kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf-cgroup.h>
   6 #include <linux/bpf_trace.h>
   7 #include <linux/bpf_lirc.h>
   8 #include <linux/bpf_verifier.h>
   9 #include <linux/bsearch.h>
  10 #include <linux/btf.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mmzone.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/fdtable.h>
  18 #include <linux/file.h>
  19 #include <linux/fs.h>
  20 #include <linux/license.h>
  21 #include <linux/filter.h>
  22 #include <linux/kernel.h>
  23 #include <linux/idr.h>
  24 #include <linux/cred.h>
  25 #include <linux/timekeeping.h>
  26 #include <linux/ctype.h>
  27 #include <linux/nospec.h>
  28 #include <linux/audit.h>
  29 #include <uapi/linux/btf.h>
  30 #include <linux/pgtable.h>
  31 #include <linux/bpf_lsm.h>
  32 #include <linux/poll.h>
  33 #include <linux/sort.h>
  34 #include <linux/bpf-netns.h>
  35 #include <linux/rcupdate_trace.h>
  36 #include <linux/memcontrol.h>
  37 #include <linux/trace_events.h>
  38
  39 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  40                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  41                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  42 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  43 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  44 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  45                         IS_FD_HASH(map))
  46
  47 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  48
  49 DEFINE_PER_CPU(int, bpf_prog_active);
  50 static DEFINE_IDR(prog_idr);
  51 static DEFINE_SPINLOCK(prog_idr_lock);
  52 static DEFINE_IDR(map_idr);
  53 static DEFINE_SPINLOCK(map_idr_lock);
  54 static DEFINE_IDR(link_idr);
  55 static DEFINE_SPINLOCK(link_idr_lock);
  56
  57 int sysctl_unprivileged_bpf_disabled __read_mostly =
  58         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  59
  60 static const struct bpf_map_ops * const bpf_map_types[] = {
  61 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  62 #define BPF_MAP_TYPE(_id, _ops) \
  63         [_id] = &_ops,
  64 #define BPF_LINK_TYPE(_id, _name)
  65 #include <linux/bpf_types.h>
  66 #undef BPF_PROG_TYPE
  67 #undef BPF_MAP_TYPE
  68 #undef BPF_LINK_TYPE
  69 };
  70
  71 /*
  72  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  73  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  74  * we don't know about yet.
  75  *
  76  * There is a ToCToU between this function call and the following
  77  * copy_from_user() call. However, this is not a concern since this function is
  78  * meant to be a future-proofing of bits.
  79  */
  80 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  81                              size_t expected_size,
  82                              size_t actual_size)
  83 {
  84         int res;
  85
  86         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  87                 return -E2BIG;
  88
  89         if (actual_size <= expected_size)
  90                 return 0;
  91
  92         if (uaddr.is_kernel)
  93                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  94                                  actual_size - expected_size) == NULL;
  95         else
  96                 res = check_zeroed_user(uaddr.user + expected_size,
  97                                         actual_size - expected_size);
  98         if (res < 0)
  99                 return res;
 100         return res ? 0 : -E2BIG;
 101 }
 102
 103 const struct bpf_map_ops bpf_map_offload_ops = {
 104         .map_meta_equal = bpf_map_meta_equal,
 105         .map_alloc = bpf_map_offload_map_alloc,
 106         .map_free = bpf_map_offload_map_free,
 107         .map_check_btf = map_check_no_btf,
 108 };
 109
 110 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 111 {
 112         const struct bpf_map_ops *ops;
 113         u32 type = attr->map_type;
 114         struct bpf_map *map;
 115         int err;
 116
 117         if (type >= ARRAY_SIZE(bpf_map_types))
 118                 return ERR_PTR(-EINVAL);
 119         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 120         ops = bpf_map_types[type];
 121         if (!ops)
 122                 return ERR_PTR(-EINVAL);
 123
 124         if (ops->map_alloc_check) {
 125                 err = ops->map_alloc_check(attr);
 126                 if (err)
 127                         return ERR_PTR(err);
 128         }
 129         if (attr->map_ifindex)
 130                 ops = &bpf_map_offload_ops;
 131         map = ops->map_alloc(attr);
 132         if (IS_ERR(map))
 133                 return map;
 134         map->ops = ops;
 135         map->map_type = type;
 136         return map;
 137 }
 138
 139 static void bpf_map_write_active_inc(struct bpf_map *map)
 140 {
 141         atomic64_inc(&map->writecnt);
 142 }
 143
 144 static void bpf_map_write_active_dec(struct bpf_map *map)
 145 {
 146         atomic64_dec(&map->writecnt);
 147 }
 148
 149 bool bpf_map_write_active(const struct bpf_map *map)
 150 {
 151         return atomic64_read(&map->writecnt) != 0;
 152 }
 153
 154 static u32 bpf_map_value_size(const struct bpf_map *map)
 155 {
 156         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 157             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 158             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 159             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 160                 return round_up(map->value_size, 8) * num_possible_cpus();
 161         else if (IS_FD_MAP(map))
 162                 return sizeof(u32);
 163         else
 164                 return  map->value_size;
 165 }
 166
 167 static void maybe_wait_bpf_programs(struct bpf_map *map)
 168 {
 169         /* Wait for any running BPF programs to complete so that
 170          * userspace, when we return to it, knows that all programs
 171          * that could be running use the new map value.
 172          */
 173         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 174             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 175                 synchronize_rcu();
 176 }
 177
 178 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
 179                                 void *value, __u64 flags)
 180 {
 181         int err;
 182
 183         /* Need to create a kthread, thus must support schedule */
 184         if (bpf_map_is_dev_bound(map)) {
 185                 return bpf_map_offload_update_elem(map, key, value, flags);
 186         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 187                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 188                 return map->ops->map_update_elem(map, key, value, flags);
 189         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 190                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 191                 return sock_map_update_elem_sys(map, key, value, flags);
 192         } else if (IS_FD_PROG_ARRAY(map)) {
 193                 return bpf_fd_array_map_update_elem(map, f.file, key, value,
 194                                                     flags);
 195         }
 196
 197         bpf_disable_instrumentation();
 198         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 199             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 200                 err = bpf_percpu_hash_update(map, key, value, flags);
 201         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 202                 err = bpf_percpu_array_update(map, key, value, flags);
 203         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 204                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 205                                                        flags);
 206         } else if (IS_FD_ARRAY(map)) {
 207                 rcu_read_lock();
 208                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 209                                                    flags);
 210                 rcu_read_unlock();
 211         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 212                 rcu_read_lock();
 213                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 214                                                   flags);
 215                 rcu_read_unlock();
 216         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 217                 /* rcu_read_lock() is not needed */
 218                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 219                                                          flags);
 220         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 221                    map->map_type == BPF_MAP_TYPE_STACK ||
 222                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 223                 err = map->ops->map_push_elem(map, value, flags);
 224         } else {
 225                 rcu_read_lock();
 226                 err = map->ops->map_update_elem(map, key, value, flags);
 227                 rcu_read_unlock();
 228         }
 229         bpf_enable_instrumentation();
 230         maybe_wait_bpf_programs(map);
 231
 232         return err;
 233 }
 234
 235 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 236                               __u64 flags)
 237 {
 238         void *ptr;
 239         int err;
 240
 241         if (bpf_map_is_dev_bound(map))
 242                 return bpf_map_offload_lookup_elem(map, key, value);
 243
 244         bpf_disable_instrumentation();
 245         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 246             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 247                 err = bpf_percpu_hash_copy(map, key, value);
 248         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 249                 err = bpf_percpu_array_copy(map, key, value);
 250         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 251                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 252         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 253                 err = bpf_stackmap_copy(map, key, value);
 254         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 255                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 256         } else if (IS_FD_HASH(map)) {
 257                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 258         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 259                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 260         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 261                    map->map_type == BPF_MAP_TYPE_STACK ||
 262                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 263                 err = map->ops->map_peek_elem(map, value);
 264         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 265                 /* struct_ops map requires directly updating "value" */
 266                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 267         } else {
 268                 rcu_read_lock();
 269                 if (map->ops->map_lookup_elem_sys_only)
 270                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 271                 else
 272                         ptr = map->ops->map_lookup_elem(map, key);
 273                 if (IS_ERR(ptr)) {
 274                         err = PTR_ERR(ptr);
 275                 } else if (!ptr) {
 276                         err = -ENOENT;
 277                 } else {
 278                         err = 0;
 279                         if (flags & BPF_F_LOCK)
 280                                 /* lock 'ptr' and copy everything but lock */
 281                                 copy_map_value_locked(map, value, ptr, true);
 282                         else
 283                                 copy_map_value(map, value, ptr);
 284                         /* mask lock and timer, since value wasn't zero inited */
 285                         check_and_init_map_value(map, value);
 286                 }
 287                 rcu_read_unlock();
 288         }
 289
 290         bpf_enable_instrumentation();
 291         maybe_wait_bpf_programs(map);
 292
 293         return err;
 294 }
 295
 296 /* Please, do not use this function outside from the map creation path
 297  * (e.g. in map update path) without taking care of setting the active
 298  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 299  */
 300 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 301 {
 302         /* We really just want to fail instead of triggering OOM killer
 303          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 304          * which is used for lower order allocation requests.
 305          *
 306          * It has been observed that higher order allocation requests done by
 307          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 308          * to reclaim memory from the page cache, thus we set
 309          * __GFP_RETRY_MAYFAIL to avoid such situations.
 310          */
 311
 312         const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
 313         unsigned int flags = 0;
 314         unsigned long align = 1;
 315         void *area;
 316
 317         if (size >= SIZE_MAX)
 318                 return NULL;
 319
 320         /* kmalloc()'ed memory can't be mmap()'ed */
 321         if (mmapable) {
 322                 BUG_ON(!PAGE_ALIGNED(size));
 323                 align = SHMLBA;
 324                 flags = VM_USERMAP;
 325         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 326                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 327                                     numa_node);
 328                 if (area != NULL)
 329                         return area;
 330         }
 331
 332         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 333                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 334                         flags, numa_node, __builtin_return_address(0));
 335 }
 336
 337 void *bpf_map_area_alloc(u64 size, int numa_node)
 338 {
 339         return __bpf_map_area_alloc(size, numa_node, false);
 340 }
 341
 342 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 343 {
 344         return __bpf_map_area_alloc(size, numa_node, true);
 345 }
 346
 347 void bpf_map_area_free(void *area)
 348 {
 349         kvfree(area);
 350 }
 351
 352 static u32 bpf_map_flags_retain_permanent(u32 flags)
 353 {
 354         /* Some map creation flags are not tied to the map object but
 355          * rather to the map fd instead, so they have no meaning upon
 356          * map object inspection since multiple file descriptors with
 357          * different (access) properties can exist here. Thus, given
 358          * this has zero meaning for the map itself, lets clear these
 359          * from here.
 360          */
 361         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 362 }
 363
 364 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 365 {
 366         map->map_type = attr->map_type;
 367         map->key_size = attr->key_size;
 368         map->value_size = attr->value_size;
 369         map->max_entries = attr->max_entries;
 370         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 371         map->numa_node = bpf_map_attr_numa_node(attr);
 372         map->map_extra = attr->map_extra;
 373 }
 374
 375 static int bpf_map_alloc_id(struct bpf_map *map)
 376 {
 377         int id;
 378
 379         idr_preload(GFP_KERNEL);
 380         spin_lock_bh(&map_idr_lock);
 381         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 382         if (id > 0)
 383                 map->id = id;
 384         spin_unlock_bh(&map_idr_lock);
 385         idr_preload_end();
 386
 387         if (WARN_ON_ONCE(!id))
 388                 return -ENOSPC;
 389
 390         return id > 0 ? 0 : id;
 391 }
 392
 393 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 394 {
 395         unsigned long flags;
 396
 397         /* Offloaded maps are removed from the IDR store when their device
 398          * disappears - even if someone holds an fd to them they are unusable,
 399          * the memory is gone, all ops will fail; they are simply waiting for
 400          * refcnt to drop to be freed.
 401          */
 402         if (!map->id)
 403                 return;
 404
 405         if (do_idr_lock)
 406                 spin_lock_irqsave(&map_idr_lock, flags);
 407         else
 408                 __acquire(&map_idr_lock);
 409
 410         idr_remove(&map_idr, map->id);
 411         map->id = 0;
 412
 413         if (do_idr_lock)
 414                 spin_unlock_irqrestore(&map_idr_lock, flags);
 415         else
 416                 __release(&map_idr_lock);
 417 }
 418
 419 #ifdef CONFIG_MEMCG_KMEM
 420 static void bpf_map_save_memcg(struct bpf_map *map)
 421 {
 422         /* Currently if a map is created by a process belonging to the root
 423          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 424          * So we have to check map->objcg for being NULL each time it's
 425          * being used.
 426          */
 427         map->objcg = get_obj_cgroup_from_current();
 428 }
 429
 430 static void bpf_map_release_memcg(struct bpf_map *map)
 431 {
 432         if (map->objcg)
 433                 obj_cgroup_put(map->objcg);
 434 }
 435
 436 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 437 {
 438         if (map->objcg)
 439                 return get_mem_cgroup_from_objcg(map->objcg);
 440
 441         return root_mem_cgroup;
 442 }
 443
 444 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 445                            int node)
 446 {
 447         struct mem_cgroup *memcg, *old_memcg;
 448         void *ptr;
 449
 450         memcg = bpf_map_get_memcg(map);
 451         old_memcg = set_active_memcg(memcg);
 452         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 453         set_active_memcg(old_memcg);
 454         mem_cgroup_put(memcg);
 455
 456         return ptr;
 457 }
 458
 459 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 460 {
 461         struct mem_cgroup *memcg, *old_memcg;
 462         void *ptr;
 463
 464         memcg = bpf_map_get_memcg(map);
 465         old_memcg = set_active_memcg(memcg);
 466         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 467         set_active_memcg(old_memcg);
 468         mem_cgroup_put(memcg);
 469
 470         return ptr;
 471 }
 472
 473 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 474                                     size_t align, gfp_t flags)
 475 {
 476         struct mem_cgroup *memcg, *old_memcg;
 477         void __percpu *ptr;
 478
 479         memcg = bpf_map_get_memcg(map);
 480         old_memcg = set_active_memcg(memcg);
 481         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 482         set_active_memcg(old_memcg);
 483         mem_cgroup_put(memcg);
 484
 485         return ptr;
 486 }
 487
 488 #else
 489 static void bpf_map_save_memcg(struct bpf_map *map)
 490 {
 491 }
 492
 493 static void bpf_map_release_memcg(struct bpf_map *map)
 494 {
 495 }
 496 #endif
 497
 498 static int bpf_map_kptr_off_cmp(const void *a, const void *b)
 499 {
 500         const struct bpf_map_value_off_desc *off_desc1 = a, *off_desc2 = b;
 501
 502         if (off_desc1->offset < off_desc2->offset)
 503                 return -1;
 504         else if (off_desc1->offset > off_desc2->offset)
 505                 return 1;
 506         return 0;
 507 }
 508
 509 struct bpf_map_value_off_desc *bpf_map_kptr_off_contains(struct bpf_map *map, u32 offset)
 510 {
 511         /* Since members are iterated in btf_find_field in increasing order,
 512          * offsets appended to kptr_off_tab are in increasing order, so we can
 513          * do bsearch to find exact match.
 514          */
 515         struct bpf_map_value_off *tab;
 516
 517         if (!map_value_has_kptrs(map))
 518                 return NULL;
 519         tab = map->kptr_off_tab;
 520         return bsearch(&offset, tab->off, tab->nr_off, sizeof(tab->off[0]), bpf_map_kptr_off_cmp);
 521 }
 522
 523 void bpf_map_free_kptr_off_tab(struct bpf_map *map)
 524 {
 525         struct bpf_map_value_off *tab = map->kptr_off_tab;
 526         int i;
 527
 528         if (!map_value_has_kptrs(map))
 529                 return;
 530         for (i = 0; i < tab->nr_off; i++) {
 531                 if (tab->off[i].kptr.module)
 532                         module_put(tab->off[i].kptr.module);
 533                 btf_put(tab->off[i].kptr.btf);
 534         }
 535         kfree(tab);
 536         map->kptr_off_tab = NULL;
 537 }
 538
 539 struct bpf_map_value_off *bpf_map_copy_kptr_off_tab(const struct bpf_map *map)
 540 {
 541         struct bpf_map_value_off *tab = map->kptr_off_tab, *new_tab;
 542         int size, i;
 543
 544         if (!map_value_has_kptrs(map))
 545                 return ERR_PTR(-ENOENT);
 546         size = offsetof(struct bpf_map_value_off, off[tab->nr_off]);
 547         new_tab = kmemdup(tab, size, GFP_KERNEL | __GFP_NOWARN);
 548         if (!new_tab)
 549                 return ERR_PTR(-ENOMEM);
 550         /* Do a deep copy of the kptr_off_tab */
 551         for (i = 0; i < tab->nr_off; i++) {
 552                 btf_get(tab->off[i].kptr.btf);
 553                 if (tab->off[i].kptr.module && !try_module_get(tab->off[i].kptr.module)) {
 554                         while (i--) {
 555                                 if (tab->off[i].kptr.module)
 556                                         module_put(tab->off[i].kptr.module);
 557                                 btf_put(tab->off[i].kptr.btf);
 558                         }
 559                         kfree(new_tab);
 560                         return ERR_PTR(-ENXIO);
 561                 }
 562         }
 563         return new_tab;
 564 }
 565
 566 bool bpf_map_equal_kptr_off_tab(const struct bpf_map *map_a, const struct bpf_map *map_b)
 567 {
 568         struct bpf_map_value_off *tab_a = map_a->kptr_off_tab, *tab_b = map_b->kptr_off_tab;
 569         bool a_has_kptr = map_value_has_kptrs(map_a), b_has_kptr = map_value_has_kptrs(map_b);
 570         int size;
 571
 572         if (!a_has_kptr && !b_has_kptr)
 573                 return true;
 574         if (a_has_kptr != b_has_kptr)
 575                 return false;
 576         if (tab_a->nr_off != tab_b->nr_off)
 577                 return false;
 578         size = offsetof(struct bpf_map_value_off, off[tab_a->nr_off]);
 579         return !memcmp(tab_a, tab_b, size);
 580 }
 581
 582 /* Caller must ensure map_value_has_kptrs is true. Note that this function can
 583  * be called on a map value while the map_value is visible to BPF programs, as
 584  * it ensures the correct synchronization, and we already enforce the same using
 585  * the bpf_kptr_xchg helper on the BPF program side for referenced kptrs.
 586  */
 587 void bpf_map_free_kptrs(struct bpf_map *map, void *map_value)
 588 {
 589         struct bpf_map_value_off *tab = map->kptr_off_tab;
 590         unsigned long *btf_id_ptr;
 591         int i;
 592
 593         for (i = 0; i < tab->nr_off; i++) {
 594                 struct bpf_map_value_off_desc *off_desc = &tab->off[i];
 595                 unsigned long old_ptr;
 596
 597                 btf_id_ptr = map_value + off_desc->offset;
 598                 if (off_desc->type == BPF_KPTR_UNREF) {
 599                         u64 *p = (u64 *)btf_id_ptr;
 600
 601                         WRITE_ONCE(*p, 0);
 602                         continue;
 603                 }
 604                 old_ptr = xchg(btf_id_ptr, 0);
 605                 off_desc->kptr.dtor((void *)old_ptr);
 606         }
 607 }
 608
 609 /* called from workqueue */
 610 static void bpf_map_free_deferred(struct work_struct *work)
 611 {
 612         struct bpf_map *map = container_of(work, struct bpf_map, work);
 613
 614         security_bpf_map_free(map);
 615         kfree(map->off_arr);
 616         bpf_map_release_memcg(map);
 617         /* implementation dependent freeing, map_free callback also does
 618          * bpf_map_free_kptr_off_tab, if needed.
 619          */
 620         map->ops->map_free(map);
 621 }
 622
 623 static void bpf_map_put_uref(struct bpf_map *map)
 624 {
 625         if (atomic64_dec_and_test(&map->usercnt)) {
 626                 if (map->ops->map_release_uref)
 627                         map->ops->map_release_uref(map);
 628         }
 629 }
 630
 631 /* decrement map refcnt and schedule it for freeing via workqueue
 632  * (unrelying map implementation ops->map_free() might sleep)
 633  */
 634 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 635 {
 636         if (atomic64_dec_and_test(&map->refcnt)) {
 637                 /* bpf_map_free_id() must be called first */
 638                 bpf_map_free_id(map, do_idr_lock);
 639                 btf_put(map->btf);
 640                 INIT_WORK(&map->work, bpf_map_free_deferred);
 641                 /* Avoid spawning kworkers, since they all might contend
 642                  * for the same mutex like slab_mutex.
