kernel/bpf/syscall.c

   1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  *
   7  * This program is distributed in the hope that it will be useful, but
   8  * WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  10  * General Public License for more details.
  11  */
  12 #include <linux/bpf.h>
  13 #include <linux/syscalls.h>
  14 #include <linux/slab.h>
  15 #include <linux/anon_inodes.h>
  16 #include <linux/file.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20
  21 DEFINE_PER_CPU(int, bpf_prog_active);
  22
  23 int sysctl_unprivileged_bpf_disabled __read_mostly;
  24
  25 static LIST_HEAD(bpf_map_types);
  26
  27 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
  28 {
  29         struct bpf_map_type_list *tl;
  30         struct bpf_map *map;
  31
  32         list_for_each_entry(tl, &bpf_map_types, list_node) {
  33                 if (tl->type == attr->map_type) {
  34                         map = tl->ops->map_alloc(attr);
  35                         if (IS_ERR(map))
  36                                 return map;
  37                         map->ops = tl->ops;
  38                         map->map_type = attr->map_type;
  39                         return map;
  40                 }
  41         }
  42         return ERR_PTR(-EINVAL);
  43 }
  44
  45 /* boot time registration of different map implementations */
  46 void bpf_register_map_type(struct bpf_map_type_list *tl)
  47 {
  48         list_add(&tl->list_node, &bpf_map_types);
  49 }
  50
  51 int bpf_map_precharge_memlock(u32 pages)
  52 {
  53         struct user_struct *user = get_current_user();
  54         unsigned long memlock_limit, cur;
  55
  56         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  57         cur = atomic_long_read(&user->locked_vm);
  58         free_uid(user);
  59         if (cur + pages > memlock_limit)
  60                 return -EPERM;
  61         return 0;
  62 }
  63
  64 static int bpf_map_charge_memlock(struct bpf_map *map)
  65 {
  66         struct user_struct *user = get_current_user();
  67         unsigned long memlock_limit;
  68
  69         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  70
  71         atomic_long_add(map->pages, &user->locked_vm);
  72
  73         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
  74                 atomic_long_sub(map->pages, &user->locked_vm);
  75                 free_uid(user);
  76                 return -EPERM;
  77         }
  78         map->user = user;
  79         return 0;
  80 }
  81
  82 static void bpf_map_uncharge_memlock(struct bpf_map *map)
  83 {
  84         struct user_struct *user = map->user;
  85
  86         atomic_long_sub(map->pages, &user->locked_vm);
  87         free_uid(user);
  88 }
  89
  90 /* called from workqueue */
  91 static void bpf_map_free_deferred(struct work_struct *work)
  92 {
  93         struct bpf_map *map = container_of(work, struct bpf_map, work);
  94
  95         bpf_map_uncharge_memlock(map);
  96         /* implementation dependent freeing */
  97         map->ops->map_free(map);
  98 }
  99
 100 static void bpf_map_put_uref(struct bpf_map *map)
 101 {
 102         if (atomic_dec_and_test(&map->usercnt)) {
 103                 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
 104                         bpf_fd_array_map_clear(map);
 105         }
 106 }
 107
 108 /* decrement map refcnt and schedule it for freeing via workqueue
 109  * (unrelying map implementation ops->map_free() might sleep)
 110  */
 111 void bpf_map_put(struct bpf_map *map)
 112 {
 113         if (atomic_dec_and_test(&map->refcnt)) {
 114                 INIT_WORK(&map->work, bpf_map_free_deferred);
 115                 schedule_work(&map->work);
 116         }
 117 }
 118
 119 void bpf_map_put_with_uref(struct bpf_map *map)
 120 {
 121         bpf_map_put_uref(map);
 122         bpf_map_put(map);
 123 }
 124
 125 static int bpf_map_release(struct inode *inode, struct file *filp)
 126 {
 127         struct bpf_map *map = filp->private_data;
 128
 129         if (map->ops->map_release)
 130                 map->ops->map_release(map, filp);
 131
 132         bpf_map_put_with_uref(map);
 133         return 0;
 134 }
 135
 136 #ifdef CONFIG_PROC_FS
 137 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 138 {
 139         const struct bpf_map *map = filp->private_data;
