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_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26 #include <linux/audit.h>
  27 #include <uapi/linux/btf.h>
  28 #include <linux/bpf_lsm.h>
  29
  30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  31                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  32                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  33 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  36                         IS_FD_HASH(map))
  37
  38 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  39
  40 DEFINE_PER_CPU(int, bpf_prog_active);
  41 static DEFINE_IDR(prog_idr);
  42 static DEFINE_SPINLOCK(prog_idr_lock);
  43 static DEFINE_IDR(map_idr);
  44 static DEFINE_SPINLOCK(map_idr_lock);
  45
  46 int sysctl_unprivileged_bpf_disabled __read_mostly;
  47
  48 static const struct bpf_map_ops * const bpf_map_types[] = {
  49 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  50 #define BPF_MAP_TYPE(_id, _ops) \
  51         [_id] = &_ops,
  52 #include <linux/bpf_types.h>
  53 #undef BPF_PROG_TYPE
  54 #undef BPF_MAP_TYPE
  55 };
  56
  57 /*
  58  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  59  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  60  * we don't know about yet.
  61  *
  62  * There is a ToCToU between this function call and the following
  63  * copy_from_user() call. However, this is not a concern since this function is
  64  * meant to be a future-proofing of bits.
  65  */
  66 int bpf_check_uarg_tail_zero(void __user *uaddr,
  67                              size_t expected_size,
  68                              size_t actual_size)
  69 {
  70         unsigned char __user *addr;
  71         unsigned char __user *end;
  72         unsigned char val;
  73         int err;
  74
  75         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  76                 return -E2BIG;
  77
  78         if (unlikely(!access_ok(uaddr, actual_size)))
  79                 return -EFAULT;
  80
  81         if (actual_size <= expected_size)
  82                 return 0;
  83
  84         addr = uaddr + expected_size;
  85         end  = uaddr + actual_size;
  86
  87         for (; addr < end; addr++) {
  88                 err = get_user(val, addr);
  89                 if (err)
  90                         return err;
  91                 if (val)
  92                         return -E2BIG;
  93         }
  94
  95         return 0;
  96 }
  97
  98 const struct bpf_map_ops bpf_map_offload_ops = {
  99         .map_alloc = bpf_map_offload_map_alloc,
 100         .map_free = bpf_map_offload_map_free,
 101         .map_check_btf = map_check_no_btf,
 102 };
 103
 104 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 105 {
 106         const struct bpf_map_ops *ops;
 107         u32 type = attr->map_type;
 108         struct bpf_map *map;
 109         int err;
 110
 111         if (type >= ARRAY_SIZE(bpf_map_types))
 112                 return ERR_PTR(-EINVAL);
 113         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 114         ops = bpf_map_types[type];
 115         if (!ops)
 116                 return ERR_PTR(-EINVAL);
 117
 118         if (ops->map_alloc_check) {
 119                 err = ops->map_alloc_check(attr);
 120                 if (err)
 121                         return ERR_PTR(err);
 122         }
 123         if (attr->map_ifindex)
 124                 ops = &bpf_map_offload_ops;
 125         map = ops->map_alloc(attr);
 126         if (IS_ERR(map))
 127                 return map;
 128         map->ops = ops;
 129         map->map_type = type;
 130         return map;
 131 }
 132
 133 static u32 bpf_map_value_size(struct bpf_map *map)
 134 {
 135         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 136             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 137             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 138             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 139                 return round_up(map->value_size, 8) * num_possible_cpus();
 140         else if (IS_FD_MAP(map))
 141                 return sizeof(u32);
 142         else
 143                 return  map->value_size;
 144 }
 145
 146 static void maybe_wait_bpf_programs(struct bpf_map *map)
 147 {
 148         /* Wait for any running BPF programs to complete so that
 149          * userspace, when we return to it, knows that all programs
 150          * that could be running use the new map value.
 151          */
 152         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 153             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 154                 synchronize_rcu();
 155 }
 156
 157 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
 158                                 void *value, __u64 flags)
 159 {
 160         int err;
 161
 162         /* Need to create a kthread, thus must support schedule */
 163         if (bpf_map_is_dev_bound(map)) {
 164                 return bpf_map_offload_update_elem(map, key, value, flags);
 165         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 166                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 167                    map->map_type == BPF_MAP_TYPE_SOCKMAP ||
 168                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 169                 return map->ops->map_update_elem(map, key, value, flags);
 170         } else if (IS_FD_PROG_ARRAY(map)) {
 171                 return bpf_fd_array_map_update_elem(map, f.file, key, value,
 172                                                     flags);
 173         }
 174
 175         bpf_disable_instrumentation();
 176         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 177             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 178                 err = bpf_percpu_hash_update(map, key, value, flags);
 179         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 180                 err = bpf_percpu_array_update(map, key, value, flags);
 181         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 182                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 183                                                        flags);
 184         } else if (IS_FD_ARRAY(map)) {
 185                 rcu_read_lock();
 186                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 187                                                    flags);
 188                 rcu_read_unlock();
 189         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 190                 rcu_read_lock();
 191                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 192                                                   flags);
 193                 rcu_read_unlock();
 194         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 195                 /* rcu_read_lock() is not needed */
 196                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 197                                                          flags);
 198         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 199                    map->map_type == BPF_MAP_TYPE_STACK) {
 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         maybe_wait_bpf_programs(map);
 208
 209         return err;
 210 }
 211
 212 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 213                               __u64 flags)
 214 {
 215         void *ptr;
 216         int err;
 217
 218         if (bpf_map_is_dev_bound(map))
 219                 return bpf_map_offload_lookup_elem(map, key, value);
 220
 221         bpf_disable_instrumentation();
 222         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 223             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 224                 err = bpf_percpu_hash_copy(map, key, value);
 225         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 226                 err = bpf_percpu_array_copy(map, key, value);
 227         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 228                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 229         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 230                 err = bpf_stackmap_copy(map, key, value);
 231         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 232                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 233         } else if (IS_FD_HASH(map)) {
 234                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 235         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 236                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 237         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 238                    map->map_type == BPF_MAP_TYPE_STACK) {
 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, since value wasn't zero inited */
 261                         check_and_init_map_lock(map, value);
 262                 }
 263                 rcu_read_unlock();
 264         }
 265
 266         bpf_enable_instrumentation();
 267         maybe_wait_bpf_programs(map);
 268
 269         return err;
 270 }
 271
 272 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 273 {
 274         /* We really just want to fail instead of triggering OOM killer
 275          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 276          * which is used for lower order allocation requests.
 277          *
 278          * It has been observed that higher order allocation requests done by
 279          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 280          * to reclaim memory from the page cache, thus we set
 281          * __GFP_RETRY_MAYFAIL to avoid such situations.
 282          */
 283
 284         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 285         void *area;
 286
 287         if (size >= SIZE_MAX)
 288                 return NULL;
 289
 290         /* kmalloc()'ed memory can't be mmap()'ed */
 291         if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 292                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 293                                     numa_node);
 294                 if (area != NULL)
 295                         return area;
 296         }
 297         if (mmapable) {
 298                 BUG_ON(!PAGE_ALIGNED(size));
 299                 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
 300                                                __GFP_RETRY_MAYFAIL | flags);
 301         }
 302         return __vmalloc_node_flags_caller(size, numa_node,
 303                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 304                                            flags, __builtin_return_address(0));
 305 }
 306
 307 void *bpf_map_area_alloc(u64 size, int numa_node)
 308 {
 309         return __bpf_map_area_alloc(size, numa_node, false);
 310 }
 311
 312 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 313 {
 314         return __bpf_map_area_alloc(size, numa_node, true);
 315 }
 316
 317 void bpf_map_area_free(void *area)
 318 {
 319         kvfree(area);
 320 }
 321
 322 static u32 bpf_map_flags_retain_permanent(u32 flags)
 323 {
 324         /* Some map creation flags are not tied to the map object but
 325          * rather to the map fd instead, so they have no meaning upon
 326          * map object inspection since multiple file descriptors with
 327          * different (access) properties can exist here. Thus, given
 328          * this has zero meaning for the map itself, lets clear these
 329          * from here.
 330          */
 331         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 332 }
 333
 334 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 335 {
 336         map->map_type = attr->map_type;
 337         map->key_size = attr->key_size;
 338         map->value_size = attr->value_size;
 339         map->max_entries = attr->max_entries;
 340         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 341         map->numa_node = bpf_map_attr_numa_node(attr);
 342 }
 343
 344 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 345 {
 346         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 347
 348         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 349                 atomic_long_sub(pages, &user->locked_vm);
 350                 return -EPERM;
 351         }
 352         return 0;
 353 }
 354
 355 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 356 {
 357         if (user)
 358                 atomic_long_sub(pages, &user->locked_vm);
 359 }
 360
 361 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
 362 {
 363         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 364         struct user_struct *user;
 365         int ret;
 366
 367         if (size >= U32_MAX - PAGE_SIZE)
 368                 return -E2BIG;
 369
 370         user = get_current_user();
 371         ret = bpf_charge_memlock(user, pages);
 372         if (ret) {
 373                 free_uid(user);
 374                 return ret;
 375         }
 376
 377         mem->pages = pages;
 378         mem->user = user;
 379
 380         return 0;
 381 }
 382
 383 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 384 {
 385         bpf_uncharge_memlock(mem->user, mem->pages);
 386         free_uid(mem->user);
 387 }
 388
 389 void bpf_map_charge_move(struct bpf_map_memory *dst,
 390                          struct bpf_map_memory *src)
 391 {
 392         *dst = *src;
 393
 394         /* Make sure src will not be used for the redundant uncharging. */
 395         memset(src, 0, sizeof(struct bpf_map_memory));
 396 }
 397
 398 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 399 {
 400         int ret;
 401
 402         ret = bpf_charge_memlock(map->memory.user, pages);
 403         if (ret)
 404                 return ret;
 405         map->memory.pages += pages;
 406         return ret;
 407 }
 408
 409 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 410 {
 411         bpf_uncharge_memlock(map->memory.user, pages);
 412         map->memory.pages -= pages;
 413 }
 414
 415 static int bpf_map_alloc_id(struct bpf_map *map)
 416 {
 417         int id;
 418
 419         idr_preload(GFP_KERNEL);
 420         spin_lock_bh(&map_idr_lock);
 421         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 422         if (id > 0)
 423                 map->id = id;
 424         spin_unlock_bh(&map_idr_lock);
 425         idr_preload_end();
 426
 427         if (WARN_ON_ONCE(!id))
 428                 return -ENOSPC;
 429
 430         return id > 0 ? 0 : id;
 431 }
 432
 433 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 434 {
 435         unsigned long flags;
 436
 437         /* Offloaded maps are removed from the IDR store when their device
 438          * disappears - even if someone holds an fd to them they are unusable,
 439          * the memory is gone, all ops will fail; they are simply waiting for
 440          * refcnt to drop to be freed.
 441          */
 442         if (!map->id)
 443                 return;
 444
 445         if (do_idr_lock)
 446                 spin_lock_irqsave(&map_idr_lock, flags);
 447         else
 448                 __acquire(&map_idr_lock);
 449
 450         idr_remove(&map_idr, map->id);
 451         map->id = 0;
 452
 453         if (do_idr_lock)
 454                 spin_unlock_irqrestore(&map_idr_lock, flags);
 455         else
 456                 __release(&map_idr_lock);
 457 }
 458
 459 /* called from workqueue */
 460 static void bpf_map_free_deferred(struct work_struct *work)
 461 {
 462         struct bpf_map *map = container_of(work, struct bpf_map, work);
 463         struct bpf_map_memory mem;
 464
 465         bpf_map_charge_move(&mem, &map->memory);
 466         security_bpf_map_free(map);
 467         /* implementation dependent freeing */
 468         map->ops->map_free(map);
 469         bpf_map_charge_finish(&mem);
 470 }
 471
 472 static void bpf_map_put_uref(struct bpf_map *map)
 473 {
 474         if (atomic64_dec_and_test(&map->usercnt)) {
 475                 if (map->ops->map_release_uref)
 476                         map->ops->map_release_uref(map);
 477         }
 478 }
 479
 480 /* decrement map refcnt and schedule it for freeing via workqueue
 481  * (unrelying map implementation ops->map_free() might sleep)
 482  */
 483 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 484 {
 485         if (atomic64_dec_and_test(&map->refcnt)) {
 486                 /* bpf_map_free_id() must be called first */
 487                 bpf_map_free_id(map, do_idr_lock);
 488                 btf_put(map->btf);
 489                 INIT_WORK(&map->work, bpf_map_free_deferred);
 490                 schedule_work(&map->work);
 491         }
 492 }
 493
 494 void bpf_map_put(struct bpf_map *map)
 495 {
 496         __bpf_map_put(map, true);
 497 }
 498 EXPORT_SYMBOL_GPL(bpf_map_put);
 499
 500 void bpf_map_put_with_uref(struct bpf_map *map)
 501 {
 502         bpf_map_put_uref(map);
 503         bpf_map_put(map);
 504 }
 505
 506 static int bpf_map_release(struct inode *inode, struct file *filp)
 507 {
 508         struct bpf_map *map = filp->private_data;
 509
 510         if (map->ops->map_release)
 511                 map->ops->map_release(map, filp);
 512
 513         bpf_map_put_with_uref(map);
 514         return 0;
 515 }
 516
 517 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 518 {
 519         fmode_t mode = f.file->f_mode;
 520
 521         /* Our file permissions may have been overridden by global
 522          * map permissions facing syscall side.
