         .map_free = bpf_map_offload_map_free,
 108         .map_check_btf = map_check_no_btf,
 109         .map_mem_usage = bpf_map_offload_map_mem_usage,
 110 };
 111
 112 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 113 {
 114         const struct bpf_map_ops *ops;
 115         u32 type = attr->map_type;
 116         struct bpf_map *map;
 117         int err;
 118
 119         if (type >= ARRAY_SIZE(bpf_map_types))
 120                 return ERR_PTR(-EINVAL);
 121         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 122         ops = bpf_map_types[type];
 123         if (!ops)
 124                 return ERR_PTR(-EINVAL);
 125
 126         if (ops->map_alloc_check) {
 127                 err = ops->map_alloc_check(attr);
 128                 if (err)
 129                         return ERR_PTR(err);
 130         }
 131         if (attr->map_ifindex)
 132                 ops = &bpf_map_offload_ops;
 133         if (!ops->map_mem_usage)
 134                 return ERR_PTR(-EINVAL);
 135         map = ops->map_alloc(attr);
 136         if (IS_ERR(map))
 137                 return map;
 138         map->ops = ops;
 139         map->map_type = type;
 140         return map;
 141 }
 142
 143 static void bpf_map_write_active_inc(struct bpf_map *map)
 144 {
 145         atomic64_inc(&map->writecnt);
 146 }
 147
 148 static void bpf_map_write_active_dec(struct bpf_map *map)
 149 {
 150         atomic64_dec(&map->writecnt);
 151 }
 152
 153 bool bpf_map_write_active(const struct bpf_map *map)
 154 {
 155         return atomic64_read(&map->writecnt) != 0;
 156 }
 157
 158 static u32 bpf_map_value_size(const struct bpf_map *map)
 159 {
 160         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 161             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 162             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 163             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 164                 return round_up(map->value_size, 8) * num_possible_cpus();
 165         else if (IS_FD_MAP(map))
 166                 return sizeof(u32);
 167         else
 168                 return  map->value_size;
 169 }
 170
 171 static void maybe_wait_bpf_programs(struct bpf_map *map)
 172 {
 173         /* Wait for any running BPF programs to complete so that
 174          * userspace, when we return to it, knows that all programs
 175          * that could be running use the new map value.
 176          */
 177         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 178             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 179                 synchronize_rcu();
 180 }
 181
 182 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 183                                 void *key, void *value, __u64 flags)
 184 {
 185         int err;
 186
 187         /* Need to create a kthread, thus must support schedule */
 188         if (bpf_map_is_offloaded(map)) {
 189                 return bpf_map_offload_update_elem(map, key, value, flags);
 190         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 191                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 192                 return map->ops->map_update_elem(map, key, value, flags);
 193         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 194                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 195                 return sock_map_update_elem_sys(map, key, value, flags);
 196         } else if (IS_FD_PROG_ARRAY(map)) {
 197                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 198                                                     flags);
 199         }
 200
 201         bpf_disable_instrumentation();
 202         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 203             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 204                 err = bpf_percpu_hash_update(map, key, value, flags);
 205         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 206                 err = bpf_percpu_array_update(map, key, value, flags);
 207         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 208                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 209                                                        flags);
 210         } else if (IS_FD_ARRAY(map)) {
 211                 rcu_read_lock();
 212                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 213                                                    flags);
 214                 rcu_read_unlock();
 215         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 216                 rcu_read_lock();
 217                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 218                                                   flags);
 219                 rcu_read_unlock();
 220         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 221                 /* rcu_read_lock() is not needed */
 222                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 223                                                          flags);
 224         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 225                    map->map_type == BPF_MAP_TYPE_STACK ||
 226                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 227                 err = map->ops->map_push_elem(map, value, flags);
 228         } else {
 229                 rcu_read_lock();
 230                 err = map->ops->map_update_elem(map, key, value, flags);
 231                 rcu_read_unlock();
 232         }
 233         bpf_enable_instrumentation();
 234         maybe_wait_bpf_programs(map);
 235
 236         return err;
 237 }
 238
 239 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 240                               __u64 flags)
 241 {
 242         void *ptr;
 243         int err;
 244
 245         if (bpf_map_is_offloaded(map))
 246                 return bpf_map_offload_lookup_elem(map, key, value);
 247
 248         bpf_disable_instrumentation();
 249         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 250             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 251                 err = bpf_percpu_hash_copy(map, key, value);
 252         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 253                 err = bpf_percpu_array_copy(map, key, value);
 254         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 255                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 256         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 257                 err = bpf_stackmap_copy(map, key, value);
 258         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 259                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 260         } else if (IS_FD_HASH(map)) {
 261                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 262         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 263                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 264         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 265                    map->map_type == BPF_MAP_TYPE_STACK ||
 266                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 267                 err = map->ops->map_peek_elem(map, value);
 268         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 269                 /* struct_ops map requires directly updating "value" */
 270                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 271         } else {
 272                 rcu_read_lock();
 273                 if (map->ops->map_lookup_elem_sys_only)
 274                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 275                 else
 276                         ptr = map->ops->map_lookup_elem(map, key);
 277                 if (IS_ERR(ptr)) {
 278                         err = PTR_ERR(ptr);
 279                 } else if (!ptr) {
 280                         err = -ENOENT;
 281                 } else {
 282                         err = 0;
 283                         if (flags & BPF_F_LOCK)
 284                                 /* lock 'ptr' and copy everything but lock */
 285                                 copy_map_value_locked(map, value, ptr, true);
 286                         else
 287                                 copy_map_value(map, value, ptr);
 288                         /* mask lock and timer, since value wasn't zero inited */
 289                         check_and_init_map_value(map, value);
 290                 }
 291                 rcu_read_unlock();
 292         }
 293
 294         bpf_enable_instrumentation();
 295         maybe_wait_bpf_programs(map);
 296
 297         return err;
 298 }
 299
 300 /* Please, do not use this function outside from the map creation path
 301  * (e.g. in map update path) without taking care of setting the active
 302  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 303  */
 304 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 305 {
 306         /* We really just want to fail instead of triggering OOM killer
 307          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 308          * which is used for lower order allocation requests.
 309          *
 310          * It has been observed that higher order allocation requests done by
 311          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 312          * to reclaim memory from the page cache, thus we set
 313          * __GFP_RETRY_MAYFAIL to avoid such situations.
 314          */
 315
 316         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 317         unsigned int flags = 0;
 318         unsigned long align = 1;
 319         void *area;
 320
 321         if (size >= SIZE_MAX)
 322                 return NULL;
 323
 324         /* kmalloc()'ed memory can't be mmap()'ed */
 325         if (mmapable) {
 326                 BUG_ON(!PAGE_ALIGNED(size));
 327                 align = SHMLBA;
 328                 flags = VM_USERMAP;
 329         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 330                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 331                                     numa_node);
 332                 if (area != NULL)
 333                         return area;
 334         }
 335
 336         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 337                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 338                         flags, numa_node, __builtin_return_address(0));
 339 }
 340
 341 void *bpf_map_area_alloc(u64 size, int numa_node)
 342 {
 343         return __bpf_map_area_alloc(size, numa_node, false);
 344 }
 345
 346 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 347 {
 348         return __bpf_map_area_alloc(size, numa_node, true);
 349 }
 350
 351 void bpf_map_area_free(void *area)
 352 {
 353         kvfree(area);
 354 }
 355
 356 static u32 bpf_map_flags_retain_permanent(u32 flags)
 357 {
 358         /* Some map creation flags are not tied to the map object but
 359          * rather to the map fd instead, so they have no meaning upon
 360          * map object inspection since multiple file descriptors with
 361          * different (access) properties can exist here. Thus, given
 362          * this has zero meaning for the map itself, lets clear these
 363          * from here.
 364          */
 365         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 366 }
 367
 368 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 369 {
 370         map->map_type = attr->map_type;
 371         map->key_size = attr->key_size;
 372         map->value_size = attr->value_size;
 373         map->max_entries = attr->max_entries;
 374         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 375         map->numa_node = bpf_map_attr_numa_node(attr);
 376         map->map_extra = attr->map_extra;
 377 }
 378
 379 static int bpf_map_alloc_id(struct bpf_map *map)
 380 {
 381         int id;
 382
 383         idr_preload(GFP_KERNEL);
 384         spin_lock_bh(&map_idr_lock);
 385         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 386         if (id > 0)
 387                 map->id = id;
 388         spin_unlock_bh(&map_idr_lock);
 389         idr_preload_end();
 390
 391         if (WARN_ON_ONCE(!id))
 392                 return -ENOSPC;
 393
 394         return id > 0 ? 0 : id;
 395 }
 396
 397 void bpf_map_free_id(struct bpf_map *map)
 398 {
 399         unsigned long flags;
 400
 401         /* Offloaded maps are removed from the IDR store when their device
 402          * disappears - even if someone holds an fd to them they are unusable,
 403          * the memory is gone, all ops will fail; they are simply waiting for
 404          * refcnt to drop to be freed.
 405          */
 406         if (!map->id)
 407                 return;
 408
 409         spin_lock_irqsave(&map_idr_lock, flags);
 410
 411         idr_remove(&map_idr, map->id);
 412         map->id = 0;
 413
 414         spin_unlock_irqrestore(&map_idr_lock, flags);
 415 }
 416
 417 #ifdef CONFIG_MEMCG_KMEM
 418 static void bpf_map_save_memcg(struct bpf_map *map)
 419 {
 420         /* Currently if a map is created by a process belonging to the root
 421          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 422          * So we have to check map->objcg for being NULL each time it's
 423          * being used.
 424          */
 425         if (memcg_bpf_enabled())
 426                 map->objcg = get_obj_cgroup_from_current();
 427 }
 428
 429 static void bpf_map_release_memcg(struct bpf_map *map)
 430 {
 431         if (map->objcg)
 432                 obj_cgroup_put(map->objcg);
 433 }
 434
 435 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 436 {
 437         if (map->objcg)
 438                 return get_mem_cgroup_from_objcg(map->objcg);
 439
 440         return root_mem_cgroup;
 441 }
 442
 443 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 444                            int node)
 445 {
 446         struct mem_cgroup *memcg, *old_memcg;
 447         void *ptr;
 448
 449         memcg = bpf_map_get_memcg(map);
 450         old_memcg = set_active_memcg(memcg);
 451         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 452         set_active_memcg(old_memcg);
 453         mem_cgroup_put(memcg);
 454
 455         return ptr;
 456 }
 457
 458 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 459 {
 460         struct mem_cgroup *memcg, *old_memcg;
 461         void *ptr;
 462
 463         memcg = bpf_map_get_memcg(map);
 464         old_memcg = set_active_memcg(memcg);
 465         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 466         set_active_memcg(old_memcg);
 467         mem_cgroup_put(memcg);
 468
 469         return ptr;
 470 }
 471
 472 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 473                        gfp_t flags)
 474 {
 475         struct mem_cgroup *memcg, *old_memcg;
 476         void *ptr;
 477
 478         memcg = bpf_map_get_memcg(map);
 479         old_memcg = set_active_memcg(memcg);
 480         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 481         set_active_memcg(old_memcg);
 482         mem_cgroup_put(memcg);
 483
 484         return ptr;
 485 }
 486
 487 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 488                                     size_t align, gfp_t flags)
 489 {
 490         struct mem_cgroup *memcg, *old_memcg;
 491         void __percpu *ptr;
 492
 493         memcg = bpf_map_get_memcg(map);
 494         old_memcg = set_active_memcg(memcg);
 495         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 496         set_active_memcg(old_memcg);
 497         mem_cgroup_put(memcg);
 498
 499         return ptr;
 500 }
 501
 502 #else
 503 static void bpf_map_save_memcg(struct bpf_map *map)
 504 {
 505 }
 506
 507 static void bpf_map_release_memcg(struct bpf_map *map)
 508 {
 509 }
 510 #endif
 511
 512 static int btf_field_cmp(const void *a, const void *b)
 513 {
 514         const struct btf_field *f1 = a, *f2 = b;
 515
 516         if (f1->offset < f2->offset)
 517                 return -1;
 518         else if (f1->offset > f2->offset)
 519                 return 1;
 520         return 0;
 521 }
 522
 523 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 524                                   u32 field_mask)
 525 {
 526         struct btf_field *field;
 527
 528         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 529                 return NULL;
 530         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 531         if (!field || !(field->type & field_mask))
 532                 return NULL;
 533         return field;
 534 }
 535
 536 void btf_record_free(struct btf_record *rec)
 537 {
 538         int i;
 539
 540         if (IS_ERR_OR_NULL(rec))
 541                 return;
 542         for (i = 0; i < rec->cnt; i++) {
 543                 switch (rec->fields[i].type) {
 544                 case BPF_KPTR_UNREF:
 545                 case BPF_KPTR_REF:
 546                         if (rec->fields[i].kptr.module)
 547                                 module_put(rec->fields[i].kptr.module);
 548                         btf_put(rec->fields[i].kptr.btf);
 549                         break;
 550                 case BPF_LIST_HEAD:
 551                 case BPF_LIST_NODE:
 552                 case BPF_RB_ROOT:
 553                 case BPF_RB_NODE:
 554                 case BPF_SPIN_LOCK:
 555                 case BPF_TIMER:
 556                 case BPF_REFCOUNT:
 557                         /* Nothing to release */
 558                         break;
 559                 default:
 560                         WARN_ON_ONCE(1);
 561                         continue;
 562                 }
 563         }
 564         kfree(rec);
 565 }
 566
 567 void bpf_map_free_record(struct bpf_map *map)
 568 {
 569         btf_record_free(map->record);
 570         map->record = NULL;
 571 }
 572
 573 struct btf_record *btf_record_dup(const struct btf_record *rec)
 574 {
 575         const struct btf_field *fields;
 576         struct btf_record *new_rec;
 577         int ret, size, i;
 578
 579         if (IS_ERR_OR_NULL(rec))
 580                 return NULL;
 581         size = offsetof(struct btf_record, fields[rec->cnt]);
 582         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 583         if (!new_rec)
 584                 return ERR_PTR(-ENOMEM);
 585         /* Do a deep copy of the btf_record */
 586         fields = rec->fields;
 587         new_rec->cnt = 0;
 588         for (i = 0; i < rec->cnt; i++) {
 589                 switch (fields[i].type) {
 590                 case BPF_KPTR_UNREF:
 591                 case BPF_KPTR_REF:
 592                         btf_get(fields[i].kptr.btf);
 593                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 594                                 ret = -ENXIO;
 595                                 goto free;
 596                         }
 597                         break;
 598                 case BPF_LIST_HEAD:
 599                 case BPF_LIST_NODE:
 600                 case BPF_RB_ROOT:
 601                 case BPF_RB_NODE:
 602                 case BPF_SPIN_LOCK:
 603                 case BPF_TIMER:
 604                 case BPF_REFCOUNT:
 605                         /* Nothing to acquire */
 606                         break;
 607                 default:
 608                         ret = -EFAULT;
 609                         WARN_ON_ONCE(1);
 610                         goto free;
 611                 }
 612                 new_rec->cnt++;
 613         }
 614         return new_rec;
 615 free:
 616         btf_record_free(new_rec);
 617         return ERR_PTR(ret);
 618 }
 619
 620 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 621 {
 622         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 623         int size;
 624
 625         if (!a_has_fields && !b_has_fields)
 626                 return true;
 627         if (a_has_fields != b_has_fields)
 628                 return false;
 629         if (rec_a->cnt != rec_b->cnt)
 630                 return false;
 631         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 632         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 633          * members are zeroed out. So memcmp is safe to do without worrying
 634          * about padding/unused fields.
 635          *
 636          * While spin_lock, timer, and kptr have no relation to map BTF,
 637          * list_head metadata is specific to map BTF, the btf and value_rec
 638          * members in particular. btf is the map BTF, while value_rec points to
 639          * btf_record in that map BTF.
 640          *
 641          * So while by default, we don't rely on the map BTF (which the records
 642          * were parsed from) matching for both records, which is not backwards
 643          * compatible, in case list_head is part of it, we implicitly rely on
 644          * that by way of depending on memcmp succeeding for it.
 645          */
 646         return !memcmp(rec_a, rec_b, size);
 647 }
 648
 649 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 650 {
 651         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 652                 return;
 653         bpf_timer_cancel_and_free(obj + rec->timer_off);
 654 }
 655
 656 extern void __bpf_obj_drop_impl(void *p, const struct btf_record *rec);
 657
 658 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 659 {
 660         const struct btf_field *fields;
 661         int i;
 662
 663         if (IS_ERR_OR_NULL(rec))
 664                 return;
 665         fields = rec->fields;
 666         for (i = 0; i < rec->cnt; i++) {
 667                 struct btf_struct_meta *pointee_struct_meta;
 668                 const struct btf_field *field = &fields[i];
 669                 void *field_ptr = obj + field->offset;
 670                 void *xchgd_field;
 671
 672                 switch (fields[i].type) {
 673                 case BPF_SPIN_LOCK:
 674                         break;
 675                 case BPF_TIMER:
 676                         bpf_timer_cancel_and_free(field_ptr);
 677                         break;
 678                 case BPF_KPTR_UNREF:
 679                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 680                         break;
 681                 case BPF_KPTR_REF:
 682                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 683                         if (!xchgd_field)
 684                                 break;
 685
 686                         if (!btf_is_kernel(field->kptr.btf)) {
 687                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 688                                                                            field->kptr.btf_id);
 689                                 WARN_ON_ONCE(!pointee_struct_meta);
 690                                 migrate_disable();
 691                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 692                                                                  pointee_struct_meta->record :
 693                                                                  NULL);
 694                                 migrate_enable();
 695                         } else {
 696                                 field->kptr.dtor(xchgd_field);
 697                         }
 698                         break;
 699                 case BPF_LIST_HEAD:
 700                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 701                                 continue;
 702                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 703                         break;
 704                 case BPF_RB_ROOT:
 705                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 706                                 continue;
 707                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 708                         break;
 709                 case BPF_LIST_NODE:
 710                 case BPF_RB_NODE:
 711                 case BPF_REFCOUNT:
 712                         break;
 713                 default:
 714                         WARN_ON_ONCE(1);
 715                         continue;
 716                 }
 717         }
 718 }
 719
 720 /* called from workqueue */
 721 static void bpf_map_free_deferred(struct work_struct *work)
 722 {
 723         struct bpf_map *map = container_of(work, struct bpf_map, work);
 724         struct btf_record *rec = map->record;
 725
 726         security_bpf_map_free(map);
 727         bpf_map_release_memcg(map);
 728         /* implementation dependent freeing */
 729         map->ops->map_free(map);
 730         /* Delay freeing of btf_record for maps, as map_free
 731          * callback usually needs access to them. It is better to do it here
 732          * than require each callback to do the free itself manually.
 733          *
 734          * Note that the btf_record stashed in map->inner_map_meta->record was
 735          * already freed using the map_free callback for map in map case which
 736          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 737          * template bpf_map struct used during verification.
 738          */
 739         btf_record_free(rec);
 740 }
 741
 742 static void bpf_map_put_uref(struct bpf_map *map)
 743 {
 744         if (atomic64_dec_and_test(&map->usercnt)) {
 745                 if (map->ops->map_release_uref)
 746                         map->ops->map_release_uref(map);
 747         }
 748 }
 749
 750 /* decrement map refcnt and schedule it for freeing via workqueue
 751  * (underlying map implementation ops->map_free() might sleep)
 752  */
 753 void bpf_map_put(struct bpf_map *map)
 754 {
 755         if (atomic64_dec_and_test(&map->refcnt)) {
 756                 /* bpf_map_free_id() must be called first */
 757                 bpf_map_free_id(map);
 758                 btf_put(map->btf);
 759                 INIT_WORK(&map->work, bpf_map_free_deferred);
 760                 /* Avoid spawning kworkers, since they all might contend
 761                  * for the same mutex like slab_mutex.
