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         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1935                 fdput(f);
1936                 return -EPERM;
1937         }
1938
1939         mutex_lock(&map->freeze_mutex);
1940         if (bpf_map_write_active(map)) {
1941                 err = -EBUSY;
1942                 goto err_put;
1943         }
1944         if (READ_ONCE(map->frozen)) {
1945                 err = -EBUSY;
1946                 goto err_put;
1947         }
1948
1949         WRITE_ONCE(map->frozen, true);
1950 err_put:
1951         mutex_unlock(&map->freeze_mutex);
1952         fdput(f);
1953         return err;
1954 }
1955
1956 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1957 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1958         [_id] = & _name ## _prog_ops,
1959 #define BPF_MAP_TYPE(_id, _ops)
1960 #define BPF_LINK_TYPE(_id, _name)
1961 #include <linux/bpf_types.h>
1962 #undef BPF_PROG_TYPE
1963 #undef BPF_MAP_TYPE
1964 #undef BPF_LINK_TYPE
1965 };
1966
1967 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1968 {
1969         const struct bpf_prog_ops *ops;
1970
1971         if (type >= ARRAY_SIZE(bpf_prog_types))
1972                 return -EINVAL;
1973         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1974         ops = bpf_prog_types[type];
1975         if (!ops)
1976                 return -EINVAL;
1977
1978         if (!bpf_prog_is_offloaded(prog->aux))
1979                 prog->aux->ops = ops;
1980         else
1981                 prog->aux->ops = &bpf_offload_prog_ops;
1982         prog->type = type;
1983         return 0;
1984 }
1985
1986 enum bpf_audit {
1987         BPF_AUDIT_LOAD,
1988         BPF_AUDIT_UNLOAD,
1989         BPF_AUDIT_MAX,
1990 };
1991
1992 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1993         [BPF_AUDIT_LOAD]   = "LOAD",
1994         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1995 };
1996
1997 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1998 {
1999         struct audit_context *ctx = NULL;
2000         struct audit_buffer *ab;
2001
2002         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2003                 return;
2004         if (audit_enabled == AUDIT_OFF)
2005                 return;
2006         if (!in_irq() && !irqs_disabled())
2007                 ctx = audit_context();
2008         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2009         if (unlikely(!ab))
2010                 return;
2011         audit_log_format(ab, "prog-id=%u op=%s",
2012                          prog->aux->id, bpf_audit_str[op]);
2013         audit_log_end(ab);
2014 }
2015
2016 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2017 {
2018         int id;
2019
2020         idr_preload(GFP_KERNEL);
2021         spin_lock_bh(&prog_idr_lock);
2022         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2023         if (id > 0)
2024                 prog->aux->id = id;
2025         spin_unlock_bh(&prog_idr_lock);
2026         idr_preload_end();
2027
2028         /* id is in [1, INT_MAX) */
2029         if (WARN_ON_ONCE(!id))
2030                 return -ENOSPC;
2031
2032         return id > 0 ? 0 : id;
2033 }
2034
2035 void bpf_prog_free_id(struct bpf_prog *prog)
2036 {
2037         unsigned long flags;
2038
2039         /* cBPF to eBPF migrations are currently not in the idr store.
2040          * Offloaded programs are removed from the store when their device
2041          * disappears - even if someone grabs an fd to them they are unusable,
2042          * simply waiting for refcnt to drop to be freed.
2043          */
2044         if (!prog->aux->id)
2045                 return;
2046
2047         spin_lock_irqsave(&prog_idr_lock, flags);
2048         idr_remove(&prog_idr, prog->aux->id);
2049         prog->aux->id = 0;
2050         spin_unlock_irqrestore(&prog_idr_lock, flags);
2051 }
2052
2053 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2054 {
2055         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2056
2057         kvfree(aux->func_info);
2058         kfree(aux->func_info_aux);
2059         free_uid(aux->user);
2060         security_bpf_prog_free(aux);
2061         bpf_prog_free(aux->prog);
2062 }
2063
2064 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2065 {
2066         bpf_prog_kallsyms_del_all(prog);
2067         btf_put(prog->aux->btf);
2068         module_put(prog->aux->mod);
2069         kvfree(prog->aux->jited_linfo);
2070         kvfree(prog->aux->linfo);
2071         kfree(prog->aux->kfunc_tab);
2072         if (prog->aux->attach_btf)
2073                 btf_put(prog->aux->attach_btf);
2074
2075         if (deferred) {
2076                 if (prog->aux->sleepable)
2077                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2078                 else
2079                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2080         } else {
2081                 __bpf_prog_put_rcu(&prog->aux->rcu);
2082         }
2083 }
2084
2085 static void bpf_prog_put_deferred(struct work_struct *work)
2086 {
2087         struct bpf_prog_aux *aux;
2088         struct bpf_prog *prog;
2089
2090         aux = container_of(work, struct bpf_prog_aux, work);
2091         prog = aux->prog;
2092         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2093         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2094         bpf_prog_free_id(prog);
2095         __bpf_prog_put_noref(prog, true);
2096 }
2097
2098 static void __bpf_prog_put(struct bpf_prog *prog)
2099 {
2100         struct bpf_prog_aux *aux = prog->aux;
2101
2102         if (atomic64_dec_and_test(&aux->refcnt)) {
2103                 if (in_irq() || irqs_disabled()) {
2104                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2105                         schedule_work(&aux->work);
2106                 } else {
2107                         bpf_prog_put_deferred(&aux->work);
2108                 }
2109         }
2110 }
2111
2112 void bpf_prog_put(struct bpf_prog *prog)
2113 {
2114         __bpf_prog_put(prog);
2115 }
2116 EXPORT_SYMBOL_GPL(bpf_prog_put);
2117
2118 static int bpf_prog_release(struct inode *inode, struct file *filp)
2119 {
2120         struct bpf_prog *prog = filp->private_data;
2121
2122         bpf_prog_put(prog);
2123         return 0;
2124 }
2125
2126 struct bpf_prog_kstats {
2127         u64 nsecs;
2128         u64 cnt;
2129         u64 misses;
2130 };
2131
2132 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2133 {
2134         struct bpf_prog_stats *stats;
2135         unsigned int flags;
2136
2137         stats = this_cpu_ptr(prog->stats);
2138         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2139         u64_stats_inc(&stats->misses);
2140         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2141 }
2142
2143 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2144                                struct bpf_prog_kstats *stats)
2145 {
2146         u64 nsecs = 0, cnt = 0, misses = 0;
2147         int cpu;
2148
2149         for_each_possible_cpu(cpu) {
2150                 const struct bpf_prog_stats *st;
2151                 unsigned int start;
2152                 u64 tnsecs, tcnt, tmisses;
2153
2154                 st = per_cpu_ptr(prog->stats, cpu);
2155                 do {
2156                         start = u64_stats_fetch_begin(&st->syncp);
2157                         tnsecs = u64_stats_read(&st->nsecs);
2158                         tcnt = u64_stats_read(&st->cnt);
2159                         tmisses = u64_stats_read(&st->misses);
2160                 } while (u64_stats_fetch_retry(&st->syncp, start));
2161                 nsecs += tnsecs;
2162                 cnt += tcnt;
2163                 misses += tmisses;
2164         }
2165         stats->nsecs = nsecs;
2166         stats->cnt = cnt;
2167         stats->misses = misses;
2168 }
2169
2170 #ifdef CONFIG_PROC_FS
2171 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2172 {
2173         const struct bpf_prog *prog = filp->private_data;
2174         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2175         struct bpf_prog_kstats stats;
2176
2177         bpf_prog_get_stats(prog, &stats);
2178         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2179         seq_printf(m,
2180                    "prog_type:\t%u\n"
2181                    "prog_jited:\t%u\n"
2182                    "prog_tag:\t%s\n"
2183                    "memlock:\t%llu\n"
2184                    "prog_id:\t%u\n"
2185                    "run_time_ns:\t%llu\n"
2186                    "run_cnt:\t%llu\n"
2187                    "recursion_misses:\t%llu\n"
2188                    "verified_insns:\t%u\n",
2189                    prog->type,
2190                    prog->jited,
2191                    prog_tag,
2192                    prog->pages * 1ULL << PAGE_SHIFT,
2193                    prog->aux->id,
2194                    stats.nsecs,
2195                    stats.cnt,
2196                    stats.misses,
2197                    prog->aux->verified_insns);
2198 }
2199 #endif
2200
2201 const struct file_operations bpf_prog_fops = {
2202 #ifdef CONFIG_PROC_FS
2203         .show_fdinfo    = bpf_prog_show_fdinfo,
2204 #endif
2205         .release        = bpf_prog_release,
2206         .read           = bpf_dummy_read,
2207         .write          = bpf_dummy_write,
2208 };
2209
2210 int bpf_prog_new_fd(struct bpf_prog *prog)
2211 {
2212         int ret;
2213
2214         ret = security_bpf_prog(prog);
2215         if (ret < 0)
2216                 return ret;
2217
2218         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2219                                 O_RDWR | O_CLOEXEC);
2220 }
2221
2222 static struct bpf_prog *____bpf_prog_get(struct fd f)
2223 {
2224         if (!f.file)
2225                 return ERR_PTR(-EBADF);
2226         if (f.file->f_op != &bpf_prog_fops) {
2227                 fdput(f);
2228                 return ERR_PTR(-EINVAL);
2229         }
2230
2231         return f.file->private_data;
2232 }
2233
2234 void bpf_prog_add(struct bpf_prog *prog, int i)
2235 {
2236         atomic64_add(i, &prog->aux->refcnt);
2237 }
2238 EXPORT_SYMBOL_GPL(bpf_prog_add);
2239
2240 void bpf_prog_sub(struct bpf_prog *prog, int i)
2241 {
2242         /* Only to be used for undoing previous bpf_prog_add() in some
2243          * error path. We still know that another entity in our call
2244          * path holds a reference to the program, thus atomic_sub() can
2245          * be safely used in such cases!
2246          */
2247         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2248 }
2249 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2250
2251 void bpf_prog_inc(struct bpf_prog *prog)
2252 {
2253         atomic64_inc(&prog->aux->refcnt);
2254 }
2255 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2256
2257 /* prog_idr_lock should have been held */
2258 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2259 {
2260         int refold;
2261
2262         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2263
2264         if (!refold)
2265                 return ERR_PTR(-ENOENT);
2266
2267         return prog;
2268 }
2269 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2270
2271 bool bpf_prog_get_ok(struct bpf_prog *prog,
2272                             enum bpf_prog_type *attach_type, bool attach_drv)
2273 {
2274         /* not an attachment, just a refcount inc, always allow */
2275         if (!attach_type)
2276                 return true;
2277
2278         if (prog->type != *attach_type)
2279                 return false;
2280         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2281                 return false;
2282
2283         return true;
2284 }
2285
2286 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2287                                        bool attach_drv)
2288 {
2289         struct fd f = fdget(ufd);
2290         struct bpf_prog *prog;
2291
2292         prog = ____bpf_prog_get(f);
2293         if (IS_ERR(prog))
2294                 return prog;
2295         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2296                 prog = ERR_PTR(-EINVAL);
2297                 goto out;
2298         }
2299
2300         bpf_prog_inc(prog);
2301 out:
2302         fdput(f);
2303         return prog;
2304 }
2305
2306 struct bpf_prog *bpf_prog_get(u32 ufd)
2307 {
2308         return __bpf_prog_get(ufd, NULL, false);
2309 }
2310
2311 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2312                                        bool attach_drv)
2313 {
2314         return __bpf_prog_get(ufd, &type, attach_drv);
2315 }
2316 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2317
2318 /* Initially all BPF programs could be loaded w/o specifying
2319  * expected_attach_type. Later for some of them specifying expected_attach_type
2320  * at load time became required so that program could be validated properly.
2321  * Programs of types that are allowed to be loaded both w/ and w/o (for
2322  * backward compatibility) expected_attach_type, should have the default attach
2323  * type assigned to expected_attach_type for the latter case, so that it can be
2324  * validated later at attach time.
2325  *
2326  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2327  * prog type requires it but has some attach types that have to be backward
2328  * compatible.
2329  */
2330 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2331 {
2332         switch (attr->prog_type) {
2333         case BPF_PROG_TYPE_CGROUP_SOCK:
2334                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2335                  * exist so checking for non-zero is the way to go here.
