kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf-cgroup.h>
   6 #include <linux/bpf_trace.h>
   7 #include <linux/bpf_lirc.h>
   8 #include <linux/bpf_verifier.h>
   9 #include <linux/bsearch.h>
  10 #include <linux/btf.h>
  11 #include <linux/syscalls.h>
  12 #include <linux/slab.h>
  13 #include <linux/sched/signal.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/mmzone.h>
  16 #include <linux/anon_inodes.h>
  17 #include <linux/fdtable.h>
  18 #include <linux/file.h>
  19 #include <linux/fs.h>
  20 #include <linux/license.h>
  21 #include <linux/filter.h>
  22 #include <linux/kernel.h>
  23 #include <linux/idr.h>
  24 #include <linux/cred.h>
  25 #include <linux/timekeeping.h>
  26 #include <linux/ctype.h>
  27 #include <linux/nospec.h>
  28 #include <linux/audit.h>
  29 #include <uapi/linux/btf.h>
  30 #include <linux/pgtable.h>
  31 #include <linux/bpf_lsm.h>
  32 #include <linux/poll.h>
  33 #include <linux/sort.h>
  34 #include <linux/bpf-netns.h>
  35 #include <linux/rcupdate_trace.h>
  36 #include <linux/memcontrol.h>
  37 #include <linux/trace_events.h>
  38
  39 #include <net/netfilter/nf_bpf_link.h>
  40 #include <net/netkit.h>
  41 #include <net/tcx.h>
  42
  43 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  44                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  45                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  46 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  47 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  48 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  49                         IS_FD_HASH(map))
  50
  51 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  52
  53 DEFINE_PER_CPU(int, bpf_prog_active);
  54 static DEFINE_IDR(prog_idr);
  55 static DEFINE_SPINLOCK(prog_idr_lock);
  56 static DEFINE_IDR(map_idr);
  57 static DEFINE_SPINLOCK(map_idr_lock);
  58 static DEFINE_IDR(link_idr);
  59 static DEFINE_SPINLOCK(link_idr_lock);
  60
  61 int sysctl_unprivileged_bpf_disabled __read_mostly =
  62         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  63
  64 static const struct bpf_map_ops * const bpf_map_types[] = {
  65 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  66 #define BPF_MAP_TYPE(_id, _ops) \
  67         [_id] = &_ops,
  68 #define BPF_LINK_TYPE(_id, _name)
  69 #include <linux/bpf_types.h>
  70 #undef BPF_PROG_TYPE
  71 #undef BPF_MAP_TYPE
  72 #undef BPF_LINK_TYPE
  73 };
  74
  75 /*
  76  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  77  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  78  * we don't know about yet.
  79  *
  80  * There is a ToCToU between this function call and the following
  81  * copy_from_user() call. However, this is not a concern since this function is
  82  * meant to be a future-proofing of bits.
  83  */
  84 int bpf_check_uarg_tail_zero(bpfptr_t uaddr,
  85                              size_t expected_size,
  86                              size_t actual_size)
  87 {
  88         int res;
  89
  90         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  91                 return -E2BIG;
  92
  93         if (actual_size <= expected_size)
  94                 return 0;
  95
  96         if (uaddr.is_kernel)
  97                 res = memchr_inv(uaddr.kernel + expected_size, 0,
  98                                  actual_size - expected_size) == NULL;
  99         else
 100                 res = check_zeroed_user(uaddr.user + expected_size,
 101                                         actual_size - expected_size);
 102         if (res < 0)
 103                 return res;
 104         return res ? 0 : -E2BIG;
 105 }
 106
 107 const struct bpf_map_ops bpf_map_offload_ops = {
 108         .map_meta_equal = bpf_map_meta_equal,
 109         .map_alloc = bpf_map_offload_map_alloc,
 110         .map_free = bpf_map_offload_map_free,
 111         .map_check_btf = map_check_no_btf,
 112         .map_mem_usage = bpf_map_offload_map_mem_usage,
 113 };
 114
 115 static void bpf_map_write_active_inc(struct bpf_map *map)
 116 {
 117         atomic64_inc(&map->writecnt);
 118 }
 119
 120 static void bpf_map_write_active_dec(struct bpf_map *map)
 121 {
 122         atomic64_dec(&map->writecnt);
 123 }
 124
 125 bool bpf_map_write_active(const struct bpf_map *map)
 126 {
 127         return atomic64_read(&map->writecnt) != 0;
 128 }
 129
 130 static u32 bpf_map_value_size(const struct bpf_map *map)
 131 {
 132         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 133             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 134             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 135             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 136                 return round_up(map->value_size, 8) * num_possible_cpus();
 137         else if (IS_FD_MAP(map))
 138                 return sizeof(u32);
 139         else
 140                 return  map->value_size;
 141 }
 142
 143 static void maybe_wait_bpf_programs(struct bpf_map *map)
 144 {
 145         /* Wait for any running non-sleepable BPF programs to complete so that
 146          * userspace, when we return to it, knows that all non-sleepable
 147          * programs that could be running use the new map value. For sleepable
 148          * BPF programs, synchronize_rcu_tasks_trace() should be used to wait
 149          * for the completions of these programs, but considering the waiting
 150          * time can be very long and userspace may think it will hang forever,
 151          * so don't handle sleepable BPF programs now.
 152          */
 153         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 154             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 155                 synchronize_rcu();
 156 }
 157
 158 static int bpf_map_update_value(struct bpf_map *map, struct file *map_file,
 159                                 void *key, void *value, __u64 flags)
 160 {
 161         int err;
 162
 163         /* Need to create a kthread, thus must support schedule */
 164         if (bpf_map_is_offloaded(map)) {
 165                 return bpf_map_offload_update_elem(map, key, value, flags);
 166         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 167                    map->map_type == BPF_MAP_TYPE_ARENA ||
 168                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 169                 return map->ops->map_update_elem(map, key, value, flags);
 170         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 171                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 172                 return sock_map_update_elem_sys(map, key, value, flags);
 173         } else if (IS_FD_PROG_ARRAY(map)) {
 174                 return bpf_fd_array_map_update_elem(map, map_file, key, value,
 175                                                     flags);
 176         }
 177
 178         bpf_disable_instrumentation();
 179         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 180             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 181                 err = bpf_percpu_hash_update(map, key, value, flags);
 182         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 183                 err = bpf_percpu_array_update(map, key, value, flags);
 184         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 185                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 186                                                        flags);
 187         } else if (IS_FD_ARRAY(map)) {
 188                 err = bpf_fd_array_map_update_elem(map, map_file, key, value,
 189                                                    flags);
 190         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 191                 err = bpf_fd_htab_map_update_elem(map, map_file, key, value,
 192                                                   flags);
 193         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 194                 /* rcu_read_lock() is not needed */
 195                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 196                                                          flags);
 197         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 198                    map->map_type == BPF_MAP_TYPE_STACK ||
 199                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 200                 err = map->ops->map_push_elem(map, value, flags);
 201         } else {
 202                 rcu_read_lock();
 203                 err = map->ops->map_update_elem(map, key, value, flags);
 204                 rcu_read_unlock();
 205         }
 206         bpf_enable_instrumentation();
 207
 208         return err;
 209 }
 210
 211 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 212                               __u64 flags)
 213 {
 214         void *ptr;
 215         int err;
 216
 217         if (bpf_map_is_offloaded(map))
 218                 return bpf_map_offload_lookup_elem(map, key, value);
 219
 220         bpf_disable_instrumentation();
 221         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 222             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 223                 err = bpf_percpu_hash_copy(map, key, value);
 224         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 225                 err = bpf_percpu_array_copy(map, key, value);
 226         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 227                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 228         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 229                 err = bpf_stackmap_copy(map, key, value);
 230         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 231                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 232         } else if (IS_FD_HASH(map)) {
 233                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 234         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 235                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 236         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 237                    map->map_type == BPF_MAP_TYPE_STACK ||
 238                    map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
 239                 err = map->ops->map_peek_elem(map, value);
 240         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 241                 /* struct_ops map requires directly updating "value" */
 242                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 243         } else {
 244                 rcu_read_lock();
 245                 if (map->ops->map_lookup_elem_sys_only)
 246                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 247                 else
 248                         ptr = map->ops->map_lookup_elem(map, key);
 249                 if (IS_ERR(ptr)) {
 250                         err = PTR_ERR(ptr);
 251                 } else if (!ptr) {
 252                         err = -ENOENT;
 253                 } else {
 254                         err = 0;
 255                         if (flags & BPF_F_LOCK)
 256                                 /* lock 'ptr' and copy everything but lock */
 257                                 copy_map_value_locked(map, value, ptr, true);
 258                         else
 259                                 copy_map_value(map, value, ptr);
 260                         /* mask lock and timer, since value wasn't zero inited */
 261                         check_and_init_map_value(map, value);
 262                 }
 263                 rcu_read_unlock();
 264         }
 265
 266         bpf_enable_instrumentation();
 267
 268         return err;
 269 }
 270
 271 /* Please, do not use this function outside from the map creation path
 272  * (e.g. in map update path) without taking care of setting the active
 273  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 274  */
 275 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 276 {
 277         /* We really just want to fail instead of triggering OOM killer
 278          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 279          * which is used for lower order allocation requests.
 280          *
 281          * It has been observed that higher order allocation requests done by
 282          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 283          * to reclaim memory from the page cache, thus we set
 284          * __GFP_RETRY_MAYFAIL to avoid such situations.
 285          */
 286
 287         gfp_t gfp = bpf_memcg_flags(__GFP_NOWARN | __GFP_ZERO);
 288         unsigned int flags = 0;
 289         unsigned long align = 1;
 290         void *area;
 291
 292         if (size >= SIZE_MAX)
 293                 return NULL;
 294
 295         /* kmalloc()'ed memory can't be mmap()'ed */
 296         if (mmapable) {
 297                 BUG_ON(!PAGE_ALIGNED(size));
 298                 align = SHMLBA;
 299                 flags = VM_USERMAP;
 300         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 301                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 302                                     numa_node);
 303                 if (area != NULL)
 304                         return area;
 305         }
 306
 307         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 308                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 309                         flags, numa_node, __builtin_return_address(0));
 310 }
 311
 312 void *bpf_map_area_alloc(u64 size, int numa_node)
 313 {
 314         return __bpf_map_area_alloc(size, numa_node, false);
 315 }
 316
 317 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 318 {
 319         return __bpf_map_area_alloc(size, numa_node, true);
 320 }
 321
 322 void bpf_map_area_free(void *area)
 323 {
 324         kvfree(area);
 325 }
 326
 327 static u32 bpf_map_flags_retain_permanent(u32 flags)
 328 {
 329         /* Some map creation flags are not tied to the map object but
 330          * rather to the map fd instead, so they have no meaning upon
 331          * map object inspection since multiple file descriptors with
 332          * different (access) properties can exist here. Thus, given
 333          * this has zero meaning for the map itself, lets clear these
 334          * from here.
 335          */
 336         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 337 }
 338
 339 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 340 {
 341         map->map_type = attr->map_type;
 342         map->key_size = attr->key_size;
 343         map->value_size = attr->value_size;
 344         map->max_entries = attr->max_entries;
 345         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 346         map->numa_node = bpf_map_attr_numa_node(attr);
 347         map->map_extra = attr->map_extra;
 348 }
 349
 350 static int bpf_map_alloc_id(struct bpf_map *map)
 351 {
 352         int id;
 353
 354         idr_preload(GFP_KERNEL);
 355         spin_lock_bh(&map_idr_lock);
 356         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 357         if (id > 0)
 358                 map->id = id;
 359         spin_unlock_bh(&map_idr_lock);
 360         idr_preload_end();
 361
 362         if (WARN_ON_ONCE(!id))
 363                 return -ENOSPC;
 364
 365         return id > 0 ? 0 : id;
 366 }
 367
 368 void bpf_map_free_id(struct bpf_map *map)
 369 {
 370         unsigned long flags;
 371
 372         /* Offloaded maps are removed from the IDR store when their device
 373          * disappears - even if someone holds an fd to them they are unusable,
 374          * the memory is gone, all ops will fail; they are simply waiting for
 375          * refcnt to drop to be freed.
 376          */
 377         if (!map->id)
 378                 return;
 379
 380         spin_lock_irqsave(&map_idr_lock, flags);
 381
 382         idr_remove(&map_idr, map->id);
 383         map->id = 0;
 384
 385         spin_unlock_irqrestore(&map_idr_lock, flags);
 386 }
 387
 388 #ifdef CONFIG_MEMCG
 389 static void bpf_map_save_memcg(struct bpf_map *map)
 390 {
 391         /* Currently if a map is created by a process belonging to the root
 392          * memory cgroup, get_obj_cgroup_from_current() will return NULL.
 393          * So we have to check map->objcg for being NULL each time it's
 394          * being used.
 395          */
 396         if (memcg_bpf_enabled())
 397                 map->objcg = get_obj_cgroup_from_current();
 398 }
 399
 400 static void bpf_map_release_memcg(struct bpf_map *map)
 401 {
 402         if (map->objcg)
 403                 obj_cgroup_put(map->objcg);
 404 }
 405
 406 static struct mem_cgroup *bpf_map_get_memcg(const struct bpf_map *map)
 407 {
 408         if (map->objcg)
 409                 return get_mem_cgroup_from_objcg(map->objcg);
 410
 411         return root_mem_cgroup;
 412 }
 413
 414 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 415                            int node)
 416 {
 417         struct mem_cgroup *memcg, *old_memcg;
 418         void *ptr;
 419
 420         memcg = bpf_map_get_memcg(map);
 421         old_memcg = set_active_memcg(memcg);
 422         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 423         set_active_memcg(old_memcg);
 424         mem_cgroup_put(memcg);
 425
 426         return ptr;
 427 }
 428
 429 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 430 {
 431         struct mem_cgroup *memcg, *old_memcg;
 432         void *ptr;
 433
 434         memcg = bpf_map_get_memcg(map);
 435         old_memcg = set_active_memcg(memcg);
 436         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 437         set_active_memcg(old_memcg);
 438         mem_cgroup_put(memcg);
 439
 440         return ptr;
 441 }
 442
 443 void *bpf_map_kvcalloc(struct bpf_map *map, size_t n, size_t size,
 444                        gfp_t flags)
 445 {
 446         struct mem_cgroup *memcg, *old_memcg;
 447         void *ptr;
 448
 449         memcg = bpf_map_get_memcg(map);
 450         old_memcg = set_active_memcg(memcg);
 451         ptr = kvcalloc(n, size, flags | __GFP_ACCOUNT);
 452         set_active_memcg(old_memcg);
 453         mem_cgroup_put(memcg);
 454
 455         return ptr;
 456 }
 457
 458 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 459                                     size_t align, gfp_t flags)
 460 {
 461         struct mem_cgroup *memcg, *old_memcg;
 462         void __percpu *ptr;
 463
 464         memcg = bpf_map_get_memcg(map);
 465         old_memcg = set_active_memcg(memcg);
 466         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 467         set_active_memcg(old_memcg);
 468         mem_cgroup_put(memcg);
 469
 470         return ptr;
 471 }
 472
 473 #else
 474 static void bpf_map_save_memcg(struct bpf_map *map)
 475 {
 476 }
 477
 478 static void bpf_map_release_memcg(struct bpf_map *map)
 479 {
 480 }
 481 #endif
 482
 483 int bpf_map_alloc_pages(const struct bpf_map *map, gfp_t gfp, int nid,
 484                         unsigned long nr_pages, struct page **pages)
 485 {
 486         unsigned long i, j;
 487         struct page *pg;
 488         int ret = 0;
 489 #ifdef CONFIG_MEMCG
 490         struct mem_cgroup *memcg, *old_memcg;
 491
 492         memcg = bpf_map_get_memcg(map);
 493         old_memcg = set_active_memcg(memcg);
 494 #endif
 495         for (i = 0; i < nr_pages; i++) {
 496                 pg = alloc_pages_node(nid, gfp | __GFP_ACCOUNT, 0);
 497
 498                 if (pg) {
 499                         pages[i] = pg;
 500                         continue;
 501                 }
 502                 for (j = 0; j < i; j++)
 503                         __free_page(pages[j]);
 504                 ret = -ENOMEM;
 505                 break;
 506         }
 507
 508 #ifdef CONFIG_MEMCG
 509         set_active_memcg(old_memcg);
 510         mem_cgroup_put(memcg);
 511 #endif
 512         return ret;
 513 }
 514
 515
 516 static int btf_field_cmp(const void *a, const void *b)
 517 {
 518         const struct btf_field *f1 = a, *f2 = b;
 519
 520         if (f1->offset < f2->offset)
 521                 return -1;
 522         else if (f1->offset > f2->offset)
 523                 return 1;
 524         return 0;
 525 }
 526
 527 struct btf_field *btf_record_find(const struct btf_record *rec, u32 offset,
 528                                   u32 field_mask)
 529 {
 530         struct btf_field *field;
 531
 532         if (IS_ERR_OR_NULL(rec) || !(rec->field_mask & field_mask))
 533                 return NULL;
 534         field = bsearch(&offset, rec->fields, rec->cnt, sizeof(rec->fields[0]), btf_field_cmp);
 535         if (!field || !(field->type & field_mask))
 536                 return NULL;
 537         return field;
 538 }
 539
 540 void btf_record_free(struct btf_record *rec)
 541 {
 542         int i;
 543
 544         if (IS_ERR_OR_NULL(rec))
 545                 return;
 546         for (i = 0; i < rec->cnt; i++) {
 547                 switch (rec->fields[i].type) {
 548                 case BPF_KPTR_UNREF:
 549                 case BPF_KPTR_REF:
 550                 case BPF_KPTR_PERCPU:
 551                         if (rec->fields[i].kptr.module)
 552                                 module_put(rec->fields[i].kptr.module);
 553                         if (btf_is_kernel(rec->fields[i].kptr.btf))
 554                                 btf_put(rec->fields[i].kptr.btf);
 555                         break;
 556                 case BPF_LIST_HEAD:
 557                 case BPF_LIST_NODE:
 558                 case BPF_RB_ROOT:
 559                 case BPF_RB_NODE:
 560                 case BPF_SPIN_LOCK:
 561                 case BPF_TIMER:
 562                 case BPF_REFCOUNT:
 563                 case BPF_WORKQUEUE:
 564                         /* Nothing to release */
 565                         break;
 566                 default:
 567                         WARN_ON_ONCE(1);
 568                         continue;
 569                 }
 570         }
 571         kfree(rec);
 572 }
 573
 574 void bpf_map_free_record(struct bpf_map *map)
 575 {
 576         btf_record_free(map->record);
 577         map->record = NULL;
 578 }
 579
 580 struct btf_record *btf_record_dup(const struct btf_record *rec)
 581 {
 582         const struct btf_field *fields;
 583         struct btf_record *new_rec;
 584         int ret, size, i;
 585
 586         if (IS_ERR_OR_NULL(rec))
 587                 return NULL;
 588         size = offsetof(struct btf_record, fields[rec->cnt]);
 589         new_rec = kmemdup(rec, size, GFP_KERNEL | __GFP_NOWARN);
 590         if (!new_rec)
 591                 return ERR_PTR(-ENOMEM);
 592         /* Do a deep copy of the btf_record */
 593         fields = rec->fields;
 594         new_rec->cnt = 0;
 595         for (i = 0; i < rec->cnt; i++) {
 596                 switch (fields[i].type) {
 597                 case BPF_KPTR_UNREF:
 598                 case BPF_KPTR_REF:
 599                 case BPF_KPTR_PERCPU:
 600                         if (btf_is_kernel(fields[i].kptr.btf))
 601                                 btf_get(fields[i].kptr.btf);
 602                         if (fields[i].kptr.module && !try_module_get(fields[i].kptr.module)) {
 603                                 ret = -ENXIO;
 604                                 goto free;
 605                         }
 606                         break;
 607                 case BPF_LIST_HEAD:
 608                 case BPF_LIST_NODE:
 609                 case BPF_RB_ROOT:
 610                 case BPF_RB_NODE:
 611                 case BPF_SPIN_LOCK:
 612                 case BPF_TIMER:
 613                 case BPF_REFCOUNT:
 614                 case BPF_WORKQUEUE:
 615                         /* Nothing to acquire */
 616                         break;
 617                 default:
 618                         ret = -EFAULT;
 619                         WARN_ON_ONCE(1);
 620                         goto free;
 621                 }
 622                 new_rec->cnt++;
 623         }
 624         return new_rec;
 625 free:
 626         btf_record_free(new_rec);
 627         return ERR_PTR(ret);
 628 }
 629
 630 bool btf_record_equal(const struct btf_record *rec_a, const struct btf_record *rec_b)
 631 {
 632         bool a_has_fields = !IS_ERR_OR_NULL(rec_a), b_has_fields = !IS_ERR_OR_NULL(rec_b);
 633         int size;
 634
 635         if (!a_has_fields && !b_has_fields)
 636                 return true;
 637         if (a_has_fields != b_has_fields)
 638                 return false;
 639         if (rec_a->cnt != rec_b->cnt)
 640                 return false;
 641         size = offsetof(struct btf_record, fields[rec_a->cnt]);
 642         /* btf_parse_fields uses kzalloc to allocate a btf_record, so unused
 643          * members are zeroed out. So memcmp is safe to do without worrying
 644          * about padding/unused fields.
 645          *
 646          * While spin_lock, timer, and kptr have no relation to map BTF,
 647          * list_head metadata is specific to map BTF, the btf and value_rec
 648          * members in particular. btf is the map BTF, while value_rec points to
 649          * btf_record in that map BTF.
 650          *
 651          * So while by default, we don't rely on the map BTF (which the records
 652          * were parsed from) matching for both records, which is not backwards
 653          * compatible, in case list_head is part of it, we implicitly rely on
 654          * that by way of depending on memcmp succeeding for it.
 655          */
 656         return !memcmp(rec_a, rec_b, size);
 657 }
 658
 659 void bpf_obj_free_timer(const struct btf_record *rec, void *obj)
 660 {
 661         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_TIMER)))
 662                 return;
 663         bpf_timer_cancel_and_free(obj + rec->timer_off);
 664 }
 665
 666 void bpf_obj_free_workqueue(const struct btf_record *rec, void *obj)
 667 {
 668         if (WARN_ON_ONCE(!btf_record_has_field(rec, BPF_WORKQUEUE)))
 669                 return;
 670         bpf_wq_cancel_and_free(obj + rec->wq_off);
 671 }
 672
 673 void bpf_obj_free_fields(const struct btf_record *rec, void *obj)
 674 {
 675         const struct btf_field *fields;
 676         int i;
 677
 678         if (IS_ERR_OR_NULL(rec))
 679                 return;
 680         fields = rec->fields;
 681         for (i = 0; i < rec->cnt; i++) {
 682                 struct btf_struct_meta *pointee_struct_meta;
 683                 const struct btf_field *field = &fields[i];
 684                 void *field_ptr = obj + field->offset;
 685                 void *xchgd_field;
 686
 687                 switch (fields[i].type) {
 688                 case BPF_SPIN_LOCK:
 689                         break;
 690                 case BPF_TIMER:
 691                         bpf_timer_cancel_and_free(field_ptr);
 692                         break;
 693                 case BPF_WORKQUEUE:
 694                         bpf_wq_cancel_and_free(field_ptr);
 695                         break;
 696                 case BPF_KPTR_UNREF:
 697                         WRITE_ONCE(*(u64 *)field_ptr, 0);
 698                         break;
 699                 case BPF_KPTR_REF:
 700                 case BPF_KPTR_PERCPU:
 701                         xchgd_field = (void *)xchg((unsigned long *)field_ptr, 0);
 702                         if (!xchgd_field)
 703                                 break;
 704
 705                         if (!btf_is_kernel(field->kptr.btf)) {
 706                                 pointee_struct_meta = btf_find_struct_meta(field->kptr.btf,
 707                                                                            field->kptr.btf_id);
 708                                 migrate_disable();
 709                                 __bpf_obj_drop_impl(xchgd_field, pointee_struct_meta ?
 710                                                                  pointee_struct_meta->record : NULL,
 711                                                                  fields[i].type == BPF_KPTR_PERCPU);
 712                                 migrate_enable();
 713                         } else {
 714                                 field->kptr.dtor(xchgd_field);
 715                         }
 716                         break;
 717                 case BPF_LIST_HEAD:
 718                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 719                                 continue;
 720                         bpf_list_head_free(field, field_ptr, obj + rec->spin_lock_off);
 721                         break;
 722                 case BPF_RB_ROOT:
 723                         if (WARN_ON_ONCE(rec->spin_lock_off < 0))
 724                                 continue;
 725                         bpf_rb_root_free(field, field_ptr, obj + rec->spin_lock_off);
 726                         break;
 727                 case BPF_LIST_NODE:
 728                 case BPF_RB_NODE:
 729                 case BPF_REFCOUNT:
 730                         break;
 731                 default:
 732                         WARN_ON_ONCE(1);
 733                         continue;
 734                 }
 735         }
 736 }
 737
 738 static void bpf_map_free(struct bpf_map *map)
 739 {
 740         struct btf_record *rec = map->record;
 741         struct btf *btf = map->btf;
 742
 743         /* implementation dependent freeing */
 744         map->ops->map_free(map);
 745         /* Delay freeing of btf_record for maps, as map_free
 746          * callback usually needs access to them. It is better to do it here
 747          * than require each callback to do the free itself manually.
 748          *
 749          * Note that the btf_record stashed in map->inner_map_meta->record was
 750          * already freed using the map_free callback for map in map case which
 751          * eventually calls bpf_map_free_meta, since inner_map_meta is only a
 752          * template bpf_map struct used during verification.
 753          */
 754         btf_record_free(rec);
 755         /* Delay freeing of btf for maps, as map_free callback may need
 756          * struct_meta info which will be freed with btf_put().