 762                  */
 763                 queue_work(system_unbound_wq, &map->work);
 764         }
 765 }
 766 EXPORT_SYMBOL_GPL(bpf_map_put);
 767
 768 void bpf_map_put_with_uref(struct bpf_map *map)
 769 {
 770         bpf_map_put_uref(map);
 771         bpf_map_put(map);
 772 }
 773
 774 static int bpf_map_release(struct inode *inode, struct file *filp)
 775 {
 776         struct bpf_map *map = filp->private_data;
 777
 778         if (map->ops->map_release)
 779                 map->ops->map_release(map, filp);
 780
 781         bpf_map_put_with_uref(map);
 782         return 0;
 783 }
 784
 785 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 786 {
 787         fmode_t mode = f.file->f_mode;
 788
 789         /* Our file permissions may have been overridden by global
 790          * map permissions facing syscall side.
 791          */
 792         if (READ_ONCE(map->frozen))
 793                 mode &= ~FMODE_CAN_WRITE;
 794         return mode;
 795 }
 796
 797 #ifdef CONFIG_PROC_FS
 798 /* Show the memory usage of a bpf map */
 799 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 800 {
 801         return map->ops->map_mem_usage(map);
 802 }
 803
 804 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 805 {
 806         struct bpf_map *map = filp->private_data;
 807         u32 type = 0, jited = 0;
 808
 809         if (map_type_contains_progs(map)) {
 810                 spin_lock(&map->owner.lock);
 811                 type  = map->owner.type;
 812                 jited = map->owner.jited;
 813                 spin_unlock(&map->owner.lock);
 814         }
 815
 816         seq_printf(m,
 817                    "map_type:\t%u\n"
 818                    "key_size:\t%u\n"
 819                    "value_size:\t%u\n"
 820                    "max_entries:\t%u\n"
 821                    "map_flags:\t%#x\n"
 822                    "map_extra:\t%#llx\n"
 823                    "memlock:\t%llu\n"
 824                    "map_id:\t%u\n"
 825                    "frozen:\t%u\n",
 826                    map->map_type,
 827                    map->key_size,
 828                    map->value_size,
 829                    map->max_entries,
 830                    map->map_flags,
 831                    (unsigned long long)map->map_extra,
 832                    bpf_map_memory_usage(map),
 833                    map->id,
 834                    READ_ONCE(map->frozen));
 835         if (type) {
 836                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 837                 seq_printf(m, "owner_jited:\t%u\n", jited);
 838         }
 839 }
 840 #endif
 841
 842 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 843                               loff_t *ppos)
 844 {
 845         /* We need this handler such that alloc_file() enables
 846          * f_mode with FMODE_CAN_READ.
 847          */
 848         return -EINVAL;
 849 }
 850
 851 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 852                                size_t siz, loff_t *ppos)
 853 {
 854         /* We need this handler such that alloc_file() enables
 855          * f_mode with FMODE_CAN_WRITE.
 856          */
 857         return -EINVAL;
 858 }
 859
 860 /* called for any extra memory-mapped regions (except initial) */
 861 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 862 {
 863         struct bpf_map *map = vma->vm_file->private_data;
 864
 865         if (vma->vm_flags & VM_MAYWRITE)
 866                 bpf_map_write_active_inc(map);
 867 }
 868
 869 /* called for all unmapped memory region (including initial) */
 870 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 871 {
 872         struct bpf_map *map = vma->vm_file->private_data;
 873
 874         if (vma->vm_flags & VM_MAYWRITE)
 875                 bpf_map_write_active_dec(map);
 876 }
 877
 878 static const struct vm_operations_struct bpf_map_default_vmops = {
 879         .open           = bpf_map_mmap_open,
 880         .close          = bpf_map_mmap_close,
 881 };
 882
 883 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 884 {
 885         struct bpf_map *map = filp->private_data;
 886         int err;
 887
 888         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 889                 return -ENOTSUPP;
 890
 891         if (!(vma->vm_flags & VM_SHARED))
 892                 return -EINVAL;
 893
 894         mutex_lock(&map->freeze_mutex);
 895
 896         if (vma->vm_flags & VM_WRITE) {
 897                 if (map->frozen) {
 898                         err = -EPERM;
 899                         goto out;
 900                 }
 901                 /* map is meant to be read-only, so do not allow mapping as
 902                  * writable, because it's possible to leak a writable page
 903                  * reference and allows user-space to still modify it after
 904                  * freezing, while verifier will assume contents do not change
 905                  */
 906                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 907                         err = -EACCES;
 908                         goto out;
 909                 }
 910         }
 911
 912         /* set default open/close callbacks */
 913         vma->vm_ops = &bpf_map_default_vmops;
 914         vma->vm_private_data = map;
 915         vm_flags_clear(vma, VM_MAYEXEC);
 916         if (!(vma->vm_flags & VM_WRITE))
 917                 /* disallow re-mapping with PROT_WRITE */
 918                 vm_flags_clear(vma, VM_MAYWRITE);
 919
 920         err = map->ops->map_mmap(map, vma);
 921         if (err)
 922                 goto out;
 923
 924         if (vma->vm_flags & VM_MAYWRITE)
 925                 bpf_map_write_active_inc(map);
 926 out:
 927         mutex_unlock(&map->freeze_mutex);
 928         return err;
 929 }
 930
 931 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 932 {
 933         struct bpf_map *map = filp->private_data;
 934
 935         if (map->ops->map_poll)
 936                 return map->ops->map_poll(map, filp, pts);
 937
 938         return EPOLLERR;
 939 }
 940
 941 const struct file_operations bpf_map_fops = {
 942 #ifdef CONFIG_PROC_FS
 943         .show_fdinfo    = bpf_map_show_fdinfo,
 944 #endif
 945         .release        = bpf_map_release,
 946         .read           = bpf_dummy_read,
 947         .write          = bpf_dummy_write,
 948         .mmap           = bpf_map_mmap,
 949         .poll           = bpf_map_poll,
 950 };
 951
 952 int bpf_map_new_fd(struct bpf_map *map, int flags)
 953 {
 954         int ret;
 955
 956         ret = security_bpf_map(map, OPEN_FMODE(flags));
 957         if (ret < 0)
 958                 return ret;
 959
 960         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 961                                 flags | O_CLOEXEC);
 962 }
 963
 964 int bpf_get_file_flag(int flags)
 965 {
 966         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 967                 return -EINVAL;
 968         if (flags & BPF_F_RDONLY)
 969                 return O_RDONLY;
 970         if (flags & BPF_F_WRONLY)
 971                 return O_WRONLY;
 972         return O_RDWR;
 973 }
 974
 975 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 976 #define CHECK_ATTR(CMD) \
 977         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 978                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 979                    sizeof(*attr) - \
 980                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 981                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 982
 983 /* dst and src must have at least "size" number of bytes.
 984  * Return strlen on success and < 0 on error.
 985  */
 986 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 987 {
 988         const char *end = src + size;
 989         const char *orig_src = src;
 990
 991         memset(dst, 0, size);
 992         /* Copy all isalnum(), '_' and '.' chars. */
 993         while (src < end && *src) {
 994                 if (!isalnum(*src) &&
 995                     *src != '_' && *src != '.')
 996                         return -EINVAL;
 997                 *dst++ = *src++;
 998         }
 999
1000         /* No '\0' found in "size" number of bytes */
1001         if (src == end)
1002                 return -EINVAL;
1003
1004         return src - orig_src;
1005 }
1006
1007 int map_check_no_btf(const struct bpf_map *map,
1008                      const struct btf *btf,
1009                      const struct btf_type *key_type,
1010                      const struct btf_type *value_type)
1011 {
1012         return -ENOTSUPP;
1013 }
1014
1015 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
1016                          u32 btf_key_id, u32 btf_value_id)
1017 {
1018         const struct btf_type *key_type, *value_type;
1019         u32 key_size, value_size;
1020         int ret = 0;
1021
1022         /* Some maps allow key to be unspecified. */
1023         if (btf_key_id) {
1024                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1025                 if (!key_type || key_size != map->key_size)
1026                         return -EINVAL;
1027         } else {
1028                 key_type = btf_type_by_id(btf, 0);
1029                 if (!map->ops->map_check_btf)
1030                         return -EINVAL;
1031         }
1032
1033         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1034         if (!value_type || value_size != map->value_size)
1035                 return -EINVAL;
1036
1037         map->record = btf_parse_fields(btf, value_type,
1038                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1039                                        BPF_RB_ROOT | BPF_REFCOUNT,
1040                                        map->value_size);
1041         if (!IS_ERR_OR_NULL(map->record)) {
1042                 int i;
1043
1044                 if (!bpf_capable()) {
1045                         ret = -EPERM;
1046                         goto free_map_tab;
1047                 }
1048                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1049                         ret = -EACCES;
1050                         goto free_map_tab;
1051                 }
1052                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1053                         switch (map->record->field_mask & (1 << i)) {
1054                         case 0:
1055                                 continue;
1056                         case BPF_SPIN_LOCK:
1057                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1058                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1059                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1060                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1061                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1062                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1063                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1064                                         ret = -EOPNOTSUPP;
1065                                         goto free_map_tab;
1066                                 }
1067                                 break;
1068                         case BPF_TIMER:
1069                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1070                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1071                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1072                                         ret = -EOPNOTSUPP;
1073                                         goto free_map_tab;
1074                                 }
1075                                 break;
1076                         case BPF_KPTR_UNREF:
1077                         case BPF_KPTR_REF:
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         int numa_node = bpf_map_attr_numa_node(attr);
1131         struct bpf_map *map;
1132         int f_flags;
1133         int err;
1134
1135         err = CHECK_ATTR(BPF_MAP_CREATE);
1136         if (err)
1137                 return -EINVAL;
1138
1139         if (attr->btf_vmlinux_value_type_id) {
1140                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1141                     attr->btf_key_type_id || attr->btf_value_type_id)
1142                         return -EINVAL;
1143         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1144                 return -EINVAL;
1145         }
1146
1147         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1148             attr->map_extra != 0)
1149                 return -EINVAL;
1150
1151         f_flags = bpf_get_file_flag(attr->map_flags);
1152         if (f_flags < 0)
1153                 return f_flags;
1154
1155         if (numa_node != NUMA_NO_NODE &&
1156             ((unsigned int)numa_node >= nr_node_ids ||
1157              !node_online(numa_node)))
1158                 return -EINVAL;
1159
1160         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1161         map = find_and_alloc_map(attr);
1162         if (IS_ERR(map))
1163                 return PTR_ERR(map);
1164
1165         err = bpf_obj_name_cpy(map->name, attr->map_name,
1166                                sizeof(attr->map_name));
1167         if (err < 0)
1168                 goto free_map;
1169
1170         atomic64_set(&map->refcnt, 1);
1171         atomic64_set(&map->usercnt, 1);
1172         mutex_init(&map->freeze_mutex);
1173         spin_lock_init(&map->owner.lock);
1174
1175         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1176             /* Even the map's value is a kernel's struct,
1177              * the bpf_prog.o must have BTF to begin with
1178              * to figure out the corresponding kernel's
1179              * counter part.  Thus, attr->btf_fd has
1180              * to be valid also.
1181              */
1182             attr->btf_vmlinux_value_type_id) {
1183                 struct btf *btf;
1184
1185                 btf = btf_get_by_fd(attr->btf_fd);
1186                 if (IS_ERR(btf)) {
1187                         err = PTR_ERR(btf);
1188                         goto free_map;
1189                 }
1190                 if (btf_is_kernel(btf)) {
1191                         btf_put(btf);
1192                         err = -EACCES;
1193                         goto free_map;
1194                 }
1195                 map->btf = btf;
1196
1197                 if (attr->btf_value_type_id) {
1198                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1199                                             attr->btf_value_type_id);
1200                         if (err)
1201                                 goto free_map;
1202                 }
1203
1204                 map->btf_key_type_id = attr->btf_key_type_id;
1205                 map->btf_value_type_id = attr->btf_value_type_id;
1206                 map->btf_vmlinux_value_type_id =
1207                         attr->btf_vmlinux_value_type_id;
1208         }
1209
1210         err = security_bpf_map_alloc(map);
1211         if (err)
1212                 goto free_map;
1213
1214         err = bpf_map_alloc_id(map);
1215         if (err)
1216                 goto free_map_sec;
1217
1218         bpf_map_save_memcg(map);
1219
1220         err = bpf_map_new_fd(map, f_flags);
1221         if (err < 0) {
1222                 /* failed to allocate fd.
1223                  * bpf_map_put_with_uref() is needed because the above
1224                  * bpf_map_alloc_id() has published the map
1225                  * to the userspace and the userspace may
1226                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1227                  */
1228                 bpf_map_put_with_uref(map);
1229                 return err;
1230         }
1231
1232         return err;
1233
1234 free_map_sec:
1235         security_bpf_map_free(map);
1236 free_map:
1237         btf_put(map->btf);
1238         map->ops->map_free(map);
1239         return err;
1240 }
1241
1242 /* if error is returned, fd is released.
1243  * On success caller should complete fd access with matching fdput()
1244  */
1245 struct bpf_map *__bpf_map_get(struct fd f)
1246 {
1247         if (!f.file)
1248                 return ERR_PTR(-EBADF);
1249         if (f.file->f_op != &bpf_map_fops) {
1250                 fdput(f);
1251                 return ERR_PTR(-EINVAL);
1252         }
1253
1254         return f.file->private_data;
1255 }
1256
1257 void bpf_map_inc(struct bpf_map *map)
1258 {
1259         atomic64_inc(&map->refcnt);
1260 }
1261 EXPORT_SYMBOL_GPL(bpf_map_inc);
1262
1263 void bpf_map_inc_with_uref(struct bpf_map *map)
1264 {
1265         atomic64_inc(&map->refcnt);
1266         atomic64_inc(&map->usercnt);
1267 }
1268 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1269
1270 struct bpf_map *bpf_map_get(u32 ufd)
1271 {
1272         struct fd f = fdget(ufd);
1273         struct bpf_map *map;
1274
1275         map = __bpf_map_get(f);
1276         if (IS_ERR(map))
1277                 return map;
1278
1279         bpf_map_inc(map);
1280         fdput(f);
1281
1282         return map;
1283 }
1284 EXPORT_SYMBOL(bpf_map_get);
1285
1286 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1287 {
1288         struct fd f = fdget(ufd);
1289         struct bpf_map *map;
1290
1291         map = __bpf_map_get(f);
1292         if (IS_ERR(map))
1293                 return map;
1294
1295         bpf_map_inc_with_uref(map);
1296         fdput(f);
1297
1298         return map;
1299 }
1300
1301 /* map_idr_lock should have been held or the map should have been
1302  * protected by rcu read lock.