 140
 141         seq_printf(m,
 142                    "map_type:\t%u\n"
 143                    "key_size:\t%u\n"
 144                    "value_size:\t%u\n"
 145                    "max_entries:\t%u\n"
 146                    "map_flags:\t%#x\n",
 147                    map->map_type,
 148                    map->key_size,
 149                    map->value_size,
 150                    map->max_entries,
 151                    map->map_flags);
 152 }
 153 #endif
 154
 155 static const struct file_operations bpf_map_fops = {
 156 #ifdef CONFIG_PROC_FS
 157         .show_fdinfo    = bpf_map_show_fdinfo,
 158 #endif
 159         .release        = bpf_map_release,
 160 };
 161
 162 int bpf_map_new_fd(struct bpf_map *map)
 163 {
 164         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 165                                 O_RDWR | O_CLOEXEC);
 166 }
 167
 168 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 169 #define CHECK_ATTR(CMD) \
 170         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 171                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 172                    sizeof(*attr) - \
 173                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 174                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 175
 176 #define BPF_MAP_CREATE_LAST_FIELD map_flags
 177 /* called via syscall */
 178 static int map_create(union bpf_attr *attr)
 179 {
 180         struct bpf_map *map;
 181         int err;
 182
 183         err = CHECK_ATTR(BPF_MAP_CREATE);
 184         if (err)
 185                 return -EINVAL;
 186
 187         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 188         map = find_and_alloc_map(attr);
 189         if (IS_ERR(map))
 190                 return PTR_ERR(map);
 191
 192         atomic_set(&map->refcnt, 1);
 193         atomic_set(&map->usercnt, 1);
 194
 195         err = bpf_map_charge_memlock(map);
 196         if (err)
 197                 goto free_map;
 198
 199         err = bpf_map_new_fd(map);
 200         if (err < 0)
 201                 /* failed to allocate fd */
 202                 goto free_map;
 203
 204         return err;
 205
 206 free_map:
 207         map->ops->map_free(map);
 208         return err;
 209 }
 210
 211 /* if error is returned, fd is released.
 212  * On success caller should complete fd access with matching fdput()
 213  */
 214 struct bpf_map *__bpf_map_get(struct fd f)
 215 {
 216         if (!f.file)
 217                 return ERR_PTR(-EBADF);
 218         if (f.file->f_op != &bpf_map_fops) {
 219                 fdput(f);
 220                 return ERR_PTR(-EINVAL);
 221         }
 222
 223         return f.file->private_data;
 224 }
 225
 226 /* prog's and map's refcnt limit */
 227 #define BPF_MAX_REFCNT 32768
 228
 229 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 230 {
 231         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 232                 atomic_dec(&map->refcnt);
 233                 return ERR_PTR(-EBUSY);
 234         }
 235         if (uref)
 236                 atomic_inc(&map->usercnt);
 237         return map;
 238 }
 239
 240 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 241 {
 242         struct fd f = fdget(ufd);
 243         struct bpf_map *map;
 244
 245         map = __bpf_map_get(f);
 246         if (IS_ERR(map))
 247                 return map;
 248
 249         map = bpf_map_inc(map, true);
 250         fdput(f);
 251
 252         return map;
 253 }
 254
 255 /* helper to convert user pointers passed inside __aligned_u64 fields */
 256 static void __user *u64_to_ptr(__u64 val)
 257 {
 258         return (void __user *) (unsigned long) val;
 259 }
 260
 261 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 262 {
 263         return -ENOTSUPP;
 264 }
 265
 266 /* last field in 'union bpf_attr' used by this command */
 267 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 268
 269 static int map_lookup_elem(union bpf_attr *attr)
 270 {
 271         void __user *ukey = u64_to_ptr(attr->key);
 272         void __user *uvalue = u64_to_ptr(attr->value);
 273         int ufd = attr->map_fd;
 274         struct bpf_map *map;
 275         void *key, *value, *ptr;
 276         u32 value_size;
 277         struct fd f;
 278         int err;
 279
 280         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 281                 return -EINVAL;