 757          */
 758         btf_put(btf);
 759 }
 760
 761 /* called from workqueue */
 762 static void bpf_map_free_deferred(struct work_struct *work)
 763 {
 764         struct bpf_map *map = container_of(work, struct bpf_map, work);
 765
 766         security_bpf_map_free(map);
 767         bpf_map_release_memcg(map);
 768         bpf_map_free(map);
 769 }
 770
 771 static void bpf_map_put_uref(struct bpf_map *map)
 772 {
 773         if (atomic64_dec_and_test(&map->usercnt)) {
 774                 if (map->ops->map_release_uref)
 775                         map->ops->map_release_uref(map);
 776         }
 777 }
 778
 779 static void bpf_map_free_in_work(struct bpf_map *map)
 780 {
 781         INIT_WORK(&map->work, bpf_map_free_deferred);
 782         /* Avoid spawning kworkers, since they all might contend
 783          * for the same mutex like slab_mutex.
 784          */
 785         queue_work(system_unbound_wq, &map->work);
 786 }
 787
 788 static void bpf_map_free_rcu_gp(struct rcu_head *rcu)
 789 {
 790         bpf_map_free_in_work(container_of(rcu, struct bpf_map, rcu));
 791 }
 792
 793 static void bpf_map_free_mult_rcu_gp(struct rcu_head *rcu)
 794 {
 795         if (rcu_trace_implies_rcu_gp())
 796                 bpf_map_free_rcu_gp(rcu);
 797         else
 798                 call_rcu(rcu, bpf_map_free_rcu_gp);
 799 }
 800
 801 /* decrement map refcnt and schedule it for freeing via workqueue
 802  * (underlying map implementation ops->map_free() might sleep)
 803  */
 804 void bpf_map_put(struct bpf_map *map)
 805 {
 806         if (atomic64_dec_and_test(&map->refcnt)) {
 807                 /* bpf_map_free_id() must be called first */
 808                 bpf_map_free_id(map);
 809
 810                 WARN_ON_ONCE(atomic64_read(&map->sleepable_refcnt));
 811                 if (READ_ONCE(map->free_after_mult_rcu_gp))
 812                         call_rcu_tasks_trace(&map->rcu, bpf_map_free_mult_rcu_gp);
 813                 else if (READ_ONCE(map->free_after_rcu_gp))
 814                         call_rcu(&map->rcu, bpf_map_free_rcu_gp);
 815                 else
 816                         bpf_map_free_in_work(map);
 817         }
 818 }
 819 EXPORT_SYMBOL_GPL(bpf_map_put);
 820
 821 void bpf_map_put_with_uref(struct bpf_map *map)
 822 {
 823         bpf_map_put_uref(map);
 824         bpf_map_put(map);
 825 }
 826
 827 static int bpf_map_release(struct inode *inode, struct file *filp)
 828 {
 829         struct bpf_map *map = filp->private_data;
 830
 831         if (map->ops->map_release)
 832                 map->ops->map_release(map, filp);
 833
 834         bpf_map_put_with_uref(map);
 835         return 0;
 836 }
 837
 838 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 839 {
 840         fmode_t mode = fd_file(f)->f_mode;
 841
 842         /* Our file permissions may have been overridden by global
 843          * map permissions facing syscall side.
 844          */
 845         if (READ_ONCE(map->frozen))
 846                 mode &= ~FMODE_CAN_WRITE;
 847         return mode;
 848 }
 849
 850 #ifdef CONFIG_PROC_FS
 851 /* Show the memory usage of a bpf map */
 852 static u64 bpf_map_memory_usage(const struct bpf_map *map)
 853 {
 854         return map->ops->map_mem_usage(map);
 855 }
 856
 857 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 858 {
 859         struct bpf_map *map = filp->private_data;
 860         u32 type = 0, jited = 0;
 861
 862         if (map_type_contains_progs(map)) {
 863                 spin_lock(&map->owner.lock);
 864                 type  = map->owner.type;
 865                 jited = map->owner.jited;
 866                 spin_unlock(&map->owner.lock);
 867         }
 868
 869         seq_printf(m,
 870                    "map_type:\t%u\n"
 871                    "key_size:\t%u\n"
 872                    "value_size:\t%u\n"
 873                    "max_entries:\t%u\n"
 874                    "map_flags:\t%#x\n"
 875                    "map_extra:\t%#llx\n"
 876                    "memlock:\t%llu\n"
 877                    "map_id:\t%u\n"
 878                    "frozen:\t%u\n",
 879                    map->map_type,
 880                    map->key_size,
 881                    map->value_size,
 882                    map->max_entries,
 883                    map->map_flags,
 884                    (unsigned long long)map->map_extra,
 885                    bpf_map_memory_usage(map),
 886                    map->id,
 887                    READ_ONCE(map->frozen));
 888         if (type) {
 889                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 890                 seq_printf(m, "owner_jited:\t%u\n", jited);
 891         }
 892 }
 893 #endif
 894
 895 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 896                               loff_t *ppos)
 897 {
 898         /* We need this handler such that alloc_file() enables
 899          * f_mode with FMODE_CAN_READ.
 900          */
 901         return -EINVAL;
 902 }
 903
 904 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 905                                size_t siz, loff_t *ppos)
 906 {
 907         /* We need this handler such that alloc_file() enables
 908          * f_mode with FMODE_CAN_WRITE.
 909          */
 910         return -EINVAL;
 911 }
 912
 913 /* called for any extra memory-mapped regions (except initial) */
 914 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 915 {
 916         struct bpf_map *map = vma->vm_file->private_data;
 917
 918         if (vma->vm_flags & VM_MAYWRITE)
 919                 bpf_map_write_active_inc(map);
 920 }
 921
 922 /* called for all unmapped memory region (including initial) */
 923 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 924 {
 925         struct bpf_map *map = vma->vm_file->private_data;
 926
 927         if (vma->vm_flags & VM_MAYWRITE)
 928                 bpf_map_write_active_dec(map);
 929 }
 930
 931 static const struct vm_operations_struct bpf_map_default_vmops = {
 932         .open           = bpf_map_mmap_open,
 933         .close          = bpf_map_mmap_close,
 934 };
 935
 936 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 937 {
 938         struct bpf_map *map = filp->private_data;
 939         int err;
 940
 941         if (!map->ops->map_mmap || !IS_ERR_OR_NULL(map->record))
 942                 return -ENOTSUPP;
 943
 944         if (!(vma->vm_flags & VM_SHARED))
 945                 return -EINVAL;
 946
 947         mutex_lock(&map->freeze_mutex);
 948
 949         if (vma->vm_flags & VM_WRITE) {
 950                 if (map->frozen) {
 951                         err = -EPERM;
 952                         goto out;
 953                 }
 954                 /* map is meant to be read-only, so do not allow mapping as
 955                  * writable, because it's possible to leak a writable page
 956                  * reference and allows user-space to still modify it after
 957                  * freezing, while verifier will assume contents do not change
 958                  */
 959                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 960                         err = -EACCES;
 961                         goto out;
 962                 }
 963         }
 964
 965         /* set default open/close callbacks */
 966         vma->vm_ops = &bpf_map_default_vmops;
 967         vma->vm_private_data = map;
 968         vm_flags_clear(vma, VM_MAYEXEC);
 969         if (!(vma->vm_flags & VM_WRITE))
 970                 /* disallow re-mapping with PROT_WRITE */
 971                 vm_flags_clear(vma, VM_MAYWRITE);
 972
 973         err = map->ops->map_mmap(map, vma);
 974         if (err)
 975                 goto out;
 976
 977         if (vma->vm_flags & VM_MAYWRITE)
 978                 bpf_map_write_active_inc(map);
 979 out:
 980         mutex_unlock(&map->freeze_mutex);
 981         return err;
 982 }
 983
 984 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 985 {
 986         struct bpf_map *map = filp->private_data;
 987
 988         if (map->ops->map_poll)
 989                 return map->ops->map_poll(map, filp, pts);
 990
 991         return EPOLLERR;
 992 }
 993
 994 static unsigned long bpf_get_unmapped_area(struct file *filp, unsigned long addr,
 995                                            unsigned long len, unsigned long pgoff,
 996                                            unsigned long flags)
 997 {
 998         struct bpf_map *map = filp->private_data;
 999
1000         if (map->ops->map_get_unmapped_area)
1001                 return map->ops->map_get_unmapped_area(filp, addr, len, pgoff, flags);
1002 #ifdef CONFIG_MMU
1003         return mm_get_unmapped_area(current->mm, filp, addr, len, pgoff, flags);
1004 #else
1005         return addr;
1006 #endif
1007 }
1008
1009 const struct file_operations bpf_map_fops = {
1010 #ifdef CONFIG_PROC_FS
1011         .show_fdinfo    = bpf_map_show_fdinfo,
1012 #endif
1013         .release        = bpf_map_release,
1014         .read           = bpf_dummy_read,
1015         .write          = bpf_dummy_write,
1016         .mmap           = bpf_map_mmap,
1017         .poll           = bpf_map_poll,
1018         .get_unmapped_area = bpf_get_unmapped_area,
1019 };
1020
1021 int bpf_map_new_fd(struct bpf_map *map, int flags)
1022 {
1023         int ret;
1024
1025         ret = security_bpf_map(map, OPEN_FMODE(flags));
1026         if (ret < 0)
1027                 return ret;
1028
1029         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
1030                                 flags | O_CLOEXEC);
1031 }
1032
1033 int bpf_get_file_flag(int flags)
1034 {
1035         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
1036                 return -EINVAL;
1037         if (flags & BPF_F_RDONLY)
1038                 return O_RDONLY;
1039         if (flags & BPF_F_WRONLY)
1040                 return O_WRONLY;
1041         return O_RDWR;
1042 }
1043
1044 /* helper macro to check that unused fields 'union bpf_attr' are zero */
1045 #define CHECK_ATTR(CMD) \
1046         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
1047                    sizeof(attr->CMD##_LAST_FIELD), 0, \
1048                    sizeof(*attr) - \
1049                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
1050                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
1051
1052 /* dst and src must have at least "size" number of bytes.
1053  * Return strlen on success and < 0 on error.
1054  */
1055 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
1056 {
1057         const char *end = src + size;
1058         const char *orig_src = src;
1059
1060         memset(dst, 0, size);
1061         /* Copy all isalnum(), '_' and '.' chars. */
1062         while (src < end && *src) {
1063                 if (!isalnum(*src) &&
1064                     *src != '_' && *src != '.')
1065                         return -EINVAL;
1066                 *dst++ = *src++;
1067         }
1068
1069         /* No '\0' found in "size" number of bytes */
1070         if (src == end)
1071                 return -EINVAL;
1072
1073         return src - orig_src;
1074 }
1075
1076 int map_check_no_btf(const struct bpf_map *map,
1077                      const struct btf *btf,
1078                      const struct btf_type *key_type,
1079                      const struct btf_type *value_type)
1080 {
1081         return -ENOTSUPP;
1082 }
1083
1084 static int map_check_btf(struct bpf_map *map, struct bpf_token *token,
1085                          const struct btf *btf, u32 btf_key_id, u32 btf_value_id)
1086 {
1087         const struct btf_type *key_type, *value_type;
1088         u32 key_size, value_size;
1089         int ret = 0;
1090
1091         /* Some maps allow key to be unspecified. */
1092         if (btf_key_id) {
1093                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
1094                 if (!key_type || key_size != map->key_size)
1095                         return -EINVAL;
1096         } else {
1097                 key_type = btf_type_by_id(btf, 0);
1098                 if (!map->ops->map_check_btf)
1099                         return -EINVAL;
1100         }
1101
1102         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
1103         if (!value_type || value_size != map->value_size)
1104                 return -EINVAL;
1105
1106         map->record = btf_parse_fields(btf, value_type,
1107                                        BPF_SPIN_LOCK | BPF_TIMER | BPF_KPTR | BPF_LIST_HEAD |
1108                                        BPF_RB_ROOT | BPF_REFCOUNT | BPF_WORKQUEUE,
1109                                        map->value_size);
1110         if (!IS_ERR_OR_NULL(map->record)) {
1111                 int i;
1112
1113                 if (!bpf_token_capable(token, CAP_BPF)) {
1114                         ret = -EPERM;
1115                         goto free_map_tab;
1116                 }
1117                 if (map->map_flags & (BPF_F_RDONLY_PROG | BPF_F_WRONLY_PROG)) {
1118                         ret = -EACCES;
1119                         goto free_map_tab;
1120                 }
1121                 for (i = 0; i < sizeof(map->record->field_mask) * 8; i++) {
1122                         switch (map->record->field_mask & (1 << i)) {
1123                         case 0:
1124                                 continue;
1125                         case BPF_SPIN_LOCK:
1126                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1127                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1128                                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
1129                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1130                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1131                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1132                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1133                                         ret = -EOPNOTSUPP;
1134                                         goto free_map_tab;
1135                                 }
1136                                 break;
1137                         case BPF_TIMER:
1138                         case BPF_WORKQUEUE:
1139                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1140                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1141                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1142                                         ret = -EOPNOTSUPP;
1143                                         goto free_map_tab;
1144                                 }
1145                                 break;
1146                         case BPF_KPTR_UNREF:
1147                         case BPF_KPTR_REF:
1148                         case BPF_KPTR_PERCPU:
1149                         case BPF_REFCOUNT:
1150                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1151                                     map->map_type != BPF_MAP_TYPE_PERCPU_HASH &&
1152                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1153                                     map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH &&
1154                                     map->map_type != BPF_MAP_TYPE_ARRAY &&
1155                                     map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY &&
1156                                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
1157                                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
1158                                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE &&
1159                                     map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) {
1160                                         ret = -EOPNOTSUPP;
1161                                         goto free_map_tab;
1162                                 }
1163                                 break;
1164                         case BPF_LIST_HEAD:
1165                         case BPF_RB_ROOT:
1166                                 if (map->map_type != BPF_MAP_TYPE_HASH &&
1167                                     map->map_type != BPF_MAP_TYPE_LRU_HASH &&
1168                                     map->map_type != BPF_MAP_TYPE_ARRAY) {
1169                                         ret = -EOPNOTSUPP;
1170                                         goto free_map_tab;
1171                                 }
1172                                 break;
1173                         default:
1174                                 /* Fail if map_type checks are missing for a field type */
1175                                 ret = -EOPNOTSUPP;
1176                                 goto free_map_tab;
1177                         }
1178                 }
1179         }
1180
1181         ret = btf_check_and_fixup_fields(btf, map->record);
1182         if (ret < 0)
1183                 goto free_map_tab;
1184
1185         if (map->ops->map_check_btf) {
1186                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
1187                 if (ret < 0)
1188                         goto free_map_tab;
1189         }
1190
1191         return ret;
1192 free_map_tab:
1193         bpf_map_free_record(map);
1194         return ret;
1195 }
1196
1197 static bool bpf_net_capable(void)
1198 {
1199         return capable(CAP_NET_ADMIN) || capable(CAP_SYS_ADMIN);
1200 }
1201
1202 #define BPF_MAP_CREATE_LAST_FIELD map_token_fd
1203 /* called via syscall */
1204 static int map_create(union bpf_attr *attr)
1205 {
1206         const struct bpf_map_ops *ops;
1207         struct bpf_token *token = NULL;
1208         int numa_node = bpf_map_attr_numa_node(attr);
1209         u32 map_type = attr->map_type;
1210         struct bpf_map *map;
1211         bool token_flag;
1212         int f_flags;
1213         int err;
1214
1215         err = CHECK_ATTR(BPF_MAP_CREATE);
1216         if (err)
1217                 return -EINVAL;
1218
1219         /* check BPF_F_TOKEN_FD flag, remember if it's set, and then clear it
1220          * to avoid per-map type checks tripping on unknown flag
1221          */
1222         token_flag = attr->map_flags & BPF_F_TOKEN_FD;
1223         attr->map_flags &= ~BPF_F_TOKEN_FD;
1224
1225         if (attr->btf_vmlinux_value_type_id) {
1226                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
1227                     attr->btf_key_type_id || attr->btf_value_type_id)
1228                         return -EINVAL;
1229         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
1230                 return -EINVAL;
1231         }
1232
1233         if (attr->map_type != BPF_MAP_TYPE_BLOOM_FILTER &&
1234             attr->map_type != BPF_MAP_TYPE_ARENA &&
1235             attr->map_extra != 0)
1236                 return -EINVAL;
1237
1238         f_flags = bpf_get_file_flag(attr->map_flags);
1239         if (f_flags < 0)
1240                 return f_flags;
1241
1242         if (numa_node != NUMA_NO_NODE &&
1243             ((unsigned int)numa_node >= nr_node_ids ||
1244              !node_online(numa_node)))
1245                 return -EINVAL;
1246
1247         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
1248         map_type = attr->map_type;
1249         if (map_type >= ARRAY_SIZE(bpf_map_types))
1250                 return -EINVAL;
1251         map_type = array_index_nospec(map_type, ARRAY_SIZE(bpf_map_types));
1252         ops = bpf_map_types[map_type];
1253         if (!ops)
1254                 return -EINVAL;
1255
1256         if (ops->map_alloc_check) {
1257                 err = ops->map_alloc_check(attr);
1258                 if (err)
1259                         return err;
1260         }
1261         if (attr->map_ifindex)
1262                 ops = &bpf_map_offload_ops;
1263         if (!ops->map_mem_usage)
1264                 return -EINVAL;
1265
1266         if (token_flag) {
1267                 token = bpf_token_get_from_fd(attr->map_token_fd);
1268                 if (IS_ERR(token))
1269                         return PTR_ERR(token);
1270
1271                 /* if current token doesn't grant map creation permissions,
1272                  * then we can't use this token, so ignore it and rely on
1273                  * system-wide capabilities checks
1274                  */
1275                 if (!bpf_token_allow_cmd(token, BPF_MAP_CREATE) ||
1276                     !bpf_token_allow_map_type(token, attr->map_type)) {
1277                         bpf_token_put(token);
1278                         token = NULL;
1279                 }
1280         }
1281
1282         err = -EPERM;
1283
1284         /* Intent here is for unprivileged_bpf_disabled to block BPF map
1285          * creation for unprivileged users; other actions depend
1286          * on fd availability and access to bpffs, so are dependent on
1287          * object creation success. Even with unprivileged BPF disabled,
1288          * capability checks are still carried out.
1289          */
1290         if (sysctl_unprivileged_bpf_disabled && !bpf_token_capable(token, CAP_BPF))
1291                 goto put_token;
1292
1293         /* check privileged map type permissions */
1294         switch (map_type) {
1295         case BPF_MAP_TYPE_ARRAY:
1296         case BPF_MAP_TYPE_PERCPU_ARRAY:
1297         case BPF_MAP_TYPE_PROG_ARRAY:
1298         case BPF_MAP_TYPE_PERF_EVENT_ARRAY:
1299         case BPF_MAP_TYPE_CGROUP_ARRAY:
1300         case BPF_MAP_TYPE_ARRAY_OF_MAPS:
1301         case BPF_MAP_TYPE_HASH:
1302         case BPF_MAP_TYPE_PERCPU_HASH:
1303         case BPF_MAP_TYPE_HASH_OF_MAPS:
1304         case BPF_MAP_TYPE_RINGBUF:
1305         case BPF_MAP_TYPE_USER_RINGBUF:
1306         case BPF_MAP_TYPE_CGROUP_STORAGE:
1307         case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE:
1308                 /* unprivileged */
1309                 break;
1310         case BPF_MAP_TYPE_SK_STORAGE:
1311         case BPF_MAP_TYPE_INODE_STORAGE:
1312         case BPF_MAP_TYPE_TASK_STORAGE:
1313         case BPF_MAP_TYPE_CGRP_STORAGE:
1314         case BPF_MAP_TYPE_BLOOM_FILTER:
1315         case BPF_MAP_TYPE_LPM_TRIE:
1316         case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY:
1317         case BPF_MAP_TYPE_STACK_TRACE:
1318         case BPF_MAP_TYPE_QUEUE:
1319         case BPF_MAP_TYPE_STACK:
1320         case BPF_MAP_TYPE_LRU_HASH:
1321         case BPF_MAP_TYPE_LRU_PERCPU_HASH:
1322         case BPF_MAP_TYPE_STRUCT_OPS:
1323         case BPF_MAP_TYPE_CPUMAP:
1324         case BPF_MAP_TYPE_ARENA:
1325                 if (!bpf_token_capable(token, CAP_BPF))
1326                         goto put_token;
1327                 break;
1328         case BPF_MAP_TYPE_SOCKMAP:
1329         case BPF_MAP_TYPE_SOCKHASH:
1330         case BPF_MAP_TYPE_DEVMAP:
1331         case BPF_MAP_TYPE_DEVMAP_HASH:
1332         case BPF_MAP_TYPE_XSKMAP:
1333                 if (!bpf_token_capable(token, CAP_NET_ADMIN))
1334                         goto put_token;
1335                 break;
1336         default:
1337                 WARN(1, "unsupported map type %d", map_type);
1338                 goto put_token;
1339         }
1340
1341         map = ops->map_alloc(attr);
1342         if (IS_ERR(map)) {
1343                 err = PTR_ERR(map);
1344                 goto put_token;
1345         }
1346         map->ops = ops;
1347         map->map_type = map_type;
1348
1349         err = bpf_obj_name_cpy(map->name, attr->map_name,
1350                                sizeof(attr->map_name));
1351         if (err < 0)
1352                 goto free_map;
1353
1354         atomic64_set(&map->refcnt, 1);
1355         atomic64_set(&map->usercnt, 1);
1356         mutex_init(&map->freeze_mutex);
1357         spin_lock_init(&map->owner.lock);
1358
1359         if (attr->btf_key_type_id || attr->btf_value_type_id ||
1360             /* Even the map's value is a kernel's struct,
1361              * the bpf_prog.o must have BTF to begin with
1362              * to figure out the corresponding kernel's
1363              * counter part.  Thus, attr->btf_fd has
1364              * to be valid also.
1365              */
1366             attr->btf_vmlinux_value_type_id) {
1367                 struct btf *btf;
1368
1369                 btf = btf_get_by_fd(attr->btf_fd);
1370                 if (IS_ERR(btf)) {
1371                         err = PTR_ERR(btf);
1372                         goto free_map;
1373                 }
1374                 if (btf_is_kernel(btf)) {
1375                         btf_put(btf);
1376                         err = -EACCES;
1377                         goto free_map;
1378                 }
1379                 map->btf = btf;
1380
1381                 if (attr->btf_value_type_id) {
1382                         err = map_check_btf(map, token, btf, attr->btf_key_type_id,
1383                                             attr->btf_value_type_id);
1384                         if (err)
1385                                 goto free_map;
1386                 }
1387
1388                 map->btf_key_type_id = attr->btf_key_type_id;
1389                 map->btf_value_type_id = attr->btf_value_type_id;
1390                 map->btf_vmlinux_value_type_id =
1391                         attr->btf_vmlinux_value_type_id;
1392         }
1393
1394         err = security_bpf_map_create(map, attr, token);
1395         if (err)
1396                 goto free_map_sec;
1397
1398         err = bpf_map_alloc_id(map);
1399         if (err)
1400                 goto free_map_sec;
1401
1402         bpf_map_save_memcg(map);
1403         bpf_token_put(token);
1404
1405         err = bpf_map_new_fd(map, f_flags);
1406         if (err < 0) {
1407                 /* failed to allocate fd.
1408                  * bpf_map_put_with_uref() is needed because the above
1409                  * bpf_map_alloc_id() has published the map
1410                  * to the userspace and the userspace may
1411                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
1412                  */
1413                 bpf_map_put_with_uref(map);
1414                 return err;
1415         }
1416
1417         return err;
1418
1419 free_map_sec:
1420         security_bpf_map_free(map);
1421 free_map:
1422         bpf_map_free(map);
1423 put_token:
1424         bpf_token_put(token);
1425         return err;
1426 }
1427
1428 void bpf_map_inc(struct bpf_map *map)
1429 {
1430         atomic64_inc(&map->refcnt);
1431 }
1432 EXPORT_SYMBOL_GPL(bpf_map_inc);
1433
1434 void bpf_map_inc_with_uref(struct bpf_map *map)
1435 {
1436         atomic64_inc(&map->refcnt);
1437         atomic64_inc(&map->usercnt);
1438 }
1439 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
1440
1441 struct bpf_map *bpf_map_get(u32 ufd)
1442 {
1443         CLASS(fd, f)(ufd);
1444         struct bpf_map *map = __bpf_map_get(f);
1445
1446         if (!IS_ERR(map))
1447                 bpf_map_inc(map);
1448
1449         return map;
1450 }
1451 EXPORT_SYMBOL(bpf_map_get);
1452
1453 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
1454 {
1455         CLASS(fd, f)(ufd);
1456         struct bpf_map *map = __bpf_map_get(f);
1457
1458         if (!IS_ERR(map))
1459                 bpf_map_inc_with_uref(map);
1460
1461         return map;
1462 }
1463
1464 /* map_idr_lock should have been held or the map should have been
1465  * protected by rcu read lock.
