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