 643                  */
 644                 queue_work(system_unbound_wq, &map->work);
 645         }
 646 }
 647
 648 void bpf_map_put(struct bpf_map *map)
 649 {
 650         __bpf_map_put(map, true);
 651 }
 652 EXPORT_SYMBOL_GPL(bpf_map_put);
 653
 654 void bpf_map_put_with_uref(struct bpf_map *map)
 655 {
 656         bpf_map_put_uref(map);
 657         bpf_map_put(map);
 658 }
 659
 660 static int bpf_map_release(struct inode *inode, struct file *filp)
 661 {
 662         struct bpf_map *map = filp->private_data;
 663
 664         if (map->ops->map_release)
 665                 map->ops->map_release(map, filp);
 666
 667         bpf_map_put_with_uref(map);
 668         return 0;
 669 }
 670
 671 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 672 {
 673         fmode_t mode = f.file->f_mode;
 674
 675         /* Our file permissions may have been overridden by global
 676          * map permissions facing syscall side.
 677          */
 678         if (READ_ONCE(map->frozen))
 679                 mode &= ~FMODE_CAN_WRITE;
 680         return mode;
 681 }
 682
 683 #ifdef CONFIG_PROC_FS
 684 /* Provides an approximation of the map's memory footprint.
 685  * Used only to provide a backward compatibility and display
 686  * a reasonable "memlock" info.
 687  */
 688 static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
 689 {
 690         unsigned long size;
 691
 692         size = round_up(map->key_size + bpf_map_value_size(map), 8);
 693
 694         return round_up(map->max_entries * size, PAGE_SIZE);
 695 }
 696
 697 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 698 {
 699         struct bpf_map *map = filp->private_data;
 700         u32 type = 0, jited = 0;
 701
 702         if (map_type_contains_progs(map)) {
 703                 spin_lock(&map->owner.lock);
 704                 type  = map->owner.type;
 705                 jited = map->owner.jited;
 706                 spin_unlock(&map->owner.lock);
 707         }
 708
 709         seq_printf(m,
 710                    "map_type:\t%u\n"
 711                    "key_size:\t%u\n"
 712                    "value_size:\t%u\n"
 713                    "max_entries:\t%u\n"
 714                    "map_flags:\t%#x\n"
 715                    "map_extra:\t%#llx\n"
 716                    "memlock:\t%lu\n"
 717                    "map_id:\t%u\n"
 718                    "frozen:\t%u\n",
 719                    map->map_type,
 720                    map->key_size,
 721                    map->value_size,
 722                    map->max_entries,
 723                    map->map_flags,
 724                    (unsigned long long)map->map_extra,
 725                    bpf_map_memory_footprint(map),
 726                    map->id,
 727                    READ_ONCE(map->frozen));
 728         if (type) {
 729                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 730                 seq_printf(m, "owner_jited:\t%u\n", jited);
 731         }
 732 }
 733 #endif
 734
 735 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 736                               loff_t *ppos)
 737 {
 738         /* We need this handler such that alloc_file() enables
 739          * f_mode with FMODE_CAN_READ.
 740          */
 741         return -EINVAL;
 742 }
 743
 744 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 745                                size_t siz, loff_t *ppos)
 746 {
 747         /* We need this handler such that alloc_file() enables
 748          * f_mode with FMODE_CAN_WRITE.
 749          */
 750         return -EINVAL;
 751 }
 752
 753 /* called for any extra memory-mapped regions (except initial) */
 754 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 755 {
 756         struct bpf_map *map = vma->vm_file->private_data;
 757
 758         if (vma->vm_flags & VM_MAYWRITE)
 759                 bpf_map_write_active_inc(map);
 760 }
 761
 762 /* called for all unmapped memory region (including initial) */
 763 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 764 {
 765         struct bpf_map *map = vma->vm_file->private_data;
 766
 767         if (vma->vm_flags & VM_MAYWRITE)
 768                 bpf_map_write_active_dec(map);
 769 }
 770
 771 static const struct vm_operations_struct bpf_map_default_vmops = {
 772         .open           = bpf_map_mmap_open,
 773         .close          = bpf_map_mmap_close,
 774 };
 775
 776 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 777 {
 778         struct bpf_map *map = filp->private_data;
 779         int err;
 780
 781         if (!map->ops->map_mmap || map_value_has_spin_lock(map) ||
 782             map_value_has_timer(map) || map_value_has_kptrs(map))
 783                 return -ENOTSUPP;
 784
 785         if (!(vma->vm_flags & VM_SHARED))
 786                 return -EINVAL;
 787
 788         mutex_lock(&map->freeze_mutex);
 789
 790         if (vma->vm_flags & VM_WRITE) {
 791                 if (map->frozen) {
 792                         err = -EPERM;
 793                         goto out;
 794                 }
 795                 /* map is meant to be read-only, so do not allow mapping as
 796                  * writable, because it's possible to leak a writable page
 797                  * reference and allows user-space to still modify it after
 798                  * freezing, while verifier will assume contents do not change
 799                  */
 800                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 801                         err = -EACCES;
 802                         goto out;
 803                 }
 804         }
 805
 806         /* set default open/close callbacks */
 807         vma->vm_ops = &bpf_map_default_vmops;
 808         vma->vm_private_data = map;
 809         vma->vm_flags &= ~VM_MAYEXEC;
 810         if (!(vma->vm_flags & VM_WRITE))
 811                 /* disallow re-mapping with PROT_WRITE */
 812                 vma->vm_flags &= ~VM_MAYWRITE;
 813
 814         err = map->ops->map_mmap(map, vma);
 815         if (err)
 816                 goto out;
 817
 818         if (vma->vm_flags & VM_MAYWRITE)
 819                 bpf_map_write_active_inc(map);
 820 out:
 821         mutex_unlock(&map->freeze_mutex);
 822         return err;
 823 }
 824
 825 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 826 {
 827         struct bpf_map *map = filp->private_data;
 828
 829         if (map->ops->map_poll)
 830                 return map->ops->map_poll(map, filp, pts);
 831
 832         return EPOLLERR;
 833 }
 834
 835 const struct file_operations bpf_map_fops = {
 836 #ifdef CONFIG_PROC_FS
 837         .show_fdinfo    = bpf_map_show_fdinfo,
 838 #endif
 839         .release        = bpf_map_release,
 840         .read           = bpf_dummy_read,
 841         .write          = bpf_dummy_write,
 842         .mmap           = bpf_map_mmap,
 843         .poll           = bpf_map_poll,
 844 };
 845
 846 int bpf_map_new_fd(struct bpf_map *map, int flags)
 847 {
 848         int ret;
 849
 850         ret = security_bpf_map(map, OPEN_FMODE(flags));
 851         if (ret < 0)
 852                 return ret;
 853
 854         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 855                                 flags | O_CLOEXEC);
 856 }
 857
 858 int bpf_get_file_flag(int flags)
 859 {
 860         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 861                 return -EINVAL;
 862         if (flags & BPF_F_RDONLY)
 863                 return O_RDONLY;
 864         if (flags & BPF_F_WRONLY)
 865                 return O_WRONLY;
 866         return O_RDWR;
 867 }
 868
 869 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 870 #define CHECK_ATTR(CMD) \
 871         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 872                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 873                    sizeof(*attr) - \
 874                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 875                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 876
 877 /* dst and src must have at least "size" number of bytes.
 878  * Return strlen on success and < 0 on error.
 879  */
 880 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 881 {
 882         const char *end = src + size;
 883         const char *orig_src = src;
 884
 885         memset(dst, 0, size);
 886         /* Copy all isalnum(), '_' and '.' chars. */
 887         while (src < end && *src) {
 888                 if (!isalnum(*src) &&
 889                     *src != '_' && *src != '.')
 890                         return -EINVAL;
 891                 *dst++ = *src++;
 892         }
 893
 894         /* No '\0' found in "size" number of bytes */
 895         if (src == end)
 896                 return -EINVAL;
 897
 898         return src - orig_src;
 899 }
 900
 901 int map_check_no_btf(const struct bpf_map *map,
 902                      const struct btf *btf,
 903                      const struct btf_type *key_type,
 904                      const struct btf_type *value_type)
 905 {
 906         return -ENOTSUPP;
 907 }
 908
 909 static int map_off_arr_cmp(const void *_a, const void *_b, const void *priv)
 910 {
 911         const u32 a = *(const u32 *)_a;
 912         const u32 b = *(const u32 *)_b;
 913
 914         if (a < b)
 915                 return -1;
 916         else if (a > b)
 917                 return 1;
 918         return 0;
 919 }
 920
 921 static void map_off_arr_swap(void *_a, void *_b, int size, const void *priv)
 922 {
 923         struct bpf_map *map = (struct bpf_map *)priv;
 924         u32 *off_base = map->off_arr->field_off;
 925         u32 *a = _a, *b = _b;
 926         u8 *sz_a, *sz_b;
 927
 928         sz_a = map->off_arr->field_sz + (a - off_base);
 929         sz_b = map->off_arr->field_sz + (b - off_base);
 930
 931         swap(*a, *b);
 932         swap(*sz_a, *sz_b);
 933 }
 934
 935 static int bpf_map_alloc_off_arr(struct bpf_map *map)
 936 {
 937         bool has_spin_lock = map_value_has_spin_lock(map);
 938         bool has_timer = map_value_has_timer(map);
 939         bool has_kptrs = map_value_has_kptrs(map);
 940         struct bpf_map_off_arr *off_arr;
 941         u32 i;
 942
 943         if (!has_spin_lock && !has_timer && !has_kptrs) {
 944                 map->off_arr = NULL;
 945                 return 0;
 946         }
 947
 948         off_arr = kmalloc(sizeof(*map->off_arr), GFP_KERNEL | __GFP_NOWARN);
 949         if (!off_arr)
 950                 return -ENOMEM;
 951         map->off_arr = off_arr;
 952
 953         off_arr->cnt = 0;
 954         if (has_spin_lock) {
 955                 i = off_arr->cnt;
 956
 957                 off_arr->field_off[i] = map->spin_lock_off;
 958                 off_arr->field_sz[i] = sizeof(struct bpf_spin_lock);
 959                 off_arr->cnt++;
 960         }
 961         if (has_timer) {
 962                 i = off_arr->cnt;
 963
 964                 off_arr->field_off[i] = map->timer_off;
 965                 off_arr->field_sz[i] = sizeof(struct bpf_timer);
 966                 off_arr->cnt++;
 967         }
 968         if (has_kptrs) {
 969                 struct bpf_map_value_off *tab = map->kptr_off_tab;
 970                 u32 *off = &off_arr->field_off[off_arr->cnt];
 971                 u8 *sz = &off_arr->field_sz[off_arr->cnt];
 972
 973                 for (i = 0; i < tab->nr_off; i++) {
 974                         *off++ = tab->off[i].offset;
 975                         *sz++ = sizeof(u64);
 976                 }
 977                 off_arr->cnt += tab->nr_off;
 978         }
 979
 980         if (off_arr->cnt == 1)
 981                 return 0;
 982         sort_r(off_arr->field_off, off_arr->cnt, sizeof(off_arr->field_off[0]),
 983                map_off_arr_cmp, map_off_arr_swap, map);
 984         return 0;
 985 }
 986
 987 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 988                          u32 btf_key_id, u32 btf_value_id)
 989 {
 990         const struct btf_type *key_type, *value_type;
 991         u32 key_size, value_size;
 992         int ret = 0;
 993
 994         /* Some maps allow key to be unspecified. */
 995         if (btf_key_id) {
 996                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 997                 if (!key_type || key_size != map->key_size)
 998                         return -EINVAL;
 999         } else {
1000                 key_type = btf_type_by_id(btf, 0);
1001                 if (!map->ops->map_check_btf)
1002                         return -EINVAL;
1003         }
1004
1005         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1006         if (!value_type || value_size != map->value_size)
1007                 return -EINVAL;
1008
1009         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
1010
1011         if (map_value_has_spin_lock(map)) {
1012                 if (map->map_flags & BPF_F_RDONLY_PROG)
1013                         return -EACCES;
1014                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1015                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1016                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1017                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1018                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1019                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
1020                         return -ENOTSUPP;
1021                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
1022                     map->value_size) {
1023                         WARN_ONCE(1,
1024                                   "verifier bug spin_lock_off %d value_size %d\n",
1025                                   map->spin_lock_off, map->value_size);
1026                         return -EFAULT;
1027                 }
1028         }
1029
1030         map->timer_off = btf_find_timer(btf, value_type);
1031         if (map_value_has_timer(map)) {
1032                 if (map->map_flags & BPF_F_RDONLY_PROG)
1033                         return -EACCES;
1034                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1035                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1036                     map->map_type != BPF_MAP_TYPE_ARRAY)
1037                         return -EOPNOTSUPP;
1038         }
1039
1040         map->kptr_off_tab = btf_parse_kptrs(btf, value_type);
1041         if (map_value_has_kptrs(map)) {
1042                 if (!bpf_capable()) {
1043                         ret = -EPERM;
1044                         goto free_map_tab;
1045                 }
1046                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1047                         ret = -EACCES;
1048                         goto free_map_tab;
1049                 }
1050                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1051                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1052                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1053                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY) {
1054                         ret = -EOPNOTSUPP;
1055                         goto free_map_tab;
1056                 }
1057         }
1058
1059         if (map->ops->map_check_btf) {
1060                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1061                 if (ret < 0)
1062                         goto free_map_tab;
1063         }
1064
1065         return ret;
1066 free_map_tab:
1067         bpf_map_free_kptr_off_tab(map);
1068         return ret;
1069 }
1070
1071 #define BPF_MAP_CREATE_LAST_FIELD map_extra
1072 /* called via syscall */
1073 static int map_create(union bpf_attr *attr)
1074 {
1075         int numa_node = bpf_map_attr_numa_node(attr);
1076         struct bpf_map *map;
1077         int f_flags;
1078         int err;
1079
1080         err = CHECK_ATTR(BPF_MAP_CREATE);
1081         if (err)
1082                 return -EINVAL;
1083
1084         if (attr->btf_vmlinux_value_type_id) {
1085                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1086                     attr->btf_key_type_id || attr->btf_value_type_id)
1087                         return -EINVAL;
1088         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1089                 return -EINVAL;
1090         }
1091
1092         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1093             attr->map_extra != 0)
1094                 return -EINVAL;
1095
1096         f_flags = bpf_get_file_flag(attr->map_flags);
1097         if (f_flags < 0)
1098                 return f_flags;
1099
1100         if (numa_node != NUMA_NO_NODE &&
1101             ((unsigned int)numa_node >= nr_node_ids ||
1102              !node_online(numa_node)))
1103                 return -EINVAL;
1104
1105         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1106         map = find_and_alloc_map(attr);
1107         if (IS_ERR(map))
1108                 return PTR_ERR(map);
1109
1110         err = bpf_obj_name_cpy(map->name, attr->map_name,
1111                                sizeof(attr->map_name));
1112         if (err < 0)
1113                 goto free_map;
1114
1115         atomic64_set(&map->refcnt, 1);
1116         atomic64_set(&map->usercnt, 1);
1117         mutex_init(&map->freeze_mutex);
1118         spin_lock_init(&map->owner.lock);
1119
1120         map->spin_lock_off = -EINVAL;
1121         map->timer_off = -EINVAL;
1122         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1123             /* Even the map's value is a kernel's struct,
1124              * the bpf_prog.o must have BTF to begin with
1125              * to figure out the corresponding kernel's
1126              * counter part.  Thus, attr->btf_fd has
1127              * to be valid also.
1128              */
1129             attr->btf_vmlinux_value_type_id) {
1130                 struct btf *btf;
1131
1132                 btf = btf_get_by_fd(attr->btf_fd);
1133                 if (IS_ERR(btf)) {
1134                         err = PTR_ERR(btf);
1135                         goto free_map;
1136                 }
1137                 if (btf_is_kernel(btf)) {
1138                         btf_put(btf);
1139                         err = -EACCES;
1140                         goto free_map;
1141                 }
1142                 map->btf = btf;
1143
1144                 if (attr->btf_value_type_id) {
1145                         err = map_check_btf(map, btf, attr->btf_key_type_id,
1146                                             attr->btf_value_type_id);
1147                         if (err)
1148                                 goto free_map;
1149                 }
1150
1151                 map->btf_key_type_id = attr->btf_key_type_id;
1152                 map->btf_value_type_id = attr->btf_value_type_id;
1153                 map->btf_vmlinux_value_type_id =
1154                         attr->btf_vmlinux_value_type_id;
1155         }
1156
1157         err = bpf_map_alloc_off_arr(map);
1158         if (err)
1159                 goto free_map;
1160
1161         err = security_bpf_map_alloc(map);
1162         if (err)
1163                 goto free_map_off_arr;
1164
1165         err = bpf_map_alloc_id(map);
1166         if (err)
1167                 goto free_map_sec;
1168
1169         bpf_map_save_memcg(map);
1170
1171         err = bpf_map_new_fd(map, f_flags);
1172         if (err < 0) {
1173                 /* failed to allocate fd.
1174                  * bpf_map_put_with_uref() is needed because the above
1175                  * bpf_map_alloc_id() has published the map
1176                  * to the userspace and the userspace may
1177                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1178                  */
1179                 bpf_map_put_with_uref(map);
1180                 return err;
1181         }
1182
1183         return err;
1184
1185 free_map_sec:
1186         security_bpf_map_free(map);
1187 free_map_off_arr:
1188         kfree(map->off_arr);
1189 free_map:
1190         btf_put(map->btf);
1191         map->ops->map_free(map);
1192         return err;
1193 }
1194
1195 /* if error is returned, fd is released.
