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 static LIST_HEAD(bpf_map_types);
  22
  23 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
  24 {
  25         struct bpf_map_type_list *tl;
  26         struct bpf_map *map;
  27
  28         list_for_each_entry(tl, &bpf_map_types, list_node) {
  29                 if (tl->type == attr->map_type) {
  30                         map = tl->ops->map_alloc(attr);
  31                         if (IS_ERR(map))
  32                                 return map;
  33                         map->ops = tl->ops;
  34                         map->map_type = attr->map_type;
  35                         return map;
  36                 }
  37         }
  38         return ERR_PTR(-EINVAL);
  39 }
  40
  41 /* boot time registration of different map implementations */
  42 void bpf_register_map_type(struct bpf_map_type_list *tl)
  43 {
  44         list_add(&tl->list_node, &bpf_map_types);
  45 }
  46
  47 /* called from workqueue */
  48 static void bpf_map_free_deferred(struct work_struct *work)
  49 {
  50         struct bpf_map *map = container_of(work, struct bpf_map, work);
  51
  52         /* implementation dependent freeing */
  53         map->ops->map_free(map);
  54 }
  55
  56 /* decrement map refcnt and schedule it for freeing via workqueue
  57  * (unrelying map implementation ops->map_free() might sleep)
  58  */
  59 void bpf_map_put(struct bpf_map *map)
  60 {
  61         if (atomic_dec_and_test(&map->refcnt)) {
  62                 INIT_WORK(&map->work, bpf_map_free_deferred);
  63                 schedule_work(&map->work);
  64         }
  65 }
  66
  67 static int bpf_map_release(struct inode *inode, struct file *filp)
  68 {
  69         struct bpf_map *map = filp->private_data;
  70
  71         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  72                 /* prog_array stores refcnt-ed bpf_prog pointers
  73                  * release them all when user space closes prog_array_fd
  74                  */
  75                 bpf_prog_array_map_clear(map);
  76
  77         bpf_map_put(map);
  78         return 0;
  79 }
  80
  81 static const struct file_operations bpf_map_fops = {
  82         .release = bpf_map_release,
  83 };
  84
  85 /* helper macro to check that unused fields 'union bpf_attr' are zero */
  86 #define CHECK_ATTR(CMD) \
  87         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
  88                    sizeof(attr->CMD##_LAST_FIELD), 0, \
  89                    sizeof(*attr) - \
  90                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
  91                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
  92
  93 #define BPF_MAP_CREATE_LAST_FIELD max_entries
  94 /* called via syscall */
  95 static int map_create(union bpf_attr *attr)
  96 {
  97         struct bpf_map *map;
  98         int err;
  99
 100         err = CHECK_ATTR(BPF_MAP_CREATE);
 101         if (err)
 102                 return -EINVAL;
 103
 104         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 105         map = find_and_alloc_map(attr);
 106         if (IS_ERR(map))
 107                 return PTR_ERR(map);
 108
 109         atomic_set(&map->refcnt, 1);
 110
 111         err = anon_inode_getfd("bpf-map", &bpf_map_fops, map, O_RDWR | O_CLOEXEC);
 112
 113         if (err < 0)
 114                 /* failed to allocate fd */
 115                 goto free_map;
 116
 117         return err;
 118
 119 free_map:
 120         map->ops->map_free(map);
 121         return err;
 122 }
 123
 124 /* if error is returned, fd is released.