2336                  */
2337                 if (!attr->expected_attach_type)
2338                         attr->expected_attach_type =
2339                                 BPF_CGROUP_INET_SOCK_CREATE;
2340                 break;
2341         case BPF_PROG_TYPE_SK_REUSEPORT:
2342                 if (!attr->expected_attach_type)
2343                         attr->expected_attach_type =
2344                                 BPF_SK_REUSEPORT_SELECT;
2345                 break;
2346         }
2347 }
2348
2349 static int
2350 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2351                            enum bpf_attach_type expected_attach_type,
2352                            struct btf *attach_btf, u32 btf_id,
2353                            struct bpf_prog *dst_prog)
2354 {
2355         if (btf_id) {
2356                 if (btf_id > BTF_MAX_TYPE)
2357                         return -EINVAL;
2358
2359                 if (!attach_btf && !dst_prog)
2360                         return -EINVAL;
2361
2362                 switch (prog_type) {
2363                 case BPF_PROG_TYPE_TRACING:
2364                 case BPF_PROG_TYPE_LSM:
2365                 case BPF_PROG_TYPE_STRUCT_OPS:
2366                 case BPF_PROG_TYPE_EXT:
2367                         break;
2368                 default:
2369                         return -EINVAL;
2370                 }
2371         }
2372
2373         if (attach_btf && (!btf_id || dst_prog))
2374                 return -EINVAL;
2375
2376         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2377             prog_type != BPF_PROG_TYPE_EXT)
2378                 return -EINVAL;
2379
2380         switch (prog_type) {
2381         case BPF_PROG_TYPE_CGROUP_SOCK:
2382                 switch (expected_attach_type) {
2383                 case BPF_CGROUP_INET_SOCK_CREATE:
2384                 case BPF_CGROUP_INET_SOCK_RELEASE:
2385                 case BPF_CGROUP_INET4_POST_BIND:
2386                 case BPF_CGROUP_INET6_POST_BIND:
2387                         return 0;
2388                 default:
2389                         return -EINVAL;
2390                 }
2391         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2392                 switch (expected_attach_type) {
2393                 case BPF_CGROUP_INET4_BIND:
2394                 case BPF_CGROUP_INET6_BIND:
2395                 case BPF_CGROUP_INET4_CONNECT:
2396                 case BPF_CGROUP_INET6_CONNECT:
2397                 case BPF_CGROUP_INET4_GETPEERNAME:
2398                 case BPF_CGROUP_INET6_GETPEERNAME:
2399                 case BPF_CGROUP_INET4_GETSOCKNAME:
2400                 case BPF_CGROUP_INET6_GETSOCKNAME:
2401                 case BPF_CGROUP_UDP4_SENDMSG:
2402                 case BPF_CGROUP_UDP6_SENDMSG:
2403                 case BPF_CGROUP_UDP4_RECVMSG:
2404                 case BPF_CGROUP_UDP6_RECVMSG:
2405                         return 0;
2406                 default:
2407                         return -EINVAL;
2408                 }
2409         case BPF_PROG_TYPE_CGROUP_SKB:
2410                 switch (expected_attach_type) {
2411                 case BPF_CGROUP_INET_INGRESS:
2412                 case BPF_CGROUP_INET_EGRESS:
2413                         return 0;
2414                 default:
2415                         return -EINVAL;
2416                 }
2417         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2418                 switch (expected_attach_type) {
2419                 case BPF_CGROUP_SETSOCKOPT:
2420                 case BPF_CGROUP_GETSOCKOPT:
2421                         return 0;
2422                 default:
2423                         return -EINVAL;
2424                 }
2425         case BPF_PROG_TYPE_SK_LOOKUP:
2426                 if (expected_attach_type == BPF_SK_LOOKUP)
2427                         return 0;
2428                 return -EINVAL;
2429         case BPF_PROG_TYPE_SK_REUSEPORT:
2430                 switch (expected_attach_type) {
2431                 case BPF_SK_REUSEPORT_SELECT:
2432                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2433                         return 0;
2434                 default:
2435                         return -EINVAL;
2436                 }
2437         case BPF_PROG_TYPE_NETFILTER:
2438                 if (expected_attach_type == BPF_NETFILTER)
2439                         return 0;
2440                 return -EINVAL;
2441         case BPF_PROG_TYPE_SYSCALL:
2442         case BPF_PROG_TYPE_EXT:
2443                 if (expected_attach_type)
2444                         return -EINVAL;
2445                 fallthrough;
2446         default:
2447                 return 0;
2448         }
2449 }
2450
2451 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2452 {
2453         switch (prog_type) {
2454         case BPF_PROG_TYPE_SCHED_CLS:
2455         case BPF_PROG_TYPE_SCHED_ACT:
2456         case BPF_PROG_TYPE_XDP:
2457         case BPF_PROG_TYPE_LWT_IN:
2458         case BPF_PROG_TYPE_LWT_OUT:
2459         case BPF_PROG_TYPE_LWT_XMIT:
2460         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2461         case BPF_PROG_TYPE_SK_SKB:
2462         case BPF_PROG_TYPE_SK_MSG:
2463         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2464         case BPF_PROG_TYPE_CGROUP_DEVICE:
2465         case BPF_PROG_TYPE_CGROUP_SOCK:
2466         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2467         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2468         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2469         case BPF_PROG_TYPE_SOCK_OPS:
2470         case BPF_PROG_TYPE_EXT: /* extends any prog */
2471         case BPF_PROG_TYPE_NETFILTER:
2472                 return true;
2473         case BPF_PROG_TYPE_CGROUP_SKB:
2474                 /* always unpriv */
2475         case BPF_PROG_TYPE_SK_REUSEPORT:
2476                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2477         default:
2478                 return false;
2479         }
2480 }
2481
2482 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2483 {
2484         switch (prog_type) {
2485         case BPF_PROG_TYPE_KPROBE:
2486         case BPF_PROG_TYPE_TRACEPOINT:
2487         case BPF_PROG_TYPE_PERF_EVENT:
2488         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2489         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2490         case BPF_PROG_TYPE_TRACING:
2491         case BPF_PROG_TYPE_LSM:
2492         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2493         case BPF_PROG_TYPE_EXT: /* extends any prog */
2494                 return true;
2495         default:
2496                 return false;
2497         }
2498 }
2499
2500 /* last field in 'union bpf_attr' used by this command */
2501 #define BPF_PROG_LOAD_LAST_FIELD log_true_size
2502
2503 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2504 {
2505         enum bpf_prog_type type = attr->prog_type;
2506         struct bpf_prog *prog, *dst_prog = NULL;
2507         struct btf *attach_btf = NULL;
2508         int err;
2509         char license[128];
2510         bool is_gpl;
2511
2512         if (CHECK_ATTR(BPF_PROG_LOAD))
2513                 return -EINVAL;
2514
2515         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2516                                  BPF_F_ANY_ALIGNMENT |
2517                                  BPF_F_TEST_STATE_FREQ |
2518                                  BPF_F_SLEEPABLE |
2519                                  BPF_F_TEST_RND_HI32 |
2520                                  BPF_F_XDP_HAS_FRAGS |
2521                                  BPF_F_XDP_DEV_BOUND_ONLY))
2522                 return -EINVAL;
2523
2524         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2525             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2526             !bpf_capable())
2527                 return -EPERM;
2528
2529         /* copy eBPF program license from user space */
2530         if (strncpy_from_bpfptr(license,
2531                                 make_bpfptr(attr->license, uattr.is_kernel),
2532                                 sizeof(license) - 1) < 0)
2533                 return -EFAULT;
2534         license[sizeof(license) - 1] = 0;
2535
2536         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2537         is_gpl = license_is_gpl_compatible(license);
2538
2539         if (attr->insn_cnt == 0 ||
2540             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2541                 return -E2BIG;
2542         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2543             type != BPF_PROG_TYPE_CGROUP_SKB &&
2544             !bpf_capable())
2545                 return -EPERM;
2546
2547         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2548                 return -EPERM;
2549         if (is_perfmon_prog_type(type) && !perfmon_capable())
2550                 return -EPERM;
2551
2552         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2553          * or btf, we need to check which one it is
2554          */
2555         if (attr->attach_prog_fd) {
2556                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2557                 if (IS_ERR(dst_prog)) {
2558                         dst_prog = NULL;
2559                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2560                         if (IS_ERR(attach_btf))
2561                                 return -EINVAL;
2562                         if (!btf_is_kernel(attach_btf)) {
2563                                 /* attaching through specifying bpf_prog's BTF
2564                                  * objects directly might be supported eventually
2565                                  */
2566                                 btf_put(attach_btf);
2567                                 return -ENOTSUPP;
2568                         }
2569                 }
2570         } else if (attr->attach_btf_id) {
2571                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2572                 attach_btf = bpf_get_btf_vmlinux();
2573                 if (IS_ERR(attach_btf))
2574                         return PTR_ERR(attach_btf);
2575                 if (!attach_btf)
2576                         return -EINVAL;
2577                 btf_get(attach_btf);
2578         }
2579
2580         bpf_prog_load_fixup_attach_type(attr);
2581         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2582                                        attach_btf, attr->attach_btf_id,
2583                                        dst_prog)) {
2584                 if (dst_prog)
2585                         bpf_prog_put(dst_prog);
2586                 if (attach_btf)
2587                         btf_put(attach_btf);
2588                 return -EINVAL;
2589         }
2590
2591         /* plain bpf_prog allocation */
2592         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2593         if (!prog) {
2594                 if (dst_prog)
2595                         bpf_prog_put(dst_prog);
2596                 if (attach_btf)
2597                         btf_put(attach_btf);
2598                 return -ENOMEM;
2599         }
2600
2601         prog->expected_attach_type = attr->expected_attach_type;
2602         prog->aux->attach_btf = attach_btf;
2603         prog->aux->attach_btf_id = attr->attach_btf_id;
2604         prog->aux->dst_prog = dst_prog;
2605         prog->aux->dev_bound = !!attr->prog_ifindex;
2606         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2607         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2608
2609         err = security_bpf_prog_alloc(prog->aux);
2610         if (err)
2611                 goto free_prog;
2612
2613         prog->aux->user = get_current_user();
2614         prog->len = attr->insn_cnt;
2615
2616         err = -EFAULT;
2617         if (copy_from_bpfptr(prog->insns,
2618                              make_bpfptr(attr->insns, uattr.is_kernel),
2619                              bpf_prog_insn_size(prog)) != 0)
2620                 goto free_prog_sec;
2621
2622         prog->orig_prog = NULL;
2623         prog->jited = 0;
2624
2625         atomic64_set(&prog->aux->refcnt, 1);
2626         prog->gpl_compatible = is_gpl ? 1 : 0;
2627
2628         if (bpf_prog_is_dev_bound(prog->aux)) {
2629                 err = bpf_prog_dev_bound_init(prog, attr);
2630                 if (err)
2631                         goto free_prog_sec;
2632         }
2633
2634         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2635             bpf_prog_is_dev_bound(dst_prog->aux)) {
2636                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2637                 if (err)
2638                         goto free_prog_sec;
2639         }
2640
2641         /* find program type: socket_filter vs tracing_filter */
2642         err = find_prog_type(type, prog);
2643         if (err < 0)
2644                 goto free_prog_sec;
2645
2646         prog->aux->load_time = ktime_get_boottime_ns();
2647         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2648                                sizeof(attr->prog_name));
2649         if (err < 0)
2650                 goto free_prog_sec;
2651
2652         /* run eBPF verifier */
2653         err = bpf_check(&prog, attr, uattr, uattr_size);
2654         if (err < 0)
2655                 goto free_used_maps;
2656
2657         prog = bpf_prog_select_runtime(prog, &err);
2658         if (err < 0)
2659                 goto free_used_maps;
2660
2661         err = bpf_prog_alloc_id(prog);
2662         if (err)
2663                 goto free_used_maps;
2664
2665         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2666          * effectively publicly exposed. However, retrieving via
2667          * bpf_prog_get_fd_by_id() will take another reference,
2668          * therefore it cannot be gone underneath us.
2669          *
2670          * Only for the time /after/ successful bpf_prog_new_fd()
2671          * and before returning to userspace, we might just hold
2672          * one reference and any parallel close on that fd could
2673          * rip everything out. Hence, below notifications must
2674          * happen before bpf_prog_new_fd().
2675          *
2676          * Also, any failure handling from this point onwards must
2677          * be using bpf_prog_put() given the program is exposed.
2678          */
2679         bpf_prog_kallsyms_add(prog);
2680         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2681         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2682
2683         err = bpf_prog_new_fd(prog);
2684         if (err < 0)
2685                 bpf_prog_put(prog);
2686         return err;
2687
2688 free_used_maps:
2689         /* In case we have subprogs, we need to wait for a grace
2690          * period before we can tear down JIT memory since symbols
2691          * are already exposed under kallsyms.
2692          */
2693         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2694         return err;
2695 free_prog_sec:
2696         free_uid(prog->aux->user);
2697         security_bpf_prog_free(prog->aux);
2698 free_prog:
2699         if (prog->aux->attach_btf)
2700                 btf_put(prog->aux->attach_btf);
2701         bpf_prog_free(prog);
2702         return err;
2703 }
2704
2705 #define BPF_OBJ_LAST_FIELD path_fd
2706
2707 static int bpf_obj_pin(const union bpf_attr *attr)
2708 {
2709         int path_fd;
2710
2711         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2712                 return -EINVAL;
2713
2714         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2715         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2716                 return -EINVAL;
2717
2718         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2719         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2720                                 u64_to_user_ptr(attr->pathname));
2721 }
2722
2723 static int bpf_obj_get(const union bpf_attr *attr)
2724 {
2725         int path_fd;
2726
2727         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2728             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2729                 return -EINVAL;
2730
2731         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2732         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2733                 return -EINVAL;
2734
2735         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2736         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2737                                 attr->file_flags);
2738 }
2739
2740 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2741                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2742 {
2743         atomic64_set(&link->refcnt, 1);
2744         link->type = type;
2745         link->id = 0;
2746         link->ops = ops;
2747         link->prog = prog;
2748 }
2749
2750 static void bpf_link_free_id(int id)
2751 {
2752         if (!id)
2753                 return;
2754
2755         spin_lock_bh(&link_idr_lock);
2756         idr_remove(&link_idr, id);
2757         spin_unlock_bh(&link_idr_lock);
2758 }
2759
2760 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2761  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2762  * anon_inode's release() call. This helper marksbpf_link as
2763  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2764  * is not decremented, it's the responsibility of a calling code that failed
2765  * to complete bpf_link initialization.