1196  * On success caller should complete fd access with matching fdput()
1197  */
1198 struct bpf_map *__bpf_map_get(struct fd f)
1199 {
1200         if (!f.file)
1201                 return ERR_PTR(-EBADF);
1202         if (f.file->f_op != &bpf_map_fops) {
1203                 fdput(f);
1204                 return ERR_PTR(-EINVAL);
1205         }
1206
1207         return f.file->private_data;
1208 }
1209
1210 void bpf_map_inc(struct bpf_map *map)
1211 {
1212         atomic64_inc(&map->refcnt);
1213 }
1214 EXPORT_SYMBOL_GPL(bpf_map_inc);
1215
1216 void bpf_map_inc_with_uref(struct bpf_map *map)
1217 {
1218         atomic64_inc(&map->refcnt);
1219         atomic64_inc(&map->usercnt);
1220 }
1221 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1222
1223 struct bpf_map *bpf_map_get(u32 ufd)
1224 {
1225         struct fd f = fdget(ufd);
1226         struct bpf_map *map;
1227
1228         map = __bpf_map_get(f);
1229         if (IS_ERR(map))
1230                 return map;
1231
1232         bpf_map_inc(map);
1233         fdput(f);
1234
1235         return map;
1236 }
1237 EXPORT_SYMBOL(bpf_map_get);
1238
1239 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1240 {
1241         struct fd f = fdget(ufd);
1242         struct bpf_map *map;
1243
1244         map = __bpf_map_get(f);
1245         if (IS_ERR(map))
1246                 return map;
1247
1248         bpf_map_inc_with_uref(map);
1249         fdput(f);
1250
1251         return map;
1252 }
1253
1254 /* map_idr_lock should have been held */
1255 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1256 {
1257         int refold;
1258
1259         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1260         if (!refold)
1261                 return ERR_PTR(-ENOENT);
1262         if (uref)
1263                 atomic64_inc(&map->usercnt);
1264
1265         return map;
1266 }
1267
1268 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1269 {
1270         spin_lock_bh(&map_idr_lock);
1271         map = __bpf_map_inc_not_zero(map, false);
1272         spin_unlock_bh(&map_idr_lock);
1273
1274         return map;
1275 }
1276 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1277
1278 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1279 {
1280         return -ENOTSUPP;
1281 }
1282
1283 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1284 {
1285         if (key_size)
1286                 return vmemdup_user(ukey, key_size);
1287
1288         if (ukey)
1289                 return ERR_PTR(-EINVAL);
1290
1291         return NULL;
1292 }
1293
1294 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1295 {
1296         if (key_size)
1297                 return kvmemdup_bpfptr(ukey, key_size);
1298
1299         if (!bpfptr_is_null(ukey))
1300                 return ERR_PTR(-EINVAL);
1301
1302         return NULL;
1303 }
1304
1305 /* last field in 'union bpf_attr' used by this command */
1306 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1307
1308 static int map_lookup_elem(union bpf_attr *attr)
1309 {
1310         void __user *ukey = u64_to_user_ptr(attr->key);
1311         void __user *uvalue = u64_to_user_ptr(attr->value);
1312         int ufd = attr->map_fd;
1313         struct bpf_map *map;
1314         void *key, *value;
1315         u32 value_size;
1316         struct fd f;
1317         int err;
1318
1319         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1320                 return -EINVAL;
1321
1322         if (attr->flags & ~BPF_F_LOCK)
1323                 return -EINVAL;
1324
1325         f = fdget(ufd);
1326         map = __bpf_map_get(f);
1327         if (IS_ERR(map))
1328                 return PTR_ERR(map);
1329         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1330                 err = -EPERM;
1331                 goto err_put;
1332         }
1333
1334         if ((attr->flags & BPF_F_LOCK) &&
1335             !map_value_has_spin_lock(map)) {
1336                 err = -EINVAL;
1337                 goto err_put;
1338         }
1339
1340         key = __bpf_copy_key(ukey, map->key_size);
1341         if (IS_ERR(key)) {
1342                 err = PTR_ERR(key);
1343                 goto err_put;
1344         }
1345
1346         value_size = bpf_map_value_size(map);
1347
1348         err = -ENOMEM;
1349         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1350         if (!value)
1351                 goto free_key;
1352
1353         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1354                 if (copy_from_user(value, uvalue, value_size))
1355                         err = -EFAULT;
1356                 else
1357                         err = bpf_map_copy_value(map, key, value, attr->flags);
1358                 goto free_value;
1359         }
1360
1361         err = bpf_map_copy_value(map, key, value, attr->flags);
1362         if (err)
1363                 goto free_value;
1364
1365         err = -EFAULT;
1366         if (copy_to_user(uvalue, value, value_size) != 0)
1367                 goto free_value;
1368
1369         err = 0;
1370
1371 free_value:
1372         kvfree(value);
1373 free_key:
1374         kvfree(key);
1375 err_put:
1376         fdput(f);
1377         return err;
1378 }
1379
1380
1381 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1382
1383 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1384 {
1385         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1386         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1387         int ufd = attr->map_fd;
1388         struct bpf_map *map;
1389         void *key, *value;
1390         u32 value_size;
1391         struct fd f;
1392         int err;
1393
1394         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1395                 return -EINVAL;
1396
1397         f = fdget(ufd);
1398         map = __bpf_map_get(f);
1399         if (IS_ERR(map))
1400                 return PTR_ERR(map);
1401         bpf_map_write_active_inc(map);
1402         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1403                 err = -EPERM;
1404                 goto err_put;
1405         }
1406
1407         if ((attr->flags & BPF_F_LOCK) &&
1408             !map_value_has_spin_lock(map)) {
1409                 err = -EINVAL;
1410                 goto err_put;
1411         }
1412
1413         key = ___bpf_copy_key(ukey, map->key_size);
1414         if (IS_ERR(key)) {
1415                 err = PTR_ERR(key);
1416                 goto err_put;
1417         }
1418
1419         value_size = bpf_map_value_size(map);
1420         value = kvmemdup_bpfptr(uvalue, value_size);
1421         if (IS_ERR(value)) {
1422                 err = PTR_ERR(value);
1423                 goto free_key;
1424         }
1425
1426         err = bpf_map_update_value(map, f, key, value, attr->flags);
1427
1428         kvfree(value);
1429 free_key:
1430         kvfree(key);
1431 err_put:
1432         bpf_map_write_active_dec(map);
1433         fdput(f);
1434         return err;
1435 }
1436
1437 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1438
1439 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1440 {
1441         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1442         int ufd = attr->map_fd;
1443         struct bpf_map *map;
1444         struct fd f;
1445         void *key;
1446         int err;
1447
1448         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1449                 return -EINVAL;
1450
1451         f = fdget(ufd);
1452         map = __bpf_map_get(f);
1453         if (IS_ERR(map))
1454                 return PTR_ERR(map);
1455         bpf_map_write_active_inc(map);
1456         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1457                 err = -EPERM;
1458                 goto err_put;
1459         }
1460
1461         key = ___bpf_copy_key(ukey, map->key_size);
1462         if (IS_ERR(key)) {
1463                 err = PTR_ERR(key);
1464                 goto err_put;
1465         }
1466
1467         if (bpf_map_is_dev_bound(map)) {
1468                 err = bpf_map_offload_delete_elem(map, key);
1469                 goto out;
1470         } else if (IS_FD_PROG_ARRAY(map) ||
1471                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1472                 /* These maps require sleepable context */
1473                 err = map->ops->map_delete_elem(map, key);
1474                 goto out;
1475         }
1476
1477         bpf_disable_instrumentation();
1478         rcu_read_lock();
1479         err = map->ops->map_delete_elem(map, key);
1480         rcu_read_unlock();
1481         bpf_enable_instrumentation();
1482         maybe_wait_bpf_programs(map);
1483 out:
1484         kvfree(key);
1485 err_put:
1486         bpf_map_write_active_dec(map);
1487         fdput(f);
1488         return err;
1489 }
1490
1491 /* last field in 'union bpf_attr' used by this command */
1492 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1493
1494 static int map_get_next_key(union bpf_attr *attr)
1495 {
1496         void __user *ukey = u64_to_user_ptr(attr->key);
1497         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1498         int ufd = attr->map_fd;
1499         struct bpf_map *map;
1500         void *key, *next_key;
1501         struct fd f;
1502         int err;
1503
1504         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1505                 return -EINVAL;
1506
1507         f = fdget(ufd);
1508         map = __bpf_map_get(f);
1509         if (IS_ERR(map))
1510                 return PTR_ERR(map);
1511         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1512                 err = -EPERM;
1513                 goto err_put;
1514         }
1515
1516         if (ukey) {
1517                 key = __bpf_copy_key(ukey, map->key_size);
1518                 if (IS_ERR(key)) {
1519                         err = PTR_ERR(key);
1520                         goto err_put;
1521                 }
1522         } else {
1523                 key = NULL;
1524         }
1525
1526         err = -ENOMEM;
1527         next_key = kvmalloc(map->key_size, GFP_USER);
1528         if (!next_key)
1529                 goto free_key;
1530
1531         if (bpf_map_is_dev_bound(map)) {
1532                 err = bpf_map_offload_get_next_key(map, key, next_key);
1533                 goto out;
1534         }
1535
1536         rcu_read_lock();
1537         err = map->ops->map_get_next_key(map, key, next_key);
1538         rcu_read_unlock();
1539 out:
1540         if (err)
1541                 goto free_next_key;
1542
1543         err = -EFAULT;
1544         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1545                 goto free_next_key;
1546
1547         err = 0;
1548
1549 free_next_key:
1550         kvfree(next_key);
1551 free_key:
1552         kvfree(key);
1553 err_put:
1554         fdput(f);
1555         return err;
1556 }
1557
1558 int generic_map_delete_batch(struct bpf_map *map,
1559                              const union bpf_attr *attr,
1560                              union bpf_attr __user *uattr)
1561 {
1562         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1563         u32 cp, max_count;
1564         int err = 0;
1565         void *key;
1566
1567         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1568                 return -EINVAL;
1569
1570         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1571             !map_value_has_spin_lock(map)) {
1572                 return -EINVAL;
1573         }
1574
1575         max_count = attr->batch.count;
1576         if (!max_count)
1577                 return 0;
1578
1579         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1580         if (!key)
1581                 return -ENOMEM;
1582
1583         for (cp = 0; cp < max_count; cp++) {
1584                 err = -EFAULT;
1585                 if (copy_from_user(key, keys + cp * map->key_size,
1586                                    map->key_size))
1587                         break;
1588
1589                 if (bpf_map_is_dev_bound(map)) {
1590                         err = bpf_map_offload_delete_elem(map, key);
1591                         break;
1592                 }
1593
1594                 bpf_disable_instrumentation();
1595                 rcu_read_lock();
1596                 err = map->ops->map_delete_elem(map, key);
1597                 rcu_read_unlock();
1598                 bpf_enable_instrumentation();
1599                 if (err)
1600                         break;
1601                 cond_resched();
1602         }
1603         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1604                 err = -EFAULT;
1605
1606         kvfree(key);
1607
1608         maybe_wait_bpf_programs(map);
1609         return err;
1610 }
1611
1612 int generic_map_update_batch(struct bpf_map *map,
1613                              const union bpf_attr *attr,
1614                              union bpf_attr __user *uattr)
1615 {
1616         void __user *values = u64_to_user_ptr(attr->batch.values);
1617         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1618         u32 value_size, cp, max_count;
1619         int ufd = attr->batch.map_fd;
1620         void *key, *value;
1621         struct fd f;
1622         int err = 0;
1623
1624         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1625                 return -EINVAL;
1626
1627         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1628             !map_value_has_spin_lock(map)) {
1629                 return -EINVAL;
1630         }
1631
1632         value_size = bpf_map_value_size(map);
1633
1634         max_count = attr->batch.count;
1635         if (!max_count)
1636                 return 0;
1637
1638         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1639         if (!key)
1640                 return -ENOMEM;
1641
1642         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1643         if (!value) {
1644                 kvfree(key);
1645                 return -ENOMEM;
1646         }
1647
1648         f = fdget(ufd); /* bpf_map_do_batch() guarantees ufd is valid */
1649         for (cp = 0; cp < max_count; cp++) {
1650                 err = -EFAULT;
1651                 if (copy_from_user(key, keys + cp * map->key_size,
1652                     map->key_size) ||
1653                     copy_from_user(value, values + cp * value_size, value_size))
1654                         break;
1655
1656                 err = bpf_map_update_value(map, f, key, value,
1657                                            attr->batch.elem_flags);
1658
1659                 if (err)
1660                         break;
1661                 cond_resched();
1662         }
1663
1664         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1665                 err = -EFAULT;
1666
1667         kvfree(value);
1668         kvfree(key);
1669         fdput(f);
1670         return err;
1671 }
1672
1673 #define MAP_LOOKUP_RETRIES 3
1674
1675 int generic_map_lookup_batch(struct bpf_map *map,
1676                                     const union bpf_attr *attr,
1677                                     union bpf_attr __user *uattr)
1678 {
1679         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1680         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1681         void __user *values = u64_to_user_ptr(attr->batch.values);
1682         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1683         void *buf, *buf_prevkey, *prev_key, *key, *value;
1684         int err, retry = MAP_LOOKUP_RETRIES;
1685         u32 value_size, cp, max_count;
1686
1687         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1688                 return -EINVAL;
1689
1690         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1691             !map_value_has_spin_lock(map))
1692                 return -EINVAL;
1693
1694         value_size = bpf_map_value_size(map);
1695
1696         max_count = attr->batch.count;
1697         if (!max_count)
1698                 return 0;
1699
1700         if (put_user(0, &uattr->batch.count))
1701                 return -EFAULT;
1702
1703         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1704         if (!buf_prevkey)
1705                 return -ENOMEM;
1706
1707         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1708         if (!buf) {
1709                 kvfree(buf_prevkey);
1710                 return -ENOMEM;
1711         }
1712
1713         err = -EFAULT;
1714         prev_key = NULL;
1715         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1716                 goto free_buf;
1717         key = buf;
1718         value = key + map->key_size;
1719         if (ubatch)
1720                 prev_key = buf_prevkey;
1721
1722         for (cp = 0; cp < max_count;) {
1723                 rcu_read_lock();
1724                 err = map->ops->map_get_next_key(map, prev_key, key);
1725                 rcu_read_unlock();
1726                 if (err)
1727                         break;
1728                 err = bpf_map_copy_value(map, key, value,
1729                                          attr->batch.elem_flags);
1730
1731                 if (err == -ENOENT) {
1732                         if (retry) {
1733                                 retry--;
1734                                 continue;
1735                         }
1736                         err = -EINTR;
1737                         break;
1738                 }
1739
1740                 if (err)
1741                         goto free_buf;
1742
1743                 if (copy_to_user(keys + cp * map->key_size, key,
1744                                  map->key_size)) {
1745                         err = -EFAULT;
1746                         goto free_buf;
1747                 }
1748                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1749                         err = -EFAULT;
1750                         goto free_buf;
1751                 }
1752
1753                 if (!prev_key)
1754                         prev_key = buf_prevkey;
1755
1756                 swap(prev_key, key);
1757                 retry = MAP_LOOKUP_RETRIES;
1758                 cp++;
1759                 cond_resched();
1760         }
1761
1762         if (err == -EFAULT)
1763                 goto free_buf;
1764
1765         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1766                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1767                 err = -EFAULT;
1768
1769 free_buf:
1770         kvfree(buf_prevkey);
1771         kvfree(buf);
1772         return err;
1773 }
1774
1775 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1776
1777 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1778 {
1779         void __user *ukey = u64_to_user_ptr(attr->key);
1780         void __user *uvalue = u64_to_user_ptr(attr->value);
1781         int ufd = attr->map_fd;
1782         struct bpf_map *map;
1783         void *key, *value;
1784         u32 value_size;
1785         struct fd f;
1786         int err;
1787
1788         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1789                 return -EINVAL;
1790
1791         if (attr->flags & ~BPF_F_LOCK)
1792                 return -EINVAL;
1793
1794         f = fdget(ufd);
1795         map = __bpf_map_get(f);
1796         if (IS_ERR(map))
1797                 return PTR_ERR(map);
1798         bpf_map_write_active_inc(map);
1799         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1800             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1801                 err = -EPERM;
1802                 goto err_put;
1803         }
1804
1805         if (attr->flags &&
1806             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1807              map->map_type == BPF_MAP_TYPE_STACK)) {
1808                 err = -EINVAL;
1809                 goto err_put;
1810         }
1811
1812         if ((attr->flags & BPF_F_LOCK) &&
1813             !map_value_has_spin_lock(map)) {
1814                 err = -EINVAL;
1815                 goto err_put;
1816         }
1817
1818         key = __bpf_copy_key(ukey, map->key_size);
1819         if (IS_ERR(key)) {
1820                 err = PTR_ERR(key);
1821                 goto err_put;
1822         }
1823
1824         value_size = bpf_map_value_size(map);
1825
1826         err = -ENOMEM;
1827         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1828         if (!value)
1829                 goto free_key;
1830
1831         err = -ENOTSUPP;
1832         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1833             map->map_type == BPF_MAP_TYPE_STACK) {
1834                 err = map->ops->map_pop_elem(map, value);
1835         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
1836                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1837                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
1838                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1839                 if (!bpf_map_is_dev_bound(map)) {
1840                         bpf_disable_instrumentation();
1841                         rcu_read_lock();
1842                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
1843                         rcu_read_unlock();
1844                         bpf_enable_instrumentation();
1845                 }
1846         }
1847
1848         if (err)
1849                 goto free_value;
1850
1851         if (copy_to_user(uvalue, value, value_size) != 0) {
1852                 err = -EFAULT;
1853                 goto free_value;
1854         }
1855
1856         err = 0;
1857
1858 free_value:
1859         kvfree(value);
1860 free_key:
1861         kvfree(key);
1862 err_put:
1863         bpf_map_write_active_dec(map);
1864         fdput(f);
1865         return err;
1866 }
1867
1868 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1869
1870 static int map_freeze(const union bpf_attr *attr)
1871 {
1872         int err = 0, ufd = attr->map_fd;
1873         struct bpf_map *map;
1874         struct fd f;
1875
1876         if (CHECK_ATTR(BPF_MAP_FREEZE))
1877                 return -EINVAL;
1878
1879         f = fdget(ufd);
1880         map = __bpf_map_get(f);
1881         if (IS_ERR(map))
1882                 return PTR_ERR(map);
1883
1884         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS ||
1885             map_value_has_timer(map) || map_value_has_kptrs(map)) {
1886                 fdput(f);
1887                 return -ENOTSUPP;
1888         }
1889
1890         mutex_lock(&map->freeze_mutex);
1891         if (bpf_map_write_active(map)) {
1892                 err = -EBUSY;
1893                 goto err_put;
1894         }
1895         if (READ_ONCE(map->frozen)) {
1896                 err = -EBUSY;
1897                 goto err_put;
1898         }
1899         if (!bpf_capable()) {
1900                 err = -EPERM;
1901                 goto err_put;
1902         }
1903
1904         WRITE_ONCE(map->frozen, true);
1905 err_put:
1906         mutex_unlock(&map->freeze_mutex);
1907         fdput(f);
1908         return err;
1909 }
1910
1911 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1912 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1913         [_id] = & _name ## _prog_ops,
1914 #define BPF_MAP_TYPE(_id, _ops)
1915 #define BPF_LINK_TYPE(_id, _name)
1916 #include <linux/bpf_types.h>
1917 #undef BPF_PROG_TYPE
1918 #undef BPF_MAP_TYPE
1919 #undef BPF_LINK_TYPE
1920 };
1921
1922 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1923 {
1924         const struct bpf_prog_ops *ops;
1925
1926         if (type >= ARRAY_SIZE(bpf_prog_types))
1927                 return -EINVAL;
1928         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1929         ops = bpf_prog_types[type];
1930         if (!ops)
1931                 return -EINVAL;
1932
1933         if (!bpf_prog_is_dev_bound(prog->aux))
1934                 prog->aux->ops = ops;
1935         else
1936                 prog->aux->ops = &bpf_offload_prog_ops;
1937         prog->type = type;
1938         return 0;
1939 }
1940
1941 enum bpf_audit {
1942         BPF_AUDIT_LOAD,
1943         BPF_AUDIT_UNLOAD,
1944         BPF_AUDIT_MAX,
1945 };
1946
1947 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1948         [BPF_AUDIT_LOAD]   = "LOAD",
1949         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1950 };
1951
1952 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1953 {
1954         struct audit_context *ctx = NULL;
1955         struct audit_buffer *ab;
1956
1957         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1958                 return;
1959         if (audit_enabled == AUDIT_OFF)
1960                 return;
1961         if (op == BPF_AUDIT_LOAD)
1962                 ctx = audit_context();
1963         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1964         if (unlikely(!ab))
1965                 return;
1966         audit_log_format(ab, "prog-id=%u op=%s",
1967                          prog->aux->id, bpf_audit_str[op]);
1968         audit_log_end(ab);
1969 }
1970
1971 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1972 {
1973         int id;
1974
1975         idr_preload(GFP_KERNEL);
1976         spin_lock_bh(&prog_idr_lock);
1977         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1978         if (id > 0)
1979                 prog->aux->id = id;
1980         spin_unlock_bh(&prog_idr_lock);
1981         idr_preload_end();
1982
1983         /* id is in [1, INT_MAX) */
1984         if (WARN_ON_ONCE(!id))
1985                 return -ENOSPC;
1986
1987         return id > 0 ? 0 : id;
1988 }
1989
1990 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1991 {
1992         unsigned long flags;
1993
1994         /* cBPF to eBPF migrations are currently not in the idr store.
1995          * Offloaded programs are removed from the store when their device
1996          * disappears - even if someone grabs an fd to them they are unusable,
1997          * simply waiting for refcnt to drop to be freed.