 125  * On success caller should complete fd access with matching fdput()
 126  */
 127 struct bpf_map *bpf_map_get(struct fd f)
 128 {
 129         struct bpf_map *map;
 130
 131         if (!f.file)
 132                 return ERR_PTR(-EBADF);
 133
 134         if (f.file->f_op != &bpf_map_fops) {
 135                 fdput(f);
 136                 return ERR_PTR(-EINVAL);
 137         }
 138
 139         map = f.file->private_data;
 140
 141         return map;
 142 }
 143
 144 /* helper to convert user pointers passed inside __aligned_u64 fields */
 145 static void __user *u64_to_ptr(__u64 val)
 146 {
 147         return (void __user *) (unsigned long) val;
 148 }
 149
 150 /* last field in 'union bpf_attr' used by this command */
 151 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 152
 153 static int map_lookup_elem(union bpf_attr *attr)
 154 {
 155         void __user *ukey = u64_to_ptr(attr->key);
 156         void __user *uvalue = u64_to_ptr(attr->value);
 157         int ufd = attr->map_fd;
 158         struct fd f = fdget(ufd);
 159         struct bpf_map *map;
 160         void *key, *value, *ptr;
 161         int err;
 162
 163         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 164                 return -EINVAL;
 165
 166         map = bpf_map_get(f);
 167         if (IS_ERR(map))
 168                 return PTR_ERR(map);
 169
 170         err = -ENOMEM;
 171         key = kmalloc(map->key_size, GFP_USER);
 172         if (!key)
 173                 goto err_put;
 174
 175         err = -EFAULT;
 176         if (copy_from_user(key, ukey, map->key_size) != 0)
 177                 goto free_key;
 178
 179         err = -ENOMEM;
 180         value = kmalloc(map->value_size, GFP_USER);
 181         if (!value)
 182                 goto free_key;
 183
 184         rcu_read_lock();
 185         ptr = map->ops->map_lookup_elem(map, key);
 186         if (ptr)
 187                 memcpy(value, ptr, map->value_size);
 188         rcu_read_unlock();
 189
 190         err = -ENOENT;
 191         if (!ptr)
 192                 goto free_value;
 193
 194         err = -EFAULT;
 195         if (copy_to_user(uvalue, value, map->value_size) != 0)
 196                 goto free_value;
 197
 198         err = 0;
 199
 200 free_value:
 201         kfree(value);
 202 free_key:
 203         kfree(key);
 204 err_put:
 205         fdput(f);
 206         return err;
 207 }
 208
 209 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 210
 211 static int map_update_elem(union bpf_attr *attr)
 212 {
 213         void __user *ukey = u64_to_ptr(attr->key);
 214         void __user *uvalue = u64_to_ptr(attr->value);
 215         int ufd = attr->map_fd;
 216         struct fd f = fdget(ufd);
 217         struct bpf_map *map;
 218         void *key, *value;
 219         int err;
 220
 221         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 222                 return -EINVAL;
 223
 224         map = bpf_map_get(f);
 225         if (IS_ERR(map))
 226                 return PTR_ERR(map);
 227
 228         err = -ENOMEM;
 229         key = kmalloc(map->key_size, GFP_USER);
 230         if (!key)
 231                 goto err_put;
 232
 233         err = -EFAULT;
 234         if (copy_from_user(key, ukey, map->key_size) != 0)
 235                 goto free_key;
 236
 237         err = -ENOMEM;
 238         value = kmalloc(map->value_size, GFP_USER);
 239         if (!value)
 240                 goto free_key;
 241
 242         err = -EFAULT;
 243         if (copy_from_user(value, uvalue, map->value_size) != 0)
 244                 goto free_value;
 245
 246         /* eBPF program that use maps are running under rcu_read_lock(),
 247          * therefore all map accessors rely on this fact, so do the same here
 248          */
 249         rcu_read_lock();
 250         err = map->ops->map_update_elem(map, key, value, attr->flags);