1303  */
1304 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1305 {
1306         int refold;
1307
1308         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1309         if (!refold)
1310                 return ERR_PTR(-ENOENT);
1311         if (uref)
1312                 atomic64_inc(&map->usercnt);
1313
1314         return map;
1315 }
1316
1317 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1318 {
1319         spin_lock_bh(&map_idr_lock);
1320         map = __bpf_map_inc_not_zero(map, false);
1321         spin_unlock_bh(&map_idr_lock);
1322
1323         return map;
1324 }
1325 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1326
1327 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1328 {
1329         return -ENOTSUPP;
1330 }
1331
1332 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1333 {
1334         if (key_size)
1335                 return vmemdup_user(ukey, key_size);
1336
1337         if (ukey)
1338                 return ERR_PTR(-EINVAL);
1339
1340         return NULL;
1341 }
1342
1343 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1344 {
1345         if (key_size)
1346                 return kvmemdup_bpfptr(ukey, key_size);
1347
1348         if (!bpfptr_is_null(ukey))
1349                 return ERR_PTR(-EINVAL);
1350
1351         return NULL;
1352 }
1353
1354 /* last field in 'union bpf_attr' used by this command */
1355 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1356
1357 static int map_lookup_elem(union bpf_attr *attr)
1358 {
1359         void __user *ukey = u64_to_user_ptr(attr->key);
1360         void __user *uvalue = u64_to_user_ptr(attr->value);
1361         int ufd = attr->map_fd;
1362         struct bpf_map *map;
1363         void *key, *value;
1364         u32 value_size;
1365         struct fd f;
1366         int err;
1367
1368         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1369                 return -EINVAL;
1370
1371         if (attr->flags & ~BPF_F_LOCK)
1372                 return -EINVAL;
1373
1374         f = fdget(ufd);
1375         map = __bpf_map_get(f);
1376         if (IS_ERR(map))
1377                 return PTR_ERR(map);
1378         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1379                 err = -EPERM;
1380                 goto err_put;
1381         }
1382
1383         if ((attr->flags & BPF_F_LOCK) &&
1384             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1385                 err = -EINVAL;
1386                 goto err_put;
1387         }
1388
1389         key = __bpf_copy_key(ukey, map->key_size);
1390         if (IS_ERR(key)) {
1391                 err = PTR_ERR(key);
1392                 goto err_put;
1393         }
1394
1395         value_size = bpf_map_value_size(map);
1396
1397         err = -ENOMEM;
1398         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1399         if (!value)
1400                 goto free_key;
1401
1402         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1403                 if (copy_from_user(value, uvalue, value_size))
1404                         err = -EFAULT;
1405                 else
1406                         err = bpf_map_copy_value(map, key, value, attr->flags);
1407                 goto free_value;
1408         }
1409
1410         err = bpf_map_copy_value(map, key, value, attr->flags);
1411         if (err)
1412                 goto free_value;
1413
1414         err = -EFAULT;
1415         if (copy_to_user(uvalue, value, value_size) != 0)
1416                 goto free_value;
1417
1418         err = 0;
1419
1420 free_value:
1421         kvfree(value);
1422 free_key:
1423         kvfree(key);
1424 err_put:
1425         fdput(f);
1426         return err;
1427 }
1428
1429
1430 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1431
1432 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1433 {
1434         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1435         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1436         int ufd = attr->map_fd;
1437         struct bpf_map *map;
1438         void *key, *value;
1439         u32 value_size;
1440         struct fd f;
1441         int err;
1442
1443         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1444                 return -EINVAL;
1445
1446         f = fdget(ufd);
1447         map = __bpf_map_get(f);
1448         if (IS_ERR(map))
1449                 return PTR_ERR(map);
1450         bpf_map_write_active_inc(map);
1451         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1452                 err = -EPERM;
1453                 goto err_put;
1454         }
1455
1456         if ((attr->flags & BPF_F_LOCK) &&
1457             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1458                 err = -EINVAL;
1459                 goto err_put;
1460         }
1461
1462         key = ___bpf_copy_key(ukey, map->key_size);
1463         if (IS_ERR(key)) {
1464                 err = PTR_ERR(key);
1465                 goto err_put;
1466         }
1467
1468         value_size = bpf_map_value_size(map);
1469         value = kvmemdup_bpfptr(uvalue, value_size);
1470         if (IS_ERR(value)) {
1471                 err = PTR_ERR(value);
1472                 goto free_key;
1473         }
1474
1475         err = bpf_map_update_value(map, f.file, key, value, attr->flags);
1476
1477         kvfree(value);
1478 free_key:
1479         kvfree(key);
1480 err_put:
1481         bpf_map_write_active_dec(map);
1482         fdput(f);
1483         return err;
1484 }
1485
1486 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1487
1488 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1489 {
1490         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1491         int ufd = attr->map_fd;
1492         struct bpf_map *map;
1493         struct fd f;
1494         void *key;
1495         int err;
1496
1497         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1498                 return -EINVAL;
1499
1500         f = fdget(ufd);
1501         map = __bpf_map_get(f);
1502         if (IS_ERR(map))
1503                 return PTR_ERR(map);
1504         bpf_map_write_active_inc(map);
1505         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1506                 err = -EPERM;
1507                 goto err_put;
1508         }
1509
1510         key = ___bpf_copy_key(ukey, map->key_size);
1511         if (IS_ERR(key)) {
1512                 err = PTR_ERR(key);
1513                 goto err_put;
1514         }
1515
1516         if (bpf_map_is_offloaded(map)) {
1517                 err = bpf_map_offload_delete_elem(map, key);
1518                 goto out;
1519         } else if (IS_FD_PROG_ARRAY(map) ||
1520                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1521                 /* These maps require sleepable context */
1522                 err = map->ops->map_delete_elem(map, key);
1523                 goto out;
1524         }
1525
1526         bpf_disable_instrumentation();
1527         rcu_read_lock();
1528         err = map->ops->map_delete_elem(map, key);
1529         rcu_read_unlock();
1530         bpf_enable_instrumentation();
1531         maybe_wait_bpf_programs(map);
1532 out:
1533         kvfree(key);
1534 err_put:
1535         bpf_map_write_active_dec(map);
1536         fdput(f);
1537         return err;
1538 }
1539
1540 /* last field in 'union bpf_attr' used by this command */
1541 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1542
1543 static int map_get_next_key(union bpf_attr *attr)
1544 {
1545         void __user *ukey = u64_to_user_ptr(attr->key);
1546         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1547         int ufd = attr->map_fd;
1548         struct bpf_map *map;
1549         void *key, *next_key;
1550         struct fd f;
1551         int err;
1552
1553         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1554                 return -EINVAL;
1555
1556         f = fdget(ufd);
1557         map = __bpf_map_get(f);
1558         if (IS_ERR(map))
1559                 return PTR_ERR(map);
1560         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1561                 err = -EPERM;
1562                 goto err_put;
1563         }
1564
1565         if (ukey) {
1566                 key = __bpf_copy_key(ukey, map->key_size);
1567                 if (IS_ERR(key)) {
1568                         err = PTR_ERR(key);
1569                         goto err_put;
1570                 }
1571         } else {
1572                 key = NULL;
1573         }
1574
1575         err = -ENOMEM;
1576         next_key = kvmalloc(map->key_size, GFP_USER);
1577         if (!next_key)
1578                 goto free_key;
1579
1580         if (bpf_map_is_offloaded(map)) {
1581                 err = bpf_map_offload_get_next_key(map, key, next_key);
1582                 goto out;
1583         }
1584
1585         rcu_read_lock();
1586         err = map->ops->map_get_next_key(map, key, next_key);
1587         rcu_read_unlock();
1588 out:
1589         if (err)
1590                 goto free_next_key;
1591
1592         err = -EFAULT;
1593         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1594                 goto free_next_key;
1595
1596         err = 0;
1597
1598 free_next_key:
1599         kvfree(next_key);
1600 free_key:
1601         kvfree(key);
1602 err_put:
1603         fdput(f);
1604         return err;
1605 }
1606
1607 int generic_map_delete_batch(struct bpf_map *map,
1608                              const union bpf_attr *attr,
1609                              union bpf_attr __user *uattr)
1610 {
1611         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1612         u32 cp, max_count;
1613         int err = 0;
1614         void *key;
1615
1616         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1617                 return -EINVAL;
1618
1619         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1620             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1621                 return -EINVAL;
1622         }
1623
1624         max_count = attr->batch.count;
1625         if (!max_count)
1626                 return 0;
1627
1628         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1629         if (!key)
1630                 return -ENOMEM;
1631
1632         for (cp = 0; cp < max_count; cp++) {
1633                 err = -EFAULT;
1634                 if (copy_from_user(key, keys + cp * map->key_size,
1635                                    map->key_size))
1636                         break;
1637
1638                 if (bpf_map_is_offloaded(map)) {
1639                         err = bpf_map_offload_delete_elem(map, key);
1640                         break;
1641                 }
1642
1643                 bpf_disable_instrumentation();
1644                 rcu_read_lock();
1645                 err = map->ops->map_delete_elem(map, key);
1646                 rcu_read_unlock();
1647                 bpf_enable_instrumentation();
1648                 if (err)
1649                         break;
1650                 cond_resched();
1651         }
1652         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1653                 err = -EFAULT;
1654
1655         kvfree(key);
1656
1657         maybe_wait_bpf_programs(map);
1658         return err;
1659 }
1660
1661 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1662                              const union bpf_attr *attr,
1663                              union bpf_attr __user *uattr)
1664 {
1665         void __user *values = u64_to_user_ptr(attr->batch.values);
1666         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1667         u32 value_size, cp, max_count;
1668         void *key, *value;
1669         int err = 0;
1670
1671         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1672                 return -EINVAL;
1673
1674         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1675             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1676                 return -EINVAL;
1677         }
1678
1679         value_size = bpf_map_value_size(map);
1680
1681         max_count = attr->batch.count;
1682         if (!max_count)
1683                 return 0;
1684
1685         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1686         if (!key)
1687                 return -ENOMEM;
1688
1689         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1690         if (!value) {
1691                 kvfree(key);
1692                 return -ENOMEM;
1693         }
1694
1695         for (cp = 0; cp < max_count; cp++) {
1696                 err = -EFAULT;
1697                 if (copy_from_user(key, keys + cp * map->key_size,
1698                     map->key_size) ||
1699                     copy_from_user(value, values + cp * value_size, value_size))
1700                         break;
1701
1702                 err = bpf_map_update_value(map, map_file, key, value,
1703                                            attr->batch.elem_flags);
1704
1705                 if (err)
1706                         break;
1707                 cond_resched();
1708         }
1709
1710         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1711                 err = -EFAULT;
1712
1713         kvfree(value);
1714         kvfree(key);
1715         return err;
1716 }
1717
1718 #define MAP_LOOKUP_RETRIES 3
1719
1720 int generic_map_lookup_batch(struct bpf_map *map,
1721                                     const union bpf_attr *attr,
1722                                     union bpf_attr __user *uattr)
1723 {
1724         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1725         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1726         void __user *values = u64_to_user_ptr(attr->batch.values);
1727         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1728         void *buf, *buf_prevkey, *prev_key, *key, *value;
1729         int err, retry = MAP_LOOKUP_RETRIES;
1730         u32 value_size, cp, max_count;
1731
1732         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1733                 return -EINVAL;
1734
1735         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1736             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1737                 return -EINVAL;
1738
1739         value_size = bpf_map_value_size(map);
1740
1741         max_count = attr->batch.count;
1742         if (!max_count)
1743                 return 0;
1744
1745         if (put_user(0, &uattr->batch.count))
1746                 return -EFAULT;
1747
1748         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1749         if (!buf_prevkey)
1750                 return -ENOMEM;
1751
1752         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1753         if (!buf) {
1754                 kvfree(buf_prevkey);
1755                 return -ENOMEM;
1756         }
1757
1758         err = -EFAULT;
1759         prev_key = NULL;
1760         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1761                 goto free_buf;
1762         key = buf;
1763         value = key + map->key_size;
1764         if (ubatch)
1765                 prev_key = buf_prevkey;
1766
1767         for (cp = 0; cp < max_count;) {
1768                 rcu_read_lock();
1769                 err = map->ops->map_get_next_key(map, prev_key, key);
1770                 rcu_read_unlock();
1771                 if (err)
1772                         break;
1773                 err = bpf_map_copy_value(map, key, value,
1774                                          attr->batch.elem_flags);
1775
1776                 if (err == -ENOENT) {
1777                         if (retry) {
1778                                 retry--;
1779                                 continue;
1780                         }
1781                         err = -EINTR;
1782                         break;
1783                 }
1784
1785                 if (err)
1786                         goto free_buf;
1787
1788                 if (copy_to_user(keys + cp * map->key_size, key,
1789                                  map->key_size)) {
1790                         err = -EFAULT;
1791                         goto free_buf;
1792                 }
1793                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1794                         err = -EFAULT;
1795                         goto free_buf;
1796                 }
1797
1798                 if (!prev_key)
1799                         prev_key = buf_prevkey;
1800
1801                 swap(prev_key, key);
1802                 retry = MAP_LOOKUP_RETRIES;
1803                 cp++;
1804                 cond_resched();
1805         }
1806
1807         if (err == -EFAULT)
1808                 goto free_buf;
1809
1810         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1811                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1812                 err = -EFAULT;
1813
1814 free_buf:
1815         kvfree(buf_prevkey);
1816         kvfree(buf);
1817         return err;
1818 }
1819
1820 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1821
1822 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1823 {
1824         void __user *ukey = u64_to_user_ptr(attr->key);
1825         void __user *uvalue = u64_to_user_ptr(attr->value);
1826         int ufd = attr->map_fd;
1827         struct bpf_map *map;
1828         void *key, *value;
1829         u32 value_size;
1830         struct fd f;
1831         int err;
1832
1833         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1834                 return -EINVAL;
1835
1836         if (attr->flags & ~BPF_F_LOCK)
1837                 return -EINVAL;
1838
1839         f = fdget(ufd);
1840         map = __bpf_map_get(f);
1841         if (IS_ERR(map))
1842                 return PTR_ERR(map);
1843         bpf_map_write_active_inc(map);
1844         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1845             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1846                 err = -EPERM;
1847                 goto err_put;
1848         }
1849
1850         if (attr->flags &&
1851             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1852              map->map_type == BPF_MAP_TYPE_STACK)) {
1853                 err = -EINVAL;
1854                 goto err_put;
1855         }
1856
1857         if ((attr->flags & BPF_F_LOCK) &&
1858             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1859                 err = -EINVAL;
1860                 goto err_put;
1861         }
1862
1863         key = __bpf_copy_key(ukey, map->key_size);
1864         if (IS_ERR(key)) {
1865                 err = PTR_ERR(key);
1866                 goto err_put;
1867         }
1868
1869         value_size = bpf_map_value_size(map);
1870
1871         err = -ENOMEM;
1872         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1873         if (!value)
1874                 goto free_key;
1875
1876         err = -ENOTSUPP;
1877         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1878             map->map_type == BPF_MAP_TYPE_STACK) {
1879                 err = map->ops->map_pop_elem(map, value);
1880         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1881                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1882                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1883                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1884                 if (!bpf_map_is_offloaded(map)) {
1885                         bpf_disable_instrumentation();
1886                         rcu_read_lock();
1887                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1888                         rcu_read_unlock();
1889                         bpf_enable_instrumentation();
1890                 }
1891         }
1892
1893         if (err)
1894                 goto free_value;
1895
1896         if (copy_to_user(uvalue, value, value_size) != 0) {
1897                 err = -EFAULT;
1898                 goto free_value;
1899         }
1900
1901         err = 0;
1902
1903 free_value:
1904         kvfree(value);
1905 free_key:
1906         kvfree(key);
1907 err_put:
1908         bpf_map_write_active_dec(map);
1909         fdput(f);
1910         return err;
1911 }
1912
1913 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1914
1915 static int map_freeze(const union bpf_attr *attr)
1916 {
1917         int err = 0, ufd = attr->map_fd;
1918         struct bpf_map *map;
1919         struct fd f;
1920
1921         if (CHECK_ATTR(BPF_MAP_FREEZE))
1922                 return -EINVAL;
1923
1924         f = fdget(ufd);
1925         map = __bpf_map_get(f);
1926         if (IS_ERR(map))
1927                 return PTR_ERR(map);
1928
1929         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record)) {
1930                 fdput(f);
1931                 return -ENOTSUPP;
1932         }
1933
1934         mutex_lock(&map->freeze_mutex);
1935         if (bpf_map_write_active(map)) {
1936                 err = -EBUSY;
1937                 goto err_put;
1938         }
1939         if (READ_ONCE(map->frozen)) {
1940                 err = -EBUSY;
1941                 goto err_put;
1942         }
1943         if (!bpf_capable()) {
1944                 err = -EPERM;
1945                 goto err_put;
1946         }
1947
1948         WRITE_ONCE(map->frozen, true);
1949 err_put:
1950         mutex_unlock(&map->freeze_mutex);
1951         fdput(f);
1952         return err;
1953 }
1954
1955 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1956 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1957         [_id] = & _name ## _prog_ops,
1958 #define BPF_MAP_TYPE(_id, _ops)
1959 #define BPF_LINK_TYPE(_id, _name)
1960 #include <linux/bpf_types.h>
1961 #undef BPF_PROG_TYPE
1962 #undef BPF_MAP_TYPE
1963 #undef BPF_LINK_TYPE
1964 };
1965
1966 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1967 {
1968         const struct bpf_prog_ops *ops;
1969
1970         if (type >= ARRAY_SIZE(bpf_prog_types))
1971                 return -EINVAL;
1972         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1973         ops = bpf_prog_types[type];
1974         if (!ops)
1975                 return -EINVAL;
1976
1977         if (!bpf_prog_is_offloaded(prog->aux))
1978                 prog->aux->ops = ops;
1979         else
1980                 prog->aux->ops = &bpf_offload_prog_ops;
1981         prog->type = type;
1982         return 0;
1983 }
1984
1985 enum bpf_audit {
1986         BPF_AUDIT_LOAD,
1987         BPF_AUDIT_UNLOAD,
1988         BPF_AUDIT_MAX,
1989 };
1990
1991 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1992         [BPF_AUDIT_LOAD]   = "LOAD",
1993         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1994 };
1995
1996 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1997 {
1998         struct audit_context *ctx = NULL;
1999         struct audit_buffer *ab;
2000
2001         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2002                 return;
2003         if (audit_enabled == AUDIT_OFF)
2004                 return;
2005         if (!in_irq() && !irqs_disabled())
2006                 ctx = audit_context();
2007         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2008         if (unlikely(!ab))
2009                 return;
2010         audit_log_format(ab, "prog-id=%u op=%s",
2011                          prog->aux->id, bpf_audit_str[op]);
2012         audit_log_end(ab);
2013 }
2014
2015 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2016 {
2017         int id;
2018
2019         idr_preload(GFP_KERNEL);
2020         spin_lock_bh(&prog_idr_lock);
2021         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2022         if (id > 0)
2023                 prog->aux->id = id;
2024         spin_unlock_bh(&prog_idr_lock);
2025         idr_preload_end();
2026
2027         /* id is in [1, INT_MAX) */
2028         if (WARN_ON_ONCE(!id))
2029                 return -ENOSPC;
2030
2031         return id > 0 ? 0 : id;
2032 }
2033
2034 void bpf_prog_free_id(struct bpf_prog *prog)
2035 {
2036         unsigned long flags;
2037
2038         /* cBPF to eBPF migrations are currently not in the idr store.
2039          * Offloaded programs are removed from the store when their device
2040          * disappears - even if someone grabs an fd to them they are unusable,
2041          * simply waiting for refcnt to drop to be freed.
2042          */
2043         if (!prog->aux->id)
2044                 return;
2045
2046         spin_lock_irqsave(&prog_idr_lock, flags);
2047         idr_remove(&prog_idr, prog->aux->id);
2048         prog->aux->id = 0;
2049         spin_unlock_irqrestore(&prog_idr_lock, flags);
2050 }
2051
2052 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2053 {
2054         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2055
2056         kvfree(aux->func_info);
2057         kfree(aux->func_info_aux);
2058         free_uid(aux->user);
2059         security_bpf_prog_free(aux);
2060         bpf_prog_free(aux->prog);
2061 }
2062
2063 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2064 {
2065         bpf_prog_kallsyms_del_all(prog);
2066         btf_put(prog->aux->btf);
2067         module_put(prog->aux->mod);
2068         kvfree(prog->aux->jited_linfo);
2069         kvfree(prog->aux->linfo);
2070         kfree(prog->aux->kfunc_tab);
2071         if (prog->aux->attach_btf)
2072                 btf_put(prog->aux->attach_btf);
2073
2074         if (deferred) {
2075                 if (prog->aux->sleepable)
2076                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2077                 else
2078                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2079         } else {
2080                 __bpf_prog_put_rcu(&prog->aux->rcu);
2081         }
2082 }
2083
2084 static void bpf_prog_put_deferred(struct work_struct *work)
2085 {
2086         struct bpf_prog_aux *aux;
2087         struct bpf_prog *prog;
2088
2089         aux = container_of(work, struct bpf_prog_aux, work);
2090         prog = aux->prog;
2091         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2092         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2093         bpf_prog_free_id(prog);
2094         __bpf_prog_put_noref(prog, true);
2095 }
2096
2097 static void __bpf_prog_put(struct bpf_prog *prog)
2098 {
2099         struct bpf_prog_aux *aux = prog->aux;
2100
2101         if (atomic64_dec_and_test(&aux->refcnt)) {
2102                 if (in_irq() || irqs_disabled()) {
2103                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2104                         schedule_work(&aux->work);
2105                 } else {
2106                         bpf_prog_put_deferred(&aux->work);
2107                 }
2108         }
2109 }
2110
2111 void bpf_prog_put(struct bpf_prog *prog)
2112 {
2113         __bpf_prog_put(prog);
2114 }
2115 EXPORT_SYMBOL_GPL(bpf_prog_put);
2116
2117 static int bpf_prog_release(struct inode *inode, struct file *filp)
2118 {
2119         struct bpf_prog *prog = filp->private_data;
2120
2121         bpf_prog_put(prog);
2122         return 0;
2123 }
2124
2125 struct bpf_prog_kstats {
2126         u64 nsecs;
2127         u64 cnt;
2128         u64 misses;
2129 };
2130
2131 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2132 {
2133         struct bpf_prog_stats *stats;
2134         unsigned int flags;
2135
2136         stats = this_cpu_ptr(prog->stats);
2137         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2138         u64_stats_inc(&stats->misses);
2139         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2140 }
2141
2142 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2143                                struct bpf_prog_kstats *stats)
2144 {
2145         u64 nsecs = 0, cnt = 0, misses = 0;
2146         int cpu;
2147
2148         for_each_possible_cpu(cpu) {
2149                 const struct bpf_prog_stats *st;
2150                 unsigned int start;
2151                 u64 tnsecs, tcnt, tmisses;
2152
2153                 st = per_cpu_ptr(prog->stats, cpu);
2154                 do {
2155                         start = u64_stats_fetch_begin(&st->syncp);
2156                         tnsecs = u64_stats_read(&st->nsecs);
2157                         tcnt = u64_stats_read(&st->cnt);
2158                         tmisses = u64_stats_read(&st->misses);
2159                 } while (u64_stats_fetch_retry(&st->syncp, start));
2160                 nsecs += tnsecs;
2161                 cnt += tcnt;
2162                 misses += tmisses;
2163         }
2164         stats->nsecs = nsecs;
2165         stats->cnt = cnt;
2166         stats->misses = misses;
2167 }
2168
2169 #ifdef CONFIG_PROC_FS
2170 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2171 {
2172         const struct bpf_prog *prog = filp->private_data;
2173         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2174         struct bpf_prog_kstats stats;
2175
2176         bpf_prog_get_stats(prog, &stats);
2177         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2178         seq_printf(m,
2179                    "prog_type:\t%u\n"
2180                    "prog_jited:\t%u\n"
2181                    "prog_tag:\t%s\n"
2182                    "memlock:\t%llu\n"
2183                    "prog_id:\t%u\n"
2184                    "run_time_ns:\t%llu\n"
2185                    "run_cnt:\t%llu\n"
2186                    "recursion_misses:\t%llu\n"
2187                    "verified_insns:\t%u\n",
2188                    prog->type,
2189                    prog->jited,
2190                    prog_tag,
2191                    prog->pages * 1ULL << PAGE_SHIFT,
2192                    prog->aux->id,
2193                    stats.nsecs,
2194                    stats.cnt,
2195                    stats.misses,
2196                    prog->aux->verified_insns);
2197 }
2198 #endif
2199
2200 const struct file_operations bpf_prog_fops = {
2201 #ifdef CONFIG_PROC_FS
2202         .show_fdinfo    = bpf_prog_show_fdinfo,
2203 #endif
2204         .release        = bpf_prog_release,
2205         .read           = bpf_dummy_read,
2206         .write          = bpf_dummy_write,
2207 };
2208
2209 int bpf_prog_new_fd(struct bpf_prog *prog)
2210 {
2211         int ret;
2212
2213         ret = security_bpf_prog(prog);
2214         if (ret < 0)
2215                 return ret;
2216
2217         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2218                                 O_RDWR | O_CLOEXEC);
2219 }
2220
2221 static struct bpf_prog *____bpf_prog_get(struct fd f)
2222 {
2223         if (!f.file)
2224                 return ERR_PTR(-EBADF);
2225         if (f.file->f_op != &bpf_prog_fops) {
2226                 fdput(f);
2227                 return ERR_PTR(-EINVAL);
2228         }
2229
2230         return f.file->private_data;
2231 }
2232
2233 void bpf_prog_add(struct bpf_prog *prog, int i)
2234 {
2235         atomic64_add(i, &prog->aux->refcnt);
2236 }
2237 EXPORT_SYMBOL_GPL(bpf_prog_add);
2238
2239 void bpf_prog_sub(struct bpf_prog *prog, int i)
2240 {
2241         /* Only to be used for undoing previous bpf_prog_add() in some
2242          * error path. We still know that another entity in our call
2243          * path holds a reference to the program, thus atomic_sub() can
2244          * be safely used in such cases!