1466  */
1467 struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
1468 {
1469         int refold;
1470
1471         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
1472         if (!refold)
1473                 return ERR_PTR(-ENOENT);
1474         if (uref)
1475                 atomic64_inc(&map->usercnt);
1476
1477         return map;
1478 }
1479
1480 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
1481 {
1482         spin_lock_bh(&map_idr_lock);
1483         map = __bpf_map_inc_not_zero(map, false);
1484         spin_unlock_bh(&map_idr_lock);
1485
1486         return map;
1487 }
1488 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
1489
1490 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
1491 {
1492         return -ENOTSUPP;
1493 }
1494
1495 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
1496 {
1497         if (key_size)
1498                 return vmemdup_user(ukey, key_size);
1499
1500         if (ukey)
1501                 return ERR_PTR(-EINVAL);
1502
1503         return NULL;
1504 }
1505
1506 static void *___bpf_copy_key(bpfptr_t ukey, u64 key_size)
1507 {
1508         if (key_size)
1509                 return kvmemdup_bpfptr(ukey, key_size);
1510
1511         if (!bpfptr_is_null(ukey))
1512                 return ERR_PTR(-EINVAL);
1513
1514         return NULL;
1515 }
1516
1517 /* last field in 'union bpf_attr' used by this command */
1518 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1519
1520 static int map_lookup_elem(union bpf_attr *attr)
1521 {
1522         void __user *ukey = u64_to_user_ptr(attr->key);
1523         void __user *uvalue = u64_to_user_ptr(attr->value);
1524         struct bpf_map *map;
1525         void *key, *value;
1526         u32 value_size;
1527         int err;
1528
1529         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1530                 return -EINVAL;
1531
1532         if (attr->flags & ~BPF_F_LOCK)
1533                 return -EINVAL;
1534
1535         CLASS(fd, f)(attr->map_fd);
1536         map = __bpf_map_get(f);
1537         if (IS_ERR(map))
1538                 return PTR_ERR(map);
1539         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1540                 return -EPERM;
1541
1542         if ((attr->flags & BPF_F_LOCK) &&
1543             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1544                 return -EINVAL;
1545
1546         key = __bpf_copy_key(ukey, map->key_size);
1547         if (IS_ERR(key))
1548                 return PTR_ERR(key);
1549
1550         value_size = bpf_map_value_size(map);
1551
1552         err = -ENOMEM;
1553         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1554         if (!value)
1555                 goto free_key;
1556
1557         if (map->map_type == BPF_MAP_TYPE_BLOOM_FILTER) {
1558                 if (copy_from_user(value, uvalue, value_size))
1559                         err = -EFAULT;
1560                 else
1561                         err = bpf_map_copy_value(map, key, value, attr->flags);
1562                 goto free_value;
1563         }
1564
1565         err = bpf_map_copy_value(map, key, value, attr->flags);
1566         if (err)
1567                 goto free_value;
1568
1569         err = -EFAULT;
1570         if (copy_to_user(uvalue, value, value_size) != 0)
1571                 goto free_value;
1572
1573         err = 0;
1574
1575 free_value:
1576         kvfree(value);
1577 free_key:
1578         kvfree(key);
1579         return err;
1580 }
1581
1582
1583 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1584
1585 static int map_update_elem(union bpf_attr *attr, bpfptr_t uattr)
1586 {
1587         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1588         bpfptr_t uvalue = make_bpfptr(attr->value, uattr.is_kernel);
1589         struct bpf_map *map;
1590         void *key, *value;
1591         u32 value_size;
1592         int err;
1593
1594         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1595                 return -EINVAL;
1596
1597         CLASS(fd, f)(attr->map_fd);
1598         map = __bpf_map_get(f);
1599         if (IS_ERR(map))
1600                 return PTR_ERR(map);
1601         bpf_map_write_active_inc(map);
1602         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1603                 err = -EPERM;
1604                 goto err_put;
1605         }
1606
1607         if ((attr->flags & BPF_F_LOCK) &&
1608             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1609                 err = -EINVAL;
1610                 goto err_put;
1611         }
1612
1613         key = ___bpf_copy_key(ukey, map->key_size);
1614         if (IS_ERR(key)) {
1615                 err = PTR_ERR(key);
1616                 goto err_put;
1617         }
1618
1619         value_size = bpf_map_value_size(map);
1620         value = kvmemdup_bpfptr(uvalue, value_size);
1621         if (IS_ERR(value)) {
1622                 err = PTR_ERR(value);
1623                 goto free_key;
1624         }
1625
1626         err = bpf_map_update_value(map, fd_file(f), key, value, attr->flags);
1627         if (!err)
1628                 maybe_wait_bpf_programs(map);
1629
1630         kvfree(value);
1631 free_key:
1632         kvfree(key);
1633 err_put:
1634         bpf_map_write_active_dec(map);
1635         return err;
1636 }
1637
1638 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1639
1640 static int map_delete_elem(union bpf_attr *attr, bpfptr_t uattr)
1641 {
1642         bpfptr_t ukey = make_bpfptr(attr->key, uattr.is_kernel);
1643         struct bpf_map *map;
1644         void *key;
1645         int err;
1646
1647         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1648                 return -EINVAL;
1649
1650         CLASS(fd, f)(attr->map_fd);
1651         map = __bpf_map_get(f);
1652         if (IS_ERR(map))
1653                 return PTR_ERR(map);
1654         bpf_map_write_active_inc(map);
1655         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1656                 err = -EPERM;
1657                 goto err_put;
1658         }
1659
1660         key = ___bpf_copy_key(ukey, map->key_size);
1661         if (IS_ERR(key)) {
1662                 err = PTR_ERR(key);
1663                 goto err_put;
1664         }
1665
1666         if (bpf_map_is_offloaded(map)) {
1667                 err = bpf_map_offload_delete_elem(map, key);
1668                 goto out;
1669         } else if (IS_FD_PROG_ARRAY(map) ||
1670                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1671                 /* These maps require sleepable context */
1672                 err = map->ops->map_delete_elem(map, key);
1673                 goto out;
1674         }
1675
1676         bpf_disable_instrumentation();
1677         rcu_read_lock();
1678         err = map->ops->map_delete_elem(map, key);
1679         rcu_read_unlock();
1680         bpf_enable_instrumentation();
1681         if (!err)
1682                 maybe_wait_bpf_programs(map);
1683 out:
1684         kvfree(key);
1685 err_put:
1686         bpf_map_write_active_dec(map);
1687         return err;
1688 }
1689
1690 /* last field in 'union bpf_attr' used by this command */
1691 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1692
1693 static int map_get_next_key(union bpf_attr *attr)
1694 {
1695         void __user *ukey = u64_to_user_ptr(attr->key);
1696         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1697         struct bpf_map *map;
1698         void *key, *next_key;
1699         int err;
1700
1701         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1702                 return -EINVAL;
1703
1704         CLASS(fd, f)(attr->map_fd);
1705         map = __bpf_map_get(f);
1706         if (IS_ERR(map))
1707                 return PTR_ERR(map);
1708         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ))
1709                 return -EPERM;
1710
1711         if (ukey) {
1712                 key = __bpf_copy_key(ukey, map->key_size);
1713                 if (IS_ERR(key))
1714                         return PTR_ERR(key);
1715         } else {
1716                 key = NULL;
1717         }
1718
1719         err = -ENOMEM;
1720         next_key = kvmalloc(map->key_size, GFP_USER);
1721         if (!next_key)
1722                 goto free_key;
1723
1724         if (bpf_map_is_offloaded(map)) {
1725                 err = bpf_map_offload_get_next_key(map, key, next_key);
1726                 goto out;
1727         }
1728
1729         rcu_read_lock();
1730         err = map->ops->map_get_next_key(map, key, next_key);
1731         rcu_read_unlock();
1732 out:
1733         if (err)
1734                 goto free_next_key;
1735
1736         err = -EFAULT;
1737         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1738                 goto free_next_key;
1739
1740         err = 0;
1741
1742 free_next_key:
1743         kvfree(next_key);
1744 free_key:
1745         kvfree(key);
1746         return err;
1747 }
1748
1749 int generic_map_delete_batch(struct bpf_map *map,
1750                              const union bpf_attr *attr,
1751                              union bpf_attr __user *uattr)
1752 {
1753         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1754         u32 cp, max_count;
1755         int err = 0;
1756         void *key;
1757
1758         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1759                 return -EINVAL;
1760
1761         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1762             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1763                 return -EINVAL;
1764         }
1765
1766         max_count = attr->batch.count;
1767         if (!max_count)
1768                 return 0;
1769
1770         if (put_user(0, &uattr->batch.count))
1771                 return -EFAULT;
1772
1773         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1774         if (!key)
1775                 return -ENOMEM;
1776
1777         for (cp = 0; cp < max_count; cp++) {
1778                 err = -EFAULT;
1779                 if (copy_from_user(key, keys + cp * map->key_size,
1780                                    map->key_size))
1781                         break;
1782
1783                 if (bpf_map_is_offloaded(map)) {
1784                         err = bpf_map_offload_delete_elem(map, key);
1785                         break;
1786                 }
1787
1788                 bpf_disable_instrumentation();
1789                 rcu_read_lock();
1790                 err = map->ops->map_delete_elem(map, key);
1791                 rcu_read_unlock();
1792                 bpf_enable_instrumentation();
1793                 if (err)
1794                         break;
1795                 cond_resched();
1796         }
1797         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1798                 err = -EFAULT;
1799
1800         kvfree(key);
1801
1802         return err;
1803 }
1804
1805 int generic_map_update_batch(struct bpf_map *map, struct file *map_file,
1806                              const union bpf_attr *attr,
1807                              union bpf_attr __user *uattr)
1808 {
1809         void __user *values = u64_to_user_ptr(attr->batch.values);
1810         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1811         u32 value_size, cp, max_count;
1812         void *key, *value;
1813         int err = 0;
1814
1815         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1816                 return -EINVAL;
1817
1818         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1819             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
1820                 return -EINVAL;
1821         }
1822
1823         value_size = bpf_map_value_size(map);
1824
1825         max_count = attr->batch.count;
1826         if (!max_count)
1827                 return 0;
1828
1829         if (put_user(0, &uattr->batch.count))
1830                 return -EFAULT;
1831
1832         key = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1833         if (!key)
1834                 return -ENOMEM;
1835
1836         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
1837         if (!value) {
1838                 kvfree(key);
1839                 return -ENOMEM;
1840         }
1841
1842         for (cp = 0; cp < max_count; cp++) {
1843                 err = -EFAULT;
1844                 if (copy_from_user(key, keys + cp * map->key_size,
1845                     map->key_size) ||
1846                     copy_from_user(value, values + cp * value_size, value_size))
1847                         break;
1848
1849                 err = bpf_map_update_value(map, map_file, key, value,
1850                                            attr->batch.elem_flags);
1851
1852                 if (err)
1853                         break;
1854                 cond_resched();
1855         }
1856
1857         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1858                 err = -EFAULT;
1859
1860         kvfree(value);
1861         kvfree(key);
1862
1863         return err;
1864 }
1865
1866 #define MAP_LOOKUP_RETRIES 3
1867
1868 int generic_map_lookup_batch(struct bpf_map *map,
1869                                     const union bpf_attr *attr,
1870                                     union bpf_attr __user *uattr)
1871 {
1872         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1873         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1874         void __user *values = u64_to_user_ptr(attr->batch.values);
1875         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1876         void *buf, *buf_prevkey, *prev_key, *key, *value;
1877         int err, retry = MAP_LOOKUP_RETRIES;
1878         u32 value_size, cp, max_count;
1879
1880         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1881                 return -EINVAL;
1882
1883         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1884             !btf_record_has_field(map->record, BPF_SPIN_LOCK))
1885                 return -EINVAL;
1886
1887         value_size = bpf_map_value_size(map);
1888
1889         max_count = attr->batch.count;
1890         if (!max_count)
1891                 return 0;
1892
1893         if (put_user(0, &uattr->batch.count))
1894                 return -EFAULT;
1895
1896         buf_prevkey = kvmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1897         if (!buf_prevkey)
1898                 return -ENOMEM;
1899
1900         buf = kvmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1901         if (!buf) {
1902                 kvfree(buf_prevkey);
1903                 return -ENOMEM;
1904         }
1905
1906         err = -EFAULT;
1907         prev_key = NULL;
1908         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1909                 goto free_buf;
1910         key = buf;
1911         value = key + map->key_size;
1912         if (ubatch)
1913                 prev_key = buf_prevkey;
1914
1915         for (cp = 0; cp < max_count;) {
1916                 rcu_read_lock();
1917                 err = map->ops->map_get_next_key(map, prev_key, key);
1918                 rcu_read_unlock();
1919                 if (err)
1920                         break;
1921                 err = bpf_map_copy_value(map, key, value,
1922                                          attr->batch.elem_flags);
1923
1924                 if (err == -ENOENT) {
1925                         if (retry) {
1926                                 retry--;
1927                                 continue;
1928                         }
1929                         err = -EINTR;
1930                         break;
1931                 }
1932
1933                 if (err)
1934                         goto free_buf;
1935
1936                 if (copy_to_user(keys + cp * map->key_size, key,
1937                                  map->key_size)) {
1938                         err = -EFAULT;
1939                         goto free_buf;
1940                 }
1941                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1942                         err = -EFAULT;
1943                         goto free_buf;
1944                 }
1945
1946                 if (!prev_key)
1947                         prev_key = buf_prevkey;
1948
1949                 swap(prev_key, key);
1950                 retry = MAP_LOOKUP_RETRIES;
1951                 cp++;
1952                 cond_resched();
1953         }
1954
1955         if (err == -EFAULT)
1956                 goto free_buf;
1957
1958         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1959                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1960                 err = -EFAULT;
1961
1962 free_buf:
1963         kvfree(buf_prevkey);
1964         kvfree(buf);
1965         return err;
1966 }
1967
1968 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD flags
1969
1970 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1971 {
1972         void __user *ukey = u64_to_user_ptr(attr->key);
1973         void __user *uvalue = u64_to_user_ptr(attr->value);
1974         struct bpf_map *map;
1975         void *key, *value;
1976         u32 value_size;
1977         int err;
1978
1979         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1980                 return -EINVAL;
1981
1982         if (attr->flags & ~BPF_F_LOCK)
1983                 return -EINVAL;
1984
1985         CLASS(fd, f)(attr->map_fd);
1986         map = __bpf_map_get(f);
1987         if (IS_ERR(map))
1988                 return PTR_ERR(map);
1989         bpf_map_write_active_inc(map);
1990         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1991             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1992                 err = -EPERM;
1993                 goto err_put;
1994         }
1995
1996         if (attr->flags &&
1997             (map->map_type == BPF_MAP_TYPE_QUEUE ||
1998              map->map_type == BPF_MAP_TYPE_STACK)) {
1999                 err = -EINVAL;
2000                 goto err_put;
2001         }
2002
2003         if ((attr->flags & BPF_F_LOCK) &&
2004             !btf_record_has_field(map->record, BPF_SPIN_LOCK)) {
2005                 err = -EINVAL;
2006                 goto err_put;
2007         }
2008
2009         key = __bpf_copy_key(ukey, map->key_size);
2010         if (IS_ERR(key)) {
2011                 err = PTR_ERR(key);
2012                 goto err_put;
2013         }
2014
2015         value_size = bpf_map_value_size(map);
2016
2017         err = -ENOMEM;
2018         value = kvmalloc(value_size, GFP_USER | __GFP_NOWARN);
2019         if (!value)
2020                 goto free_key;
2021
2022         err = -ENOTSUPP;
2023         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
2024             map->map_type == BPF_MAP_TYPE_STACK) {
2025                 err = map->ops->map_pop_elem(map, value);
2026         } else if (map->map_type == BPF_MAP_TYPE_HASH ||
2027                    map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
2028                    map->map_type == BPF_MAP_TYPE_LRU_HASH ||
2029                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
2030                 if (!bpf_map_is_offloaded(map)) {
2031                         bpf_disable_instrumentation();
2032                         rcu_read_lock();
2033                         err = map->ops->map_lookup_and_delete_elem(map, key, value, attr->flags);
2034                         rcu_read_unlock();
2035                         bpf_enable_instrumentation();
2036                 }
2037         }
2038
2039         if (err)
2040                 goto free_value;
2041
2042         if (copy_to_user(uvalue, value, value_size) != 0) {
2043                 err = -EFAULT;
2044                 goto free_value;
2045         }
2046
2047         err = 0;
2048
2049 free_value:
2050         kvfree(value);
2051 free_key:
2052         kvfree(key);
2053 err_put:
2054         bpf_map_write_active_dec(map);
2055         return err;
2056 }
2057
2058 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
2059
2060 static int map_freeze(const union bpf_attr *attr)
2061 {
2062         int err = 0;
2063         struct bpf_map *map;
2064
2065         if (CHECK_ATTR(BPF_MAP_FREEZE))
2066                 return -EINVAL;
2067
2068         CLASS(fd, f)(attr->map_fd);
2069         map = __bpf_map_get(f);
2070         if (IS_ERR(map))
2071                 return PTR_ERR(map);
2072
2073         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS || !IS_ERR_OR_NULL(map->record))
2074                 return -ENOTSUPP;
2075
2076         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE))
2077                 return -EPERM;
2078
2079         mutex_lock(&map->freeze_mutex);
2080         if (bpf_map_write_active(map)) {
2081                 err = -EBUSY;
2082                 goto err_put;
2083         }
2084         if (READ_ONCE(map->frozen)) {
2085                 err = -EBUSY;
2086                 goto err_put;
2087         }
2088
2089         WRITE_ONCE(map->frozen, true);
2090 err_put:
2091         mutex_unlock(&map->freeze_mutex);
2092         return err;
2093 }
2094
2095 static const struct bpf_prog_ops * const bpf_prog_types[] = {
2096 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
2097         [_id] = & _name ## _prog_ops,
2098 #define BPF_MAP_TYPE(_id, _ops)
2099 #define BPF_LINK_TYPE(_id, _name)
2100 #include <linux/bpf_types.h>
2101 #undef BPF_PROG_TYPE
2102 #undef BPF_MAP_TYPE
2103 #undef BPF_LINK_TYPE
2104 };
2105
2106 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
2107 {
2108         const struct bpf_prog_ops *ops;
2109
2110         if (type >= ARRAY_SIZE(bpf_prog_types))
2111                 return -EINVAL;
2112         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
2113         ops = bpf_prog_types[type];
2114         if (!ops)
2115                 return -EINVAL;
2116
2117         if (!bpf_prog_is_offloaded(prog->aux))
2118                 prog->aux->ops = ops;
2119         else
2120                 prog->aux->ops = &bpf_offload_prog_ops;
2121         prog->type = type;
2122         return 0;
2123 }
2124
2125 enum bpf_audit {
2126         BPF_AUDIT_LOAD,
2127         BPF_AUDIT_UNLOAD,
2128         BPF_AUDIT_MAX,
2129 };
2130
2131 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
2132         [BPF_AUDIT_LOAD]   = "LOAD",
2133         [BPF_AUDIT_UNLOAD] = "UNLOAD",
2134 };
2135
2136 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
2137 {
2138         struct audit_context *ctx = NULL;
2139         struct audit_buffer *ab;
2140
2141         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
2142                 return;
2143         if (audit_enabled == AUDIT_OFF)
2144                 return;
2145         if (!in_irq() && !irqs_disabled())
2146                 ctx = audit_context();
2147         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
2148         if (unlikely(!ab))
2149                 return;
2150         audit_log_format(ab, "prog-id=%u op=%s",
2151                          prog->aux->id, bpf_audit_str[op]);
2152         audit_log_end(ab);
2153 }
2154
2155 static int bpf_prog_alloc_id(struct bpf_prog *prog)
2156 {
2157         int id;
2158
2159         idr_preload(GFP_KERNEL);
2160         spin_lock_bh(&prog_idr_lock);
2161         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
2162         if (id > 0)
2163                 prog->aux->id = id;
2164         spin_unlock_bh(&prog_idr_lock);
2165         idr_preload_end();
2166
2167         /* id is in [1, INT_MAX) */
2168         if (WARN_ON_ONCE(!id))
2169                 return -ENOSPC;
2170
2171         return id > 0 ? 0 : id;
2172 }
2173
2174 void bpf_prog_free_id(struct bpf_prog *prog)
2175 {
2176         unsigned long flags;
2177
2178         /* cBPF to eBPF migrations are currently not in the idr store.
2179          * Offloaded programs are removed from the store when their device
2180          * disappears - even if someone grabs an fd to them they are unusable,
2181          * simply waiting for refcnt to drop to be freed.
2182          */
2183         if (!prog->aux->id)
2184                 return;
2185
2186         spin_lock_irqsave(&prog_idr_lock, flags);
2187         idr_remove(&prog_idr, prog->aux->id);
2188         prog->aux->id = 0;
2189         spin_unlock_irqrestore(&prog_idr_lock, flags);
2190 }
2191
2192 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
2193 {
2194         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
2195
2196         kvfree(aux->func_info);
2197         kfree(aux->func_info_aux);
2198         free_uid(aux->user);
2199         security_bpf_prog_free(aux->prog);
2200         bpf_prog_free(aux->prog);
2201 }
2202
2203 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
2204 {
2205         bpf_prog_kallsyms_del_all(prog);
2206         btf_put(prog->aux->btf);
2207         module_put(prog->aux->mod);
2208         kvfree(prog->aux->jited_linfo);
2209         kvfree(prog->aux->linfo);
2210         kfree(prog->aux->kfunc_tab);
2211         if (prog->aux->attach_btf)
2212                 btf_put(prog->aux->attach_btf);
2213
2214         if (deferred) {
2215                 if (prog->sleepable)
2216                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
2217                 else
2218                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
2219         } else {
2220                 __bpf_prog_put_rcu(&prog->aux->rcu);
2221         }
2222 }
2223
2224 static void bpf_prog_put_deferred(struct work_struct *work)
2225 {
2226         struct bpf_prog_aux *aux;
2227         struct bpf_prog *prog;
2228
2229         aux = container_of(work, struct bpf_prog_aux, work);
2230         prog = aux->prog;
2231         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
2232         bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
2233         bpf_prog_free_id(prog);
2234         __bpf_prog_put_noref(prog, true);
2235 }
2236
2237 static void __bpf_prog_put(struct bpf_prog *prog)
2238 {
2239         struct bpf_prog_aux *aux = prog->aux;
2240
2241         if (atomic64_dec_and_test(&aux->refcnt)) {
2242                 if (in_irq() || irqs_disabled()) {
2243                         INIT_WORK(&aux->work, bpf_prog_put_deferred);
2244                         schedule_work(&aux->work);
2245                 } else {
2246                         bpf_prog_put_deferred(&aux->work);
2247                 }
2248         }
2249 }
2250
2251 void bpf_prog_put(struct bpf_prog *prog)
2252 {
2253         __bpf_prog_put(prog);
2254 }
2255 EXPORT_SYMBOL_GPL(bpf_prog_put);
2256
2257 static int bpf_prog_release(struct inode *inode, struct file *filp)
2258 {
2259         struct bpf_prog *prog = filp->private_data;
2260
2261         bpf_prog_put(prog);
2262         return 0;
2263 }
2264
2265 struct bpf_prog_kstats {
2266         u64 nsecs;
2267         u64 cnt;
2268         u64 misses;
2269 };
2270
2271 void notrace bpf_prog_inc_misses_counter(struct bpf_prog *prog)
2272 {
2273         struct bpf_prog_stats *stats;
2274         unsigned int flags;
2275
2276         stats = this_cpu_ptr(prog->stats);
2277         flags = u64_stats_update_begin_irqsave(&stats->syncp);
2278         u64_stats_inc(&stats->misses);
2279         u64_stats_update_end_irqrestore(&stats->syncp, flags);
2280 }
2281
2282 static void bpf_prog_get_stats(const struct bpf_prog *prog,
2283                                struct bpf_prog_kstats *stats)
2284 {
2285         u64 nsecs = 0, cnt = 0, misses = 0;
2286         int cpu;
2287
2288         for_each_possible_cpu(cpu) {
2289                 const struct bpf_prog_stats *st;
2290                 unsigned int start;
2291                 u64 tnsecs, tcnt, tmisses;
2292
2293                 st = per_cpu_ptr(prog->stats, cpu);
2294                 do {
2295                         start = u64_stats_fetch_begin(&st->syncp);
2296                         tnsecs = u64_stats_read(&st->nsecs);
2297                         tcnt = u64_stats_read(&st->cnt);
2298                         tmisses = u64_stats_read(&st->misses);
2299                 } while (u64_stats_fetch_retry(&st->syncp, start));
2300                 nsecs += tnsecs;
2301                 cnt += tcnt;
2302                 misses += tmisses;
2303         }
2304         stats->nsecs = nsecs;
2305         stats->cnt = cnt;
2306         stats->misses = misses;
2307 }
2308
2309 #ifdef CONFIG_PROC_FS
2310 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
2311 {
2312         const struct bpf_prog *prog = filp->private_data;
2313         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2314         struct bpf_prog_kstats stats;
2315
2316         bpf_prog_get_stats(prog, &stats);
2317         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2318         seq_printf(m,
2319                    "prog_type:\t%u\n"
2320                    "prog_jited:\t%u\n"
2321                    "prog_tag:\t%s\n"
2322                    "memlock:\t%llu\n"
2323                    "prog_id:\t%u\n"
2324                    "run_time_ns:\t%llu\n"
2325                    "run_cnt:\t%llu\n"
2326                    "recursion_misses:\t%llu\n"
2327                    "verified_insns:\t%u\n",
2328                    prog->type,
2329                    prog->jited,
2330                    prog_tag,
2331                    prog->pages * 1ULL << PAGE_SHIFT,
2332                    prog->aux->id,
2333                    stats.nsecs,
2334                    stats.cnt,
2335                    stats.misses,
2336                    prog->aux->verified_insns);
2337 }
2338 #endif
2339
2340 const struct file_operations bpf_prog_fops = {
2341 #ifdef CONFIG_PROC_FS
2342         .show_fdinfo    = bpf_prog_show_fdinfo,
2343 #endif
2344         .release        = bpf_prog_release,
2345         .read           = bpf_dummy_read,
2346         .write          = bpf_dummy_write,
2347 };
2348
2349 int bpf_prog_new_fd(struct bpf_prog *prog)
2350 {
2351         int ret;
2352
2353         ret = security_bpf_prog(prog);
2354         if (ret < 0)
2355                 return ret;
2356
2357         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
2358                                 O_RDWR | O_CLOEXEC);
2359 }
2360
2361 void bpf_prog_add(struct bpf_prog *prog, int i)
2362 {
2363         atomic64_add(i, &prog->aux->refcnt);
2364 }
2365 EXPORT_SYMBOL_GPL(bpf_prog_add);
2366
2367 void bpf_prog_sub(struct bpf_prog *prog, int i)
2368 {
2369         /* Only to be used for undoing previous bpf_prog_add() in some
2370          * error path. We still know that another entity in our call
2371          * path holds a reference to the program, thus atomic_sub() can
2372          * be safely used in such cases!