1998          */
1999         if (!prog->aux->id)
2000                 return;
2001
2002         if (do_idr_lock)
2003                 spin_lock_irqsave(&prog_idr_lock, flags);
2004         else
2005                 __acquire(&prog_idr_lock);
2006
2007         idr_remove(&prog_idr, prog->aux->id);
2008         prog->aux->id = 0;
2009
2010         if (do_idr_lock)
2011                 spin_unlock_irqrestore(&prog_idr_lock, flags);
2012         else
2013                 __release(&prog_idr_lock);
2014 }
2015
2016 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2017 {
2018         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2019
2020         kvfree(aux->func_info);
2021         kfree(aux->func_info_aux);
2022         free_uid(aux->user);
2023         security_bpf_prog_free(aux);
2024         bpf_prog_free(aux->prog);
2025 }
2026
2027 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2028 {
2029         bpf_prog_kallsyms_del_all(prog);
2030         btf_put(prog->aux->btf);
2031         kvfree(prog->aux->jited_linfo);
2032         kvfree(prog->aux->linfo);
2033         kfree(prog->aux->kfunc_tab);
2034         if (prog->aux->attach_btf)
2035                 btf_put(prog->aux->attach_btf);
2036
2037         if (deferred) {
2038                 if (prog->aux->sleepable)
2039                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2040                 else
2041                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2042         } else {
2043                 __bpf_prog_put_rcu(&prog->aux->rcu);
2044         }
2045 }
2046
2047 static void bpf_prog_put_deferred(struct work_struct *work)
2048 {
2049         struct bpf_prog_aux *aux;
2050         struct bpf_prog *prog;
2051
2052         aux = container_of(work, struct bpf_prog_aux, work);
2053         prog = aux->prog;
2054         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2055         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2056         __bpf_prog_put_noref(prog, true);
2057 }
2058
2059 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
2060 {
2061         struct bpf_prog_aux *aux = prog->aux;
2062
2063         if (atomic64_dec_and_test(&aux->refcnt)) {
2064                 /* bpf_prog_free_id() must be called first */
2065                 bpf_prog_free_id(prog, do_idr_lock);
2066
2067                 if (in_irq() || irqs_disabled()) {
2068                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2069                         schedule_work(&aux->work);
2070                 } else {
2071                         bpf_prog_put_deferred(&aux->work);
2072                 }
2073         }
2074 }
2075
2076 void bpf_prog_put(struct bpf_prog *prog)
2077 {
2078         __bpf_prog_put(prog, true);
2079 }
2080 EXPORT_SYMBOL_GPL(bpf_prog_put);
2081
2082 static int bpf_prog_release(struct inode *inode, struct file *filp)
2083 {
2084         struct bpf_prog *prog = filp->private_data;
2085
2086         bpf_prog_put(prog);
2087         return 0;
2088 }
2089
2090 struct bpf_prog_kstats {
2091         u64 nsecs;
2092         u64 cnt;
2093         u64 misses;
2094 };
2095
2096 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2097 {
2098         struct bpf_prog_stats *stats;
2099         unsigned int flags;
2100
2101         stats = this_cpu_ptr(prog->stats);
2102         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2103         u64_stats_inc(&stats->misses);
2104         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2105 }
2106
2107 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2108                                struct bpf_prog_kstats *stats)
2109 {
2110         u64 nsecs = 0, cnt = 0, misses = 0;
2111         int cpu;
2112
2113         for_each_possible_cpu(cpu) {
2114                 const struct bpf_prog_stats *st;
2115                 unsigned int start;
2116                 u64 tnsecs, tcnt, tmisses;
2117
2118                 st = per_cpu_ptr(prog->stats, cpu);
2119                 do {
2120                         start = u64_stats_fetch_begin_irq(&st->syncp);
2121                         tnsecs = u64_stats_read(&st->nsecs);
2122                         tcnt = u64_stats_read(&st->cnt);
2123                         tmisses = u64_stats_read(&st->misses);
2124                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
2125                 nsecs += tnsecs;
2126                 cnt += tcnt;
2127                 misses += tmisses;
2128         }
2129         stats->nsecs = nsecs;
2130         stats->cnt = cnt;
2131         stats->misses = misses;
2132 }
2133
2134 #ifdef CONFIG_PROC_FS
2135 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2136 {
2137         const struct bpf_prog *prog = filp->private_data;
2138         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2139         struct bpf_prog_kstats stats;
2140
2141         bpf_prog_get_stats(prog, &stats);
2142         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2143         seq_printf(m,
2144                    "prog_type:\t%u\n"
2145                    "prog_jited:\t%u\n"
2146                    "prog_tag:\t%s\n"
2147                    "memlock:\t%llu\n"
2148                    "prog_id:\t%u\n"
2149                    "run_time_ns:\t%llu\n"
2150                    "run_cnt:\t%llu\n"
2151                    "recursion_misses:\t%llu\n"
2152                    "verified_insns:\t%u\n",
2153                    prog->type,
2154                    prog->jited,
2155                    prog_tag,
2156                    prog->pages * 1ULL << PAGE_SHIFT,
2157                    prog->aux->id,
2158                    stats.nsecs,
2159                    stats.cnt,
2160                    stats.misses,
2161                    prog->aux->verified_insns);
2162 }
2163 #endif
2164
2165 const struct file_operations bpf_prog_fops = {
2166 #ifdef CONFIG_PROC_FS
2167         .show_fdinfo    = bpf_prog_show_fdinfo,
2168 #endif
2169         .release        = bpf_prog_release,
2170         .read           = bpf_dummy_read,
2171         .write          = bpf_dummy_write,
2172 };
2173
2174 int bpf_prog_new_fd(struct bpf_prog *prog)
2175 {
2176         int ret;
2177
2178         ret = security_bpf_prog(prog);
2179         if (ret < 0)
2180                 return ret;
2181
2182         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2183                                 O_RDWR | O_CLOEXEC);
2184 }
2185
2186 static struct bpf_prog *____bpf_prog_get(struct fd f)
2187 {
2188         if (!f.file)
2189                 return ERR_PTR(-EBADF);
2190         if (f.file->f_op != &bpf_prog_fops) {
2191                 fdput(f);
2192                 return ERR_PTR(-EINVAL);
2193         }
2194
2195         return f.file->private_data;
2196 }
2197
2198 void bpf_prog_add(struct bpf_prog *prog, int i)
2199 {
2200         atomic64_add(i, &prog->aux->refcnt);
2201 }
2202 EXPORT_SYMBOL_GPL(bpf_prog_add);
2203
2204 void bpf_prog_sub(struct bpf_prog *prog, int i)
2205 {
2206         /* Only to be used for undoing previous bpf_prog_add() in some
2207          * error path. We still know that another entity in our call
2208          * path holds a reference to the program, thus atomic_sub() can
2209          * be safely used in such cases!
2210          */
2211         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2212 }
2213 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2214
2215 void bpf_prog_inc(struct bpf_prog *prog)
2216 {
2217         atomic64_inc(&prog->aux->refcnt);
2218 }
2219 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2220
2221 /* prog_idr_lock should have been held */
2222 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2223 {
2224         int refold;
2225
2226         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2227
2228         if (!refold)
2229                 return ERR_PTR(-ENOENT);
2230
2231         return prog;
2232 }
2233 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2234
2235 bool bpf_prog_get_ok(struct bpf_prog *prog,
2236                             enum bpf_prog_type *attach_type, bool attach_drv)
2237 {
2238         /* not an attachment, just a refcount inc, always allow */
2239         if (!attach_type)
2240                 return true;
2241
2242         if (prog->type != *attach_type)
2243                 return false;
2244         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
2245                 return false;
2246
2247         return true;
2248 }
2249
2250 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2251                                        bool attach_drv)
2252 {
2253         struct fd f = fdget(ufd);
2254         struct bpf_prog *prog;
2255
2256         prog = ____bpf_prog_get(f);
2257         if (IS_ERR(prog))
2258                 return prog;
2259         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
2260                 prog = ERR_PTR(-EINVAL);
2261                 goto out;
2262         }
2263
2264         bpf_prog_inc(prog);
2265 out:
2266         fdput(f);
2267         return prog;
2268 }
2269
2270 struct bpf_prog *bpf_prog_get(u32 ufd)
2271 {
2272         return __bpf_prog_get(ufd, NULL, false);
2273 }
2274
2275 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2276                                        bool attach_drv)
2277 {
2278         return __bpf_prog_get(ufd, &type, attach_drv);
2279 }
2280 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2281
2282 /* Initially all BPF programs could be loaded w/o specifying
2283  * expected_attach_type. Later for some of them specifying expected_attach_type
2284  * at load time became required so that program could be validated properly.
2285  * Programs of types that are allowed to be loaded both w/ and w/o (for
2286  * backward compatibility) expected_attach_type, should have the default attach
2287  * type assigned to expected_attach_type for the latter case, so that it can be
2288  * validated later at attach time.
2289  *
2290  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2291  * prog type requires it but has some attach types that have to be backward
2292  * compatible.
2293  */
2294 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2295 {
2296         switch (attr->prog_type) {
2297         case BPF_PROG_TYPE_CGROUP_SOCK:
2298                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2299                  * exist so checking for non-zero is the way to go here.
2300                  */
2301                 if (!attr->expected_attach_type)
2302                         attr->expected_attach_type =
2303                                 BPF_CGROUP_INET_SOCK_CREATE;
2304                 break;
2305         case BPF_PROG_TYPE_SK_REUSEPORT:
2306                 if (!attr->expected_attach_type)
2307                         attr->expected_attach_type =
2308                                 BPF_SK_REUSEPORT_SELECT;
2309                 break;
2310         }
2311 }
2312
2313 static int
2314 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2315                            enum bpf_attach_type expected_attach_type,
2316                            struct btf *attach_btf, u32 btf_id,
2317                            struct bpf_prog *dst_prog)
2318 {
2319         if (btf_id) {
2320                 if (btf_id > BTF_MAX_TYPE)
2321                         return -EINVAL;
2322
2323                 if (!attach_btf && !dst_prog)
2324                         return -EINVAL;
2325
2326                 switch (prog_type) {
2327                 case BPF_PROG_TYPE_TRACING:
2328                 case BPF_PROG_TYPE_LSM:
2329                 case BPF_PROG_TYPE_STRUCT_OPS:
2330                 case BPF_PROG_TYPE_EXT:
2331                         break;
2332                 default:
2333                         return -EINVAL;
2334                 }
2335         }
2336
2337         if (attach_btf && (!btf_id || dst_prog))
2338                 return -EINVAL;
2339
2340         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2341             prog_type != BPF_PROG_TYPE_EXT)
2342                 return -EINVAL;
2343
2344         switch (prog_type) {
2345         case BPF_PROG_TYPE_CGROUP_SOCK:
2346                 switch (expected_attach_type) {
2347                 case BPF_CGROUP_INET_SOCK_CREATE:
2348                 case BPF_CGROUP_INET_SOCK_RELEASE:
2349                 case BPF_CGROUP_INET4_POST_BIND:
2350                 case BPF_CGROUP_INET6_POST_BIND:
2351                         return 0;
2352                 default:
2353                         return -EINVAL;
2354                 }
2355         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2356                 switch (expected_attach_type) {
2357                 case BPF_CGROUP_INET4_BIND:
2358                 case BPF_CGROUP_INET6_BIND:
2359                 case BPF_CGROUP_INET4_CONNECT:
2360                 case BPF_CGROUP_INET6_CONNECT:
2361                 case BPF_CGROUP_INET4_GETPEERNAME:
2362                 case BPF_CGROUP_INET6_GETPEERNAME:
2363                 case BPF_CGROUP_INET4_GETSOCKNAME:
2364                 case BPF_CGROUP_INET6_GETSOCKNAME:
2365                 case BPF_CGROUP_UDP4_SENDMSG:
2366                 case BPF_CGROUP_UDP6_SENDMSG:
2367                 case BPF_CGROUP_UDP4_RECVMSG:
2368                 case BPF_CGROUP_UDP6_RECVMSG:
2369                         return 0;
2370                 default:
2371                         return -EINVAL;
2372                 }
2373         case BPF_PROG_TYPE_CGROUP_SKB:
2374                 switch (expected_attach_type) {
2375                 case BPF_CGROUP_INET_INGRESS:
2376                 case BPF_CGROUP_INET_EGRESS:
2377                         return 0;
2378                 default:
2379                         return -EINVAL;
2380                 }
2381         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2382                 switch (expected_attach_type) {
2383                 case BPF_CGROUP_SETSOCKOPT:
2384                 case BPF_CGROUP_GETSOCKOPT:
2385                         return 0;
2386                 default:
2387                         return -EINVAL;
2388                 }
2389         case BPF_PROG_TYPE_SK_LOOKUP:
2390                 if (expected_attach_type == BPF_SK_LOOKUP)
2391                         return 0;
2392                 return -EINVAL;
2393         case BPF_PROG_TYPE_SK_REUSEPORT:
2394                 switch (expected_attach_type) {
2395                 case BPF_SK_REUSEPORT_SELECT:
2396                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2397                         return 0;
2398                 default:
2399                         return -EINVAL;
2400                 }
2401         case BPF_PROG_TYPE_SYSCALL:
2402         case BPF_PROG_TYPE_EXT:
2403                 if (expected_attach_type)
2404                         return -EINVAL;
2405                 fallthrough;
2406         default:
2407                 return 0;
2408         }
2409 }
2410
2411 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2412 {
2413         switch (prog_type) {
2414         case BPF_PROG_TYPE_SCHED_CLS:
2415         case BPF_PROG_TYPE_SCHED_ACT:
2416         case BPF_PROG_TYPE_XDP:
2417         case BPF_PROG_TYPE_LWT_IN:
2418         case BPF_PROG_TYPE_LWT_OUT:
2419         case BPF_PROG_TYPE_LWT_XMIT:
2420         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2421         case BPF_PROG_TYPE_SK_SKB:
2422         case BPF_PROG_TYPE_SK_MSG:
2423         case BPF_PROG_TYPE_LIRC_MODE2:
2424         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2425         case BPF_PROG_TYPE_CGROUP_DEVICE:
2426         case BPF_PROG_TYPE_CGROUP_SOCK:
2427         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2428         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2429         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2430         case BPF_PROG_TYPE_SOCK_OPS:
2431         case BPF_PROG_TYPE_EXT: /* extends any prog */
2432                 return true;
2433         case BPF_PROG_TYPE_CGROUP_SKB:
2434                 /* always unpriv */
2435         case BPF_PROG_TYPE_SK_REUSEPORT:
2436                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2437         default:
2438                 return false;
2439         }
2440 }
2441
2442 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2443 {
2444         switch (prog_type) {
2445         case BPF_PROG_TYPE_KPROBE:
2446         case BPF_PROG_TYPE_TRACEPOINT:
2447         case BPF_PROG_TYPE_PERF_EVENT:
2448         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2449         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2450         case BPF_PROG_TYPE_TRACING:
2451         case BPF_PROG_TYPE_LSM:
2452         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2453         case BPF_PROG_TYPE_EXT: /* extends any prog */
2454                 return true;
2455         default:
2456                 return false;
2457         }
2458 }
2459
2460 /* last field in 'union bpf_attr' used by this command */
2461 #define BPF_PROG_LOAD_LAST_FIELD core_relo_rec_size
2462
2463 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr)
2464 {
2465         enum bpf_prog_type type = attr->prog_type;
2466         struct bpf_prog *prog, *dst_prog = NULL;
2467         struct btf *attach_btf = NULL;
2468         int err;
2469         char license[128];
2470         bool is_gpl;
2471
2472         if (CHECK_ATTR(BPF_PROG_LOAD))
2473                 return -EINVAL;
2474
2475         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2476                                  BPF_F_ANY_ALIGNMENT |
2477                                  BPF_F_TEST_STATE_FREQ |
2478                                  BPF_F_SLEEPABLE |
2479                                  BPF_F_TEST_RND_HI32 |
2480                                  BPF_F_XDP_HAS_FRAGS))
2481                 return -EINVAL;
2482
2483         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2484             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2485             !bpf_capable())
2486                 return -EPERM;
2487
2488         /* copy eBPF program license from user space */
2489         if (strncpy_from_bpfptr(license,
2490                                 make_bpfptr(attr->license, uattr.is_kernel),
2491                                 sizeof(license) - 1) < 0)
2492                 return -EFAULT;
2493         license[sizeof(license) - 1] = 0;
2494
2495         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2496         is_gpl = license_is_gpl_compatible(license);
2497
2498         if (attr->insn_cnt == 0 ||
2499             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2500                 return -E2BIG;
2501         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2502             type != BPF_PROG_TYPE_CGROUP_SKB &&
2503             !bpf_capable())
2504                 return -EPERM;
2505
2506         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2507                 return -EPERM;
2508         if (is_perfmon_prog_type(type) && !perfmon_capable())
2509                 return -EPERM;
2510
2511         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2512          * or btf, we need to check which one it is
2513          */
2514         if (attr->attach_prog_fd) {
2515                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2516                 if (IS_ERR(dst_prog)) {
2517                         dst_prog = NULL;
2518                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2519                         if (IS_ERR(attach_btf))
2520                                 return -EINVAL;
2521                         if (!btf_is_kernel(attach_btf)) {
2522                                 /* attaching through specifying bpf_prog's BTF
2523                                  * objects directly might be supported eventually
2524                                  */
2525                                 btf_put(attach_btf);
2526                                 return -ENOTSUPP;
2527                         }
2528                 }
2529         } else if (attr->attach_btf_id) {
2530                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2531                 attach_btf = bpf_get_btf_vmlinux();
2532                 if (IS_ERR(attach_btf))
2533                         return PTR_ERR(attach_btf);
2534                 if (!attach_btf)
2535                         return -EINVAL;
2536                 btf_get(attach_btf);
2537         }
2538
2539         bpf_prog_load_fixup_attach_type(attr);
2540         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2541                                        attach_btf, attr->attach_btf_id,
2542                                        dst_prog)) {
2543                 if (dst_prog)
2544                         bpf_prog_put(dst_prog);
2545                 if (attach_btf)
2546                         btf_put(attach_btf);
2547                 return -EINVAL;
2548         }
2549
2550         /* plain bpf_prog allocation */
2551         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2552         if (!prog) {
2553                 if (dst_prog)
2554                         bpf_prog_put(dst_prog);
2555                 if (attach_btf)
2556                         btf_put(attach_btf);
2557                 return -ENOMEM;
2558         }
2559
2560         prog->expected_attach_type = attr->expected_attach_type;
2561         prog->aux->attach_btf = attach_btf;
2562         prog->aux->attach_btf_id = attr->attach_btf_id;
2563         prog->aux->dst_prog = dst_prog;
2564         prog->aux->offload_requested = !!attr->prog_ifindex;
2565         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2566         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2567
2568         err = security_bpf_prog_alloc(prog->aux);
2569         if (err)
2570                 goto free_prog;
2571
2572         prog->aux->user = get_current_user();
2573         prog->len = attr->insn_cnt;
2574
2575         err = -EFAULT;
2576         if (copy_from_bpfptr(prog->insns,
2577                              make_bpfptr(attr->insns, uattr.is_kernel),
2578                              bpf_prog_insn_size(prog)) != 0)
2579                 goto free_prog_sec;
2580
2581         prog->orig_prog = NULL;
2582         prog->jited = 0;
2583
2584         atomic64_set(&prog->aux->refcnt, 1);
2585         prog->gpl_compatible = is_gpl ? 1 : 0;
2586
2587         if (bpf_prog_is_dev_bound(prog->aux)) {
2588                 err = bpf_prog_offload_init(prog, attr);
2589                 if (err)
2590                         goto free_prog_sec;
2591         }
2592
2593         /* find program type: socket_filter vs tracing_filter */
2594         err = find_prog_type(type, prog);
2595         if (err < 0)
2596                 goto free_prog_sec;
2597
2598         prog->aux->load_time = ktime_get_boottime_ns();
2599         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2600                                sizeof(attr->prog_name));
2601         if (err < 0)
2602                 goto free_prog_sec;
2603
2604         /* run eBPF verifier */
2605         err = bpf_check(&prog, attr, uattr);
2606         if (err < 0)
2607                 goto free_used_maps;
2608
2609         prog = bpf_prog_select_runtime(prog, &err);
2610         if (err < 0)
2611                 goto free_used_maps;
2612
2613         err = bpf_prog_alloc_id(prog);
2614         if (err)
2615                 goto free_used_maps;
2616
2617         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2618          * effectively publicly exposed. However, retrieving via
2619          * bpf_prog_get_fd_by_id() will take another reference,
2620          * therefore it cannot be gone underneath us.
2621          *
2622          * Only for the time /after/ successful bpf_prog_new_fd()
2623          * and before returning to userspace, we might just hold
2624          * one reference and any parallel close on that fd could
2625          * rip everything out. Hence, below notifications must
2626          * happen before bpf_prog_new_fd().
2627          *
2628          * Also, any failure handling from this point onwards must
2629          * be using bpf_prog_put() given the program is exposed.
2630          */
2631         bpf_prog_kallsyms_add(prog);
2632         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2633         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2634
2635         err = bpf_prog_new_fd(prog);
2636         if (err < 0)
2637                 bpf_prog_put(prog);
2638         return err;
2639
2640 free_used_maps:
2641         /* In case we have subprogs, we need to wait for a grace
2642          * period before we can tear down JIT memory since symbols
2643          * are already exposed under kallsyms.
2644          */
2645         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2646         return err;
2647 free_prog_sec:
2648         free_uid(prog->aux->user);
2649         security_bpf_prog_free(prog->aux);
2650 free_prog:
2651         if (prog->aux->attach_btf)
2652                 btf_put(prog->aux->attach_btf);
2653         bpf_prog_free(prog);
2654         return err;
2655 }
2656
2657 #define BPF_OBJ_LAST_FIELD file_flags
2658
2659 static int bpf_obj_pin(const union bpf_attr *attr)
2660 {
2661         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2662                 return -EINVAL;
2663
2664         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2665 }
2666
2667 static int bpf_obj_get(const union bpf_attr *attr)
2668 {
2669         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2670             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2671                 return -EINVAL;
2672
2673         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2674                                 attr->file_flags);
2675 }
2676
2677 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2678                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2679 {
2680         atomic64_set(&link->refcnt, 1);
2681         link->type = type;
2682         link->id = 0;
2683         link->ops = ops;
2684         link->prog = prog;
2685 }
2686
2687 static void bpf_link_free_id(int id)
2688 {
2689         if (!id)
2690                 return;
2691
2692         spin_lock_bh(&link_idr_lock);
2693         idr_remove(&link_idr, id);
2694         spin_unlock_bh(&link_idr_lock);
2695 }
2696
2697 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2698  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2699  * anon_inode's release() call. This helper marksbpf_link as
2700  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2701  * is not decremented, it's the responsibility of a calling code that failed
2702  * to complete bpf_link initialization.