 523          */
 524         if (READ_ONCE(map->frozen))
 525                 mode &= ~FMODE_CAN_WRITE;
 526         return mode;
 527 }
 528
 529 #ifdef CONFIG_PROC_FS
 530 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 531 {
 532         const struct bpf_map *map = filp->private_data;
 533         const struct bpf_array *array;
 534         u32 type = 0, jited = 0;
 535
 536         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 537                 array = container_of(map, struct bpf_array, map);
 538                 type  = array->aux->type;
 539                 jited = array->aux->jited;
 540         }
 541
 542         seq_printf(m,
 543                    "map_type:\t%u\n"
 544                    "key_size:\t%u\n"
 545                    "value_size:\t%u\n"
 546                    "max_entries:\t%u\n"
 547                    "map_flags:\t%#x\n"
 548                    "memlock:\t%llu\n"
 549                    "map_id:\t%u\n"
 550                    "frozen:\t%u\n",
 551                    map->map_type,
 552                    map->key_size,
 553                    map->value_size,
 554                    map->max_entries,
 555                    map->map_flags,
 556                    map->memory.pages * 1ULL << PAGE_SHIFT,
 557                    map->id,
 558                    READ_ONCE(map->frozen));
 559         if (type) {
 560                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 561                 seq_printf(m, "owner_jited:\t%u\n", jited);
 562         }
 563 }
 564 #endif
 565
 566 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 567                               loff_t *ppos)
 568 {
 569         /* We need this handler such that alloc_file() enables
 570          * f_mode with FMODE_CAN_READ.
 571          */
 572         return -EINVAL;
 573 }
 574
 575 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 576                                size_t siz, loff_t *ppos)
 577 {
 578         /* We need this handler such that alloc_file() enables
 579          * f_mode with FMODE_CAN_WRITE.
 580          */
 581         return -EINVAL;
 582 }
 583
 584 /* called for any extra memory-mapped regions (except initial) */
 585 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 586 {
 587         struct bpf_map *map = vma->vm_file->private_data;
 588
 589         bpf_map_inc_with_uref(map);
 590
 591         if (vma->vm_flags & VM_WRITE) {
 592                 mutex_lock(&map->freeze_mutex);
 593                 map->writecnt++;
 594                 mutex_unlock(&map->freeze_mutex);
 595         }
 596 }
 597
 598 /* called for all unmapped memory region (including initial) */
 599 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 600 {
 601         struct bpf_map *map = vma->vm_file->private_data;
 602
 603         if (vma->vm_flags & VM_WRITE) {
 604                 mutex_lock(&map->freeze_mutex);
 605                 map->writecnt--;
 606                 mutex_unlock(&map->freeze_mutex);
 607         }
 608
 609         bpf_map_put_with_uref(map);
 610 }
 611
 612 static const struct vm_operations_struct bpf_map_default_vmops = {
 613         .open           = bpf_map_mmap_open,
 614         .close          = bpf_map_mmap_close,
 615 };
 616
 617 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 618 {
 619         struct bpf_map *map = filp->private_data;
 620         int err;
 621
 622         if (!map->ops->map_mmap || map_value_has_spin_lock(map))
 623                 return -ENOTSUPP;
 624
 625         if (!(vma->vm_flags & VM_SHARED))
 626                 return -EINVAL;
 627
 628         mutex_lock(&map->freeze_mutex);
 629
 630         if ((vma->vm_flags & VM_WRITE) && map->frozen) {
 631                 err = -EPERM;
 632                 goto out;
 633         }
 634
 635         /* set default open/close callbacks */
 636         vma->vm_ops = &bpf_map_default_vmops;
 637         vma->vm_private_data = map;
 638
 639         err = map->ops->map_mmap(map, vma);
 640         if (err)
 641                 goto out;
 642
 643         bpf_map_inc_with_uref(map);
 644
 645         if (vma->vm_flags & VM_WRITE)
 646                 map->writecnt++;
 647 out:
 648         mutex_unlock(&map->freeze_mutex);
 649         return err;
 650 }
 651
 652 const struct file_operations bpf_map_fops = {
 653 #ifdef CONFIG_PROC_FS
 654         .show_fdinfo    = bpf_map_show_fdinfo,
 655 #endif
 656         .release        = bpf_map_release,
 657         .read           = bpf_dummy_read,
 658         .write          = bpf_dummy_write,
 659         .mmap           = bpf_map_mmap,
 660 };
 661
 662 int bpf_map_new_fd(struct bpf_map *map, int flags)
 663 {
 664         int ret;
 665
 666         ret = security_bpf_map(map, OPEN_FMODE(flags));
 667         if (ret < 0)
 668                 return ret;
 669
 670         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 671                                 flags | O_CLOEXEC);
 672 }
 673
 674 int bpf_get_file_flag(int flags)
 675 {
 676         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 677                 return -EINVAL;
 678         if (flags & BPF_F_RDONLY)
 679                 return O_RDONLY;
 680         if (flags & BPF_F_WRONLY)
 681                 return O_WRONLY;
 682         return O_RDWR;
 683 }
 684
 685 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 686 #define CHECK_ATTR(CMD) \
 687         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 688                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 689                    sizeof(*attr) - \
 690                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 691                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 692
 693 /* dst and src must have at least "size" number of bytes.
 694  * Return strlen on success and < 0 on error.
 695  */
 696 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 697 {
 698         const char *end = src + size;
 699         const char *orig_src = src;
 700
 701         memset(dst, 0, size);
 702         /* Copy all isalnum(), '_' and '.' chars. */
 703         while (src < end && *src) {
 704                 if (!isalnum(*src) &&
 705                     *src != '_' && *src != '.')
 706                         return -EINVAL;
 707                 *dst++ = *src++;
 708         }
 709
 710         /* No '\0' found in "size" number of bytes */
 711         if (src == end)
 712                 return -EINVAL;
 713
 714         return src - orig_src;
 715 }
 716
 717 int map_check_no_btf(const struct bpf_map *map,
 718                      const struct btf *btf,
 719                      const struct btf_type *key_type,
 720                      const struct btf_type *value_type)
 721 {
 722         return -ENOTSUPP;
 723 }
 724
 725 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 726                          u32 btf_key_id, u32 btf_value_id)
 727 {
 728         const struct btf_type *key_type, *value_type;
 729         u32 key_size, value_size;
 730         int ret = 0;
 731
 732         /* Some maps allow key to be unspecified. */
 733         if (btf_key_id) {
 734                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 735                 if (!key_type || key_size != map->key_size)
 736                         return -EINVAL;
 737         } else {
 738                 key_type = btf_type_by_id(btf, 0);
 739                 if (!map->ops->map_check_btf)
 740                         return -EINVAL;
 741         }
 742
 743         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 744         if (!value_type || value_size != map->value_size)
 745                 return -EINVAL;
 746
 747         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 748
 749         if (map_value_has_spin_lock(map)) {
 750                 if (map->map_flags & BPF_F_RDONLY_PROG)
 751                         return -EACCES;
 752                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 753                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 754                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 755                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 756                         return -ENOTSUPP;
 757                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 758                     map->value_size) {
 759                         WARN_ONCE(1,
 760                                   "verifier bug spin_lock_off %d value_size %d\n",
 761                                   map->spin_lock_off, map->value_size);
 762                         return -EFAULT;
 763                 }
 764         }
 765
 766         if (map->ops->map_check_btf)
 767                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 768
 769         return ret;
 770 }
 771
 772 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
 773 /* called via syscall */
 774 static int map_create(union bpf_attr *attr)
 775 {
 776         int numa_node = bpf_map_attr_numa_node(attr);
 777         struct bpf_map_memory mem;
 778         struct bpf_map *map;
 779         int f_flags;
 780         int err;
 781
 782         err = CHECK_ATTR(BPF_MAP_CREATE);
 783         if (err)
 784                 return -EINVAL;
 785
 786         if (attr->btf_vmlinux_value_type_id) {
 787                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
 788                     attr->btf_key_type_id || attr->btf_value_type_id)
 789                         return -EINVAL;
 790         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
 791                 return -EINVAL;
 792         }
 793
 794         f_flags = bpf_get_file_flag(attr->map_flags);
 795         if (f_flags < 0)
 796                 return f_flags;
 797
 798         if (numa_node != NUMA_NO_NODE &&
 799             ((unsigned int)numa_node >= nr_node_ids ||
 800              !node_online(numa_node)))
 801                 return -EINVAL;
 802
 803         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 804         map = find_and_alloc_map(attr);
 805         if (IS_ERR(map))
 806                 return PTR_ERR(map);
 807
 808         err = bpf_obj_name_cpy(map->name, attr->map_name,
 809                                sizeof(attr->map_name));
 810         if (err < 0)
 811                 goto free_map;
 812
 813         atomic64_set(&map->refcnt, 1);
 814         atomic64_set(&map->usercnt, 1);
 815         mutex_init(&map->freeze_mutex);
 816
 817         map->spin_lock_off = -EINVAL;
 818         if (attr->btf_key_type_id || attr->btf_value_type_id ||
 819             /* Even the map's value is a kernel's struct,
 820              * the bpf_prog.o must have BTF to begin with
 821              * to figure out the corresponding kernel's
 822              * counter part.  Thus, attr->btf_fd has
 823              * to be valid also.
 824              */
 825             attr->btf_vmlinux_value_type_id) {
 826                 struct btf *btf;
 827
 828                 btf = btf_get_by_fd(attr->btf_fd);
 829                 if (IS_ERR(btf)) {
 830                         err = PTR_ERR(btf);
 831                         goto free_map;
 832                 }
 833                 map->btf = btf;
 834
 835                 if (attr->btf_value_type_id) {
 836                         err = map_check_btf(map, btf, attr->btf_key_type_id,
 837                                             attr->btf_value_type_id);
 838                         if (err)
 839                                 goto free_map;
 840                 }
 841
 842                 map->btf_key_type_id = attr->btf_key_type_id;
 843                 map->btf_value_type_id = attr->btf_value_type_id;
 844                 map->btf_vmlinux_value_type_id =
 845                         attr->btf_vmlinux_value_type_id;
 846         }
 847
 848         err = security_bpf_map_alloc(map);
 849         if (err)
 850                 goto free_map;
 851
 852         err = bpf_map_alloc_id(map);
 853         if (err)
 854                 goto free_map_sec;
 855
 856         err = bpf_map_new_fd(map, f_flags);
 857         if (err < 0) {
 858                 /* failed to allocate fd.
 859                  * bpf_map_put_with_uref() is needed because the above
 860                  * bpf_map_alloc_id() has published the map
 861                  * to the userspace and the userspace may
 862                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 863                  */
 864                 bpf_map_put_with_uref(map);
 865                 return err;
 866         }
 867
 868         return err;
 869
 870 free_map_sec:
 871         security_bpf_map_free(map);
 872 free_map:
 873         btf_put(map->btf);
 874         bpf_map_charge_move(&mem, &map->memory);
 875         map->ops->map_free(map);
 876         bpf_map_charge_finish(&mem);
 877         return err;
 878 }
 879
 880 /* if error is returned, fd is released.