2373          */
2374         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
2375 }
2376 EXPORT_SYMBOL_GPL(bpf_prog_sub);
2377
2378 void bpf_prog_inc(struct bpf_prog *prog)
2379 {
2380         atomic64_inc(&prog->aux->refcnt);
2381 }
2382 EXPORT_SYMBOL_GPL(bpf_prog_inc);
2383
2384 /* prog_idr_lock should have been held */
2385 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
2386 {
2387         int refold;
2388
2389         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
2390
2391         if (!refold)
2392                 return ERR_PTR(-ENOENT);
2393
2394         return prog;
2395 }
2396 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
2397
2398 bool bpf_prog_get_ok(struct bpf_prog *prog,
2399                             enum bpf_prog_type *attach_type, bool attach_drv)
2400 {
2401         /* not an attachment, just a refcount inc, always allow */
2402         if (!attach_type)
2403                 return true;
2404
2405         if (prog->type != *attach_type)
2406                 return false;
2407         if (bpf_prog_is_offloaded(prog->aux) && !attach_drv)
2408                 return false;
2409
2410         return true;
2411 }
2412
2413 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
2414                                        bool attach_drv)
2415 {
2416         CLASS(fd, f)(ufd);
2417         struct bpf_prog *prog;
2418
2419         if (fd_empty(f))
2420                 return ERR_PTR(-EBADF);
2421         if (fd_file(f)->f_op != &bpf_prog_fops)
2422                 return ERR_PTR(-EINVAL);
2423
2424         prog = fd_file(f)->private_data;
2425         if (!bpf_prog_get_ok(prog, attach_type, attach_drv))
2426                 return ERR_PTR(-EINVAL);
2427
2428         bpf_prog_inc(prog);
2429         return prog;
2430 }
2431
2432 struct bpf_prog *bpf_prog_get(u32 ufd)
2433 {
2434         return __bpf_prog_get(ufd, NULL, false);
2435 }
2436
2437 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
2438                                        bool attach_drv)
2439 {
2440         return __bpf_prog_get(ufd, &type, attach_drv);
2441 }
2442 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
2443
2444 /* Initially all BPF programs could be loaded w/o specifying
2445  * expected_attach_type. Later for some of them specifying expected_attach_type
2446  * at load time became required so that program could be validated properly.
2447  * Programs of types that are allowed to be loaded both w/ and w/o (for
2448  * backward compatibility) expected_attach_type, should have the default attach
2449  * type assigned to expected_attach_type for the latter case, so that it can be
2450  * validated later at attach time.
2451  *
2452  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
2453  * prog type requires it but has some attach types that have to be backward
2454  * compatible.
2455  */
2456 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
2457 {
2458         switch (attr->prog_type) {
2459         case BPF_PROG_TYPE_CGROUP_SOCK:
2460                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
2461                  * exist so checking for non-zero is the way to go here.
2462                  */
2463                 if (!attr->expected_attach_type)
2464                         attr->expected_attach_type =
2465                                 BPF_CGROUP_INET_SOCK_CREATE;
2466                 break;
2467         case BPF_PROG_TYPE_SK_REUSEPORT:
2468                 if (!attr->expected_attach_type)
2469                         attr->expected_attach_type =
2470                                 BPF_SK_REUSEPORT_SELECT;
2471                 break;
2472         }
2473 }
2474
2475 static int
2476 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
2477                            enum bpf_attach_type expected_attach_type,
2478                            struct btf *attach_btf, u32 btf_id,
2479                            struct bpf_prog *dst_prog)
2480 {
2481         if (btf_id) {
2482                 if (btf_id > BTF_MAX_TYPE)
2483                         return -EINVAL;
2484
2485                 if (!attach_btf && !dst_prog)
2486                         return -EINVAL;
2487
2488                 switch (prog_type) {
2489                 case BPF_PROG_TYPE_TRACING:
2490                 case BPF_PROG_TYPE_LSM:
2491                 case BPF_PROG_TYPE_STRUCT_OPS:
2492                 case BPF_PROG_TYPE_EXT:
2493                         break;
2494                 default:
2495                         return -EINVAL;
2496                 }
2497         }
2498
2499         if (attach_btf && (!btf_id || dst_prog))
2500                 return -EINVAL;
2501
2502         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
2503             prog_type != BPF_PROG_TYPE_EXT)
2504                 return -EINVAL;
2505
2506         switch (prog_type) {
2507         case BPF_PROG_TYPE_CGROUP_SOCK:
2508                 switch (expected_attach_type) {
2509                 case BPF_CGROUP_INET_SOCK_CREATE:
2510                 case BPF_CGROUP_INET_SOCK_RELEASE:
2511                 case BPF_CGROUP_INET4_POST_BIND:
2512                 case BPF_CGROUP_INET6_POST_BIND:
2513                         return 0;
2514                 default:
2515                         return -EINVAL;
2516                 }
2517         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2518                 switch (expected_attach_type) {
2519                 case BPF_CGROUP_INET4_BIND:
2520                 case BPF_CGROUP_INET6_BIND:
2521                 case BPF_CGROUP_INET4_CONNECT:
2522                 case BPF_CGROUP_INET6_CONNECT:
2523                 case BPF_CGROUP_UNIX_CONNECT:
2524                 case BPF_CGROUP_INET4_GETPEERNAME:
2525                 case BPF_CGROUP_INET6_GETPEERNAME:
2526                 case BPF_CGROUP_UNIX_GETPEERNAME:
2527                 case BPF_CGROUP_INET4_GETSOCKNAME:
2528                 case BPF_CGROUP_INET6_GETSOCKNAME:
2529                 case BPF_CGROUP_UNIX_GETSOCKNAME:
2530                 case BPF_CGROUP_UDP4_SENDMSG:
2531                 case BPF_CGROUP_UDP6_SENDMSG:
2532                 case BPF_CGROUP_UNIX_SENDMSG:
2533                 case BPF_CGROUP_UDP4_RECVMSG:
2534                 case BPF_CGROUP_UDP6_RECVMSG:
2535                 case BPF_CGROUP_UNIX_RECVMSG:
2536                         return 0;
2537                 default:
2538                         return -EINVAL;
2539                 }
2540         case BPF_PROG_TYPE_CGROUP_SKB:
2541                 switch (expected_attach_type) {
2542                 case BPF_CGROUP_INET_INGRESS:
2543                 case BPF_CGROUP_INET_EGRESS:
2544                         return 0;
2545                 default:
2546                         return -EINVAL;
2547                 }
2548         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2549                 switch (expected_attach_type) {
2550                 case BPF_CGROUP_SETSOCKOPT:
2551                 case BPF_CGROUP_GETSOCKOPT:
2552                         return 0;
2553                 default:
2554                         return -EINVAL;
2555                 }
2556         case BPF_PROG_TYPE_SK_LOOKUP:
2557                 if (expected_attach_type == BPF_SK_LOOKUP)
2558                         return 0;
2559                 return -EINVAL;
2560         case BPF_PROG_TYPE_SK_REUSEPORT:
2561                 switch (expected_attach_type) {
2562                 case BPF_SK_REUSEPORT_SELECT:
2563                 case BPF_SK_REUSEPORT_SELECT_OR_MIGRATE:
2564                         return 0;
2565                 default:
2566                         return -EINVAL;
2567                 }
2568         case BPF_PROG_TYPE_NETFILTER:
2569                 if (expected_attach_type == BPF_NETFILTER)
2570                         return 0;
2571                 return -EINVAL;
2572         case BPF_PROG_TYPE_SYSCALL:
2573         case BPF_PROG_TYPE_EXT:
2574                 if (expected_attach_type)
2575                         return -EINVAL;
2576                 fallthrough;
2577         default:
2578                 return 0;
2579         }
2580 }
2581
2582 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2583 {
2584         switch (prog_type) {
2585         case BPF_PROG_TYPE_SCHED_CLS:
2586         case BPF_PROG_TYPE_SCHED_ACT:
2587         case BPF_PROG_TYPE_XDP:
2588         case BPF_PROG_TYPE_LWT_IN:
2589         case BPF_PROG_TYPE_LWT_OUT:
2590         case BPF_PROG_TYPE_LWT_XMIT:
2591         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2592         case BPF_PROG_TYPE_SK_SKB:
2593         case BPF_PROG_TYPE_SK_MSG:
2594         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2595         case BPF_PROG_TYPE_CGROUP_DEVICE:
2596         case BPF_PROG_TYPE_CGROUP_SOCK:
2597         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2598         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2599         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2600         case BPF_PROG_TYPE_SOCK_OPS:
2601         case BPF_PROG_TYPE_EXT: /* extends any prog */
2602         case BPF_PROG_TYPE_NETFILTER:
2603                 return true;
2604         case BPF_PROG_TYPE_CGROUP_SKB:
2605                 /* always unpriv */
2606         case BPF_PROG_TYPE_SK_REUSEPORT:
2607                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2608         default:
2609                 return false;
2610         }
2611 }
2612
2613 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2614 {
2615         switch (prog_type) {
2616         case BPF_PROG_TYPE_KPROBE:
2617         case BPF_PROG_TYPE_TRACEPOINT:
2618         case BPF_PROG_TYPE_PERF_EVENT:
2619         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2620         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2621         case BPF_PROG_TYPE_TRACING:
2622         case BPF_PROG_TYPE_LSM:
2623         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2624         case BPF_PROG_TYPE_EXT: /* extends any prog */
2625                 return true;
2626         default:
2627                 return false;
2628         }
2629 }
2630
2631 /* last field in 'union bpf_attr' used by this command */
2632 #define BPF_PROG_LOAD_LAST_FIELD prog_token_fd
2633
2634 static int bpf_prog_load(union bpf_attr *attr, bpfptr_t uattr, u32 uattr_size)
2635 {
2636         enum bpf_prog_type type = attr->prog_type;
2637         struct bpf_prog *prog, *dst_prog = NULL;
2638         struct btf *attach_btf = NULL;
2639         struct bpf_token *token = NULL;
2640         bool bpf_cap;
2641         int err;
2642         char license[128];
2643
2644         if (CHECK_ATTR(BPF_PROG_LOAD))
2645                 return -EINVAL;
2646
2647         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2648                                  BPF_F_ANY_ALIGNMENT |
2649                                  BPF_F_TEST_STATE_FREQ |
2650                                  BPF_F_SLEEPABLE |
2651                                  BPF_F_TEST_RND_HI32 |
2652                                  BPF_F_XDP_HAS_FRAGS |
2653                                  BPF_F_XDP_DEV_BOUND_ONLY |
2654                                  BPF_F_TEST_REG_INVARIANTS |
2655                                  BPF_F_TOKEN_FD))
2656                 return -EINVAL;
2657
2658         bpf_prog_load_fixup_attach_type(attr);
2659
2660         if (attr->prog_flags & BPF_F_TOKEN_FD) {
2661                 token = bpf_token_get_from_fd(attr->prog_token_fd);
2662                 if (IS_ERR(token))
2663                         return PTR_ERR(token);
2664                 /* if current token doesn't grant prog loading permissions,
2665                  * then we can't use this token, so ignore it and rely on
2666                  * system-wide capabilities checks
2667                  */
2668                 if (!bpf_token_allow_cmd(token, BPF_PROG_LOAD) ||
2669                     !bpf_token_allow_prog_type(token, attr->prog_type,
2670                                                attr->expected_attach_type)) {
2671                         bpf_token_put(token);
2672                         token = NULL;
2673                 }
2674         }
2675
2676         bpf_cap = bpf_token_capable(token, CAP_BPF);
2677         err = -EPERM;
2678
2679         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2680             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2681             !bpf_cap)
2682                 goto put_token;
2683
2684         /* Intent here is for unprivileged_bpf_disabled to block BPF program
2685          * creation for unprivileged users; other actions depend
2686          * on fd availability and access to bpffs, so are dependent on
2687          * object creation success. Even with unprivileged BPF disabled,
2688          * capability checks are still carried out for these
2689          * and other operations.
2690          */
2691         if (sysctl_unprivileged_bpf_disabled && !bpf_cap)
2692                 goto put_token;
2693
2694         if (attr->insn_cnt == 0 ||
2695             attr->insn_cnt > (bpf_cap ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS)) {
2696                 err = -E2BIG;
2697                 goto put_token;
2698         }
2699         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2700             type != BPF_PROG_TYPE_CGROUP_SKB &&
2701             !bpf_cap)
2702                 goto put_token;
2703
2704         if (is_net_admin_prog_type(type) && !bpf_token_capable(token, CAP_NET_ADMIN))
2705                 goto put_token;
2706         if (is_perfmon_prog_type(type) && !bpf_token_capable(token, CAP_PERFMON))
2707                 goto put_token;
2708
2709         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2710          * or btf, we need to check which one it is
2711          */
2712         if (attr->attach_prog_fd) {
2713                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2714                 if (IS_ERR(dst_prog)) {
2715                         dst_prog = NULL;
2716                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2717                         if (IS_ERR(attach_btf)) {
2718                                 err = -EINVAL;
2719                                 goto put_token;
2720                         }
2721                         if (!btf_is_kernel(attach_btf)) {
2722                                 /* attaching through specifying bpf_prog's BTF
2723                                  * objects directly might be supported eventually
2724                                  */
2725                                 btf_put(attach_btf);
2726                                 err = -ENOTSUPP;
2727                                 goto put_token;
2728                         }
2729                 }
2730         } else if (attr->attach_btf_id) {
2731                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2732                 attach_btf = bpf_get_btf_vmlinux();
2733                 if (IS_ERR(attach_btf)) {
2734                         err = PTR_ERR(attach_btf);
2735                         goto put_token;
2736                 }
2737                 if (!attach_btf) {
2738                         err = -EINVAL;
2739                         goto put_token;
2740                 }
2741                 btf_get(attach_btf);
2742         }
2743
2744         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2745                                        attach_btf, attr->attach_btf_id,
2746                                        dst_prog)) {
2747                 if (dst_prog)
2748                         bpf_prog_put(dst_prog);
2749                 if (attach_btf)
2750                         btf_put(attach_btf);
2751                 err = -EINVAL;
2752                 goto put_token;
2753         }
2754
2755         /* plain bpf_prog allocation */
2756         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2757         if (!prog) {
2758                 if (dst_prog)
2759                         bpf_prog_put(dst_prog);
2760                 if (attach_btf)
2761                         btf_put(attach_btf);
2762                 err = -EINVAL;
2763                 goto put_token;
2764         }
2765
2766         prog->expected_attach_type = attr->expected_attach_type;
2767         prog->sleepable = !!(attr->prog_flags & BPF_F_SLEEPABLE);
2768         prog->aux->attach_btf = attach_btf;
2769         prog->aux->attach_btf_id = attr->attach_btf_id;
2770         prog->aux->dst_prog = dst_prog;
2771         prog->aux->dev_bound = !!attr->prog_ifindex;
2772         prog->aux->xdp_has_frags = attr->prog_flags & BPF_F_XDP_HAS_FRAGS;
2773
2774         /* move token into prog->aux, reuse taken refcnt */
2775         prog->aux->token = token;
2776         token = NULL;
2777
2778         prog->aux->user = get_current_user();
2779         prog->len = attr->insn_cnt;
2780
2781         err = -EFAULT;
2782         if (copy_from_bpfptr(prog->insns,
2783                              make_bpfptr(attr->insns, uattr.is_kernel),
2784                              bpf_prog_insn_size(prog)) != 0)
2785                 goto free_prog;
2786         /* copy eBPF program license from user space */
2787         if (strncpy_from_bpfptr(license,
2788                                 make_bpfptr(attr->license, uattr.is_kernel),
2789                                 sizeof(license) - 1) < 0)
2790                 goto free_prog;
2791         license[sizeof(license) - 1] = 0;
2792
2793         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2794         prog->gpl_compatible = license_is_gpl_compatible(license) ? 1 : 0;
2795
2796         prog->orig_prog = NULL;
2797         prog->jited = 0;
2798
2799         atomic64_set(&prog->aux->refcnt, 1);
2800
2801         if (bpf_prog_is_dev_bound(prog->aux)) {
2802                 err = bpf_prog_dev_bound_init(prog, attr);
2803                 if (err)
2804                         goto free_prog;
2805         }
2806
2807         if (type == BPF_PROG_TYPE_EXT && dst_prog &&
2808             bpf_prog_is_dev_bound(dst_prog->aux)) {
2809                 err = bpf_prog_dev_bound_inherit(prog, dst_prog);
2810                 if (err)
2811                         goto free_prog;
2812         }
2813
2814         /*
2815          * Bookkeeping for managing the program attachment chain.
2816          *
2817          * It might be tempting to set attach_tracing_prog flag at the attachment
2818          * time, but this will not prevent from loading bunch of tracing prog
2819          * first, then attach them one to another.
2820          *
2821          * The flag attach_tracing_prog is set for the whole program lifecycle, and
2822          * doesn't have to be cleared in bpf_tracing_link_release, since tracing
2823          * programs cannot change attachment target.
2824          */
2825         if (type == BPF_PROG_TYPE_TRACING && dst_prog &&
2826             dst_prog->type == BPF_PROG_TYPE_TRACING) {
2827                 prog->aux->attach_tracing_prog = true;
2828         }
2829
2830         /* find program type: socket_filter vs tracing_filter */
2831         err = find_prog_type(type, prog);
2832         if (err < 0)
2833                 goto free_prog;
2834
2835         prog->aux->load_time = ktime_get_boottime_ns();
2836         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2837                                sizeof(attr->prog_name));
2838         if (err < 0)
2839                 goto free_prog;
2840
2841         err = security_bpf_prog_load(prog, attr, token);
2842         if (err)
2843                 goto free_prog_sec;
2844
2845         /* run eBPF verifier */
2846         err = bpf_check(&prog, attr, uattr, uattr_size);
2847         if (err < 0)
2848                 goto free_used_maps;
2849
2850         prog = bpf_prog_select_runtime(prog, &err);
2851         if (err < 0)
2852                 goto free_used_maps;
2853
2854         err = bpf_prog_alloc_id(prog);
2855         if (err)
2856                 goto free_used_maps;
2857
2858         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2859          * effectively publicly exposed. However, retrieving via
2860          * bpf_prog_get_fd_by_id() will take another reference,
2861          * therefore it cannot be gone underneath us.
2862          *
2863          * Only for the time /after/ successful bpf_prog_new_fd()
2864          * and before returning to userspace, we might just hold
2865          * one reference and any parallel close on that fd could
2866          * rip everything out. Hence, below notifications must
2867          * happen before bpf_prog_new_fd().
2868          *
2869          * Also, any failure handling from this point onwards must
2870          * be using bpf_prog_put() given the program is exposed.
2871          */
2872         bpf_prog_kallsyms_add(prog);
2873         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2874         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2875
2876         err = bpf_prog_new_fd(prog);
2877         if (err < 0)
2878                 bpf_prog_put(prog);
2879         return err;
2880
2881 free_used_maps:
2882         /* In case we have subprogs, we need to wait for a grace
2883          * period before we can tear down JIT memory since symbols
2884          * are already exposed under kallsyms.
2885          */
2886         __bpf_prog_put_noref(prog, prog->aux->real_func_cnt);
2887         return err;
2888
2889 free_prog_sec:
2890         security_bpf_prog_free(prog);
2891 free_prog:
2892         free_uid(prog->aux->user);
2893         if (prog->aux->attach_btf)
2894                 btf_put(prog->aux->attach_btf);
2895         bpf_prog_free(prog);
2896 put_token:
2897         bpf_token_put(token);
2898         return err;
2899 }
2900
2901 #define BPF_OBJ_LAST_FIELD path_fd
2902
2903 static int bpf_obj_pin(const union bpf_attr *attr)
2904 {
2905         int path_fd;
2906
2907         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags & ~BPF_F_PATH_FD)
2908                 return -EINVAL;
2909
2910         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2911         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2912                 return -EINVAL;
2913
2914         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2915         return bpf_obj_pin_user(attr->bpf_fd, path_fd,
2916                                 u64_to_user_ptr(attr->pathname));
2917 }
2918
2919 static int bpf_obj_get(const union bpf_attr *attr)
2920 {
2921         int path_fd;
2922
2923         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2924             attr->file_flags & ~(BPF_OBJ_FLAG_MASK | BPF_F_PATH_FD))
2925                 return -EINVAL;
2926
2927         /* path_fd has to be accompanied by BPF_F_PATH_FD flag */
2928         if (!(attr->file_flags & BPF_F_PATH_FD) && attr->path_fd)
2929                 return -EINVAL;
2930
2931         path_fd = attr->file_flags & BPF_F_PATH_FD ? attr->path_fd : AT_FDCWD;
2932         return bpf_obj_get_user(path_fd, u64_to_user_ptr(attr->pathname),
2933                                 attr->file_flags);
2934 }
2935
2936 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2937                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2938 {
2939         WARN_ON(ops->dealloc && ops->dealloc_deferred);
2940         atomic64_set(&link->refcnt, 1);
2941         link->type = type;
2942         link->id = 0;
2943         link->ops = ops;
2944         link->prog = prog;
2945 }
2946
2947 static void bpf_link_free_id(int id)
2948 {
2949         if (!id)
2950                 return;
2951
2952         spin_lock_bh(&link_idr_lock);
2953         idr_remove(&link_idr, id);
2954         spin_unlock_bh(&link_idr_lock);
2955 }
2956
2957 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2958  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2959  * anon_inode's release() call. This helper marks bpf_link as
2960  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2961  * is not decremented, it's the responsibility of a calling code that failed
2962  * to complete bpf_link initialization.
2963  * This helper eventually calls link's dealloc callback, but does not call
2964  * link's release callback.
2965  */
2966 void bpf_link_cleanup(struct bpf_link_primer *primer)
2967 {
2968         primer->link->prog = NULL;
2969         bpf_link_free_id(primer->id);
2970         fput(primer->file);
2971         put_unused_fd(primer->fd);
2972 }
2973
2974 void bpf_link_inc(struct bpf_link *link)
2975 {
2976         atomic64_inc(&link->refcnt);
2977 }
2978
2979 static void bpf_link_defer_dealloc_rcu_gp(struct rcu_head *rcu)
2980 {
2981         struct bpf_link *link = container_of(rcu, struct bpf_link, rcu);
2982
2983         /* free bpf_link and its containing memory */
2984         link->ops->dealloc_deferred(link);
2985 }
2986
2987 static void bpf_link_defer_dealloc_mult_rcu_gp(struct rcu_head *rcu)
2988 {
2989         if (rcu_trace_implies_rcu_gp())
2990                 bpf_link_defer_dealloc_rcu_gp(rcu);
2991         else
2992                 call_rcu(rcu, bpf_link_defer_dealloc_rcu_gp);
2993 }
2994
2995 /* bpf_link_free is guaranteed to be called from process context */
2996 static void bpf_link_free(struct bpf_link *link)
2997 {
2998         const struct bpf_link_ops *ops = link->ops;
2999         bool sleepable = false;
3000
3001         bpf_link_free_id(link->id);
3002         if (link->prog) {
3003                 sleepable = link->prog->sleepable;
3004                 /* detach BPF program, clean up used resources */
3005                 ops->release(link);
3006                 bpf_prog_put(link->prog);
3007         }
3008         if (ops->dealloc_deferred) {
3009                 /* schedule BPF link deallocation; if underlying BPF program
3010                  * is sleepable, we need to first wait for RCU tasks trace
3011                  * sync, then go through "classic" RCU grace period
3012                  */
3013                 if (sleepable)
3014                         call_rcu_tasks_trace(&link->rcu, bpf_link_defer_dealloc_mult_rcu_gp);
3015                 else
3016                         call_rcu(&link->rcu, bpf_link_defer_dealloc_rcu_gp);
3017         } else if (ops->dealloc)
3018                 ops->dealloc(link);
3019 }
3020
3021 static void bpf_link_put_deferred(struct work_struct *work)
3022 {
3023         struct bpf_link *link = container_of(work, struct bpf_link, work);
3024
3025         bpf_link_free(link);
3026 }
3027
3028 /* bpf_link_put might be called from atomic context. It needs to be called
3029  * from sleepable context in order to acquire sleeping locks during the process.
3030  */
3031 void bpf_link_put(struct bpf_link *link)
3032 {
3033         if (!atomic64_dec_and_test(&link->refcnt))
3034                 return;
3035
3036         INIT_WORK(&link->work, bpf_link_put_deferred);
3037         schedule_work(&link->work);
3038 }
3039 EXPORT_SYMBOL(bpf_link_put);
3040
3041 static void bpf_link_put_direct(struct bpf_link *link)
3042 {
3043         if (!atomic64_dec_and_test(&link->refcnt))
3044                 return;
3045         bpf_link_free(link);
3046 }
3047
3048 static int bpf_link_release(struct inode *inode, struct file *filp)
3049 {
3050         struct bpf_link *link = filp->private_data;
3051
3052         bpf_link_put_direct(link);
3053         return 0;
3054 }
3055
3056 #ifdef CONFIG_PROC_FS
3057 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
3058 #define BPF_MAP_TYPE(_id, _ops)
3059 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
3060 static const char *bpf_link_type_strs[] = {
3061         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
3062 #include <linux/bpf_types.h>
3063 };
3064 #undef BPF_PROG_TYPE
3065 #undef BPF_MAP_TYPE
3066 #undef BPF_LINK_TYPE
3067
3068 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
3069 {
3070         const struct bpf_link *link = filp->private_data;
3071         const struct bpf_prog *prog = link->prog;
3072         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
3073
3074         seq_printf(m,
3075                    "link_type:\t%s\n"
3076                    "link_id:\t%u\n",
3077                    bpf_link_type_strs[link->type],
3078                    link->id);
3079         if (prog) {
3080                 bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
3081                 seq_printf(m,
3082                            "prog_tag:\t%s\n"
3083                            "prog_id:\t%u\n",
3084                            prog_tag,
3085                            prog->aux->id);
3086         }
3087         if (link->ops->show_fdinfo)
3088                 link->ops->show_fdinfo(link, m);
3089 }
3090 #endif
3091
3092 static __poll_t bpf_link_poll(struct file *file, struct poll_table_struct *pts)
3093 {
3094         struct bpf_link *link = file->private_data;
3095
3096         return link->ops->poll(file, pts);
3097 }
3098
3099 static const struct file_operations bpf_link_fops = {
3100 #ifdef CONFIG_PROC_FS
3101         .show_fdinfo    = bpf_link_show_fdinfo,
3102 #endif
3103         .release        = bpf_link_release,
3104         .read           = bpf_dummy_read,
3105         .write          = bpf_dummy_write,
3106 };
3107
3108 static const struct file_operations bpf_link_fops_poll = {
3109 #ifdef CONFIG_PROC_FS
3110         .show_fdinfo    = bpf_link_show_fdinfo,
3111 #endif
3112         .release        = bpf_link_release,
3113         .read           = bpf_dummy_read,
3114         .write          = bpf_dummy_write,
3115         .poll           = bpf_link_poll,
3116 };
3117
3118 static int bpf_link_alloc_id(struct bpf_link *link)
3119 {
3120         int id;
3121
3122         idr_preload(GFP_KERNEL);
3123         spin_lock_bh(&link_idr_lock);
3124         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
3125         spin_unlock_bh(&link_idr_lock);
3126         idr_preload_end();
3127
3128         return id;
3129 }
3130
3131 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
3132  * reserving unused FD and allocating ID from link_idr. This is to be paired
3133  * with bpf_link_settle() to install FD and ID and expose bpf_link to
3134  * user-space, if bpf_link is successfully attached. If not, bpf_link and
3135  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
3136  * transient state is passed around in struct bpf_link_primer.