2703  */
2704 void bpf_link_cleanup(struct bpf_link_primer *primer)
2705 {
2706         primer->link->prog = NULL;
2707         bpf_link_free_id(primer->id);
2708         fput(primer->file);
2709         put_unused_fd(primer->fd);
2710 }
2711
2712 void bpf_link_inc(struct bpf_link *link)
2713 {
2714         atomic64_inc(&link->refcnt);
2715 }
2716
2717 /* bpf_link_free is guaranteed to be called from process context */
2718 static void bpf_link_free(struct bpf_link *link)
2719 {
2720         bpf_link_free_id(link->id);
2721         if (link->prog) {
2722                 /* detach BPF program, clean up used resources */
2723                 link->ops->release(link);
2724                 bpf_prog_put(link->prog);
2725         }
2726         /* free bpf_link and its containing memory */
2727         link->ops->dealloc(link);
2728 }
2729
2730 static void bpf_link_put_deferred(struct work_struct *work)
2731 {
2732         struct bpf_link *link = container_of(work, struct bpf_link, work);
2733
2734         bpf_link_free(link);
2735 }
2736
2737 /* bpf_link_put can be called from atomic context, but ensures that resources
2738  * are freed from process context
2739  */
2740 void bpf_link_put(struct bpf_link *link)
2741 {
2742         if (!atomic64_dec_and_test(&link->refcnt))
2743                 return;
2744
2745         if (in_atomic()) {
2746                 INIT_WORK(&link->work, bpf_link_put_deferred);
2747                 schedule_work(&link->work);
2748         } else {
2749                 bpf_link_free(link);
2750         }
2751 }
2752 EXPORT_SYMBOL(bpf_link_put);
2753
2754 static int bpf_link_release(struct inode *inode, struct file *filp)
2755 {
2756         struct bpf_link *link = filp->private_data;
2757
2758         bpf_link_put(link);
2759         return 0;
2760 }
2761
2762 #ifdef CONFIG_PROC_FS
2763 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2764 #define BPF_MAP_TYPE(_id, _ops)
2765 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2766 static const char *bpf_link_type_strs[] = {
2767         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2768 #include <linux/bpf_types.h>
2769 };
2770 #undef BPF_PROG_TYPE
2771 #undef BPF_MAP_TYPE
2772 #undef BPF_LINK_TYPE
2773
2774 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2775 {
2776         const struct bpf_link *link = filp->private_data;
2777         const struct bpf_prog *prog = link->prog;
2778         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2779
2780         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2781         seq_printf(m,
2782                    "link_type:\t%s\n"
2783                    "link_id:\t%u\n"
2784                    "prog_tag:\t%s\n"
2785                    "prog_id:\t%u\n",
2786                    bpf_link_type_strs[link->type],
2787                    link->id,
2788                    prog_tag,
2789                    prog->aux->id);
2790         if (link->ops->show_fdinfo)
2791                 link->ops->show_fdinfo(link, m);
2792 }
2793 #endif
2794
2795 static const struct file_operations bpf_link_fops = {
2796 #ifdef CONFIG_PROC_FS
2797         .show_fdinfo    = bpf_link_show_fdinfo,
2798 #endif
2799         .release        = bpf_link_release,
2800         .read           = bpf_dummy_read,
2801         .write          = bpf_dummy_write,
2802 };
2803
2804 static int bpf_link_alloc_id(struct bpf_link *link)
2805 {
2806         int id;
2807
2808         idr_preload(GFP_KERNEL);
2809         spin_lock_bh(&link_idr_lock);
2810         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2811         spin_unlock_bh(&link_idr_lock);
2812         idr_preload_end();
2813
2814         return id;
2815 }
2816
2817 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2818  * reserving unused FD and allocating ID from link_idr. This is to be paired
2819  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2820  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2821  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2822  * transient state is passed around in struct bpf_link_primer.
2823  * This is preferred way to create and initialize bpf_link, especially when
2824  * there are complicated and expensive operations in between creating bpf_link
2825  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2826  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2827  * expensive (and potentially failing) roll back operations in a rare case
2828  * that file, FD, or ID can't be allocated.
2829  */
2830 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2831 {
2832         struct file *file;
2833         int fd, id;
2834
2835         fd = get_unused_fd_flags(O_CLOEXEC);
2836         if (fd < 0)
2837                 return fd;
2838
2839
2840         id = bpf_link_alloc_id(link);
2841         if (id < 0) {
2842                 put_unused_fd(fd);
2843                 return id;
2844         }
2845
2846         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2847         if (IS_ERR(file)) {
2848                 bpf_link_free_id(id);
2849                 put_unused_fd(fd);
2850                 return PTR_ERR(file);
2851         }
2852
2853         primer->link = link;
2854         primer->file = file;
2855         primer->fd = fd;
2856         primer->id = id;
2857         return 0;
2858 }
2859
2860 int bpf_link_settle(struct bpf_link_primer *primer)
2861 {
2862         /* make bpf_link fetchable by ID */
2863         spin_lock_bh(&link_idr_lock);
2864         primer->link->id = primer->id;
2865         spin_unlock_bh(&link_idr_lock);
2866         /* make bpf_link fetchable by FD */
2867         fd_install(primer->fd, primer->file);
2868         /* pass through installed FD */
2869         return primer->fd;
2870 }
2871
2872 int bpf_link_new_fd(struct bpf_link *link)
2873 {
2874         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2875 }
2876
2877 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2878 {
2879         struct fd f = fdget(ufd);
2880         struct bpf_link *link;
2881
2882         if (!f.file)
2883                 return ERR_PTR(-EBADF);
2884         if (f.file->f_op != &bpf_link_fops) {
2885                 fdput(f);
2886                 return ERR_PTR(-EINVAL);
2887         }
2888
2889         link = f.file->private_data;
2890         bpf_link_inc(link);
2891         fdput(f);
2892
2893         return link;
2894 }
2895 EXPORT_SYMBOL(bpf_link_get_from_fd);
2896
2897 static void bpf_tracing_link_release(struct bpf_link *link)
2898 {
2899         struct bpf_tracing_link *tr_link =
2900                 container_of(link, struct bpf_tracing_link, link.link);
2901
2902         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
2903                                                 tr_link->trampoline));
2904
2905         bpf_trampoline_put(tr_link->trampoline);
2906
2907         /* tgt_prog is NULL if target is a kernel function */
2908         if (tr_link->tgt_prog)
2909                 bpf_prog_put(tr_link->tgt_prog);
2910 }
2911
2912 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2913 {
2914         struct bpf_tracing_link *tr_link =
2915                 container_of(link, struct bpf_tracing_link, link.link);
2916
2917         kfree(tr_link);
2918 }
2919
2920 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2921                                          struct seq_file *seq)
2922 {
2923         struct bpf_tracing_link *tr_link =
2924                 container_of(link, struct bpf_tracing_link, link.link);
2925
2926         seq_printf(seq,
2927                    "attach_type:\t%d\n",
2928                    tr_link->attach_type);
2929 }
2930
2931 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2932                                            struct bpf_link_info *info)
2933 {
2934         struct bpf_tracing_link *tr_link =
2935                 container_of(link, struct bpf_tracing_link, link.link);
2936
2937         info->tracing.attach_type = tr_link->attach_type;
2938         bpf_trampoline_unpack_key(tr_link->trampoline->key,
2939                                   &info->tracing.target_obj_id,
2940                                   &info->tracing.target_btf_id);
2941
2942         return 0;
2943 }
2944
2945 static const struct bpf_link_ops bpf_tracing_link_lops = {
2946         .release = bpf_tracing_link_release,
2947         .dealloc = bpf_tracing_link_dealloc,
2948         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2949         .fill_link_info = bpf_tracing_link_fill_link_info,
2950 };
2951
2952 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2953                                    int tgt_prog_fd,
2954                                    u32 btf_id,
2955                                    u64 bpf_cookie)
2956 {
2957         struct bpf_link_primer link_primer;
2958         struct bpf_prog *tgt_prog = NULL;
2959         struct bpf_trampoline *tr = NULL;
2960         struct bpf_tracing_link *link;
2961         u64 key = 0;
2962         int err;
2963
2964         switch (prog->type) {
2965         case BPF_PROG_TYPE_TRACING:
2966                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2967                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2968                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2969                         err = -EINVAL;
2970                         goto out_put_prog;
2971                 }
2972                 break;
2973         case BPF_PROG_TYPE_EXT:
2974                 if (prog->expected_attach_type != 0) {
2975                         err = -EINVAL;
2976                         goto out_put_prog;
2977                 }
2978                 break;
2979         case BPF_PROG_TYPE_LSM:
2980                 if (prog->expected_attach_type != BPF_LSM_MAC) {
2981                         err = -EINVAL;
2982                         goto out_put_prog;
2983                 }
2984                 break;
2985         default:
2986                 err = -EINVAL;
2987                 goto out_put_prog;
2988         }
2989
2990         if (!!tgt_prog_fd != !!btf_id) {
2991                 err = -EINVAL;
2992                 goto out_put_prog;
2993         }
2994
2995         if (tgt_prog_fd) {
2996                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2997                 if (prog->type != BPF_PROG_TYPE_EXT) {
2998                         err = -EINVAL;
2999                         goto out_put_prog;
3000                 }
3001
3002                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3003                 if (IS_ERR(tgt_prog)) {
3004                         err = PTR_ERR(tgt_prog);
3005                         tgt_prog = NULL;
3006                         goto out_put_prog;
3007                 }
3008
3009                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3010         }
3011
3012         link = kzalloc(sizeof(*link), GFP_USER);
3013         if (!link) {
3014                 err = -ENOMEM;
3015                 goto out_put_prog;
3016         }
3017         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3018                       &bpf_tracing_link_lops, prog);
3019         link->attach_type = prog->expected_attach_type;
3020         link->link.cookie = bpf_cookie;
3021
3022         mutex_lock(&prog->aux->dst_mutex);
3023
3024         /* There are a few possible cases here:
3025          *
3026          * - if prog->aux->dst_trampoline is set, the program was just loaded
3027          *   and not yet attached to anything, so we can use the values stored
3028          *   in prog->aux
3029          *
3030          * - if prog->aux->dst_trampoline is NULL, the program has already been
3031          *   attached to a target and its initial target was cleared (below)
3032          *
3033          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3034          *   target_btf_id using the link_create API.
3035          *
3036          * - if tgt_prog == NULL when this function was called using the old
3037          *   raw_tracepoint_open API, and we need a target from prog->aux
3038          *
3039          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3040          *   was detached and is going for re-attachment.
3041          */
3042         if (!prog->aux->dst_trampoline && !tgt_prog) {
3043                 /*
3044                  * Allow re-attach for TRACING and LSM programs. If it's
3045                  * currently linked, bpf_trampoline_link_prog will fail.
3046                  * EXT programs need to specify tgt_prog_fd, so they
3047                  * re-attach in separate code path.
3048                  */
3049                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3050                     prog->type != BPF_PROG_TYPE_LSM) {
3051                         err = -EINVAL;
3052                         goto out_unlock;
3053                 }
3054                 btf_id = prog->aux->attach_btf_id;
3055                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3056         }
3057
3058         if (!prog->aux->dst_trampoline ||
3059             (key && key != prog->aux->dst_trampoline->key)) {
3060                 /* If there is no saved target, or the specified target is
3061                  * different from the destination specified at load time, we
3062                  * need a new trampoline and a check for compatibility
3063                  */
3064                 struct bpf_attach_target_info tgt_info = {};
3065
3066                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3067                                               &tgt_info);
3068                 if (err)
3069                         goto out_unlock;
3070
3071                 tr = bpf_trampoline_get(key, &tgt_info);
3072                 if (!tr) {
3073                         err = -ENOMEM;
3074                         goto out_unlock;
3075                 }
3076         } else {
3077                 /* The caller didn't specify a target, or the target was the
3078                  * same as the destination supplied during program load. This
3079                  * means we can reuse the trampoline and reference from program
3080                  * load time, and there is no need to allocate a new one. This
3081                  * can only happen once for any program, as the saved values in
3082                  * prog->aux are cleared below.
3083                  */
3084                 tr = prog->aux->dst_trampoline;
3085                 tgt_prog = prog->aux->dst_prog;
3086         }
3087
3088         err = bpf_link_prime(&link->link.link, &link_primer);
3089         if (err)
3090                 goto out_unlock;
3091
3092         err = bpf_trampoline_link_prog(&link->link, tr);
3093         if (err) {
3094                 bpf_link_cleanup(&link_primer);
3095                 link = NULL;
3096                 goto out_unlock;
3097         }
3098
3099         link->tgt_prog = tgt_prog;
3100         link->trampoline = tr;
3101
3102         /* Always clear the trampoline and target prog from prog->aux to make
3103          * sure the original attach destination is not kept alive after a
3104          * program is (re-)attached to another target.
3105          */
3106         if (prog->aux->dst_prog &&
3107             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3108                 /* got extra prog ref from syscall, or attaching to different prog */
3109                 bpf_prog_put(prog->aux->dst_prog);
3110         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3111                 /* we allocated a new trampoline, so free the old one */
3112                 bpf_trampoline_put(prog->aux->dst_trampoline);
3113
3114         prog->aux->dst_prog = NULL;
3115         prog->aux->dst_trampoline = NULL;
3116         mutex_unlock(&prog->aux->dst_mutex);
3117
3118         return bpf_link_settle(&link_primer);
3119 out_unlock:
3120         if (tr && tr != prog->aux->dst_trampoline)
3121                 bpf_trampoline_put(tr);
3122         mutex_unlock(&prog->aux->dst_mutex);
3123         kfree(link);
3124 out_put_prog:
3125         if (tgt_prog_fd && tgt_prog)
3126                 bpf_prog_put(tgt_prog);
3127         return err;
3128 }
3129
3130 struct bpf_raw_tp_link {
3131         struct bpf_link link;
3132         struct bpf_raw_event_map *btp;
3133 };
3134
3135 static void bpf_raw_tp_link_release(struct bpf_link *link)
3136 {
3137         struct bpf_raw_tp_link *raw_tp =
3138                 container_of(link, struct bpf_raw_tp_link, link);
3139
3140         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
3141         bpf_put_raw_tracepoint(raw_tp->btp);
3142 }
3143
3144 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3145 {
3146         struct bpf_raw_tp_link *raw_tp =
3147                 container_of(link, struct bpf_raw_tp_link, link);
3148
3149         kfree(raw_tp);
3150 }
3151
3152 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3153                                         struct seq_file *seq)
3154 {
3155         struct bpf_raw_tp_link *raw_tp_link =
3156                 container_of(link, struct bpf_raw_tp_link, link);
3157
3158         seq_printf(seq,
3159                    "tp_name:\t%s\n",
3160                    raw_tp_link->btp->tp->name);
3161 }
3162
3163 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3164                                           struct bpf_link_info *info)
3165 {
3166         struct bpf_raw_tp_link *raw_tp_link =
3167                 container_of(link, struct bpf_raw_tp_link, link);
3168         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3169         const char *tp_name = raw_tp_link->btp->tp->name;
3170         u32 ulen = info->raw_tracepoint.tp_name_len;
3171         size_t tp_len = strlen(tp_name);
3172
3173         if (!ulen ^ !ubuf)
3174                 return -EINVAL;
3175
3176         info->raw_tracepoint.tp_name_len = tp_len + 1;
3177
3178         if (!ubuf)
3179                 return 0;
3180
3181         if (ulen >= tp_len + 1) {
3182                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
3183                         return -EFAULT;
3184         } else {
3185                 char zero = '\0';
3186
3187                 if (copy_to_user(ubuf, tp_name, ulen - 1))
3188                         return -EFAULT;
3189                 if (put_user(zero, ubuf + ulen - 1))
3190                         return -EFAULT;
3191                 return -ENOSPC;
3192         }
3193
3194         return 0;
3195 }
3196
3197 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3198         .release = bpf_raw_tp_link_release,
3199         .dealloc = bpf_raw_tp_link_dealloc,
3200         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3201         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3202 };
3203
3204 #ifdef CONFIG_PERF_EVENTS
3205 struct bpf_perf_link {
3206         struct bpf_link link;
3207         struct file *perf_file;
3208 };
3209
3210 static void bpf_perf_link_release(struct bpf_link *link)
3211 {
3212         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3213         struct perf_event *event = perf_link->perf_file->private_data;
3214
3215         perf_event_free_bpf_prog(event);
3216         fput(perf_link->perf_file);
3217 }
3218
3219 static void bpf_perf_link_dealloc(struct bpf_link *link)
3220 {
3221         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3222
3223         kfree(perf_link);
3224 }
3225
3226 static const struct bpf_link_ops bpf_perf_link_lops = {
3227         .release = bpf_perf_link_release,
3228         .dealloc = bpf_perf_link_dealloc,
3229 };
3230
3231 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3232 {
3233         struct bpf_link_primer link_primer;
3234         struct bpf_perf_link *link;
3235         struct perf_event *event;
3236         struct file *perf_file;
3237         int err;
3238
3239         if (attr->link_create.flags)
3240                 return -EINVAL;
3241
3242         perf_file = perf_event_get(attr->link_create.target_fd);
3243         if (IS_ERR(perf_file))
3244                 return PTR_ERR(perf_file);
3245
3246         link = kzalloc(sizeof(*link), GFP_USER);
3247         if (!link) {
3248                 err = -ENOMEM;
3249                 goto out_put_file;
3250         }
3251         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3252         link->perf_file = perf_file;
3253
3254         err = bpf_link_prime(&link->link, &link_primer);
3255         if (err) {
3256                 kfree(link);
3257                 goto out_put_file;
3258         }
3259
3260         event = perf_file->private_data;
3261         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3262         if (err) {
3263                 bpf_link_cleanup(&link_primer);
3264                 goto out_put_file;
3265         }
3266         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3267         bpf_prog_inc(prog);
3268
3269         return bpf_link_settle(&link_primer);
3270
3271 out_put_file:
3272         fput(perf_file);
3273         return err;
3274 }
3275 #else
3276 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3277 {
3278         return -EOPNOTSUPP;
3279 }
3280 #endif /* CONFIG_PERF_EVENTS */
3281
3282 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3283                                   const char __user *user_tp_name)
3284 {
3285         struct bpf_link_primer link_primer;
3286         struct bpf_raw_tp_link *link;
3287         struct bpf_raw_event_map *btp;
3288         const char *tp_name;
3289         char buf[128];
3290         int err;
3291
3292         switch (prog->type) {
3293         case BPF_PROG_TYPE_TRACING:
3294         case BPF_PROG_TYPE_EXT:
3295         case BPF_PROG_TYPE_LSM:
3296                 if (user_tp_name)
3297                         /* The attach point for this category of programs
3298                          * should be specified via btf_id during program load.
3299                          */
3300                         return -EINVAL;
3301                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3302                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3303                         tp_name = prog->aux->attach_func_name;
3304                         break;
3305                 }
3306                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3307         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3308         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3309                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3310                         return -EFAULT;
3311                 buf[sizeof(buf) - 1] = 0;
3312                 tp_name = buf;
3313                 break;
3314         default:
3315                 return -EINVAL;
3316         }
3317
3318         btp = bpf_get_raw_tracepoint(tp_name);
3319         if (!btp)
3320                 return -ENOENT;
3321
3322         link = kzalloc(sizeof(*link), GFP_USER);
3323         if (!link) {
3324                 err = -ENOMEM;
3325                 goto out_put_btp;
3326         }
3327         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3328                       &bpf_raw_tp_link_lops, prog);
3329         link->btp = btp;
3330
3331         err = bpf_link_prime(&link->link, &link_primer);
3332         if (err) {
3333                 kfree(link);
3334                 goto out_put_btp;
3335         }
3336
3337         err = bpf_probe_register(link->btp, prog);
3338         if (err) {
3339                 bpf_link_cleanup(&link_primer);
3340                 goto out_put_btp;
3341         }
3342
3343         return bpf_link_settle(&link_primer);
3344
3345 out_put_btp:
3346         bpf_put_raw_tracepoint(btp);
3347         return err;
3348 }
3349
3350 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
3351
3352 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3353 {
3354         struct bpf_prog *prog;
3355         int fd;
3356
3357         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3358                 return -EINVAL;
3359
3360         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3361         if (IS_ERR(prog))
3362                 return PTR_ERR(prog);
3363
3364         fd = bpf_raw_tp_link_attach(prog, u64_to_user_ptr(attr->raw_tracepoint.name));
3365         if (fd < 0)
3366                 bpf_prog_put(prog);
3367         return fd;
3368 }
3369
3370 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3371                                              enum bpf_attach_type attach_type)
3372 {
3373         switch (prog->type) {
3374         case BPF_PROG_TYPE_CGROUP_SOCK:
3375         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3376         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3377         case BPF_PROG_TYPE_SK_LOOKUP:
3378                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3379         case BPF_PROG_TYPE_CGROUP_SKB:
3380                 if (!capable(CAP_NET_ADMIN))
3381                         /* cg-skb progs can be loaded by unpriv user.
3382                          * check permissions at attach time.
3383                          */
3384                         return -EPERM;
3385                 return prog->enforce_expected_attach_type &&
3386                         prog->expected_attach_type != attach_type ?