 282
 283         f = fdget(ufd);
 284         map = __bpf_map_get(f);
 285         if (IS_ERR(map))
 286                 return PTR_ERR(map);
 287
 288         err = -ENOMEM;
 289         key = kmalloc(map->key_size, GFP_USER);
 290         if (!key)
 291                 goto err_put;
 292
 293         err = -EFAULT;
 294         if (copy_from_user(key, ukey, map->key_size) != 0)
 295                 goto free_key;
 296
 297         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 298             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 299                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 300         else
 301                 value_size = map->value_size;
 302
 303         err = -ENOMEM;
 304         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 305         if (!value)
 306                 goto free_key;
 307
 308         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
 309                 err = bpf_percpu_hash_copy(map, key, value);
 310         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 311                 err = bpf_percpu_array_copy(map, key, value);
 312         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 313                 err = bpf_stackmap_copy(map, key, value);
 314         } else {
 315                 rcu_read_lock();
 316                 ptr = map->ops->map_lookup_elem(map, key);
 317                 if (ptr)
 318                         memcpy(value, ptr, value_size);
 319                 rcu_read_unlock();
 320                 err = ptr ? 0 : -ENOENT;
 321         }
 322
 323         if (err)
 324                 goto free_value;
 325
 326         err = -EFAULT;
 327         if (copy_to_user(uvalue, value, value_size) != 0)
 328                 goto free_value;
 329
 330         err = 0;
 331
 332 free_value:
 333         kfree(value);
 334 free_key:
 335         kfree(key);
 336 err_put:
 337         fdput(f);
 338         return err;
 339 }
 340
 341 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 342
 343 static int map_update_elem(union bpf_attr *attr)
 344 {
 345         void __user *ukey = u64_to_ptr(attr->key);
 346         void __user *uvalue = u64_to_ptr(attr->value);
 347         int ufd = attr->map_fd;
 348         struct bpf_map *map;
 349         void *key, *value;
 350         u32 value_size;
 351         struct fd f;
 352         int err;
 353
 354         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 355                 return -EINVAL;
 356
 357         f = fdget(ufd);
 358         map = __bpf_map_get(f);
 359         if (IS_ERR(map))
 360                 return PTR_ERR(map);
 361
 362         err = -ENOMEM;
 363         key = kmalloc(map->key_size, GFP_USER);
 364         if (!key)
 365                 goto err_put;
 366
 367         err = -EFAULT;
 368         if (copy_from_user(key, ukey, map->key_size) != 0)
 369                 goto free_key;
 370
 371         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 372             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 373                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 374         else
 375                 value_size = map->value_size;
 376
 377         err = -ENOMEM;
 378         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 379         if (!value)
 380                 goto free_key;
 381
 382         err = -EFAULT;
 383         if (copy_from_user(value, uvalue, value_size) != 0)
 384                 goto free_value;
 385
 386         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 387          * inside bpf map update or delete otherwise deadlocks are possible
 388          */
 389         preempt_disable();
 390         __this_cpu_inc(bpf_prog_active);
 391         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
 392                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 393         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 394                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 395         } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
 396                    map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 397                 rcu_read_lock();
 398                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 399                                                    attr->flags);