3137  * This is preferred way to create and initialize bpf_link, especially when
3138  * there are complicated and expensive operations in between creating bpf_link
3139  * itself and attaching it to BPF hook. By using bpf_link_prime() and
3140  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
3141  * expensive (and potentially failing) roll back operations in a rare case
3142  * that file, FD, or ID can't be allocated.
3143  */
3144 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
3145 {
3146         struct file *file;
3147         int fd, id;
3148
3149         fd = get_unused_fd_flags(O_CLOEXEC);
3150         if (fd < 0)
3151                 return fd;
3152
3153
3154         id = bpf_link_alloc_id(link);
3155         if (id < 0) {
3156                 put_unused_fd(fd);
3157                 return id;
3158         }
3159
3160         file = anon_inode_getfile("bpf_link",
3161                                   link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3162                                   link, O_CLOEXEC);
3163         if (IS_ERR(file)) {
3164                 bpf_link_free_id(id);
3165                 put_unused_fd(fd);
3166                 return PTR_ERR(file);
3167         }
3168
3169         primer->link = link;
3170         primer->file = file;
3171         primer->fd = fd;
3172         primer->id = id;
3173         return 0;
3174 }
3175
3176 int bpf_link_settle(struct bpf_link_primer *primer)
3177 {
3178         /* make bpf_link fetchable by ID */
3179         spin_lock_bh(&link_idr_lock);
3180         primer->link->id = primer->id;
3181         spin_unlock_bh(&link_idr_lock);
3182         /* make bpf_link fetchable by FD */
3183         fd_install(primer->fd, primer->file);
3184         /* pass through installed FD */
3185         return primer->fd;
3186 }
3187
3188 int bpf_link_new_fd(struct bpf_link *link)
3189 {
3190         return anon_inode_getfd("bpf-link",
3191                                 link->ops->poll ? &bpf_link_fops_poll : &bpf_link_fops,
3192                                 link, O_CLOEXEC);
3193 }
3194
3195 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
3196 {
3197         CLASS(fd, f)(ufd);
3198         struct bpf_link *link;
3199
3200         if (fd_empty(f))
3201                 return ERR_PTR(-EBADF);
3202         if (fd_file(f)->f_op != &bpf_link_fops && fd_file(f)->f_op != &bpf_link_fops_poll)
3203                 return ERR_PTR(-EINVAL);
3204
3205         link = fd_file(f)->private_data;
3206         bpf_link_inc(link);
3207         return link;
3208 }
3209 EXPORT_SYMBOL(bpf_link_get_from_fd);
3210
3211 static void bpf_tracing_link_release(struct bpf_link *link)
3212 {
3213         struct bpf_tracing_link *tr_link =
3214                 container_of(link, struct bpf_tracing_link, link.link);
3215
3216         WARN_ON_ONCE(bpf_trampoline_unlink_prog(&tr_link->link,
3217                                                 tr_link->trampoline));
3218
3219         bpf_trampoline_put(tr_link->trampoline);
3220
3221         /* tgt_prog is NULL if target is a kernel function */
3222         if (tr_link->tgt_prog)
3223                 bpf_prog_put(tr_link->tgt_prog);
3224 }
3225
3226 static void bpf_tracing_link_dealloc(struct bpf_link *link)
3227 {
3228         struct bpf_tracing_link *tr_link =
3229                 container_of(link, struct bpf_tracing_link, link.link);
3230
3231         kfree(tr_link);
3232 }
3233
3234 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
3235                                          struct seq_file *seq)
3236 {
3237         struct bpf_tracing_link *tr_link =
3238                 container_of(link, struct bpf_tracing_link, link.link);
3239         u32 target_btf_id, target_obj_id;
3240
3241         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3242                                   &target_obj_id, &target_btf_id);
3243         seq_printf(seq,
3244                    "attach_type:\t%d\n"
3245                    "target_obj_id:\t%u\n"
3246                    "target_btf_id:\t%u\n",
3247                    tr_link->attach_type,
3248                    target_obj_id,
3249                    target_btf_id);
3250 }
3251
3252 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
3253                                            struct bpf_link_info *info)
3254 {
3255         struct bpf_tracing_link *tr_link =
3256                 container_of(link, struct bpf_tracing_link, link.link);
3257
3258         info->tracing.attach_type = tr_link->attach_type;
3259         bpf_trampoline_unpack_key(tr_link->trampoline->key,
3260                                   &info->tracing.target_obj_id,
3261                                   &info->tracing.target_btf_id);
3262
3263         return 0;
3264 }
3265
3266 static const struct bpf_link_ops bpf_tracing_link_lops = {
3267         .release = bpf_tracing_link_release,
3268         .dealloc = bpf_tracing_link_dealloc,
3269         .show_fdinfo = bpf_tracing_link_show_fdinfo,
3270         .fill_link_info = bpf_tracing_link_fill_link_info,
3271 };
3272
3273 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
3274                                    int tgt_prog_fd,
3275                                    u32 btf_id,
3276                                    u64 bpf_cookie)
3277 {
3278         struct bpf_link_primer link_primer;
3279         struct bpf_prog *tgt_prog = NULL;
3280         struct bpf_trampoline *tr = NULL;
3281         struct bpf_tracing_link *link;
3282         u64 key = 0;
3283         int err;
3284
3285         switch (prog->type) {
3286         case BPF_PROG_TYPE_TRACING:
3287                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
3288                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
3289                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
3290                         err = -EINVAL;
3291                         goto out_put_prog;
3292                 }
3293                 break;
3294         case BPF_PROG_TYPE_EXT:
3295                 if (prog->expected_attach_type != 0) {
3296                         err = -EINVAL;
3297                         goto out_put_prog;
3298                 }
3299                 break;
3300         case BPF_PROG_TYPE_LSM:
3301                 if (prog->expected_attach_type != BPF_LSM_MAC) {
3302                         err = -EINVAL;
3303                         goto out_put_prog;
3304                 }
3305                 break;
3306         default:
3307                 err = -EINVAL;
3308                 goto out_put_prog;
3309         }
3310
3311         if (!!tgt_prog_fd != !!btf_id) {
3312                 err = -EINVAL;
3313                 goto out_put_prog;
3314         }
3315
3316         if (tgt_prog_fd) {
3317                 /*
3318                  * For now we only allow new targets for BPF_PROG_TYPE_EXT. If this
3319                  * part would be changed to implement the same for
3320                  * BPF_PROG_TYPE_TRACING, do not forget to update the way how
3321                  * attach_tracing_prog flag is set.
3322                  */
3323                 if (prog->type != BPF_PROG_TYPE_EXT) {
3324                         err = -EINVAL;
3325                         goto out_put_prog;
3326                 }
3327
3328                 tgt_prog = bpf_prog_get(tgt_prog_fd);
3329                 if (IS_ERR(tgt_prog)) {
3330                         err = PTR_ERR(tgt_prog);
3331                         tgt_prog = NULL;
3332                         goto out_put_prog;
3333                 }
3334
3335                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
3336         }
3337
3338         link = kzalloc(sizeof(*link), GFP_USER);
3339         if (!link) {
3340                 err = -ENOMEM;
3341                 goto out_put_prog;
3342         }
3343         bpf_link_init(&link->link.link, BPF_LINK_TYPE_TRACING,
3344                       &bpf_tracing_link_lops, prog);
3345         link->attach_type = prog->expected_attach_type;
3346         link->link.cookie = bpf_cookie;
3347
3348         mutex_lock(&prog->aux->dst_mutex);
3349
3350         /* There are a few possible cases here:
3351          *
3352          * - if prog->aux->dst_trampoline is set, the program was just loaded
3353          *   and not yet attached to anything, so we can use the values stored
3354          *   in prog->aux
3355          *
3356          * - if prog->aux->dst_trampoline is NULL, the program has already been
3357          *   attached to a target and its initial target was cleared (below)
3358          *
3359          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
3360          *   target_btf_id using the link_create API.
3361          *
3362          * - if tgt_prog == NULL when this function was called using the old
3363          *   raw_tracepoint_open API, and we need a target from prog->aux
3364          *
3365          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
3366          *   was detached and is going for re-attachment.
3367          *
3368          * - if prog->aux->dst_trampoline is NULL and tgt_prog and prog->aux->attach_btf
3369          *   are NULL, then program was already attached and user did not provide
3370          *   tgt_prog_fd so we have no way to find out or create trampoline
3371          */
3372         if (!prog->aux->dst_trampoline && !tgt_prog) {
3373                 /*
3374                  * Allow re-attach for TRACING and LSM programs. If it's
3375                  * currently linked, bpf_trampoline_link_prog will fail.
3376                  * EXT programs need to specify tgt_prog_fd, so they
3377                  * re-attach in separate code path.
3378                  */
3379                 if (prog->type != BPF_PROG_TYPE_TRACING &&
3380                     prog->type != BPF_PROG_TYPE_LSM) {
3381                         err = -EINVAL;
3382                         goto out_unlock;
3383                 }
3384                 /* We can allow re-attach only if we have valid attach_btf. */
3385                 if (!prog->aux->attach_btf) {
3386                         err = -EINVAL;
3387                         goto out_unlock;
3388                 }
3389                 btf_id = prog->aux->attach_btf_id;
3390                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
3391         }
3392
3393         if (!prog->aux->dst_trampoline ||
3394             (key && key != prog->aux->dst_trampoline->key)) {
3395                 /* If there is no saved target, or the specified target is
3396                  * different from the destination specified at load time, we
3397                  * need a new trampoline and a check for compatibility
3398                  */
3399                 struct bpf_attach_target_info tgt_info = {};
3400
3401                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
3402                                               &tgt_info);
3403                 if (err)
3404                         goto out_unlock;
3405
3406                 if (tgt_info.tgt_mod) {
3407                         module_put(prog->aux->mod);
3408                         prog->aux->mod = tgt_info.tgt_mod;
3409                 }
3410
3411                 tr = bpf_trampoline_get(key, &tgt_info);
3412                 if (!tr) {
3413                         err = -ENOMEM;
3414                         goto out_unlock;
3415                 }
3416         } else {
3417                 /* The caller didn't specify a target, or the target was the
3418                  * same as the destination supplied during program load. This
3419                  * means we can reuse the trampoline and reference from program
3420                  * load time, and there is no need to allocate a new one. This
3421                  * can only happen once for any program, as the saved values in
3422                  * prog->aux are cleared below.
3423                  */
3424                 tr = prog->aux->dst_trampoline;
3425                 tgt_prog = prog->aux->dst_prog;
3426         }
3427
3428         err = bpf_link_prime(&link->link.link, &link_primer);
3429         if (err)
3430                 goto out_unlock;
3431
3432         err = bpf_trampoline_link_prog(&link->link, tr);
3433         if (err) {
3434                 bpf_link_cleanup(&link_primer);
3435                 link = NULL;
3436                 goto out_unlock;
3437         }
3438
3439         link->tgt_prog = tgt_prog;
3440         link->trampoline = tr;
3441
3442         /* Always clear the trampoline and target prog from prog->aux to make
3443          * sure the original attach destination is not kept alive after a
3444          * program is (re-)attached to another target.
3445          */
3446         if (prog->aux->dst_prog &&
3447             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
3448                 /* got extra prog ref from syscall, or attaching to different prog */
3449                 bpf_prog_put(prog->aux->dst_prog);
3450         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
3451                 /* we allocated a new trampoline, so free the old one */
3452                 bpf_trampoline_put(prog->aux->dst_trampoline);
3453
3454         prog->aux->dst_prog = NULL;
3455         prog->aux->dst_trampoline = NULL;
3456         mutex_unlock(&prog->aux->dst_mutex);
3457
3458         return bpf_link_settle(&link_primer);
3459 out_unlock:
3460         if (tr && tr != prog->aux->dst_trampoline)
3461                 bpf_trampoline_put(tr);
3462         mutex_unlock(&prog->aux->dst_mutex);
3463         kfree(link);
3464 out_put_prog:
3465         if (tgt_prog_fd && tgt_prog)
3466                 bpf_prog_put(tgt_prog);
3467         return err;
3468 }
3469
3470 static void bpf_raw_tp_link_release(struct bpf_link *link)
3471 {
3472         struct bpf_raw_tp_link *raw_tp =
3473                 container_of(link, struct bpf_raw_tp_link, link);
3474
3475         bpf_probe_unregister(raw_tp->btp, raw_tp);
3476         bpf_put_raw_tracepoint(raw_tp->btp);
3477 }
3478
3479 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
3480 {
3481         struct bpf_raw_tp_link *raw_tp =
3482                 container_of(link, struct bpf_raw_tp_link, link);
3483
3484         kfree(raw_tp);
3485 }
3486
3487 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
3488                                         struct seq_file *seq)
3489 {
3490         struct bpf_raw_tp_link *raw_tp_link =
3491                 container_of(link, struct bpf_raw_tp_link, link);
3492
3493         seq_printf(seq,
3494                    "tp_name:\t%s\n",
3495                    raw_tp_link->btp->tp->name);
3496 }
3497
3498 static int bpf_copy_to_user(char __user *ubuf, const char *buf, u32 ulen,
3499                             u32 len)
3500 {
3501         if (ulen >= len + 1) {
3502                 if (copy_to_user(ubuf, buf, len + 1))
3503                         return -EFAULT;
3504         } else {
3505                 char zero = '\0';
3506
3507                 if (copy_to_user(ubuf, buf, ulen - 1))
3508                         return -EFAULT;
3509                 if (put_user(zero, ubuf + ulen - 1))
3510                         return -EFAULT;
3511                 return -ENOSPC;
3512         }
3513
3514         return 0;
3515 }
3516
3517 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
3518                                           struct bpf_link_info *info)
3519 {
3520         struct bpf_raw_tp_link *raw_tp_link =
3521                 container_of(link, struct bpf_raw_tp_link, link);
3522         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
3523         const char *tp_name = raw_tp_link->btp->tp->name;
3524         u32 ulen = info->raw_tracepoint.tp_name_len;
3525         size_t tp_len = strlen(tp_name);
3526
3527         if (!ulen ^ !ubuf)
3528                 return -EINVAL;
3529
3530         info->raw_tracepoint.tp_name_len = tp_len + 1;
3531
3532         if (!ubuf)
3533                 return 0;
3534
3535         return bpf_copy_to_user(ubuf, tp_name, ulen, tp_len);
3536 }
3537
3538 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
3539         .release = bpf_raw_tp_link_release,
3540         .dealloc_deferred = bpf_raw_tp_link_dealloc,
3541         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
3542         .fill_link_info = bpf_raw_tp_link_fill_link_info,
3543 };
3544
3545 #ifdef CONFIG_PERF_EVENTS
3546 struct bpf_perf_link {
3547         struct bpf_link link;
3548         struct file *perf_file;
3549 };
3550
3551 static void bpf_perf_link_release(struct bpf_link *link)
3552 {
3553         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3554         struct perf_event *event = perf_link->perf_file->private_data;
3555
3556         perf_event_free_bpf_prog(event);
3557         fput(perf_link->perf_file);
3558 }
3559
3560 static void bpf_perf_link_dealloc(struct bpf_link *link)
3561 {
3562         struct bpf_perf_link *perf_link = container_of(link, struct bpf_perf_link, link);
3563
3564         kfree(perf_link);
3565 }
3566
3567 static int bpf_perf_link_fill_common(const struct perf_event *event,
3568                                      char __user *uname, u32 ulen,
3569                                      u64 *probe_offset, u64 *probe_addr,
3570                                      u32 *fd_type, unsigned long *missed)
3571 {
3572         const char *buf;
3573         u32 prog_id;
3574         size_t len;
3575         int err;
3576
3577         if (!ulen ^ !uname)
3578                 return -EINVAL;
3579
3580         err = bpf_get_perf_event_info(event, &prog_id, fd_type, &buf,
3581                                       probe_offset, probe_addr, missed);
3582         if (err)
3583                 return err;
3584         if (!uname)
3585                 return 0;
3586         if (buf) {
3587                 len = strlen(buf);
3588                 err = bpf_copy_to_user(uname, buf, ulen, len);
3589                 if (err)
3590                         return err;
3591         } else {
3592                 char zero = '\0';
3593
3594                 if (put_user(zero, uname))
3595                         return -EFAULT;
3596         }
3597         return 0;
3598 }
3599
3600 #ifdef CONFIG_KPROBE_EVENTS
3601 static int bpf_perf_link_fill_kprobe(const struct perf_event *event,
3602                                      struct bpf_link_info *info)
3603 {
3604         unsigned long missed;
3605         char __user *uname;
3606         u64 addr, offset;
3607         u32 ulen, type;
3608         int err;
3609
3610         uname = u64_to_user_ptr(info->perf_event.kprobe.func_name);
3611         ulen = info->perf_event.kprobe.name_len;
3612         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3613                                         &type, &missed);
3614         if (err)
3615                 return err;
3616         if (type == BPF_FD_TYPE_KRETPROBE)
3617                 info->perf_event.type = BPF_PERF_EVENT_KRETPROBE;
3618         else
3619                 info->perf_event.type = BPF_PERF_EVENT_KPROBE;
3620
3621         info->perf_event.kprobe.offset = offset;
3622         info->perf_event.kprobe.missed = missed;
3623         if (!kallsyms_show_value(current_cred()))
3624                 addr = 0;
3625         info->perf_event.kprobe.addr = addr;
3626         info->perf_event.kprobe.cookie = event->bpf_cookie;
3627         return 0;
3628 }
3629 #endif
3630
3631 #ifdef CONFIG_UPROBE_EVENTS
3632 static int bpf_perf_link_fill_uprobe(const struct perf_event *event,
3633                                      struct bpf_link_info *info)
3634 {
3635         char __user *uname;
3636         u64 addr, offset;
3637         u32 ulen, type;
3638         int err;
3639
3640         uname = u64_to_user_ptr(info->perf_event.uprobe.file_name);
3641         ulen = info->perf_event.uprobe.name_len;
3642         err = bpf_perf_link_fill_common(event, uname, ulen, &offset, &addr,
3643                                         &type, NULL);
3644         if (err)
3645                 return err;
3646
3647         if (type == BPF_FD_TYPE_URETPROBE)
3648                 info->perf_event.type = BPF_PERF_EVENT_URETPROBE;
3649         else
3650                 info->perf_event.type = BPF_PERF_EVENT_UPROBE;
3651         info->perf_event.uprobe.offset = offset;
3652         info->perf_event.uprobe.cookie = event->bpf_cookie;
3653         return 0;
3654 }
3655 #endif
3656
3657 static int bpf_perf_link_fill_probe(const struct perf_event *event,
3658                                     struct bpf_link_info *info)
3659 {
3660 #ifdef CONFIG_KPROBE_EVENTS
3661         if (event->tp_event->flags & TRACE_EVENT_FL_KPROBE)
3662                 return bpf_perf_link_fill_kprobe(event, info);
3663 #endif
3664 #ifdef CONFIG_UPROBE_EVENTS
3665         if (event->tp_event->flags & TRACE_EVENT_FL_UPROBE)
3666                 return bpf_perf_link_fill_uprobe(event, info);
3667 #endif
3668         return -EOPNOTSUPP;
3669 }
3670
3671 static int bpf_perf_link_fill_tracepoint(const struct perf_event *event,
3672                                          struct bpf_link_info *info)
3673 {
3674         char __user *uname;
3675         u32 ulen;
3676
3677         uname = u64_to_user_ptr(info->perf_event.tracepoint.tp_name);
3678         ulen = info->perf_event.tracepoint.name_len;
3679         info->perf_event.type = BPF_PERF_EVENT_TRACEPOINT;
3680         info->perf_event.tracepoint.cookie = event->bpf_cookie;
3681         return bpf_perf_link_fill_common(event, uname, ulen, NULL, NULL, NULL, NULL);
3682 }
3683
3684 static int bpf_perf_link_fill_perf_event(const struct perf_event *event,
3685                                          struct bpf_link_info *info)
3686 {
3687         info->perf_event.event.type = event->attr.type;
3688         info->perf_event.event.config = event->attr.config;
3689         info->perf_event.event.cookie = event->bpf_cookie;
3690         info->perf_event.type = BPF_PERF_EVENT_EVENT;
3691         return 0;
3692 }
3693
3694 static int bpf_perf_link_fill_link_info(const struct bpf_link *link,
3695                                         struct bpf_link_info *info)
3696 {
3697         struct bpf_perf_link *perf_link;
3698         const struct perf_event *event;
3699
3700         perf_link = container_of(link, struct bpf_perf_link, link);
3701         event = perf_get_event(perf_link->perf_file);
3702         if (IS_ERR(event))
3703                 return PTR_ERR(event);
3704
3705         switch (event->prog->type) {
3706         case BPF_PROG_TYPE_PERF_EVENT:
3707                 return bpf_perf_link_fill_perf_event(event, info);
3708         case BPF_PROG_TYPE_TRACEPOINT:
3709                 return bpf_perf_link_fill_tracepoint(event, info);
3710         case BPF_PROG_TYPE_KPROBE:
3711                 return bpf_perf_link_fill_probe(event, info);
3712         default:
3713                 return -EOPNOTSUPP;
3714         }
3715 }
3716
3717 static const struct bpf_link_ops bpf_perf_link_lops = {
3718         .release = bpf_perf_link_release,
3719         .dealloc = bpf_perf_link_dealloc,
3720         .fill_link_info = bpf_perf_link_fill_link_info,
3721 };
3722
3723 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3724 {
3725         struct bpf_link_primer link_primer;
3726         struct bpf_perf_link *link;
3727         struct perf_event *event;
3728         struct file *perf_file;
3729         int err;
3730
3731         if (attr->link_create.flags)
3732                 return -EINVAL;
3733
3734         perf_file = perf_event_get(attr->link_create.target_fd);
3735         if (IS_ERR(perf_file))
3736                 return PTR_ERR(perf_file);
3737
3738         link = kzalloc(sizeof(*link), GFP_USER);
3739         if (!link) {
3740                 err = -ENOMEM;
3741                 goto out_put_file;
3742         }
3743         bpf_link_init(&link->link, BPF_LINK_TYPE_PERF_EVENT, &bpf_perf_link_lops, prog);
3744         link->perf_file = perf_file;
3745
3746         err = bpf_link_prime(&link->link, &link_primer);
3747         if (err) {
3748                 kfree(link);
3749                 goto out_put_file;
3750         }
3751
3752         event = perf_file->private_data;
3753         err = perf_event_set_bpf_prog(event, prog, attr->link_create.perf_event.bpf_cookie);
3754         if (err) {
3755                 bpf_link_cleanup(&link_primer);
3756                 goto out_put_file;
3757         }
3758         /* perf_event_set_bpf_prog() doesn't take its own refcnt on prog */
3759         bpf_prog_inc(prog);
3760
3761         return bpf_link_settle(&link_primer);
3762
3763 out_put_file:
3764         fput(perf_file);
3765         return err;
3766 }
3767 #else
3768 static int bpf_perf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3769 {
3770         return -EOPNOTSUPP;
3771 }
3772 #endif /* CONFIG_PERF_EVENTS */
3773
3774 static int bpf_raw_tp_link_attach(struct bpf_prog *prog,
3775                                   const char __user *user_tp_name, u64 cookie)
3776 {
3777         struct bpf_link_primer link_primer;
3778         struct bpf_raw_tp_link *link;
3779         struct bpf_raw_event_map *btp;
3780         const char *tp_name;
3781         char buf[128];
3782         int err;
3783
3784         switch (prog->type) {
3785         case BPF_PROG_TYPE_TRACING:
3786         case BPF_PROG_TYPE_EXT:
3787         case BPF_PROG_TYPE_LSM:
3788                 if (user_tp_name)
3789                         /* The attach point for this category of programs
3790                          * should be specified via btf_id during program load.