 881  * On success caller should complete fd access with matching fdput()
 882  */
 883 struct bpf_map *__bpf_map_get(struct fd f)
 884 {
 885         if (!f.file)
 886                 return ERR_PTR(-EBADF);
 887         if (f.file->f_op != &bpf_map_fops) {
 888                 fdput(f);
 889                 return ERR_PTR(-EINVAL);
 890         }
 891
 892         return f.file->private_data;
 893 }
 894
 895 void bpf_map_inc(struct bpf_map *map)
 896 {
 897         atomic64_inc(&map->refcnt);
 898 }
 899 EXPORT_SYMBOL_GPL(bpf_map_inc);
 900
 901 void bpf_map_inc_with_uref(struct bpf_map *map)
 902 {
 903         atomic64_inc(&map->refcnt);
 904         atomic64_inc(&map->usercnt);
 905 }
 906 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
 907
 908 struct bpf_map *bpf_map_get(u32 ufd)
 909 {
 910         struct fd f = fdget(ufd);
 911         struct bpf_map *map;
 912
 913         map = __bpf_map_get(f);
 914         if (IS_ERR(map))
 915                 return map;
 916
 917         bpf_map_inc(map);
 918         fdput(f);
 919
 920         return map;
 921 }
 922
 923 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 924 {
 925         struct fd f = fdget(ufd);
 926         struct bpf_map *map;
 927
 928         map = __bpf_map_get(f);
 929         if (IS_ERR(map))
 930                 return map;
 931
 932         bpf_map_inc_with_uref(map);
 933         fdput(f);
 934
 935         return map;
 936 }
 937
 938 /* map_idr_lock should have been held */
 939 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 940 {
 941         int refold;
 942
 943         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
 944         if (!refold)
 945                 return ERR_PTR(-ENOENT);
 946         if (uref)
 947                 atomic64_inc(&map->usercnt);
 948
 949         return map;
 950 }
 951
 952 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
 953 {
 954         spin_lock_bh(&map_idr_lock);
 955         map = __bpf_map_inc_not_zero(map, false);
 956         spin_unlock_bh(&map_idr_lock);
 957
 958         return map;
 959 }
 960 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 961
 962 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 963 {
 964         return -ENOTSUPP;
 965 }
 966
 967 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 968 {
 969         if (key_size)
 970                 return memdup_user(ukey, key_size);
 971
 972         if (ukey)
 973                 return ERR_PTR(-EINVAL);
 974
 975         return NULL;
 976 }
 977
 978 /* last field in 'union bpf_attr' used by this command */
 979 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 980
 981 static int map_lookup_elem(union bpf_attr *attr)
 982 {
 983         void __user *ukey = u64_to_user_ptr(attr->key);
 984         void __user *uvalue = u64_to_user_ptr(attr->value);
 985         int ufd = attr->map_fd;
 986         struct bpf_map *map;
 987         void *key, *value;
 988         u32 value_size;
 989         struct fd f;
 990         int err;
 991
 992         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 993                 return -EINVAL;
 994
 995         if (attr->flags & ~BPF_F_LOCK)
 996                 return -EINVAL;
 997
 998         f = fdget(ufd);
 999         map = __bpf_map_get(f);
1000         if (IS_ERR(map))
1001                 return PTR_ERR(map);
1002         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1003                 err = -EPERM;
1004                 goto err_put;
1005         }
1006
1007         if ((attr->flags & BPF_F_LOCK) &&
1008             !map_value_has_spin_lock(map)) {
1009                 err = -EINVAL;
1010                 goto err_put;
1011         }
1012
1013         key = __bpf_copy_key(ukey, map->key_size);
1014         if (IS_ERR(key)) {
1015                 err = PTR_ERR(key);
1016                 goto err_put;
1017         }
1018
1019         value_size = bpf_map_value_size(map);
1020
1021         err = -ENOMEM;
1022         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1023         if (!value)
1024                 goto free_key;
1025
1026         err = bpf_map_copy_value(map, key, value, attr->flags);
1027         if (err)
1028                 goto free_value;
1029
1030         err = -EFAULT;
1031         if (copy_to_user(uvalue, value, value_size) != 0)
1032                 goto free_value;
1033
1034         err = 0;
1035
1036 free_value:
1037         kfree(value);
1038 free_key:
1039         kfree(key);
1040 err_put:
1041         fdput(f);
1042         return err;
1043 }
1044
1045
1046 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1047
1048 static int map_update_elem(union bpf_attr *attr)
1049 {
1050         void __user *ukey = u64_to_user_ptr(attr->key);
1051         void __user *uvalue = u64_to_user_ptr(attr->value);
1052         int ufd = attr->map_fd;
1053         struct bpf_map *map;
1054         void *key, *value;
1055         u32 value_size;
1056         struct fd f;
1057         int err;
1058
1059         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1060                 return -EINVAL;
1061
1062         f = fdget(ufd);
1063         map = __bpf_map_get(f);
1064         if (IS_ERR(map))
1065                 return PTR_ERR(map);
1066         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1067                 err = -EPERM;
1068                 goto err_put;
1069         }
1070
1071         if ((attr->flags & BPF_F_LOCK) &&
1072             !map_value_has_spin_lock(map)) {
1073                 err = -EINVAL;
1074                 goto err_put;
1075         }
1076
1077         key = __bpf_copy_key(ukey, map->key_size);
1078         if (IS_ERR(key)) {
1079                 err = PTR_ERR(key);
1080                 goto err_put;
1081         }
1082
1083         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1084             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1085             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1086             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1087                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1088         else
1089                 value_size = map->value_size;
1090
1091         err = -ENOMEM;
1092         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1093         if (!value)
1094                 goto free_key;
1095
1096         err = -EFAULT;
1097         if (copy_from_user(value, uvalue, value_size) != 0)
1098                 goto free_value;
1099
1100         err = bpf_map_update_value(map, f, key, value, attr->flags);
1101
1102 free_value:
1103         kfree(value);
1104 free_key:
1105         kfree(key);
1106 err_put:
1107         fdput(f);
1108         return err;
1109 }
1110
1111 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1112
1113 static int map_delete_elem(union bpf_attr *attr)
1114 {
1115         void __user *ukey = u64_to_user_ptr(attr->key);
1116         int ufd = attr->map_fd;
1117         struct bpf_map *map;
1118         struct fd f;
1119         void *key;
1120         int err;
1121
1122         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1123                 return -EINVAL;
1124
1125         f = fdget(ufd);
1126         map = __bpf_map_get(f);
1127         if (IS_ERR(map))
1128                 return PTR_ERR(map);
1129         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1130                 err = -EPERM;
1131                 goto err_put;
1132         }
1133
1134         key = __bpf_copy_key(ukey, map->key_size);
1135         if (IS_ERR(key)) {
1136                 err = PTR_ERR(key);
1137                 goto err_put;
1138         }
1139
1140         if (bpf_map_is_dev_bound(map)) {
1141                 err = bpf_map_offload_delete_elem(map, key);
1142                 goto out;
1143         } else if (IS_FD_PROG_ARRAY(map) ||
1144                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1145                 /* These maps require sleepable context */
1146                 err = map->ops->map_delete_elem(map, key);
1147                 goto out;
1148         }
1149
1150         bpf_disable_instrumentation();
1151         rcu_read_lock();
1152         err = map->ops->map_delete_elem(map, key);
1153         rcu_read_unlock();
1154         bpf_enable_instrumentation();
1155         maybe_wait_bpf_programs(map);
1156 out:
1157         kfree(key);
1158 err_put:
1159         fdput(f);
1160         return err;
1161 }
1162
1163 /* last field in 'union bpf_attr' used by this command */
1164 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1165
1166 static int map_get_next_key(union bpf_attr *attr)
1167 {
1168         void __user *ukey = u64_to_user_ptr(attr->key);
1169         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1170         int ufd = attr->map_fd;
1171         struct bpf_map *map;
1172         void *key, *next_key;
1173         struct fd f;
1174         int err;
1175
1176         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1177                 return -EINVAL;
1178
1179         f = fdget(ufd);
1180         map = __bpf_map_get(f);
1181         if (IS_ERR(map))
1182                 return PTR_ERR(map);
1183         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1184                 err = -EPERM;
1185                 goto err_put;
1186         }
1187
1188         if (ukey) {
1189                 key = __bpf_copy_key(ukey, map->key_size);
1190                 if (IS_ERR(key)) {
1191                         err = PTR_ERR(key);
1192                         goto err_put;
1193                 }
1194         } else {
1195                 key = NULL;
1196         }
1197
1198         err = -ENOMEM;
1199         next_key = kmalloc(map->key_size, GFP_USER);
1200         if (!next_key)
1201                 goto free_key;
1202
1203         if (bpf_map_is_dev_bound(map)) {
1204                 err = bpf_map_offload_get_next_key(map, key, next_key);
1205                 goto out;
1206         }
1207
1208         rcu_read_lock();
1209         err = map->ops->map_get_next_key(map, key, next_key);
1210         rcu_read_unlock();
1211 out:
1212         if (err)
1213                 goto free_next_key;
1214
1215         err = -EFAULT;
1216         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1217                 goto free_next_key;
1218
1219         err = 0;
1220
1221 free_next_key:
1222         kfree(next_key);
1223 free_key:
1224         kfree(key);
1225 err_put:
1226         fdput(f);
1227         return err;
1228 }
1229
1230 int generic_map_delete_batch(struct bpf_map *map,
1231                              const union bpf_attr *attr,
1232                              union bpf_attr __user *uattr)
1233 {
1234         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1235         u32 cp, max_count;
1236         int err = 0;
1237         void *key;
1238
1239         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1240                 return -EINVAL;
1241
1242         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1243             !map_value_has_spin_lock(map)) {
1244                 return -EINVAL;
1245         }
1246
1247         max_count = attr->batch.count;
1248         if (!max_count)
1249                 return 0;
1250
1251         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1252         if (!key)
1253                 return -ENOMEM;
1254
1255         for (cp = 0; cp < max_count; cp++) {
1256                 err = -EFAULT;
1257                 if (copy_from_user(key, keys + cp * map->key_size,
1258                                    map->key_size))
1259                         break;
1260
1261                 if (bpf_map_is_dev_bound(map)) {
1262                         err = bpf_map_offload_delete_elem(map, key);
1263                         break;
1264                 }
1265
1266                 bpf_disable_instrumentation();
1267                 rcu_read_lock();
1268                 err = map->ops->map_delete_elem(map, key);
1269                 rcu_read_unlock();
1270                 bpf_enable_instrumentation();
1271                 maybe_wait_bpf_programs(map);
1272                 if (err)
1273                         break;
1274         }
1275         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1276                 err = -EFAULT;
1277
1278         kfree(key);
1279         return err;
1280 }
1281
1282 int generic_map_update_batch(struct bpf_map *map,
1283                              const union bpf_attr *attr,
1284                              union bpf_attr __user *uattr)
1285 {
1286         void __user *values = u64_to_user_ptr(attr->batch.values);
1287         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1288         u32 value_size, cp, max_count;
1289         int ufd = attr->map_fd;
1290         void *key, *value;
1291         struct fd f;
1292         int err = 0;
1293
1294         f = fdget(ufd);
1295         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1296                 return -EINVAL;
1297
1298         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1299             !map_value_has_spin_lock(map)) {
1300                 return -EINVAL;
1301         }
1302
1303         value_size = bpf_map_value_size(map);
1304
1305         max_count = attr->batch.count;
1306         if (!max_count)
1307                 return 0;
1308
1309         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1310         if (!key)
1311                 return -ENOMEM;
1312
1313         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1314         if (!value) {
1315                 kfree(key);
1316                 return -ENOMEM;
1317         }
1318
1319         for (cp = 0; cp < max_count; cp++) {
1320                 err = -EFAULT;
1321                 if (copy_from_user(key, keys + cp * map->key_size,
1322                     map->key_size) ||
1323                     copy_from_user(value, values + cp * value_size, value_size))
1324                         break;
1325
1326                 err = bpf_map_update_value(map, f, key, value,
1327                                            attr->batch.elem_flags);
1328
1329                 if (err)
1330                         break;
1331         }
1332
1333         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1334                 err = -EFAULT;
1335
1336         kfree(value);
1337         kfree(key);
1338         return err;
1339 }
1340
1341 #define MAP_LOOKUP_RETRIES 3
1342
1343 int generic_map_lookup_batch(struct bpf_map *map,
1344                                     const union bpf_attr *attr,
1345                                     union bpf_attr __user *uattr)
1346 {
1347         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1348         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1349         void __user *values = u64_to_user_ptr(attr->batch.values);
1350         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1351         void *buf, *buf_prevkey, *prev_key, *key, *value;
1352         int err, retry = MAP_LOOKUP_RETRIES;
1353         u32 value_size, cp, max_count;
1354
1355         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1356                 return -EINVAL;
1357
1358         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1359             !map_value_has_spin_lock(map))
1360                 return -EINVAL;
1361
1362         value_size = bpf_map_value_size(map);
1363
1364         max_count = attr->batch.count;
1365         if (!max_count)
1366                 return 0;
1367
1368         if (put_user(0, &uattr->batch.count))
1369                 return -EFAULT;
1370
1371         buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1372         if (!buf_prevkey)
1373                 return -ENOMEM;
1374
1375         buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1376         if (!buf) {
1377                 kvfree(buf_prevkey);
1378                 return -ENOMEM;
1379         }
1380
1381         err = -EFAULT;
1382         prev_key = NULL;
1383         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1384                 goto free_buf;
1385         key = buf;
1386         value = key + map->key_size;
1387         if (ubatch)
1388                 prev_key = buf_prevkey;
1389
1390         for (cp = 0; cp < max_count;) {
1391                 rcu_read_lock();
1392                 err = map->ops->map_get_next_key(map, prev_key, key);
1393                 rcu_read_unlock();
1394                 if (err)
1395                         break;
1396                 err = bpf_map_copy_value(map, key, value,
1397                                          attr->batch.elem_flags);
1398
1399                 if (err == -ENOENT) {
1400                         if (retry) {
1401                                 retry--;
1402                                 continue;
1403                         }
1404                         err = -EINTR;
1405                         break;
1406                 }
1407
1408                 if (err)
1409                         goto free_buf;
1410
1411                 if (copy_to_user(keys + cp * map->key_size, key,
1412                                  map->key_size)) {
1413                         err = -EFAULT;
1414                         goto free_buf;
1415                 }
1416                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1417                         err = -EFAULT;
1418                         goto free_buf;
1419                 }
1420
1421                 if (!prev_key)
1422                         prev_key = buf_prevkey;
1423
1424                 swap(prev_key, key);
1425                 retry = MAP_LOOKUP_RETRIES;
1426                 cp++;
1427         }
1428
1429         if (err == -EFAULT)
1430                 goto free_buf;
1431
1432         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1433                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1434                 err = -EFAULT;
1435
1436 free_buf:
1437         kfree(buf_prevkey);
1438         kfree(buf);
1439         return err;
1440 }
1441
1442 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1443
1444 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1445 {
1446         void __user *ukey = u64_to_user_ptr(attr->key);
1447         void __user *uvalue = u64_to_user_ptr(attr->value);
1448         int ufd = attr->map_fd;
1449         struct bpf_map *map;
1450         void *key, *value;
1451         u32 value_size;
1452         struct fd f;
1453         int err;
1454
1455         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1456                 return -EINVAL;
1457
1458         f = fdget(ufd);
1459         map = __bpf_map_get(f);
1460         if (IS_ERR(map))
1461                 return PTR_ERR(map);
1462         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1463                 err = -EPERM;
1464                 goto err_put;
1465         }
1466
1467         key = __bpf_copy_key(ukey, map->key_size);
1468         if (IS_ERR(key)) {
1469                 err = PTR_ERR(key);
1470                 goto err_put;
1471         }
1472
1473         value_size = map->value_size;
1474
1475         err = -ENOMEM;
1476         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1477         if (!value)
1478                 goto free_key;
1479
1480         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1481             map->map_type == BPF_MAP_TYPE_STACK) {
1482                 err = map->ops->map_pop_elem(map, value);
1483         } else {
1484                 err = -ENOTSUPP;
1485         }
1486
1487         if (err)
1488                 goto free_value;
1489
1490         if (copy_to_user(uvalue, value, value_size) != 0)
1491                 goto free_value;
1492
1493         err = 0;
1494
1495 free_value:
1496         kfree(value);
1497 free_key:
1498         kfree(key);
1499 err_put:
1500         fdput(f);
1501         return err;
1502 }
1503
1504 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1505
1506 static int map_freeze(const union bpf_attr *attr)
1507 {
1508         int err = 0, ufd = attr->map_fd;
1509         struct bpf_map *map;
1510         struct fd f;
1511
1512         if (CHECK_ATTR(BPF_MAP_FREEZE))
1513                 return -EINVAL;
1514
1515         f = fdget(ufd);
1516         map = __bpf_map_get(f);
1517         if (IS_ERR(map))
1518                 return PTR_ERR(map);
1519
1520         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1521                 fdput(f);
1522                 return -ENOTSUPP;
1523         }
1524
1525         mutex_lock(&map->freeze_mutex);
1526
1527         if (map->writecnt) {
1528                 err = -EBUSY;
1529                 goto err_put;
1530         }
1531         if (READ_ONCE(map->frozen)) {
1532                 err = -EBUSY;
1533                 goto err_put;
1534         }
1535         if (!capable(CAP_SYS_ADMIN)) {
1536                 err = -EPERM;
1537                 goto err_put;
1538         }
1539
1540         WRITE_ONCE(map->frozen, true);
1541 err_put:
1542         mutex_unlock(&map->freeze_mutex);
1543         fdput(f);
1544         return err;
1545 }
1546
1547 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1548 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1549         [_id] = & _name ## _prog_ops,
1550 #define BPF_MAP_TYPE(_id, _ops)
1551 #include <linux/bpf_types.h>
1552 #undef BPF_PROG_TYPE
1553 #undef BPF_MAP_TYPE
1554 };
1555
1556 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1557 {
1558         const struct bpf_prog_ops *ops;
1559
1560         if (type >= ARRAY_SIZE(bpf_prog_types))
1561                 return -EINVAL;
1562         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1563         ops = bpf_prog_types[type];
1564         if (!ops)
1565                 return -EINVAL;
1566
1567         if (!bpf_prog_is_dev_bound(prog->aux))
1568                 prog->aux->ops = ops;
1569         else
1570                 prog->aux->ops = &bpf_offload_prog_ops;
1571         prog->type = type;
1572         return 0;
1573 }
1574
1575 enum bpf_audit {
1576         BPF_AUDIT_LOAD,
1577         BPF_AUDIT_UNLOAD,
1578         BPF_AUDIT_MAX,
1579 };
1580
1581 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1582         [BPF_AUDIT_LOAD]   = "LOAD",
1583         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1584 };
1585
1586 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1587 {
1588         struct audit_context *ctx = NULL;
1589         struct audit_buffer *ab;
1590
1591         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1592                 return;
1593         if (audit_enabled == AUDIT_OFF)
1594                 return;
1595         if (op == BPF_AUDIT_LOAD)
1596                 ctx = audit_context();
1597         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1598         if (unlikely(!ab))
1599                 return;
1600         audit_log_format(ab, "prog-id=%u op=%s",
1601                          prog->aux->id, bpf_audit_str[op]);
1602         audit_log_end(ab);
1603 }
1604
1605 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1606 {
1607         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1608         unsigned long user_bufs;
1609
1610         if (user) {
1611                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1612                 if (user_bufs > memlock_limit) {
1613                         atomic_long_sub(pages, &user->locked_vm);
1614                         return -EPERM;
1615                 }
1616         }
1617
1618         return 0;
1619 }
1620
1621 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1622 {
1623         if (user)
1624                 atomic_long_sub(pages, &user->locked_vm);
1625 }
1626
1627 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1628 {
1629         struct user_struct *user = get_current_user();
1630         int ret;
1631
1632         ret = __bpf_prog_charge(user, prog->pages);
1633         if (ret) {
1634                 free_uid(user);
1635                 return ret;
1636         }
1637
1638         prog->aux->user = user;
1639         return 0;
1640 }
1641
1642 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1643 {
1644         struct user_struct *user = prog->aux->user;
1645
1646         __bpf_prog_uncharge(user, prog->pages);
1647         free_uid(user);
1648 }
1649
1650 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1651 {
1652         int id;
1653
1654         idr_preload(GFP_KERNEL);
1655         spin_lock_bh(&prog_idr_lock);
1656         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1657         if (id > 0)
1658                 prog->aux->id = id;
1659         spin_unlock_bh(&prog_idr_lock);
1660         idr_preload_end();
1661
1662         /* id is in [1, INT_MAX) */
1663         if (WARN_ON_ONCE(!id))
1664                 return -ENOSPC;
1665
1666         return id > 0 ? 0 : id;
1667 }
1668
1669 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1670 {
1671         /* cBPF to eBPF migrations are currently not in the idr store.