2245          */
2246         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2247 }
2248 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2249
2250 void bpf_prog_inc(struct bpf_prog *prog)
2251 {
2252         atomic64_inc(&prog->aux->refcnt);
2253 }
2254 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2255
2256 /* prog_idr_lock should have been held */
2257 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2258 {
2259         int refold;
2260
2261         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2262
2263         if (!refold)
2264                 return ERR_PTR(-ENOENT);
2265
2266         return prog;
2267 }
2268 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2269
2270 bool bpf_prog_get_ok(struct bpf_prog *prog,
2271                             enum bpf_prog_type *attach_type, bool attach_drv)
2272 {
2273         /* not an attachment, just a refcount inc, always allow */
2274         if (!attach_type)
2275                 return true;
2276
2277         if (prog->type != *attach_type)
2278                 return false;
2279         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2280                 return false;
2281
2282         return true;
2283 }
2284
2285 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2286                                        bool attach_drv)
2287 {
2288         struct fd f = fdget(ufd);
2289         struct bpf_prog *prog;
2290
2291         prog = ____bpf_prog_get(f);
2292         if (IS_ERR(prog))
2293                 return prog;
2294         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2295                 prog = ERR_PTR(-EINVAL);
2296                 goto out;
2297         }
2298
2299         bpf_prog_inc(prog);
2300 out:
2301         fdput(f);
2302         return prog;
2303 }
2304
2305 struct bpf_prog *bpf_prog_get(u32 ufd)
2306 {
2307         return __bpf_prog_get(ufd, NULL, false);
2308 }
2309
2310 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2311                                        bool attach_drv)
2312 {
2313         return __bpf_prog_get(ufd, &type, attach_drv);
2314 }
2315 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2316
2317 /* Initially all BPF programs could be loaded w/o specifying
2318  * expected_attach_type. Later for some of them specifying expected_attach_type
2319  * at load time became required so that program could be validated properly.
2320  * Programs of types that are allowed to be loaded both w/ and w/o (for
2321  * backward compatibility) expected_attach_type, should have the default attach
2322  * type assigned to expected_attach_type for the latter case, so that it can be
2323  * validated later at attach time.
2324  *
2325  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2326  * prog type requires it but has some attach types that have to be backward
2327  * compatible.
2328  */
2329 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2330 {
2331         switch (attr->prog_type) {
2332         case BPF_PROG_TYPE_CGROUP_SOCK:
2333                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2334                  * exist so checking for non-zero is the way to go here.
2335                  */
2336                 if (!attr->expected_attach_type)
2337                         attr->expected_attach_type =
2338                                 BPF_CGROUP_INET_SOCK_CREATE;
2339                 break;
2340         case BPF_PROG_TYPE_SK_REUSEPORT:
2341                 if (!attr->expected_attach_type)
2342                         attr->expected_attach_type =
2343                                 BPF_SK_REUSEPORT_SELECT;
2344                 break;
2345         }
2346 }
2347
2348 static int
2349 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2350                            enum bpf_attach_type expected_attach_type,
2351                            struct btf *attach_btf, u32 btf_id,
2352                            struct bpf_prog *dst_prog)
2353 {
2354         if (btf_id) {
2355                 if (btf_id > BTF_MAX_TYPE)
2356                         return -EINVAL;
2357
2358                 if (!attach_btf && !dst_prog)
2359                         return -EINVAL;
2360
2361                 switch (prog_type) {
2362                 case BPF_PROG_TYPE_TRACING:
2363                 case BPF_PROG_TYPE_LSM:
2364                 case BPF_PROG_TYPE_STRUCT_OPS:
2365                 case BPF_PROG_TYPE_EXT:
2366                         break;
2367                 default:
2368                         return -EINVAL;
2369                 }
2370         }
2371
2372         if (attach_btf && (!btf_id || dst_prog))
2373                 return -EINVAL;
2374
2375         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2376             prog_type != BPF_PROG_TYPE_EXT)
2377                 return -EINVAL;
2378
2379         switch (prog_type) {
2380         case BPF_PROG_TYPE_CGROUP_SOCK:
2381                 switch (expected_attach_type) {
2382                 case BPF_CGROUP_INET_SOCK_CREATE:
2383                 case BPF_CGROUP_INET_SOCK_RELEASE:
2384                 case BPF_CGROUP_INET4_POST_BIND:
2385                 case BPF_CGROUP_INET6_POST_BIND:
2386                         return 0;
2387                 default:
2388                         return -EINVAL;
2389                 }
2390         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2391                 switch (expected_attach_type) {
2392                 case BPF_CGROUP_INET4_BIND:
2393                 case BPF_CGROUP_INET6_BIND:
2394                 case BPF_CGROUP_INET4_CONNECT:
2395                 case BPF_CGROUP_INET6_CONNECT:
2396                 case BPF_CGROUP_INET4_GETPEERNAME:
2397                 case BPF_CGROUP_INET6_GETPEERNAME:
2398                 case BPF_CGROUP_INET4_GETSOCKNAME:
2399                 case BPF_CGROUP_INET6_GETSOCKNAME:
2400                 case BPF_CGROUP_UDP4_SENDMSG:
2401                 case BPF_CGROUP_UDP6_SENDMSG:
2402                 case BPF_CGROUP_UDP4_RECVMSG:
2403                 case BPF_CGROUP_UDP6_RECVMSG:
2404                         return 0;
2405                 default:
2406                         return -EINVAL;
2407                 }
2408         case BPF_PROG_TYPE_CGROUP_SKB:
2409                 switch (expected_attach_type) {
2410                 case BPF_CGROUP_INET_INGRESS:
2411                 case BPF_CGROUP_INET_EGRESS:
2412                         return 0;
2413                 default:
2414                         return -EINVAL;
2415                 }
2416         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2417                 switch (expected_attach_type) {
2418                 case BPF_CGROUP_SETSOCKOPT:
2419                 case BPF_CGROUP_GETSOCKOPT:
2420                         return 0;
2421                 default:
2422                         return -EINVAL;
2423                 }
2424         case BPF_PROG_TYPE_SK_LOOKUP:
2425                 if (expected_attach_type == BPF_SK_LOOKUP)
2426                         return 0;
2427                 return -EINVAL;
2428         case BPF_PROG_TYPE_SK_REUSEPORT:
2429                 switch (expected_attach_type) {
2430                 case BPF_SK_REUSEPORT_SELECT:
2431                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2432                         return 0;
2433                 default:
2434                         return -EINVAL;
2435                 }
2436         case BPF_PROG_TYPE_SYSCALL:
2437         case BPF_PROG_TYPE_EXT:
2438                 if (expected_attach_type)
2439                         return -EINVAL;
2440                 fallthrough;
2441         default:
2442                 return 0;
2443         }
2444 }
2445
2446 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2447 {
2448         switch (prog_type) {
2449         case BPF_PROG_TYPE_SCHED_CLS:
2450         case BPF_PROG_TYPE_SCHED_ACT:
2451         case BPF_PROG_TYPE_XDP:
2452         case BPF_PROG_TYPE_LWT_IN:
2453         case BPF_PROG_TYPE_LWT_OUT:
2454         case BPF_PROG_TYPE_LWT_XMIT:
2455         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2456         case BPF_PROG_TYPE_SK_SKB:
2457         case BPF_PROG_TYPE_SK_MSG:
2458         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2459         case BPF_PROG_TYPE_CGROUP_DEVICE:
2460         case BPF_PROG_TYPE_CGROUP_SOCK:
2461         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2462         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2463         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2464         case BPF_PROG_TYPE_SOCK_OPS:
2465         case BPF_PROG_TYPE_EXT: /* extends any prog */
2466         case BPF_PROG_TYPE_NETFILTER:
2467                 return true;
2468         case BPF_PROG_TYPE_CGROUP_SKB:
2469                 /* always unpriv */
2470         case BPF_PROG_TYPE_SK_REUSEPORT:
2471                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2472         default:
2473                 return false;
2474         }
2475 }
2476
2477 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2478 {
2479         switch (prog_type) {
2480         case BPF_PROG_TYPE_KPROBE:
2481         case BPF_PROG_TYPE_TRACEPOINT:
2482         case BPF_PROG_TYPE_PERF_EVENT:
2483         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2484         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2485         case BPF_PROG_TYPE_TRACING:
2486         case BPF_PROG_TYPE_LSM:
2487         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2488         case BPF_PROG_TYPE_EXT: /* extends any prog */
2489                 return true;
2490         default:
2491                 return false;
2492         }
2493 }
2494
2495 /* last field in 'union bpf_attr' used by this command */
2496 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2497
2498 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2499 {
2500         enum bpf_prog_type type = attr->prog_type;
2501         struct bpf_prog *prog, *dst_prog = NULL;
2502         struct btf *attach_btf = NULL;
2503         int err;
2504         char license[128];
2505         bool is_gpl;
2506
2507         if (CHECK_ATTR(BPF_PROG_LOAD))
2508                 return -EINVAL;
2509
2510         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2511                                  BPF_F_ANY_ALIGNMENT |
2512                                  BPF_F_TEST_STATE_FREQ |
2513                                  BPF_F_SLEEPABLE |
2514                                  BPF_F_TEST_RND_HI32 |
2515                                  BPF_F_XDP_HAS_FRAGS |
2516                                  BPF_F_XDP_DEV_BOUND_ONLY))
2517                 return -EINVAL;
2518
2519         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2520             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2521             !bpf_capable())
2522                 return -EPERM;
2523
2524         /* copy eBPF program license from user space */
2525         if (strncpy_from_bpfptr(license,
2526                                 make_bpfptr(attr->license, uattr.is_kernel),
2527                                 sizeof(license) - 1) < 0)
2528                 return -EFAULT;
2529         license[sizeof(license) - 1] = 0;
2530
2531         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2532         is_gpl = license_is_gpl_compatible(license);
2533
2534         if (attr->insn_cnt == 0 ||
2535             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2536                 return -E2BIG;
2537         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2538             type != BPF_PROG_TYPE_CGROUP_SKB &&
2539             !bpf_capable())
2540                 return -EPERM;
2541
2542         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2543                 return -EPERM;
2544         if (is_perfmon_prog_type(type) && !perfmon_capable())
2545                 return -EPERM;
2546
2547         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2548          * or btf, we need to check which one it is
2549          */
2550         if (attr->attach_prog_fd) {
2551                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2552                 if (IS_ERR(dst_prog)) {
2553                         dst_prog = NULL;
2554                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2555                         if (IS_ERR(attach_btf))
2556                                 return -EINVAL;
2557                         if (!btf_is_kernel(attach_btf)) {
2558                                 /* attaching through specifying bpf_prog's BTF
2559                                  * objects directly might be supported eventually
2560                                  */
2561                                 btf_put(attach_btf);
2562                                 return -ENOTSUPP;
2563                         }
2564                 }
2565         } else if (attr->attach_btf_id) {
2566                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2567                 attach_btf = bpf_get_btf_vmlinux();
2568                 if (IS_ERR(attach_btf))
2569                         return PTR_ERR(attach_btf);
2570                 if (!attach_btf)
2571                         return -EINVAL;
2572                 btf_get(attach_btf);
2573         }
2574
2575         bpf_prog_load_fixup_attach_type(attr);
2576         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2577                                        attach_btf, attr->attach_btf_id,
2578                                        dst_prog)) {
2579                 if (dst_prog)
2580                         bpf_prog_put(dst_prog);
2581                 if (attach_btf)
2582                         btf_put(attach_btf);
2583                 return -EINVAL;
2584         }
2585
2586         /* plain bpf_prog allocation */
2587         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2588         if (!prog) {
2589                 if (dst_prog)
2590                         bpf_prog_put(dst_prog);
2591                 if (attach_btf)
2592                         btf_put(attach_btf);
2593                 return -ENOMEM;
2594         }
2595
2596         prog->expected_attach_type = attr->expected_attach_type;
2597         prog->aux->attach_btf = attach_btf;
2598         prog->aux->attach_btf_id = attr->attach_btf_id;
2599         prog->aux->dst_prog = dst_prog;
2600         prog->aux->dev_bound = !!attr->prog_ifindex;
2601         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2602         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2603
2604         err = security_bpf_prog_alloc(prog->aux);
2605         if (err)
2606                 goto free_prog;
2607
2608         prog->aux->user = get_current_user();
2609         prog->len = attr->insn_cnt;
2610
2611         err = -EFAULT;
2612         if (copy_from_bpfptr(prog->insns,
2613                              make_bpfptr(attr->insns, uattr.is_kernel),
2614                              bpf_prog_insn_size(prog)) != 0)
2615                 goto free_prog_sec;
2616
2617         prog->orig_prog = NULL;
2618         prog->jited = 0;
2619
2620         atomic64_set(&prog->aux->refcnt, 1);
2621         prog->gpl_compatible = is_gpl ? 1 : 0;
2622
2623         if (bpf_prog_is_dev_bound(prog->aux)) {
2624                 err = bpf_prog_dev_bound_init(prog, attr);
2625                 if (err)
2626                         goto free_prog_sec;
2627         }
2628
2629         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2630             bpf_prog_is_dev_bound(dst_prog->aux)) {
2631                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2632                 if (err)
2633                         goto free_prog_sec;
2634         }
2635
2636         /* find program type: socket_filter vs tracing_filter */
2637         err = find_prog_type(type, prog);
2638         if (err < 0)
2639                 goto free_prog_sec;
2640
2641         prog->aux->load_time = ktime_get_boottime_ns();
2642         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2643                                sizeof(attr->prog_name));
2644         if (err < 0)
2645                 goto free_prog_sec;
2646
2647         /* run eBPF verifier */
2648         err = bpf_check(&prog, attr, uattr, uattr_size);
2649         if (err < 0)
2650                 goto free_used_maps;
2651
2652         prog = bpf_prog_select_runtime(prog, &err);
2653         if (err < 0)
2654                 goto free_used_maps;
2655
2656         err = bpf_prog_alloc_id(prog);
2657         if (err)
2658                 goto free_used_maps;
2659
2660         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2661          * effectively publicly exposed. However, retrieving via
2662          * bpf_prog_get_fd_by_id() will take another reference,
2663          * therefore it cannot be gone underneath us.
2664          *
2665          * Only for the time /after/ successful bpf_prog_new_fd()
2666          * and before returning to userspace, we might just hold
2667          * one reference and any parallel close on that fd could
2668          * rip everything out. Hence, below notifications must
2669          * happen before bpf_prog_new_fd().
2670          *
2671          * Also, any failure handling from this point onwards must
2672          * be using bpf_prog_put() given the program is exposed.
2673          */
2674         bpf_prog_kallsyms_add(prog);
2675         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2676         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2677
2678         err = bpf_prog_new_fd(prog);
2679         if (err < 0)
2680                 bpf_prog_put(prog);
2681         return err;
2682
2683 free_used_maps:
2684         /* In case we have subprogs, we need to wait for a grace
2685          * period before we can tear down JIT memory since symbols
2686          * are already exposed under kallsyms.
2687          */
2688         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2689         return err;
2690 free_prog_sec:
2691         free_uid(prog->aux->user);
2692         security_bpf_prog_free(prog->aux);
2693 free_prog:
2694         if (prog->aux->attach_btf)
2695                 btf_put(prog->aux->attach_btf);
2696         bpf_prog_free(prog);
2697         return err;
2698 }
2699
2700 #define BPF_OBJ_LAST_FIELD file_flags
2701
2702 static int bpf_obj_pin(const union bpf_attr *attr)
2703 {
2704         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2705                 return -EINVAL;
2706
2707         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2708 }
2709
2710 static int bpf_obj_get(const union bpf_attr *attr)
2711 {
2712         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2713             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2714                 return -EINVAL;
2715
2716         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2717                                 attr->file_flags);
2718 }
2719
2720 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2721                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2722 {
2723         atomic64_set(&link->refcnt, 1);
2724         link->type = type;
2725         link->id = 0;
2726         link->ops = ops;
2727         link->prog = prog;
2728 }
2729
2730 static void bpf_link_free_id(int id)
2731 {
2732         if (!id)
2733                 return;
2734
2735         spin_lock_bh(&link_idr_lock);
2736         idr_remove(&link_idr, id);
2737         spin_unlock_bh(&link_idr_lock);
2738 }
2739
2740 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2741  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2742  * anon_inode's release() call. This helper marksbpf_link as
2743  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2744  * is not decremented, it's the responsibility of a calling code that failed
2745  * to complete bpf_link initialization.
2746  */
2747 void bpf_link_cleanup(struct bpf_link_primer *primer)
2748 {
2749         primer->link->prog = NULL;
2750         bpf_link_free_id(primer->id);
2751         fput(primer->file);
2752         put_unused_fd(primer->fd);
2753 }
2754
2755 void bpf_link_inc(struct bpf_link *link)
2756 {
2757         atomic64_inc(&link->refcnt);
2758 }
2759
2760 /* bpf_link_free is guaranteed to be called from process context */
2761 static void bpf_link_free(struct bpf_link *link)
2762 {
2763         bpf_link_free_id(link->id);
2764         if (link->prog) {
2765                 /* detach BPF program, clean up used resources */
2766                 link->ops->release(link);
2767                 bpf_prog_put(link->prog);
2768         }
2769         /* free bpf_link and its containing memory */
2770         link->ops->dealloc(link);
2771 }
2772
2773 static void bpf_link_put_deferred(struct work_struct *work)
2774 {
2775         struct bpf_link *link = container_of(work, struct bpf_link, work);
2776
2777         bpf_link_free(link);
2778 }
2779
2780 /* bpf_link_put can be called from atomic context, but ensures that resources
2781  * are freed from process context
2782  */
2783 void bpf_link_put(struct bpf_link *link)
2784 {
2785         if (!atomic64_dec_and_test(&link->refcnt))
2786                 return;
2787
2788         if (in_atomic()) {
2789                 INIT_WORK(&link->work, bpf_link_put_deferred);
2790                 schedule_work(&link->work);
2791         } else {
2792                 bpf_link_free(link);
2793         }
2794 }
2795 EXPORT_SYMBOL(bpf_link_put);
2796
2797 static int bpf_link_release(struct inode *inode, struct file *filp)
2798 {
2799         struct bpf_link *link = filp->private_data;
2800
2801         bpf_link_put(link);
2802         return 0;
2803 }
2804
2805 #ifdef CONFIG_PROC_FS
2806 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2807 #define BPF_MAP_TYPE(_id, _ops)
2808 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2809 static const char *bpf_link_type_strs[] = {
2810         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2811 #include <linux/bpf_types.h>
2812 };
2813 #undef BPF_PROG_TYPE
2814 #undef BPF_MAP_TYPE
2815 #undef BPF_LINK_TYPE
2816
2817 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2818 {
2819         const struct bpf_link *link = filp->private_data;
2820         const struct bpf_prog *prog = link->prog;
2821         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2822
2823         seq_printf(m,
2824                    "link_type:\t%s\n"
2825                    "link_id:\t%u\n",
2826                    bpf_link_type_strs[link->type],
2827                    link->id);
2828         if (prog) {
2829                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2830                 seq_printf(m,
2831                            "prog_tag:\t%s\n"
2832                            "prog_id:\t%u\n",
2833                            prog_tag,
2834                            prog->aux->id);
2835         }
2836         if (link->ops->show_fdinfo)
2837                 link->ops->show_fdinfo(link, m);
2838 }
2839 #endif
2840
2841 static const struct file_operations bpf_link_fops = {
2842 #ifdef CONFIG_PROC_FS
2843         .show_fdinfo    = bpf_link_show_fdinfo,
2844 #endif
2845         .release        = bpf_link_release,
2846         .read           = bpf_dummy_read,
2847         .write          = bpf_dummy_write,
2848 };
2849
2850 static int bpf_link_alloc_id(struct bpf_link *link)
2851 {
2852         int id;
2853
2854         idr_preload(GFP_KERNEL);
2855         spin_lock_bh(&link_idr_lock);
2856         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2857         spin_unlock_bh(&link_idr_lock);
2858         idr_preload_end();
2859
2860         return id;
2861 }
2862
2863 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2864  * reserving unused FD and allocating ID from link_idr. This is to be paired
2865  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2866  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2867  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2868  * transient state is passed around in struct bpf_link_primer.