3791                          */
3792                         return -EINVAL;
3793                 if (prog->type == BPF_PROG_TYPE_TRACING &&
3794                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
3795                         tp_name = prog->aux->attach_func_name;
3796                         break;
3797                 }
3798                 return bpf_tracing_prog_attach(prog, 0, 0, 0);
3799         case BPF_PROG_TYPE_RAW_TRACEPOINT:
3800         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
3801                 if (strncpy_from_user(buf, user_tp_name, sizeof(buf) - 1) < 0)
3802                         return -EFAULT;
3803                 buf[sizeof(buf) - 1] = 0;
3804                 tp_name = buf;
3805                 break;
3806         default:
3807                 return -EINVAL;
3808         }
3809
3810         btp = bpf_get_raw_tracepoint(tp_name);
3811         if (!btp)
3812                 return -ENOENT;
3813
3814         link = kzalloc(sizeof(*link), GFP_USER);
3815         if (!link) {
3816                 err = -ENOMEM;
3817                 goto out_put_btp;
3818         }
3819         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
3820                       &bpf_raw_tp_link_lops, prog);
3821         link->btp = btp;
3822         link->cookie = cookie;
3823
3824         err = bpf_link_prime(&link->link, &link_primer);
3825         if (err) {
3826                 kfree(link);
3827                 goto out_put_btp;
3828         }
3829
3830         err = bpf_probe_register(link->btp, link);
3831         if (err) {
3832                 bpf_link_cleanup(&link_primer);
3833                 goto out_put_btp;
3834         }
3835
3836         return bpf_link_settle(&link_primer);
3837
3838 out_put_btp:
3839         bpf_put_raw_tracepoint(btp);
3840         return err;
3841 }
3842
3843 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.cookie
3844
3845 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
3846 {
3847         struct bpf_prog *prog;
3848         void __user *tp_name;
3849         __u64 cookie;
3850         int fd;
3851
3852         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
3853                 return -EINVAL;
3854
3855         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
3856         if (IS_ERR(prog))
3857                 return PTR_ERR(prog);
3858
3859         tp_name = u64_to_user_ptr(attr->raw_tracepoint.name);
3860         cookie = attr->raw_tracepoint.cookie;
3861         fd = bpf_raw_tp_link_attach(prog, tp_name, cookie);
3862         if (fd < 0)
3863                 bpf_prog_put(prog);
3864         return fd;
3865 }
3866
3867 static enum bpf_prog_type
3868 attach_type_to_prog_type(enum bpf_attach_type attach_type)
3869 {
3870         switch (attach_type) {
3871         case BPF_CGROUP_INET_INGRESS:
3872         case BPF_CGROUP_INET_EGRESS:
3873                 return BPF_PROG_TYPE_CGROUP_SKB;
3874         case BPF_CGROUP_INET_SOCK_CREATE:
3875         case BPF_CGROUP_INET_SOCK_RELEASE:
3876         case BPF_CGROUP_INET4_POST_BIND:
3877         case BPF_CGROUP_INET6_POST_BIND:
3878                 return BPF_PROG_TYPE_CGROUP_SOCK;
3879         case BPF_CGROUP_INET4_BIND:
3880         case BPF_CGROUP_INET6_BIND:
3881         case BPF_CGROUP_INET4_CONNECT:
3882         case BPF_CGROUP_INET6_CONNECT:
3883         case BPF_CGROUP_UNIX_CONNECT:
3884         case BPF_CGROUP_INET4_GETPEERNAME:
3885         case BPF_CGROUP_INET6_GETPEERNAME:
3886         case BPF_CGROUP_UNIX_GETPEERNAME:
3887         case BPF_CGROUP_INET4_GETSOCKNAME:
3888         case BPF_CGROUP_INET6_GETSOCKNAME:
3889         case BPF_CGROUP_UNIX_GETSOCKNAME:
3890         case BPF_CGROUP_UDP4_SENDMSG:
3891         case BPF_CGROUP_UDP6_SENDMSG:
3892         case BPF_CGROUP_UNIX_SENDMSG:
3893         case BPF_CGROUP_UDP4_RECVMSG:
3894         case BPF_CGROUP_UDP6_RECVMSG:
3895         case BPF_CGROUP_UNIX_RECVMSG:
3896                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
3897         case BPF_CGROUP_SOCK_OPS:
3898                 return BPF_PROG_TYPE_SOCK_OPS;
3899         case BPF_CGROUP_DEVICE:
3900                 return BPF_PROG_TYPE_CGROUP_DEVICE;
3901         case BPF_SK_MSG_VERDICT:
3902                 return BPF_PROG_TYPE_SK_MSG;
3903         case BPF_SK_SKB_STREAM_PARSER:
3904         case BPF_SK_SKB_STREAM_VERDICT:
3905         case BPF_SK_SKB_VERDICT:
3906                 return BPF_PROG_TYPE_SK_SKB;
3907         case BPF_LIRC_MODE2:
3908                 return BPF_PROG_TYPE_LIRC_MODE2;
3909         case BPF_FLOW_DISSECTOR:
3910                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
3911         case BPF_CGROUP_SYSCTL:
3912                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
3913         case BPF_CGROUP_GETSOCKOPT:
3914         case BPF_CGROUP_SETSOCKOPT:
3915                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
3916         case BPF_TRACE_ITER:
3917         case BPF_TRACE_RAW_TP:
3918         case BPF_TRACE_FENTRY:
3919         case BPF_TRACE_FEXIT:
3920         case BPF_MODIFY_RETURN:
3921                 return BPF_PROG_TYPE_TRACING;
3922         case BPF_LSM_MAC:
3923                 return BPF_PROG_TYPE_LSM;
3924         case BPF_SK_LOOKUP:
3925                 return BPF_PROG_TYPE_SK_LOOKUP;
3926         case BPF_XDP:
3927                 return BPF_PROG_TYPE_XDP;
3928         case BPF_LSM_CGROUP:
3929                 return BPF_PROG_TYPE_LSM;
3930         case BPF_TCX_INGRESS:
3931         case BPF_TCX_EGRESS:
3932         case BPF_NETKIT_PRIMARY:
3933         case BPF_NETKIT_PEER:
3934                 return BPF_PROG_TYPE_SCHED_CLS;
3935         default:
3936                 return BPF_PROG_TYPE_UNSPEC;
3937         }
3938 }
3939
3940 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
3941                                              enum bpf_attach_type attach_type)
3942 {
3943         enum bpf_prog_type ptype;
3944
3945         switch (prog->type) {
3946         case BPF_PROG_TYPE_CGROUP_SOCK:
3947         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3948         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3949         case BPF_PROG_TYPE_SK_LOOKUP:
3950                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
3951         case BPF_PROG_TYPE_CGROUP_SKB:
3952                 if (!bpf_token_capable(prog->aux->token, CAP_NET_ADMIN))
3953                         /* cg-skb progs can be loaded by unpriv user.
3954                          * check permissions at attach time.
3955                          */
3956                         return -EPERM;
3957
3958                 ptype = attach_type_to_prog_type(attach_type);
3959                 if (prog->type != ptype)
3960                         return -EINVAL;
3961
3962                 return prog->enforce_expected_attach_type &&
3963                         prog->expected_attach_type != attach_type ?
3964                         -EINVAL : 0;
3965         case BPF_PROG_TYPE_EXT:
3966                 return 0;
3967         case BPF_PROG_TYPE_NETFILTER:
3968                 if (attach_type != BPF_NETFILTER)
3969                         return -EINVAL;
3970                 return 0;
3971         case BPF_PROG_TYPE_PERF_EVENT:
3972         case BPF_PROG_TYPE_TRACEPOINT:
3973                 if (attach_type != BPF_PERF_EVENT)
3974                         return -EINVAL;
3975                 return 0;
3976         case BPF_PROG_TYPE_KPROBE:
3977                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI &&
3978                     attach_type != BPF_TRACE_KPROBE_MULTI)
3979                         return -EINVAL;
3980                 if (prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION &&
3981                     attach_type != BPF_TRACE_KPROBE_SESSION)
3982                         return -EINVAL;
3983                 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI &&
3984                     attach_type != BPF_TRACE_UPROBE_MULTI)
3985                         return -EINVAL;
3986                 if (attach_type != BPF_PERF_EVENT &&
3987                     attach_type != BPF_TRACE_KPROBE_MULTI &&
3988                     attach_type != BPF_TRACE_KPROBE_SESSION &&
3989                     attach_type != BPF_TRACE_UPROBE_MULTI)
3990                         return -EINVAL;
3991                 return 0;
3992         case BPF_PROG_TYPE_SCHED_CLS:
3993                 if (attach_type != BPF_TCX_INGRESS &&
3994                     attach_type != BPF_TCX_EGRESS &&
3995                     attach_type != BPF_NETKIT_PRIMARY &&
3996                     attach_type != BPF_NETKIT_PEER)
3997                         return -EINVAL;
3998                 return 0;
3999         default:
4000                 ptype = attach_type_to_prog_type(attach_type);
4001                 if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type)
4002                         return -EINVAL;
4003                 return 0;
4004         }
4005 }
4006
4007 #define BPF_PROG_ATTACH_LAST_FIELD expected_revision
4008
4009 #define BPF_F_ATTACH_MASK_BASE  \
4010         (BPF_F_ALLOW_OVERRIDE | \
4011          BPF_F_ALLOW_MULTI |    \
4012          BPF_F_REPLACE)
4013
4014 #define BPF_F_ATTACH_MASK_MPROG \
4015         (BPF_F_REPLACE |        \
4016          BPF_F_BEFORE |         \
4017          BPF_F_AFTER |          \
4018          BPF_F_ID |             \
4019          BPF_F_LINK)
4020
4021 static int bpf_prog_attach(const union bpf_attr *attr)
4022 {
4023         enum bpf_prog_type ptype;
4024         struct bpf_prog *prog;
4025         int ret;
4026
4027         if (CHECK_ATTR(BPF_PROG_ATTACH))
4028                 return -EINVAL;
4029
4030         ptype = attach_type_to_prog_type(attr->attach_type);
4031         if (ptype == BPF_PROG_TYPE_UNSPEC)
4032                 return -EINVAL;
4033         if (bpf_mprog_supported(ptype)) {
4034                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4035                         return -EINVAL;
4036         } else {
4037                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_BASE)
4038                         return -EINVAL;
4039                 if (attr->relative_fd ||
4040                     attr->expected_revision)
4041                         return -EINVAL;
4042         }
4043
4044         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4045         if (IS_ERR(prog))
4046                 return PTR_ERR(prog);
4047
4048         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
4049                 bpf_prog_put(prog);
4050                 return -EINVAL;
4051         }
4052
4053         switch (ptype) {
4054         case BPF_PROG_TYPE_SK_SKB:
4055         case BPF_PROG_TYPE_SK_MSG:
4056                 ret = sock_map_get_from_fd(attr, prog);
4057                 break;
4058         case BPF_PROG_TYPE_LIRC_MODE2:
4059                 ret = lirc_prog_attach(attr, prog);
4060                 break;
4061         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4062                 ret = netns_bpf_prog_attach(attr, prog);
4063                 break;
4064         case BPF_PROG_TYPE_CGROUP_DEVICE:
4065         case BPF_PROG_TYPE_CGROUP_SKB:
4066         case BPF_PROG_TYPE_CGROUP_SOCK:
4067         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4068         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4069         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4070         case BPF_PROG_TYPE_SOCK_OPS:
4071         case BPF_PROG_TYPE_LSM:
4072                 if (ptype == BPF_PROG_TYPE_LSM &&
4073                     prog->expected_attach_type != BPF_LSM_CGROUP)
4074                         ret = -EINVAL;
4075                 else
4076                         ret = cgroup_bpf_prog_attach(attr, ptype, prog);
4077                 break;
4078         case BPF_PROG_TYPE_SCHED_CLS:
4079                 if (attr->attach_type == BPF_TCX_INGRESS ||
4080                     attr->attach_type == BPF_TCX_EGRESS)
4081                         ret = tcx_prog_attach(attr, prog);
4082                 else
4083                         ret = netkit_prog_attach(attr, prog);
4084                 break;
4085         default:
4086                 ret = -EINVAL;
4087         }
4088
4089         if (ret)
4090                 bpf_prog_put(prog);
4091         return ret;
4092 }
4093
4094 #define BPF_PROG_DETACH_LAST_FIELD expected_revision
4095
4096 static int bpf_prog_detach(const union bpf_attr *attr)
4097 {
4098         struct bpf_prog *prog = NULL;
4099         enum bpf_prog_type ptype;
4100         int ret;
4101
4102         if (CHECK_ATTR(BPF_PROG_DETACH))
4103                 return -EINVAL;
4104
4105         ptype = attach_type_to_prog_type(attr->attach_type);
4106         if (bpf_mprog_supported(ptype)) {
4107                 if (ptype == BPF_PROG_TYPE_UNSPEC)
4108                         return -EINVAL;
4109                 if (attr->attach_flags & ~BPF_F_ATTACH_MASK_MPROG)
4110                         return -EINVAL;
4111                 if (attr->attach_bpf_fd) {
4112                         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
4113                         if (IS_ERR(prog))
4114                                 return PTR_ERR(prog);
4115                 }
4116         } else if (attr->attach_flags ||
4117                    attr->relative_fd ||
4118                    attr->expected_revision) {
4119                 return -EINVAL;
4120         }
4121
4122         switch (ptype) {
4123         case BPF_PROG_TYPE_SK_MSG:
4124         case BPF_PROG_TYPE_SK_SKB:
4125                 ret = sock_map_prog_detach(attr, ptype);
4126                 break;
4127         case BPF_PROG_TYPE_LIRC_MODE2:
4128                 ret = lirc_prog_detach(attr);
4129                 break;
4130         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4131                 ret = netns_bpf_prog_detach(attr, ptype);
4132                 break;
4133         case BPF_PROG_TYPE_CGROUP_DEVICE:
4134         case BPF_PROG_TYPE_CGROUP_SKB:
4135         case BPF_PROG_TYPE_CGROUP_SOCK:
4136         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4137         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4138         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4139         case BPF_PROG_TYPE_SOCK_OPS:
4140         case BPF_PROG_TYPE_LSM:
4141                 ret = cgroup_bpf_prog_detach(attr, ptype);
4142                 break;
4143         case BPF_PROG_TYPE_SCHED_CLS:
4144                 if (attr->attach_type == BPF_TCX_INGRESS ||
4145                     attr->attach_type == BPF_TCX_EGRESS)
4146                         ret = tcx_prog_detach(attr, prog);
4147                 else
4148                         ret = netkit_prog_detach(attr, prog);
4149                 break;
4150         default:
4151                 ret = -EINVAL;
4152         }
4153
4154         if (prog)
4155                 bpf_prog_put(prog);
4156         return ret;
4157 }
4158
4159 #define BPF_PROG_QUERY_LAST_FIELD query.revision
4160
4161 static int bpf_prog_query(const union bpf_attr *attr,
4162                           union bpf_attr __user *uattr)
4163 {
4164         if (!bpf_net_capable())
4165                 return -EPERM;
4166         if (CHECK_ATTR(BPF_PROG_QUERY))
4167                 return -EINVAL;
4168         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
4169                 return -EINVAL;
4170
4171         switch (attr->query.attach_type) {
4172         case BPF_CGROUP_INET_INGRESS:
4173         case BPF_CGROUP_INET_EGRESS:
4174         case BPF_CGROUP_INET_SOCK_CREATE:
4175         case BPF_CGROUP_INET_SOCK_RELEASE:
4176         case BPF_CGROUP_INET4_BIND:
4177         case BPF_CGROUP_INET6_BIND:
4178         case BPF_CGROUP_INET4_POST_BIND:
4179         case BPF_CGROUP_INET6_POST_BIND:
4180         case BPF_CGROUP_INET4_CONNECT:
4181         case BPF_CGROUP_INET6_CONNECT:
4182         case BPF_CGROUP_UNIX_CONNECT:
4183         case BPF_CGROUP_INET4_GETPEERNAME:
4184         case BPF_CGROUP_INET6_GETPEERNAME:
4185         case BPF_CGROUP_UNIX_GETPEERNAME:
4186         case BPF_CGROUP_INET4_GETSOCKNAME:
4187         case BPF_CGROUP_INET6_GETSOCKNAME:
4188         case BPF_CGROUP_UNIX_GETSOCKNAME:
4189         case BPF_CGROUP_UDP4_SENDMSG:
4190         case BPF_CGROUP_UDP6_SENDMSG:
4191         case BPF_CGROUP_UNIX_SENDMSG:
4192         case BPF_CGROUP_UDP4_RECVMSG:
4193         case BPF_CGROUP_UDP6_RECVMSG:
4194         case BPF_CGROUP_UNIX_RECVMSG:
4195         case BPF_CGROUP_SOCK_OPS:
4196         case BPF_CGROUP_DEVICE:
4197         case BPF_CGROUP_SYSCTL:
4198         case BPF_CGROUP_GETSOCKOPT:
4199         case BPF_CGROUP_SETSOCKOPT:
4200         case BPF_LSM_CGROUP:
4201                 return cgroup_bpf_prog_query(attr, uattr);
4202         case BPF_LIRC_MODE2:
4203                 return lirc_prog_query(attr, uattr);
4204         case BPF_FLOW_DISSECTOR:
4205         case BPF_SK_LOOKUP:
4206                 return netns_bpf_prog_query(attr, uattr);
4207         case BPF_SK_SKB_STREAM_PARSER:
4208         case BPF_SK_SKB_STREAM_VERDICT:
4209         case BPF_SK_MSG_VERDICT:
4210         case BPF_SK_SKB_VERDICT:
4211                 return sock_map_bpf_prog_query(attr, uattr);
4212         case BPF_TCX_INGRESS:
4213         case BPF_TCX_EGRESS:
4214                 return tcx_prog_query(attr, uattr);
4215         case BPF_NETKIT_PRIMARY:
4216         case BPF_NETKIT_PEER:
4217                 return netkit_prog_query(attr, uattr);
4218         default:
4219                 return -EINVAL;
4220         }
4221 }
4222
4223 #define BPF_PROG_TEST_RUN_LAST_FIELD test.batch_size
4224
4225 static int bpf_prog_test_run(const union bpf_attr *attr,
4226                              union bpf_attr __user *uattr)
4227 {
4228         struct bpf_prog *prog;
4229         int ret = -ENOTSUPP;
4230
4231         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
4232                 return -EINVAL;
4233
4234         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
4235             (!attr->test.ctx_size_in && attr->test.ctx_in))
4236                 return -EINVAL;
4237
4238         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
4239             (!attr->test.ctx_size_out && attr->test.ctx_out))
4240                 return -EINVAL;
4241
4242         prog = bpf_prog_get(attr->test.prog_fd);
4243         if (IS_ERR(prog))
4244                 return PTR_ERR(prog);
4245
4246         if (prog->aux->ops->test_run)
4247                 ret = prog->aux->ops->test_run(prog, attr, uattr);
4248
4249         bpf_prog_put(prog);
4250         return ret;
4251 }
4252
4253 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
4254
4255 static int bpf_obj_get_next_id(const union bpf_attr *attr,
4256                                union bpf_attr __user *uattr,
4257                                struct idr *idr,
4258                                spinlock_t *lock)
4259 {
4260         u32 next_id = attr->start_id;
4261         int err = 0;
4262
4263         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
4264                 return -EINVAL;
4265
4266         if (!capable(CAP_SYS_ADMIN))
4267                 return -EPERM;
4268
4269         next_id++;
4270         spin_lock_bh(lock);
4271         if (!idr_get_next(idr, &next_id))
4272                 err = -ENOENT;
4273         spin_unlock_bh(lock);
4274
4275         if (!err)
4276                 err = put_user(next_id, &uattr->next_id);
4277
4278         return err;
4279 }
4280
4281 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
4282 {
4283         struct bpf_map *map;
4284
4285         spin_lock_bh(&map_idr_lock);
4286 again:
4287         map = idr_get_next(&map_idr, id);
4288         if (map) {
4289                 map = __bpf_map_inc_not_zero(map, false);
4290                 if (IS_ERR(map)) {
4291                         (*id)++;
4292                         goto again;
4293                 }
4294         }
4295         spin_unlock_bh(&map_idr_lock);
4296
4297         return map;
4298 }
4299
4300 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
4301 {
4302         struct bpf_prog *prog;
4303
4304         spin_lock_bh(&prog_idr_lock);
4305 again:
4306         prog = idr_get_next(&prog_idr, id);
4307         if (prog) {
4308                 prog = bpf_prog_inc_not_zero(prog);
4309                 if (IS_ERR(prog)) {
4310                         (*id)++;
4311                         goto again;
4312                 }
4313         }
4314         spin_unlock_bh(&prog_idr_lock);
4315
4316         return prog;
4317 }
4318
4319 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
4320
4321 struct bpf_prog *bpf_prog_by_id(u32 id)
4322 {
4323         struct bpf_prog *prog;
4324
4325         if (!id)
4326                 return ERR_PTR(-ENOENT);
4327
4328         spin_lock_bh(&prog_idr_lock);
4329         prog = idr_find(&prog_idr, id);
4330         if (prog)
4331                 prog = bpf_prog_inc_not_zero(prog);
4332         else
4333                 prog = ERR_PTR(-ENOENT);
4334         spin_unlock_bh(&prog_idr_lock);
4335         return prog;
4336 }
4337
4338 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
4339 {
4340         struct bpf_prog *prog;
4341         u32 id = attr->prog_id;
4342         int fd;
4343
4344         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
4345                 return -EINVAL;
4346
4347         if (!capable(CAP_SYS_ADMIN))
4348                 return -EPERM;
4349
4350         prog = bpf_prog_by_id(id);
4351         if (IS_ERR(prog))
4352                 return PTR_ERR(prog);
4353
4354         fd = bpf_prog_new_fd(prog);
4355         if (fd < 0)
4356                 bpf_prog_put(prog);
4357
4358         return fd;
4359 }
4360
4361 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
4362
4363 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
4364 {
4365         struct bpf_map *map;
4366         u32 id = attr->map_id;
4367         int f_flags;
4368         int fd;
4369
4370         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
4371             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
4372                 return -EINVAL;
4373
4374         if (!capable(CAP_SYS_ADMIN))
4375                 return -EPERM;
4376
4377         f_flags = bpf_get_file_flag(attr->open_flags);
4378         if (f_flags < 0)
4379                 return f_flags;
4380
4381         spin_lock_bh(&map_idr_lock);
4382         map = idr_find(&map_idr, id);
4383         if (map)
4384                 map = __bpf_map_inc_not_zero(map, true);
4385         else
4386                 map = ERR_PTR(-ENOENT);
4387         spin_unlock_bh(&map_idr_lock);
4388
4389         if (IS_ERR(map))
4390                 return PTR_ERR(map);
4391
4392         fd = bpf_map_new_fd(map, f_flags);
4393         if (fd < 0)
4394                 bpf_map_put_with_uref(map);
4395
4396         return fd;
4397 }
4398
4399 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
4400                                               unsigned long addr, u32 *off,
4401                                               u32 *type)
4402 {
4403         const struct bpf_map *map;
4404         int i;
4405
4406         mutex_lock(&prog->aux->used_maps_mutex);
4407         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
4408                 map = prog->aux->used_maps[i];
4409                 if (map == (void *)addr) {
4410                         *type = BPF_PSEUDO_MAP_FD;
4411                         goto out;
4412                 }
4413                 if (!map->ops->map_direct_value_meta)
4414                         continue;
4415                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
4416                         *type = BPF_PSEUDO_MAP_VALUE;
4417                         goto out;
4418                 }
4419         }
4420         map = NULL;
4421
4422 out:
4423         mutex_unlock(&prog->aux->used_maps_mutex);
4424         return map;
4425 }
4426
4427 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
4428                                               const struct cred *f_cred)
4429 {
4430         const struct bpf_map *map;
4431         struct bpf_insn *insns;
4432         u32 off, type;
4433         u64 imm;
4434         u8 code;
4435         int i;
4436
4437         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
4438                         GFP_USER);
4439         if (!insns)
4440                 return insns;
4441
4442         for (i = 0; i < prog->len; i++) {
4443                 code = insns[i].code;
4444
4445                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
4446                         insns[i].code = BPF_JMP | BPF_CALL;
4447                         insns[i].imm = BPF_FUNC_tail_call;
4448                         /* fall-through */
4449                 }
4450                 if (code == (BPF_JMP | BPF_CALL) ||
4451                     code == (BPF_JMP | BPF_CALL_ARGS)) {
4452                         if (code == (BPF_JMP | BPF_CALL_ARGS))
4453                                 insns[i].code = BPF_JMP | BPF_CALL;
4454                         if (!bpf_dump_raw_ok(f_cred))
4455                                 insns[i].imm = 0;
4456                         continue;
4457                 }
4458                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
4459                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
4460                         continue;
4461                 }
4462
4463                 if ((BPF_CLASS(code) == BPF_LDX || BPF_CLASS(code) == BPF_STX ||
4464                      BPF_CLASS(code) == BPF_ST) && BPF_MODE(code) == BPF_PROBE_MEM32) {
4465                         insns[i].code = BPF_CLASS(code) | BPF_SIZE(code) | BPF_MEM;
4466                         continue;
4467                 }
4468
4469                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
4470                         continue;
4471
4472                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
4473                 map = bpf_map_from_imm(prog, imm, &off, &type);
4474                 if (map) {
4475                         insns[i].src_reg = type;
4476                         insns[i].imm = map->id;
4477                         insns[i + 1].imm = off;
4478                         continue;
4479                 }
4480         }
4481
4482         return insns;
4483 }
4484
4485 static int set_info_rec_size(struct bpf_prog_info *info)
4486 {
4487         /*
4488          * Ensure info.*_rec_size is the same as kernel expected size
4489          *
4490          * or
4491          *
4492          * Only allow zero *_rec_size if both _rec_size and _cnt are
4493          * zero.  In this case, the kernel will set the expected
4494          * _rec_size back to the info.
