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_fd_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 bpf_map *map;
 159         void *key, *value, *ptr;
 160         struct fd f;
 161         int err;
 162
 163         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 164                 return -EINVAL;
 165
 166         f = fdget(ufd);
 167         map = bpf_map_get(f);
 168         if (IS_ERR(map))
 169                 return PTR_ERR(map);
 170
 171         err = -ENOMEM;
 172         key = kmalloc(map->key_size, GFP_USER);
 173         if (!key)
 174                 goto err_put;
 175
 176         err = -EFAULT;
 177         if (copy_from_user(key, ukey, map->key_size) != 0)
 178                 goto free_key;
 179
 180         err = -ENOMEM;
 181         value = kmalloc(map->value_size, GFP_USER);
 182         if (!value)
 183                 goto free_key;
 184
 185         rcu_read_lock();
 186         ptr = map->ops->map_lookup_elem(map, key);
 187         if (ptr)
 188                 memcpy(value, ptr, map->value_size);
 189         rcu_read_unlock();
 190
 191         err = -ENOENT;
 192         if (!ptr)
 193                 goto free_value;
 194
 195         err = -EFAULT;
 196         if (copy_to_user(uvalue, value, map->value_size) != 0)
 197                 goto free_value;
 198
 199         err = 0;
 200
 201 free_value:
 202         kfree(value);
 203 free_key:
 204         kfree(key);
 205 err_put:
 206         fdput(f);
 207         return err;
 208 }
 209
 210 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 211
 212 static int map_update_elem(union bpf_attr *attr)
 213 {
 214         void __user *ukey = u64_to_ptr(attr->key);
 215         void __user *uvalue = u64_to_ptr(attr->value);
 216         int ufd = attr->map_fd;
 217         struct bpf_map *map;
 218         void *key, *value;
 219         struct fd f;
 220         int err;
 221
 222         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 223                 return -EINVAL;
 224
 225         f = fdget(ufd);
 226         map = bpf_map_get(f);
 227         if (IS_ERR(map))
 228                 return PTR_ERR(map);
 229
 230         err = -ENOMEM;
 231         key = kmalloc(map->key_size, GFP_USER);
 232         if (!key)
 233                 goto err_put;
 234
 235         err = -EFAULT;
 236         if (copy_from_user(key, ukey, map->key_size) != 0)
 237                 goto free_key;
 238
 239         err = -ENOMEM;
 240         value = kmalloc(map->value_size, GFP_USER);
 241         if (!value)
 242                 goto free_key;
 243
 244         err = -EFAULT;
 245         if (copy_from_user(value, uvalue, map->value_size) != 0)
 246                 goto free_value;
 247
 248         /* eBPF program that use maps are running under rcu_read_lock(),
 249          * therefore all map accessors rely on this fact, so do the same here
 250          */
 251         rcu_read_lock();
 252         err = map->ops->map_update_elem(map, key, value, attr->flags);
 253         rcu_read_unlock();
 254
 255 free_value:
 256         kfree(value);
 257 free_key:
 258         kfree(key);
 259 err_put:
 260         fdput(f);
 261         return err;
 262 }
 263
 264 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 265
 266 static int map_delete_elem(union bpf_attr *attr)
 267 {
 268         void __user *ukey = u64_to_ptr(attr->key);
 269         int ufd = attr->map_fd;
 270         struct bpf_map *map;
 271         struct fd f;
 272         void *key;
 273         int err;
 274
 275         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 276                 return -EINVAL;
 277
 278         f = fdget(ufd);
 279         map = bpf_map_get(f);
 280         if (IS_ERR(map))
 281                 return PTR_ERR(map);
 282
 283         err = -ENOMEM;
 284         key = kmalloc(map->key_size, GFP_USER);
 285         if (!key)
 286                 goto err_put;
 287
 288         err = -EFAULT;
 289         if (copy_from_user(key, ukey, map->key_size) != 0)
 290                 goto free_key;
 291
 292         rcu_read_lock();
 293         err = map->ops->map_delete_elem(map, key);
 294         rcu_read_unlock();
 295
 296 free_key:
 297         kfree(key);
 298 err_put:
 299         fdput(f);
 300         return err;
 301 }
 302
 303 /* last field in 'union bpf_attr' used by this command */
 304 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 305
 306 static int map_get_next_key(union bpf_attr *attr)
 307 {
 308         void __user *ukey = u64_to_ptr(attr->key);
 309         void __user *unext_key = u64_to_ptr(attr->next_key);
 310         int ufd = attr->map_fd;
 311         struct bpf_map *map;
 312         void *key, *next_key;
 313         struct fd f;
 314         int err;
 315
 316         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 317                 return -EINVAL;
 318
 319         f = fdget(ufd);
 320         map = bpf_map_get(f);
 321         if (IS_ERR(map))
 322                 return PTR_ERR(map);
 323
 324         err = -ENOMEM;
 325         key = kmalloc(map->key_size, GFP_USER);
 326         if (!key)
 327                 goto err_put;
 328
 329         err = -EFAULT;
 330         if (copy_from_user(key, ukey, map->key_size) != 0)
 331                 goto free_key;
 332
 333         err = -ENOMEM;