 400                 rcu_read_unlock();
 401         } else {
 402                 rcu_read_lock();
 403                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 404                 rcu_read_unlock();
 405         }
 406         __this_cpu_dec(bpf_prog_active);
 407         preempt_enable();
 408
 409 free_value:
 410         kfree(value);
 411 free_key:
 412         kfree(key);
 413 err_put:
 414         fdput(f);
 415         return err;
 416 }
 417
 418 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 419
 420 static int map_delete_elem(union bpf_attr *attr)
 421 {
 422         void __user *ukey = u64_to_ptr(attr->key);
 423         int ufd = attr->map_fd;
 424         struct bpf_map *map;
 425         struct fd f;
 426         void *key;
 427         int err;
 428
 429         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 430                 return -EINVAL;
 431
 432         f = fdget(ufd);
 433         map = __bpf_map_get(f);
 434         if (IS_ERR(map))
 435                 return PTR_ERR(map);
 436
 437         err = -ENOMEM;
 438         key = kmalloc(map->key_size, GFP_USER);
 439         if (!key)
 440                 goto err_put;
 441
 442         err = -EFAULT;
 443         if (copy_from_user(key, ukey, map->key_size) != 0)
 444                 goto free_key;
 445
 446         preempt_disable();
 447         __this_cpu_inc(bpf_prog_active);
 448         rcu_read_lock();
 449         err = map->ops->map_delete_elem(map, key);
 450         rcu_read_unlock();
 451         __this_cpu_dec(bpf_prog_active);
 452         preempt_enable();
 453
 454 free_key:
 455         kfree(key);
 456 err_put:
 457         fdput(f);
 458         return err;
 459 }
 460
 461 /* last field in 'union bpf_attr' used by this command */
 462 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 463
 464 static int map_get_next_key(union bpf_attr *attr)
 465 {
 466         void __user *ukey = u64_to_ptr(attr->key);
 467         void __user *unext_key = u64_to_ptr(attr->next_key);
 468         int ufd = attr->map_fd;
 469         struct bpf_map *map;
 470         void *key, *next_key;
 471         struct fd f;
 472         int err;
 473
 474         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 475                 return -EINVAL;
 476
 477         f = fdget(ufd);
 478         map = __bpf_map_get(f);
 479         if (IS_ERR(map))
 480                 return PTR_ERR(map);
 481
 482         err = -ENOMEM;
 483         key = kmalloc(map->key_size, GFP_USER);
 484         if (!key)
 485                 goto err_put;
 486
 487         err = -EFAULT;
 488         if (copy_from_user(key, ukey, map->key_size) != 0)
 489                 goto free_key;
 490
 491         err = -ENOMEM;
 492         next_key = kmalloc(map->key_size, GFP_USER);
 493         if (!next_key)
 494                 goto free_key;
 495
 496         rcu_read_lock();
 497         err = map->ops->map_get_next_key(map, key, next_key);
 498         rcu_read_unlock();
 499         if (err)
 500                 goto free_next_key;
 501
 502         err = -EFAULT;
 503         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 504                 goto free_next_key;
 505
 506         err = 0;
 507
 508 free_next_key:
 509         kfree(next_key);
 510 free_key:
 511         kfree(key);
 512 err_put:
 513         fdput(f);
 514         return err;
 515 }
 516
 517 static LIST_HEAD(bpf_prog_types);
 518
 519 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 520 {
 521         struct bpf_prog_type_list *tl;
 522
 523         list_for_each_entry(tl, &bpf_prog_types, list_node) {
 524                 if (tl->type == type) {
 525                         prog->aux->ops = tl->ops;
 526                         prog->type = type;
 527                         return 0;
 528                 }
 529         }
 530
 531         return -EINVAL;
 532 }
 533
 534 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
 535 {
 536         list_add(&tl->list_node, &bpf_prog_types);
 537 }
 538
 539 /* fixup insn->imm field of bpf_call instructions:
 540  * if (insn->imm == BPF_FUNC_map_lookup_elem)
 541  *      insn->imm = bpf_map_lookup_elem - __bpf_call_base;
 542  * else if (insn->imm == BPF_FUNC_map_update_elem)
 543  *      insn->imm = bpf_map_update_elem - __bpf_call_base;
 544  * else ...