4495          */
4496
4497         if ((info->nr_func_info || info->func_info_rec_size) &&
4498             info->func_info_rec_size != sizeof(struct bpf_func_info))
4499                 return -EINVAL;
4500
4501         if ((info->nr_line_info || info->line_info_rec_size) &&
4502             info->line_info_rec_size != sizeof(struct bpf_line_info))
4503                 return -EINVAL;
4504
4505         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
4506             info->jited_line_info_rec_size != sizeof(__u64))
4507                 return -EINVAL;
4508
4509         info->func_info_rec_size = sizeof(struct bpf_func_info);
4510         info->line_info_rec_size = sizeof(struct bpf_line_info);
4511         info->jited_line_info_rec_size = sizeof(__u64);
4512
4513         return 0;
4514 }
4515
4516 static int bpf_prog_get_info_by_fd(struct file *file,
4517                                    struct bpf_prog *prog,
4518                                    const union bpf_attr *attr,
4519                                    union bpf_attr __user *uattr)
4520 {
4521         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4522         struct btf *attach_btf = bpf_prog_get_target_btf(prog);
4523         struct bpf_prog_info info;
4524         u32 info_len = attr->info.info_len;
4525         struct bpf_prog_kstats stats;
4526         char __user *uinsns;
4527         u32 ulen;
4528         int err;
4529
4530         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4531         if (err)
4532                 return err;
4533         info_len = min_t(u32, sizeof(info), info_len);
4534
4535         memset(&info, 0, sizeof(info));
4536         if (copy_from_user(&info, uinfo, info_len))
4537                 return -EFAULT;
4538
4539         info.type = prog->type;
4540         info.id = prog->aux->id;
4541         info.load_time = prog->aux->load_time;
4542         info.created_by_uid = from_kuid_munged(current_user_ns(),
4543                                                prog->aux->user->uid);
4544         info.gpl_compatible = prog->gpl_compatible;
4545
4546         memcpy(info.tag, prog->tag, sizeof(prog->tag));
4547         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
4548
4549         mutex_lock(&prog->aux->used_maps_mutex);
4550         ulen = info.nr_map_ids;
4551         info.nr_map_ids = prog->aux->used_map_cnt;
4552         ulen = min_t(u32, info.nr_map_ids, ulen);
4553         if (ulen) {
4554                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
4555                 u32 i;
4556
4557                 for (i = 0; i < ulen; i++)
4558                         if (put_user(prog->aux->used_maps[i]->id,
4559                                      &user_map_ids[i])) {
4560                                 mutex_unlock(&prog->aux->used_maps_mutex);
4561                                 return -EFAULT;
4562                         }
4563         }
4564         mutex_unlock(&prog->aux->used_maps_mutex);
4565
4566         err = set_info_rec_size(&info);
4567         if (err)
4568                 return err;
4569
4570         bpf_prog_get_stats(prog, &stats);
4571         info.run_time_ns = stats.nsecs;
4572         info.run_cnt = stats.cnt;
4573         info.recursion_misses = stats.misses;
4574
4575         info.verified_insns = prog->aux->verified_insns;
4576
4577         if (!bpf_capable()) {
4578                 info.jited_prog_len = 0;
4579                 info.xlated_prog_len = 0;
4580                 info.nr_jited_ksyms = 0;
4581                 info.nr_jited_func_lens = 0;
4582                 info.nr_func_info = 0;
4583                 info.nr_line_info = 0;
4584                 info.nr_jited_line_info = 0;
4585                 goto done;
4586         }
4587
4588         ulen = info.xlated_prog_len;
4589         info.xlated_prog_len = bpf_prog_insn_size(prog);
4590         if (info.xlated_prog_len && ulen) {
4591                 struct bpf_insn *insns_sanitized;
4592                 bool fault;
4593
4594                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
4595                         info.xlated_prog_insns = 0;
4596                         goto done;
4597                 }
4598                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
4599                 if (!insns_sanitized)
4600                         return -ENOMEM;
4601                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
4602                 ulen = min_t(u32, info.xlated_prog_len, ulen);
4603                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
4604                 kfree(insns_sanitized);
4605                 if (fault)
4606                         return -EFAULT;
4607         }
4608
4609         if (bpf_prog_is_offloaded(prog->aux)) {
4610                 err = bpf_prog_offload_info_fill(&info, prog);
4611                 if (err)
4612                         return err;
4613                 goto done;
4614         }
4615
4616         /* NOTE: the following code is supposed to be skipped for offload.
4617          * bpf_prog_offload_info_fill() is the place to fill similar fields
4618          * for offload.
4619          */
4620         ulen = info.jited_prog_len;
4621         if (prog->aux->func_cnt) {
4622                 u32 i;
4623
4624                 info.jited_prog_len = 0;
4625                 for (i = 0; i < prog->aux->func_cnt; i++)
4626                         info.jited_prog_len += prog->aux->func[i]->jited_len;
4627         } else {
4628                 info.jited_prog_len = prog->jited_len;
4629         }
4630
4631         if (info.jited_prog_len && ulen) {
4632                 if (bpf_dump_raw_ok(file->f_cred)) {
4633                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
4634                         ulen = min_t(u32, info.jited_prog_len, ulen);
4635
4636                         /* for multi-function programs, copy the JITed
4637                          * instructions for all the functions
4638                          */
4639                         if (prog->aux->func_cnt) {
4640                                 u32 len, free, i;
4641                                 u8 *img;
4642
4643                                 free = ulen;
4644                                 for (i = 0; i < prog->aux->func_cnt; i++) {
4645                                         len = prog->aux->func[i]->jited_len;
4646                                         len = min_t(u32, len, free);
4647                                         img = (u8 *) prog->aux->func[i]->bpf_func;
4648                                         if (copy_to_user(uinsns, img, len))
4649                                                 return -EFAULT;
4650                                         uinsns += len;
4651                                         free -= len;
4652                                         if (!free)
4653                                                 break;
4654                                 }
4655                         } else {
4656                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
4657                                         return -EFAULT;
4658                         }
4659                 } else {
4660                         info.jited_prog_insns = 0;
4661                 }
4662         }
4663
4664         ulen = info.nr_jited_ksyms;
4665         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
4666         if (ulen) {
4667                 if (bpf_dump_raw_ok(file->f_cred)) {
4668                         unsigned long ksym_addr;
4669                         u64 __user *user_ksyms;
4670                         u32 i;
4671
4672                         /* copy the address of the kernel symbol
4673                          * corresponding to each function
4674                          */
4675                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
4676                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
4677                         if (prog->aux->func_cnt) {
4678                                 for (i = 0; i < ulen; i++) {
4679                                         ksym_addr = (unsigned long)
4680                                                 prog->aux->func[i]->bpf_func;
4681                                         if (put_user((u64) ksym_addr,
4682                                                      &user_ksyms[i]))
4683                                                 return -EFAULT;
4684                                 }
4685                         } else {
4686                                 ksym_addr = (unsigned long) prog->bpf_func;
4687                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
4688                                         return -EFAULT;
4689                         }
4690                 } else {
4691                         info.jited_ksyms = 0;
4692                 }
4693         }
4694
4695         ulen = info.nr_jited_func_lens;
4696         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
4697         if (ulen) {
4698                 if (bpf_dump_raw_ok(file->f_cred)) {
4699                         u32 __user *user_lens;
4700                         u32 func_len, i;
4701
4702                         /* copy the JITed image lengths for each function */
4703                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
4704                         user_lens = u64_to_user_ptr(info.jited_func_lens);
4705                         if (prog->aux->func_cnt) {
4706                                 for (i = 0; i < ulen; i++) {
4707                                         func_len =
4708                                                 prog->aux->func[i]->jited_len;
4709                                         if (put_user(func_len, &user_lens[i]))
4710                                                 return -EFAULT;
4711                                 }
4712                         } else {
4713                                 func_len = prog->jited_len;
4714                                 if (put_user(func_len, &user_lens[0]))
4715                                         return -EFAULT;
4716                         }
4717                 } else {
4718                         info.jited_func_lens = 0;
4719                 }
4720         }
4721
4722         if (prog->aux->btf)
4723                 info.btf_id = btf_obj_id(prog->aux->btf);
4724         info.attach_btf_id = prog->aux->attach_btf_id;
4725         if (attach_btf)
4726                 info.attach_btf_obj_id = btf_obj_id(attach_btf);
4727
4728         ulen = info.nr_func_info;
4729         info.nr_func_info = prog->aux->func_info_cnt;
4730         if (info.nr_func_info && ulen) {
4731                 char __user *user_finfo;
4732
4733                 user_finfo = u64_to_user_ptr(info.func_info);
4734                 ulen = min_t(u32, info.nr_func_info, ulen);
4735                 if (copy_to_user(user_finfo, prog->aux->func_info,
4736                                  info.func_info_rec_size * ulen))
4737                         return -EFAULT;
4738         }
4739
4740         ulen = info.nr_line_info;
4741         info.nr_line_info = prog->aux->nr_linfo;
4742         if (info.nr_line_info && ulen) {
4743                 __u8 __user *user_linfo;
4744
4745                 user_linfo = u64_to_user_ptr(info.line_info);
4746                 ulen = min_t(u32, info.nr_line_info, ulen);
4747                 if (copy_to_user(user_linfo, prog->aux->linfo,
4748                                  info.line_info_rec_size * ulen))
4749                         return -EFAULT;
4750         }
4751
4752         ulen = info.nr_jited_line_info;
4753         if (prog->aux->jited_linfo)
4754                 info.nr_jited_line_info = prog->aux->nr_linfo;
4755         else
4756                 info.nr_jited_line_info = 0;
4757         if (info.nr_jited_line_info && ulen) {
4758                 if (bpf_dump_raw_ok(file->f_cred)) {
4759                         unsigned long line_addr;
4760                         __u64 __user *user_linfo;
4761                         u32 i;
4762
4763                         user_linfo = u64_to_user_ptr(info.jited_line_info);
4764                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
4765                         for (i = 0; i < ulen; i++) {
4766                                 line_addr = (unsigned long)prog->aux->jited_linfo[i];
4767                                 if (put_user((__u64)line_addr, &user_linfo[i]))
4768                                         return -EFAULT;
4769                         }
4770                 } else {
4771                         info.jited_line_info = 0;
4772                 }
4773         }
4774
4775         ulen = info.nr_prog_tags;
4776         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
4777         if (ulen) {
4778                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
4779                 u32 i;
4780
4781                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
4782                 ulen = min_t(u32, info.nr_prog_tags, ulen);
4783                 if (prog->aux->func_cnt) {
4784                         for (i = 0; i < ulen; i++) {
4785                                 if (copy_to_user(user_prog_tags[i],
4786                                                  prog->aux->func[i]->tag,
4787                                                  BPF_TAG_SIZE))
4788                                         return -EFAULT;
4789                         }
4790                 } else {
4791                         if (copy_to_user(user_prog_tags[0],
4792                                          prog->tag, BPF_TAG_SIZE))
4793                                 return -EFAULT;
4794                 }
4795         }
4796
4797 done:
4798         if (copy_to_user(uinfo, &info, info_len) ||
4799             put_user(info_len, &uattr->info.info_len))
4800                 return -EFAULT;
4801
4802         return 0;
4803 }
4804
4805 static int bpf_map_get_info_by_fd(struct file *file,
4806                                   struct bpf_map *map,
4807                                   const union bpf_attr *attr,
4808                                   union bpf_attr __user *uattr)
4809 {
4810         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4811         struct bpf_map_info info;
4812         u32 info_len = attr->info.info_len;
4813         int err;
4814
4815         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4816         if (err)
4817                 return err;
4818         info_len = min_t(u32, sizeof(info), info_len);
4819
4820         memset(&info, 0, sizeof(info));
4821         info.type = map->map_type;
4822         info.id = map->id;
4823         info.key_size = map->key_size;
4824         info.value_size = map->value_size;
4825         info.max_entries = map->max_entries;
4826         info.map_flags = map->map_flags;
4827         info.map_extra = map->map_extra;
4828         memcpy(info.name, map->name, sizeof(map->name));
4829
4830         if (map->btf) {
4831                 info.btf_id = btf_obj_id(map->btf);
4832                 info.btf_key_type_id = map->btf_key_type_id;
4833                 info.btf_value_type_id = map->btf_value_type_id;
4834         }
4835         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
4836         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS)
4837                 bpf_map_struct_ops_info_fill(&info, map);
4838
4839         if (bpf_map_is_offloaded(map)) {
4840                 err = bpf_map_offload_info_fill(&info, map);
4841                 if (err)
4842                         return err;
4843         }
4844
4845         if (copy_to_user(uinfo, &info, info_len) ||
4846             put_user(info_len, &uattr->info.info_len))
4847                 return -EFAULT;
4848
4849         return 0;
4850 }
4851
4852 static int bpf_btf_get_info_by_fd(struct file *file,
4853                                   struct btf *btf,
4854                                   const union bpf_attr *attr,
4855                                   union bpf_attr __user *uattr)
4856 {
4857         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4858         u32 info_len = attr->info.info_len;
4859         int err;
4860
4861         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(*uinfo), info_len);
4862         if (err)
4863                 return err;
4864
4865         return btf_get_info_by_fd(btf, attr, uattr);
4866 }
4867
4868 static int bpf_link_get_info_by_fd(struct file *file,
4869                                   struct bpf_link *link,
4870                                   const union bpf_attr *attr,
4871                                   union bpf_attr __user *uattr)
4872 {
4873         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
4874         struct bpf_link_info info;
4875         u32 info_len = attr->info.info_len;
4876         int err;
4877
4878         err = bpf_check_uarg_tail_zero(USER_BPFPTR(uinfo), sizeof(info), info_len);
4879         if (err)
4880                 return err;
4881         info_len = min_t(u32, sizeof(info), info_len);
4882
4883         memset(&info, 0, sizeof(info));
4884         if (copy_from_user(&info, uinfo, info_len))
4885                 return -EFAULT;
4886
4887         info.type = link->type;
4888         info.id = link->id;
4889         if (link->prog)
4890                 info.prog_id = link->prog->aux->id;
4891
4892         if (link->ops->fill_link_info) {
4893                 err = link->ops->fill_link_info(link, &info);
4894                 if (err)
4895                         return err;
4896         }
4897
4898         if (copy_to_user(uinfo, &info, info_len) ||
4899             put_user(info_len, &uattr->info.info_len))
4900                 return -EFAULT;
4901
4902         return 0;
4903 }
4904
4905
4906 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
4907
4908 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
4909                                   union bpf_attr __user *uattr)
4910 {
4911         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
4912                 return -EINVAL;
4913
4914         CLASS(fd, f)(attr->info.bpf_fd);
4915         if (fd_empty(f))
4916                 return -EBADFD;
4917
4918         if (fd_file(f)->f_op == &bpf_prog_fops)
4919                 return bpf_prog_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
4920                                               uattr);
4921         else if (fd_file(f)->f_op == &bpf_map_fops)
4922                 return bpf_map_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr,
4923                                              uattr);
4924         else if (fd_file(f)->f_op == &btf_fops)
4925                 return bpf_btf_get_info_by_fd(fd_file(f), fd_file(f)->private_data, attr, uattr);
4926         else if (fd_file(f)->f_op == &bpf_link_fops || fd_file(f)->f_op == &bpf_link_fops_poll)
4927                 return bpf_link_get_info_by_fd(fd_file(f), fd_file(f)->private_data,
4928                                               attr, uattr);
4929         return -EINVAL;
4930 }
4931
4932 #define BPF_BTF_LOAD_LAST_FIELD btf_token_fd
4933
4934 static int bpf_btf_load(const union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size)
4935 {
4936         struct bpf_token *token = NULL;
4937
4938         if (CHECK_ATTR(BPF_BTF_LOAD))
4939                 return -EINVAL;
4940
4941         if (attr->btf_flags & ~BPF_F_TOKEN_FD)
4942                 return -EINVAL;
4943
4944         if (attr->btf_flags & BPF_F_TOKEN_FD) {
4945                 token = bpf_token_get_from_fd(attr->btf_token_fd);
4946                 if (IS_ERR(token))
4947                         return PTR_ERR(token);
4948                 if (!bpf_token_allow_cmd(token, BPF_BTF_LOAD)) {
4949                         bpf_token_put(token);
4950                         token = NULL;
4951                 }
4952         }
4953
4954         if (!bpf_token_capable(token, CAP_BPF)) {
4955                 bpf_token_put(token);
4956                 return -EPERM;
4957         }
4958
4959         bpf_token_put(token);
4960
4961         return btf_new_fd(attr, uattr, uattr_size);
4962 }
4963
4964 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
4965
4966 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
4967 {
4968         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
4969                 return -EINVAL;
4970
4971         if (!capable(CAP_SYS_ADMIN))
4972                 return -EPERM;
4973
4974         return btf_get_fd_by_id(attr->btf_id);
4975 }
4976
4977 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
4978                                     union bpf_attr __user *uattr,
4979                                     u32 prog_id, u32 fd_type,
4980                                     const char *buf, u64 probe_offset,
4981                                     u64 probe_addr)
4982 {
4983         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
4984         u32 len = buf ? strlen(buf) : 0, input_len;
4985         int err = 0;
4986
4987         if (put_user(len, &uattr->task_fd_query.buf_len))
4988                 return -EFAULT;
4989         input_len = attr->task_fd_query.buf_len;
4990         if (input_len && ubuf) {
4991                 if (!len) {
4992                         /* nothing to copy, just make ubuf NULL terminated */
4993                         char zero = '\0';
4994
4995                         if (put_user(zero, ubuf))
4996                                 return -EFAULT;
4997                 } else if (input_len >= len + 1) {
4998                         /* ubuf can hold the string with NULL terminator */
4999                         if (copy_to_user(ubuf, buf, len + 1))
5000                                 return -EFAULT;
5001                 } else {
5002                         /* ubuf cannot hold the string with NULL terminator,
5003                          * do a partial copy with NULL terminator.
5004                          */
5005                         char zero = '\0';
5006
5007                         err = -ENOSPC;
5008                         if (copy_to_user(ubuf, buf, input_len - 1))
5009                                 return -EFAULT;
5010                         if (put_user(zero, ubuf + input_len - 1))
5011                                 return -EFAULT;
5012                 }
5013         }
5014
5015         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
5016             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
5017             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
5018             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
5019                 return -EFAULT;
5020
5021         return err;
5022 }
5023
5024 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
5025
5026 static int bpf_task_fd_query(const union bpf_attr *attr,
5027                              union bpf_attr __user *uattr)
5028 {
5029         pid_t pid = attr->task_fd_query.pid;
5030         u32 fd = attr->task_fd_query.fd;
5031         const struct perf_event *event;
5032         struct task_struct *task;
5033         struct file *file;
5034         int err;
5035
5036         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
5037                 return -EINVAL;
5038
5039         if (!capable(CAP_SYS_ADMIN))
5040                 return -EPERM;
5041
5042         if (attr->task_fd_query.flags != 0)
5043                 return -EINVAL;
5044
5045         rcu_read_lock();
5046         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
5047         rcu_read_unlock();
5048         if (!task)
5049                 return -ENOENT;
5050
5051         err = 0;
5052         file = fget_task(task, fd);
5053         put_task_struct(task);
5054         if (!file)
5055                 return -EBADF;
5056
5057         if (file->f_op == &bpf_link_fops || file->f_op == &bpf_link_fops_poll) {
5058                 struct bpf_link *link = file->private_data;
5059
5060                 if (link->ops == &bpf_raw_tp_link_lops) {
5061                         struct bpf_raw_tp_link *raw_tp =
5062                                 container_of(link, struct bpf_raw_tp_link, link);
5063                         struct bpf_raw_event_map *btp = raw_tp->btp;
5064
5065                         err = bpf_task_fd_query_copy(attr, uattr,
5066                                                      raw_tp->link.prog->aux->id,
5067                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
5068                                                      btp->tp->name, 0, 0);
5069                         goto put_file;
5070                 }
5071                 goto out_not_supp;
5072         }
5073
5074         event = perf_get_event(file);
5075         if (!IS_ERR(event)) {
5076                 u64 probe_offset, probe_addr;
5077                 u32 prog_id, fd_type;
5078                 const char *buf;
5079
5080                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
5081                                               &buf, &probe_offset,
5082                                               &probe_addr, NULL);
5083                 if (!err)
5084                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
5085                                                      fd_type, buf,
5086                                                      probe_offset,
5087                                                      probe_addr);
5088                 goto put_file;
5089         }
5090
5091 out_not_supp:
5092         err = -ENOTSUPP;
5093 put_file:
5094         fput(file);
5095         return err;
5096 }
5097
5098 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
5099
5100 #define BPF_DO_BATCH(fn, ...)                   \
5101         do {                                    \
5102                 if (!fn) {                      \
5103                         err = -ENOTSUPP;        \
5104                         goto err_put;           \
5105                 }                               \
5106                 err = fn(__VA_ARGS__);          \
5107         } while (0)
5108
5109 static int bpf_map_do_batch(const union bpf_attr *attr,
5110                             union bpf_attr __user *uattr,
5111                             int cmd)
5112 {
5113         bool has_read  = cmd == BPF_MAP_LOOKUP_BATCH ||
5114                          cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH;
5115         bool has_write = cmd != BPF_MAP_LOOKUP_BATCH;
5116         struct bpf_map *map;
5117         int err;
5118
5119         if (CHECK_ATTR(BPF_MAP_BATCH))
5120                 return -EINVAL;
5121
5122         CLASS(fd, f)(attr->batch.map_fd);
5123
5124         map = __bpf_map_get(f);
5125         if (IS_ERR(map))
5126                 return PTR_ERR(map);
5127         if (has_write)
5128                 bpf_map_write_active_inc(map);
5129         if (has_read && !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
5130                 err = -EPERM;
5131                 goto err_put;
5132         }
5133         if (has_write && !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
5134                 err = -EPERM;
5135                 goto err_put;
5136         }
5137
5138         if (cmd == BPF_MAP_LOOKUP_BATCH)
5139                 BPF_DO_BATCH(map->ops->map_lookup_batch, map, attr, uattr);
5140         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
5141                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch, map, attr, uattr);
5142         else if (cmd == BPF_MAP_UPDATE_BATCH)
5143                 BPF_DO_BATCH(map->ops->map_update_batch, map, fd_file(f), attr, uattr);
5144         else
5145                 BPF_DO_BATCH(map->ops->map_delete_batch, map, attr, uattr);
5146 err_put:
5147         if (has_write) {
5148                 maybe_wait_bpf_programs(map);
5149                 bpf_map_write_active_dec(map);
5150         }
5151         return err;
5152 }
5153
5154 #define BPF_LINK_CREATE_LAST_FIELD link_create.uprobe_multi.pid
5155 static int link_create(union bpf_attr *attr, bpfptr_t uattr)
5156 {
5157         struct bpf_prog *prog;
5158         int ret;
5159
5160         if (CHECK_ATTR(BPF_LINK_CREATE))
5161                 return -EINVAL;
5162
5163         if (attr->link_create.attach_type == BPF_STRUCT_OPS)
5164                 return bpf_struct_ops_link_create(attr);
5165
5166         prog = bpf_prog_get(attr->link_create.prog_fd);
5167         if (IS_ERR(prog))
5168                 return PTR_ERR(prog);
5169
5170         ret = bpf_prog_attach_check_attach_type(prog,
5171                                                 attr->link_create.attach_type);
5172         if (ret)
5173                 goto out;
5174
5175         switch (prog->type) {
5176         case BPF_PROG_TYPE_CGROUP_SKB:
5177         case BPF_PROG_TYPE_CGROUP_SOCK:
5178         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
5179         case BPF_PROG_TYPE_SOCK_OPS:
5180         case BPF_PROG_TYPE_CGROUP_DEVICE:
5181         case BPF_PROG_TYPE_CGROUP_SYSCTL:
5182         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
5183                 ret = cgroup_bpf_link_attach(attr, prog);
5184                 break;
5185         case BPF_PROG_TYPE_EXT:
5186                 ret = bpf_tracing_prog_attach(prog,
5187                                               attr->link_create.target_fd,
5188                                               attr->link_create.target_btf_id,
5189                                               attr->link_create.tracing.cookie);