 334         next_key = kmalloc(map->key_size, GFP_USER);
 335         if (!next_key)
 336                 goto free_key;
 337
 338         rcu_read_lock();
 339         err = map->ops->map_get_next_key(map, key, next_key);
 340         rcu_read_unlock();
 341         if (err)
 342                 goto free_next_key;
 343
 344         err = -EFAULT;
 345         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 346                 goto free_next_key;
 347
 348         err = 0;
 349
 350 free_next_key:
 351         kfree(next_key);
 352 free_key:
 353         kfree(key);
 354 err_put:
 355         fdput(f);
 356         return err;
 357 }
 358
 359 static LIST_HEAD(bpf_prog_types);
 360
 361 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 362 {
 363         struct bpf_prog_type_list *tl;
 364
 365         list_for_each_entry(tl, &bpf_prog_types, list_node) {
 366                 if (tl->type == type) {
 367                         prog->aux->ops = tl->ops;
 368                         prog->type = type;
 369                         return 0;
 370                 }
 371         }
 372
 373         return -EINVAL;
 374 }
 375
 376 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
 377 {
 378         list_add(&tl->list_node, &bpf_prog_types);
 379 }
 380
 381 /* fixup insn->imm field of bpf_call instructions:
 382  * if (insn->imm == BPF_FUNC_map_lookup_elem)
 383  *      insn->imm = bpf_map_lookup_elem - __bpf_call_base;
 384  * else if (insn->imm == BPF_FUNC_map_update_elem)
 385  *      insn->imm = bpf_map_update_elem - __bpf_call_base;
 386  * else ...
 387  *
 388  * this function is called after eBPF program passed verification
 389  */
 390 static void fixup_bpf_calls(struct bpf_prog *prog)
 391 {
 392         const struct bpf_func_proto *fn;
 393         int i;
 394
 395         for (i = 0; i < prog->len; i++) {
 396                 struct bpf_insn *insn = &prog->insnsi[i];
 397
 398                 if (insn->code == (BPF_JMP | BPF_CALL)) {
 399                         /* we reach here when program has bpf_call instructions
 400                          * and it passed bpf_check(), means that
 401                          * ops->get_func_proto must have been supplied, check it
 402                          */
 403                         BUG_ON(!prog->aux->ops->get_func_proto);
 404
 405                         if (insn->imm == BPF_FUNC_tail_call) {
 406                                 /* mark bpf_tail_call as different opcode
 407                                  * to avoid conditional branch in
 408                                  * interpeter for every normal call
 409                                  * and to prevent accidental JITing by
 410                                  * JIT compiler that doesn't support
 411                                  * bpf_tail_call yet
 412                                  */
 413                                 insn->imm = 0;
 414                                 insn->code |= BPF_X;
 415                                 continue;
 416                         }
 417
 418                         fn = prog->aux->ops->get_func_proto(insn->imm);
 419                         /* all functions that have prototype and verifier allowed
 420                          * programs to call them, must be real in-kernel functions
 421                          */
 422                         BUG_ON(!fn->func);
 423                         insn->imm = fn->func - __bpf_call_base;
 424                 }
 425         }
 426 }
 427
 428 /* drop refcnt on maps used by eBPF program and free auxilary data */
 429 static void free_used_maps(struct bpf_prog_aux *aux)
 430 {
 431         int i;
 432
 433         for (i = 0; i < aux->used_map_cnt; i++)
 434                 bpf_map_put(aux->used_maps[i]);
 435
 436         kfree(aux->used_maps);
 437 }
 438
 439 static void __prog_put_rcu(struct rcu_head *rcu)
 440 {
 441         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 442
 443         free_used_maps(aux);
 444         bpf_prog_free(aux->prog);
 445 }
 446
 447 /* version of bpf_prog_put() that is called after a grace period */
 448 void bpf_prog_put_rcu(struct bpf_prog *prog)
 449 {
 450         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 451                 prog->aux->prog = prog;
 452                 call_rcu(&prog->aux->rcu, __prog_put_rcu);
 453         }
 454 }
 455
 456 void bpf_prog_put(struct bpf_prog *prog)
 457 {
 458         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 459                 free_used_maps(prog->aux);
 460                 bpf_prog_free(prog);
 461         }
 462 }
 463 EXPORT_SYMBOL_GPL(bpf_prog_put);
 464
 465 static int bpf_prog_release(struct inode *inode, struct file *filp)
 466 {
 467         struct bpf_prog *prog = filp->private_data;
 468
 469         bpf_prog_put_rcu(prog);
 470         return 0;
 471 }
 472
 473 static const struct file_operations bpf_prog_fops = {
 474         .release = bpf_prog_release,
 475 };
 476
 477 static struct bpf_prog *get_prog(struct fd f)
 478 {
 479         struct bpf_prog *prog;
 480
 481         if (!f.file)
 482                 return ERR_PTR(-EBADF);
 483
 484         if (f.file->f_op != &bpf_prog_fops) {
 485                 fdput(f);
 486                 return ERR_PTR(-EINVAL);