2869  * This is preferred way to create and initialize bpf_link, especially when
2870  * there are complicated and expensive operations in between creating bpf_link
2871  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2872  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2873  * expensive (and potentially failing) roll back operations in a rare case
2874  * that file, FD, or ID can't be allocated.
2875  */
2876 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2877 {
2878         struct file *file;
2879         int fd, id;
2880
2881         fd = get_unused_fd_flags(O_CLOEXEC);
2882         if (fd < 0)
2883                 return fd;
2884
2885
2886         id = bpf_link_alloc_id(link);
2887         if (id < 0) {
2888                 put_unused_fd(fd);
2889                 return id;
2890         }
2891
2892         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2893         if (IS_ERR(file)) {
2894                 bpf_link_free_id(id);
2895                 put_unused_fd(fd);
2896                 return PTR_ERR(file);
2897         }
2898
2899         primer->link = link;
2900         primer->file = file;
2901         primer->fd = fd;
2902         primer->id = id;
2903         return 0;
2904 }
2905
2906 int bpf_link_settle(struct bpf_link_primer *primer)
2907 {
2908         /* make bpf_link fetchable by ID */
2909         spin_lock_bh(&link_idr_lock);
2910         primer->link->id = primer->id;
2911         spin_unlock_bh(&link_idr_lock);
2912         /* make bpf_link fetchable by FD */
2913         fd_install(primer->fd, primer->file);
2914         /* pass through installed FD */
2915         return primer->fd;
2916 }
2917
2918 int bpf_link_new_fd(struct bpf_link *link)
2919 {
2920         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2921 }
2922
2923 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2924 {
2925         struct fd f = fdget(ufd);
2926         struct bpf_link *link;
2927
2928         if (!f.file)
2929                 return ERR_PTR(-EBADF);
2930         if (f.file->f_op != &bpf_link_fops) {
2931                 fdput(f);
2932                 return ERR_PTR(-EINVAL);
2933         }
2934
2935         link = f.file->private_data;
2936         bpf_link_inc(link);
2937         fdput(f);
2938
2939         return link;
2940 }
2941 EXPORT_SYMBOL(bpf_link_get_from_fd);
2942
2943 static void bpf_tracing_link_release(struct bpf_link *link)
2944 {
2945         struct bpf_tracing_link *tr_link =
2946                 container_of(link, struct bpf_tracing_link, link.link);
2947
2948         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2949                                                 tr_link->trampoline));
2950
2951         bpf_trampoline_put(tr_link->trampoline);
2952
2953         /* tgt_prog is NULL if target is a kernel function */
2954         if (tr_link->tgt_prog)
2955                 bpf_prog_put(tr_link->tgt_prog);
2956 }
2957
2958 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2959 {
2960         struct bpf_tracing_link *tr_link =
2961                 container_of(link, struct bpf_tracing_link, link.link);
2962
2963         kfree(tr_link);
2964 }
2965
2966 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2967                                          struct seq_file *seq)
2968 {
2969         struct bpf_tracing_link *tr_link =
2970                 container_of(link, struct bpf_tracing_link, link.link);
2971
2972         seq_printf(seq,
2973                    "attach_type:\t%d\n",
2974                    tr_link->attach_type);
2975 }
2976
2977 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2978                                            struct bpf_link_info *info)
2979 {
2980         struct bpf_tracing_link *tr_link =
2981                 container_of(link, struct bpf_tracing_link, link.link);
2982
2983         info->tracing.attach_type = tr_link->attach_type;
2984         bpf_trampoline_unpack_key(tr_link->trampoline->key,
2985                                   &info->tracing.target_obj_id,
2986                                   &info->tracing.target_btf_id);
2987
2988         return 0;
2989 }
2990
2991 static const struct bpf_link_ops bpf_tracing_link_lops = {
2992         .release = bpf_tracing_link_release,
2993         .dealloc = bpf_tracing_link_dealloc,
2994         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2995         .fill_link_info = bpf_tracing_link_fill_link_info,
2996 };
2997
2998 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2999                                    int tgt_prog_fd,
3000                                    u32 btf_id,
3001                                    u64 bpf_cookie)
3002 {
3003         struct bpf_link_primer link_primer;
3004         struct bpf_prog *tgt_prog = NULL;
3005         struct bpf_trampoline *tr = NULL;
3006         struct bpf_tracing_link *link;
3007         u64 key = 0;
3008         int err;
3009
3010         switch (prog->type) {
3011         case BPF_PROG_TYPE_TRACING:
3012                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3013                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3014                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3015                         err = -EINVAL;
3016                         goto out_put_prog;
3017                 }
3018                 break;
3019         case BPF_PROG_TYPE_EXT:
3020                 if (prog->expected_attach_type != 0) {
3021                         err = -EINVAL;
3022                         goto out_put_prog;
3023                 }
3024                 break;
3025         case BPF_PROG_TYPE_LSM:
3026                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3027                         err = -EINVAL;
3028                         goto out_put_prog;
3029                 }
3030                 break;
3031         default:
3032                 err = -EINVAL;
3033                 goto out_put_prog;
3034         }
3035
3036         if (!!tgt_prog_fd != !!btf_id) {
3037                 err = -EINVAL;
3038                 goto out_put_prog;
3039         }
3040
3041         if (tgt_prog_fd) {
3042                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3043                 if (prog->type != BPF_PROG_TYPE_EXT) {
3044                         err = -EINVAL;
3045                         goto out_put_prog;
3046                 }
3047
3048                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3049                 if (IS_ERR(tgt_prog)) {
3050                         err = PTR_ERR(tgt_prog);
3051                         tgt_prog = NULL;
3052                         goto out_put_prog;
3053                 }
3054
3055                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3056         }
3057
3058         link = kzalloc(sizeof(*link), GFP_USER);
3059         if (!link) {
3060                 err = -ENOMEM;
3061                 goto out_put_prog;
3062         }
3063         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3064                       &bpf_tracing_link_lops, prog);
3065         link->attach_type = prog->expected_attach_type;
3066         link->link.cookie = bpf_cookie;
3067
3068         mutex_lock(&prog->aux->dst_mutex);
3069
3070         /* There are a few possible cases here:
3071          *
3072          * - if prog->aux->dst_trampoline is set, the program was just loaded
3073          *   and not yet attached to anything, so we can use the values stored
3074          *   in prog->aux
3075          *
3076          * - if prog->aux->dst_trampoline is NULL, the program has already been
3077          *   attached to a target and its initial target was cleared (below)
3078          *
3079          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3080          *   target_btf_id using the link_create API.
3081          *
3082          * - if tgt_prog == NULL when this function was called using the old
3083          *   raw_tracepoint_open API, and we need a target from prog->aux
3084          *
3085          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3086          *   was detached and is going for re-attachment.
3087          */
3088         if (!prog->aux->dst_trampoline && !tgt_prog) {
3089                 /*
3090                  * Allow re-attach for TRACING and LSM programs. If it's
3091                  * currently linked, bpf_trampoline_link_prog will fail.
3092                  * EXT programs need to specify tgt_prog_fd, so they
3093                  * re-attach in separate code path.
3094                  */
3095                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3096                     prog->type != BPF_PROG_TYPE_LSM) {
3097                         err = -EINVAL;
3098                         goto out_unlock;
3099                 }
3100                 btf_id = prog->aux->attach_btf_id;
3101                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3102         }
3103
3104         if (!prog->aux->dst_trampoline ||
3105             (key && key != prog->aux->dst_trampoline->key)) {
3106                 /* If there is no saved target, or the specified target is
3107                  * different from the destination specified at load time, we
3108                  * need a new trampoline and a check for compatibility
3109                  */
3110                 struct bpf_attach_target_info tgt_info = {};
3111
3112                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3113                                               &tgt_info);
3114                 if (err)
3115                         goto out_unlock;
3116
3117                 if (tgt_info.tgt_mod) {
3118                         module_put(prog->aux->mod);
3119                         prog->aux->mod = tgt_info.tgt_mod;
3120                 }
3121
3122                 tr = bpf_trampoline_get(key, &tgt_info);
3123                 if (!tr) {
3124                         err = -ENOMEM;
3125                         goto out_unlock;
3126                 }
3127         } else {
3128                 /* The caller didn't specify a target, or the target was the
3129                  * same as the destination supplied during program load. This
3130                  * means we can reuse the trampoline and reference from program
3131                  * load time, and there is no need to allocate a new one. This
3132                  * can only happen once for any program, as the saved values in
3133                  * prog->aux are cleared below.
3134                  */
3135                 tr = prog->aux->dst_trampoline;
3136                 tgt_prog = prog->aux->dst_prog;
3137         }
3138
3139         err = bpf_link_prime(&link->link.link, &link_primer);
3140         if (err)
3141                 goto out_unlock;
3142
3143         err = bpf_trampoline_link_prog(&link->link, tr);
3144         if (err) {
3145                 bpf_link_cleanup(&link_primer);
3146                 link = NULL;
3147                 goto out_unlock;
3148         }
3149
3150         link->tgt_prog = tgt_prog;
3151         link->trampoline = tr;
3152
3153         /* Always clear the trampoline and target prog from prog->aux to make
3154          * sure the original attach destination is not kept alive after a
3155          * program is (re-)attached to another target.
3156          */
3157         if (prog->aux->dst_prog &&
3158             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3159                 /* got extra prog ref from syscall, or attaching to different prog */
3160                 bpf_prog_put(prog->aux->dst_prog);
3161         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3162                 /* we allocated a new trampoline, so free the old one */
3163                 bpf_trampoline_put(prog->aux->dst_trampoline);
3164
3165         prog->aux->dst_prog = NULL;
3166         prog->aux->dst_trampoline = NULL;
3167         mutex_unlock(&prog->aux->dst_mutex);
3168
3169         return bpf_link_settle(&link_primer);
3170 out_unlock:
3171         if (tr && tr != prog->aux->dst_trampoline)
3172                 bpf_trampoline_put(tr);
3173         mutex_unlock(&prog->aux->dst_mutex);
3174         kfree(link);
3175 out_put_prog:
3176         if (tgt_prog_fd && tgt_prog)
3177                 bpf_prog_put(tgt_prog);
3178         return err;
3179 }
3180
3181 struct bpf_raw_tp_link {
3182         struct bpf_link link;
3183         struct bpf_raw_event_map *btp;
3184 };
3185
3186 static void bpf_raw_tp_link_release(struct bpf_link *link)
3187 {
3188         struct bpf_raw_tp_link *raw_tp =
3189                 container_of(link, struct bpf_raw_tp_link, link);
3190
3191         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3192         bpf_put_raw_tracepoint(raw_tp->btp);
3193 }
3194
3195 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3196 {
3197         struct bpf_raw_tp_link *raw_tp =
3198                 container_of(link, struct bpf_raw_tp_link, link);
3199
3200         kfree(raw_tp);
3201 }
3202
3203 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3204                                         struct seq_file *seq)
3205 {
3206         struct bpf_raw_tp_link *raw_tp_link =
3207                 container_of(link, struct bpf_raw_tp_link, link);
3208
3209         seq_printf(seq,
3210                    "tp_name:\t%s\n",
3211                    raw_tp_link->btp->tp->name);
3212 }
3213
3214 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3215                                           struct bpf_link_info *info)
3216 {
3217         struct bpf_raw_tp_link *raw_tp_link =
3218                 container_of(link, struct bpf_raw_tp_link, link);
3219         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3220         const char *tp_name = raw_tp_link->btp->tp->name;
3221         u32 ulen = info->raw_tracepoint.tp_name_len;
3222         size_t tp_len = strlen(tp_name);
3223
3224         if (!ulen ^ !ubuf)
3225                 return -EINVAL;
3226
3227         info->raw_tracepoint.tp_name_len = tp_len + 1;
3228
3229         if (!ubuf)
3230                 return 0;
3231
3232         if (ulen >= tp_len + 1) {
3233                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3234                         return -EFAULT;
3235         } else {
3236                 char zero = '\0';
3237
3238                 if (copy_to_user(ubuf, tp_name, ulen - 1))
3239                         return -EFAULT;
3240                 if (put_user(zero, ubuf + ulen - 1))
3241                         return -EFAULT;
3242                 return -ENOSPC;
3243         }
3244
3245         return 0;
3246 }
3247
3248 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3249         .release = bpf_raw_tp_link_release,
3250         .dealloc = bpf_raw_tp_link_dealloc,
3251         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3252         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3253 };
3254
3255 #ifdef CONFIG_PERF_EVENTS
3256 struct bpf_perf_link {
3257         struct bpf_link link;
3258         struct file *perf_file;
3259 };
3260
3261 static void bpf_perf_link_release(struct bpf_link *link)
3262 {
3263         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3264         struct perf_event *event = perf_link->perf_file->private_data;
3265
3266         perf_event_free_bpf_prog(event);
3267         fput(perf_link->perf_file);
3268 }
3269
3270 static void bpf_perf_link_dealloc(struct bpf_link *link)
3271 {
3272         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3273
3274         kfree(perf_link);
3275 }
3276
3277 static const struct bpf_link_ops bpf_perf_link_lops = {
3278         .release = bpf_perf_link_release,
3279         .dealloc = bpf_perf_link_dealloc,
3280 };
3281
3282 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3283 {
3284         struct bpf_link_primer link_primer;
3285         struct bpf_perf_link *link;
3286         struct perf_event *event;
3287         struct file *perf_file;
3288         int err;
3289
3290         if (attr->link_create.flags)
3291                 return -EINVAL;
3292
3293         perf_file = perf_event_get(attr->link_create.target_fd);
3294         if (IS_ERR(perf_file))
3295                 return PTR_ERR(perf_file);
3296
3297         link = kzalloc(sizeof(*link), GFP_USER);
3298         if (!link) {
3299                 err = -ENOMEM;
3300                 goto out_put_file;
3301         }
3302         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3303         link->perf_file = perf_file;
3304
3305         err = bpf_link_prime(&link->link, &link_primer);
3306         if (err) {
3307                 kfree(link);
3308                 goto out_put_file;
3309         }
3310
3311         event = perf_file->private_data;
3312         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3313         if (err) {
3314                 bpf_link_cleanup(&link_primer);
3315                 goto out_put_file;
3316         }
3317         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3318         bpf_prog_inc(prog);
3319
3320         return bpf_link_settle(&link_primer);
3321
3322 out_put_file:
3323         fput(perf_file);
3324         return err;
3325 }
3326 #else
3327 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3328 {
3329         return -EOPNOTSUPP;
3330 }
3331 #endif /* CONFIG_PERF_EVENTS */
3332
3333 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3334                                   const char __user *user_tp_name)
3335 {
3336         struct bpf_link_primer link_primer;
3337         struct bpf_raw_tp_link *link;
3338         struct bpf_raw_event_map *btp;
3339         const char *tp_name;
3340         char buf[128];
3341         int err;
3342
3343         switch (prog->type) {
3344         case BPF_PROG_TYPE_TRACING:
3345         case BPF_PROG_TYPE_EXT:
3346         case BPF_PROG_TYPE_LSM:
3347                 if (user_tp_name)
3348                         /* The attach point for this category of programs
3349                          * should be specified via btf_id during program load.
3350                          */
3351                         return -EINVAL;
3352                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3353                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3354                         tp_name = prog->aux->attach_func_name;
3355                         break;
3356                 }
3357                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3358         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3359         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3360                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3361                         return -EFAULT;
3362                 buf[sizeof(buf) - 1] = 0;
3363                 tp_name = buf;
3364                 break;
3365         default:
3366                 return -EINVAL;
3367         }
3368
3369         btp = bpf_get_raw_tracepoint(tp_name);
3370         if (!btp)
3371                 return -ENOENT;
3372
3373         link = kzalloc(sizeof(*link), GFP_USER);
3374         if (!link) {
3375                 err = -ENOMEM;
3376                 goto out_put_btp;
3377         }
3378         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3379                       &bpf_raw_tp_link_lops, prog);
3380         link->btp = btp;
3381
3382         err = bpf_link_prime(&link->link, &link_primer);
3383         if (err) {
3384                 kfree(link);
3385                 goto out_put_btp;
3386         }
3387
3388         err = bpf_probe_register(link->btp, prog);
3389         if (err) {
3390                 bpf_link_cleanup(&link_primer);
3391                 goto out_put_btp;
3392         }
3393
3394         return bpf_link_settle(&link_primer);
3395
3396 out_put_btp:
3397         bpf_put_raw_tracepoint(btp);
3398         return err;
3399 }
3400
3401 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3402
3403 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3404 {
3405         struct bpf_prog *prog;
3406         int fd;
3407
3408         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3409                 return -EINVAL;
3410
3411         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3412         if (IS_ERR(prog))
3413                 return PTR_ERR(prog);
3414
3415         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3416         if (fd < 0)
3417                 bpf_prog_put(prog);
3418         return fd;
3419 }
3420
3421 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3422                                              enum bpf_attach_type attach_type)
3423 {
3424         switch (prog->type) {
3425         case BPF_PROG_TYPE_CGROUP_SOCK:
3426         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3427         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3428         case BPF_PROG_TYPE_SK_LOOKUP:
3429                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3430         case BPF_PROG_TYPE_CGROUP_SKB:
3431                 if (!capable(CAP_NET_ADMIN))
3432                         /* cg-skb progs can be loaded by unpriv user.
3433                          * check permissions at attach time.
3434                          */
3435                         return -EPERM;
3436                 return prog->enforce_expected_attach_type &&
3437                         prog->expected_attach_type != attach_type ?