 545  *
 546  * this function is called after eBPF program passed verification
 547  */
 548 static void fixup_bpf_calls(struct bpf_prog *prog)
 549 {
 550         const struct bpf_func_proto *fn;
 551         int i;
 552
 553         for (i = 0; i < prog->len; i++) {
 554                 struct bpf_insn *insn = &prog->insnsi[i];
 555
 556                 if (insn->code == (BPF_JMP | BPF_CALL)) {
 557                         /* we reach here when program has bpf_call instructions
 558                          * and it passed bpf_check(), means that
 559                          * ops->get_func_proto must have been supplied, check it
 560                          */
 561                         BUG_ON(!prog->aux->ops->get_func_proto);
 562
 563                         if (insn->imm == BPF_FUNC_get_route_realm)
 564                                 prog->dst_needed = 1;
 565                         if (insn->imm == BPF_FUNC_get_prandom_u32)
 566                                 bpf_user_rnd_init_once();
 567                         if (insn->imm == BPF_FUNC_tail_call) {
 568                                 /* mark bpf_tail_call as different opcode
 569                                  * to avoid conditional branch in
 570                                  * interpeter for every normal call
 571                                  * and to prevent accidental JITing by
 572                                  * JIT compiler that doesn't support
 573                                  * bpf_tail_call yet
 574                                  */
 575                                 insn->imm = 0;
 576                                 insn->code |= BPF_X;
 577                                 continue;
 578                         }
 579
 580                         fn = prog->aux->ops->get_func_proto(insn->imm);
 581                         /* all functions that have prototype and verifier allowed
 582                          * programs to call them, must be real in-kernel functions
 583                          */
 584                         BUG_ON(!fn->func);
 585                         insn->imm = fn->func - __bpf_call_base;
 586                 }
 587         }
 588 }
 589
 590 /* drop refcnt on maps used by eBPF program and free auxilary data */
 591 static void free_used_maps(struct bpf_prog_aux *aux)
 592 {
 593         int i;
 594
 595         for (i = 0; i < aux->used_map_cnt; i++)
 596                 bpf_map_put(aux->used_maps[i]);
 597
 598         kfree(aux->used_maps);
 599 }
 600
 601 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
 602 {
 603         struct user_struct *user = get_current_user();
 604         unsigned long memlock_limit;
 605
 606         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 607
 608         atomic_long_add(prog->pages, &user->locked_vm);
 609         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 610                 atomic_long_sub(prog->pages, &user->locked_vm);
 611                 free_uid(user);
 612                 return -EPERM;
 613         }
 614         prog->aux->user = user;
 615         return 0;
 616 }
 617
 618 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
 619 {
 620         struct user_struct *user = prog->aux->user;
 621
 622         atomic_long_sub(prog->pages, &user->locked_vm);
 623         free_uid(user);
 624 }
 625
 626 static void __prog_put_common(struct rcu_head *rcu)
 627 {
 628         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 629
 630         free_used_maps(aux);
 631         bpf_prog_uncharge_memlock(aux->prog);
 632         bpf_prog_free(aux->prog);
 633 }
 634
 635 /* version of bpf_prog_put() that is called after a grace period */
 636 void bpf_prog_put_rcu(struct bpf_prog *prog)
 637 {
 638         if (atomic_dec_and_test(&prog->aux->refcnt))
 639                 call_rcu(&prog->aux->rcu, __prog_put_common);
 640 }
 641
 642 void bpf_prog_put(struct bpf_prog *prog)
 643 {