1672          * Offloaded programs are removed from the store when their device
1673          * disappears - even if someone grabs an fd to them they are unusable,
1674          * simply waiting for refcnt to drop to be freed.
1675          */
1676         if (!prog->aux->id)
1677                 return;
1678
1679         if (do_idr_lock)
1680                 spin_lock_bh(&prog_idr_lock);
1681         else
1682                 __acquire(&prog_idr_lock);
1683
1684         idr_remove(&prog_idr, prog->aux->id);
1685         prog->aux->id = 0;
1686
1687         if (do_idr_lock)
1688                 spin_unlock_bh(&prog_idr_lock);
1689         else
1690                 __release(&prog_idr_lock);
1691 }
1692
1693 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1694 {
1695         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1696
1697         kvfree(aux->func_info);
1698         kfree(aux->func_info_aux);
1699         bpf_prog_uncharge_memlock(aux->prog);
1700         security_bpf_prog_free(aux);
1701         bpf_prog_free(aux->prog);
1702 }
1703
1704 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1705 {
1706         bpf_prog_kallsyms_del_all(prog);
1707         btf_put(prog->aux->btf);
1708         bpf_prog_free_linfo(prog);
1709
1710         if (deferred)
1711                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1712         else
1713                 __bpf_prog_put_rcu(&prog->aux->rcu);
1714 }
1715
1716 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1717 {
1718         if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1719                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1720                 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1721                 /* bpf_prog_free_id() must be called first */
1722                 bpf_prog_free_id(prog, do_idr_lock);
1723                 __bpf_prog_put_noref(prog, true);
1724         }
1725 }
1726
1727 void bpf_prog_put(struct bpf_prog *prog)
1728 {
1729         __bpf_prog_put(prog, true);
1730 }
1731 EXPORT_SYMBOL_GPL(bpf_prog_put);
1732
1733 static int bpf_prog_release(struct inode *inode, struct file *filp)
1734 {
1735         struct bpf_prog *prog = filp->private_data;
1736
1737         bpf_prog_put(prog);
1738         return 0;
1739 }
1740
1741 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1742                                struct bpf_prog_stats *stats)
1743 {
1744         u64 nsecs = 0, cnt = 0;
1745         int cpu;
1746
1747         for_each_possible_cpu(cpu) {
1748                 const struct bpf_prog_stats *st;
1749                 unsigned int start;
1750                 u64 tnsecs, tcnt;
1751
1752                 st = per_cpu_ptr(prog->aux->stats, cpu);
1753                 do {
1754                         start = u64_stats_fetch_begin_irq(&st->syncp);
1755                         tnsecs = st->nsecs;
1756                         tcnt = st->cnt;
1757                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1758                 nsecs += tnsecs;
1759                 cnt += tcnt;
1760         }
1761         stats->nsecs = nsecs;
1762         stats->cnt = cnt;
1763 }
1764
1765 #ifdef CONFIG_PROC_FS
1766 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1767 {
1768         const struct bpf_prog *prog = filp->private_data;
1769         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1770         struct bpf_prog_stats stats;
1771
1772         bpf_prog_get_stats(prog, &stats);
1773         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1774         seq_printf(m,
1775                    "prog_type:\t%u\n"
1776                    "prog_jited:\t%u\n"
1777                    "prog_tag:\t%s\n"
1778                    "memlock:\t%llu\n"
1779                    "prog_id:\t%u\n"
1780                    "run_time_ns:\t%llu\n"
1781                    "run_cnt:\t%llu\n",
1782                    prog->type,
1783                    prog->jited,
1784                    prog_tag,
1785                    prog->pages * 1ULL << PAGE_SHIFT,
1786                    prog->aux->id,
1787                    stats.nsecs,
1788                    stats.cnt);
1789 }
1790 #endif
1791
1792 const struct file_operations bpf_prog_fops = {
1793 #ifdef CONFIG_PROC_FS
1794         .show_fdinfo    = bpf_prog_show_fdinfo,
1795 #endif
1796         .release        = bpf_prog_release,
1797         .read           = bpf_dummy_read,
1798         .write          = bpf_dummy_write,
1799 };
1800
1801 int bpf_prog_new_fd(struct bpf_prog *prog)
1802 {
1803         int ret;
1804
1805         ret = security_bpf_prog(prog);
1806         if (ret < 0)
1807                 return ret;
1808
1809         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1810                                 O_RDWR | O_CLOEXEC);
1811 }
1812
1813 static struct bpf_prog *____bpf_prog_get(struct fd f)
1814 {
1815         if (!f.file)
1816                 return ERR_PTR(-EBADF);
1817         if (f.file->f_op != &bpf_prog_fops) {
1818                 fdput(f);
1819                 return ERR_PTR(-EINVAL);
1820         }
1821
1822         return f.file->private_data;
1823 }
1824
1825 void bpf_prog_add(struct bpf_prog *prog, int i)
1826 {
1827         atomic64_add(i, &prog->aux->refcnt);
1828 }
1829 EXPORT_SYMBOL_GPL(bpf_prog_add);
1830
1831 void bpf_prog_sub(struct bpf_prog *prog, int i)
1832 {
1833         /* Only to be used for undoing previous bpf_prog_add() in some
1834          * error path. We still know that another entity in our call
1835          * path holds a reference to the program, thus atomic_sub() can
1836          * be safely used in such cases!
1837          */
1838         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1839 }
1840 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1841
1842 void bpf_prog_inc(struct bpf_prog *prog)
1843 {
1844         atomic64_inc(&prog->aux->refcnt);
1845 }
1846 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1847
1848 /* prog_idr_lock should have been held */
1849 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1850 {
1851         int refold;
1852
1853         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1854
1855         if (!refold)
1856                 return ERR_PTR(-ENOENT);
1857
1858         return prog;
1859 }
1860 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1861
1862 bool bpf_prog_get_ok(struct bpf_prog *prog,
1863                             enum bpf_prog_type *attach_type, bool attach_drv)
1864 {
1865         /* not an attachment, just a refcount inc, always allow */
1866         if (!attach_type)
1867                 return true;
1868
1869         if (prog->type != *attach_type)
1870                 return false;
1871         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1872                 return false;
1873
1874         return true;
1875 }
1876
1877 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1878                                        bool attach_drv)
1879 {
1880         struct fd f = fdget(ufd);
1881         struct bpf_prog *prog;
1882
1883         prog = ____bpf_prog_get(f);
1884         if (IS_ERR(prog))
1885                 return prog;
1886         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1887                 prog = ERR_PTR(-EINVAL);
1888                 goto out;
1889         }
1890
1891         bpf_prog_inc(prog);
1892 out:
1893         fdput(f);
1894         return prog;
1895 }
1896
1897 struct bpf_prog *bpf_prog_get(u32 ufd)
1898 {
1899         return __bpf_prog_get(ufd, NULL, false);
1900 }
1901
1902 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1903                                        bool attach_drv)
1904 {
1905         return __bpf_prog_get(ufd, &type, attach_drv);
1906 }
1907 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1908
1909 /* Initially all BPF programs could be loaded w/o specifying
1910  * expected_attach_type. Later for some of them specifying expected_attach_type
1911  * at load time became required so that program could be validated properly.
1912  * Programs of types that are allowed to be loaded both w/ and w/o (for
1913  * backward compatibility) expected_attach_type, should have the default attach
1914  * type assigned to expected_attach_type for the latter case, so that it can be
1915  * validated later at attach time.
1916  *
1917  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1918  * prog type requires it but has some attach types that have to be backward
1919  * compatible.
1920  */
1921 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1922 {
1923         switch (attr->prog_type) {
1924         case BPF_PROG_TYPE_CGROUP_SOCK:
1925                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1926                  * exist so checking for non-zero is the way to go here.