3438                         -EINVAL : 0;
3439         default:
3440                 return 0;
3441         }
3442 }
3443
3444 static enum bpf_prog_type
3445 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3446 {
3447         switch (attach_type) {
3448         case BPF_CGROUP_INET_INGRESS:
3449         case BPF_CGROUP_INET_EGRESS:
3450                 return BPF_PROG_TYPE_CGROUP_SKB;
3451         case BPF_CGROUP_INET_SOCK_CREATE:
3452         case BPF_CGROUP_INET_SOCK_RELEASE:
3453         case BPF_CGROUP_INET4_POST_BIND:
3454         case BPF_CGROUP_INET6_POST_BIND:
3455                 return BPF_PROG_TYPE_CGROUP_SOCK;
3456         case BPF_CGROUP_INET4_BIND:
3457         case BPF_CGROUP_INET6_BIND:
3458         case BPF_CGROUP_INET4_CONNECT:
3459         case BPF_CGROUP_INET6_CONNECT:
3460         case BPF_CGROUP_INET4_GETPEERNAME:
3461         case BPF_CGROUP_INET6_GETPEERNAME:
3462         case BPF_CGROUP_INET4_GETSOCKNAME:
3463         case BPF_CGROUP_INET6_GETSOCKNAME:
3464         case BPF_CGROUP_UDP4_SENDMSG:
3465         case BPF_CGROUP_UDP6_SENDMSG:
3466         case BPF_CGROUP_UDP4_RECVMSG:
3467         case BPF_CGROUP_UDP6_RECVMSG:
3468                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3469         case BPF_CGROUP_SOCK_OPS:
3470                 return BPF_PROG_TYPE_SOCK_OPS;
3471         case BPF_CGROUP_DEVICE:
3472                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3473         case BPF_SK_MSG_VERDICT:
3474                 return BPF_PROG_TYPE_SK_MSG;
3475         case BPF_SK_SKB_STREAM_PARSER:
3476         case BPF_SK_SKB_STREAM_VERDICT:
3477         case BPF_SK_SKB_VERDICT:
3478                 return BPF_PROG_TYPE_SK_SKB;
3479         case BPF_LIRC_MODE2:
3480                 return BPF_PROG_TYPE_LIRC_MODE2;
3481         case BPF_FLOW_DISSECTOR:
3482                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3483         case BPF_CGROUP_SYSCTL:
3484                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3485         case BPF_CGROUP_GETSOCKOPT:
3486         case BPF_CGROUP_SETSOCKOPT:
3487                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3488         case BPF_TRACE_ITER:
3489         case BPF_TRACE_RAW_TP:
3490         case BPF_TRACE_FENTRY:
3491         case BPF_TRACE_FEXIT:
3492         case BPF_MODIFY_RETURN:
3493                 return BPF_PROG_TYPE_TRACING;
3494         case BPF_LSM_MAC:
3495                 return BPF_PROG_TYPE_LSM;
3496         case BPF_SK_LOOKUP:
3497                 return BPF_PROG_TYPE_SK_LOOKUP;
3498         case BPF_XDP:
3499                 return BPF_PROG_TYPE_XDP;
3500         case BPF_LSM_CGROUP:
3501                 return BPF_PROG_TYPE_LSM;
3502         default:
3503                 return BPF_PROG_TYPE_UNSPEC;
3504         }
3505 }
3506
3507 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3508
3509 #define BPF_F_ATTACH_MASK \
3510         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3511
3512 static int bpf_prog_attach(const union bpf_attr *attr)
3513 {
3514         enum bpf_prog_type ptype;
3515         struct bpf_prog *prog;
3516         int ret;
3517
3518         if (CHECK_ATTR(BPF_PROG_ATTACH))
3519                 return -EINVAL;
3520
3521         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3522                 return -EINVAL;
3523
3524         ptype = attach_type_to_prog_type(attr->attach_type);
3525         if (ptype == BPF_PROG_TYPE_UNSPEC)
3526                 return -EINVAL;
3527
3528         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3529         if (IS_ERR(prog))
3530                 return PTR_ERR(prog);
3531
3532         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3533                 bpf_prog_put(prog);
3534                 return -EINVAL;
3535         }
3536
3537         switch (ptype) {
3538         case BPF_PROG_TYPE_SK_SKB:
3539         case BPF_PROG_TYPE_SK_MSG:
3540                 ret = sock_map_get_from_fd(attr, prog);
3541                 break;
3542         case BPF_PROG_TYPE_LIRC_MODE2:
3543                 ret = lirc_prog_attach(attr, prog);
3544                 break;
3545         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3546                 ret = netns_bpf_prog_attach(attr, prog);
3547                 break;
3548         case BPF_PROG_TYPE_CGROUP_DEVICE:
3549         case BPF_PROG_TYPE_CGROUP_SKB:
3550         case BPF_PROG_TYPE_CGROUP_SOCK:
3551         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3552         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3553         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3554         case BPF_PROG_TYPE_SOCK_OPS:
3555         case BPF_PROG_TYPE_LSM:
3556                 if (ptype == BPF_PROG_TYPE_LSM &&
3557                     prog->expected_attach_type != BPF_LSM_CGROUP)
3558                         ret = -EINVAL;
3559                 else
3560                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3561                 break;
3562         default:
3563                 ret = -EINVAL;
3564         }
3565
3566         if (ret)
3567                 bpf_prog_put(prog);
3568         return ret;
3569 }
3570
3571 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3572
3573 static int bpf_prog_detach(const union bpf_attr *attr)
3574 {
3575         enum bpf_prog_type ptype;
3576
3577         if (CHECK_ATTR(BPF_PROG_DETACH))
3578                 return -EINVAL;
3579
3580         ptype = attach_type_to_prog_type(attr->attach_type);
3581
3582         switch (ptype) {
3583         case BPF_PROG_TYPE_SK_MSG:
3584         case BPF_PROG_TYPE_SK_SKB:
3585                 return sock_map_prog_detach(attr, ptype);
3586         case BPF_PROG_TYPE_LIRC_MODE2:
3587                 return lirc_prog_detach(attr);
3588         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3589                 return netns_bpf_prog_detach(attr, ptype);
3590         case BPF_PROG_TYPE_CGROUP_DEVICE:
3591         case BPF_PROG_TYPE_CGROUP_SKB:
3592         case BPF_PROG_TYPE_CGROUP_SOCK:
3593         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3594         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3595         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3596         case BPF_PROG_TYPE_SOCK_OPS:
3597         case BPF_PROG_TYPE_LSM:
3598                 return cgroup_bpf_prog_detach(attr, ptype);
3599         default:
3600                 return -EINVAL;
3601         }
3602 }
3603
3604 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3605
3606 static int bpf_prog_query(const union bpf_attr *attr,
3607                           union bpf_attr __user *uattr)
3608 {
3609         if (!capable(CAP_NET_ADMIN))
3610                 return -EPERM;
3611         if (CHECK_ATTR(BPF_PROG_QUERY))
3612                 return -EINVAL;
3613         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3614                 return -EINVAL;
3615
3616         switch (attr->query.attach_type) {
3617         case BPF_CGROUP_INET_INGRESS:
3618         case BPF_CGROUP_INET_EGRESS:
3619         case BPF_CGROUP_INET_SOCK_CREATE:
3620         case BPF_CGROUP_INET_SOCK_RELEASE:
3621         case BPF_CGROUP_INET4_BIND:
3622         case BPF_CGROUP_INET6_BIND:
3623         case BPF_CGROUP_INET4_POST_BIND:
3624         case BPF_CGROUP_INET6_POST_BIND:
3625         case BPF_CGROUP_INET4_CONNECT:
3626         case BPF_CGROUP_INET6_CONNECT:
3627         case BPF_CGROUP_INET4_GETPEERNAME:
3628         case BPF_CGROUP_INET6_GETPEERNAME:
3629         case BPF_CGROUP_INET4_GETSOCKNAME:
3630         case BPF_CGROUP_INET6_GETSOCKNAME:
3631         case BPF_CGROUP_UDP4_SENDMSG:
3632         case BPF_CGROUP_UDP6_SENDMSG:
3633         case BPF_CGROUP_UDP4_RECVMSG:
3634         case BPF_CGROUP_UDP6_RECVMSG:
3635         case BPF_CGROUP_SOCK_OPS:
3636         case BPF_CGROUP_DEVICE:
3637         case BPF_CGROUP_SYSCTL:
3638         case BPF_CGROUP_GETSOCKOPT:
3639         case BPF_CGROUP_SETSOCKOPT:
3640         case BPF_LSM_CGROUP:
3641                 return cgroup_bpf_prog_query(attr, uattr);
3642         case BPF_LIRC_MODE2:
3643                 return lirc_prog_query(attr, uattr);
3644         case BPF_FLOW_DISSECTOR:
3645         case BPF_SK_LOOKUP:
3646                 return netns_bpf_prog_query(attr, uattr);
3647         case BPF_SK_SKB_STREAM_PARSER:
3648         case BPF_SK_SKB_STREAM_VERDICT:
3649         case BPF_SK_MSG_VERDICT:
3650         case BPF_SK_SKB_VERDICT:
3651                 return sock_map_bpf_prog_query(attr, uattr);
3652         default:
3653                 return -EINVAL;
3654         }
3655 }
3656
3657 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3658
3659 static int bpf_prog_test_run(const union bpf_attr *attr,
3660                              union bpf_attr __user *uattr)
3661 {
3662         struct bpf_prog *prog;
3663         int ret = -ENOTSUPP;
3664
3665         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3666                 return -EINVAL;
3667
3668         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3669             (!attr->test.ctx_size_in && attr->test.ctx_in))
3670                 return -EINVAL;
3671
3672         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3673             (!attr->test.ctx_size_out && attr->test.ctx_out))
3674                 return -EINVAL;
3675
3676         prog = bpf_prog_get(attr->test.prog_fd);
3677         if (IS_ERR(prog))
3678                 return PTR_ERR(prog);
3679
3680         if (prog->aux->ops->test_run)
3681                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3682
3683         bpf_prog_put(prog);
3684         return ret;
3685 }
3686
3687 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3688
3689 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3690                                union bpf_attr __user *uattr,
3691                                struct idr *idr,
3692                                spinlock_t *lock)
3693 {
3694         u32 next_id = attr->start_id;
3695         int err = 0;
3696
3697         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3698                 return -EINVAL;
3699
3700         if (!capable(CAP_SYS_ADMIN))
3701                 return -EPERM;
3702
3703         next_id++;
3704         spin_lock_bh(lock);
3705         if (!idr_get_next(idr, &next_id))
3706                 err = -ENOENT;
3707         spin_unlock_bh(lock);
3708
3709         if (!err)
3710                 err = put_user(next_id, &uattr->next_id);
3711
3712         return err;
3713 }
3714
3715 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3716 {
3717         struct bpf_map *map;
3718
3719         spin_lock_bh(&map_idr_lock);
3720 again:
3721         map = idr_get_next(&map_idr, id);
3722         if (map) {
3723                 map = __bpf_map_inc_not_zero(map, false);
3724                 if (IS_ERR(map)) {
3725                         (*id)++;
3726                         goto again;
3727                 }
3728         }
3729         spin_unlock_bh(&map_idr_lock);
3730
3731         return map;
3732 }
3733
3734 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3735 {
3736         struct bpf_prog *prog;
3737
3738         spin_lock_bh(&prog_idr_lock);
3739 again:
3740         prog = idr_get_next(&prog_idr, id);
3741         if (prog) {
3742                 prog = bpf_prog_inc_not_zero(prog);
3743                 if (IS_ERR(prog)) {
3744                         (*id)++;
3745                         goto again;
3746                 }
3747         }
3748         spin_unlock_bh(&prog_idr_lock);
3749
3750         return prog;
3751 }
3752
3753 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3754
3755 struct bpf_prog *bpf_prog_by_id(u32 id)
3756 {
3757         struct bpf_prog *prog;
3758
3759         if (!id)
3760                 return ERR_PTR(-ENOENT);
3761
3762         spin_lock_bh(&prog_idr_lock);
3763         prog = idr_find(&prog_idr, id);
3764         if (prog)
3765                 prog = bpf_prog_inc_not_zero(prog);
3766         else
3767                 prog = ERR_PTR(-ENOENT);
3768         spin_unlock_bh(&prog_idr_lock);
3769         return prog;
3770 }
3771
3772 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3773 {
3774         struct bpf_prog *prog;
3775         u32 id = attr->prog_id;
3776         int fd;
3777
3778         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3779                 return -EINVAL;
3780
3781         if (!capable(CAP_SYS_ADMIN))
3782                 return -EPERM;
3783
3784         prog = bpf_prog_by_id(id);
3785         if (IS_ERR(prog))
3786                 return PTR_ERR(prog);
3787
3788         fd = bpf_prog_new_fd(prog);
3789         if (fd < 0)
3790                 bpf_prog_put(prog);
3791
3792         return fd;
3793 }
3794
3795 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3796
3797 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3798 {
3799         struct bpf_map *map;
3800         u32 id = attr->map_id;
3801         int f_flags;
3802         int fd;
3803
3804         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3805             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3806                 return -EINVAL;
3807
3808         if (!capable(CAP_SYS_ADMIN))
3809                 return -EPERM;
3810
3811         f_flags = bpf_get_file_flag(attr->open_flags);
3812         if (f_flags < 0)
3813                 return f_flags;
3814
3815         spin_lock_bh(&map_idr_lock);
3816         map = idr_find(&map_idr, id);
3817         if (map)
3818                 map = __bpf_map_inc_not_zero(map, true);
3819         else
3820                 map = ERR_PTR(-ENOENT);
3821         spin_unlock_bh(&map_idr_lock);
3822
3823         if (IS_ERR(map))
3824                 return PTR_ERR(map);
3825
3826         fd = bpf_map_new_fd(map, f_flags);
3827         if (fd < 0)
3828                 bpf_map_put_with_uref(map);
3829
3830         return fd;
3831 }
3832
3833 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3834                                               unsigned long addr, u32 *off,
3835                                               u32 *type)
3836 {
3837         const struct bpf_map *map;
3838         int i;
3839
3840         mutex_lock(&prog->aux->used_maps_mutex);
3841         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3842                 map = prog->aux->used_maps[i];
3843                 if (map == (void *)addr) {
3844                         *type = BPF_PSEUDO_MAP_FD;
3845                         goto out;
3846                 }
3847                 if (!map->ops->map_direct_value_meta)
3848                         continue;
3849                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3850                         *type = BPF_PSEUDO_MAP_VALUE;
3851                         goto out;
3852                 }
3853         }
3854         map = NULL;
3855
3856 out:
3857         mutex_unlock(&prog->aux->used_maps_mutex);
3858         return map;
3859 }
3860
3861 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3862                                               const struct cred *f_cred)
3863 {
3864         const struct bpf_map *map;
3865         struct bpf_insn *insns;
3866         u32 off, type;
3867         u64 imm;
3868         u8 code;
3869         int i;
3870
3871         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3872                         GFP_USER);
3873         if (!insns)
3874                 return insns;
3875
3876         for (i = 0; i < prog->len; i++) {
3877                 code = insns[i].code;
3878
3879                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3880                         insns[i].code = BPF_JMP | BPF_CALL;
3881                         insns[i].imm = BPF_FUNC_tail_call;
3882                         /* fall-through */
3883                 }
3884                 if (code == (BPF_JMP | BPF_CALL) ||
3885                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3886                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3887                                 insns[i].code = BPF_JMP | BPF_CALL;
3888                         if (!bpf_dump_raw_ok(f_cred))
3889                                 insns[i].imm = 0;
3890                         continue;
3891                 }
3892                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3893                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3894                         continue;
3895                 }
3896
3897                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3898                         continue;
3899
3900                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3901                 map = bpf_map_from_imm(prog, imm, &off, &type);
3902                 if (map) {
3903                         insns[i].src_reg = type;
3904                         insns[i].imm = map->id;
3905                         insns[i + 1].imm = off;
3906                         continue;
3907                 }
3908         }
3909
3910         return insns;
3911 }
3912
3913 static int set_info_rec_size(struct bpf_prog_info *info)
3914 {
3915         /*
3916          * Ensure info.*_rec_size is the same as kernel expected size
3917          *
3918          * or
3919          *
3920          * Only allow zero *_rec_size if both _rec_size and _cnt are
3921          * zero.  In this case, the kernel will set the expected
3922          * _rec_size back to the info.
3923          */
3924
3925         if ((info->nr_func_info || info->func_info_rec_size) &&
3926             info->func_info_rec_size != sizeof(struct bpf_func_info))
3927                 return -EINVAL;
3928
3929         if ((info->nr_line_info || info->line_info_rec_size) &&
3930             info->line_info_rec_size != sizeof(struct bpf_line_info))
3931                 return -EINVAL;
3932
3933         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3934             info->jited_line_info_rec_size != sizeof(__u64))
3935                 return -EINVAL;
3936
3937         info->func_info_rec_size = sizeof(struct bpf_func_info);
3938         info->line_info_rec_size = sizeof(struct bpf_line_info);
3939         info->jited_line_info_rec_size = sizeof(__u64);
3940
3941         return 0;
3942 }
3943
3944 static int bpf_prog_get_info_by_fd(struct file *file,
3945                                    struct bpf_prog *prog,
3946                                    const union bpf_attr *attr,
3947                                    union bpf_attr __user *uattr)
3948 {
3949         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3950         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3951         struct bpf_prog_info info;
3952         u32 info_len = attr->info.info_len;
3953         struct bpf_prog_kstats stats;
3954         char __user *uinsns;
3955         u32 ulen;
3956         int err;
3957
3958         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3959         if (err)
3960                 return err;
3961         info_len = min_t(u32, sizeof(info), info_len);
3962
3963         memset(&info, 0, sizeof(info));
3964         if (copy_from_user(&info, uinfo, info_len))
3965                 return -EFAULT;
3966
3967         info.type = prog->type;
3968         info.id = prog->aux->id;
3969         info.load_time = prog->aux->load_time;
3970         info.created_by_uid = from_kuid_munged(current_user_ns(),
3971                                                prog->aux->user->uid);
3972         info.gpl_compatible = prog->gpl_compatible;
3973
3974         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3975         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3976
3977         mutex_lock(&prog->aux->used_maps_mutex);
3978         ulen = info.nr_map_ids;
3979         info.nr_map_ids = prog->aux->used_map_cnt;
3980         ulen = min_t(u32, info.nr_map_ids, ulen);
3981         if (ulen) {
3982                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3983                 u32 i;
3984
3985                 for (i = 0; i < ulen; i++)
3986                         if (put_user(prog->aux->used_maps[i]->id,
3987                                      &user_map_ids[i])) {
3988                                 mutex_unlock(&prog->aux->used_maps_mutex);
3989                                 return -EFAULT;
3990                         }
3991         }
3992         mutex_unlock(&prog->aux->used_maps_mutex);
3993
3994         err = set_info_rec_size(&info);
3995         if (err)
3996                 return err;
3997
3998         bpf_prog_get_stats(prog, &stats);
3999         info.run_time_ns = stats.nsecs;
4000         info.run_cnt = stats.cnt;
4001         info.recursion_misses = stats.misses;
4002
4003         info.verified_insns = prog->aux->verified_insns;
4004
4005         if (!bpf_capable()) {
4006                 info.jited_prog_len = 0;
4007                 info.xlated_prog_len = 0;
4008                 info.nr_jited_ksyms = 0;
4009                 info.nr_jited_func_lens = 0;
4010                 info.nr_func_info = 0;
4011                 info.nr_line_info = 0;
4012                 info.nr_jited_line_info = 0;
4013                 goto done;
4014         }
4015
4016         ulen = info.xlated_prog_len;
4017         info.xlated_prog_len = bpf_prog_insn_size(prog);
4018         if (info.xlated_prog_len && ulen) {
4019                 struct bpf_insn *insns_sanitized;
4020                 bool fault;
4021
4022                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4023                         info.xlated_prog_insns = 0;
4024                         goto done;
4025                 }
4026                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4027                 if (!insns_sanitized)
4028                         return -ENOMEM;
4029                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4030                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4031                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4032                 kfree(insns_sanitized);
4033                 if (fault)
4034                         return -EFAULT;
4035         }
4036
4037         if (bpf_prog_is_offloaded(prog->aux)) {
4038                 err = bpf_prog_offload_info_fill(&info, prog);
4039                 if (err)
4040                         return err;
4041                 goto done;
4042         }
4043
4044         /* NOTE: the following code is supposed to be skipped for offload.
4045          * bpf_prog_offload_info_fill() is the place to fill similar fields
4046          * for offload.