5190                 break;
5191         case BPF_PROG_TYPE_LSM:
5192         case BPF_PROG_TYPE_TRACING:
5193                 if (attr->link_create.attach_type != prog->expected_attach_type) {
5194                         ret = -EINVAL;
5195                         goto out;
5196                 }
5197                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
5198                         ret = bpf_raw_tp_link_attach(prog, NULL, attr->link_create.tracing.cookie);
5199                 else if (prog->expected_attach_type == BPF_TRACE_ITER)
5200                         ret = bpf_iter_link_attach(attr, uattr, prog);
5201                 else if (prog->expected_attach_type == BPF_LSM_CGROUP)
5202                         ret = cgroup_bpf_link_attach(attr, prog);
5203                 else
5204                         ret = bpf_tracing_prog_attach(prog,
5205                                                       attr->link_create.target_fd,
5206                                                       attr->link_create.target_btf_id,
5207                                                       attr->link_create.tracing.cookie);
5208                 break;
5209         case BPF_PROG_TYPE_FLOW_DISSECTOR:
5210         case BPF_PROG_TYPE_SK_LOOKUP:
5211                 ret = netns_bpf_link_create(attr, prog);
5212                 break;
5213         case BPF_PROG_TYPE_SK_MSG:
5214         case BPF_PROG_TYPE_SK_SKB:
5215                 ret = sock_map_link_create(attr, prog);
5216                 break;
5217 #ifdef CONFIG_NET
5218         case BPF_PROG_TYPE_XDP:
5219                 ret = bpf_xdp_link_attach(attr, prog);
5220                 break;
5221         case BPF_PROG_TYPE_SCHED_CLS:
5222                 if (attr->link_create.attach_type == BPF_TCX_INGRESS ||
5223                     attr->link_create.attach_type == BPF_TCX_EGRESS)
5224                         ret = tcx_link_attach(attr, prog);
5225                 else
5226                         ret = netkit_link_attach(attr, prog);
5227                 break;
5228         case BPF_PROG_TYPE_NETFILTER:
5229                 ret = bpf_nf_link_attach(attr, prog);
5230                 break;
5231 #endif
5232         case BPF_PROG_TYPE_PERF_EVENT:
5233         case BPF_PROG_TYPE_TRACEPOINT:
5234                 ret = bpf_perf_link_attach(attr, prog);
5235                 break;
5236         case BPF_PROG_TYPE_KPROBE:
5237                 if (attr->link_create.attach_type == BPF_PERF_EVENT)
5238                         ret = bpf_perf_link_attach(attr, prog);
5239                 else if (attr->link_create.attach_type == BPF_TRACE_KPROBE_MULTI ||
5240                          attr->link_create.attach_type == BPF_TRACE_KPROBE_SESSION)
5241                         ret = bpf_kprobe_multi_link_attach(attr, prog);
5242                 else if (attr->link_create.attach_type == BPF_TRACE_UPROBE_MULTI)
5243                         ret = bpf_uprobe_multi_link_attach(attr, prog);
5244                 break;
5245         default:
5246                 ret = -EINVAL;
5247         }
5248
5249 out:
5250         if (ret < 0)
5251                 bpf_prog_put(prog);
5252         return ret;
5253 }
5254
5255 static int link_update_map(struct bpf_link *link, union bpf_attr *attr)
5256 {
5257         struct bpf_map *new_map, *old_map = NULL;
5258         int ret;
5259
5260         new_map = bpf_map_get(attr->link_update.new_map_fd);
5261         if (IS_ERR(new_map))
5262                 return PTR_ERR(new_map);
5263
5264         if (attr->link_update.flags & BPF_F_REPLACE) {
5265                 old_map = bpf_map_get(attr->link_update.old_map_fd);
5266                 if (IS_ERR(old_map)) {
5267                         ret = PTR_ERR(old_map);
5268                         goto out_put;
5269                 }
5270         } else if (attr->link_update.old_map_fd) {
5271                 ret = -EINVAL;
5272                 goto out_put;
5273         }
5274
5275         ret = link->ops->update_map(link, new_map, old_map);
5276
5277         if (old_map)
5278                 bpf_map_put(old_map);
5279 out_put:
5280         bpf_map_put(new_map);
5281         return ret;
5282 }
5283
5284 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
5285
5286 static int link_update(union bpf_attr *attr)
5287 {
5288         struct bpf_prog *old_prog = NULL, *new_prog;
5289         struct bpf_link *link;
5290         u32 flags;
5291         int ret;
5292
5293         if (CHECK_ATTR(BPF_LINK_UPDATE))
5294                 return -EINVAL;
5295
5296         flags = attr->link_update.flags;
5297         if (flags & ~BPF_F_REPLACE)
5298                 return -EINVAL;
5299
5300         link = bpf_link_get_from_fd(attr->link_update.link_fd);
5301         if (IS_ERR(link))
5302                 return PTR_ERR(link);
5303
5304         if (link->ops->update_map) {
5305                 ret = link_update_map(link, attr);
5306                 goto out_put_link;
5307         }
5308
5309         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
5310         if (IS_ERR(new_prog)) {
5311                 ret = PTR_ERR(new_prog);
5312                 goto out_put_link;
5313         }
5314
5315         if (flags & BPF_F_REPLACE) {
5316                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
5317                 if (IS_ERR(old_prog)) {
5318                         ret = PTR_ERR(old_prog);
5319                         old_prog = NULL;
5320                         goto out_put_progs;
5321                 }
5322         } else if (attr->link_update.old_prog_fd) {
5323                 ret = -EINVAL;
5324                 goto out_put_progs;
5325         }
5326
5327         if (link->ops->update_prog)
5328                 ret = link->ops->update_prog(link, new_prog, old_prog);
5329         else
5330                 ret = -EINVAL;
5331
5332 out_put_progs:
5333         if (old_prog)
5334                 bpf_prog_put(old_prog);
5335         if (ret)
5336                 bpf_prog_put(new_prog);
5337 out_put_link:
5338         bpf_link_put_direct(link);
5339         return ret;
5340 }
5341
5342 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
5343
5344 static int link_detach(union bpf_attr *attr)
5345 {
5346         struct bpf_link *link;
5347         int ret;
5348
5349         if (CHECK_ATTR(BPF_LINK_DETACH))
5350                 return -EINVAL;
5351
5352         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
5353         if (IS_ERR(link))
5354                 return PTR_ERR(link);
5355
5356         if (link->ops->detach)
5357                 ret = link->ops->detach(link);
5358         else
5359                 ret = -EOPNOTSUPP;
5360
5361         bpf_link_put_direct(link);
5362         return ret;
5363 }
5364
5365 struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
5366 {
5367         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
5368 }
5369 EXPORT_SYMBOL(bpf_link_inc_not_zero);
5370
5371 struct bpf_link *bpf_link_by_id(u32 id)
5372 {
5373         struct bpf_link *link;
5374
5375         if (!id)
5376                 return ERR_PTR(-ENOENT);
5377
5378         spin_lock_bh(&link_idr_lock);
5379         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
5380         link = idr_find(&link_idr, id);
5381         if (link) {
5382                 if (link->id)
5383                         link = bpf_link_inc_not_zero(link);
5384                 else
5385                         link = ERR_PTR(-EAGAIN);
5386         } else {
5387                 link = ERR_PTR(-ENOENT);
5388         }
5389         spin_unlock_bh(&link_idr_lock);
5390         return link;
5391 }
5392
5393 struct bpf_link *bpf_link_get_curr_or_next(u32 *id)
5394 {
5395         struct bpf_link *link;
5396
5397         spin_lock_bh(&link_idr_lock);
5398 again:
5399         link = idr_get_next(&link_idr, id);
5400         if (link) {
5401                 link = bpf_link_inc_not_zero(link);
5402                 if (IS_ERR(link)) {
5403                         (*id)++;
5404                         goto again;
5405                 }
5406         }
5407         spin_unlock_bh(&link_idr_lock);
5408
5409         return link;
5410 }
5411
5412 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
5413
5414 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
5415 {
5416         struct bpf_link *link;
5417         u32 id = attr->link_id;
5418         int fd;
5419
5420         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
5421                 return -EINVAL;
5422
5423         if (!capable(CAP_SYS_ADMIN))
5424                 return -EPERM;
5425
5426         link = bpf_link_by_id(id);
5427         if (IS_ERR(link))
5428                 return PTR_ERR(link);
5429
5430         fd = bpf_link_new_fd(link);
5431         if (fd < 0)
5432                 bpf_link_put_direct(link);
5433
5434         return fd;
5435 }
5436
5437 DEFINE_MUTEX(bpf_stats_enabled_mutex);
5438
5439 static int bpf_stats_release(struct inode *inode, struct file *file)
5440 {
5441         mutex_lock(&bpf_stats_enabled_mutex);
5442         static_key_slow_dec(&bpf_stats_enabled_key.key);
5443         mutex_unlock(&bpf_stats_enabled_mutex);
5444         return 0;
5445 }
5446
5447 static const struct file_operations bpf_stats_fops = {
5448         .release = bpf_stats_release,
5449 };
5450
5451 static int bpf_enable_runtime_stats(void)
5452 {
5453         int fd;
5454
5455         mutex_lock(&bpf_stats_enabled_mutex);
5456
5457         /* Set a very high limit to avoid overflow */
5458         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
5459                 mutex_unlock(&bpf_stats_enabled_mutex);
5460                 return -EBUSY;
5461         }
5462
5463         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
5464         if (fd >= 0)
5465                 static_key_slow_inc(&bpf_stats_enabled_key.key);
5466
5467         mutex_unlock(&bpf_stats_enabled_mutex);
5468         return fd;
5469 }
5470
5471 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
5472
5473 static int bpf_enable_stats(union bpf_attr *attr)
5474 {
5475
5476         if (CHECK_ATTR(BPF_ENABLE_STATS))
5477                 return -EINVAL;
5478
5479         if (!capable(CAP_SYS_ADMIN))
5480                 return -EPERM;
5481
5482         switch (attr->enable_stats.type) {
5483         case BPF_STATS_RUN_TIME:
5484                 return bpf_enable_runtime_stats();
5485         default:
5486                 break;
5487         }
5488         return -EINVAL;
5489 }
5490
5491 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
5492
5493 static int bpf_iter_create(union bpf_attr *attr)
5494 {
5495         struct bpf_link *link;
5496         int err;
5497
5498         if (CHECK_ATTR(BPF_ITER_CREATE))
5499                 return -EINVAL;
5500
5501         if (attr->iter_create.flags)
5502                 return -EINVAL;
5503
5504         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
5505         if (IS_ERR(link))
5506                 return PTR_ERR(link);
5507
5508         err = bpf_iter_new_fd(link);
5509         bpf_link_put_direct(link);
5510
5511         return err;
5512 }
5513
5514 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
5515
5516 static int bpf_prog_bind_map(union bpf_attr *attr)
5517 {
5518         struct bpf_prog *prog;
5519         struct bpf_map *map;
5520         struct bpf_map **used_maps_old, **used_maps_new;
5521         int i, ret = 0;
5522
5523         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
5524                 return -EINVAL;
5525
5526         if (attr->prog_bind_map.flags)
5527                 return -EINVAL;
5528
5529         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
5530         if (IS_ERR(prog))
5531                 return PTR_ERR(prog);
5532
5533         map = bpf_map_get(attr->prog_bind_map.map_fd);
5534         if (IS_ERR(map)) {
5535                 ret = PTR_ERR(map);
5536                 goto out_prog_put;
5537         }
5538
5539         mutex_lock(&prog->aux->used_maps_mutex);
5540
5541         used_maps_old = prog->aux->used_maps;
5542
5543         for (i = 0; i < prog->aux->used_map_cnt; i++)
5544                 if (used_maps_old[i] == map) {
5545                         bpf_map_put(map);
5546                         goto out_unlock;
5547                 }
5548
5549         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
5550                                       sizeof(used_maps_new[0]),
5551                                       GFP_KERNEL);
5552         if (!used_maps_new) {
5553                 ret = -ENOMEM;
5554                 goto out_unlock;
5555         }
5556
5557         /* The bpf program will not access the bpf map, but for the sake of
5558          * simplicity, increase sleepable_refcnt for sleepable program as well.
5559          */
5560         if (prog->sleepable)
5561                 atomic64_inc(&map->sleepable_refcnt);
5562         memcpy(used_maps_new, used_maps_old,
5563                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
5564         used_maps_new[prog->aux->used_map_cnt] = map;
5565
5566         prog->aux->used_map_cnt++;
5567         prog->aux->used_maps = used_maps_new;
5568
5569         kfree(used_maps_old);
5570
5571 out_unlock:
5572         mutex_unlock(&prog->aux->used_maps_mutex);
5573
5574         if (ret)
5575                 bpf_map_put(map);
5576 out_prog_put:
5577         bpf_prog_put(prog);
5578         return ret;
5579 }
5580
5581 #define BPF_TOKEN_CREATE_LAST_FIELD token_create.bpffs_fd
5582
5583 static int token_create(union bpf_attr *attr)
5584 {
5585         if (CHECK_ATTR(BPF_TOKEN_CREATE))
5586                 return -EINVAL;
5587
5588         /* no flags are supported yet */
5589         if (attr->token_create.flags)
5590                 return -EINVAL;
5591
5592         return bpf_token_create(attr);
5593 }
5594
5595 static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
5596 {
5597         union bpf_attr attr;
5598         int err;
5599
5600         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
5601         if (err)
5602                 return err;
5603         size = min_t(u32, size, sizeof(attr));
5604
5605         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
5606         memset(&attr, 0, sizeof(attr));
5607         if (copy_from_bpfptr(&attr, uattr, size) != 0)
5608                 return -EFAULT;
5609
5610         err = security_bpf(cmd, &attr, size);
5611         if (err < 0)
5612                 return err;
5613
5614         switch (cmd) {
5615         case BPF_MAP_CREATE:
5616                 err = map_create(&attr);
5617                 break;
5618         case BPF_MAP_LOOKUP_ELEM:
5619                 err = map_lookup_elem(&attr);
5620                 break;
5621         case BPF_MAP_UPDATE_ELEM:
5622                 err = map_update_elem(&attr, uattr);
5623                 break;
5624         case BPF_MAP_DELETE_ELEM:
5625                 err = map_delete_elem(&attr, uattr);
5626                 break;
5627         case BPF_MAP_GET_NEXT_KEY:
5628                 err = map_get_next_key(&attr);
5629                 break;
5630         case BPF_MAP_FREEZE:
5631                 err = map_freeze(&attr);
5632                 break;
5633         case BPF_PROG_LOAD:
5634                 err = bpf_prog_load(&attr, uattr, size);
5635                 break;
5636         case BPF_OBJ_PIN:
5637                 err = bpf_obj_pin(&attr);
5638                 break;
5639         case BPF_OBJ_GET:
5640                 err = bpf_obj_get(&attr);
5641                 break;
5642         case BPF_PROG_ATTACH:
5643                 err = bpf_prog_attach(&attr);
5644                 break;
5645         case BPF_PROG_DETACH:
5646                 err = bpf_prog_detach(&attr);
5647                 break;
5648         case BPF_PROG_QUERY:
5649                 err = bpf_prog_query(&attr, uattr.user);
5650                 break;
5651         case BPF_PROG_TEST_RUN:
5652                 err = bpf_prog_test_run(&attr, uattr.user);
5653                 break;
5654         case BPF_PROG_GET_NEXT_ID:
5655                 err = bpf_obj_get_next_id(&attr, uattr.user,
5656                                           &prog_idr, &prog_idr_lock);
5657                 break;
5658         case BPF_MAP_GET_NEXT_ID:
5659                 err = bpf_obj_get_next_id(&attr, uattr.user,
5660                                           &map_idr, &map_idr_lock);
5661                 break;
5662         case BPF_BTF_GET_NEXT_ID:
5663                 err = bpf_obj_get_next_id(&attr, uattr.user,
5664                                           &btf_idr, &btf_idr_lock);
5665                 break;
5666         case BPF_PROG_GET_FD_BY_ID:
5667                 err = bpf_prog_get_fd_by_id(&attr);
5668                 break;
5669         case BPF_MAP_GET_FD_BY_ID:
5670                 err = bpf_map_get_fd_by_id(&attr);
5671                 break;
5672         case BPF_OBJ_GET_INFO_BY_FD:
5673                 err = bpf_obj_get_info_by_fd(&attr, uattr.user);
5674                 break;
5675         case BPF_RAW_TRACEPOINT_OPEN:
5676                 err = bpf_raw_tracepoint_open(&attr);
5677                 break;
5678         case BPF_BTF_LOAD:
5679                 err = bpf_btf_load(&attr, uattr, size);
5680                 break;
5681         case BPF_BTF_GET_FD_BY_ID:
5682                 err = bpf_btf_get_fd_by_id(&attr);
5683                 break;
5684         case BPF_TASK_FD_QUERY:
5685                 err = bpf_task_fd_query(&attr, uattr.user);
5686                 break;
5687         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
5688                 err = map_lookup_and_delete_elem(&attr);
5689                 break;
5690         case BPF_MAP_LOOKUP_BATCH:
5691                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_LOOKUP_BATCH);
5692                 break;
5693         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
5694                 err = bpf_map_do_batch(&attr, uattr.user,
5695                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
5696                 break;
5697         case BPF_MAP_UPDATE_BATCH:
5698                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_UPDATE_BATCH);
5699                 break;
5700         case BPF_MAP_DELETE_BATCH:
5701                 err = bpf_map_do_batch(&attr, uattr.user, BPF_MAP_DELETE_BATCH);
5702                 break;
5703         case BPF_LINK_CREATE:
5704                 err = link_create(&attr, uattr);
5705                 break;
5706         case BPF_LINK_UPDATE:
5707                 err = link_update(&attr);
5708                 break;
5709         case BPF_LINK_GET_FD_BY_ID:
5710                 err = bpf_link_get_fd_by_id(&attr);
5711                 break;
5712         case BPF_LINK_GET_NEXT_ID:
5713                 err = bpf_obj_get_next_id(&attr, uattr.user,
5714                                           &link_idr, &link_idr_lock);
5715                 break;
5716         case BPF_ENABLE_STATS:
5717                 err = bpf_enable_stats(&attr);
5718                 break;
5719         case BPF_ITER_CREATE:
5720                 err = bpf_iter_create(&attr);
5721                 break;
5722         case BPF_LINK_DETACH:
5723                 err = link_detach(&attr);
5724                 break;
5725         case BPF_PROG_BIND_MAP:
5726                 err = bpf_prog_bind_map(&attr);
5727                 break;
5728         case BPF_TOKEN_CREATE:
5729                 err = token_create(&attr);
5730                 break;
5731         default:
5732                 err = -EINVAL;
5733                 break;
5734         }
5735
5736         return err;
5737 }
5738
5739 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
5740 {
5741         return __sys_bpf(cmd, USER_BPFPTR(uattr), size);
5742 }
5743
5744 static bool syscall_prog_is_valid_access(int off, int size,
5745                                          enum bpf_access_type type,
5746                                          const struct bpf_prog *prog,
5747                                          struct bpf_insn_access_aux *info)
5748 {
5749         if (off < 0 || off >= U16_MAX)
5750                 return false;
5751         if (off % size != 0)
5752                 return false;
5753         return true;
5754 }
5755
5756 BPF_CALL_3(bpf_sys_bpf, int, cmd, union bpf_attr *, attr, u32, attr_size)
5757 {
5758         switch (cmd) {
5759         case BPF_MAP_CREATE:
5760         case BPF_MAP_DELETE_ELEM:
5761         case BPF_MAP_UPDATE_ELEM:
5762         case BPF_MAP_FREEZE:
5763         case BPF_MAP_GET_FD_BY_ID:
5764         case BPF_PROG_LOAD:
5765         case BPF_BTF_LOAD:
5766         case BPF_LINK_CREATE:
5767         case BPF_RAW_TRACEPOINT_OPEN:
5768                 break;
5769         default:
5770                 return -EINVAL;
5771         }
5772         return __sys_bpf(cmd, KERNEL_BPFPTR(attr), attr_size);
5773 }
5774
5775
5776 /* To shut up -Wmissing-prototypes.
5777  * This function is used by the kernel light skeleton
5778  * to load bpf programs when modules are loaded or during kernel boot.
5779  * See tools/lib/bpf/skel_internal.h
5780  */
5781 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size);
5782
5783 int kern_sys_bpf(int cmd, union bpf_attr *attr, unsigned int size)
5784 {
5785         struct bpf_prog * __maybe_unused prog;
5786         struct bpf_tramp_run_ctx __maybe_unused run_ctx;
5787
5788         switch (cmd) {
5789 #ifdef CONFIG_BPF_JIT /* __bpf_prog_enter_sleepable used by trampoline and JIT */
5790         case BPF_PROG_TEST_RUN:
5791                 if (attr->test.data_in || attr->test.data_out ||
5792                     attr->test.ctx_out || attr->test.duration ||
5793                     attr->test.repeat || attr->test.flags)
5794                         return -EINVAL;
5795
5796                 prog = bpf_prog_get_type(attr->test.prog_fd, BPF_PROG_TYPE_SYSCALL);
5797                 if (IS_ERR(prog))
5798                         return PTR_ERR(prog);
5799
5800                 if (attr->test.ctx_size_in < prog->aux->max_ctx_offset ||
5801                     attr->test.ctx_size_in > U16_MAX) {
5802                         bpf_prog_put(prog);
5803                         return -EINVAL;
5804                 }
5805
5806                 run_ctx.bpf_cookie = 0;
5807                 if (!__bpf_prog_enter_sleepable_recur(prog, &run_ctx)) {
5808                         /* recursion detected */
5809                         __bpf_prog_exit_sleepable_recur(prog, 0, &run_ctx);
5810                         bpf_prog_put(prog);
5811                         return -EBUSY;
5812                 }
5813                 attr->test.retval = bpf_prog_run(prog, (void *) (long) attr->test.ctx_in);
5814                 __bpf_prog_exit_sleepable_recur(prog, 0 /* bpf_prog_run does runtime stats */,
5815                                                 &run_ctx);
5816                 bpf_prog_put(prog);
5817                 return 0;
5818 #endif
5819         default:
5820                 return ____bpf_sys_bpf(cmd, attr, size);
5821         }
5822 }
5823 EXPORT_SYMBOL(kern_sys_bpf);
5824
5825 static const struct bpf_func_proto bpf_sys_bpf_proto = {
5826         .func           = bpf_sys_bpf,
5827         .gpl_only       = false,
5828         .ret_type       = RET_INTEGER,
5829         .arg1_type      = ARG_ANYTHING,
5830         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5831         .arg3_type      = ARG_CONST_SIZE,
5832 };
5833
5834 const struct bpf_func_proto * __weak
5835 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5836 {
5837         return bpf_base_func_proto(func_id, prog);
5838 }
5839
5840 BPF_CALL_1(bpf_sys_close, u32, fd)
5841 {
5842         /* When bpf program calls this helper there should not be
5843          * an fdget() without matching completed fdput().
5844          * This helper is allowed in the following callchain only:
5845          * sys_bpf->prog_test_run->bpf_prog->bpf_sys_close
5846          */
5847         return close_fd(fd);
5848 }
5849
5850 static const struct bpf_func_proto bpf_sys_close_proto = {
5851         .func           = bpf_sys_close,
5852         .gpl_only       = false,
5853         .ret_type       = RET_INTEGER,
5854         .arg1_type      = ARG_ANYTHING,
5855 };
5856
5857 BPF_CALL_4(bpf_kallsyms_lookup_name, const char *, name, int, name_sz, int, flags, u64 *, res)
5858 {
5859         *res = 0;
5860         if (flags)
5861                 return -EINVAL;
5862
5863         if (name_sz <= 1 || name[name_sz - 1])
5864                 return -EINVAL;
5865
5866         if (!bpf_dump_raw_ok(current_cred()))
5867                 return -EPERM;
5868
5869         *res = kallsyms_lookup_name(name);
5870         return *res ? 0 : -ENOENT;
5871 }
5872
5873 static const struct bpf_func_proto bpf_kallsyms_lookup_name_proto = {
5874         .func           = bpf_kallsyms_lookup_name,
5875         .gpl_only       = false,
5876         .ret_type       = RET_INTEGER,
5877         .arg1_type      = ARG_PTR_TO_MEM,
5878         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
5879         .arg3_type      = ARG_ANYTHING,
5880         .arg4_type      = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_ALIGNED,
5881         .arg4_size      = sizeof(u64),
5882 };
5883
5884 static const struct bpf_func_proto *
5885 syscall_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5886 {
5887         switch (func_id) {
5888         case BPF_FUNC_sys_bpf:
5889                 return !bpf_token_capable(prog->aux->token, CAP_PERFMON)
5890                        ? NULL : &bpf_sys_bpf_proto;
5891         case BPF_FUNC_btf_find_by_name_kind:
5892                 return &bpf_btf_find_by_name_kind_proto;
5893         case BPF_FUNC_sys_close:
5894                 return &bpf_sys_close_proto;
5895         case BPF_FUNC_kallsyms_lookup_name:
5896                 return &bpf_kallsyms_lookup_name_proto;
5897         default:
5898                 return tracing_prog_func_proto(func_id, prog);
5899         }
5900 }
5901
5902 const struct bpf_verifier_ops bpf_syscall_verifier_ops = {
5903         .get_func_proto  = syscall_prog_func_proto,
5904         .is_valid_access = syscall_prog_is_valid_access,
5905 };
5906
5907 const struct bpf_prog_ops bpf_syscall_prog_ops = {
5908         .test_run = bpf_prog_test_run_syscall,
5909 };
5910
5911 #ifdef CONFIG_SYSCTL
5912 static int bpf_stats_handler(const struct ctl_table *table, int write,
5913                              void *buffer, size_t *lenp, loff_t *ppos)
5914 {
5915         struct static_key *key = (struct static_key *)table->data;
5916         static int saved_val;
5917         int val, ret;
5918         struct ctl_table tmp = {
5919                 .data   = &val,
5920                 .maxlen = sizeof(val),
5921                 .mode   = table->mode,
5922                 .extra1 = SYSCTL_ZERO,
5923                 .extra2 = SYSCTL_ONE,
5924         };
5925
5926         if (write && !capable(CAP_SYS_ADMIN))
5927                 return -EPERM;
5928
5929         mutex_lock(&bpf_stats_enabled_mutex);
5930         val = saved_val;
5931         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5932         if (write && !ret && val != saved_val) {
5933                 if (val)
5934                         static_key_slow_inc(key);
5935                 else
5936                         static_key_slow_dec(key);
5937                 saved_val = val;
5938         }
5939         mutex_unlock(&bpf_stats_enabled_mutex);
5940         return ret;
5941 }
5942
5943 void __weak unpriv_ebpf_notify(int new_state)
5944 {
5945 }
5946
5947 static int bpf_unpriv_handler(const struct ctl_table *table, int write,
5948                               void *buffer, size_t *lenp, loff_t *ppos)
5949 {
5950         int ret, unpriv_enable = *(int *)table->data;
5951         bool locked_state = unpriv_enable == 1;
5952         struct ctl_table tmp = *table;
5953
5954         if (write && !capable(CAP_SYS_ADMIN))
5955                 return -EPERM;
5956
5957         tmp.data = &unpriv_enable;
5958         ret = proc_dointvec_minmax(&tmp, write, buffer, lenp, ppos);
5959         if (write && !ret) {
5960                 if (locked_state && unpriv_enable != 1)
5961                         return -EPERM;
5962                 *(int *)table->data = unpriv_enable;
5963         }
5964
5965         if (write)
5966                 unpriv_ebpf_notify(unpriv_enable);
5967
5968         return ret;
5969 }
5970
5971 static struct ctl_table bpf_syscall_table[] = {
5972         {
5973                 .procname       = "unprivileged_bpf_disabled",
5974                 .data           = &sysctl_unprivileged_bpf_disabled,
5975                 .maxlen         = sizeof(sysctl_unprivileged_bpf_disabled),
5976                 .mode           = 0644,
5977                 .proc_handler   = bpf_unpriv_handler,
5978                 .extra1         = SYSCTL_ZERO,
5979                 .extra2         = SYSCTL_TWO,
5980         },
5981         {
5982                 .procname       = "bpf_stats_enabled",
5983                 .data           = &bpf_stats_enabled_key.key,
5984                 .mode           = 0644,
5985                 .proc_handler   = bpf_stats_handler,
5986         },
5987 };
5988
5989 static int __init bpf_syscall_sysctl_init(void)
5990 {
5991         register_sysctl_init("kernel", bpf_syscall_table);
5992         return 0;
5993 }
5994 late_initcall(bpf_syscall_sysctl_init);
5995 #endif /* CONFIG_SYSCTL */