2766  */
2767 void bpf_link_cleanup(struct bpf_link_primer *primer)
2768 {
2769         primer->link->prog = NULL;
2770         bpf_link_free_id(primer->id);
2771         fput(primer->file);
2772         put_unused_fd(primer->fd);
2773 }
2774
2775 void bpf_link_inc(struct bpf_link *link)
2776 {
2777         atomic64_inc(&link->refcnt);
2778 }
2779
2780 /* bpf_link_free is guaranteed to be called from process context */
2781 static void bpf_link_free(struct bpf_link *link)
2782 {
2783         bpf_link_free_id(link->id);
2784         if (link->prog) {
2785                 /* detach BPF program, clean up used resources */
2786                 link->ops->release(link);
2787                 bpf_prog_put(link->prog);
2788         }
2789         /* free bpf_link and its containing memory */
2790         link->ops->dealloc(link);
2791 }
2792
2793 static void bpf_link_put_deferred(struct work_struct *work)
2794 {
2795         struct bpf_link *link = container_of(work, struct bpf_link, work);
2796
2797         bpf_link_free(link);
2798 }
2799
2800 /* bpf_link_put can be called from atomic context, but ensures that resources
2801  * are freed from process context
2802  */
2803 void bpf_link_put(struct bpf_link *link)
2804 {
2805         if (!atomic64_dec_and_test(&link->refcnt))
2806                 return;
2807
2808         if (in_atomic()) {
2809                 INIT_WORK(&link->work, bpf_link_put_deferred);
2810                 schedule_work(&link->work);
2811         } else {
2812                 bpf_link_free(link);
2813         }
2814 }
2815 EXPORT_SYMBOL(bpf_link_put);
2816
2817 static int bpf_link_release(struct inode *inode, struct file *filp)
2818 {
2819         struct bpf_link *link = filp->private_data;
2820
2821         bpf_link_put(link);
2822         return 0;
2823 }
2824
2825 #ifdef CONFIG_PROC_FS
2826 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2827 #define BPF_MAP_TYPE(_id, _ops)
2828 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2829 static const char *bpf_link_type_strs[] = {
2830         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2831 #include <linux/bpf_types.h>
2832 };
2833 #undef BPF_PROG_TYPE
2834 #undef BPF_MAP_TYPE
2835 #undef BPF_LINK_TYPE
2836
2837 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2838 {
2839         const struct bpf_link *link = filp->private_data;
2840         const struct bpf_prog *prog = link->prog;
2841         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2842
2843         seq_printf(m,
2844                    "link_type:\t%s\n"
2845                    "link_id:\t%u\n",
2846                    bpf_link_type_strs[link->type],
2847                    link->id);
2848         if (prog) {
2849                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2850                 seq_printf(m,
2851                            "prog_tag:\t%s\n"
2852                            "prog_id:\t%u\n",
2853                            prog_tag,
2854                            prog->aux->id);
2855         }
2856         if (link->ops->show_fdinfo)
2857                 link->ops->show_fdinfo(link, m);
2858 }
2859 #endif
2860
2861 static const struct file_operations bpf_link_fops = {
2862 #ifdef CONFIG_PROC_FS
2863         .show_fdinfo    = bpf_link_show_fdinfo,
2864 #endif
2865         .release        = bpf_link_release,
2866         .read           = bpf_dummy_read,
2867         .write          = bpf_dummy_write,
2868 };
2869
2870 static int bpf_link_alloc_id(struct bpf_link *link)
2871 {
2872         int id;
2873
2874         idr_preload(GFP_KERNEL);
2875         spin_lock_bh(&link_idr_lock);
2876         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2877         spin_unlock_bh(&link_idr_lock);
2878         idr_preload_end();
2879
2880         return id;
2881 }
2882
2883 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2884  * reserving unused FD and allocating ID from link_idr. This is to be paired
2885  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2886  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2887  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2888  * transient state is passed around in struct bpf_link_primer.
2889  * This is preferred way to create and initialize bpf_link, especially when
2890  * there are complicated and expensive operations in between creating bpf_link
2891  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2892  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2893  * expensive (and potentially failing) roll back operations in a rare case
2894  * that file, FD, or ID can't be allocated.
2895  */
2896 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2897 {
2898         struct file *file;
2899         int fd, id;
2900
2901         fd = get_unused_fd_flags(O_CLOEXEC);
2902         if (fd < 0)
2903                 return fd;
2904
2905
2906         id = bpf_link_alloc_id(link);
2907         if (id < 0) {
2908                 put_unused_fd(fd);
2909                 return id;
2910         }
2911
2912         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2913         if (IS_ERR(file)) {
2914                 bpf_link_free_id(id);
2915                 put_unused_fd(fd);
2916                 return PTR_ERR(file);
2917         }
2918
2919         primer->link = link;
2920         primer->file = file;
2921         primer->fd = fd;
2922         primer->id = id;
2923         return 0;
2924 }
2925
2926 int bpf_link_settle(struct bpf_link_primer *primer)
2927 {
2928         /* make bpf_link fetchable by ID */
2929         spin_lock_bh(&link_idr_lock);
2930         primer->link->id = primer->id;
2931         spin_unlock_bh(&link_idr_lock);
2932         /* make bpf_link fetchable by FD */
2933         fd_install(primer->fd, primer->file);
2934         /* pass through installed FD */
2935         return primer->fd;
2936 }
2937
2938 int bpf_link_new_fd(struct bpf_link *link)
2939 {
2940         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2941 }
2942
2943 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2944 {
2945         struct fd f = fdget(ufd);
2946         struct bpf_link *link;
2947
2948         if (!f.file)
2949                 return ERR_PTR(-EBADF);
2950         if (f.file->f_op != &bpf_link_fops) {
2951                 fdput(f);
2952                 return ERR_PTR(-EINVAL);
2953         }
2954
2955         link = f.file->private_data;
2956         bpf_link_inc(link);
2957         fdput(f);
2958
2959         return link;
2960 }
2961 EXPORT_SYMBOL(bpf_link_get_from_fd);
2962
2963 static void bpf_tracing_link_release(struct bpf_link *link)
2964 {
2965         struct bpf_tracing_link *tr_link =
2966                 container_of(link, struct bpf_tracing_link, link.link);
2967
2968         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2969                                                 tr_link->trampoline));
2970
2971         bpf_trampoline_put(tr_link->trampoline);
2972
2973         /* tgt_prog is NULL if target is a kernel function */
2974         if (tr_link->tgt_prog)
2975                 bpf_prog_put(tr_link->tgt_prog);
2976 }
2977
2978 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2979 {
2980         struct bpf_tracing_link *tr_link =
2981                 container_of(link, struct bpf_tracing_link, link.link);
2982
2983         kfree(tr_link);
2984 }
2985
2986 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2987                                          struct seq_file *seq)
2988 {
2989         struct bpf_tracing_link *tr_link =
2990                 container_of(link, struct bpf_tracing_link, link.link);
2991         u32 target_btf_id, target_obj_id;
2992
2993         bpf_trampoline_unpack_key(tr_link->trampoline->key,
2994                                   &target_obj_id, &target_btf_id);
2995         seq_printf(seq,
2996                    "attach_type:\t%d\n"
2997                    "target_obj_id:\t%u\n"
2998                    "target_btf_id:\t%u\n",
2999                    tr_link->attach_type,
3000                    target_obj_id,
3001                    target_btf_id);
3002 }
3003
3004 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3005                                            struct bpf_link_info *info)
3006 {
3007         struct bpf_tracing_link *tr_link =
3008                 container_of(link, struct bpf_tracing_link, link.link);
3009
3010         info->tracing.attach_type = tr_link->attach_type;
3011         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3012                                   &info->tracing.target_obj_id,
3013                                   &info->tracing.target_btf_id);
3014
3015         return 0;
3016 }
3017
3018 static const struct bpf_link_ops bpf_tracing_link_lops = {
3019         .release = bpf_tracing_link_release,
3020         .dealloc = bpf_tracing_link_dealloc,
3021         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3022         .fill_link_info = bpf_tracing_link_fill_link_info,
3023 };
3024
3025 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3026                                    int tgt_prog_fd,
3027                                    u32 btf_id,
3028                                    u64 bpf_cookie)
3029 {
3030         struct bpf_link_primer link_primer;
3031         struct bpf_prog *tgt_prog = NULL;
3032         struct bpf_trampoline *tr = NULL;
3033         struct bpf_tracing_link *link;
3034         u64 key = 0;
3035         int err;
3036
3037         switch (prog->type) {
3038         case BPF_PROG_TYPE_TRACING:
3039                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3040                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3041                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3042                         err = -EINVAL;
3043                         goto out_put_prog;
3044                 }
3045                 break;
3046         case BPF_PROG_TYPE_EXT:
3047                 if (prog->expected_attach_type != 0) {
3048                         err = -EINVAL;
3049                         goto out_put_prog;
3050                 }
3051                 break;
3052         case BPF_PROG_TYPE_LSM:
3053                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3054                         err = -EINVAL;
3055                         goto out_put_prog;
3056                 }
3057                 break;
3058         default:
3059                 err = -EINVAL;
3060                 goto out_put_prog;
3061         }
3062
3063         if (!!tgt_prog_fd != !!btf_id) {
3064                 err = -EINVAL;
3065                 goto out_put_prog;
3066         }
3067
3068         if (tgt_prog_fd) {
3069                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
3070                 if (prog->type != BPF_PROG_TYPE_EXT) {
3071                         err = -EINVAL;
3072                         goto out_put_prog;
3073                 }
3074
3075                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3076                 if (IS_ERR(tgt_prog)) {
3077                         err = PTR_ERR(tgt_prog);
3078                         tgt_prog = NULL;
3079                         goto out_put_prog;
3080                 }
3081
3082                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3083         }
3084
3085         link = kzalloc(sizeof(*link), GFP_USER);
3086         if (!link) {
3087                 err = -ENOMEM;
3088                 goto out_put_prog;
3089         }
3090         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3091                       &bpf_tracing_link_lops, prog);
3092         link->attach_type = prog->expected_attach_type;
3093         link->link.cookie = bpf_cookie;
3094
3095         mutex_lock(&prog->aux->dst_mutex);
3096
3097         /* There are a few possible cases here:
3098          *
3099          * - if prog->aux->dst_trampoline is set, the program was just loaded
3100          *   and not yet attached to anything, so we can use the values stored
3101          *   in prog->aux
3102          *
3103          * - if prog->aux->dst_trampoline is NULL, the program has already been
3104          *   attached to a target and its initial target was cleared (below)
3105          *
3106          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3107          *   target_btf_id using the link_create API.
3108          *
3109          * - if tgt_prog == NULL when this function was called using the old
3110          *   raw_tracepoint_open API, and we need a target from prog->aux
3111          *
3112          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3113          *   was detached and is going for re-attachment.
3114          */
3115         if (!prog->aux->dst_trampoline && !tgt_prog) {
3116                 /*
3117                  * Allow re-attach for TRACING and LSM programs. If it's
3118                  * currently linked, bpf_trampoline_link_prog will fail.
3119                  * EXT programs need to specify tgt_prog_fd, so they
3120                  * re-attach in separate code path.
3121                  */
3122                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3123                     prog->type != BPF_PROG_TYPE_LSM) {
3124                         err = -EINVAL;
3125                         goto out_unlock;
3126                 }
3127                 btf_id = prog->aux->attach_btf_id;
3128                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3129         }
3130
3131         if (!prog->aux->dst_trampoline ||
3132             (key && key != prog->aux->dst_trampoline->key)) {
3133                 /* If there is no saved target, or the specified target is
3134                  * different from the destination specified at load time, we
3135                  * need a new trampoline and a check for compatibility
3136                  */
3137                 struct bpf_attach_target_info tgt_info = {};
3138
3139                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3140                                               &tgt_info);
3141                 if (err)
3142                         goto out_unlock;
3143
3144                 if (tgt_info.tgt_mod) {
3145                         module_put(prog->aux->mod);
3146                         prog->aux->mod = tgt_info.tgt_mod;
3147                 }
3148
3149                 tr = bpf_trampoline_get(key, &tgt_info);
3150                 if (!tr) {
3151                         err = -ENOMEM;
3152                         goto out_unlock;
3153                 }
3154         } else {
3155                 /* The caller didn't specify a target, or the target was the
3156                  * same as the destination supplied during program load. This
3157                  * means we can reuse the trampoline and reference from program
3158                  * load time, and there is no need to allocate a new one. This
3159                  * can only happen once for any program, as the saved values in
3160                  * prog->aux are cleared below.
3161                  */
3162                 tr = prog->aux->dst_trampoline;
3163                 tgt_prog = prog->aux->dst_prog;
3164         }
3165
3166         err = bpf_link_prime(&link->link.link, &link_primer);
3167         if (err)
3168                 goto out_unlock;
3169
3170         err = bpf_trampoline_link_prog(&link->link, tr);
3171         if (err) {
3172                 bpf_link_cleanup(&link_primer);
3173                 link = NULL;
3174                 goto out_unlock;
3175         }
3176
3177         link->tgt_prog = tgt_prog;
3178         link->trampoline = tr;
3179
3180         /* Always clear the trampoline and target prog from prog->aux to make
3181          * sure the original attach destination is not kept alive after a
3182          * program is (re-)attached to another target.
