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