 251         rcu_read_unlock();
 252
 253 free_value:
 254         kfree(value);
 255 free_key:
 256         kfree(key);
 257 err_put:
 258         fdput(f);
 259         return err;
 260 }
 261
 262 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 263
 264 static int map_delete_elem(union bpf_attr *attr)
 265 {
 266         void __user *ukey = u64_to_ptr(attr->key);
 267         int ufd = attr->map_fd;
 268         struct fd f = fdget(ufd);
 269         struct bpf_map *map;
 270         void *key;
 271         int err;
 272
 273         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 274                 return -EINVAL;
 275
 276         map = bpf_map_get(f);
 277         if (IS_ERR(map))
 278                 return PTR_ERR(map);
 279
 280         err = -ENOMEM;
 281         key = kmalloc(map->key_size, GFP_USER);
 282         if (!key)
 283                 goto err_put;
 284
 285         err = -EFAULT;
 286         if (copy_from_user(key, ukey, map->key_size) != 0)
 287                 goto free_key;
 288
 289         rcu_read_lock();
 290         err = map->ops->map_delete_elem(map, key);
 291         rcu_read_unlock();
 292
 293 free_key:
 294         kfree(key);
 295 err_put:
 296         fdput(f);
 297         return err;
 298 }
 299
 300 /* last field in 'union bpf_attr' used by this command */
 301 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 302
 303 static int map_get_next_key(union bpf_attr *attr)
 304 {
 305         void __user *ukey = u64_to_ptr(attr->key);
 306         void __user *unext_key = u64_to_ptr(attr->next_key);
 307         int ufd = attr->map_fd;
 308         struct fd f = fdget(ufd);
 309         struct bpf_map *map;
 310         void *key, *next_key;
 311         int err;
 312
 313         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 314                 return -EINVAL;
 315
 316         map = bpf_map_get(f);
 317         if (IS_ERR(map))
 318                 return PTR_ERR(map);
 319
 320         err = -ENOMEM;
 321         key = kmalloc(map->key_size, GFP_USER);
 322         if (!key)
 323                 goto err_put;
 324
 325         err = -EFAULT;
 326         if (copy_from_user(key, ukey, map->key_size) != 0)
 327                 goto free_key;
 328
 329         err = -ENOMEM;
 330         next_key = kmalloc(map->key_size, GFP_USER);
 331         if (!next_key)
 332                 goto free_key;
 333
 334         rcu_read_lock();
 335         err = map->ops->map_get_next_key(map, key, next_key);
 336         rcu_read_unlock();
 337         if (err)
 338                 goto free_next_key;
 339
 340         err = -EFAULT;
 341         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 342                 goto free_next_key;
 343
 344         err = 0;
 345
 346 free_next_key:
 347         kfree(next_key);
 348 free_key:
 349         kfree(key);
 350 err_put:
 351         fdput(f);
 352         return err;
 353 }
 354
 355 static LIST_HEAD(bpf_prog_types);
 356
 357 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 358 {
 359         struct bpf_prog_type_list *tl;
 360
 361         list_for_each_entry(tl, &bpf_prog_types, list_node) {
 362                 if (tl->type == type) {
 363                         prog->aux->ops = tl->ops;
 364                         prog->type = type;
 365                         return 0;
 366                 }
 367         }
 368
 369         return -EINVAL;
 370 }
 371
 372 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
 373 {
 374         list_add(&tl->list_node, &bpf_prog_types);
 375 }
 376
 377 /* fixup insn->imm field of bpf_call instructions:
 378  * if (insn->imm == BPF_FUNC_map_lookup_elem)
 379  *      insn->imm = bpf_map_lookup_elem - __bpf_call_base;
 380  * else if (insn->imm == BPF_FUNC_map_update_elem)
 381  *      insn->imm = bpf_map_update_elem - __bpf_call_base;
 382  * else ...