4047          */
4048         ulen = info.jited_prog_len;
4049         if (prog->aux->func_cnt) {
4050                 u32 i;
4051
4052                 info.jited_prog_len = 0;
4053                 for (i = 0; i < prog->aux->func_cnt; i++)
4054                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4055         } else {
4056                 info.jited_prog_len = prog->jited_len;
4057         }
4058
4059         if (info.jited_prog_len && ulen) {
4060                 if (bpf_dump_raw_ok(file->f_cred)) {
4061                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4062                         ulen = min_t(u32, info.jited_prog_len, ulen);
4063
4064                         /* for multi-function programs, copy the JITed
4065                          * instructions for all the functions
4066                          */
4067                         if (prog->aux->func_cnt) {
4068                                 u32 len, free, i;
4069                                 u8 *img;
4070
4071                                 free = ulen;
4072                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4073                                         len = prog->aux->func[i]->jited_len;
4074                                         len = min_t(u32, len, free);
4075                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4076                                         if (copy_to_user(uinsns, img, len))
4077                                                 return -EFAULT;
4078                                         uinsns += len;
4079                                         free -= len;
4080                                         if (!free)
4081                                                 break;
4082                                 }
4083                         } else {
4084                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4085                                         return -EFAULT;
4086                         }
4087                 } else {
4088                         info.jited_prog_insns = 0;
4089                 }
4090         }
4091
4092         ulen = info.nr_jited_ksyms;
4093         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4094         if (ulen) {
4095                 if (bpf_dump_raw_ok(file->f_cred)) {
4096                         unsigned long ksym_addr;
4097                         u64 __user *user_ksyms;
4098                         u32 i;
4099
4100                         /* copy the address of the kernel symbol
4101                          * corresponding to each function
4102                          */
4103                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4104                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4105                         if (prog->aux->func_cnt) {
4106                                 for (i = 0; i < ulen; i++) {
4107                                         ksym_addr = (unsigned long)
4108                                                 prog->aux->func[i]->bpf_func;
4109                                         if (put_user((u64) ksym_addr,
4110                                                      &user_ksyms[i]))
4111                                                 return -EFAULT;
4112                                 }
4113                         } else {
4114                                 ksym_addr = (unsigned long) prog->bpf_func;
4115                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4116                                         return -EFAULT;
4117                         }
4118                 } else {
4119                         info.jited_ksyms = 0;
4120                 }
4121         }
4122
4123         ulen = info.nr_jited_func_lens;
4124         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4125         if (ulen) {
4126                 if (bpf_dump_raw_ok(file->f_cred)) {
4127                         u32 __user *user_lens;
4128                         u32 func_len, i;
4129
4130                         /* copy the JITed image lengths for each function */
4131                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4132                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4133                         if (prog->aux->func_cnt) {
4134                                 for (i = 0; i < ulen; i++) {
4135                                         func_len =
4136                                                 prog->aux->func[i]->jited_len;
4137                                         if (put_user(func_len, &user_lens[i]))
4138                                                 return -EFAULT;
4139                                 }
4140                         } else {
4141                                 func_len = prog->jited_len;
4142                                 if (put_user(func_len, &user_lens[0]))
4143                                         return -EFAULT;
4144                         }
4145                 } else {
4146                         info.jited_func_lens = 0;
4147                 }
4148         }
4149
4150         if (prog->aux->btf)
4151                 info.btf_id = btf_obj_id(prog->aux->btf);
4152         info.attach_btf_id = prog->aux->attach_btf_id;
4153         if (attach_btf)
4154                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4155
4156         ulen = info.nr_func_info;
4157         info.nr_func_info = prog->aux->func_info_cnt;
4158         if (info.nr_func_info && ulen) {
4159                 char __user *user_finfo;
4160
4161                 user_finfo = u64_to_user_ptr(info.func_info);
4162                 ulen = min_t(u32, info.nr_func_info, ulen);
4163                 if (copy_to_user(user_finfo, prog->aux->func_info,
4164                                  info.func_info_rec_size * ulen))
4165                         return -EFAULT;
4166         }
4167
4168         ulen = info.nr_line_info;
4169         info.nr_line_info = prog->aux->nr_linfo;
4170         if (info.nr_line_info && ulen) {
4171                 __u8 __user *user_linfo;
4172
4173                 user_linfo = u64_to_user_ptr(info.line_info);
4174                 ulen = min_t(u32, info.nr_line_info, ulen);
4175                 if (copy_to_user(user_linfo, prog->aux->linfo,
4176                                  info.line_info_rec_size * ulen))
4177                         return -EFAULT;
4178         }
4179
4180         ulen = info.nr_jited_line_info;
4181         if (prog->aux->jited_linfo)
4182                 info.nr_jited_line_info = prog->aux->nr_linfo;
4183         else
4184                 info.nr_jited_line_info = 0;
4185         if (info.nr_jited_line_info && ulen) {
4186                 if (bpf_dump_raw_ok(file->f_cred)) {
4187                         unsigned long line_addr;
4188                         __u64 __user *user_linfo;
4189                         u32 i;
4190
4191                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4192                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4193                         for (i = 0; i < ulen; i++) {
4194                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4195                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4196                                         return -EFAULT;
4197                         }
4198                 } else {
4199                         info.jited_line_info = 0;
4200                 }
4201         }
4202
4203         ulen = info.nr_prog_tags;
4204         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4205         if (ulen) {
4206                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4207                 u32 i;
4208
4209                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4210                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4211                 if (prog->aux->func_cnt) {
4212                         for (i = 0; i < ulen; i++) {
4213                                 if (copy_to_user(user_prog_tags[i],
4214                                                  prog->aux->func[i]->tag,
4215                                                  BPF_TAG_SIZE))
4216                                         return -EFAULT;
4217                         }
4218                 } else {
4219                         if (copy_to_user(user_prog_tags[0],
4220                                          prog->tag, BPF_TAG_SIZE))
4221                                 return -EFAULT;
4222                 }
4223         }
4224
4225 done:
4226         if (copy_to_user(uinfo, &info, info_len) ||
4227             put_user(info_len, &uattr->info.info_len))
4228                 return -EFAULT;
4229
4230         return 0;
4231 }
4232
4233 static int bpf_map_get_info_by_fd(struct file *file,
4234                                   struct bpf_map *map,
4235                                   const union bpf_attr *attr,
4236                                   union bpf_attr __user *uattr)
4237 {
4238         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4239         struct bpf_map_info info;
4240         u32 info_len = attr->info.info_len;
4241         int err;
4242
4243         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4244         if (err)
4245                 return err;
4246         info_len = min_t(u32, sizeof(info), info_len);
4247
4248         memset(&info, 0, sizeof(info));
4249         info.type = map->map_type;
4250         info.id = map->id;
4251         info.key_size = map->key_size;
4252         info.value_size = map->value_size;
4253         info.max_entries = map->max_entries;
4254         info.map_flags = map->map_flags;
4255         info.map_extra = map->map_extra;
4256         memcpy(info.name, map->name, sizeof(map->name));
4257
4258         if (map->btf) {
4259                 info.btf_id = btf_obj_id(map->btf);
4260                 info.btf_key_type_id = map->btf_key_type_id;
4261                 info.btf_value_type_id = map->btf_value_type_id;
4262         }
4263         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4264
4265         if (bpf_map_is_offloaded(map)) {
4266                 err = bpf_map_offload_info_fill(&info, map);
4267                 if (err)
4268                         return err;
4269         }
4270
4271         if (copy_to_user(uinfo, &info, info_len) ||
4272             put_user(info_len, &uattr->info.info_len))
4273                 return -EFAULT;
4274
4275         return 0;
4276 }
4277
4278 static int bpf_btf_get_info_by_fd(struct file *file,
4279                                   struct btf *btf,
4280                                   const union bpf_attr *attr,
4281                                   union bpf_attr __user *uattr)
4282 {
4283         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4284         u32 info_len = attr->info.info_len;
4285         int err;
4286
4287         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4288         if (err)
4289                 return err;
4290
4291         return btf_get_info_by_fd(btf, attr, uattr);
4292 }
4293
4294 static int bpf_link_get_info_by_fd(struct file *file,
4295                                   struct bpf_link *link,
4296                                   const union bpf_attr *attr,
4297                                   union bpf_attr __user *uattr)
4298 {
4299         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4300         struct bpf_link_info info;
4301         u32 info_len = attr->info.info_len;
4302         int err;
4303
4304         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4305         if (err)
4306                 return err;
4307         info_len = min_t(u32, sizeof(info), info_len);
4308
4309         memset(&info, 0, sizeof(info));
4310         if (copy_from_user(&info, uinfo, info_len))
4311                 return -EFAULT;
4312
4313         info.type = link->type;
4314         info.id = link->id;
4315         if (link->prog)
4316                 info.prog_id = link->prog->aux->id;
4317
4318         if (link->ops->fill_link_info) {
4319                 err = link->ops->fill_link_info(link, &info);
4320                 if (err)
4321                         return err;
4322         }
4323
4324         if (copy_to_user(uinfo, &info, info_len) ||
4325             put_user(info_len, &uattr->info.info_len))
4326                 return -EFAULT;
4327
4328         return 0;
4329 }
4330
4331
4332 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4333
4334 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4335                                   union bpf_attr __user *uattr)
4336 {
4337         int ufd = attr->info.bpf_fd;
4338         struct fd f;
4339         int err;
4340
4341         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4342                 return -EINVAL;
4343
4344         f = fdget(ufd);
4345         if (!f.file)
4346                 return -EBADFD;
4347
4348         if (f.file->f_op == &bpf_prog_fops)
4349                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4350                                               uattr);
4351         else if (f.file->f_op == &bpf_map_fops)
4352                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4353                                              uattr);
4354         else if (f.file->f_op == &btf_fops)
4355                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4356         else if (f.file->f_op == &bpf_link_fops)
4357                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4358                                               attr, uattr);
4359         else
4360                 err = -EINVAL;
4361
4362         fdput(f);
4363         return err;
4364 }
4365
4366 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4367
4368 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4369 {
4370         if (CHECK_ATTR(BPF_BTF_LOAD))
4371                 return -EINVAL;
4372
4373         if (!bpf_capable())
4374                 return -EPERM;
4375
4376         return btf_new_fd(attr, uattr, uattr_size);
4377 }
4378
4379 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4380
4381 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4382 {
4383         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4384                 return -EINVAL;
4385
4386         if (!capable(CAP_SYS_ADMIN))
4387                 return -EPERM;
4388
4389         return btf_get_fd_by_id(attr->btf_id);
4390 }
4391
4392 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4393                                     union bpf_attr __user *uattr,
4394                                     u32 prog_id, u32 fd_type,
4395                                     const char *buf, u64 probe_offset,
4396                                     u64 probe_addr)
4397 {
4398         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4399         u32 len = buf ? strlen(buf) : 0, input_len;
4400         int err = 0;
4401
4402         if (put_user(len, &uattr->task_fd_query.buf_len))
4403                 return -EFAULT;
4404         input_len = attr->task_fd_query.buf_len;
4405         if (input_len && ubuf) {
4406                 if (!len) {
4407                         /* nothing to copy, just make ubuf NULL terminated */
4408                         char zero = '\0';
4409
4410                         if (put_user(zero, ubuf))
4411                                 return -EFAULT;
4412                 } else if (input_len >= len + 1) {
4413                         /* ubuf can hold the string with NULL terminator */
4414                         if (copy_to_user(ubuf, buf, len + 1))
4415                                 return -EFAULT;
4416                 } else {
4417                         /* ubuf cannot hold the string with NULL terminator,
4418                          * do a partial copy with NULL terminator.
4419                          */
4420                         char zero = '\0';
4421
4422                         err = -ENOSPC;
4423                         if (copy_to_user(ubuf, buf, input_len - 1))
4424                                 return -EFAULT;
4425                         if (put_user(zero, ubuf + input_len - 1))
4426                                 return -EFAULT;
4427                 }
4428         }
4429
4430         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4431             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4432             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4433             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4434                 return -EFAULT;
4435
4436         return err;
4437 }
4438
4439 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4440
4441 static int bpf_task_fd_query(const union bpf_attr *attr,
4442                              union bpf_attr __user *uattr)
4443 {
4444         pid_t pid = attr->task_fd_query.pid;
4445         u32 fd = attr->task_fd_query.fd;
4446         const struct perf_event *event;
4447         struct task_struct *task;
4448         struct file *file;
4449         int err;
4450
4451         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4452                 return -EINVAL;
4453
4454         if (!capable(CAP_SYS_ADMIN))
4455                 return -EPERM;
4456
4457         if (attr->task_fd_query.flags != 0)
4458                 return -EINVAL;
4459
4460         rcu_read_lock();
4461         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4462         rcu_read_unlock();
4463         if (!task)
4464                 return -ENOENT;
4465
4466         err = 0;
4467         file = fget_task(task, fd);
4468         put_task_struct(task);
4469         if (!file)
4470                 return -EBADF;
4471
4472         if (file->f_op == &bpf_link_fops) {
4473                 struct bpf_link *link = file->private_data;
4474
4475                 if (link->ops == &bpf_raw_tp_link_lops) {
4476                         struct bpf_raw_tp_link *raw_tp =
4477                                 container_of(link, struct bpf_raw_tp_link, link);
4478                         struct bpf_raw_event_map *btp = raw_tp->btp;
4479
4480                         err = bpf_task_fd_query_copy(attr, uattr,
4481                                                      raw_tp->link.prog->aux->id,
4482                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4483                                                      btp->tp->name, 0, 0);
4484                         goto put_file;
4485                 }
4486                 goto out_not_supp;
4487         }
4488
4489         event = perf_get_event(file);
4490         if (!IS_ERR(event)) {
4491                 u64 probe_offset, probe_addr;
4492                 u32 prog_id, fd_type;
4493                 const char *buf;
4494
4495                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4496                                               &buf, &probe_offset,
4497                                               &probe_addr);
4498                 if (!err)
4499                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4500                                                      fd_type, buf,
4501                                                      probe_offset,
4502                                                      probe_addr);
4503                 goto put_file;
4504         }
4505
4506 out_not_supp:
4507         err = -ENOTSUPP;
4508 put_file:
4509         fput(file);
4510         return err;
4511 }
4512
4513 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4514
4515 #define BPF_DO_BATCH(fn, ...)                   \
4516         do {                                    \
4517                 if (!fn) {                      \
4518                         err = -ENOTSUPP;        \
4519                         goto err_put;           \
4520                 }                               \
4521                 err = fn(__VA_ARGS__);          \
4522         } while (0)
4523
4524 static int bpf_map_do_batch(const union bpf_attr *attr,
4525                             union bpf_attr __user *uattr,
4526                             int cmd)
4527 {
4528         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4529                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4530         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4531         struct bpf_map *map;
4532         int err, ufd;
4533         struct fd f;
4534
4535         if (CHECK_ATTR(BPF_MAP_BATCH))
4536                 return -EINVAL;
4537
4538         ufd = attr->batch.map_fd;
4539         f = fdget(ufd);
4540         map = __bpf_map_get(f);
4541         if (IS_ERR(map))
4542                 return PTR_ERR(map);
4543         if (has_write)
4544                 bpf_map_write_active_inc(map);
4545         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4546                 err = -EPERM;
4547                 goto err_put;
4548         }
4549         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4550                 err = -EPERM;
4551                 goto err_put;
4552         }
4553
4554         if (cmd == BPF_MAP_LOOKUP_BATCH)
4555                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4556         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4557                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4558         else if (cmd == BPF_MAP_UPDATE_BATCH)
4559                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4560         else
4561                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4562 err_put:
4563         if (has_write)
4564                 bpf_map_write_active_dec(map);
4565         fdput(f);
4566         return err;
4567 }
4568
4569 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4570 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4571 {
4572         enum bpf_prog_type ptype;
4573         struct bpf_prog *prog;
4574         int ret;
4575
4576         if (CHECK_ATTR(BPF_LINK_CREATE))
4577                 return -EINVAL;
4578
4579         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
4580                 return bpf_struct_ops_link_create(attr);
4581
4582         prog = bpf_prog_get(attr->link_create.prog_fd);
4583         if (IS_ERR(prog))
4584                 return PTR_ERR(prog);
4585
4586         ret = bpf_prog_attach_check_attach_type(prog,
4587                                                 attr->link_create.attach_type);
4588         if (ret)
4589                 goto out;
4590
4591         switch (prog->type) {
4592         case BPF_PROG_TYPE_EXT:
4593         case BPF_PROG_TYPE_NETFILTER:
4594                 break;
4595         case BPF_PROG_TYPE_PERF_EVENT:
4596         case BPF_PROG_TYPE_TRACEPOINT:
4597                 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4598                         ret = -EINVAL;
4599                         goto out;
4600                 }
4601                 break;
4602         case BPF_PROG_TYPE_KPROBE:
4603                 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4604                     attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4605                         ret = -EINVAL;
4606                         goto out;
4607                 }
4608                 break;
4609         default:
4610                 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4611                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4612                         ret = -EINVAL;
4613                         goto out;
4614                 }
4615                 break;
4616         }
4617
4618         switch (prog->type) {
4619         case BPF_PROG_TYPE_CGROUP_SKB:
4620         case BPF_PROG_TYPE_CGROUP_SOCK:
4621         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4622         case BPF_PROG_TYPE_SOCK_OPS:
4623         case BPF_PROG_TYPE_CGROUP_DEVICE:
4624         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4625         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4626                 ret = cgroup_bpf_link_attach(attr, prog);
4627                 break;
4628         case BPF_PROG_TYPE_EXT:
4629                 ret = bpf_tracing_prog_attach(prog,
4630                                               attr->link_create.target_fd,
4631                                               attr->link_create.target_btf_id,
4632                                               attr->link_create.tracing.cookie);
4633                 break;
4634         case BPF_PROG_TYPE_LSM:
4635         case BPF_PROG_TYPE_TRACING:
4636                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4637                         ret = -EINVAL;
4638                         goto out;
4639                 }
4640                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4641                         ret = bpf_raw_tp_link_attach(prog, NULL);
4642                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4643                         ret = bpf_iter_link_attach(attr, uattr, prog);
4644                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4645                         ret = cgroup_bpf_link_attach(attr, prog);
4646                 else
4647                         ret = bpf_tracing_prog_attach(prog,
4648                                                       attr->link_create.target_fd,
4649                                                       attr->link_create.target_btf_id,
4650                                                       attr->link_create.tracing.cookie);
4651                 break;
4652         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4653         case BPF_PROG_TYPE_SK_LOOKUP:
4654                 ret = netns_bpf_link_create(attr, prog);
4655                 break;
4656 #ifdef CONFIG_NET
4657         case BPF_PROG_TYPE_XDP:
4658                 ret = bpf_xdp_link_attach(attr, prog);
4659                 break;
4660         case BPF_PROG_TYPE_NETFILTER:
4661                 ret = bpf_nf_link_attach(attr, prog);
4662                 break;
4663 #endif
4664         case BPF_PROG_TYPE_PERF_EVENT:
4665         case BPF_PROG_TYPE_TRACEPOINT:
4666                 ret = bpf_perf_link_attach(attr, prog);
4667                 break;