3183          */
3184         if (prog->aux->dst_prog &&
3185             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3186                 /* got extra prog ref from syscall, or attaching to different prog */
3187                 bpf_prog_put(prog->aux->dst_prog);
3188         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3189                 /* we allocated a new trampoline, so free the old one */
3190                 bpf_trampoline_put(prog->aux->dst_trampoline);
3191
3192         prog->aux->dst_prog = NULL;
3193         prog->aux->dst_trampoline = NULL;
3194         mutex_unlock(&prog->aux->dst_mutex);
3195
3196         return bpf_link_settle(&link_primer);
3197 out_unlock:
3198         if (tr && tr != prog->aux->dst_trampoline)
3199                 bpf_trampoline_put(tr);
3200         mutex_unlock(&prog->aux->dst_mutex);
3201         kfree(link);
3202 out_put_prog:
3203         if (tgt_prog_fd && tgt_prog)
3204                 bpf_prog_put(tgt_prog);
3205         return err;
3206 }
3207
3208 struct bpf_raw_tp_link {
3209         struct bpf_link link;
3210         struct bpf_raw_event_map *btp;
3211 };
3212
3213 static void bpf_raw_tp_link_release(struct bpf_link *link)
3214 {
3215         struct bpf_raw_tp_link *raw_tp =
3216                 container_of(link, struct bpf_raw_tp_link, link);
3217
3218         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3219         bpf_put_raw_tracepoint(raw_tp->btp);
3220 }
3221
3222 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3223 {
3224         struct bpf_raw_tp_link *raw_tp =
3225                 container_of(link, struct bpf_raw_tp_link, link);
3226
3227         kfree(raw_tp);
3228 }
3229
3230 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3231                                         struct seq_file *seq)
3232 {
3233         struct bpf_raw_tp_link *raw_tp_link =
3234                 container_of(link, struct bpf_raw_tp_link, link);
3235
3236         seq_printf(seq,
3237                    "tp_name:\t%s\n",
3238                    raw_tp_link->btp->tp->name);
3239 }
3240
3241 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3242                                           struct bpf_link_info *info)
3243 {
3244         struct bpf_raw_tp_link *raw_tp_link =
3245                 container_of(link, struct bpf_raw_tp_link, link);
3246         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3247         const char *tp_name = raw_tp_link->btp->tp->name;
3248         u32 ulen = info->raw_tracepoint.tp_name_len;
3249         size_t tp_len = strlen(tp_name);
3250
3251         if (!ulen ^ !ubuf)
3252                 return -EINVAL;
3253
3254         info->raw_tracepoint.tp_name_len = tp_len + 1;
3255
3256         if (!ubuf)
3257                 return 0;
3258
3259         if (ulen >= tp_len + 1) {
3260                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3261                         return -EFAULT;
3262         } else {
3263                 char zero = '\0';
3264
3265                 if (copy_to_user(ubuf, tp_name, ulen - 1))
3266                         return -EFAULT;
3267                 if (put_user(zero, ubuf + ulen - 1))
3268                         return -EFAULT;
3269                 return -ENOSPC;
3270         }
3271
3272         return 0;
3273 }
3274
3275 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3276         .release = bpf_raw_tp_link_release,
3277         .dealloc = bpf_raw_tp_link_dealloc,
3278         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3279         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3280 };
3281
3282 #ifdef CONFIG_PERF_EVENTS
3283 struct bpf_perf_link {
3284         struct bpf_link link;
3285         struct file *perf_file;
3286 };
3287
3288 static void bpf_perf_link_release(struct bpf_link *link)
3289 {
3290         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3291         struct perf_event *event = perf_link->perf_file->private_data;
3292
3293         perf_event_free_bpf_prog(event);
3294         fput(perf_link->perf_file);
3295 }
3296
3297 static void bpf_perf_link_dealloc(struct bpf_link *link)
3298 {
3299         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3300
3301         kfree(perf_link);
3302 }
3303
3304 static const struct bpf_link_ops bpf_perf_link_lops = {
3305         .release = bpf_perf_link_release,
3306         .dealloc = bpf_perf_link_dealloc,
3307 };
3308
3309 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3310 {
3311         struct bpf_link_primer link_primer;
3312         struct bpf_perf_link *link;
3313         struct perf_event *event;
3314         struct file *perf_file;
3315         int err;
3316
3317         if (attr->link_create.flags)
3318                 return -EINVAL;
3319
3320         perf_file = perf_event_get(attr->link_create.target_fd);
3321         if (IS_ERR(perf_file))
3322                 return PTR_ERR(perf_file);
3323
3324         link = kzalloc(sizeof(*link), GFP_USER);
3325         if (!link) {
3326                 err = -ENOMEM;
3327                 goto out_put_file;
3328         }
3329         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3330         link->perf_file = perf_file;
3331
3332         err = bpf_link_prime(&link->link, &link_primer);
3333         if (err) {
3334                 kfree(link);
3335                 goto out_put_file;
3336         }
3337
3338         event = perf_file->private_data;
3339         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3340         if (err) {
3341                 bpf_link_cleanup(&link_primer);
3342                 goto out_put_file;
3343         }
3344         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3345         bpf_prog_inc(prog);
3346
3347         return bpf_link_settle(&link_primer);
3348
3349 out_put_file:
3350         fput(perf_file);
3351         return err;
3352 }
3353 #else
3354 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3355 {
3356         return -EOPNOTSUPP;
3357 }
3358 #endif /* CONFIG_PERF_EVENTS */
3359
3360 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3361                                   const char __user *user_tp_name)
3362 {
3363         struct bpf_link_primer link_primer;
3364         struct bpf_raw_tp_link *link;
3365         struct bpf_raw_event_map *btp;
3366         const char *tp_name;
3367         char buf[128];
3368         int err;
3369
3370         switch (prog->type) {
3371         case BPF_PROG_TYPE_TRACING:
3372         case BPF_PROG_TYPE_EXT:
3373         case BPF_PROG_TYPE_LSM:
3374                 if (user_tp_name)
3375                         /* The attach point for this category of programs
3376                          * should be specified via btf_id during program load.
3377                          */
3378                         return -EINVAL;
3379                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3380                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3381                         tp_name = prog->aux->attach_func_name;
3382                         break;
3383                 }
3384                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3385         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3386         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3387                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3388                         return -EFAULT;
3389                 buf[sizeof(buf) - 1] = 0;
3390                 tp_name = buf;
3391                 break;
3392         default:
3393                 return -EINVAL;
3394         }
3395
3396         btp = bpf_get_raw_tracepoint(tp_name);
3397         if (!btp)
3398                 return -ENOENT;
3399
3400         link = kzalloc(sizeof(*link), GFP_USER);
3401         if (!link) {
3402                 err = -ENOMEM;
3403                 goto out_put_btp;
3404         }
3405         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3406                       &bpf_raw_tp_link_lops, prog);
3407         link->btp = btp;
3408
3409         err = bpf_link_prime(&link->link, &link_primer);
3410         if (err) {
3411                 kfree(link);
3412                 goto out_put_btp;
3413         }
3414
3415         err = bpf_probe_register(link->btp, prog);
3416         if (err) {
3417                 bpf_link_cleanup(&link_primer);
3418                 goto out_put_btp;
3419         }
3420
3421         return bpf_link_settle(&link_primer);
3422
3423 out_put_btp:
3424         bpf_put_raw_tracepoint(btp);
3425         return err;
3426 }
3427
3428 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3429
3430 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3431 {
3432         struct bpf_prog *prog;
3433         int fd;
3434
3435         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3436                 return -EINVAL;
3437
3438         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3439         if (IS_ERR(prog))
3440                 return PTR_ERR(prog);
3441
3442         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3443         if (fd < 0)
3444                 bpf_prog_put(prog);
3445         return fd;
3446 }
3447
3448 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3449                                              enum bpf_attach_type attach_type)
3450 {
3451         switch (prog->type) {
3452         case BPF_PROG_TYPE_CGROUP_SOCK:
3453         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3454         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3455         case BPF_PROG_TYPE_SK_LOOKUP:
3456                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3457         case BPF_PROG_TYPE_CGROUP_SKB:
3458                 if (!capable(CAP_NET_ADMIN))
3459                         /* cg-skb progs can be loaded by unpriv user.
3460                          * check permissions at attach time.
3461                          */
3462                         return -EPERM;
3463                 return prog->enforce_expected_attach_type &&
3464                         prog->expected_attach_type != attach_type ?
3465                         -EINVAL : 0;
3466         default:
3467                 return 0;
3468         }
3469 }
3470
3471 static enum bpf_prog_type
3472 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3473 {
3474         switch (attach_type) {
3475         case BPF_CGROUP_INET_INGRESS:
3476         case BPF_CGROUP_INET_EGRESS:
3477                 return BPF_PROG_TYPE_CGROUP_SKB;
3478         case BPF_CGROUP_INET_SOCK_CREATE:
3479         case BPF_CGROUP_INET_SOCK_RELEASE:
3480         case BPF_CGROUP_INET4_POST_BIND:
3481         case BPF_CGROUP_INET6_POST_BIND:
3482                 return BPF_PROG_TYPE_CGROUP_SOCK;
3483         case BPF_CGROUP_INET4_BIND:
3484         case BPF_CGROUP_INET6_BIND:
3485         case BPF_CGROUP_INET4_CONNECT:
3486         case BPF_CGROUP_INET6_CONNECT:
3487         case BPF_CGROUP_INET4_GETPEERNAME:
3488         case BPF_CGROUP_INET6_GETPEERNAME:
3489         case BPF_CGROUP_INET4_GETSOCKNAME:
3490         case BPF_CGROUP_INET6_GETSOCKNAME:
3491         case BPF_CGROUP_UDP4_SENDMSG:
3492         case BPF_CGROUP_UDP6_SENDMSG:
3493         case BPF_CGROUP_UDP4_RECVMSG:
3494         case BPF_CGROUP_UDP6_RECVMSG:
3495                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3496         case BPF_CGROUP_SOCK_OPS:
3497                 return BPF_PROG_TYPE_SOCK_OPS;
3498         case BPF_CGROUP_DEVICE:
3499                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3500         case BPF_SK_MSG_VERDICT:
3501                 return BPF_PROG_TYPE_SK_MSG;
3502         case BPF_SK_SKB_STREAM_PARSER:
3503         case BPF_SK_SKB_STREAM_VERDICT:
3504         case BPF_SK_SKB_VERDICT:
3505                 return BPF_PROG_TYPE_SK_SKB;
3506         case BPF_LIRC_MODE2:
3507                 return BPF_PROG_TYPE_LIRC_MODE2;
3508         case BPF_FLOW_DISSECTOR:
3509                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3510         case BPF_CGROUP_SYSCTL:
3511                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3512         case BPF_CGROUP_GETSOCKOPT:
3513         case BPF_CGROUP_SETSOCKOPT:
3514                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3515         case BPF_TRACE_ITER:
3516         case BPF_TRACE_RAW_TP:
3517         case BPF_TRACE_FENTRY:
3518         case BPF_TRACE_FEXIT:
3519         case BPF_MODIFY_RETURN:
3520                 return BPF_PROG_TYPE_TRACING;
3521         case BPF_LSM_MAC:
3522                 return BPF_PROG_TYPE_LSM;
3523         case BPF_SK_LOOKUP:
3524                 return BPF_PROG_TYPE_SK_LOOKUP;
3525         case BPF_XDP:
3526                 return BPF_PROG_TYPE_XDP;
3527         case BPF_LSM_CGROUP:
3528                 return BPF_PROG_TYPE_LSM;
3529         default:
3530                 return BPF_PROG_TYPE_UNSPEC;
3531         }
3532 }
3533
3534 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3535
3536 #define BPF_F_ATTACH_MASK \
3537         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3538
3539 static int bpf_prog_attach(const union bpf_attr *attr)
3540 {
3541         enum bpf_prog_type ptype;
3542         struct bpf_prog *prog;
3543         int ret;
3544
3545         if (CHECK_ATTR(BPF_PROG_ATTACH))
3546                 return -EINVAL;
3547
3548         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3549                 return -EINVAL;
3550
3551         ptype = attach_type_to_prog_type(attr->attach_type);
3552         if (ptype == BPF_PROG_TYPE_UNSPEC)
3553                 return -EINVAL;
3554
3555         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3556         if (IS_ERR(prog))
3557                 return PTR_ERR(prog);
3558
3559         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3560                 bpf_prog_put(prog);
3561                 return -EINVAL;
3562         }
3563
3564         switch (ptype) {
3565         case BPF_PROG_TYPE_SK_SKB:
3566         case BPF_PROG_TYPE_SK_MSG:
3567                 ret = sock_map_get_from_fd(attr, prog);
3568                 break;
3569         case BPF_PROG_TYPE_LIRC_MODE2:
3570                 ret = lirc_prog_attach(attr, prog);
3571                 break;
3572         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3573                 ret = netns_bpf_prog_attach(attr, prog);
3574                 break;
3575         case BPF_PROG_TYPE_CGROUP_DEVICE:
3576         case BPF_PROG_TYPE_CGROUP_SKB:
3577         case BPF_PROG_TYPE_CGROUP_SOCK:
3578         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3579         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3580         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3581         case BPF_PROG_TYPE_SOCK_OPS:
3582         case BPF_PROG_TYPE_LSM:
3583                 if (ptype == BPF_PROG_TYPE_LSM &&
3584                     prog->expected_attach_type != BPF_LSM_CGROUP)
3585                         ret = -EINVAL;
3586                 else
3587                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3588                 break;
3589         default:
3590                 ret = -EINVAL;
3591         }
3592
3593         if (ret)
3594                 bpf_prog_put(prog);
3595         return ret;
3596 }
3597
3598 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3599
3600 static int bpf_prog_detach(const union bpf_attr *attr)
3601 {
3602         enum bpf_prog_type ptype;
3603
3604         if (CHECK_ATTR(BPF_PROG_DETACH))
3605                 return -EINVAL;
3606
3607         ptype = attach_type_to_prog_type(attr->attach_type);
3608
3609         switch (ptype) {
3610         case BPF_PROG_TYPE_SK_MSG:
3611         case BPF_PROG_TYPE_SK_SKB:
3612                 return sock_map_prog_detach(attr, ptype);
3613         case BPF_PROG_TYPE_LIRC_MODE2:
3614                 return lirc_prog_detach(attr);
3615         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3616                 return netns_bpf_prog_detach(attr, ptype);
3617         case BPF_PROG_TYPE_CGROUP_DEVICE:
3618         case BPF_PROG_TYPE_CGROUP_SKB:
3619         case BPF_PROG_TYPE_CGROUP_SOCK:
3620         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3621         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3622         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3623         case BPF_PROG_TYPE_SOCK_OPS:
3624         case BPF_PROG_TYPE_LSM:
3625                 return cgroup_bpf_prog_detach(attr, ptype);
3626         default:
3627                 return -EINVAL;
3628         }
3629 }
3630
3631 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3632
3633 static int bpf_prog_query(const union bpf_attr *attr,
3634                           union bpf_attr __user *uattr)
3635 {
3636         if (!capable(CAP_NET_ADMIN))
3637                 return -EPERM;
3638         if (CHECK_ATTR(BPF_PROG_QUERY))
3639                 return -EINVAL;
3640         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3641                 return -EINVAL;
3642
3643         switch (attr->query.attach_type) {
3644         case BPF_CGROUP_INET_INGRESS:
3645         case BPF_CGROUP_INET_EGRESS:
3646         case BPF_CGROUP_INET_SOCK_CREATE:
3647         case BPF_CGROUP_INET_SOCK_RELEASE:
3648         case BPF_CGROUP_INET4_BIND:
3649         case BPF_CGROUP_INET6_BIND:
3650         case BPF_CGROUP_INET4_POST_BIND:
3651         case BPF_CGROUP_INET6_POST_BIND:
3652         case BPF_CGROUP_INET4_CONNECT:
3653         case BPF_CGROUP_INET6_CONNECT:
3654         case BPF_CGROUP_INET4_GETPEERNAME:
3655         case BPF_CGROUP_INET6_GETPEERNAME:
3656         case BPF_CGROUP_INET4_GETSOCKNAME:
3657         case BPF_CGROUP_INET6_GETSOCKNAME:
3658         case BPF_CGROUP_UDP4_SENDMSG:
3659         case BPF_CGROUP_UDP6_SENDMSG:
3660         case BPF_CGROUP_UDP4_RECVMSG:
3661         case BPF_CGROUP_UDP6_RECVMSG:
3662         case BPF_CGROUP_SOCK_OPS:
3663         case BPF_CGROUP_DEVICE:
3664         case BPF_CGROUP_SYSCTL:
3665         case BPF_CGROUP_GETSOCKOPT:
3666         case BPF_CGROUP_SETSOCKOPT:
3667         case BPF_LSM_CGROUP:
3668                 return cgroup_bpf_prog_query(attr, uattr);
3669         case BPF_LIRC_MODE2:
3670                 return lirc_prog_query(attr, uattr);
3671         case BPF_FLOW_DISSECTOR:
3672         case BPF_SK_LOOKUP:
3673                 return netns_bpf_prog_query(attr, uattr);
3674         case BPF_SK_SKB_STREAM_PARSER:
3675         case BPF_SK_SKB_STREAM_VERDICT:
3676         case BPF_SK_MSG_VERDICT:
3677         case BPF_SK_SKB_VERDICT:
3678                 return sock_map_bpf_prog_query(attr, uattr);
3679         default:
3680                 return -EINVAL;
3681         }
3682 }
3683
3684 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3685
3686 static int bpf_prog_test_run(const union bpf_attr *attr,
3687                              union bpf_attr __user *uattr)
3688 {
3689         struct bpf_prog *prog;
3690         int ret = -ENOTSUPP;
3691
3692         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3693                 return -EINVAL;
3694
3695         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3696             (!attr->test.ctx_size_in && attr->test.ctx_in))
3697                 return -EINVAL;
3698
3699         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3700             (!attr->test.ctx_size_out && attr->test.ctx_out))
3701                 return -EINVAL;
3702
3703         prog = bpf_prog_get(attr->test.prog_fd);
3704         if (IS_ERR(prog))
3705                 return PTR_ERR(prog);
3706
3707         if (prog->aux->ops->test_run)
3708                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3709
3710         bpf_prog_put(prog);
3711         return ret;
3712 }
3713
3714 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3715
3716 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3717                                union bpf_attr __user *uattr,
3718                                struct idr *idr,
3719                                spinlock_t *lock)
3720 {
3721         u32 next_id = attr->start_id;
3722         int err = 0;
3723
3724         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3725                 return -EINVAL;
3726
3727         if (!capable(CAP_SYS_ADMIN))
3728                 return -EPERM;
3729
3730         next_id++;
3731         spin_lock_bh(lock);
3732         if (!idr_get_next(idr, &next_id))
3733                 err = -ENOENT;
3734         spin_unlock_bh(lock);
3735
3736         if (!err)
3737                 err = put_user(next_id, &uattr->next_id);
3738
3739         return err;
3740 }
3741
3742 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3743 {
3744         struct bpf_map *map;
3745
3746         spin_lock_bh(&map_idr_lock);
3747 again:
3748         map = idr_get_next(&map_idr, id);
3749         if (map) {
3750                 map = __bpf_map_inc_not_zero(map, false);
3751                 if (IS_ERR(map)) {
3752                         (*id)++;
3753                         goto again;
3754                 }
3755         }
3756         spin_unlock_bh(&map_idr_lock);
3757
3758         return map;
3759 }
3760
3761 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3762 {
3763         struct bpf_prog *prog;
3764
3765         spin_lock_bh(&prog_idr_lock);
3766 again:
3767         prog = idr_get_next(&prog_idr, id);
3768         if (prog) {
3769                 prog = bpf_prog_inc_not_zero(prog);
3770                 if (IS_ERR(prog)) {
3771                         (*id)++;
3772                         goto again;
3773                 }
3774         }
3775         spin_unlock_bh(&prog_idr_lock);
3776
3777         return prog;
3778 }
3779
3780 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3781
3782 struct bpf_prog *bpf_prog_by_id(u32 id)
3783 {
3784         struct bpf_prog *prog;
3785
3786         if (!id)
3787                 return ERR_PTR(-ENOENT);
3788
3789         spin_lock_bh(&prog_idr_lock);
3790         prog = idr_find(&prog_idr, id);
3791         if (prog)
3792                 prog = bpf_prog_inc_not_zero(prog);
3793         else
3794                 prog = ERR_PTR(-ENOENT);
3795         spin_unlock_bh(&prog_idr_lock);
3796         return prog;
3797 }
3798
3799 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3800 {
3801         struct bpf_prog *prog;
3802         u32 id = attr->prog_id;
3803         int fd;
3804
3805         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3806                 return -EINVAL;
3807
3808         if (!capable(CAP_SYS_ADMIN))
3809                 return -EPERM;
3810
3811         prog = bpf_prog_by_id(id);
3812         if (IS_ERR(prog))
3813                 return PTR_ERR(prog);
3814
3815         fd = bpf_prog_new_fd(prog);
3816         if (fd < 0)
3817                 bpf_prog_put(prog);
3818
3819         return fd;
3820 }
3821
3822 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3823
3824 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3825 {
3826         struct bpf_map *map;
3827         u32 id = attr->map_id;
3828         int f_flags;
3829         int fd;
3830
3831         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3832             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3833                 return -EINVAL;
3834
3835         if (!capable(CAP_SYS_ADMIN))
3836                 return -EPERM;
3837
3838         f_flags = bpf_get_file_flag(attr->open_flags);
3839         if (f_flags < 0)
3840                 return f_flags;
3841
3842         spin_lock_bh(&map_idr_lock);
3843         map = idr_find(&map_idr, id);
3844         if (map)
3845                 map = __bpf_map_inc_not_zero(map, true);
3846         else
3847                 map = ERR_PTR(-ENOENT);
3848         spin_unlock_bh(&map_idr_lock);
3849
3850         if (IS_ERR(map))
3851                 return PTR_ERR(map);
3852
3853         fd = bpf_map_new_fd(map, f_flags);
3854         if (fd < 0)
3855                 bpf_map_put_with_uref(map);
3856
3857         return fd;
3858 }
3859
3860 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3861                                               unsigned long addr, u32 *off,
3862                                               u32 *type)
3863 {
3864         const struct bpf_map *map;
3865         int i;
3866
3867         mutex_lock(&prog->aux->used_maps_mutex);
3868         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3869                 map = prog->aux->used_maps[i];
3870                 if (map == (void *)addr) {
3871                         *type = BPF_PSEUDO_MAP_FD;
3872                         goto out;
3873                 }
3874                 if (!map->ops->map_direct_value_meta)
3875                         continue;
3876                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3877                         *type = BPF_PSEUDO_MAP_VALUE;
3878                         goto out;
3879                 }
3880         }
3881         map = NULL;
3882
3883 out:
3884         mutex_unlock(&prog->aux->used_maps_mutex);
3885         return map;
3886 }
3887
3888 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3889                                               const struct cred *f_cred)
3890 {
3891         const struct bpf_map *map;
3892         struct bpf_insn *insns;
3893         u32 off, type;
3894         u64 imm;
3895         u8 code;
3896         int i;
3897
3898         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3899                         GFP_USER);
3900         if (!insns)
3901                 return insns;
3902
3903         for (i = 0; i < prog->len; i++) {
3904                 code = insns[i].code;
3905
3906                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3907                         insns[i].code = BPF_JMP | BPF_CALL;
3908                         insns[i].imm = BPF_FUNC_tail_call;
3909                         /* fall-through */
3910                 }
3911                 if (code == (BPF_JMP | BPF_CALL) ||
3912                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3913                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3914                                 insns[i].code = BPF_JMP | BPF_CALL;
3915                         if (!bpf_dump_raw_ok(f_cred))
3916                                 insns[i].imm = 0;
3917                         continue;
3918                 }
3919                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3920                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3921                         continue;
3922                 }
3923
3924                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3925                         continue;
3926
3927                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3928                 map = bpf_map_from_imm(prog, imm, &off, &type);
3929                 if (map) {
3930                         insns[i].src_reg = type;
3931                         insns[i].imm = map->id;
3932                         insns[i + 1].imm = off;
3933                         continue;
3934                 }
3935         }
3936
3937         return insns;
3938 }
3939
3940 static int set_info_rec_size(struct bpf_prog_info *info)
3941 {
3942         /*
3943          * Ensure info.*_rec_size is the same as kernel expected size
3944          *
3945          * or
3946          *
3947          * Only allow zero *_rec_size if both _rec_size and _cnt are
3948          * zero.  In this case, the kernel will set the expected
3949          * _rec_size back to the info.
3950          */
3951
3952         if ((info->nr_func_info || info->func_info_rec_size) &&
3953             info->func_info_rec_size != sizeof(struct bpf_func_info))
3954                 return -EINVAL;
3955
3956         if ((info->nr_line_info || info->line_info_rec_size) &&
3957             info->line_info_rec_size != sizeof(struct bpf_line_info))
3958                 return -EINVAL;
3959
3960         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3961             info->jited_line_info_rec_size != sizeof(__u64))
3962                 return -EINVAL;
3963
3964         info->func_info_rec_size = sizeof(struct bpf_func_info);
3965         info->line_info_rec_size = sizeof(struct bpf_line_info);
3966         info->jited_line_info_rec_size = sizeof(__u64);
3967
3968         return 0;
3969 }
3970
3971 static int bpf_prog_get_info_by_fd(struct file *file,
3972                                    struct bpf_prog *prog,
3973                                    const union bpf_attr *attr,
3974                                    union bpf_attr __user *uattr)
3975 {
3976         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3977         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3978         struct bpf_prog_info info;
3979         u32 info_len = attr->info.info_len;
3980         struct bpf_prog_kstats stats;
3981         char __user *uinsns;
3982         u32 ulen;
3983         int err;
3984
3985         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3986         if (err)
3987                 return err;
3988         info_len = min_t(u32, sizeof(info), info_len);
3989
3990         memset(&info, 0, sizeof(info));
3991         if (copy_from_user(&info, uinfo, info_len))
3992                 return -EFAULT;
3993
3994         info.type = prog->type;
3995         info.id = prog->aux->id;
3996         info.load_time = prog->aux->load_time;
3997         info.created_by_uid = from_kuid_munged(current_user_ns(),
3998                                                prog->aux->user->uid);
3999         info.gpl_compatible = prog->gpl_compatible;
4000
4001         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4002         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4003
4004         mutex_lock(&prog->aux->used_maps_mutex);
4005         ulen = info.nr_map_ids;
4006         info.nr_map_ids = prog->aux->used_map_cnt;
4007         ulen = min_t(u32, info.nr_map_ids, ulen);
4008         if (ulen) {
4009                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4010                 u32 i;
4011
4012                 for (i = 0; i < ulen; i++)
4013                         if (put_user(prog->aux->used_maps[i]->id,
4014                                      &user_map_ids[i])) {
4015                                 mutex_unlock(&prog->aux->used_maps_mutex);
4016                                 return -EFAULT;
4017                         }
4018         }
4019         mutex_unlock(&prog->aux->used_maps_mutex);
4020
4021         err = set_info_rec_size(&info);
4022         if (err)
4023                 return err;
4024
4025         bpf_prog_get_stats(prog, &stats);
4026         info.run_time_ns = stats.nsecs;
4027         info.run_cnt = stats.cnt;
4028         info.recursion_misses = stats.misses;
4029
4030         info.verified_insns = prog->aux->verified_insns;
4031
4032         if (!bpf_capable()) {
4033                 info.jited_prog_len = 0;
4034                 info.xlated_prog_len = 0;
4035                 info.nr_jited_ksyms = 0;
4036                 info.nr_jited_func_lens = 0;
4037                 info.nr_func_info = 0;
4038                 info.nr_line_info = 0;
4039                 info.nr_jited_line_info = 0;
4040                 goto done;
4041         }
4042
4043         ulen = info.xlated_prog_len;
4044         info.xlated_prog_len = bpf_prog_insn_size(prog);
4045         if (info.xlated_prog_len && ulen) {
4046                 struct bpf_insn *insns_sanitized;
4047                 bool fault;
4048
4049                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4050                         info.xlated_prog_insns = 0;
4051                         goto done;
4052                 }
4053                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4054                 if (!insns_sanitized)
4055                         return -ENOMEM;
4056                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4057                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4058                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4059                 kfree(insns_sanitized);
4060                 if (fault)
4061                         return -EFAULT;
4062         }
4063
4064         if (bpf_prog_is_offloaded(prog->aux)) {
4065                 err = bpf_prog_offload_info_fill(&info, prog);
4066                 if (err)
4067                         return err;
4068                 goto done;
4069         }
4070
4071         /* NOTE: the following code is supposed to be skipped for offload.
4072          * bpf_prog_offload_info_fill() is the place to fill similar fields
4073          * for offload.