1927                  */
1928                 if (!attr->expected_attach_type)
1929                         attr->expected_attach_type =
1930                                 BPF_CGROUP_INET_SOCK_CREATE;
1931                 break;
1932         }
1933 }
1934
1935 static int
1936 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1937                            enum bpf_attach_type expected_attach_type,
1938                            u32 btf_id, u32 prog_fd)
1939 {
1940         if (btf_id) {
1941                 if (btf_id > BTF_MAX_TYPE)
1942                         return -EINVAL;
1943
1944                 switch (prog_type) {
1945                 case BPF_PROG_TYPE_TRACING:
1946                 case BPF_PROG_TYPE_LSM:
1947                 case BPF_PROG_TYPE_STRUCT_OPS:
1948                 case BPF_PROG_TYPE_EXT:
1949                         break;
1950                 default:
1951                         return -EINVAL;
1952                 }
1953         }
1954
1955         if (prog_fd && prog_type != BPF_PROG_TYPE_TRACING &&
1956             prog_type != BPF_PROG_TYPE_EXT)
1957                 return -EINVAL;
1958
1959         switch (prog_type) {
1960         case BPF_PROG_TYPE_CGROUP_SOCK:
1961                 switch (expected_attach_type) {
1962                 case BPF_CGROUP_INET_SOCK_CREATE:
1963                 case BPF_CGROUP_INET4_POST_BIND:
1964                 case BPF_CGROUP_INET6_POST_BIND:
1965                         return 0;
1966                 default:
1967                         return -EINVAL;
1968                 }
1969         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1970                 switch (expected_attach_type) {
1971                 case BPF_CGROUP_INET4_BIND:
1972                 case BPF_CGROUP_INET6_BIND:
1973                 case BPF_CGROUP_INET4_CONNECT:
1974                 case BPF_CGROUP_INET6_CONNECT:
1975                 case BPF_CGROUP_UDP4_SENDMSG:
1976                 case BPF_CGROUP_UDP6_SENDMSG:
1977                 case BPF_CGROUP_UDP4_RECVMSG:
1978                 case BPF_CGROUP_UDP6_RECVMSG:
1979                         return 0;
1980                 default:
1981                         return -EINVAL;
1982                 }
1983         case BPF_PROG_TYPE_CGROUP_SKB:
1984                 switch (expected_attach_type) {
1985                 case BPF_CGROUP_INET_INGRESS:
1986                 case BPF_CGROUP_INET_EGRESS:
1987                         return 0;
1988                 default:
1989                         return -EINVAL;
1990                 }
1991         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1992                 switch (expected_attach_type) {
1993                 case BPF_CGROUP_SETSOCKOPT:
1994                 case BPF_CGROUP_GETSOCKOPT:
1995                         return 0;
1996                 default:
1997                         return -EINVAL;
1998                 }
1999         case BPF_PROG_TYPE_EXT:
2000                 if (expected_attach_type)
2001                         return -EINVAL;
2002                 /* fallthrough */
2003         default:
2004                 return 0;
2005         }
2006 }
2007
2008 /* last field in 'union bpf_attr' used by this command */
2009 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2010
2011 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2012 {
2013         enum bpf_prog_type type = attr->prog_type;
2014         struct bpf_prog *prog;
2015         int err;
2016         char license[128];
2017         bool is_gpl;
2018
2019         if (CHECK_ATTR(BPF_PROG_LOAD))
2020                 return -EINVAL;
2021
2022         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2023                                  BPF_F_ANY_ALIGNMENT |
2024                                  BPF_F_TEST_STATE_FREQ |
2025                                  BPF_F_TEST_RND_HI32))
2026                 return -EINVAL;
2027
2028         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2029             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2030             !capable(CAP_SYS_ADMIN))
2031                 return -EPERM;
2032
2033         /* copy eBPF program license from user space */
2034         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2035                               sizeof(license) - 1) < 0)
2036                 return -EFAULT;
2037         license[sizeof(license) - 1] = 0;
2038
2039         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2040         is_gpl = license_is_gpl_compatible(license);
2041
2042         if (attr->insn_cnt == 0 ||
2043             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2044                 return -E2BIG;
2045         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2046             type != BPF_PROG_TYPE_CGROUP_SKB &&
2047             !capable(CAP_SYS_ADMIN))
2048                 return -EPERM;
2049
2050         bpf_prog_load_fixup_attach_type(attr);
2051         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2052                                        attr->attach_btf_id,
2053                                        attr->attach_prog_fd))
2054                 return -EINVAL;
2055
2056         /* plain bpf_prog allocation */
2057         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2058         if (!prog)
2059                 return -ENOMEM;
2060
2061         prog->expected_attach_type = attr->expected_attach_type;
2062         prog->aux->attach_btf_id = attr->attach_btf_id;
2063         if (attr->attach_prog_fd) {
2064                 struct bpf_prog *tgt_prog;
2065
2066                 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
2067                 if (IS_ERR(tgt_prog)) {
2068                         err = PTR_ERR(tgt_prog);
2069                         goto free_prog_nouncharge;
2070                 }
2071                 prog->aux->linked_prog = tgt_prog;
2072         }
2073
2074         prog->aux->offload_requested = !!attr->prog_ifindex;
2075
2076         err = security_bpf_prog_alloc(prog->aux);
2077         if (err)
2078                 goto free_prog_nouncharge;
2079
2080         err = bpf_prog_charge_memlock(prog);
2081         if (err)
2082                 goto free_prog_sec;
2083
2084         prog->len = attr->insn_cnt;
2085
2086         err = -EFAULT;
2087         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2088                            bpf_prog_insn_size(prog)) != 0)
2089                 goto free_prog;
2090
2091         prog->orig_prog = NULL;
2092         prog->jited = 0;
2093
2094         atomic64_set(&prog->aux->refcnt, 1);
2095         prog->gpl_compatible = is_gpl ? 1 : 0;
2096
2097         if (bpf_prog_is_dev_bound(prog->aux)) {
2098                 err = bpf_prog_offload_init(prog, attr);
2099                 if (err)
2100                         goto free_prog;
2101         }
2102
2103         /* find program type: socket_filter vs tracing_filter */
2104         err = find_prog_type(type, prog);
2105         if (err < 0)
2106                 goto free_prog;
2107
2108         prog->aux->load_time = ktime_get_boottime_ns();
2109         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2110                                sizeof(attr->prog_name));
2111         if (err < 0)
2112                 goto free_prog;
2113
2114         /* run eBPF verifier */
2115         err = bpf_check(&prog, attr, uattr);
2116         if (err < 0)
2117                 goto free_used_maps;
2118
2119         prog = bpf_prog_select_runtime(prog, &err);
2120         if (err < 0)
2121                 goto free_used_maps;
2122
2123         err = bpf_prog_alloc_id(prog);
2124         if (err)
2125                 goto free_used_maps;
2126
2127         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2128          * effectively publicly exposed. However, retrieving via
2129          * bpf_prog_get_fd_by_id() will take another reference,
2130          * therefore it cannot be gone underneath us.
2131          *
2132          * Only for the time /after/ successful bpf_prog_new_fd()
2133          * and before returning to userspace, we might just hold
2134          * one reference and any parallel close on that fd could
2135          * rip everything out. Hence, below notifications must
2136          * happen before bpf_prog_new_fd().
2137          *
2138          * Also, any failure handling from this point onwards must
2139          * be using bpf_prog_put() given the program is exposed.
2140          */
2141         bpf_prog_kallsyms_add(prog);
2142         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2143         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2144
2145         err = bpf_prog_new_fd(prog);
2146         if (err < 0)
2147                 bpf_prog_put(prog);
2148         return err;
2149
2150 free_used_maps:
2151         /* In case we have subprogs, we need to wait for a grace
2152          * period before we can tear down JIT memory since symbols
2153          * are already exposed under kallsyms.
2154          */
2155         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2156         return err;
2157 free_prog:
2158         bpf_prog_uncharge_memlock(prog);
2159 free_prog_sec:
2160         security_bpf_prog_free(prog->aux);
2161 free_prog_nouncharge:
2162         bpf_prog_free(prog);
2163         return err;
2164 }
2165
2166 #define BPF_OBJ_LAST_FIELD file_flags
2167
2168 static int bpf_obj_pin(const union bpf_attr *attr)
2169 {
2170         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2171                 return -EINVAL;
2172
2173         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2174 }
2175
2176 static int bpf_obj_get(const union bpf_attr *attr)
2177 {
2178         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2179             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2180                 return -EINVAL;
2181
2182         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2183                                 attr->file_flags);
2184 }
2185
2186 void bpf_link_init(struct bpf_link *link, const struct bpf_link_ops *ops,
2187                    struct bpf_prog *prog)
2188 {
2189         atomic64_set(&link->refcnt, 1);
2190         link->ops = ops;
2191         link->prog = prog;
2192 }
2193
2194 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2195  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2196  * anon_inode's release() call. This helper manages marking bpf_link as
2197  * defunct, releases anon_inode file and puts reserved FD.
2198  */
2199 void bpf_link_cleanup(struct bpf_link *link, struct file *link_file,
2200                       int link_fd)
2201 {
2202         link->prog = NULL;
2203         fput(link_file);
2204         put_unused_fd(link_fd);
2205 }
2206
2207 void bpf_link_inc(struct bpf_link *link)
2208 {
2209         atomic64_inc(&link->refcnt);
2210 }
2211
2212 /* bpf_link_free is guaranteed to be called from process context */
2213 static void bpf_link_free(struct bpf_link *link)
2214 {
2215         if (link->prog) {
2216                 /* detach BPF program, clean up used resources */
2217                 link->ops->release(link);
2218                 bpf_prog_put(link->prog);
2219         }
2220         /* free bpf_link and its containing memory */
2221         link->ops->dealloc(link);
2222 }
2223
2224 static void bpf_link_put_deferred(struct work_struct *work)
2225 {
2226         struct bpf_link *link = container_of(work, struct bpf_link, work);
2227
2228         bpf_link_free(link);
2229 }
2230
2231 /* bpf_link_put can be called from atomic context, but ensures that resources
2232  * are freed from process context
2233  */
2234 void bpf_link_put(struct bpf_link *link)
2235 {
2236         if (!atomic64_dec_and_test(&link->refcnt))
2237                 return;
2238
2239         if (in_atomic()) {
2240                 INIT_WORK(&link->work, bpf_link_put_deferred);
2241                 schedule_work(&link->work);
2242         } else {
2243                 bpf_link_free(link);
2244         }
2245 }
2246
2247 static int bpf_link_release(struct inode *inode, struct file *filp)
2248 {
2249         struct bpf_link *link = filp->private_data;
2250
2251         bpf_link_put(link);
2252         return 0;
2253 }
2254
2255 #ifdef CONFIG_PROC_FS
2256 static const struct bpf_link_ops bpf_raw_tp_lops;
2257 static const struct bpf_link_ops bpf_tracing_link_lops;
2258
2259 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2260 {
2261         const struct bpf_link *link = filp->private_data;
2262         const struct bpf_prog *prog = link->prog;
2263         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2264         const char *link_type;
2265
2266         if (link->ops == &bpf_raw_tp_lops)
2267                 link_type = "raw_tracepoint";
2268         else if (link->ops == &bpf_tracing_link_lops)
2269                 link_type = "tracing";
2270 #ifdef CONFIG_CGROUP_BPF
2271         else if (link->ops == &bpf_cgroup_link_lops)
2272                 link_type = "cgroup";
2273 #endif
2274         else
2275                 link_type = "unknown";
2276
2277         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2278         seq_printf(m,
2279                    "link_type:\t%s\n"
2280                    "prog_tag:\t%s\n"
2281                    "prog_id:\t%u\n",
2282                    link_type,
2283                    prog_tag,
2284                    prog->aux->id);
2285 }
2286 #endif
2287
2288 const struct file_operations bpf_link_fops = {
2289 #ifdef CONFIG_PROC_FS
2290         .show_fdinfo    = bpf_link_show_fdinfo,
2291 #endif
2292         .release        = bpf_link_release,
2293         .read           = bpf_dummy_read,
2294         .write          = bpf_dummy_write,
2295 };
2296
2297 int bpf_link_new_fd(struct bpf_link *link)
2298 {
2299         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2300 }
2301
2302 /* Similar to bpf_link_new_fd, create anon_inode for given bpf_link, but
2303  * instead of immediately installing fd in fdtable, just reserve it and
2304  * return. Caller then need to either install it with fd_install(fd, file) or
2305  * release with put_unused_fd(fd).
2306  * This is useful for cases when bpf_link attachment/detachment are
2307  * complicated and expensive operations and should be delayed until all the fd
2308  * reservation and anon_inode creation succeeds.
2309  */
2310 struct file *bpf_link_new_file(struct bpf_link *link, int *reserved_fd)
2311 {
2312         struct file *file;
2313         int fd;
2314
2315         fd = get_unused_fd_flags(O_CLOEXEC);
2316         if (fd < 0)
2317                 return ERR_PTR(fd);
2318
2319         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2320         if (IS_ERR(file)) {
2321                 put_unused_fd(fd);
2322                 return file;
2323         }
2324
2325         *reserved_fd = fd;
2326         return file;
2327 }
2328
2329 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2330 {
2331         struct fd f = fdget(ufd);
2332         struct bpf_link *link;
2333
2334         if (!f.file)
2335                 return ERR_PTR(-EBADF);
2336         if (f.file->f_op != &bpf_link_fops) {
2337                 fdput(f);
2338                 return ERR_PTR(-EINVAL);
2339         }
2340
2341         link = f.file->private_data;
2342         bpf_link_inc(link);
2343         fdput(f);
2344
2345         return link;
2346 }
2347
2348 struct bpf_tracing_link {
2349         struct bpf_link link;
2350 };
2351
2352 static void bpf_tracing_link_release(struct bpf_link *link)
2353 {
2354         WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog));
2355 }
2356
2357 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2358 {
2359         struct bpf_tracing_link *tr_link =
2360                 container_of(link, struct bpf_tracing_link, link);
2361
2362         kfree(tr_link);
2363 }
2364
2365 static const struct bpf_link_ops bpf_tracing_link_lops = {
2366         .release = bpf_tracing_link_release,
2367         .dealloc = bpf_tracing_link_dealloc,
2368 };
2369
2370 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
2371 {
2372         struct bpf_tracing_link *link;
2373         struct file *link_file;
2374         int link_fd, err;
2375
2376         switch (prog->type) {
2377         case BPF_PROG_TYPE_TRACING:
2378                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2379                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2380                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2381                         err = -EINVAL;
2382                         goto out_put_prog;
2383                 }
2384                 break;
2385         case BPF_PROG_TYPE_EXT:
2386                 if (prog->expected_attach_type != 0) {
2387                         err = -EINVAL;
2388                         goto out_put_prog;
2389                 }
2390                 break;
2391         case BPF_PROG_TYPE_LSM:
2392                 if (prog->expected_attach_type != BPF_LSM_MAC) {
2393                         err = -EINVAL;
2394                         goto out_put_prog;
2395                 }
2396                 break;
2397         default:
2398                 err = -EINVAL;
2399                 goto out_put_prog;
2400         }
2401
2402         link = kzalloc(sizeof(*link), GFP_USER);
2403         if (!link) {
2404                 err = -ENOMEM;
2405                 goto out_put_prog;
2406         }
2407         bpf_link_init(&link->link, &bpf_tracing_link_lops, prog);
2408
2409         link_file = bpf_link_new_file(&link->link, &link_fd);
2410         if (IS_ERR(link_file)) {
2411                 kfree(link);
2412                 err = PTR_ERR(link_file);
2413                 goto out_put_prog;
2414         }
2415
2416         err = bpf_trampoline_link_prog(prog);
2417         if (err) {
2418                 bpf_link_cleanup(&link->link, link_file, link_fd);
2419                 goto out_put_prog;
2420         }
2421
2422         fd_install(link_fd, link_file);
2423         return link_fd;
2424
2425 out_put_prog:
2426         bpf_prog_put(prog);
2427         return err;
2428 }
2429
2430 struct bpf_raw_tp_link {
2431         struct bpf_link link;
2432         struct bpf_raw_event_map *btp;
2433 };
2434
2435 static void bpf_raw_tp_link_release(struct bpf_link *link)
2436 {
2437         struct bpf_raw_tp_link *raw_tp =
2438                 container_of(link, struct bpf_raw_tp_link, link);
2439
2440         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2441         bpf_put_raw_tracepoint(raw_tp->btp);
2442 }
2443
2444 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2445 {
2446         struct bpf_raw_tp_link *raw_tp =
2447                 container_of(link, struct bpf_raw_tp_link, link);
2448
2449         kfree(raw_tp);
2450 }
2451
2452 static const struct bpf_link_ops bpf_raw_tp_lops = {
2453         .release = bpf_raw_tp_link_release,
2454         .dealloc = bpf_raw_tp_link_dealloc,
2455 };
2456
2457 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2458
2459 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2460 {
2461         struct bpf_raw_tp_link *link;
2462         struct bpf_raw_event_map *btp;
2463         struct file *link_file;
2464         struct bpf_prog *prog;
2465         const char *tp_name;
2466         char buf[128];
2467         int link_fd, err;
2468
2469         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2470                 return -EINVAL;
2471
2472         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2473         if (IS_ERR(prog))
2474                 return PTR_ERR(prog);
2475
2476         switch (prog->type) {
2477         case BPF_PROG_TYPE_TRACING:
2478         case BPF_PROG_TYPE_EXT:
2479         case BPF_PROG_TYPE_LSM:
2480                 if (attr->raw_tracepoint.name) {
2481                         /* The attach point for this category of programs
2482                          * should be specified via btf_id during program load.