3387                         -EINVAL : 0;
3388         default:
3389                 return 0;
3390         }
3391 }
3392
3393 static enum bpf_prog_type
3394 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3395 {
3396         switch (attach_type) {
3397         case BPF_CGROUP_INET_INGRESS:
3398         case BPF_CGROUP_INET_EGRESS:
3399                 return BPF_PROG_TYPE_CGROUP_SKB;
3400         case BPF_CGROUP_INET_SOCK_CREATE:
3401         case BPF_CGROUP_INET_SOCK_RELEASE:
3402         case BPF_CGROUP_INET4_POST_BIND:
3403         case BPF_CGROUP_INET6_POST_BIND:
3404                 return BPF_PROG_TYPE_CGROUP_SOCK;
3405         case BPF_CGROUP_INET4_BIND:
3406         case BPF_CGROUP_INET6_BIND:
3407         case BPF_CGROUP_INET4_CONNECT:
3408         case BPF_CGROUP_INET6_CONNECT:
3409         case BPF_CGROUP_INET4_GETPEERNAME:
3410         case BPF_CGROUP_INET6_GETPEERNAME:
3411         case BPF_CGROUP_INET4_GETSOCKNAME:
3412         case BPF_CGROUP_INET6_GETSOCKNAME:
3413         case BPF_CGROUP_UDP4_SENDMSG:
3414         case BPF_CGROUP_UDP6_SENDMSG:
3415         case BPF_CGROUP_UDP4_RECVMSG:
3416         case BPF_CGROUP_UDP6_RECVMSG:
3417                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3418         case BPF_CGROUP_SOCK_OPS:
3419                 return BPF_PROG_TYPE_SOCK_OPS;
3420         case BPF_CGROUP_DEVICE:
3421                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3422         case BPF_SK_MSG_VERDICT:
3423                 return BPF_PROG_TYPE_SK_MSG;
3424         case BPF_SK_SKB_STREAM_PARSER:
3425         case BPF_SK_SKB_STREAM_VERDICT:
3426         case BPF_SK_SKB_VERDICT:
3427                 return BPF_PROG_TYPE_SK_SKB;
3428         case BPF_LIRC_MODE2:
3429                 return BPF_PROG_TYPE_LIRC_MODE2;
3430         case BPF_FLOW_DISSECTOR:
3431                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3432         case BPF_CGROUP_SYSCTL:
3433                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3434         case BPF_CGROUP_GETSOCKOPT:
3435         case BPF_CGROUP_SETSOCKOPT:
3436                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3437         case BPF_TRACE_ITER:
3438         case BPF_TRACE_RAW_TP:
3439         case BPF_TRACE_FENTRY:
3440         case BPF_TRACE_FEXIT:
3441         case BPF_MODIFY_RETURN:
3442                 return BPF_PROG_TYPE_TRACING;
3443         case BPF_LSM_MAC:
3444                 return BPF_PROG_TYPE_LSM;
3445         case BPF_SK_LOOKUP:
3446                 return BPF_PROG_TYPE_SK_LOOKUP;
3447         case BPF_XDP:
3448                 return BPF_PROG_TYPE_XDP;
3449         case BPF_LSM_CGROUP:
3450                 return BPF_PROG_TYPE_LSM;
3451         default:
3452                 return BPF_PROG_TYPE_UNSPEC;
3453         }
3454 }
3455
3456 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
3457
3458 #define BPF_F_ATTACH_MASK \
3459         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
3460
3461 static int bpf_prog_attach(const union bpf_attr *attr)
3462 {
3463         enum bpf_prog_type ptype;
3464         struct bpf_prog *prog;
3465         int ret;
3466
3467         if (CHECK_ATTR(BPF_PROG_ATTACH))
3468                 return -EINVAL;
3469
3470         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3471                 return -EINVAL;
3472
3473         ptype = attach_type_to_prog_type(attr->attach_type);
3474         if (ptype == BPF_PROG_TYPE_UNSPEC)
3475                 return -EINVAL;
3476
3477         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3478         if (IS_ERR(prog))
3479                 return PTR_ERR(prog);
3480
3481         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3482                 bpf_prog_put(prog);
3483                 return -EINVAL;
3484         }
3485
3486         switch (ptype) {
3487         case BPF_PROG_TYPE_SK_SKB:
3488         case BPF_PROG_TYPE_SK_MSG:
3489                 ret = sock_map_get_from_fd(attr, prog);
3490                 break;
3491         case BPF_PROG_TYPE_LIRC_MODE2:
3492                 ret = lirc_prog_attach(attr, prog);
3493                 break;
3494         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3495                 ret = netns_bpf_prog_attach(attr, prog);
3496                 break;
3497         case BPF_PROG_TYPE_CGROUP_DEVICE:
3498         case BPF_PROG_TYPE_CGROUP_SKB:
3499         case BPF_PROG_TYPE_CGROUP_SOCK:
3500         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3501         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3502         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3503         case BPF_PROG_TYPE_SOCK_OPS:
3504         case BPF_PROG_TYPE_LSM:
3505                 if (ptype == BPF_PROG_TYPE_LSM &&
3506                     prog->expected_attach_type != BPF_LSM_CGROUP)
3507                         return -EINVAL;
3508
3509                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3510                 break;
3511         default:
3512                 ret = -EINVAL;
3513         }
3514
3515         if (ret)
3516                 bpf_prog_put(prog);
3517         return ret;
3518 }
3519
3520 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3521
3522 static int bpf_prog_detach(const union bpf_attr *attr)
3523 {
3524         enum bpf_prog_type ptype;
3525
3526         if (CHECK_ATTR(BPF_PROG_DETACH))
3527                 return -EINVAL;
3528
3529         ptype = attach_type_to_prog_type(attr->attach_type);
3530
3531         switch (ptype) {
3532         case BPF_PROG_TYPE_SK_MSG:
3533         case BPF_PROG_TYPE_SK_SKB:
3534                 return sock_map_prog_detach(attr, ptype);
3535         case BPF_PROG_TYPE_LIRC_MODE2:
3536                 return lirc_prog_detach(attr);
3537         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3538                 return netns_bpf_prog_detach(attr, ptype);
3539         case BPF_PROG_TYPE_CGROUP_DEVICE:
3540         case BPF_PROG_TYPE_CGROUP_SKB:
3541         case BPF_PROG_TYPE_CGROUP_SOCK:
3542         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3543         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3544         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3545         case BPF_PROG_TYPE_SOCK_OPS:
3546         case BPF_PROG_TYPE_LSM:
3547                 return cgroup_bpf_prog_detach(attr, ptype);
3548         default:
3549                 return -EINVAL;
3550         }
3551 }
3552
3553 #define BPF_PROG_QUERY_LAST_FIELD query.prog_attach_flags
3554
3555 static int bpf_prog_query(const union bpf_attr *attr,
3556                           union bpf_attr __user *uattr)
3557 {
3558         if (!capable(CAP_NET_ADMIN))
3559                 return -EPERM;
3560         if (CHECK_ATTR(BPF_PROG_QUERY))
3561                 return -EINVAL;
3562         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3563                 return -EINVAL;
3564
3565         switch (attr->query.attach_type) {
3566         case BPF_CGROUP_INET_INGRESS:
3567         case BPF_CGROUP_INET_EGRESS:
3568         case BPF_CGROUP_INET_SOCK_CREATE:
3569         case BPF_CGROUP_INET_SOCK_RELEASE:
3570         case BPF_CGROUP_INET4_BIND:
3571         case BPF_CGROUP_INET6_BIND:
3572         case BPF_CGROUP_INET4_POST_BIND:
3573         case BPF_CGROUP_INET6_POST_BIND:
3574         case BPF_CGROUP_INET4_CONNECT:
3575         case BPF_CGROUP_INET6_CONNECT:
3576         case BPF_CGROUP_INET4_GETPEERNAME:
3577         case BPF_CGROUP_INET6_GETPEERNAME:
3578         case BPF_CGROUP_INET4_GETSOCKNAME:
3579         case BPF_CGROUP_INET6_GETSOCKNAME:
3580         case BPF_CGROUP_UDP4_SENDMSG:
3581         case BPF_CGROUP_UDP6_SENDMSG:
3582         case BPF_CGROUP_UDP4_RECVMSG:
3583         case BPF_CGROUP_UDP6_RECVMSG:
3584         case BPF_CGROUP_SOCK_OPS:
3585         case BPF_CGROUP_DEVICE:
3586         case BPF_CGROUP_SYSCTL:
3587         case BPF_CGROUP_GETSOCKOPT:
3588         case BPF_CGROUP_SETSOCKOPT:
3589         case BPF_LSM_CGROUP:
3590                 return cgroup_bpf_prog_query(attr, uattr);
3591         case BPF_LIRC_MODE2:
3592                 return lirc_prog_query(attr, uattr);
3593         case BPF_FLOW_DISSECTOR:
3594         case BPF_SK_LOOKUP:
3595                 return netns_bpf_prog_query(attr, uattr);
3596         case BPF_SK_SKB_STREAM_PARSER:
3597         case BPF_SK_SKB_STREAM_VERDICT:
3598         case BPF_SK_MSG_VERDICT:
3599         case BPF_SK_SKB_VERDICT:
3600                 return sock_map_bpf_prog_query(attr, uattr);
3601         default:
3602                 return -EINVAL;
3603         }
3604 }
3605
3606 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
3607
3608 static int bpf_prog_test_run(const union bpf_attr *attr,
3609                              union bpf_attr __user *uattr)
3610 {
3611         struct bpf_prog *prog;
3612         int ret = -ENOTSUPP;
3613
3614         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3615                 return -EINVAL;
3616
3617         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3618             (!attr->test.ctx_size_in && attr->test.ctx_in))
3619                 return -EINVAL;
3620
3621         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3622             (!attr->test.ctx_size_out && attr->test.ctx_out))
3623                 return -EINVAL;
3624
3625         prog = bpf_prog_get(attr->test.prog_fd);
3626         if (IS_ERR(prog))
3627                 return PTR_ERR(prog);
3628
3629         if (prog->aux->ops->test_run)
3630                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3631
3632         bpf_prog_put(prog);
3633         return ret;
3634 }
3635
3636 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3637
3638 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3639                                union bpf_attr __user *uattr,
3640                                struct idr *idr,
3641                                spinlock_t *lock)
3642 {
3643         u32 next_id = attr->start_id;
3644         int err = 0;
3645
3646         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3647                 return -EINVAL;
3648
3649         if (!capable(CAP_SYS_ADMIN))
3650                 return -EPERM;
3651
3652         next_id++;
3653         spin_lock_bh(lock);
3654         if (!idr_get_next(idr, &next_id))
3655                 err = -ENOENT;
3656         spin_unlock_bh(lock);
3657
3658         if (!err)
3659                 err = put_user(next_id, &uattr->next_id);
3660
3661         return err;
3662 }
3663
3664 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3665 {
3666         struct bpf_map *map;
3667
3668         spin_lock_bh(&map_idr_lock);
3669 again:
3670         map = idr_get_next(&map_idr, id);
3671         if (map) {
3672                 map = __bpf_map_inc_not_zero(map, false);
3673                 if (IS_ERR(map)) {
3674                         (*id)++;
3675                         goto again;
3676                 }
3677         }
3678         spin_unlock_bh(&map_idr_lock);
3679
3680         return map;
3681 }
3682
3683 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3684 {
3685         struct bpf_prog *prog;
3686
3687         spin_lock_bh(&prog_idr_lock);
3688 again:
3689         prog = idr_get_next(&prog_idr, id);
3690         if (prog) {
3691                 prog = bpf_prog_inc_not_zero(prog);
3692                 if (IS_ERR(prog)) {
3693                         (*id)++;
3694                         goto again;
3695                 }
3696         }
3697         spin_unlock_bh(&prog_idr_lock);
3698
3699         return prog;
3700 }
3701
3702 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3703
3704 struct bpf_prog *bpf_prog_by_id(u32 id)
3705 {
3706         struct bpf_prog *prog;
3707
3708         if (!id)
3709                 return ERR_PTR(-ENOENT);
3710
3711         spin_lock_bh(&prog_idr_lock);
3712         prog = idr_find(&prog_idr, id);
3713         if (prog)
3714                 prog = bpf_prog_inc_not_zero(prog);
3715         else
3716                 prog = ERR_PTR(-ENOENT);
3717         spin_unlock_bh(&prog_idr_lock);
3718         return prog;
3719 }
3720
3721 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3722 {
3723         struct bpf_prog *prog;
3724         u32 id = attr->prog_id;
3725         int fd;
3726
3727         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3728                 return -EINVAL;
3729
3730         if (!capable(CAP_SYS_ADMIN))
3731                 return -EPERM;
3732
3733         prog = bpf_prog_by_id(id);
3734         if (IS_ERR(prog))
3735                 return PTR_ERR(prog);
3736
3737         fd = bpf_prog_new_fd(prog);
3738         if (fd < 0)
3739                 bpf_prog_put(prog);
3740
3741         return fd;
3742 }
3743
3744 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3745
3746 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3747 {
3748         struct bpf_map *map;
3749         u32 id = attr->map_id;
3750         int f_flags;
3751         int fd;
3752
3753         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3754             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3755                 return -EINVAL;
3756
3757         if (!capable(CAP_SYS_ADMIN))
3758                 return -EPERM;
3759
3760         f_flags = bpf_get_file_flag(attr->open_flags);
3761         if (f_flags < 0)
3762                 return f_flags;
3763
3764         spin_lock_bh(&map_idr_lock);
3765         map = idr_find(&map_idr, id);
3766         if (map)
3767                 map = __bpf_map_inc_not_zero(map, true);
3768         else
3769                 map = ERR_PTR(-ENOENT);
3770         spin_unlock_bh(&map_idr_lock);
3771
3772         if (IS_ERR(map))
3773                 return PTR_ERR(map);
3774
3775         fd = bpf_map_new_fd(map, f_flags);
3776         if (fd < 0)
3777                 bpf_map_put_with_uref(map);
3778
3779         return fd;
3780 }
3781
3782 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3783                                               unsigned long addr, u32 *off,
3784                                               u32 *type)
3785 {
3786         const struct bpf_map *map;
3787         int i;
3788
3789         mutex_lock(&prog->aux->used_maps_mutex);
3790         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3791                 map = prog->aux->used_maps[i];
3792                 if (map == (void *)addr) {
3793                         *type = BPF_PSEUDO_MAP_FD;
3794                         goto out;
3795                 }
3796                 if (!map->ops->map_direct_value_meta)
3797                         continue;
3798                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3799                         *type = BPF_PSEUDO_MAP_VALUE;
3800                         goto out;
3801                 }
3802         }
3803         map = NULL;
3804
3805 out:
3806         mutex_unlock(&prog->aux->used_maps_mutex);
3807         return map;
3808 }
3809
3810 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3811                                               const struct cred *f_cred)
3812 {
3813         const struct bpf_map *map;
3814         struct bpf_insn *insns;
3815         u32 off, type;
3816         u64 imm;
3817         u8 code;
3818         int i;
3819
3820         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3821                         GFP_USER);
3822         if (!insns)
3823                 return insns;
3824
3825         for (i = 0; i < prog->len; i++) {
3826                 code = insns[i].code;
3827
3828                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3829                         insns[i].code = BPF_JMP | BPF_CALL;
3830                         insns[i].imm = BPF_FUNC_tail_call;
3831                         /* fall-through */
3832                 }
3833                 if (code == (BPF_JMP | BPF_CALL) ||
3834                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3835                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3836                                 insns[i].code = BPF_JMP | BPF_CALL;
3837                         if (!bpf_dump_raw_ok(f_cred))
3838                                 insns[i].imm = 0;
3839                         continue;
3840                 }
3841                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3842                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3843                         continue;
3844                 }
3845
3846                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3847                         continue;
3848
3849                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3850                 map = bpf_map_from_imm(prog, imm, &off, &type);
3851                 if (map) {
3852                         insns[i].src_reg = type;
3853                         insns[i].imm = map->id;
3854                         insns[i + 1].imm = off;
3855                         continue;
3856                 }
3857         }
3858
3859         return insns;
3860 }
3861
3862 static int set_info_rec_size(struct bpf_prog_info *info)
3863 {
3864         /*
3865          * Ensure info.*_rec_size is the same as kernel expected size
3866          *
3867          * or
3868          *
3869          * Only allow zero *_rec_size if both _rec_size and _cnt are
3870          * zero.  In this case, the kernel will set the expected
3871          * _rec_size back to the info.
3872          */
3873
3874         if ((info->nr_func_info || info->func_info_rec_size) &&
3875             info->func_info_rec_size != sizeof(struct bpf_func_info))
3876                 return -EINVAL;
3877
3878         if ((info->nr_line_info || info->line_info_rec_size) &&
3879             info->line_info_rec_size != sizeof(struct bpf_line_info))
3880                 return -EINVAL;
3881
3882         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3883             info->jited_line_info_rec_size != sizeof(__u64))
3884                 return -EINVAL;
3885
3886         info->func_info_rec_size = sizeof(struct bpf_func_info);
3887         info->line_info_rec_size = sizeof(struct bpf_line_info);
3888         info->jited_line_info_rec_size = sizeof(__u64);
3889
3890         return 0;
3891 }
3892
3893 static int bpf_prog_get_info_by_fd(struct file *file,
3894                                    struct bpf_prog *prog,
3895                                    const union bpf_attr *attr,
3896                                    union bpf_attr __user *uattr)
3897 {
3898         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3899         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
3900         struct bpf_prog_info info;
3901         u32 info_len = attr->info.info_len;
3902         struct bpf_prog_kstats stats;
3903         char __user *uinsns;
3904         u32 ulen;
3905         int err;
3906
3907         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
3908         if (err)
3909                 return err;
3910         info_len = min_t(u32, sizeof(info), info_len);
3911
3912         memset(&info, 0, sizeof(info));
3913         if (copy_from_user(&info, uinfo, info_len))
3914                 return -EFAULT;
3915
3916         info.type = prog->type;
3917         info.id = prog->aux->id;
3918         info.load_time = prog->aux->load_time;
3919         info.created_by_uid = from_kuid_munged(current_user_ns(),
3920                                                prog->aux->user->uid);
3921         info.gpl_compatible = prog->gpl_compatible;
3922
3923         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3924         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3925
3926         mutex_lock(&prog->aux->used_maps_mutex);
3927         ulen = info.nr_map_ids;
3928         info.nr_map_ids = prog->aux->used_map_cnt;
3929         ulen = min_t(u32, info.nr_map_ids, ulen);
3930         if (ulen) {
3931                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3932                 u32 i;
3933
3934                 for (i = 0; i < ulen; i++)
3935                         if (put_user(prog->aux->used_maps[i]->id,
3936                                      &user_map_ids[i])) {
3937                                 mutex_unlock(&prog->aux->used_maps_mutex);
3938                                 return -EFAULT;
3939                         }
3940         }
3941         mutex_unlock(&prog->aux->used_maps_mutex);
3942
3943         err = set_info_rec_size(&info);
3944         if (err)
3945                 return err;
3946
3947         bpf_prog_get_stats(prog, &stats);
3948         info.run_time_ns = stats.nsecs;
3949         info.run_cnt = stats.cnt;
3950         info.recursion_misses = stats.misses;
3951
3952         info.verified_insns = prog->aux->verified_insns;
3953
3954         if (!bpf_capable()) {
3955                 info.jited_prog_len = 0;
3956                 info.xlated_prog_len = 0;
3957                 info.nr_jited_ksyms = 0;
3958                 info.nr_jited_func_lens = 0;
3959                 info.nr_func_info = 0;
3960                 info.nr_line_info = 0;
3961                 info.nr_jited_line_info = 0;
3962                 goto done;
3963         }
3964
3965         ulen = info.xlated_prog_len;
3966         info.xlated_prog_len = bpf_prog_insn_size(prog);
3967         if (info.xlated_prog_len && ulen) {
3968                 struct bpf_insn *insns_sanitized;
3969                 bool fault;
3970
3971                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3972                         info.xlated_prog_insns = 0;
3973                         goto done;
3974                 }
3975                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3976                 if (!insns_sanitized)
3977                         return -ENOMEM;
3978                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3979                 ulen = min_t(u32, info.xlated_prog_len, ulen);
3980                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3981                 kfree(insns_sanitized);
3982                 if (fault)
3983                         return -EFAULT;
3984         }
3985
3986         if (bpf_prog_is_dev_bound(prog->aux)) {
3987                 err = bpf_prog_offload_info_fill(&info, prog);
3988                 if (err)
3989                         return err;
3990                 goto done;
3991         }
3992
3993         /* NOTE: the following code is supposed to be skipped for offload.
3994          * bpf_prog_offload_info_fill() is the place to fill similar fields
3995          * for offload.