4668         case BPF_PROG_TYPE_KPROBE:
4669                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4670                         ret = bpf_perf_link_attach(attr, prog);
4671                 else
4672                         ret = bpf_kprobe_multi_link_attach(attr, prog);
4673                 break;
4674         default:
4675                 ret = -EINVAL;
4676         }
4677
4678 out:
4679         if (ret < 0)
4680                 bpf_prog_put(prog);
4681         return ret;
4682 }
4683
4684 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
4685 {
4686         struct bpf_map *new_map, *old_map = NULL;
4687         int ret;
4688
4689         new_map = bpf_map_get(attr->link_update.new_map_fd);
4690         if (IS_ERR(new_map))
4691                 return PTR_ERR(new_map);
4692
4693         if (attr->link_update.flags & BPF_F_REPLACE) {
4694                 old_map = bpf_map_get(attr->link_update.old_map_fd);
4695                 if (IS_ERR(old_map)) {
4696                         ret = PTR_ERR(old_map);
4697                         goto out_put;
4698                 }
4699         } else if (attr->link_update.old_map_fd) {
4700                 ret = -EINVAL;
4701                 goto out_put;
4702         }
4703
4704         ret = link->ops->update_map(link, new_map, old_map);
4705
4706         if (old_map)
4707                 bpf_map_put(old_map);
4708 out_put:
4709         bpf_map_put(new_map);
4710         return ret;
4711 }
4712
4713 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4714
4715 static int link_update(union bpf_attr *attr)
4716 {
4717         struct bpf_prog *old_prog = NULL, *new_prog;
4718         struct bpf_link *link;
4719         u32 flags;
4720         int ret;
4721
4722         if (CHECK_ATTR(BPF_LINK_UPDATE))
4723                 return -EINVAL;
4724
4725         flags = attr->link_update.flags;
4726         if (flags & ~BPF_F_REPLACE)
4727                 return -EINVAL;
4728
4729         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4730         if (IS_ERR(link))
4731                 return PTR_ERR(link);
4732
4733         if (link->ops->update_map) {
4734                 ret = link_update_map(link, attr);
4735                 goto out_put_link;
4736         }
4737
4738         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4739         if (IS_ERR(new_prog)) {
4740                 ret = PTR_ERR(new_prog);
4741                 goto out_put_link;
4742         }
4743
4744         if (flags & BPF_F_REPLACE) {
4745                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4746                 if (IS_ERR(old_prog)) {
4747                         ret = PTR_ERR(old_prog);
4748                         old_prog = NULL;
4749                         goto out_put_progs;
4750                 }
4751         } else if (attr->link_update.old_prog_fd) {
4752                 ret = -EINVAL;
4753                 goto out_put_progs;
4754         }
4755
4756         if (link->ops->update_prog)
4757                 ret = link->ops->update_prog(link, new_prog, old_prog);
4758         else
4759                 ret = -EINVAL;
4760
4761 out_put_progs:
4762         if (old_prog)
4763                 bpf_prog_put(old_prog);
4764         if (ret)
4765                 bpf_prog_put(new_prog);
4766 out_put_link:
4767         bpf_link_put(link);
4768         return ret;
4769 }
4770
4771 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4772
4773 static int link_detach(union bpf_attr *attr)
4774 {
4775         struct bpf_link *link;
4776         int ret;
4777
4778         if (CHECK_ATTR(BPF_LINK_DETACH))
4779                 return -EINVAL;
4780
4781         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4782         if (IS_ERR(link))
4783                 return PTR_ERR(link);
4784
4785         if (link->ops->detach)
4786                 ret = link->ops->detach(link);
4787         else
4788                 ret = -EOPNOTSUPP;
4789
4790         bpf_link_put(link);
4791         return ret;
4792 }
4793
4794 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4795 {
4796         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4797 }
4798
4799 struct bpf_link *bpf_link_by_id(u32 id)
4800 {
4801         struct bpf_link *link;
4802
4803         if (!id)
4804                 return ERR_PTR(-ENOENT);
4805
4806         spin_lock_bh(&link_idr_lock);
4807         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4808         link = idr_find(&link_idr, id);
4809         if (link) {
4810                 if (link->id)
4811                         link = bpf_link_inc_not_zero(link);
4812                 else
4813                         link = ERR_PTR(-EAGAIN);
4814         } else {
4815                 link = ERR_PTR(-ENOENT);
4816         }
4817         spin_unlock_bh(&link_idr_lock);
4818         return link;
4819 }
4820
4821 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4822 {
4823         struct bpf_link *link;
4824
4825         spin_lock_bh(&link_idr_lock);
4826 again:
4827         link = idr_get_next(&link_idr, id);
4828         if (link) {
4829                 link = bpf_link_inc_not_zero(link);
4830                 if (IS_ERR(link)) {
4831                         (*id)++;
4832                         goto again;
4833                 }
4834         }
4835         spin_unlock_bh(&link_idr_lock);
4836
4837         return link;
4838 }
4839
4840 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4841
4842 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4843 {
4844         struct bpf_link *link;
4845         u32 id = attr->link_id;
4846         int fd;
4847
4848         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4849                 return -EINVAL;
4850
4851         if (!capable(CAP_SYS_ADMIN))
4852                 return -EPERM;
4853
4854         link = bpf_link_by_id(id);
4855         if (IS_ERR(link))
4856                 return PTR_ERR(link);
4857
4858         fd = bpf_link_new_fd(link);
4859         if (fd < 0)
4860                 bpf_link_put(link);
4861
4862         return fd;
4863 }
4864
4865 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4866
4867 static int bpf_stats_release(struct inode *inode, struct file *file)
4868 {
4869         mutex_lock(&bpf_stats_enabled_mutex);
4870         static_key_slow_dec(&bpf_stats_enabled_key.key);
4871         mutex_unlock(&bpf_stats_enabled_mutex);
4872         return 0;
4873 }
4874
4875 static const struct file_operations bpf_stats_fops = {
4876         .release = bpf_stats_release,
4877 };
4878
4879 static int bpf_enable_runtime_stats(void)
4880 {
4881         int fd;
4882
4883         mutex_lock(&bpf_stats_enabled_mutex);
4884
4885         /* Set a very high limit to avoid overflow */
4886         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4887                 mutex_unlock(&bpf_stats_enabled_mutex);
4888                 return -EBUSY;
4889         }
4890
4891         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4892         if (fd >= 0)
4893                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4894
4895         mutex_unlock(&bpf_stats_enabled_mutex);
4896         return fd;
4897 }
4898
4899 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4900
4901 static int bpf_enable_stats(union bpf_attr *attr)
4902 {
4903
4904         if (CHECK_ATTR(BPF_ENABLE_STATS))
4905                 return -EINVAL;
4906
4907         if (!capable(CAP_SYS_ADMIN))
4908                 return -EPERM;
4909
4910         switch (attr->enable_stats.type) {
4911         case BPF_STATS_RUN_TIME:
4912                 return bpf_enable_runtime_stats();
4913         default:
4914                 break;
4915         }
4916         return -EINVAL;
4917 }
4918
4919 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4920
4921 static int bpf_iter_create(union bpf_attr *attr)
4922 {
4923         struct bpf_link *link;
4924         int err;
4925
4926         if (CHECK_ATTR(BPF_ITER_CREATE))
4927                 return -EINVAL;
4928
4929         if (attr->iter_create.flags)
4930                 return -EINVAL;
4931
4932         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4933         if (IS_ERR(link))
4934                 return PTR_ERR(link);
4935
4936         err = bpf_iter_new_fd(link);
4937         bpf_link_put(link);
4938
4939         return err;
4940 }
4941
4942 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4943
4944 static int bpf_prog_bind_map(union bpf_attr *attr)
4945 {
4946         struct bpf_prog *prog;
4947         struct bpf_map *map;
4948         struct bpf_map **used_maps_old, **used_maps_new;
4949         int i, ret = 0;
4950
4951         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4952                 return -EINVAL;
4953
4954         if (attr->prog_bind_map.flags)
4955                 return -EINVAL;
4956
4957         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4958         if (IS_ERR(prog))
4959                 return PTR_ERR(prog);
4960
4961         map = bpf_map_get(attr->prog_bind_map.map_fd);
4962         if (IS_ERR(map)) {
4963                 ret = PTR_ERR(map);
4964                 goto out_prog_put;
4965         }
4966
4967         mutex_lock(&prog->aux->used_maps_mutex);
4968
4969         used_maps_old = prog->aux->used_maps;
4970
4971         for (i = 0; i < prog->aux->used_map_cnt; i++)
4972                 if (used_maps_old[i] == map) {
4973                         bpf_map_put(map);
4974                         goto out_unlock;
4975                 }
4976
4977         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4978                                       sizeof(used_maps_new[0]),
4979                                       GFP_KERNEL);
4980         if (!used_maps_new) {
4981                 ret = -ENOMEM;
4982                 goto out_unlock;
4983         }
4984
4985         memcpy(used_maps_new, used_maps_old,
4986                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4987         used_maps_new[prog->aux->used_map_cnt] = map;
4988
4989         prog->aux->used_map_cnt++;
4990         prog->aux->used_maps = used_maps_new;
4991
4992         kfree(used_maps_old);
4993
4994 out_unlock:
4995         mutex_unlock(&prog->aux->used_maps_mutex);
4996
4997         if (ret)
4998                 bpf_map_put(map);
4999 out_prog_put:
5000         bpf_prog_put(prog);
5001         return ret;
5002 }
5003
5004 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5005 {
5006         union bpf_attr attr;
5007         bool capable;
5008         int err;
5009
5010         capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
5011
5012         /* Intent here is for unprivileged_bpf_disabled to block key object
5013          * creation commands for unprivileged users; other actions depend
5014          * of fd availability and access to bpffs, so are dependent on
5015          * object creation success.  Capabilities are later verified for
5016          * operations such as load and map create, so even with unprivileged
5017          * BPF disabled, capability checks are still carried out for these
5018          * and other operations.
5019          */
5020         if (!capable &&
5021             (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
5022                 return -EPERM;
5023
5024         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5025         if (err)
5026                 return err;
5027         size = min_t(u32, size, sizeof(attr));
5028
5029         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5030         memset(&attr, 0, sizeof(attr));
5031         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5032                 return -EFAULT;
5033
5034         err = security_bpf(cmd, &attr, size);
5035         if (err < 0)
5036                 return err;
5037
5038         switch (cmd) {
5039         case BPF_MAP_CREATE:
5040                 err = map_create(&attr);
5041                 break;
5042         case BPF_MAP_LOOKUP_ELEM:
5043                 err = map_lookup_elem(&attr);
5044                 break;
5045         case BPF_MAP_UPDATE_ELEM:
5046                 err = map_update_elem(&attr, uattr);
5047                 break;
5048         case BPF_MAP_DELETE_ELEM:
5049                 err = map_delete_elem(&attr, uattr);
5050                 break;
5051         case BPF_MAP_GET_NEXT_KEY:
5052                 err = map_get_next_key(&attr);
5053                 break;
5054         case BPF_MAP_FREEZE:
5055                 err = map_freeze(&attr);
5056                 break;
5057         case BPF_PROG_LOAD:
5058                 err = bpf_prog_load(&attr, uattr, size);
5059                 break;
5060         case BPF_OBJ_PIN:
5061                 err = bpf_obj_pin(&attr);
5062                 break;
5063         case BPF_OBJ_GET:
5064                 err = bpf_obj_get(&attr);
5065                 break;
5066         case BPF_PROG_ATTACH:
5067                 err = bpf_prog_attach(&attr);
5068                 break;
5069         case BPF_PROG_DETACH:
5070                 err = bpf_prog_detach(&attr);
5071                 break;
5072         case BPF_PROG_QUERY:
5073                 err = bpf_prog_query(&attr, uattr.user);
5074                 break;
5075         case BPF_PROG_TEST_RUN:
5076                 err = bpf_prog_test_run(&attr, uattr.user);
5077                 break;
5078         case BPF_PROG_GET_NEXT_ID:
5079                 err = bpf_obj_get_next_id(&attr, uattr.user,
5080                                           &prog_idr, &prog_idr_lock);
5081                 break;
5082         case BPF_MAP_GET_NEXT_ID:
5083                 err = bpf_obj_get_next_id(&attr, uattr.user,
5084                                           &map_idr, &map_idr_lock);
5085                 break;
5086         case BPF_BTF_GET_NEXT_ID:
5087                 err = bpf_obj_get_next_id(&attr, uattr.user,
5088                                           &btf_idr, &btf_idr_lock);
5089                 break;
5090         case BPF_PROG_GET_FD_BY_ID:
5091                 err = bpf_prog_get_fd_by_id(&attr);
5092                 break;
5093         case BPF_MAP_GET_FD_BY_ID:
5094                 err = bpf_map_get_fd_by_id(&attr);
5095                 break;
5096         case BPF_OBJ_GET_INFO_BY_FD:
5097                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5098                 break;
5099         case BPF_RAW_TRACEPOINT_OPEN:
5100                 err = bpf_raw_tracepoint_open(&attr);
5101                 break;
5102         case BPF_BTF_LOAD:
5103                 err = bpf_btf_load(&attr, uattr, size);
5104                 break;
5105         case BPF_BTF_GET_FD_BY_ID:
5106                 err = bpf_btf_get_fd_by_id(&attr);
5107                 break;
5108         case BPF_TASK_FD_QUERY:
5109                 err = bpf_task_fd_query(&attr, uattr.user);
5110                 break;
5111         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5112                 err = map_lookup_and_delete_elem(&attr);
5113                 break;
5114         case BPF_MAP_LOOKUP_BATCH:
5115                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5116                 break;
5117         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5118                 err = bpf_map_do_batch(&attr, uattr.user,
5119                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5120                 break;
5121         case BPF_MAP_UPDATE_BATCH:
5122                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5123                 break;
5124         case BPF_MAP_DELETE_BATCH:
5125                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5126                 break;
5127         case BPF_LINK_CREATE:
5128                 err = link_create(&attr, uattr);
5129                 break;
5130         case BPF_LINK_UPDATE:
5131                 err = link_update(&attr);
5132                 break;
5133         case BPF_LINK_GET_FD_BY_ID:
5134                 err = bpf_link_get_fd_by_id(&attr);
5135                 break;
5136         case BPF_LINK_GET_NEXT_ID:
5137                 err = bpf_obj_get_next_id(&attr, uattr.user,
5138                                           &link_idr, &link_idr_lock);
5139                 break;
5140         case BPF_ENABLE_STATS:
5141                 err = bpf_enable_stats(&attr);
5142                 break;
5143         case BPF_ITER_CREATE:
5144                 err = bpf_iter_create(&attr);
5145                 break;
5146         case BPF_LINK_DETACH:
5147                 err = link_detach(&attr);
5148                 break;
5149         case BPF_PROG_BIND_MAP:
5150                 err = bpf_prog_bind_map(&attr);
5151                 break;
5152         default:
5153                 err = -EINVAL;
5154                 break;
5155         }
5156
5157         return err;
5158 }
5159
5160 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5161 {
5162         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5163 }
5164
5165 static bool syscall_prog_is_valid_access(int off, int size,
5166                                          enum bpf_access_type type,
5167                                          const struct bpf_prog *prog,
5168                                          struct bpf_insn_access_aux *info)
5169 {
5170         if (off < 0 || off >= U16_MAX)
5171                 return false;
5172         if (off % size != 0)
5173                 return false;
5174         return true;
5175 }
5176
5177 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5178 {
5179         switch (cmd) {
5180         case BPF_MAP_CREATE:
5181         case BPF_MAP_DELETE_ELEM:
5182         case BPF_MAP_UPDATE_ELEM:
5183         case BPF_MAP_FREEZE:
5184         case BPF_MAP_GET_FD_BY_ID:
5185         case BPF_PROG_LOAD:
5186         case BPF_BTF_LOAD:
5187         case BPF_LINK_CREATE:
5188         case BPF_RAW_TRACEPOINT_OPEN:
5189                 break;
5190         default:
5191                 return -EINVAL;
5192         }
5193         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5194 }
5195
5196
5197 /* To shut up -Wmissing-prototypes.
5198  * This function is used by the kernel light skeleton
5199  * to load bpf programs when modules are loaded or during kernel boot.
5200  * See tools/lib/bpf/skel_internal.h
5201  */
5202 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5203
5204 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5205 {
5206         struct bpf_prog * __maybe_unused prog;
5207         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5208
5209         switch (cmd) {
5210 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5211         case BPF_PROG_TEST_RUN:
5212                 if (attr->test.data_in || attr->test.data_out ||
5213                     attr->test.ctx_out || attr->test.duration ||
5214                     attr->test.repeat || attr->test.flags)
5215                         return -EINVAL;
5216
5217                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5218                 if (IS_ERR(prog))
5219                         return PTR_ERR(prog);
5220
5221                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5222                     attr->test.ctx_size_in > U16_MAX) {
5223                         bpf_prog_put(prog);
5224                         return -EINVAL;
5225                 }
5226
5227                 run_ctx.bpf_cookie = 0;
5228                 run_ctx.saved_run_ctx = NULL;
5229                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5230                         /* recursion detected */
5231                         bpf_prog_put(prog);
5232                         return -EBUSY;
5233                 }
5234                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5235                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5236                                                 &run_ctx);
5237                 bpf_prog_put(prog);
5238                 return 0;
5239 #endif
5240         default:
5241                 return ____bpf_sys_bpf(cmd, attr, size);
5242         }
5243 }
5244 EXPORT_SYMBOL(kern_sys_bpf);
5245
5246 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5247         .func           = bpf_sys_bpf,
5248         .gpl_only       = false,
5249         .ret_type       = RET_INTEGER,
5250         .arg1_type      = ARG_ANYTHING,
5251         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5252         .arg3_type      = ARG_CONST_SIZE,
5253 };
5254
5255 const struct bpf_func_proto * __weak
5256 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5257 {
5258         return bpf_base_func_proto(func_id);
5259 }
5260
5261 BPF_CALL_1(bpf_sys_close, u32, fd)
5262 {
5263         /* When bpf program calls this helper there should not be
5264          * an fdget() without matching completed fdput().
5265          * This helper is allowed in the following callchain only:
5266          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5267          */
5268         return close_fd(fd);
5269 }
5270
5271 static const struct bpf_func_proto bpf_sys_close_proto = {
5272         .func           = bpf_sys_close,
5273         .gpl_only       = false,
5274         .ret_type       = RET_INTEGER,
5275         .arg1_type      = ARG_ANYTHING,
5276 };
5277
5278 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5279 {
5280         if (flags)
5281                 return -EINVAL;
5282
5283         if (name_sz <= 1 || name[name_sz - 1])
5284                 return -EINVAL;
5285
5286         if (!bpf_dump_raw_ok(current_cred()))
5287                 return -EPERM;
5288
5289         *res = kallsyms_lookup_name(name);
5290         return *res ? 0 : -ENOENT;
5291 }
5292
5293 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5294         .func           = bpf_kallsyms_lookup_name,
5295         .gpl_only       = false,
5296         .ret_type       = RET_INTEGER,
5297         .arg1_type      = ARG_PTR_TO_MEM,
5298         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5299         .arg3_type      = ARG_ANYTHING,
5300         .arg4_type      = ARG_PTR_TO_LONG,
5301 };
5302
5303 static const struct bpf_func_proto *
5304 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5305 {
5306         switch (func_id) {
5307         case BPF_FUNC_sys_bpf:
5308                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5309         case BPF_FUNC_btf_find_by_name_kind:
5310                 return &bpf_btf_find_by_name_kind_proto;
5311         case BPF_FUNC_sys_close:
5312                 return &bpf_sys_close_proto;
5313         case BPF_FUNC_kallsyms_lookup_name:
5314                 return &bpf_kallsyms_lookup_name_proto;
5315         default:
5316                 return tracing_prog_func_proto(func_id, prog);
5317         }
5318 }
5319
5320 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5321         .get_func_proto  = syscall_prog_func_proto,
5322         .is_valid_access = syscall_prog_is_valid_access,
5323 };
5324
5325 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5326         .test_run = bpf_prog_test_run_syscall,
5327 };
5328
5329 #ifdef CONFIG_SYSCTL
5330 static int bpf_stats_handler(struct ctl_table *table, int write,
5331                              void *buffer, size_t *lenp, loff_t *ppos)
5332 {
5333         struct static_key *key = (struct static_key *)table->data;
5334         static int saved_val;
5335         int val, ret;
5336         struct ctl_table tmp = {
5337                 .data   = &val,
5338                 .maxlen = sizeof(val),
5339                 .mode   = table->mode,
5340                 .extra1 = SYSCTL_ZERO,
5341                 .extra2 = SYSCTL_ONE,
5342         };
5343
5344         if (write && !capable(CAP_SYS_ADMIN))
5345                 return -EPERM;
5346
5347         mutex_lock(&bpf_stats_enabled_mutex);
5348         val = saved_val;
5349         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5350         if (write && !ret && val != saved_val) {
5351                 if (val)
5352                         static_key_slow_inc(key);
5353                 else
5354                         static_key_slow_dec(key);
5355                 saved_val = val;
5356         }
5357         mutex_unlock(&bpf_stats_enabled_mutex);
5358         return ret;
5359 }
5360
5361 void __weak unpriv_ebpf_notify(int new_state)
5362 {
5363 }
5364
5365 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5366                               void *buffer, size_t *lenp, loff_t *ppos)
5367 {
5368         int ret, unpriv_enable = *(int *)table->data;
5369         bool locked_state = unpriv_enable == 1;
5370         struct ctl_table tmp = *table;
5371
5372         if (write && !capable(CAP_SYS_ADMIN))
5373                 return -EPERM;
5374
5375         tmp.data = &unpriv_enable;
5376         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5377         if (write && !ret) {
5378                 if (locked_state && unpriv_enable != 1)
5379                         return -EPERM;
5380                 *(int *)table->data = unpriv_enable;
5381         }
5382
5383         unpriv_ebpf_notify(unpriv_enable);
5384
5385         return ret;
5386 }
5387
5388 static struct ctl_table bpf_syscall_table[] = {
5389         {
5390                 .procname       = "unprivileged_bpf_disabled",
5391                 .data           = &sysctl_unprivileged_bpf_disabled,
5392                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5393                 .mode           = 0644,
5394                 .proc_handler   = bpf_unpriv_handler,
5395                 .extra1         = SYSCTL_ZERO,
5396                 .extra2         = SYSCTL_TWO,
5397         },
5398         {
5399                 .procname       = "bpf_stats_enabled",
5400                 .data           = &bpf_stats_enabled_key.key,
5401                 .mode           = 0644,
5402                 .proc_handler   = bpf_stats_handler,
5403         },
5404         { }
5405 };
5406
5407 static int __init bpf_syscall_sysctl_init(void)
5408 {
5409         register_sysctl_init("kernel", bpf_syscall_table);
5410         return 0;
5411 }
5412 late_initcall(bpf_syscall_sysctl_init);
5413 #endif /* CONFIG_SYSCTL */