2483                          */
2484                         err = -EINVAL;
2485                         goto out_put_prog;
2486                 }
2487                 if (prog->type == BPF_PROG_TYPE_TRACING &&
2488                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2489                         tp_name = prog->aux->attach_func_name;
2490                         break;
2491                 }
2492                 return bpf_tracing_prog_attach(prog);
2493         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2494         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2495                 if (strncpy_from_user(buf,
2496                                       u64_to_user_ptr(attr->raw_tracepoint.name),
2497                                       sizeof(buf) - 1) < 0) {
2498                         err = -EFAULT;
2499                         goto out_put_prog;
2500                 }
2501                 buf[sizeof(buf) - 1] = 0;
2502                 tp_name = buf;
2503                 break;
2504         default:
2505                 err = -EINVAL;
2506                 goto out_put_prog;
2507         }
2508
2509         btp = bpf_get_raw_tracepoint(tp_name);
2510         if (!btp) {
2511                 err = -ENOENT;
2512                 goto out_put_prog;
2513         }
2514
2515         link = kzalloc(sizeof(*link), GFP_USER);
2516         if (!link) {
2517                 err = -ENOMEM;
2518                 goto out_put_btp;
2519         }
2520         bpf_link_init(&link->link, &bpf_raw_tp_lops, prog);
2521         link->btp = btp;
2522
2523         link_file = bpf_link_new_file(&link->link, &link_fd);
2524         if (IS_ERR(link_file)) {
2525                 kfree(link);
2526                 err = PTR_ERR(link_file);
2527                 goto out_put_btp;
2528         }
2529
2530         err = bpf_probe_register(link->btp, prog);
2531         if (err) {
2532                 bpf_link_cleanup(&link->link, link_file, link_fd);
2533                 goto out_put_btp;
2534         }
2535
2536         fd_install(link_fd, link_file);
2537         return link_fd;
2538
2539 out_put_btp:
2540         bpf_put_raw_tracepoint(btp);
2541 out_put_prog:
2542         bpf_prog_put(prog);
2543         return err;
2544 }
2545
2546 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2547                                              enum bpf_attach_type attach_type)
2548 {
2549         switch (prog->type) {
2550         case BPF_PROG_TYPE_CGROUP_SOCK:
2551         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2552         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2553                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2554         case BPF_PROG_TYPE_CGROUP_SKB:
2555                 return prog->enforce_expected_attach_type &&
2556                         prog->expected_attach_type != attach_type ?
2557                         -EINVAL : 0;
2558         default:
2559                 return 0;
2560         }
2561 }
2562
2563 static enum bpf_prog_type
2564 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2565 {
2566         switch (attach_type) {
2567         case BPF_CGROUP_INET_INGRESS:
2568         case BPF_CGROUP_INET_EGRESS:
2569                 return BPF_PROG_TYPE_CGROUP_SKB;
2570                 break;
2571         case BPF_CGROUP_INET_SOCK_CREATE:
2572         case BPF_CGROUP_INET4_POST_BIND:
2573         case BPF_CGROUP_INET6_POST_BIND:
2574                 return BPF_PROG_TYPE_CGROUP_SOCK;
2575         case BPF_CGROUP_INET4_BIND:
2576         case BPF_CGROUP_INET6_BIND:
2577         case BPF_CGROUP_INET4_CONNECT:
2578         case BPF_CGROUP_INET6_CONNECT:
2579         case BPF_CGROUP_UDP4_SENDMSG:
2580         case BPF_CGROUP_UDP6_SENDMSG:
2581         case BPF_CGROUP_UDP4_RECVMSG:
2582         case BPF_CGROUP_UDP6_RECVMSG:
2583                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2584         case BPF_CGROUP_SOCK_OPS:
2585                 return BPF_PROG_TYPE_SOCK_OPS;
2586         case BPF_CGROUP_DEVICE:
2587                 return BPF_PROG_TYPE_CGROUP_DEVICE;
2588         case BPF_SK_MSG_VERDICT:
2589                 return BPF_PROG_TYPE_SK_MSG;
2590         case BPF_SK_SKB_STREAM_PARSER:
2591         case BPF_SK_SKB_STREAM_VERDICT:
2592                 return BPF_PROG_TYPE_SK_SKB;
2593         case BPF_LIRC_MODE2:
2594                 return BPF_PROG_TYPE_LIRC_MODE2;
2595         case BPF_FLOW_DISSECTOR:
2596                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2597         case BPF_CGROUP_SYSCTL:
2598                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2599         case BPF_CGROUP_GETSOCKOPT:
2600         case BPF_CGROUP_SETSOCKOPT:
2601                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2602         default:
2603                 return BPF_PROG_TYPE_UNSPEC;
2604         }
2605 }
2606
2607 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2608
2609 #define BPF_F_ATTACH_MASK \
2610         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2611
2612 static int bpf_prog_attach(const union bpf_attr *attr)
2613 {
2614         enum bpf_prog_type ptype;
2615         struct bpf_prog *prog;
2616         int ret;
2617
2618         if (!capable(CAP_NET_ADMIN))
2619                 return -EPERM;
2620
2621         if (CHECK_ATTR(BPF_PROG_ATTACH))
2622                 return -EINVAL;
2623
2624         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2625                 return -EINVAL;
2626
2627         ptype = attach_type_to_prog_type(attr->attach_type);
2628         if (ptype == BPF_PROG_TYPE_UNSPEC)
2629                 return -EINVAL;
2630
2631         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2632         if (IS_ERR(prog))
2633                 return PTR_ERR(prog);
2634
2635         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2636                 bpf_prog_put(prog);
2637                 return -EINVAL;
2638         }
2639
2640         switch (ptype) {
2641         case BPF_PROG_TYPE_SK_SKB:
2642         case BPF_PROG_TYPE_SK_MSG:
2643                 ret = sock_map_get_from_fd(attr, prog);
2644                 break;
2645         case BPF_PROG_TYPE_LIRC_MODE2:
2646                 ret = lirc_prog_attach(attr, prog);
2647                 break;
2648         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2649                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2650                 break;
2651         case BPF_PROG_TYPE_CGROUP_DEVICE:
2652         case BPF_PROG_TYPE_CGROUP_SKB:
2653         case BPF_PROG_TYPE_CGROUP_SOCK:
2654         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2655         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2656         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2657         case BPF_PROG_TYPE_SOCK_OPS:
2658                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2659                 break;
2660         default:
2661                 ret = -EINVAL;
2662         }
2663
2664         if (ret)
2665                 bpf_prog_put(prog);
2666         return ret;
2667 }
2668
2669 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2670
2671 static int bpf_prog_detach(const union bpf_attr *attr)
2672 {
2673         enum bpf_prog_type ptype;
2674
2675         if (!capable(CAP_NET_ADMIN))
2676                 return -EPERM;
2677
2678         if (CHECK_ATTR(BPF_PROG_DETACH))
2679                 return -EINVAL;
2680
2681         ptype = attach_type_to_prog_type(attr->attach_type);
2682
2683         switch (ptype) {
2684         case BPF_PROG_TYPE_SK_MSG:
2685         case BPF_PROG_TYPE_SK_SKB:
2686                 return sock_map_get_from_fd(attr, NULL);
2687         case BPF_PROG_TYPE_LIRC_MODE2:
2688                 return lirc_prog_detach(attr);
2689         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2690                 return skb_flow_dissector_bpf_prog_detach(attr);
2691         case BPF_PROG_TYPE_CGROUP_DEVICE:
2692         case BPF_PROG_TYPE_CGROUP_SKB:
2693         case BPF_PROG_TYPE_CGROUP_SOCK:
2694         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2695         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2696         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2697         case BPF_PROG_TYPE_SOCK_OPS:
2698                 return cgroup_bpf_prog_detach(attr, ptype);
2699         default:
2700                 return -EINVAL;
2701         }
2702 }
2703
2704 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2705
2706 static int bpf_prog_query(const union bpf_attr *attr,
2707                           union bpf_attr __user *uattr)
2708 {
2709         if (!capable(CAP_NET_ADMIN))
2710                 return -EPERM;
2711         if (CHECK_ATTR(BPF_PROG_QUERY))
2712                 return -EINVAL;
2713         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2714                 return -EINVAL;
2715
2716         switch (attr->query.attach_type) {
2717         case BPF_CGROUP_INET_INGRESS:
2718         case BPF_CGROUP_INET_EGRESS:
2719         case BPF_CGROUP_INET_SOCK_CREATE:
2720         case BPF_CGROUP_INET4_BIND:
2721         case BPF_CGROUP_INET6_BIND:
2722         case BPF_CGROUP_INET4_POST_BIND:
2723         case BPF_CGROUP_INET6_POST_BIND:
2724         case BPF_CGROUP_INET4_CONNECT:
2725         case BPF_CGROUP_INET6_CONNECT:
2726         case BPF_CGROUP_UDP4_SENDMSG:
2727         case BPF_CGROUP_UDP6_SENDMSG:
2728         case BPF_CGROUP_UDP4_RECVMSG:
2729         case BPF_CGROUP_UDP6_RECVMSG:
2730         case BPF_CGROUP_SOCK_OPS:
2731         case BPF_CGROUP_DEVICE:
2732         case BPF_CGROUP_SYSCTL:
2733         case BPF_CGROUP_GETSOCKOPT:
2734         case BPF_CGROUP_SETSOCKOPT:
2735                 return cgroup_bpf_prog_query(attr, uattr);
2736         case BPF_LIRC_MODE2:
2737                 return lirc_prog_query(attr, uattr);
2738         case BPF_FLOW_DISSECTOR:
2739                 return skb_flow_dissector_prog_query(attr, uattr);
2740         default:
2741                 return -EINVAL;
2742         }
2743 }
2744
2745 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2746
2747 static int bpf_prog_test_run(const union bpf_attr *attr,
2748                              union bpf_attr __user *uattr)
2749 {
2750         struct bpf_prog *prog;
2751         int ret = -ENOTSUPP;
2752
2753         if (!capable(CAP_SYS_ADMIN))
2754                 return -EPERM;
2755         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2756                 return -EINVAL;
2757
2758         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2759             (!attr->test.ctx_size_in && attr->test.ctx_in))
2760                 return -EINVAL;
2761
2762         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2763             (!attr->test.ctx_size_out && attr->test.ctx_out))
2764                 return -EINVAL;
2765
2766         prog = bpf_prog_get(attr->test.prog_fd);
2767         if (IS_ERR(prog))
2768                 return PTR_ERR(prog);
2769
2770         if (prog->aux->ops->test_run)
2771                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2772
2773         bpf_prog_put(prog);
2774         return ret;
2775 }
2776
2777 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2778
2779 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2780                                union bpf_attr __user *uattr,
2781                                struct idr *idr,
2782                                spinlock_t *lock)
2783 {
2784         u32 next_id = attr->start_id;
2785         int err = 0;
2786
2787         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2788                 return -EINVAL;
2789
2790         if (!capable(CAP_SYS_ADMIN))
2791                 return -EPERM;
2792
2793         next_id++;
2794         spin_lock_bh(lock);
2795         if (!idr_get_next(idr, &next_id))
2796                 err = -ENOENT;
2797         spin_unlock_bh(lock);
2798
2799         if (!err)
2800                 err = put_user(next_id, &uattr->next_id);
2801
2802         return err;
2803 }
2804
2805 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2806
2807 struct bpf_prog *bpf_prog_by_id(u32 id)
2808 {
2809         struct bpf_prog *prog;
2810
2811         if (!id)
2812                 return ERR_PTR(-ENOENT);
2813
2814         spin_lock_bh(&prog_idr_lock);
2815         prog = idr_find(&prog_idr, id);
2816         if (prog)
2817                 prog = bpf_prog_inc_not_zero(prog);
2818         else
2819                 prog = ERR_PTR(-ENOENT);
2820         spin_unlock_bh(&prog_idr_lock);
2821         return prog;
2822 }
2823
2824 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2825 {
2826         struct bpf_prog *prog;
2827         u32 id = attr->prog_id;
2828         int fd;
2829
2830         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2831                 return -EINVAL;
2832
2833         if (!capable(CAP_SYS_ADMIN))
2834                 return -EPERM;
2835
2836         prog = bpf_prog_by_id(id);
2837         if (IS_ERR(prog))
2838                 return PTR_ERR(prog);
2839
2840         fd = bpf_prog_new_fd(prog);
2841         if (fd < 0)
2842                 bpf_prog_put(prog);
2843
2844         return fd;
2845 }
2846
2847 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2848
2849 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2850 {
2851         struct bpf_map *map;
2852         u32 id = attr->map_id;
2853         int f_flags;
2854         int fd;
2855
2856         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2857             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2858                 return -EINVAL;
2859
2860         if (!capable(CAP_SYS_ADMIN))
2861                 return -EPERM;
2862
2863         f_flags = bpf_get_file_flag(attr->open_flags);
2864         if (f_flags < 0)
2865                 return f_flags;
2866
2867         spin_lock_bh(&map_idr_lock);
2868         map = idr_find(&map_idr, id);
2869         if (map)
2870                 map = __bpf_map_inc_not_zero(map, true);
2871         else
2872                 map = ERR_PTR(-ENOENT);
2873         spin_unlock_bh(&map_idr_lock);
2874
2875         if (IS_ERR(map))
2876                 return PTR_ERR(map);
2877
2878         fd = bpf_map_new_fd(map, f_flags);
2879         if (fd < 0)
2880                 bpf_map_put_with_uref(map);
2881
2882         return fd;
2883 }
2884
2885 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2886                                               unsigned long addr, u32 *off,
2887                                               u32 *type)
2888 {
2889         const struct bpf_map *map;
2890         int i;
2891
2892         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2893                 map = prog->aux->used_maps[i];
2894                 if (map == (void *)addr) {
2895                         *type = BPF_PSEUDO_MAP_FD;
2896                         return map;
2897                 }
2898                 if (!map->ops->map_direct_value_meta)
2899                         continue;
2900                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2901                         *type = BPF_PSEUDO_MAP_VALUE;
2902                         return map;
2903                 }
2904         }
2905
2906         return NULL;
2907 }
2908
2909 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2910 {
2911         const struct bpf_map *map;
2912         struct bpf_insn *insns;
2913         u32 off, type;
2914         u64 imm;
2915         int i;
2916
2917         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2918                         GFP_USER);
2919         if (!insns)
2920                 return insns;
2921
2922         for (i = 0; i < prog->len; i++) {
2923                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2924                         insns[i].code = BPF_JMP | BPF_CALL;
2925                         insns[i].imm = BPF_FUNC_tail_call;
2926                         /* fall-through */
2927                 }
2928                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2929                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2930                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2931                                 insns[i].code = BPF_JMP | BPF_CALL;
2932                         if (!bpf_dump_raw_ok())
2933                                 insns[i].imm = 0;
2934                         continue;
2935                 }
2936
2937                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2938                         continue;
2939
2940                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2941                 map = bpf_map_from_imm(prog, imm, &off, &type);
2942                 if (map) {
2943                         insns[i].src_reg = type;
2944                         insns[i].imm = map->id;
2945                         insns[i + 1].imm = off;
2946                         continue;
2947                 }
2948         }
2949
2950         return insns;
2951 }
2952
2953 static int set_info_rec_size(struct bpf_prog_info *info)
2954 {
2955         /*
2956          * Ensure info.*_rec_size is the same as kernel expected size
2957          *
2958          * or
2959          *
2960          * Only allow zero *_rec_size if both _rec_size and _cnt are
2961          * zero.  In this case, the kernel will set the expected
2962          * _rec_size back to the info.