 644         if (atomic_dec_and_test(&prog->aux->refcnt))
 645                 __prog_put_common(&prog->aux->rcu);
 646 }
 647 EXPORT_SYMBOL_GPL(bpf_prog_put);
 648
 649 static int bpf_prog_release(struct inode *inode, struct file *filp)
 650 {
 651         struct bpf_prog *prog = filp->private_data;
 652
 653         bpf_prog_put_rcu(prog);
 654         return 0;
 655 }
 656
 657 static const struct file_operations bpf_prog_fops = {
 658         .release = bpf_prog_release,
 659 };
 660
 661 int bpf_prog_new_fd(struct bpf_prog *prog)
 662 {
 663         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
 664                                 O_RDWR | O_CLOEXEC);
 665 }
 666
 667 static struct bpf_prog *__bpf_prog_get(struct fd f)
 668 {
 669         if (!f.file)
 670                 return ERR_PTR(-EBADF);
 671         if (f.file->f_op != &bpf_prog_fops) {
 672                 fdput(f);
 673                 return ERR_PTR(-EINVAL);
 674         }
 675
 676         return f.file->private_data;
 677 }
 678
 679 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
 680 {
 681         if (atomic_inc_return(&prog->aux->refcnt) > BPF_MAX_REFCNT) {
 682                 atomic_dec(&prog->aux->refcnt);
 683                 return ERR_PTR(-EBUSY);
 684         }
 685         return prog;
 686 }
 687
 688 /* called by sockets/tracing/seccomp before attaching program to an event
 689  * pairs with bpf_prog_put()
 690  */
 691 struct bpf_prog *bpf_prog_get(u32 ufd)
 692 {
 693         struct fd f = fdget(ufd);
 694         struct bpf_prog *prog;
 695
 696         prog = __bpf_prog_get(f);
 697         if (IS_ERR(prog))
 698                 return prog;
 699
 700         prog = bpf_prog_inc(prog);
 701         fdput(f);
 702
 703         return prog;
 704 }
 705 EXPORT_SYMBOL_GPL(bpf_prog_get);
 706
 707 /* last field in 'union bpf_attr' used by this command */
 708 #define BPF_PROG_LOAD_LAST_FIELD kern_version
 709
 710 static int bpf_prog_load(union bpf_attr *attr)
 711 {
 712         enum bpf_prog_type type = attr->prog_type;
 713         struct bpf_prog *prog;
 714         int err;
 715         char license[128];
 716         bool is_gpl;
 717
 718         if (CHECK_ATTR(BPF_PROG_LOAD))
 719                 return -EINVAL;
 720
 721         /* copy eBPF program license from user space */
 722         if (strncpy_from_user(license, u64_to_ptr(attr->license),
 723                               sizeof(license) - 1) < 0)
 724                 return -EFAULT;
 725         license[sizeof(license) - 1] = 0;
 726
 727         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 728         is_gpl = license_is_gpl_compatible(license);
 729
 730         if (attr->insn_cnt >= BPF_MAXINSNS)
 731                 return -EINVAL;
 732
 733         if (type == BPF_PROG_TYPE_KPROBE &&
 734             attr->kern_version != LINUX_VERSION_CODE)
 735                 return -EINVAL;
 736
 737         if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
 738                 return -EPERM;
 739
 740         /* plain bpf_prog allocation */
 741         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
 742         if (!prog)
 743                 return -ENOMEM;
 744
 745         err = bpf_prog_charge_memlock(prog);
 746         if (err)
 747                 goto free_prog_nouncharge;
 748
 749         prog->len = attr->insn_cnt;
 750
 751         err = -EFAULT;
 752         if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
 753                            prog->len * sizeof(struct bpf_insn)) != 0)
 754                 goto free_prog;
 755
 756         prog->orig_prog = NULL;
 757         prog->jited = 0;
 758
 759         atomic_set(&prog->aux->refcnt, 1);
 760         prog->gpl_compatible = is_gpl ? 1 : 0;
 761
 762         /* find program type: socket_filter vs tracing_filter */
 763         err = find_prog_type(type, prog);
 764         if (err < 0)
 765                 goto free_prog;
 766
