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