2963          */
2964
2965         if ((info->nr_func_info || info->func_info_rec_size) &&
2966             info->func_info_rec_size != sizeof(struct bpf_func_info))
2967                 return -EINVAL;
2968
2969         if ((info->nr_line_info || info->line_info_rec_size) &&
2970             info->line_info_rec_size != sizeof(struct bpf_line_info))
2971                 return -EINVAL;
2972
2973         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2974             info->jited_line_info_rec_size != sizeof(__u64))
2975                 return -EINVAL;
2976
2977         info->func_info_rec_size = sizeof(struct bpf_func_info);
2978         info->line_info_rec_size = sizeof(struct bpf_line_info);
2979         info->jited_line_info_rec_size = sizeof(__u64);
2980
2981         return 0;
2982 }
2983
2984 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2985                                    const union bpf_attr *attr,
2986                                    union bpf_attr __user *uattr)
2987 {
2988         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2989         struct bpf_prog_info info;
2990         u32 info_len = attr->info.info_len;
2991         struct bpf_prog_stats stats;
2992         char __user *uinsns;
2993         u32 ulen;
2994         int err;
2995
2996         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2997         if (err)
2998                 return err;
2999         info_len = min_t(u32, sizeof(info), info_len);
3000
3001         memset(&info, 0, sizeof(info));
3002         if (copy_from_user(&info, uinfo, info_len))
3003                 return -EFAULT;
3004
3005         info.type = prog->type;
3006         info.id = prog->aux->id;
3007         info.load_time = prog->aux->load_time;
3008         info.created_by_uid = from_kuid_munged(current_user_ns(),
3009                                                prog->aux->user->uid);
3010         info.gpl_compatible = prog->gpl_compatible;
3011
3012         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3013         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3014
3015         ulen = info.nr_map_ids;
3016         info.nr_map_ids = prog->aux->used_map_cnt;
3017         ulen = min_t(u32, info.nr_map_ids, ulen);
3018         if (ulen) {
3019                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3020                 u32 i;
3021
3022                 for (i = 0; i < ulen; i++)
3023                         if (put_user(prog->aux->used_maps[i]->id,
3024                                      &user_map_ids[i]))
3025                                 return -EFAULT;
3026         }
3027
3028         err = set_info_rec_size(&info);
3029         if (err)
3030                 return err;
3031
3032         bpf_prog_get_stats(prog, &stats);
3033         info.run_time_ns = stats.nsecs;
3034         info.run_cnt = stats.cnt;
3035
3036         if (!capable(CAP_SYS_ADMIN)) {
3037                 info.jited_prog_len = 0;
3038                 info.xlated_prog_len = 0;
3039                 info.nr_jited_ksyms = 0;
3040                 info.nr_jited_func_lens = 0;
3041                 info.nr_func_info = 0;
3042                 info.nr_line_info = 0;
3043                 info.nr_jited_line_info = 0;
3044                 goto done;
3045         }
3046
3047         ulen = info.xlated_prog_len;
3048         info.xlated_prog_len = bpf_prog_insn_size(prog);
3049         if (info.xlated_prog_len && ulen) {
3050                 struct bpf_insn *insns_sanitized;
3051                 bool fault;
3052
3053                 if (prog->blinded && !bpf_dump_raw_ok()) {
3054                         info.xlated_prog_insns = 0;
3055                         goto done;
3056                 }
3057                 insns_sanitized = bpf_insn_prepare_dump(prog);
3058                 if (!insns_sanitized)
3059                         return -ENOMEM;
3060                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3061                 ulen = min_t(u32, info.xlated_prog_len, ulen);
3062                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3063                 kfree(insns_sanitized);
3064                 if (fault)
3065                         return -EFAULT;
3066         }
3067
3068         if (bpf_prog_is_dev_bound(prog->aux)) {
3069                 err = bpf_prog_offload_info_fill(&info, prog);
3070                 if (err)
3071                         return err;
3072                 goto done;
3073         }
3074
3075         /* NOTE: the following code is supposed to be skipped for offload.
3076          * bpf_prog_offload_info_fill() is the place to fill similar fields
3077          * for offload.
3078          */
3079         ulen = info.jited_prog_len;
3080         if (prog->aux->func_cnt) {
3081                 u32 i;
3082
3083                 info.jited_prog_len = 0;
3084                 for (i = 0; i < prog->aux->func_cnt; i++)
3085                         info.jited_prog_len += prog->aux->func[i]->jited_len;
3086         } else {
3087                 info.jited_prog_len = prog->jited_len;
3088         }
3089
3090         if (info.jited_prog_len && ulen) {
3091                 if (bpf_dump_raw_ok()) {
3092                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
3093                         ulen = min_t(u32, info.jited_prog_len, ulen);
3094
3095                         /* for multi-function programs, copy the JITed
3096                          * instructions for all the functions
3097                          */
3098                         if (prog->aux->func_cnt) {
3099                                 u32 len, free, i;
3100                                 u8 *img;
3101
3102                                 free = ulen;
3103                                 for (i = 0; i < prog->aux->func_cnt; i++) {
3104                                         len = prog->aux->func[i]->jited_len;
3105                                         len = min_t(u32, len, free);
3106                                         img = (u8 *) prog->aux->func[i]->bpf_func;
3107                                         if (copy_to_user(uinsns, img, len))
3108                                                 return -EFAULT;
3109                                         uinsns += len;
3110                                         free -= len;
3111                                         if (!free)
3112                                                 break;
3113                                 }
3114                         } else {
3115                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3116                                         return -EFAULT;
3117                         }
3118                 } else {
3119                         info.jited_prog_insns = 0;
3120                 }
3121         }
3122
3123         ulen = info.nr_jited_ksyms;
3124         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3125         if (ulen) {
3126                 if (bpf_dump_raw_ok()) {
3127                         unsigned long ksym_addr;
3128                         u64 __user *user_ksyms;
3129                         u32 i;
3130
3131                         /* copy the address of the kernel symbol
3132                          * corresponding to each function
3133                          */
3134                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3135                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3136                         if (prog->aux->func_cnt) {
3137                                 for (i = 0; i < ulen; i++) {
3138                                         ksym_addr = (unsigned long)
3139                                                 prog->aux->func[i]->bpf_func;
3140                                         if (put_user((u64) ksym_addr,
3141                                                      &user_ksyms[i]))
3142                                                 return -EFAULT;
3143                                 }
3144                         } else {
3145                                 ksym_addr = (unsigned long) prog->bpf_func;
3146                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3147                                         return -EFAULT;
3148                         }
3149                 } else {
3150                         info.jited_ksyms = 0;
3151                 }
3152         }
3153
3154         ulen = info.nr_jited_func_lens;
3155         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3156         if (ulen) {
3157                 if (bpf_dump_raw_ok()) {
3158                         u32 __user *user_lens;
3159                         u32 func_len, i;
3160
3161                         /* copy the JITed image lengths for each function */
3162                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3163                         user_lens = u64_to_user_ptr(info.jited_func_lens);
3164                         if (prog->aux->func_cnt) {
3165                                 for (i = 0; i < ulen; i++) {
3166                                         func_len =
3167                                                 prog->aux->func[i]->jited_len;
3168                                         if (put_user(func_len, &user_lens[i]))
3169                                                 return -EFAULT;
3170                                 }
3171                         } else {
3172                                 func_len = prog->jited_len;
3173                                 if (put_user(func_len, &user_lens[0]))
3174                                         return -EFAULT;
3175                         }
3176                 } else {
3177                         info.jited_func_lens = 0;
3178                 }
3179         }
3180
3181         if (prog->aux->btf)
3182                 info.btf_id = btf_id(prog->aux->btf);
3183
3184         ulen = info.nr_func_info;
3185         info.nr_func_info = prog->aux->func_info_cnt;
3186         if (info.nr_func_info && ulen) {
3187                 char __user *user_finfo;
3188
3189                 user_finfo = u64_to_user_ptr(info.func_info);
3190                 ulen = min_t(u32, info.nr_func_info, ulen);
3191                 if (copy_to_user(user_finfo, prog->aux->func_info,
3192                                  info.func_info_rec_size * ulen))
3193                         return -EFAULT;
3194         }
3195
3196         ulen = info.nr_line_info;
3197         info.nr_line_info = prog->aux->nr_linfo;
3198         if (info.nr_line_info && ulen) {
3199                 __u8 __user *user_linfo;
3200
3201                 user_linfo = u64_to_user_ptr(info.line_info);
3202                 ulen = min_t(u32, info.nr_line_info, ulen);
3203                 if (copy_to_user(user_linfo, prog->aux->linfo,
3204                                  info.line_info_rec_size * ulen))
3205                         return -EFAULT;
3206         }
3207
3208         ulen = info.nr_jited_line_info;
3209         if (prog->aux->jited_linfo)
3210                 info.nr_jited_line_info = prog->aux->nr_linfo;
3211         else
3212                 info.nr_jited_line_info = 0;
3213         if (info.nr_jited_line_info && ulen) {
3214                 if (bpf_dump_raw_ok()) {
3215                         __u64 __user *user_linfo;
3216                         u32 i;
3217
3218                         user_linfo = u64_to_user_ptr(info.jited_line_info);
3219                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
3220                         for (i = 0; i < ulen; i++) {
3221                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3222                                              &user_linfo[i]))
3223                                         return -EFAULT;
3224                         }
3225                 } else {
3226                         info.jited_line_info = 0;
3227                 }
3228         }
3229
3230         ulen = info.nr_prog_tags;
3231         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3232         if (ulen) {
3233                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3234                 u32 i;
3235
3236                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3237                 ulen = min_t(u32, info.nr_prog_tags, ulen);
3238                 if (prog->aux->func_cnt) {
3239                         for (i = 0; i < ulen; i++) {
3240                                 if (copy_to_user(user_prog_tags[i],
3241                                                  prog->aux->func[i]->tag,
3242                                                  BPF_TAG_SIZE))
3243                                         return -EFAULT;
3244                         }
3245                 } else {
3246                         if (copy_to_user(user_prog_tags[0],
3247                                          prog->tag, BPF_TAG_SIZE))
3248                                 return -EFAULT;
3249                 }
3250         }
3251
3252 done:
3253         if (copy_to_user(uinfo, &info, info_len) ||
3254             put_user(info_len, &uattr->info.info_len))
3255                 return -EFAULT;
3256
3257         return 0;
3258 }
3259
3260 static int bpf_map_get_info_by_fd(struct bpf_map *map,
3261                                   const union bpf_attr *attr,
3262                                   union bpf_attr __user *uattr)
3263 {
3264         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3265         struct bpf_map_info info;
3266         u32 info_len = attr->info.info_len;
3267         int err;
3268
3269         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3270         if (err)
3271                 return err;
3272         info_len = min_t(u32, sizeof(info), info_len);
3273
3274         memset(&info, 0, sizeof(info));
3275         info.type = map->map_type;
3276         info.id = map->id;
3277         info.key_size = map->key_size;
3278         info.value_size = map->value_size;
3279         info.max_entries = map->max_entries;
3280         info.map_flags = map->map_flags;
3281         memcpy(info.name, map->name, sizeof(map->name));
3282
3283         if (map->btf) {
3284                 info.btf_id = btf_id(map->btf);
3285                 info.btf_key_type_id = map->btf_key_type_id;
3286                 info.btf_value_type_id = map->btf_value_type_id;
3287         }
3288         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3289
3290         if (bpf_map_is_dev_bound(map)) {
3291                 err = bpf_map_offload_info_fill(&info, map);
3292                 if (err)
3293                         return err;
3294         }
3295
3296         if (copy_to_user(uinfo, &info, info_len) ||
3297             put_user(info_len, &uattr->info.info_len))
3298                 return -EFAULT;
3299
3300         return 0;
3301 }
3302
3303 static int bpf_btf_get_info_by_fd(struct btf *btf,
3304                                   const union bpf_attr *attr,
3305                                   union bpf_attr __user *uattr)
3306 {
3307         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3308         u32 info_len = attr->info.info_len;
3309         int err;
3310
3311         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3312         if (err)
3313                 return err;
3314
3315         return btf_get_info_by_fd(btf, attr, uattr);
3316 }
3317
3318 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3319
3320 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3321                                   union bpf_attr __user *uattr)
3322 {
3323         int ufd = attr->info.bpf_fd;
3324         struct fd f;
3325         int err;
3326
3327         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3328                 return -EINVAL;
3329
3330         f = fdget(ufd);
3331         if (!f.file)
3332                 return -EBADFD;
3333
3334         if (f.file->f_op == &bpf_prog_fops)
3335                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
3336                                               uattr);
3337         else if (f.file->f_op == &bpf_map_fops)
3338                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
3339                                              uattr);
3340         else if (f.file->f_op == &btf_fops)
3341                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
3342         else
3343                 err = -EINVAL;
3344
3345         fdput(f);
3346         return err;
3347 }
3348
3349 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3350
3351 static int bpf_btf_load(const union bpf_attr *attr)
3352 {
3353         if (CHECK_ATTR(BPF_BTF_LOAD))
3354                 return -EINVAL;
3355
3356         if (!capable(CAP_SYS_ADMIN))
3357                 return -EPERM;
3358
3359         return btf_new_fd(attr);
3360 }
3361
3362 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3363
3364 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3365 {
3366         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3367                 return -EINVAL;
3368
3369         if (!capable(CAP_SYS_ADMIN))
3370                 return -EPERM;
3371
3372         return btf_get_fd_by_id(attr->btf_id);
3373 }
3374
3375 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3376                                     union bpf_attr __user *uattr,
3377                                     u32 prog_id, u32 fd_type,
3378                                     const char *buf, u64 probe_offset,
3379                                     u64 probe_addr)
3380 {
3381         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3382         u32 len = buf ? strlen(buf) : 0, input_len;
3383         int err = 0;
3384
3385         if (put_user(len, &uattr->task_fd_query.buf_len))
3386                 return -EFAULT;
3387         input_len = attr->task_fd_query.buf_len;
3388         if (input_len && ubuf) {
3389                 if (!len) {
3390                         /* nothing to copy, just make ubuf NULL terminated */
3391                         char zero = '\0';
3392
3393                         if (put_user(zero, ubuf))
3394                                 return -EFAULT;
3395                 } else if (input_len >= len + 1) {
3396                         /* ubuf can hold the string with NULL terminator */
3397                         if (copy_to_user(ubuf, buf, len + 1))
3398                                 return -EFAULT;
3399                 } else {
3400                         /* ubuf cannot hold the string with NULL terminator,
3401                          * do a partial copy with NULL terminator.