4074          */
4075         ulen = info.jited_prog_len;
4076         if (prog->aux->func_cnt) {
4077                 u32 i;
4078
4079                 info.jited_prog_len = 0;
4080                 for (i = 0; i < prog->aux->func_cnt; i++)
4081                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4082         } else {
4083                 info.jited_prog_len = prog->jited_len;
4084         }
4085
4086         if (info.jited_prog_len && ulen) {
4087                 if (bpf_dump_raw_ok(file->f_cred)) {
4088                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4089                         ulen = min_t(u32, info.jited_prog_len, ulen);
4090
4091                         /* for multi-function programs, copy the JITed
4092                          * instructions for all the functions
4093                          */
4094                         if (prog->aux->func_cnt) {
4095                                 u32 len, free, i;
4096                                 u8 *img;
4097
4098                                 free = ulen;
4099                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4100                                         len = prog->aux->func[i]->jited_len;
4101                                         len = min_t(u32, len, free);
4102                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4103                                         if (copy_to_user(uinsns, img, len))
4104                                                 return -EFAULT;
4105                                         uinsns += len;
4106                                         free -= len;
4107                                         if (!free)
4108                                                 break;
4109                                 }
4110                         } else {
4111                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4112                                         return -EFAULT;
4113                         }
4114                 } else {
4115                         info.jited_prog_insns = 0;
4116                 }
4117         }
4118
4119         ulen = info.nr_jited_ksyms;
4120         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4121         if (ulen) {
4122                 if (bpf_dump_raw_ok(file->f_cred)) {
4123                         unsigned long ksym_addr;
4124                         u64 __user *user_ksyms;
4125                         u32 i;
4126
4127                         /* copy the address of the kernel symbol
4128                          * corresponding to each function
4129                          */
4130                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4131                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4132                         if (prog->aux->func_cnt) {
4133                                 for (i = 0; i < ulen; i++) {
4134                                         ksym_addr = (unsigned long)
4135                                                 prog->aux->func[i]->bpf_func;
4136                                         if (put_user((u64) ksym_addr,
4137                                                      &user_ksyms[i]))
4138                                                 return -EFAULT;
4139                                 }
4140                         } else {
4141                                 ksym_addr = (unsigned long) prog->bpf_func;
4142                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4143                                         return -EFAULT;
4144                         }
4145                 } else {
4146                         info.jited_ksyms = 0;
4147                 }
4148         }
4149
4150         ulen = info.nr_jited_func_lens;
4151         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4152         if (ulen) {
4153                 if (bpf_dump_raw_ok(file->f_cred)) {
4154                         u32 __user *user_lens;
4155                         u32 func_len, i;
4156
4157                         /* copy the JITed image lengths for each function */
4158                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4159                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4160                         if (prog->aux->func_cnt) {
4161                                 for (i = 0; i < ulen; i++) {
4162                                         func_len =
4163                                                 prog->aux->func[i]->jited_len;
4164                                         if (put_user(func_len, &user_lens[i]))
4165                                                 return -EFAULT;
4166                                 }
4167                         } else {
4168                                 func_len = prog->jited_len;
4169                                 if (put_user(func_len, &user_lens[0]))
4170                                         return -EFAULT;
4171                         }
4172                 } else {
4173                         info.jited_func_lens = 0;
4174                 }
4175         }
4176
4177         if (prog->aux->btf)
4178                 info.btf_id = btf_obj_id(prog->aux->btf);
4179         info.attach_btf_id = prog->aux->attach_btf_id;
4180         if (attach_btf)
4181                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4182
4183         ulen = info.nr_func_info;
4184         info.nr_func_info = prog->aux->func_info_cnt;
4185         if (info.nr_func_info && ulen) {
4186                 char __user *user_finfo;
4187
4188                 user_finfo = u64_to_user_ptr(info.func_info);
4189                 ulen = min_t(u32, info.nr_func_info, ulen);
4190                 if (copy_to_user(user_finfo, prog->aux->func_info,
4191                                  info.func_info_rec_size * ulen))
4192                         return -EFAULT;
4193         }
4194
4195         ulen = info.nr_line_info;
4196         info.nr_line_info = prog->aux->nr_linfo;
4197         if (info.nr_line_info && ulen) {
4198                 __u8 __user *user_linfo;
4199
4200                 user_linfo = u64_to_user_ptr(info.line_info);
4201                 ulen = min_t(u32, info.nr_line_info, ulen);
4202                 if (copy_to_user(user_linfo, prog->aux->linfo,
4203                                  info.line_info_rec_size * ulen))
4204                         return -EFAULT;
4205         }
4206
4207         ulen = info.nr_jited_line_info;
4208         if (prog->aux->jited_linfo)
4209                 info.nr_jited_line_info = prog->aux->nr_linfo;
4210         else
4211                 info.nr_jited_line_info = 0;
4212         if (info.nr_jited_line_info && ulen) {
4213                 if (bpf_dump_raw_ok(file->f_cred)) {
4214                         unsigned long line_addr;
4215                         __u64 __user *user_linfo;
4216                         u32 i;
4217
4218                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4219                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4220                         for (i = 0; i < ulen; i++) {
4221                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4222                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4223                                         return -EFAULT;
4224                         }
4225                 } else {
4226                         info.jited_line_info = 0;
4227                 }
4228         }
4229
4230         ulen = info.nr_prog_tags;
4231         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4232         if (ulen) {
4233                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4234                 u32 i;
4235
4236                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4237                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4238                 if (prog->aux->func_cnt) {
4239                         for (i = 0; i < ulen; i++) {
4240                                 if (copy_to_user(user_prog_tags[i],
4241                                                  prog->aux->func[i]->tag,
4242                                                  BPF_TAG_SIZE))
4243                                         return -EFAULT;
4244                         }
4245                 } else {
4246                         if (copy_to_user(user_prog_tags[0],
4247                                          prog->tag, BPF_TAG_SIZE))
4248                                 return -EFAULT;
4249                 }
4250         }
4251
4252 done:
4253         if (copy_to_user(uinfo, &info, info_len) ||
4254             put_user(info_len, &uattr->info.info_len))
4255                 return -EFAULT;
4256
4257         return 0;
4258 }
4259
4260 static int bpf_map_get_info_by_fd(struct file *file,
4261                                   struct bpf_map *map,
4262                                   const union bpf_attr *attr,
4263                                   union bpf_attr __user *uattr)
4264 {
4265         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4266         struct bpf_map_info info;
4267         u32 info_len = attr->info.info_len;
4268         int err;
4269
4270         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4271         if (err)
4272                 return err;
4273         info_len = min_t(u32, sizeof(info), info_len);
4274
4275         memset(&info, 0, sizeof(info));
4276         info.type = map->map_type;
4277         info.id = map->id;
4278         info.key_size = map->key_size;
4279         info.value_size = map->value_size;
4280         info.max_entries = map->max_entries;
4281         info.map_flags = map->map_flags;
4282         info.map_extra = map->map_extra;
4283         memcpy(info.name, map->name, sizeof(map->name));
4284
4285         if (map->btf) {
4286                 info.btf_id = btf_obj_id(map->btf);
4287                 info.btf_key_type_id = map->btf_key_type_id;
4288                 info.btf_value_type_id = map->btf_value_type_id;
4289         }
4290         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4291
4292         if (bpf_map_is_offloaded(map)) {
4293                 err = bpf_map_offload_info_fill(&info, map);
4294                 if (err)
4295                         return err;
4296         }
4297
4298         if (copy_to_user(uinfo, &info, info_len) ||
4299             put_user(info_len, &uattr->info.info_len))
4300                 return -EFAULT;
4301
4302         return 0;
4303 }
4304
4305 static int bpf_btf_get_info_by_fd(struct file *file,
4306                                   struct btf *btf,
4307                                   const union bpf_attr *attr,
4308                                   union bpf_attr __user *uattr)
4309 {
4310         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4311         u32 info_len = attr->info.info_len;
4312         int err;
4313
4314         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4315         if (err)
4316                 return err;
4317
4318         return btf_get_info_by_fd(btf, attr, uattr);
4319 }
4320
4321 static int bpf_link_get_info_by_fd(struct file *file,
4322                                   struct bpf_link *link,
4323                                   const union bpf_attr *attr,
4324                                   union bpf_attr __user *uattr)
4325 {
4326         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4327         struct bpf_link_info info;
4328         u32 info_len = attr->info.info_len;
4329         int err;
4330
4331         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4332         if (err)
4333                 return err;
4334         info_len = min_t(u32, sizeof(info), info_len);
4335
4336         memset(&info, 0, sizeof(info));
4337         if (copy_from_user(&info, uinfo, info_len))
4338                 return -EFAULT;
4339
4340         info.type = link->type;
4341         info.id = link->id;
4342         if (link->prog)
4343                 info.prog_id = link->prog->aux->id;
4344
4345         if (link->ops->fill_link_info) {
4346                 err = link->ops->fill_link_info(link, &info);
4347                 if (err)
4348                         return err;
4349         }
4350
4351         if (copy_to_user(uinfo, &info, info_len) ||
4352             put_user(info_len, &uattr->info.info_len))
4353                 return -EFAULT;
4354
4355         return 0;
4356 }
4357
4358
4359 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4360
4361 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4362                                   union bpf_attr __user *uattr)
4363 {
4364         int ufd = attr->info.bpf_fd;
4365         struct fd f;
4366         int err;
4367
4368         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4369                 return -EINVAL;
4370
4371         f = fdget(ufd);
4372         if (!f.file)
4373                 return -EBADFD;
4374
4375         if (f.file->f_op == &bpf_prog_fops)
4376                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4377                                               uattr);
4378         else if (f.file->f_op == &bpf_map_fops)
4379                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4380                                              uattr);
4381         else if (f.file->f_op == &btf_fops)
4382                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4383         else if (f.file->f_op == &bpf_link_fops)
4384                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4385                                               attr, uattr);
4386         else
4387                 err = -EINVAL;
4388
4389         fdput(f);
4390         return err;
4391 }
4392
4393 #define BPF_BTF_LOAD_LAST_FIELD btf_log_true_size
4394
4395 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4396 {
4397         if (CHECK_ATTR(BPF_BTF_LOAD))
4398                 return -EINVAL;
4399
4400         if (!bpf_capable())
4401                 return -EPERM;
4402
4403         return btf_new_fd(attr, uattr, uattr_size);
4404 }
4405
4406 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4407
4408 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4409 {
4410         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4411                 return -EINVAL;
4412
4413         if (!capable(CAP_SYS_ADMIN))
4414                 return -EPERM;
4415
4416         return btf_get_fd_by_id(attr->btf_id);
4417 }
4418
4419 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4420                                     union bpf_attr __user *uattr,
4421                                     u32 prog_id, u32 fd_type,
4422                                     const char *buf, u64 probe_offset,
4423                                     u64 probe_addr)
4424 {
4425         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4426         u32 len = buf ? strlen(buf) : 0, input_len;
4427         int err = 0;
4428
4429         if (put_user(len, &uattr->task_fd_query.buf_len))
4430                 return -EFAULT;
4431         input_len = attr->task_fd_query.buf_len;
4432         if (input_len && ubuf) {
4433                 if (!len) {
4434                         /* nothing to copy, just make ubuf NULL terminated */
4435                         char zero = '\0';
4436
4437                         if (put_user(zero, ubuf))
4438                                 return -EFAULT;
4439                 } else if (input_len >= len + 1) {
4440                         /* ubuf can hold the string with NULL terminator */
4441                         if (copy_to_user(ubuf, buf, len + 1))
4442                                 return -EFAULT;
4443                 } else {
4444                         /* ubuf cannot hold the string with NULL terminator,
4445                          * do a partial copy with NULL terminator.