 767         /* run eBPF verifier */
 768         err = bpf_check(&prog, attr);
 769         if (err < 0)
 770                 goto free_used_maps;
 771
 772         /* fixup BPF_CALL->imm field */
 773         fixup_bpf_calls(prog);
 774
 775         /* eBPF program is ready to be JITed */
 776         prog = bpf_prog_select_runtime(prog, &err);
 777         if (err < 0)
 778                 goto free_used_maps;
 779
 780         err = bpf_prog_new_fd(prog);
 781         if (err < 0)
 782                 /* failed to allocate fd */
 783                 goto free_used_maps;
 784
 785         return err;
 786
 787 free_used_maps:
 788         free_used_maps(prog->aux);
 789 free_prog:
 790         bpf_prog_uncharge_memlock(prog);
 791 free_prog_nouncharge:
 792         bpf_prog_free(prog);
 793         return err;
 794 }
 795
 796 #define BPF_OBJ_LAST_FIELD bpf_fd
 797
 798 static int bpf_obj_pin(const union bpf_attr *attr)
 799 {
 800         if (CHECK_ATTR(BPF_OBJ))
 801                 return -EINVAL;
 802
 803         return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname));
 804 }
 805
 806 static int bpf_obj_get(const union bpf_attr *attr)
 807 {
 808         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
 809                 return -EINVAL;
 810
 811         return bpf_obj_get_user(u64_to_ptr(attr->pathname));
 812 }
 813
 814 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 815 {
 816         union bpf_attr attr = {};
 817         int err;
 818
 819         if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
 820                 return -EPERM;
 821
 822         if (!access_ok(VERIFY_READ, uattr, 1))
 823                 return -EFAULT;
 824
 825         if (size > PAGE_SIZE)   /* silly large */
 826                 return -E2BIG;
 827
 828         /* If we're handed a bigger struct than we know of,
 829          * ensure all the unknown bits are 0 - i.e. new
 830          * user-space does not rely on any kernel feature
 831          * extensions we dont know about yet.
 832          */
 833         if (size > sizeof(attr)) {
 834                 unsigned char __user *addr;
 835                 unsigned char __user *end;
 836                 unsigned char val;
 837
 838                 addr = (void __user *)uattr + sizeof(attr);
 839                 end  = (void __user *)uattr + size;
 840
 841                 for (; addr < end; addr++) {
 842                         err = get_user(val, addr);
 843                         if (err)
 844                                 return err;
 845                         if (val)
 846                                 return -E2BIG;
 847                 }
 848                 size = sizeof(attr);
 849         }
 850
 851         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
 852         if (copy_from_user(&attr, uattr, size) != 0)
 853                 return -EFAULT;
 854
 855         switch (cmd) {
 856         case BPF_MAP_CREATE:
 857                 err = map_create(&attr);
 858                 break;
 859         case BPF_MAP_LOOKUP_ELEM:
 860                 err = map_lookup_elem(&attr);
 861                 break;
 862         case BPF_MAP_UPDATE_ELEM:
 863                 err = map_update_elem(&attr);
 864                 break;
 865         case BPF_MAP_DELETE_ELEM:
 866                 err = map_delete_elem(&attr);
 867                 break;
 868         case BPF_MAP_GET_NEXT_KEY:
 869                 err = map_get_next_key(&attr);
 870                 break;
 871         case BPF_PROG_LOAD:
 872                 err = bpf_prog_load(&attr);
 873                 break;
 874         case BPF_OBJ_PIN:
 875                 err = bpf_obj_pin(&attr);
 876                 break;
 877         case BPF_OBJ_GET:
 878                 err = bpf_obj_get(&attr);
 879                 break;
 880         default:
 881                 err = -EINVAL;
 882                 break;
 883         }
 884
 885         return err;
 886 }