3402                          */
3403                         char zero = '\0';
3404
3405                         err = -ENOSPC;
3406                         if (copy_to_user(ubuf, buf, input_len - 1))
3407                                 return -EFAULT;
3408                         if (put_user(zero, ubuf + input_len - 1))
3409                                 return -EFAULT;
3410                 }
3411         }
3412
3413         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3414             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3415             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3416             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3417                 return -EFAULT;
3418
3419         return err;
3420 }
3421
3422 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3423
3424 static int bpf_task_fd_query(const union bpf_attr *attr,
3425                              union bpf_attr __user *uattr)
3426 {
3427         pid_t pid = attr->task_fd_query.pid;
3428         u32 fd = attr->task_fd_query.fd;
3429         const struct perf_event *event;
3430         struct files_struct *files;
3431         struct task_struct *task;
3432         struct file *file;
3433         int err;
3434
3435         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3436                 return -EINVAL;
3437
3438         if (!capable(CAP_SYS_ADMIN))
3439                 return -EPERM;
3440
3441         if (attr->task_fd_query.flags != 0)
3442                 return -EINVAL;
3443
3444         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3445         if (!task)
3446                 return -ENOENT;
3447
3448         files = get_files_struct(task);
3449         put_task_struct(task);
3450         if (!files)
3451                 return -ENOENT;
3452
3453         err = 0;
3454         spin_lock(&files->file_lock);
3455         file = fcheck_files(files, fd);
3456         if (!file)
3457                 err = -EBADF;
3458         else
3459                 get_file(file);
3460         spin_unlock(&files->file_lock);
3461         put_files_struct(files);
3462
3463         if (err)
3464                 goto out;
3465
3466         if (file->f_op == &bpf_link_fops) {
3467                 struct bpf_link *link = file->private_data;
3468
3469                 if (link->ops == &bpf_raw_tp_lops) {
3470                         struct bpf_raw_tp_link *raw_tp =
3471                                 container_of(link, struct bpf_raw_tp_link, link);
3472                         struct bpf_raw_event_map *btp = raw_tp->btp;
3473
3474                         err = bpf_task_fd_query_copy(attr, uattr,
3475                                                      raw_tp->link.prog->aux->id,
3476                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
3477                                                      btp->tp->name, 0, 0);
3478                         goto put_file;
3479                 }
3480                 goto out_not_supp;
3481         }
3482
3483         event = perf_get_event(file);
3484         if (!IS_ERR(event)) {
3485                 u64 probe_offset, probe_addr;
3486                 u32 prog_id, fd_type;
3487                 const char *buf;
3488
3489                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3490                                               &buf, &probe_offset,
3491                                               &probe_addr);
3492                 if (!err)
3493                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3494                                                      fd_type, buf,
3495                                                      probe_offset,
3496                                                      probe_addr);
3497                 goto put_file;
3498         }
3499
3500 out_not_supp:
3501         err = -ENOTSUPP;
3502 put_file:
3503         fput(file);
3504 out:
3505         return err;
3506 }
3507
3508 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3509
3510 #define BPF_DO_BATCH(fn)                        \
3511         do {                                    \
3512                 if (!fn) {                      \
3513                         err = -ENOTSUPP;        \
3514                         goto err_put;           \
3515                 }                               \
3516                 err = fn(map, attr, uattr);     \
3517         } while (0)
3518
3519 static int bpf_map_do_batch(const union bpf_attr *attr,
3520                             union bpf_attr __user *uattr,
3521                             int cmd)
3522 {
3523         struct bpf_map *map;
3524         int err, ufd;
3525         struct fd f;
3526
3527         if (CHECK_ATTR(BPF_MAP_BATCH))
3528                 return -EINVAL;
3529
3530         ufd = attr->batch.map_fd;
3531         f = fdget(ufd);
3532         map = __bpf_map_get(f);
3533         if (IS_ERR(map))
3534                 return PTR_ERR(map);
3535
3536         if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3537              cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3538             !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3539                 err = -EPERM;
3540                 goto err_put;
3541         }
3542
3543         if (cmd != BPF_MAP_LOOKUP_BATCH &&
3544             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3545                 err = -EPERM;
3546                 goto err_put;
3547         }
3548
3549         if (cmd == BPF_MAP_LOOKUP_BATCH)
3550                 BPF_DO_BATCH(map->ops->map_lookup_batch);
3551         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3552                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3553         else if (cmd == BPF_MAP_UPDATE_BATCH)
3554                 BPF_DO_BATCH(map->ops->map_update_batch);
3555         else
3556                 BPF_DO_BATCH(map->ops->map_delete_batch);
3557
3558 err_put:
3559         fdput(f);
3560         return err;
3561 }
3562
3563 #define BPF_LINK_CREATE_LAST_FIELD link_create.flags
3564 static int link_create(union bpf_attr *attr)
3565 {
3566         enum bpf_prog_type ptype;
3567         struct bpf_prog *prog;
3568         int ret;
3569
3570         if (!capable(CAP_NET_ADMIN))
3571                 return -EPERM;
3572
3573         if (CHECK_ATTR(BPF_LINK_CREATE))
3574                 return -EINVAL;
3575
3576         ptype = attach_type_to_prog_type(attr->link_create.attach_type);
3577         if (ptype == BPF_PROG_TYPE_UNSPEC)
3578                 return -EINVAL;
3579
3580         prog = bpf_prog_get_type(attr->link_create.prog_fd, ptype);
3581         if (IS_ERR(prog))
3582                 return PTR_ERR(prog);
3583
3584         ret = bpf_prog_attach_check_attach_type(prog,
3585                                                 attr->link_create.attach_type);
3586         if (ret)
3587                 goto err_out;
3588
3589         switch (ptype) {
3590         case BPF_PROG_TYPE_CGROUP_SKB:
3591         case BPF_PROG_TYPE_CGROUP_SOCK:
3592         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3593         case BPF_PROG_TYPE_SOCK_OPS:
3594         case BPF_PROG_TYPE_CGROUP_DEVICE:
3595         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3596         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3597                 ret = cgroup_bpf_link_attach(attr, prog);
3598                 break;
3599         default:
3600                 ret = -EINVAL;
3601         }
3602
3603 err_out:
3604         if (ret < 0)
3605                 bpf_prog_put(prog);
3606         return ret;
3607 }
3608
3609 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
3610
3611 static int link_update(union bpf_attr *attr)
3612 {
3613         struct bpf_prog *old_prog = NULL, *new_prog;
3614         struct bpf_link *link;
3615         u32 flags;
3616         int ret;
3617
3618         if (!capable(CAP_NET_ADMIN))
3619                 return -EPERM;
3620
3621         if (CHECK_ATTR(BPF_LINK_UPDATE))
3622                 return -EINVAL;
3623
3624         flags = attr->link_update.flags;
3625         if (flags & ~BPF_F_REPLACE)
3626                 return -EINVAL;
3627
3628         link = bpf_link_get_from_fd(attr->link_update.link_fd);
3629         if (IS_ERR(link))
3630                 return PTR_ERR(link);
3631
3632         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
3633         if (IS_ERR(new_prog))
3634                 return PTR_ERR(new_prog);
3635
3636         if (flags & BPF_F_REPLACE) {
3637                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
3638                 if (IS_ERR(old_prog)) {
3639                         ret = PTR_ERR(old_prog);
3640                         old_prog = NULL;
3641                         goto out_put_progs;
3642                 }
3643         }
3644
3645 #ifdef CONFIG_CGROUP_BPF
3646         if (link->ops == &bpf_cgroup_link_lops) {
3647                 ret = cgroup_bpf_replace(link, old_prog, new_prog);
3648                 goto out_put_progs;
3649         }
3650 #endif
3651         ret = -EINVAL;
3652
3653 out_put_progs:
3654         if (old_prog)
3655                 bpf_prog_put(old_prog);
3656         if (ret)
3657                 bpf_prog_put(new_prog);
3658         return ret;
3659 }
3660
3661 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3662 {
3663         union bpf_attr attr;
3664         int err;
3665
3666         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3667                 return -EPERM;
3668
3669         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3670         if (err)
3671                 return err;
3672         size = min_t(u32, size, sizeof(attr));
3673
3674         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3675         memset(&attr, 0, sizeof(attr));
3676         if (copy_from_user(&attr, uattr, size) != 0)
3677                 return -EFAULT;
3678
3679         err = security_bpf(cmd, &attr, size);
3680         if (err < 0)
3681                 return err;
3682
3683         switch (cmd) {
3684         case BPF_MAP_CREATE:
3685                 err = map_create(&attr);
3686                 break;
3687         case BPF_MAP_LOOKUP_ELEM:
3688                 err = map_lookup_elem(&attr);
3689                 break;
3690         case BPF_MAP_UPDATE_ELEM:
3691                 err = map_update_elem(&attr);
3692                 break;
3693         case BPF_MAP_DELETE_ELEM:
3694                 err = map_delete_elem(&attr);
3695                 break;
3696         case BPF_MAP_GET_NEXT_KEY:
3697                 err = map_get_next_key(&attr);
3698                 break;
3699         case BPF_MAP_FREEZE:
3700                 err = map_freeze(&attr);
3701                 break;
3702         case BPF_PROG_LOAD:
3703                 err = bpf_prog_load(&attr, uattr);
3704                 break;
3705         case BPF_OBJ_PIN:
3706                 err = bpf_obj_pin(&attr);
3707                 break;
3708         case BPF_OBJ_GET:
3709                 err = bpf_obj_get(&attr);
3710                 break;
3711         case BPF_PROG_ATTACH:
3712                 err = bpf_prog_attach(&attr);
3713                 break;
3714         case BPF_PROG_DETACH:
3715                 err = bpf_prog_detach(&attr);
3716                 break;
3717         case BPF_PROG_QUERY:
3718                 err = bpf_prog_query(&attr, uattr);
3719                 break;
3720         case BPF_PROG_TEST_RUN:
3721                 err = bpf_prog_test_run(&attr, uattr);
3722                 break;
3723         case BPF_PROG_GET_NEXT_ID:
3724                 err = bpf_obj_get_next_id(&attr, uattr,
3725                                           &prog_idr, &prog_idr_lock);
3726                 break;
3727         case BPF_MAP_GET_NEXT_ID:
3728                 err = bpf_obj_get_next_id(&attr, uattr,
3729                                           &map_idr, &map_idr_lock);
3730                 break;
3731         case BPF_BTF_GET_NEXT_ID:
3732                 err = bpf_obj_get_next_id(&attr, uattr,
3733                                           &btf_idr, &btf_idr_lock);
3734                 break;
3735         case BPF_PROG_GET_FD_BY_ID:
3736                 err = bpf_prog_get_fd_by_id(&attr);
3737                 break;
3738         case BPF_MAP_GET_FD_BY_ID:
3739                 err = bpf_map_get_fd_by_id(&attr);
3740                 break;
3741         case BPF_OBJ_GET_INFO_BY_FD:
3742                 err = bpf_obj_get_info_by_fd(&attr, uattr);
3743                 break;
3744         case BPF_RAW_TRACEPOINT_OPEN:
3745                 err = bpf_raw_tracepoint_open(&attr);
3746                 break;
3747         case BPF_BTF_LOAD:
3748                 err = bpf_btf_load(&attr);
3749                 break;
3750         case BPF_BTF_GET_FD_BY_ID:
3751                 err = bpf_btf_get_fd_by_id(&attr);
3752                 break;
3753         case BPF_TASK_FD_QUERY:
3754                 err = bpf_task_fd_query(&attr, uattr);
3755                 break;
3756         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3757                 err = map_lookup_and_delete_elem(&attr);
3758                 break;
3759         case BPF_MAP_LOOKUP_BATCH:
3760                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
3761                 break;
3762         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
3763                 err = bpf_map_do_batch(&attr, uattr,
3764                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
3765                 break;
3766         case BPF_MAP_UPDATE_BATCH:
3767                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
3768                 break;
3769         case BPF_MAP_DELETE_BATCH:
3770                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
3771                 break;
3772         case BPF_LINK_CREATE:
3773                 err = link_create(&attr);
3774                 break;
3775         case BPF_LINK_UPDATE:
3776                 err = link_update(&attr);
3777                 break;
3778         default:
3779                 err = -EINVAL;
3780                 break;
3781         }
3782
3783         return err;
3784 }