 383  *
 384  * this function is called after eBPF program passed verification
 385  */
 386 static void fixup_bpf_calls(struct bpf_prog *prog)
 387 {
 388         const struct bpf_func_proto *fn;
 389         int i;
 390
 391         for (i = 0; i < prog->len; i++) {
 392                 struct bpf_insn *insn = &prog->insnsi[i];
 393
 394                 if (insn->code == (BPF_JMP | BPF_CALL)) {
 395                         /* we reach here when program has bpf_call instructions
 396                          * and it passed bpf_check(), means that
 397                          * ops->get_func_proto must have been supplied, check it
 398                          */
 399                         BUG_ON(!prog->aux->ops->get_func_proto);
 400
 401                         if (insn->imm == BPF_FUNC_tail_call) {
 402                                 /* mark bpf_tail_call as different opcode
 403                                  * to avoid conditional branch in
 404                                  * interpeter for every normal call
 405                                  * and to prevent accidental JITing by
 406                                  * JIT compiler that doesn't support
 407                                  * bpf_tail_call yet
 408                                  */
 409                                 insn->imm = 0;
 410                                 insn->code |= BPF_X;
 411                                 continue;
 412                         }
 413
 414                         fn = prog->aux->ops->get_func_proto(insn->imm);
 415                         /* all functions that have prototype and verifier allowed
 416                          * programs to call them, must be real in-kernel functions
 417                          */
 418                         BUG_ON(!fn->func);
 419                         insn->imm = fn->func - __bpf_call_base;
 420                 }
 421         }
 422 }
 423
 424 /* drop refcnt on maps used by eBPF program and free auxilary data */
 425 static void free_used_maps(struct bpf_prog_aux *aux)
 426 {
 427         int i;
 428
 429         for (i = 0; i < aux->used_map_cnt; i++)
 430                 bpf_map_put(aux->used_maps[i]);
 431
 432         kfree(aux->used_maps);
 433 }
 434
 435 static void __prog_put_rcu(struct rcu_head *rcu)
 436 {
 437         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 438
 439         free_used_maps(aux);
 440         bpf_prog_free(aux->prog);
 441 }
 442
 443 /* version of bpf_prog_put() that is called after a grace period */
 444 void bpf_prog_put_rcu(struct bpf_prog *prog)
 445 {
 446         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 447                 prog->aux->prog = prog;
 448                 call_rcu(&prog->aux->rcu, __prog_put_rcu);
 449         }
 450 }
 451
 452 void bpf_prog_put(struct bpf_prog *prog)
 453 {
 454         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 455                 free_used_maps(prog->aux);
 456                 bpf_prog_free(prog);
 457         }
 458 }
 459 EXPORT_SYMBOL_GPL(bpf_prog_put);
 460
 461 static int bpf_prog_release(struct inode *inode, struct file *filp)
 462 {
 463         struct bpf_prog *prog = filp->private_data;
 464
 465         bpf_prog_put_rcu(prog);
 466         return 0;
 467 }
 468
 469 static const struct file_operations bpf_prog_fops = {
 470         .release = bpf_prog_release,
 471 };
 472
 473 static struct bpf_prog *get_prog(struct fd f)
 474 {
 475         struct bpf_prog *prog;
 476
 477         if (!f.file)
 478                 return ERR_PTR(-EBADF);
 479
 480         if (f.file->f_op != &bpf_prog_fops) {
 481                 fdput(f);
 482                 return ERR_PTR(-EINVAL);
 483         }
 484
 485         prog = f.file->private_data;
 486
 487         return prog;
 488 }
 489
 490 /* called by sockets/tracing/seccomp before attaching program to an event
 491  * pairs with bpf_prog_put()
 492  */
 493 struct bpf_prog *bpf_prog_get(u32 ufd)
 494 {
 495         struct fd f = fdget(ufd);
 496         struct bpf_prog *prog;
 497
 498         prog = get_prog(f);
 499
 500         if (IS_ERR(prog))
 501                 return prog;
 502
 503         atomic_inc(&prog->aux->refcnt);
 504         fdput(f);
 505         return prog;
 506 }
 507 EXPORT_SYMBOL_GPL(bpf_prog_get);
 508
 509 /* last field in 'union bpf_attr' used by this command */
 510 #define BPF_PROG_LOAD_LAST_FIELD kern_version
 511
 512 static int bpf_prog_load(union bpf_attr *attr)
 513 {
 514         enum bpf_prog_type type = attr->prog_type;
 515         struct bpf_prog *prog;
 516         int err;
 517         char license[128];
 518         bool is_gpl;
 519
 520         if (CHECK_ATTR(BPF_PROG_LOAD))
 521                 return -EINVAL;
 522
 523         /* copy eBPF program license from user space */