3996          */
3997         ulen = info.jited_prog_len;
3998         if (prog->aux->func_cnt) {
3999                 u32 i;
4000
4001                 info.jited_prog_len = 0;
4002                 for (i = 0; i < prog->aux->func_cnt; i++)
4003                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4004         } else {
4005                 info.jited_prog_len = prog->jited_len;
4006         }
4007
4008         if (info.jited_prog_len && ulen) {
4009                 if (bpf_dump_raw_ok(file->f_cred)) {
4010                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4011                         ulen = min_t(u32, info.jited_prog_len, ulen);
4012
4013                         /* for multi-function programs, copy the JITed
4014                          * instructions for all the functions
4015                          */
4016                         if (prog->aux->func_cnt) {
4017                                 u32 len, free, i;
4018                                 u8 *img;
4019
4020                                 free = ulen;
4021                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4022                                         len = prog->aux->func[i]->jited_len;
4023                                         len = min_t(u32, len, free);
4024                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4025                                         if (copy_to_user(uinsns, img, len))
4026                                                 return -EFAULT;
4027                                         uinsns += len;
4028                                         free -= len;
4029                                         if (!free)
4030                                                 break;
4031                                 }
4032                         } else {
4033                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4034                                         return -EFAULT;
4035                         }
4036                 } else {
4037                         info.jited_prog_insns = 0;
4038                 }
4039         }
4040
4041         ulen = info.nr_jited_ksyms;
4042         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4043         if (ulen) {
4044                 if (bpf_dump_raw_ok(file->f_cred)) {
4045                         unsigned long ksym_addr;
4046                         u64 __user *user_ksyms;
4047                         u32 i;
4048
4049                         /* copy the address of the kernel symbol
4050                          * corresponding to each function
4051                          */
4052                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4053                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4054                         if (prog->aux->func_cnt) {
4055                                 for (i = 0; i < ulen; i++) {
4056                                         ksym_addr = (unsigned long)
4057                                                 prog->aux->func[i]->bpf_func;
4058                                         if (put_user((u64) ksym_addr,
4059                                                      &user_ksyms[i]))
4060                                                 return -EFAULT;
4061                                 }
4062                         } else {
4063                                 ksym_addr = (unsigned long) prog->bpf_func;
4064                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4065                                         return -EFAULT;
4066                         }
4067                 } else {
4068                         info.jited_ksyms = 0;
4069                 }
4070         }
4071
4072         ulen = info.nr_jited_func_lens;
4073         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4074         if (ulen) {
4075                 if (bpf_dump_raw_ok(file->f_cred)) {
4076                         u32 __user *user_lens;
4077                         u32 func_len, i;
4078
4079                         /* copy the JITed image lengths for each function */
4080                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4081                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4082                         if (prog->aux->func_cnt) {
4083                                 for (i = 0; i < ulen; i++) {
4084                                         func_len =
4085                                                 prog->aux->func[i]->jited_len;
4086                                         if (put_user(func_len, &user_lens[i]))
4087                                                 return -EFAULT;
4088                                 }
4089                         } else {
4090                                 func_len = prog->jited_len;
4091                                 if (put_user(func_len, &user_lens[0]))
4092                                         return -EFAULT;
4093                         }
4094                 } else {
4095                         info.jited_func_lens = 0;
4096                 }
4097         }
4098
4099         if (prog->aux->btf)
4100                 info.btf_id = btf_obj_id(prog->aux->btf);
4101         info.attach_btf_id = prog->aux->attach_btf_id;
4102         if (attach_btf)
4103                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4104
4105         ulen = info.nr_func_info;
4106         info.nr_func_info = prog->aux->func_info_cnt;
4107         if (info.nr_func_info && ulen) {
4108                 char __user *user_finfo;
4109
4110                 user_finfo = u64_to_user_ptr(info.func_info);
4111                 ulen = min_t(u32, info.nr_func_info, ulen);
4112                 if (copy_to_user(user_finfo, prog->aux->func_info,
4113                                  info.func_info_rec_size * ulen))
4114                         return -EFAULT;
4115         }
4116
4117         ulen = info.nr_line_info;
4118         info.nr_line_info = prog->aux->nr_linfo;
4119         if (info.nr_line_info && ulen) {
4120                 __u8 __user *user_linfo;
4121
4122                 user_linfo = u64_to_user_ptr(info.line_info);
4123                 ulen = min_t(u32, info.nr_line_info, ulen);
4124                 if (copy_to_user(user_linfo, prog->aux->linfo,
4125                                  info.line_info_rec_size * ulen))
4126                         return -EFAULT;
4127         }
4128
4129         ulen = info.nr_jited_line_info;
4130         if (prog->aux->jited_linfo)
4131                 info.nr_jited_line_info = prog->aux->nr_linfo;
4132         else
4133                 info.nr_jited_line_info = 0;
4134         if (info.nr_jited_line_info && ulen) {
4135                 if (bpf_dump_raw_ok(file->f_cred)) {
4136                         unsigned long line_addr;
4137                         __u64 __user *user_linfo;
4138                         u32 i;
4139
4140                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4141                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4142                         for (i = 0; i < ulen; i++) {
4143                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4144                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4145                                         return -EFAULT;
4146                         }
4147                 } else {
4148                         info.jited_line_info = 0;
4149                 }
4150         }
4151
4152         ulen = info.nr_prog_tags;
4153         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4154         if (ulen) {
4155                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4156                 u32 i;
4157
4158                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4159                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4160                 if (prog->aux->func_cnt) {
4161                         for (i = 0; i < ulen; i++) {
4162                                 if (copy_to_user(user_prog_tags[i],
4163                                                  prog->aux->func[i]->tag,
4164                                                  BPF_TAG_SIZE))
4165                                         return -EFAULT;
4166                         }
4167                 } else {
4168                         if (copy_to_user(user_prog_tags[0],
4169                                          prog->tag, BPF_TAG_SIZE))
4170                                 return -EFAULT;
4171                 }
4172         }
4173
4174 done:
4175         if (copy_to_user(uinfo, &info, info_len) ||
4176             put_user(info_len, &uattr->info.info_len))
4177                 return -EFAULT;
4178
4179         return 0;
4180 }
4181
4182 static int bpf_map_get_info_by_fd(struct file *file,
4183                                   struct bpf_map *map,
4184                                   const union bpf_attr *attr,
4185                                   union bpf_attr __user *uattr)
4186 {
4187         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4188         struct bpf_map_info info;
4189         u32 info_len = attr->info.info_len;
4190         int err;
4191
4192         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4193         if (err)
4194                 return err;
4195         info_len = min_t(u32, sizeof(info), info_len);
4196
4197         memset(&info, 0, sizeof(info));
4198         info.type = map->map_type;
4199         info.id = map->id;
4200         info.key_size = map->key_size;
4201         info.value_size = map->value_size;
4202         info.max_entries = map->max_entries;
4203         info.map_flags = map->map_flags;
4204         info.map_extra = map->map_extra;
4205         memcpy(info.name, map->name, sizeof(map->name));
4206
4207         if (map->btf) {
4208                 info.btf_id = btf_obj_id(map->btf);
4209                 info.btf_key_type_id = map->btf_key_type_id;
4210                 info.btf_value_type_id = map->btf_value_type_id;
4211         }
4212         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4213
4214         if (bpf_map_is_dev_bound(map)) {
4215                 err = bpf_map_offload_info_fill(&info, map);
4216                 if (err)
4217                         return err;
4218         }
4219
4220         if (copy_to_user(uinfo, &info, info_len) ||
4221             put_user(info_len, &uattr->info.info_len))
4222                 return -EFAULT;
4223
4224         return 0;
4225 }
4226
4227 static int bpf_btf_get_info_by_fd(struct file *file,
4228                                   struct btf *btf,
4229                                   const union bpf_attr *attr,
4230                                   union bpf_attr __user *uattr)
4231 {
4232         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4233         u32 info_len = attr->info.info_len;
4234         int err;
4235
4236         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4237         if (err)
4238                 return err;
4239
4240         return btf_get_info_by_fd(btf, attr, uattr);
4241 }
4242
4243 static int bpf_link_get_info_by_fd(struct file *file,
4244                                   struct bpf_link *link,
4245                                   const union bpf_attr *attr,
4246                                   union bpf_attr __user *uattr)
4247 {
4248         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4249         struct bpf_link_info info;
4250         u32 info_len = attr->info.info_len;
4251         int err;
4252
4253         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4254         if (err)
4255                 return err;
4256         info_len = min_t(u32, sizeof(info), info_len);
4257
4258         memset(&info, 0, sizeof(info));
4259         if (copy_from_user(&info, uinfo, info_len))
4260                 return -EFAULT;
4261
4262         info.type = link->type;
4263         info.id = link->id;
4264         info.prog_id = link->prog->aux->id;
4265
4266         if (link->ops->fill_link_info) {
4267                 err = link->ops->fill_link_info(link, &info);
4268                 if (err)
4269                         return err;
4270         }
4271
4272         if (copy_to_user(uinfo, &info, info_len) ||
4273             put_user(info_len, &uattr->info.info_len))
4274                 return -EFAULT;
4275
4276         return 0;
4277 }
4278
4279
4280 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4281
4282 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4283                                   union bpf_attr __user *uattr)
4284 {
4285         int ufd = attr->info.bpf_fd;
4286         struct fd f;
4287         int err;
4288
4289         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4290                 return -EINVAL;
4291
4292         f = fdget(ufd);
4293         if (!f.file)
4294                 return -EBADFD;
4295
4296         if (f.file->f_op == &bpf_prog_fops)
4297                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
4298                                               uattr);
4299         else if (f.file->f_op == &bpf_map_fops)
4300                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
4301                                              uattr);
4302         else if (f.file->f_op == &btf_fops)
4303                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
4304         else if (f.file->f_op == &bpf_link_fops)
4305                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
4306                                               attr, uattr);
4307         else
4308                 err = -EINVAL;
4309
4310         fdput(f);
4311         return err;
4312 }
4313
4314 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
4315
4316 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr)
4317 {
4318         if (CHECK_ATTR(BPF_BTF_LOAD))
4319                 return -EINVAL;
4320
4321         if (!bpf_capable())
4322                 return -EPERM;
4323
4324         return btf_new_fd(attr, uattr);
4325 }
4326
4327 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4328
4329 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4330 {
4331         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4332                 return -EINVAL;
4333
4334         if (!capable(CAP_SYS_ADMIN))
4335                 return -EPERM;
4336
4337         return btf_get_fd_by_id(attr->btf_id);
4338 }
4339
4340 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4341                                     union bpf_attr __user *uattr,
4342                                     u32 prog_id, u32 fd_type,
4343                                     const char *buf, u64 probe_offset,
4344                                     u64 probe_addr)
4345 {
4346         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4347         u32 len = buf ? strlen(buf) : 0, input_len;
4348         int err = 0;
4349
4350         if (put_user(len, &uattr->task_fd_query.buf_len))
4351                 return -EFAULT;
4352         input_len = attr->task_fd_query.buf_len;
4353         if (input_len && ubuf) {
4354                 if (!len) {
4355                         /* nothing to copy, just make ubuf NULL terminated */
4356                         char zero = '\0';
4357
4358                         if (put_user(zero, ubuf))
4359                                 return -EFAULT;
4360                 } else if (input_len >= len + 1) {
4361                         /* ubuf can hold the string with NULL terminator */
4362                         if (copy_to_user(ubuf, buf, len + 1))
4363                                 return -EFAULT;
4364                 } else {
4365                         /* ubuf cannot hold the string with NULL terminator,
4366                          * do a partial copy with NULL terminator.
4367                          */
4368                         char zero = '\0';
4369
4370                         err = -ENOSPC;
4371                         if (copy_to_user(ubuf, buf, input_len - 1))
4372                                 return -EFAULT;
4373                         if (put_user(zero, ubuf + input_len - 1))
4374                                 return -EFAULT;
4375                 }
4376         }
4377
4378         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
4379             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
4380             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
4381             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
4382                 return -EFAULT;
4383
4384         return err;
4385 }
4386
4387 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
4388
4389 static int bpf_task_fd_query(const union bpf_attr *attr,
4390                              union bpf_attr __user *uattr)
4391 {
4392         pid_t pid = attr->task_fd_query.pid;
4393         u32 fd = attr->task_fd_query.fd;
4394         const struct perf_event *event;
4395         struct task_struct *task;
4396         struct file *file;
4397         int err;
4398
4399         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
4400                 return -EINVAL;
4401
4402         if (!capable(CAP_SYS_ADMIN))
4403                 return -EPERM;
4404
4405         if (attr->task_fd_query.flags != 0)
4406                 return -EINVAL;
4407
4408         rcu_read_lock();
4409         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
4410         rcu_read_unlock();
4411         if (!task)
4412                 return -ENOENT;
4413
4414         err = 0;
4415         file = fget_task(task, fd);
4416         put_task_struct(task);
4417         if (!file)
4418                 return -EBADF;
4419
4420         if (file->f_op == &bpf_link_fops) {
4421                 struct bpf_link *link = file->private_data;
4422
4423                 if (link->ops == &bpf_raw_tp_link_lops) {
4424                         struct bpf_raw_tp_link *raw_tp =
4425                                 container_of(link, struct bpf_raw_tp_link, link);
4426                         struct bpf_raw_event_map *btp = raw_tp->btp;
4427
4428                         err = bpf_task_fd_query_copy(attr, uattr,
4429                                                      raw_tp->link.prog->aux->id,
4430                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
4431                                                      btp->tp->name, 0, 0);
4432                         goto put_file;
4433                 }
4434                 goto out_not_supp;
4435         }
4436
4437         event = perf_get_event(file);
4438         if (!IS_ERR(event)) {
4439                 u64 probe_offset, probe_addr;
4440                 u32 prog_id, fd_type;
4441                 const char *buf;
4442
4443                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
4444                                               &buf, &probe_offset,
4445                                               &probe_addr);
4446                 if (!err)
4447                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
4448                                                      fd_type, buf,
4449                                                      probe_offset,
4450                                                      probe_addr);
4451                 goto put_file;
4452         }
4453
4454 out_not_supp:
4455         err = -ENOTSUPP;
4456 put_file:
4457         fput(file);
4458         return err;
4459 }
4460
4461 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
4462
4463 #define BPF_DO_BATCH(fn)                        \
4464         do {                                    \
4465                 if (!fn) {                      \
4466                         err = -ENOTSUPP;        \
4467                         goto err_put;           \
4468                 }                               \
4469                 err = fn(map, attr, uattr);     \
4470         } while (0)
4471
4472 static int bpf_map_do_batch(const union bpf_attr *attr,
4473                             union bpf_attr __user *uattr,
4474                             int cmd)
4475 {
4476         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
4477                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
4478         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
4479         struct bpf_map *map;
4480         int err, ufd;
4481         struct fd f;
4482
4483         if (CHECK_ATTR(BPF_MAP_BATCH))
4484                 return -EINVAL;
4485
4486         ufd = attr->batch.map_fd;
4487         f = fdget(ufd);
4488         map = __bpf_map_get(f);
4489         if (IS_ERR(map))
4490                 return PTR_ERR(map);
4491         if (has_write)
4492                 bpf_map_write_active_inc(map);
4493         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4494                 err = -EPERM;
4495                 goto err_put;
4496         }
4497         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4498                 err = -EPERM;
4499                 goto err_put;
4500         }
4501
4502         if (cmd == BPF_MAP_LOOKUP_BATCH)
4503                 BPF_DO_BATCH(map->ops->map_lookup_batch);
4504         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4505                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4506         else if (cmd == BPF_MAP_UPDATE_BATCH)
4507                 BPF_DO_BATCH(map->ops->map_update_batch);
4508         else
4509                 BPF_DO_BATCH(map->ops->map_delete_batch);
4510 err_put:
4511         if (has_write)
4512                 bpf_map_write_active_dec(map);
4513         fdput(f);
4514         return err;
4515 }
4516
4517 #define BPF_LINK_CREATE_LAST_FIELD link_create.kprobe_multi.cookies
4518 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
4519 {
4520         enum bpf_prog_type ptype;
4521         struct bpf_prog *prog;
4522         int ret;
4523
4524         if (CHECK_ATTR(BPF_LINK_CREATE))
4525                 return -EINVAL;
4526
4527         prog = bpf_prog_get(attr->link_create.prog_fd);
4528         if (IS_ERR(prog))
4529                 return PTR_ERR(prog);
4530
4531         ret = bpf_prog_attach_check_attach_type(prog,
4532                                                 attr->link_create.attach_type);
4533         if (ret)
4534                 goto out;
4535
4536         switch (prog->type) {
4537         case BPF_PROG_TYPE_EXT:
4538                 break;
4539         case BPF_PROG_TYPE_PERF_EVENT:
4540         case BPF_PROG_TYPE_TRACEPOINT:
4541                 if (attr->link_create.attach_type != BPF_PERF_EVENT) {
4542                         ret = -EINVAL;
4543                         goto out;
4544                 }
4545                 break;
4546         case BPF_PROG_TYPE_KPROBE:
4547                 if (attr->link_create.attach_type != BPF_PERF_EVENT &&
4548                     attr->link_create.attach_type != BPF_TRACE_KPROBE_MULTI) {
4549                         ret = -EINVAL;
4550                         goto out;
4551                 }
4552                 break;
4553         default:
4554                 ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4555                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4556                         ret = -EINVAL;
4557                         goto out;
4558                 }
4559                 break;
4560         }
4561
4562         switch (prog->type) {
4563         case BPF_PROG_TYPE_CGROUP_SKB:
4564         case BPF_PROG_TYPE_CGROUP_SOCK:
4565         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4566         case BPF_PROG_TYPE_SOCK_OPS:
4567         case BPF_PROG_TYPE_CGROUP_DEVICE:
4568         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4569         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4570                 ret = cgroup_bpf_link_attach(attr, prog);
4571                 break;
4572         case BPF_PROG_TYPE_EXT:
4573                 ret = bpf_tracing_prog_attach(prog,
4574                                               attr->link_create.target_fd,
4575                                               attr->link_create.target_btf_id,
4576                                               attr->link_create.tracing.cookie);
4577                 break;
4578         case BPF_PROG_TYPE_LSM:
4579         case BPF_PROG_TYPE_TRACING:
4580                 if (attr->link_create.attach_type != prog->expected_attach_type) {
4581                         ret = -EINVAL;
4582                         goto out;
4583                 }
4584                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
4585                         ret = bpf_raw_tp_link_attach(prog, NULL);
4586                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
4587                         ret = bpf_iter_link_attach(attr, uattr, prog);
4588                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
4589                         ret = cgroup_bpf_link_attach(attr, prog);
4590                 else
4591                         ret = bpf_tracing_prog_attach(prog,
4592                                                       attr->link_create.target_fd,
4593                                                       attr->link_create.target_btf_id,
4594                                                       attr->link_create.tracing.cookie);
4595                 break;
4596         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4597         case BPF_PROG_TYPE_SK_LOOKUP:
4598                 ret = netns_bpf_link_create(attr, prog);