4446                          */
4447                         char zero = '\0';
4448
4449                         err = -ENOSPC;
4450                         if (copy_to_user(ubuf, buf, input_len - 1))
4451                                 return -EFAULT;
4452                         if (put_user(zero, ubuf + input_len - 1))
4453                                 return -EFAULT;
4454                 }
4455         }
4456
4457         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4458             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4459             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4460             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4461                 return -EFAULT;
4462
4463         return err;
4464 }
4465
4466 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4467
4468 static int bpf_task_fd_query(const union bpf_attr *attr,
4469                              union bpf_attr __user *uattr)
4470 {
4471         pid_t pid = attr->task_fd_query.pid;
4472         u32 fd = attr->task_fd_query.fd;
4473         const struct perf_event *event;
4474         struct task_struct *task;
4475         struct file *file;
4476         int err;
4477
4478         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4479                 return -EINVAL;
4480
4481         if (!capable(CAP_SYS_ADMIN))
4482                 return -EPERM;
4483
4484         if (attr->task_fd_query.flags != 0)
4485                 return -EINVAL;
4486
4487         rcu_read_lock();
4488         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4489         rcu_read_unlock();
4490         if (!task)
4491                 return -ENOENT;
4492
4493         err = 0;
4494         file = fget_task(task, fd);
4495         put_task_struct(task);
4496         if (!file)
4497                 return -EBADF;
4498
4499         if (file->f_op == &bpf_link_fops) {
4500                 struct bpf_link *link = file->private_data;
4501
4502                 if (link->ops == &bpf_raw_tp_link_lops) {
4503                         struct bpf_raw_tp_link *raw_tp =
4504                                 container_of(link, struct bpf_raw_tp_link, link);
4505                         struct bpf_raw_event_map *btp = raw_tp->btp;
4506
4507                         err = bpf_task_fd_query_copy(attr, uattr,
4508                                                      raw_tp->link.prog->aux->id,
4509                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4510                                                      btp->tp->name, 0, 0);
4511                         goto put_file;
4512                 }
4513                 goto out_not_supp;
4514         }
4515
4516         event = perf_get_event(file);
4517         if (!IS_ERR(event)) {
4518                 u64 probe_offset, probe_addr;
4519                 u32 prog_id, fd_type;
4520                 const char *buf;
4521
4522                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4523                                               &buf, &probe_offset,
4524                                               &probe_addr);
4525                 if (!err)
4526                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4527                                                      fd_type, buf,
4528                                                      probe_offset,
4529                                                      probe_addr);
4530                 goto put_file;
4531         }
4532
4533 out_not_supp:
4534         err = -ENOTSUPP;
4535 put_file:
4536         fput(file);
4537         return err;
4538 }
4539
4540 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4541
4542 #define BPF_DO_BATCH(fn, ...)                   \
4543         do {                                    \
4544                 if (!fn) {                      \
4545                         err = -ENOTSUPP;        \
4546                         goto err_put;           \
4547                 }                               \
4548                 err = fn(__VA_ARGS__);          \
4549         } while (0)
4550
4551 static int bpf_map_do_batch(const union bpf_attr *attr,
4552                             union bpf_attr __user *uattr,
4553                             int cmd)
4554 {
4555         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4556                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4557         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4558         struct bpf_map *map;
4559         int err, ufd;
4560         struct fd f;
4561
4562         if (CHECK_ATTR(BPF_MAP_BATCH))
4563                 return -EINVAL;
4564
4565         ufd = attr->batch.map_fd;
4566         f = fdget(ufd);
4567         map = __bpf_map_get(f);
4568         if (IS_ERR(map))
4569                 return PTR_ERR(map);
4570         if (has_write)
4571                 bpf_map_write_active_inc(map);
4572         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4573                 err = -EPERM;
4574                 goto err_put;
4575         }
4576         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4577                 err = -EPERM;
4578                 goto err_put;
4579         }
4580
4581         if (cmd == BPF_MAP_LOOKUP_BATCH)
4582                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
4583         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4584                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
4585         else if (cmd == BPF_MAP_UPDATE_BATCH)
4586                 BPF_DO_BATCH(map->ops->map_update_batch, map, f.file, attr, uattr);
4587         else
4588                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
4589 err_put:
4590         if (has_write)
4591                 bpf_map_write_active_dec(map);
4592         fdput(f);
4593         return err;
4594 }
4595
4596 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4597 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4598 {
4599         enum bpf_prog_type ptype;
4600         struct bpf_prog *prog;
4601         int ret;
4602
4603         if (CHECK_ATTR(BPF_LINK_CREATE))
4604                 return -EINVAL;
4605
4606         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
4607                 return bpf_struct_ops_link_create(attr);
4608
4609         prog = bpf_prog_get(attr->link_create.prog_fd);
4610         if (IS_ERR(prog))
4611                 return PTR_ERR(prog);
4612
4613         ret = bpf_prog_attach_check_attach_type(prog,
4614                                                 attr->link_create.attach_type);
4615         if (ret)
4616                 goto out;
4617
4618         switch (prog->type) {
4619         case BPF_PROG_TYPE_EXT:
4620                 break;
4621         case BPF_PROG_TYPE_NETFILTER:
4622                 if (attr->link_create.attach_type != BPF_NETFILTER) {
4623                         ret = -EINVAL;
4624                         goto out;
4625                 }
4626                 break;
4627         case BPF_PROG_TYPE_PERF_EVENT:
4628         case BPF_PROG_TYPE_TRACEPOINT:
4629                 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4630                         ret = -EINVAL;
4631                         goto out;
4632                 }
4633                 break;
4634         case BPF_PROG_TYPE_KPROBE:
4635                 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4636                     attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4637                         ret = -EINVAL;
4638                         goto out;
4639                 }
4640                 break;
4641         default:
4642                 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4643                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4644                         ret = -EINVAL;
4645                         goto out;
4646                 }
4647                 break;
4648         }
4649
4650         switch (prog->type) {
4651         case BPF_PROG_TYPE_CGROUP_SKB:
4652         case BPF_PROG_TYPE_CGROUP_SOCK:
4653         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4654         case BPF_PROG_TYPE_SOCK_OPS:
4655         case BPF_PROG_TYPE_CGROUP_DEVICE:
4656         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4657         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4658                 ret = cgroup_bpf_link_attach(attr, prog);
4659                 break;
4660         case BPF_PROG_TYPE_EXT:
4661                 ret = bpf_tracing_prog_attach(prog,
4662                                               attr->link_create.target_fd,
4663                                               attr->link_create.target_btf_id,
4664                                               attr->link_create.tracing.cookie);
4665                 break;
4666         case BPF_PROG_TYPE_LSM:
4667         case BPF_PROG_TYPE_TRACING:
4668                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4669                         ret = -EINVAL;
4670                         goto out;
4671                 }
4672                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4673                         ret = bpf_raw_tp_link_attach(prog, NULL);
4674                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4675                         ret = bpf_iter_link_attach(attr, uattr, prog);
4676                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4677                         ret = cgroup_bpf_link_attach(attr, prog);
4678                 else
4679                         ret = bpf_tracing_prog_attach(prog,
4680                                                       attr->link_create.target_fd,
4681                                                       attr->link_create.target_btf_id,
4682                                                       attr->link_create.tracing.cookie);
4683                 break;
4684         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4685         case BPF_PROG_TYPE_SK_LOOKUP:
4686                 ret = netns_bpf_link_create(attr, prog);
4687                 break;
4688 #ifdef CONFIG_NET
4689         case BPF_PROG_TYPE_XDP:
4690                 ret = bpf_xdp_link_attach(attr, prog);
4691                 break;
4692         case BPF_PROG_TYPE_NETFILTER:
4693                 ret = bpf_nf_link_attach(attr, prog);
4694                 break;
4695 #endif
4696         case BPF_PROG_TYPE_PERF_EVENT:
4697         case BPF_PROG_TYPE_TRACEPOINT:
4698                 ret = bpf_perf_link_attach(attr, prog);
4699                 break;
4700         case BPF_PROG_TYPE_KPROBE:
4701                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4702                         ret = bpf_perf_link_attach(attr, prog);
4703                 else
4704                         ret = bpf_kprobe_multi_link_attach(attr, prog);
4705                 break;
4706         default:
4707                 ret = -EINVAL;
4708         }
4709
4710 out:
4711         if (ret < 0)
4712                 bpf_prog_put(prog);
4713         return ret;
4714 }
4715
4716 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
4717 {
4718         struct bpf_map *new_map, *old_map = NULL;
4719         int ret;
4720
4721         new_map = bpf_map_get(attr->link_update.new_map_fd);
4722         if (IS_ERR(new_map))
4723                 return PTR_ERR(new_map);
4724
4725         if (attr->link_update.flags & BPF_F_REPLACE) {
4726                 old_map = bpf_map_get(attr->link_update.old_map_fd);
4727                 if (IS_ERR(old_map)) {
4728                         ret = PTR_ERR(old_map);
4729                         goto out_put;
4730                 }
4731         } else if (attr->link_update.old_map_fd) {
4732                 ret = -EINVAL;
4733                 goto out_put;
4734         }
4735
4736         ret = link->ops->update_map(link, new_map, old_map);
4737
4738         if (old_map)
4739                 bpf_map_put(old_map);
4740 out_put:
4741         bpf_map_put(new_map);
4742         return ret;
4743 }
4744
4745 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4746
4747 static int link_update(union bpf_attr *attr)
4748 {
4749         struct bpf_prog *old_prog = NULL, *new_prog;
4750         struct bpf_link *link;
4751         u32 flags;
4752         int ret;
4753
4754         if (CHECK_ATTR(BPF_LINK_UPDATE))
4755                 return -EINVAL;
4756
4757         flags = attr->link_update.flags;
4758         if (flags & ~BPF_F_REPLACE)
4759                 return -EINVAL;
4760
4761         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4762         if (IS_ERR(link))
4763                 return PTR_ERR(link);
4764
4765         if (link->ops->update_map) {
4766                 ret = link_update_map(link, attr);
4767                 goto out_put_link;
4768         }
4769
4770         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4771         if (IS_ERR(new_prog)) {
4772                 ret = PTR_ERR(new_prog);
4773                 goto out_put_link;
4774         }
4775
4776         if (flags & BPF_F_REPLACE) {
4777                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4778                 if (IS_ERR(old_prog)) {
4779                         ret = PTR_ERR(old_prog);
4780                         old_prog = NULL;
4781                         goto out_put_progs;
4782                 }
4783         } else if (attr->link_update.old_prog_fd) {
4784                 ret = -EINVAL;
4785                 goto out_put_progs;
4786         }
4787
4788         if (link->ops->update_prog)
4789                 ret = link->ops->update_prog(link, new_prog, old_prog);
4790         else
4791                 ret = -EINVAL;
4792
4793 out_put_progs:
4794         if (old_prog)
4795                 bpf_prog_put(old_prog);
4796         if (ret)
4797                 bpf_prog_put(new_prog);
4798 out_put_link:
4799         bpf_link_put(link);
4800         return ret;
4801 }
4802
4803 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4804
4805 static int link_detach(union bpf_attr *attr)
4806 {
4807         struct bpf_link *link;
4808         int ret;
4809
4810         if (CHECK_ATTR(BPF_LINK_DETACH))
4811                 return -EINVAL;
4812
4813         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4814         if (IS_ERR(link))
4815                 return PTR_ERR(link);
4816
4817         if (link->ops->detach)
4818                 ret = link->ops->detach(link);
4819         else
4820                 ret = -EOPNOTSUPP;
4821
4822         bpf_link_put(link);
4823         return ret;
4824 }
4825
4826 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4827 {
4828         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4829 }
4830
4831 struct bpf_link *bpf_link_by_id(u32 id)
4832 {
4833         struct bpf_link *link;
4834
4835         if (!id)
4836                 return ERR_PTR(-ENOENT);
4837
4838         spin_lock_bh(&link_idr_lock);
4839         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4840         link = idr_find(&link_idr, id);
4841         if (link) {
4842                 if (link->id)
4843                         link = bpf_link_inc_not_zero(link);
4844                 else
4845                         link = ERR_PTR(-EAGAIN);
4846         } else {
4847                 link = ERR_PTR(-ENOENT);
4848         }
4849         spin_unlock_bh(&link_idr_lock);
4850         return link;
4851 }
4852
4853 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4854 {
4855         struct bpf_link *link;
4856
4857         spin_lock_bh(&link_idr_lock);
4858 again:
4859         link = idr_get_next(&link_idr, id);
4860         if (link) {
4861                 link = bpf_link_inc_not_zero(link);
4862                 if (IS_ERR(link)) {
4863                         (*id)++;
4864                         goto again;
4865                 }
4866         }
4867         spin_unlock_bh(&link_idr_lock);
4868
4869         return link;
4870 }
4871
4872 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4873
4874 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4875 {
4876         struct bpf_link *link;
4877         u32 id = attr->link_id;
4878         int fd;
4879
4880         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4881                 return -EINVAL;
4882
4883         if (!capable(CAP_SYS_ADMIN))
4884                 return -EPERM;
4885
4886         link = bpf_link_by_id(id);
4887         if (IS_ERR(link))
4888                 return PTR_ERR(link);
4889
4890         fd = bpf_link_new_fd(link);
4891         if (fd < 0)
4892                 bpf_link_put(link);
4893
4894         return fd;
4895 }
4896
4897 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4898
4899 static int bpf_stats_release(struct inode *inode, struct file *file)
4900 {
4901         mutex_lock(&bpf_stats_enabled_mutex);
4902         static_key_slow_dec(&bpf_stats_enabled_key.key);
4903         mutex_unlock(&bpf_stats_enabled_mutex);
4904         return 0;
4905 }
4906
4907 static const struct file_operations bpf_stats_fops = {
4908         .release = bpf_stats_release,
4909 };
4910
4911 static int bpf_enable_runtime_stats(void)
4912 {
4913         int fd;
4914
4915         mutex_lock(&bpf_stats_enabled_mutex);
4916
4917         /* Set a very high limit to avoid overflow */
4918         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4919                 mutex_unlock(&bpf_stats_enabled_mutex);
4920                 return -EBUSY;
4921         }
4922
4923         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4924         if (fd >= 0)
4925                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4926
4927         mutex_unlock(&bpf_stats_enabled_mutex);
4928         return fd;
4929 }
4930
4931 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4932
4933 static int bpf_enable_stats(union bpf_attr *attr)
4934 {
4935
4936         if (CHECK_ATTR(BPF_ENABLE_STATS))
4937                 return -EINVAL;
4938
4939         if (!capable(CAP_SYS_ADMIN))
4940                 return -EPERM;
4941
4942         switch (attr->enable_stats.type) {
4943         case BPF_STATS_RUN_TIME:
4944                 return bpf_enable_runtime_stats();
4945         default:
4946                 break;
4947         }
4948         return -EINVAL;
4949 }
4950
4951 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4952
4953 static int bpf_iter_create(union bpf_attr *attr)
4954 {
4955         struct bpf_link *link;
4956         int err;
4957
4958         if (CHECK_ATTR(BPF_ITER_CREATE))
4959                 return -EINVAL;
4960
4961         if (attr->iter_create.flags)
4962                 return -EINVAL;
4963
4964         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4965         if (IS_ERR(link))
4966                 return PTR_ERR(link);
4967
4968         err = bpf_iter_new_fd(link);
4969         bpf_link_put(link);
4970
4971         return err;
4972 }
4973
4974 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4975
4976 static int bpf_prog_bind_map(union bpf_attr *attr)
4977 {
4978         struct bpf_prog *prog;
4979         struct bpf_map *map;
4980         struct bpf_map **used_maps_old, **used_maps_new;
4981         int i, ret = 0;
4982
4983         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4984                 return -EINVAL;
4985
4986         if (attr->prog_bind_map.flags)
4987                 return -EINVAL;
4988
4989         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4990         if (IS_ERR(prog))
4991                 return PTR_ERR(prog);
4992
4993         map = bpf_map_get(attr->prog_bind_map.map_fd);
4994         if (IS_ERR(map)) {
4995                 ret = PTR_ERR(map);
4996                 goto out_prog_put;
4997         }
4998
4999         mutex_lock(&prog->aux->used_maps_mutex);
5000
5001         used_maps_old = prog->aux->used_maps;
5002
5003         for (i = 0; i < prog->aux->used_map_cnt; i++)
5004                 if (used_maps_old[i] == map) {
5005                         bpf_map_put(map);
5006                         goto out_unlock;
5007                 }
5008
5009         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5010                                       sizeof(used_maps_new[0]),
5011                                       GFP_KERNEL);
5012         if (!used_maps_new) {
5013                 ret = -ENOMEM;
5014                 goto out_unlock;
5015         }
5016
5017         memcpy(used_maps_new, used_maps_old,
5018                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5019         used_maps_new[prog->aux->used_map_cnt] = map;
5020
5021         prog->aux->used_map_cnt++;
5022         prog->aux->used_maps = used_maps_new;
5023
5024         kfree(used_maps_old);
5025
5026 out_unlock:
5027         mutex_unlock(&prog->aux->used_maps_mutex);
5028
5029         if (ret)
5030                 bpf_map_put(map);
5031 out_prog_put:
5032         bpf_prog_put(prog);
5033         return ret;
5034 }
5035
5036 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
5037 {
5038         union bpf_attr attr;
5039         bool capable;
5040         int err;
5041
5042         capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
5043
5044         /* Intent here is for unprivileged_bpf_disabled to block key object
5045          * creation commands for unprivileged users; other actions depend
5046          * of fd availability and access to bpffs, so are dependent on
5047          * object creation success.  Capabilities are later verified for
5048          * operations such as load and map create, so even with unprivileged
5049          * BPF disabled, capability checks are still carried out for these
5050          * and other operations.