 524         if (strncpy_from_user(license, u64_to_ptr(attr->license),
 525                               sizeof(license) - 1) < 0)
 526                 return -EFAULT;
 527         license[sizeof(license) - 1] = 0;
 528
 529         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 530         is_gpl = license_is_gpl_compatible(license);
 531
 532         if (attr->insn_cnt >= BPF_MAXINSNS)
 533                 return -EINVAL;
 534
 535         if (type == BPF_PROG_TYPE_KPROBE &&
 536             attr->kern_version != LINUX_VERSION_CODE)
 537                 return -EINVAL;
 538
 539         /* plain bpf_prog allocation */
 540         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
 541         if (!prog)
 542                 return -ENOMEM;
 543
 544         prog->len = attr->insn_cnt;
 545
 546         err = -EFAULT;
 547         if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
 548                            prog->len * sizeof(struct bpf_insn)) != 0)
 549                 goto free_prog;
 550
 551         prog->orig_prog = NULL;
 552         prog->jited = false;
 553
 554         atomic_set(&prog->aux->refcnt, 1);
 555         prog->gpl_compatible = is_gpl;
 556
 557         /* find program type: socket_filter vs tracing_filter */
 558         err = find_prog_type(type, prog);
 559         if (err < 0)
 560                 goto free_prog;
 561
 562         /* run eBPF verifier */
 563         err = bpf_check(&prog, attr);
 564         if (err < 0)
 565                 goto free_used_maps;
 566
 567         /* fixup BPF_CALL->imm field */
 568         fixup_bpf_calls(prog);
 569
 570         /* eBPF program is ready to be JITed */
 571         err = bpf_prog_select_runtime(prog);
 572         if (err < 0)
 573                 goto free_used_maps;
 574
 575         err = anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, O_RDWR | O_CLOEXEC);
 576         if (err < 0)
 577                 /* failed to allocate fd */
 578                 goto free_used_maps;
 579
 580         return err;
 581
 582 free_used_maps:
 583         free_used_maps(prog->aux);
 584 free_prog:
 585         bpf_prog_free(prog);
 586         return err;
 587 }
 588
 589 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 590 {
 591         union bpf_attr attr = {};
 592         int err;
 593
 594         /* the syscall is limited to root temporarily. This restriction will be
 595          * lifted when security audit is clean. Note that eBPF+tracing must have
 596          * this restriction, since it may pass kernel data to user space
 597          */
 598         if (!capable(CAP_SYS_ADMIN))
 599                 return -EPERM;
 600
 601         if (!access_ok(VERIFY_READ, uattr, 1))
 602                 return -EFAULT;
 603
 604         if (size > PAGE_SIZE)   /* silly large */
 605                 return -E2BIG;
 606
 607         /* If we're handed a bigger struct than we know of,
 608          * ensure all the unknown bits are 0 - i.e. new
 609          * user-space does not rely on any kernel feature
 610          * extensions we dont know about yet.
 611          */
 612         if (size > sizeof(attr)) {
 613                 unsigned char __user *addr;
 614                 unsigned char __user *end;
 615                 unsigned char val;
 616
 617                 addr = (void __user *)uattr + sizeof(attr);
 618                 end  = (void __user *)uattr + size;
 619
 620                 for (; addr < end; addr++) {
 621                         err = get_user(val, addr);
 622                         if (err)
 623                                 return err;
 624                         if (val)
 625                                 return -E2BIG;
 626                 }
 627                 size = sizeof(attr);
 628         }
 629
 630         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
 631         if (copy_from_user(&attr, uattr, size) != 0)
 632                 return -EFAULT;
 633
 634         switch (cmd) {
 635         case BPF_MAP_CREATE:
 636                 err = map_create(&attr);
 637                 break;
 638         case BPF_MAP_LOOKUP_ELEM:
 639                 err = map_lookup_elem(&attr);
 640                 break;
 641         case BPF_MAP_UPDATE_ELEM:
 642                 err = map_update_elem(&attr);
 643                 break;
 644         case BPF_MAP_DELETE_ELEM:
 645                 err = map_delete_elem(&attr);
 646                 break;
 647         case BPF_MAP_GET_NEXT_KEY:
 648                 err = map_get_next_key(&attr);
 649                 break;
 650         case BPF_PROG_LOAD:
 651                 err = bpf_prog_load(&attr);
 652                 break;
 653         default:
 654                 err = -EINVAL;
 655                 break;
 656         }
 657
 658         return err;
 659 }