4599                 break;
4600 #ifdef CONFIG_NET
4601         case BPF_PROG_TYPE_XDP:
4602                 ret = bpf_xdp_link_attach(attr, prog);
4603                 break;
4604 #endif
4605         case BPF_PROG_TYPE_PERF_EVENT:
4606         case BPF_PROG_TYPE_TRACEPOINT:
4607                 ret = bpf_perf_link_attach(attr, prog);
4608                 break;
4609         case BPF_PROG_TYPE_KPROBE:
4610                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
4611                         ret = bpf_perf_link_attach(attr, prog);
4612                 else
4613                         ret = bpf_kprobe_multi_link_attach(attr, prog);
4614                 break;
4615         default:
4616                 ret = -EINVAL;
4617         }
4618
4619 out:
4620         if (ret < 0)
4621                 bpf_prog_put(prog);
4622         return ret;
4623 }
4624
4625 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4626
4627 static int link_update(union bpf_attr *attr)
4628 {
4629         struct bpf_prog *old_prog = NULL, *new_prog;
4630         struct bpf_link *link;
4631         u32 flags;
4632         int ret;
4633
4634         if (CHECK_ATTR(BPF_LINK_UPDATE))
4635                 return -EINVAL;
4636
4637         flags = attr->link_update.flags;
4638         if (flags & ~BPF_F_REPLACE)
4639                 return -EINVAL;
4640
4641         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4642         if (IS_ERR(link))
4643                 return PTR_ERR(link);
4644
4645         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4646         if (IS_ERR(new_prog)) {
4647                 ret = PTR_ERR(new_prog);
4648                 goto out_put_link;
4649         }
4650
4651         if (flags & BPF_F_REPLACE) {
4652                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4653                 if (IS_ERR(old_prog)) {
4654                         ret = PTR_ERR(old_prog);
4655                         old_prog = NULL;
4656                         goto out_put_progs;
4657                 }
4658         } else if (attr->link_update.old_prog_fd) {
4659                 ret = -EINVAL;
4660                 goto out_put_progs;
4661         }
4662
4663         if (link->ops->update_prog)
4664                 ret = link->ops->update_prog(link, new_prog, old_prog);
4665         else
4666                 ret = -EINVAL;
4667
4668 out_put_progs:
4669         if (old_prog)
4670                 bpf_prog_put(old_prog);
4671         if (ret)
4672                 bpf_prog_put(new_prog);
4673 out_put_link:
4674         bpf_link_put(link);
4675         return ret;
4676 }
4677
4678 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4679
4680 static int link_detach(union bpf_attr *attr)
4681 {
4682         struct bpf_link *link;
4683         int ret;
4684
4685         if (CHECK_ATTR(BPF_LINK_DETACH))
4686                 return -EINVAL;
4687
4688         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4689         if (IS_ERR(link))
4690                 return PTR_ERR(link);
4691
4692         if (link->ops->detach)
4693                 ret = link->ops->detach(link);
4694         else
4695                 ret = -EOPNOTSUPP;
4696
4697         bpf_link_put(link);
4698         return ret;
4699 }
4700
4701 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4702 {
4703         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4704 }
4705
4706 struct bpf_link *bpf_link_by_id(u32 id)
4707 {
4708         struct bpf_link *link;
4709
4710         if (!id)
4711                 return ERR_PTR(-ENOENT);
4712
4713         spin_lock_bh(&link_idr_lock);
4714         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4715         link = idr_find(&link_idr, id);
4716         if (link) {
4717                 if (link->id)
4718                         link = bpf_link_inc_not_zero(link);
4719                 else
4720                         link = ERR_PTR(-EAGAIN);
4721         } else {
4722                 link = ERR_PTR(-ENOENT);
4723         }
4724         spin_unlock_bh(&link_idr_lock);
4725         return link;
4726 }
4727
4728 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
4729 {
4730         struct bpf_link *link;
4731
4732         spin_lock_bh(&link_idr_lock);
4733 again:
4734         link = idr_get_next(&link_idr, id);
4735         if (link) {
4736                 link = bpf_link_inc_not_zero(link);
4737                 if (IS_ERR(link)) {
4738                         (*id)++;
4739                         goto again;
4740                 }
4741         }
4742         spin_unlock_bh(&link_idr_lock);
4743
4744         return link;
4745 }
4746
4747 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4748
4749 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4750 {
4751         struct bpf_link *link;
4752         u32 id = attr->link_id;
4753         int fd;
4754
4755         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4756                 return -EINVAL;
4757
4758         if (!capable(CAP_SYS_ADMIN))
4759                 return -EPERM;
4760
4761         link = bpf_link_by_id(id);
4762         if (IS_ERR(link))
4763                 return PTR_ERR(link);
4764
4765         fd = bpf_link_new_fd(link);
4766         if (fd < 0)
4767                 bpf_link_put(link);
4768
4769         return fd;
4770 }
4771
4772 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4773
4774 static int bpf_stats_release(struct inode *inode, struct file *file)
4775 {
4776         mutex_lock(&bpf_stats_enabled_mutex);
4777         static_key_slow_dec(&bpf_stats_enabled_key.key);
4778         mutex_unlock(&bpf_stats_enabled_mutex);
4779         return 0;
4780 }
4781
4782 static const struct file_operations bpf_stats_fops = {
4783         .release = bpf_stats_release,
4784 };
4785
4786 static int bpf_enable_runtime_stats(void)
4787 {
4788         int fd;
4789
4790         mutex_lock(&bpf_stats_enabled_mutex);
4791
4792         /* Set a very high limit to avoid overflow */
4793         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4794                 mutex_unlock(&bpf_stats_enabled_mutex);
4795                 return -EBUSY;
4796         }
4797
4798         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4799         if (fd >= 0)
4800                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4801
4802         mutex_unlock(&bpf_stats_enabled_mutex);
4803         return fd;
4804 }
4805
4806 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4807
4808 static int bpf_enable_stats(union bpf_attr *attr)
4809 {
4810
4811         if (CHECK_ATTR(BPF_ENABLE_STATS))
4812                 return -EINVAL;
4813
4814         if (!capable(CAP_SYS_ADMIN))
4815                 return -EPERM;
4816
4817         switch (attr->enable_stats.type) {
4818         case BPF_STATS_RUN_TIME:
4819                 return bpf_enable_runtime_stats();
4820         default:
4821                 break;
4822         }
4823         return -EINVAL;
4824 }
4825
4826 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4827
4828 static int bpf_iter_create(union bpf_attr *attr)
4829 {
4830         struct bpf_link *link;
4831         int err;
4832
4833         if (CHECK_ATTR(BPF_ITER_CREATE))
4834                 return -EINVAL;
4835
4836         if (attr->iter_create.flags)
4837                 return -EINVAL;
4838
4839         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4840         if (IS_ERR(link))
4841                 return PTR_ERR(link);
4842
4843         err = bpf_iter_new_fd(link);
4844         bpf_link_put(link);
4845
4846         return err;
4847 }
4848
4849 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4850
4851 static int bpf_prog_bind_map(union bpf_attr *attr)
4852 {
4853         struct bpf_prog *prog;
4854         struct bpf_map *map;
4855         struct bpf_map **used_maps_old, **used_maps_new;
4856         int i, ret = 0;
4857
4858         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4859                 return -EINVAL;
4860
4861         if (attr->prog_bind_map.flags)
4862                 return -EINVAL;
4863
4864         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4865         if (IS_ERR(prog))
4866                 return PTR_ERR(prog);
4867
4868         map = bpf_map_get(attr->prog_bind_map.map_fd);
4869         if (IS_ERR(map)) {
4870                 ret = PTR_ERR(map);
4871                 goto out_prog_put;
4872         }
4873
4874         mutex_lock(&prog->aux->used_maps_mutex);
4875
4876         used_maps_old = prog->aux->used_maps;
4877
4878         for (i = 0; i < prog->aux->used_map_cnt; i++)
4879                 if (used_maps_old[i] == map) {
4880                         bpf_map_put(map);
4881                         goto out_unlock;
4882                 }
4883
4884         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4885                                       sizeof(used_maps_new[0]),
4886                                       GFP_KERNEL);
4887         if (!used_maps_new) {
4888                 ret = -ENOMEM;
4889                 goto out_unlock;
4890         }
4891
4892         memcpy(used_maps_new, used_maps_old,
4893                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4894         used_maps_new[prog->aux->used_map_cnt] = map;
4895
4896         prog->aux->used_map_cnt++;
4897         prog->aux->used_maps = used_maps_new;
4898
4899         kfree(used_maps_old);
4900
4901 out_unlock:
4902         mutex_unlock(&prog->aux->used_maps_mutex);
4903
4904         if (ret)
4905                 bpf_map_put(map);
4906 out_prog_put:
4907         bpf_prog_put(prog);
4908         return ret;
4909 }
4910
4911 static int __sys_bpf(int cmd, bpfptr_t uattr, unsigned int size)
4912 {
4913         union bpf_attr attr;
4914         bool capable;
4915         int err;
4916
4917         capable = bpf_capable() || !sysctl_unprivileged_bpf_disabled;
4918
4919         /* Intent here is for unprivileged_bpf_disabled to block key object
4920          * creation commands for unprivileged users; other actions depend
4921          * of fd availability and access to bpffs, so are dependent on
4922          * object creation success.  Capabilities are later verified for
4923          * operations such as load and map create, so even with unprivileged
4924          * BPF disabled, capability checks are still carried out for these
4925          * and other operations.
4926          */
4927         if (!capable &&
4928             (cmd == BPF_MAP_CREATE || cmd == BPF_PROG_LOAD))
4929                 return -EPERM;
4930
4931         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4932         if (err)
4933                 return err;
4934         size = min_t(u32, size, sizeof(attr));
4935
4936         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4937         memset(&attr, 0, sizeof(attr));
4938         if (copy_from_bpfptr(&attr, uattr, size) != 0)
4939                 return -EFAULT;
4940
4941         err = security_bpf(cmd, &attr, size);
4942         if (err < 0)
4943                 return err;
4944
4945         switch (cmd) {
4946         case BPF_MAP_CREATE:
4947                 err = map_create(&attr);
4948                 break;
4949         case BPF_MAP_LOOKUP_ELEM:
4950                 err = map_lookup_elem(&attr);
4951                 break;
4952         case BPF_MAP_UPDATE_ELEM:
4953                 err = map_update_elem(&attr, uattr);
4954                 break;
4955         case BPF_MAP_DELETE_ELEM:
4956                 err = map_delete_elem(&attr, uattr);
4957                 break;
4958         case BPF_MAP_GET_NEXT_KEY:
4959                 err = map_get_next_key(&attr);
4960                 break;
4961         case BPF_MAP_FREEZE:
4962                 err = map_freeze(&attr);
4963                 break;
4964         case BPF_PROG_LOAD:
4965                 err = bpf_prog_load(&attr, uattr);
4966                 break;
4967         case BPF_OBJ_PIN:
4968                 err = bpf_obj_pin(&attr);
4969                 break;
4970         case BPF_OBJ_GET:
4971                 err = bpf_obj_get(&attr);
4972                 break;
4973         case BPF_PROG_ATTACH:
4974                 err = bpf_prog_attach(&attr);
4975                 break;
4976         case BPF_PROG_DETACH:
4977                 err = bpf_prog_detach(&attr);
4978                 break;
4979         case BPF_PROG_QUERY:
4980                 err = bpf_prog_query(&attr, uattr.user);
4981                 break;
4982         case BPF_PROG_TEST_RUN:
4983                 err = bpf_prog_test_run(&attr, uattr.user);
4984                 break;
4985         case BPF_PROG_GET_NEXT_ID:
4986                 err = bpf_obj_get_next_id(&attr, uattr.user,
4987                                           &prog_idr, &prog_idr_lock);
4988                 break;
4989         case BPF_MAP_GET_NEXT_ID:
4990                 err = bpf_obj_get_next_id(&attr, uattr.user,
4991                                           &map_idr, &map_idr_lock);
4992                 break;
4993         case BPF_BTF_GET_NEXT_ID:
4994                 err = bpf_obj_get_next_id(&attr, uattr.user,
4995                                           &btf_idr, &btf_idr_lock);
4996                 break;
4997         case BPF_PROG_GET_FD_BY_ID:
4998                 err = bpf_prog_get_fd_by_id(&attr);
4999                 break;
5000         case BPF_MAP_GET_FD_BY_ID:
5001                 err = bpf_map_get_fd_by_id(&attr);
5002                 break;
5003         case BPF_OBJ_GET_INFO_BY_FD:
5004                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5005                 break;
5006         case BPF_RAW_TRACEPOINT_OPEN:
5007                 err = bpf_raw_tracepoint_open(&attr);
5008                 break;
5009         case BPF_BTF_LOAD:
5010                 err = bpf_btf_load(&attr, uattr);
5011                 break;
5012         case BPF_BTF_GET_FD_BY_ID:
5013                 err = bpf_btf_get_fd_by_id(&attr);
5014                 break;
5015         case BPF_TASK_FD_QUERY:
5016                 err = bpf_task_fd_query(&attr, uattr.user);
5017                 break;
5018         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5019                 err = map_lookup_and_delete_elem(&attr);
5020                 break;
5021         case BPF_MAP_LOOKUP_BATCH:
5022                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5023                 break;
5024         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5025                 err = bpf_map_do_batch(&attr, uattr.user,
5026                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5027                 break;
5028         case BPF_MAP_UPDATE_BATCH:
5029                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5030                 break;
5031         case BPF_MAP_DELETE_BATCH:
5032                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5033                 break;
5034         case BPF_LINK_CREATE:
5035                 err = link_create(&attr, uattr);
5036                 break;
5037         case BPF_LINK_UPDATE:
5038                 err = link_update(&attr);
5039                 break;
5040         case BPF_LINK_GET_FD_BY_ID:
5041                 err = bpf_link_get_fd_by_id(&attr);
5042                 break;
5043         case BPF_LINK_GET_NEXT_ID:
5044                 err = bpf_obj_get_next_id(&attr, uattr.user,
5045                                           &link_idr, &link_idr_lock);
5046                 break;
5047         case BPF_ENABLE_STATS:
5048                 err = bpf_enable_stats(&attr);
5049                 break;
5050         case BPF_ITER_CREATE:
5051                 err = bpf_iter_create(&attr);
5052                 break;
5053         case BPF_LINK_DETACH:
5054                 err = link_detach(&attr);
5055                 break;
5056         case BPF_PROG_BIND_MAP:
5057                 err = bpf_prog_bind_map(&attr);
5058                 break;
5059         default:
5060                 err = -EINVAL;
5061                 break;
5062         }
5063
5064         return err;
5065 }
5066
5067 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5068 {
5069         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5070 }
5071
5072 static bool syscall_prog_is_valid_access(int off, int size,
5073                                          enum bpf_access_type type,
5074                                          const struct bpf_prog *prog,
5075                                          struct bpf_insn_access_aux *info)
5076 {
5077         if (off < 0 || off >= U16_MAX)
5078                 return false;
5079         if (off % size != 0)
5080                 return false;
5081         return true;
5082 }
5083
5084 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5085 {
5086         switch (cmd) {
5087         case BPF_MAP_CREATE:
5088         case BPF_MAP_DELETE_ELEM:
5089         case BPF_MAP_UPDATE_ELEM:
5090         case BPF_MAP_FREEZE:
5091         case BPF_MAP_GET_FD_BY_ID:
5092         case BPF_PROG_LOAD:
5093         case BPF_BTF_LOAD:
5094         case BPF_LINK_CREATE:
5095         case BPF_RAW_TRACEPOINT_OPEN:
5096                 break;
5097         default:
5098                 return -EINVAL;
5099         }
5100         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5101 }
5102
5103
5104 /* To shut up -Wmissing-prototypes.
5105  * This function is used by the kernel light skeleton
5106  * to load bpf programs when modules are loaded or during kernel boot.
5107  * See tools/lib/bpf/skel_internal.h
5108  */
5109 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5110
5111 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5112 {
5113         struct bpf_prog * __maybe_unused prog;
5114         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5115
5116         switch (cmd) {
5117 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5118         case BPF_PROG_TEST_RUN:
5119                 if (attr->test.data_in || attr->test.data_out ||
5120                     attr->test.ctx_out || attr->test.duration ||
5121                     attr->test.repeat || attr->test.flags)
5122                         return -EINVAL;
5123
5124                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5125                 if (IS_ERR(prog))
5126                         return PTR_ERR(prog);
5127
5128                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5129                     attr->test.ctx_size_in > U16_MAX) {
5130                         bpf_prog_put(prog);
5131                         return -EINVAL;
5132                 }
5133
5134                 run_ctx.bpf_cookie = 0;
5135                 run_ctx.saved_run_ctx = NULL;
5136                 if (!__bpf_prog_enter_sleepable(prog, &run_ctx)) {
5137                         /* recursion detected */
5138                         bpf_prog_put(prog);
5139                         return -EBUSY;
5140                 }
5141                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5142                 __bpf_prog_exit_sleepable(prog, 0 /* bpf_prog_run does runtime stats */, &run_ctx);
5143                 bpf_prog_put(prog);
5144                 return 0;
5145 #endif
5146         default:
5147                 return ____bpf_sys_bpf(cmd, attr, size);
5148         }
5149 }
5150 EXPORT_SYMBOL(kern_sys_bpf);
5151
5152 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5153         .func           = bpf_sys_bpf,
5154         .gpl_only       = false,
5155         .ret_type       = RET_INTEGER,
5156         .arg1_type      = ARG_ANYTHING,
5157         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5158         .arg3_type      = ARG_CONST_SIZE,
5159 };
5160
5161 const struct bpf_func_proto * __weak
5162 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5163 {
5164         return bpf_base_func_proto(func_id);
5165 }
5166
5167 BPF_CALL_1(bpf_sys_close, u32, fd)
5168 {
5169         /* When bpf program calls this helper there should not be
5170          * an fdget() without matching completed fdput().
5171          * This helper is allowed in the following callchain only:
5172          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5173          */
5174         return close_fd(fd);
5175 }
5176
5177 static const struct bpf_func_proto bpf_sys_close_proto = {
5178         .func           = bpf_sys_close,
5179         .gpl_only       = false,
5180         .ret_type       = RET_INTEGER,
5181         .arg1_type      = ARG_ANYTHING,
5182 };
5183
5184 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5185 {
5186         if (flags)
5187                 return -EINVAL;
5188
5189         if (name_sz <= 1 || name[name_sz - 1])
5190                 return -EINVAL;
5191
5192         if (!bpf_dump_raw_ok(current_cred()))
5193                 return -EPERM;
5194
5195         *res = kallsyms_lookup_name(name);
5196         return *res ? 0 : -ENOENT;
5197 }
5198
5199 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5200         .func           = bpf_kallsyms_lookup_name,
5201         .gpl_only       = false,
5202         .ret_type       = RET_INTEGER,
5203         .arg1_type      = ARG_PTR_TO_MEM,
5204         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5205         .arg3_type      = ARG_ANYTHING,
5206         .arg4_type      = ARG_PTR_TO_LONG,
5207 };
5208
5209 static const struct bpf_func_proto *
5210 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5211 {
5212         switch (func_id) {
5213         case BPF_FUNC_sys_bpf:
5214                 return !perfmon_capable() ? NULL : &bpf_sys_bpf_proto;
5215         case BPF_FUNC_btf_find_by_name_kind:
5216                 return &bpf_btf_find_by_name_kind_proto;
5217         case BPF_FUNC_sys_close:
5218                 return &bpf_sys_close_proto;
5219         case BPF_FUNC_kallsyms_lookup_name:
5220                 return &bpf_kallsyms_lookup_name_proto;
5221         default:
5222                 return tracing_prog_func_proto(func_id, prog);
5223         }
5224 }
5225
5226 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5227         .get_func_proto  = syscall_prog_func_proto,
5228         .is_valid_access = syscall_prog_is_valid_access,
5229 };
5230
5231 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5232         .test_run = bpf_prog_test_run_syscall,
5233 };
5234
5235 #ifdef CONFIG_SYSCTL
5236 static int bpf_stats_handler(struct ctl_table *table, int write,
5237                              void *buffer, size_t *lenp, loff_t *ppos)
5238 {
5239         struct static_key *key = (struct static_key *)table->data;
5240         static int saved_val;
5241         int val, ret;
5242         struct ctl_table tmp = {
5243                 .data   = &val,
5244                 .maxlen = sizeof(val),
5245                 .mode   = table->mode,
5246                 .extra1 = SYSCTL_ZERO,
5247                 .extra2 = SYSCTL_ONE,
5248         };
5249
5250         if (write && !capable(CAP_SYS_ADMIN))
5251                 return -EPERM;
5252
5253         mutex_lock(&bpf_stats_enabled_mutex);
5254         val = saved_val;
5255         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5256         if (write && !ret && val != saved_val) {
5257                 if (val)
5258                         static_key_slow_inc(key);
5259                 else
5260                         static_key_slow_dec(key);
5261                 saved_val = val;
5262         }
5263         mutex_unlock(&bpf_stats_enabled_mutex);
5264         return ret;
5265 }
5266
5267 void __weak unpriv_ebpf_notify(int new_state)
5268 {
5269 }
5270
5271 static int bpf_unpriv_handler(struct ctl_table *table, int write,
5272                               void *buffer, size_t *lenp, loff_t *ppos)
5273 {
5274         int ret, unpriv_enable = *(int *)table->data;
5275         bool locked_state = unpriv_enable == 1;
5276         struct ctl_table tmp = *table;
5277
5278         if (write && !capable(CAP_SYS_ADMIN))
5279                 return -EPERM;
5280
5281         tmp.data = &unpriv_enable;
5282         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5283         if (write && !ret) {
5284                 if (locked_state && unpriv_enable != 1)
5285                         return -EPERM;
5286                 *(int *)table->data = unpriv_enable;
5287         }
5288
5289         unpriv_ebpf_notify(unpriv_enable);
5290
5291         return ret;
5292 }
5293
5294 static struct ctl_table bpf_syscall_table[] = {
5295         {
5296                 .procname       = "unprivileged_bpf_disabled",
5297                 .data           = &sysctl_unprivileged_bpf_disabled,
5298                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5299                 .mode           = 0644,
5300                 .proc_handler   = bpf_unpriv_handler,
5301                 .extra1         = SYSCTL_ZERO,
5302                 .extra2         = SYSCTL_TWO,
5303         },
5304         {
5305                 .procname       = "bpf_stats_enabled",
5306                 .data           = &bpf_stats_enabled_key.key,
5307                 .maxlen         = sizeof(bpf_stats_enabled_key),
5308                 .mode           = 0644,
5309                 .proc_handler   = bpf_stats_handler,
5310         },
5311         { }
5312 };
5313
5314 static int __init bpf_syscall_sysctl_init(void)
5315 {
5316         register_sysctl_init("kernel", bpf_syscall_table);
5317         return 0;
5318 }
5319 late_initcall(bpf_syscall_sysctl_init);
5320 #endif /* CONFIG_SYSCTL */