5051          */
5052         if (!capable &&
5053             (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
5054                 return -EPERM;
5055
5056         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5057         if (err)
5058                 return err;
5059         size = min_t(u32, size, sizeof(attr));
5060
5061         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5062         memset(&attr, 0, sizeof(attr));
5063         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5064                 return -EFAULT;
5065
5066         err = security_bpf(cmd, &attr, size);
5067         if (err < 0)
5068                 return err;
5069
5070         switch (cmd) {
5071         case BPF_MAP_CREATE:
5072                 err = map_create(&attr);
5073                 break;
5074         case BPF_MAP_LOOKUP_ELEM:
5075                 err = map_lookup_elem(&attr);
5076                 break;
5077         case BPF_MAP_UPDATE_ELEM:
5078                 err = map_update_elem(&attr, uattr);
5079                 break;
5080         case BPF_MAP_DELETE_ELEM:
5081                 err = map_delete_elem(&attr, uattr);
5082                 break;
5083         case BPF_MAP_GET_NEXT_KEY:
5084                 err = map_get_next_key(&attr);
5085                 break;
5086         case BPF_MAP_FREEZE:
5087                 err = map_freeze(&attr);
5088                 break;
5089         case BPF_PROG_LOAD:
5090                 err = bpf_prog_load(&attr, uattr, size);
5091                 break;
5092         case BPF_OBJ_PIN:
5093                 err = bpf_obj_pin(&attr);
5094                 break;
5095         case BPF_OBJ_GET:
5096                 err = bpf_obj_get(&attr);
5097                 break;
5098         case BPF_PROG_ATTACH:
5099                 err = bpf_prog_attach(&attr);
5100                 break;
5101         case BPF_PROG_DETACH:
5102                 err = bpf_prog_detach(&attr);
5103                 break;
5104         case BPF_PROG_QUERY:
5105                 err = bpf_prog_query(&attr, uattr.user);
5106                 break;
5107         case BPF_PROG_TEST_RUN:
5108                 err = bpf_prog_test_run(&attr, uattr.user);
5109                 break;
5110         case BPF_PROG_GET_NEXT_ID:
5111                 err = bpf_obj_get_next_id(&attr, uattr.user,
5112                                           &prog_idr, &prog_idr_lock);
5113                 break;
5114         case BPF_MAP_GET_NEXT_ID:
5115                 err = bpf_obj_get_next_id(&attr, uattr.user,
5116                                           &map_idr, &map_idr_lock);
5117                 break;
5118         case BPF_BTF_GET_NEXT_ID:
5119                 err = bpf_obj_get_next_id(&attr, uattr.user,
5120                                           &btf_idr, &btf_idr_lock);
5121                 break;
5122         case BPF_PROG_GET_FD_BY_ID:
5123                 err = bpf_prog_get_fd_by_id(&attr);
5124                 break;
5125         case BPF_MAP_GET_FD_BY_ID:
5126                 err = bpf_map_get_fd_by_id(&attr);
5127                 break;
5128         case BPF_OBJ_GET_INFO_BY_FD:
5129                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5130                 break;
5131         case BPF_RAW_TRACEPOINT_OPEN:
5132                 err = bpf_raw_tracepoint_open(&attr);
5133                 break;
5134         case BPF_BTF_LOAD:
5135                 err = bpf_btf_load(&attr, uattr, size);
5136                 break;
5137         case BPF_BTF_GET_FD_BY_ID:
5138                 err = bpf_btf_get_fd_by_id(&attr);
5139                 break;
5140         case BPF_TASK_FD_QUERY:
5141                 err = bpf_task_fd_query(&attr, uattr.user);
5142                 break;
5143         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5144                 err = map_lookup_and_delete_elem(&attr);
5145                 break;
5146         case BPF_MAP_LOOKUP_BATCH:
5147                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5148                 break;
5149         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5150                 err = bpf_map_do_batch(&attr, uattr.user,
5151                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5152                 break;
5153         case BPF_MAP_UPDATE_BATCH:
5154                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5155                 break;
5156         case BPF_MAP_DELETE_BATCH:
5157                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5158                 break;
5159         case BPF_LINK_CREATE:
5160                 err = link_create(&attr, uattr);
5161                 break;
5162         case BPF_LINK_UPDATE:
5163                 err = link_update(&attr);
5164                 break;
5165         case BPF_LINK_GET_FD_BY_ID:
5166                 err = bpf_link_get_fd_by_id(&attr);
5167                 break;
5168         case BPF_LINK_GET_NEXT_ID:
5169                 err = bpf_obj_get_next_id(&attr, uattr.user,
5170                                           &link_idr, &link_idr_lock);
5171                 break;
5172         case BPF_ENABLE_STATS:
5173                 err = bpf_enable_stats(&attr);
5174                 break;
5175         case BPF_ITER_CREATE:
5176                 err = bpf_iter_create(&attr);
5177                 break;
5178         case BPF_LINK_DETACH:
5179                 err = link_detach(&attr);
5180                 break;
5181         case BPF_PROG_BIND_MAP:
5182                 err = bpf_prog_bind_map(&attr);
5183                 break;
5184         default:
5185                 err = -EINVAL;
5186                 break;
5187         }
5188
5189         return err;
5190 }
5191
5192 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5193 {
5194         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5195 }
5196
5197 static bool syscall_prog_is_valid_access(int off, int size,
5198                                          enum bpf_access_type type,
5199                                          const struct bpf_prog *prog,
5200                                          struct bpf_insn_access_aux *info)
5201 {
5202         if (off < 0 || off >= U16_MAX)
5203                 return false;
5204         if (off % size != 0)
5205                 return false;
5206         return true;
5207 }
5208
5209 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5210 {
5211         switch (cmd) {
5212         case BPF_MAP_CREATE:
5213         case BPF_MAP_DELETE_ELEM:
5214         case BPF_MAP_UPDATE_ELEM:
5215         case BPF_MAP_FREEZE:
5216         case BPF_MAP_GET_FD_BY_ID:
5217         case BPF_PROG_LOAD:
5218         case BPF_BTF_LOAD:
5219         case BPF_LINK_CREATE:
5220         case BPF_RAW_TRACEPOINT_OPEN:
5221                 break;
5222         default:
5223                 return -EINVAL;
5224         }
5225         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5226 }
5227
5228
5229 /* To shut up -Wmissing-prototypes.
5230  * This function is used by the kernel light skeleton
5231  * to load bpf programs when modules are loaded or during kernel boot.
5232  * See tools/lib/bpf/skel_internal.h
5233  */
5234 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5235
5236 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5237 {
5238         struct bpf_prog * __maybe_unused prog;
5239         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5240
5241         switch (cmd) {
5242 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5243         case BPF_PROG_TEST_RUN:
5244                 if (attr->test.data_in || attr->test.data_out ||
5245                     attr->test.ctx_out || attr->test.duration ||
5246                     attr->test.repeat || attr->test.flags)
5247                         return -EINVAL;
5248
5249                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5250                 if (IS_ERR(prog))
5251                         return PTR_ERR(prog);
5252
5253                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5254                     attr->test.ctx_size_in > U16_MAX) {
5255                         bpf_prog_put(prog);
5256                         return -EINVAL;
5257                 }
5258
5259                 run_ctx.bpf_cookie = 0;
5260                 run_ctx.saved_run_ctx = NULL;
5261                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5262                         /* recursion detected */
5263                         bpf_prog_put(prog);
5264                         return -EBUSY;
5265                 }
5266                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5267                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5268                                                 &run_ctx);
5269                 bpf_prog_put(prog);
5270                 return 0;
5271 #endif
5272         default:
5273                 return ____bpf_sys_bpf(cmd, attr, size);
5274         }
5275 }
5276 EXPORT_SYMBOL(kern_sys_bpf);
5277
5278 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5279         .func           = bpf_sys_bpf,
5280         .gpl_only       = false,
5281         .ret_type       = RET_INTEGER,
5282         .arg1_type      = ARG_ANYTHING,
5283         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5284         .arg3_type      = ARG_CONST_SIZE,
5285 };
5286
5287 const struct bpf_func_proto * __weak
5288 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5289 {
5290         return bpf_base_func_proto(func_id);
5291 }
5292
5293 BPF_CALL_1(bpf_sys_close, u32, fd)
5294 {
5295         /* When bpf program calls this helper there should not be
5296          * an fdget() without matching completed fdput().
5297          * This helper is allowed in the following callchain only:
5298          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5299          */
5300         return close_fd(fd);
5301 }
5302
5303 static const struct bpf_func_proto bpf_sys_close_proto = {
5304         .func           = bpf_sys_close,
5305         .gpl_only       = false,
5306         .ret_type       = RET_INTEGER,
5307         .arg1_type      = ARG_ANYTHING,
5308 };
5309
5310 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5311 {
5312         if (flags)
5313                 return -EINVAL;
5314
5315         if (name_sz <= 1 || name[name_sz - 1])
5316                 return -EINVAL;
5317
5318         if (!bpf_dump_raw_ok(current_cred()))
5319                 return -EPERM;
5320
5321         *res = kallsyms_lookup_name(name);
5322         return *res ? 0 : -ENOENT;
5323 }
5324
5325 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5326         .func           = bpf_kallsyms_lookup_name,
5327         .gpl_only       = false,
5328         .ret_type       = RET_INTEGER,
5329         .arg1_type      = ARG_PTR_TO_MEM,
5330         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5331         .arg3_type      = ARG_ANYTHING,
5332         .arg4_type      = ARG_PTR_TO_LONG,
5333 };
5334
5335 static const struct bpf_func_proto *
5336 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5337 {
5338         switch (func_id) {
5339         case BPF_FUNC_sys_bpf:
5340                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5341         case BPF_FUNC_btf_find_by_name_kind:
5342                 return &bpf_btf_find_by_name_kind_proto;
5343         case BPF_FUNC_sys_close:
5344                 return &bpf_sys_close_proto;
5345         case BPF_FUNC_kallsyms_lookup_name:
5346                 return &bpf_kallsyms_lookup_name_proto;
5347         default:
5348                 return tracing_prog_func_proto(func_id, prog);
5349         }
5350 }
5351
5352 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5353         .get_func_proto  = syscall_prog_func_proto,
5354         .is_valid_access = syscall_prog_is_valid_access,
5355 };
5356
5357 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5358         .test_run = bpf_prog_test_run_syscall,
5359 };
5360
5361 #ifdef CONFIG_SYSCTL
5362 static int bpf_stats_handler(struct ctl_table *table, int write,
5363                              void *buffer, size_t *lenp, loff_t *ppos)
5364 {
5365         struct static_key *key = (struct static_key *)table->data;
5366         static int saved_val;
5367         int val, ret;
5368         struct ctl_table tmp = {
5369                 .data   = &val,
5370                 .maxlen = sizeof(val),
5371                 .mode   = table->mode,
5372                 .extra1 = SYSCTL_ZERO,
5373                 .extra2 = SYSCTL_ONE,
5374         };
5375
5376         if (write && !capable(CAP_SYS_ADMIN))
5377                 return -EPERM;
5378
5379         mutex_lock(&bpf_stats_enabled_mutex);
5380         val = saved_val;
5381         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5382         if (write && !ret && val != saved_val) {
5383                 if (val)
5384                         static_key_slow_inc(key);
5385                 else
5386                         static_key_slow_dec(key);
5387                 saved_val = val;
5388         }
5389         mutex_unlock(&bpf_stats_enabled_mutex);
5390         return ret;
5391 }
5392
5393 void __weak unpriv_ebpf_notify(int new_state)
5394 {
5395 }
5396
5397 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5398                               void *buffer, size_t *lenp, loff_t *ppos)
5399 {
5400         int ret, unpriv_enable = *(int *)table->data;
5401         bool locked_state = unpriv_enable == 1;
5402         struct ctl_table tmp = *table;
5403
5404         if (write && !capable(CAP_SYS_ADMIN))
5405                 return -EPERM;
5406
5407         tmp.data = &unpriv_enable;
5408         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5409         if (write && !ret) {
5410                 if (locked_state && unpriv_enable != 1)
5411                         return -EPERM;
5412                 *(int *)table->data = unpriv_enable;
5413         }
5414
5415         if (write)
5416                 unpriv_ebpf_notify(unpriv_enable);
5417
5418         return ret;
5419 }
5420
5421 static struct ctl_table bpf_syscall_table[] = {
5422         {
5423                 .procname       = "unprivileged_bpf_disabled",
5424                 .data           = &sysctl_unprivileged_bpf_disabled,
5425                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5426                 .mode           = 0644,
5427                 .proc_handler   = bpf_unpriv_handler,
5428                 .extra1         = SYSCTL_ZERO,
5429                 .extra2         = SYSCTL_TWO,
5430         },
5431         {
5432                 .procname       = "bpf_stats_enabled",
5433                 .data           = &bpf_stats_enabled_key.key,
5434                 .mode           = 0644,
5435                 .proc_handler   = bpf_stats_handler,
5436         },
5437         { }
5438 };
5439
5440 static int __init bpf_syscall_sysctl_init(void)
5441 {
5442         register_sysctl_init("kernel", bpf_syscall_table);
5443         return 0;
5444 }
5445 late_initcall(bpf_syscall_sysctl_init);
5446 #endif /* CONFIG_SYSCTL */