 487         }
 488
 489         prog = f.file->private_data;
 490
 491         return prog;
 492 }
 493
 494 /* called by sockets/tracing/seccomp before attaching program to an event
 495  * pairs with bpf_prog_put()
 496  */
 497 struct bpf_prog *bpf_prog_get(u32 ufd)
 498 {
 499         struct fd f = fdget(ufd);
 500         struct bpf_prog *prog;
 501
 502         prog = get_prog(f);
 503
 504         if (IS_ERR(prog))
 505                 return prog;
 506
 507         atomic_inc(&prog->aux->refcnt);
 508         fdput(f);
 509         return prog;
 510 }
 511 EXPORT_SYMBOL_GPL(bpf_prog_get);
 512
 513 /* last field in 'union bpf_attr' used by this command */
 514 #define BPF_PROG_LOAD_LAST_FIELD kern_version
 515
 516 static int bpf_prog_load(union bpf_attr *attr)
 517 {
 518         enum bpf_prog_type type = attr->prog_type;
 519         struct bpf_prog *prog;
 520         int err;
 521         char license[128];
 522         bool is_gpl;
 523
 524         if (CHECK_ATTR(BPF_PROG_LOAD))
 525                 return -EINVAL;
 526
 527         /* copy eBPF program license from user space */
 528         if (strncpy_from_user(license, u64_to_ptr(attr->license),
 529                               sizeof(license) - 1) < 0)
 530                 return -EFAULT;
 531         license[sizeof(license) - 1] = 0;
 532
 533         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 534         is_gpl = license_is_gpl_compatible(license);
 535
 536         if (attr->insn_cnt >= BPF_MAXINSNS)
 537                 return -EINVAL;
 538
 539         if (type == BPF_PROG_TYPE_KPROBE &&
 540             attr->kern_version != LINUX_VERSION_CODE)
 541                 return -EINVAL;
 542
 543         /* plain bpf_prog allocation */
 544         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
 545         if (!prog)
 546                 return -ENOMEM;
 547
 548         prog->len = attr->insn_cnt;
 549
 550         err = -EFAULT;
 551         if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
 552                            prog->len * sizeof(struct bpf_insn)) != 0)
 553                 goto free_prog;
 554
 555         prog->orig_prog = NULL;
 556         prog->jited = false;
 557
 558         atomic_set(&prog->aux->refcnt, 1);
 559         prog->gpl_compatible = is_gpl;
 560
 561         /* find program type: socket_filter vs tracing_filter */
 562         err = find_prog_type(type, prog);
 563         if (err < 0)
 564                 goto free_prog;
 565
 566         /* run eBPF verifier */
 567         err = bpf_check(&prog, attr);
 568         if (err < 0)
 569                 goto free_used_maps;
 570
 571         /* fixup BPF_CALL->imm field */
 572         fixup_bpf_calls(prog);
 573
 574         /* eBPF program is ready to be JITed */
 575         err = bpf_prog_select_runtime(prog);
 576         if (err < 0)
 577                 goto free_used_maps;
 578
 579         err = anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog, O_RDWR | O_CLOEXEC);
 580         if (err < 0)
 581                 /* failed to allocate fd */
 582                 goto free_used_maps;
 583
 584         return err;
 585
 586 free_used_maps:
 587         free_used_maps(prog->aux);
 588 free_prog:
 589         bpf_prog_free(prog);
 590         return err;
 591 }
 592
 593 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 594 {
 595         union bpf_attr attr = {};
 596         int err;
 597
 598         /* the syscall is limited to root temporarily. This restriction will be
 599          * lifted when security audit is clean. Note that eBPF+tracing must have
 600          * this restriction, since it may pass kernel data to user space
 601          */
 602         if (!capable(CAP_SYS_ADMIN))
 603                 return -EPERM;
 604
 605         if (!access_ok(VERIFY_READ, uattr, 1))
 606                 return -EFAULT;
 607
 608         if (size > PAGE_SIZE)   /* silly large */
 609                 return -E2BIG;
 610
 611         /* If we're handed a bigger struct than we know of,
 612          * ensure all the unknown bits are 0 - i.e. new
 613          * user-space does not rely on any kernel feature
 614          * extensions we dont know about yet.
 615          */
 616         if (size > sizeof(attr)) {
 617                 unsigned char __user *addr;
 618                 unsigned char __user *end;
 619                 unsigned char val;
 620
 621                 addr = (void __user *)uattr + sizeof(attr);
 622                 end  = (void __user *)uattr + size;
 623
 624                 for (; addr < end; addr++) {
 625                         err = get_user(val, addr);
 626                         if (err)
 627                                 return err;
 628                         if (val)
 629                                 return -E2BIG;
 630                 }
 631                 size = sizeof(attr);
 632         }
 633
 634         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
 635         if (copy_from_user(&attr, uattr, size) != 0)
 636                 return -EFAULT;
 637
 638         switch (cmd) {
 639         case BPF_MAP_CREATE:
 640                 err = map_create(&attr);
 641                 break;
 642         case BPF_MAP_LOOKUP_ELEM:
 643                 err = map_lookup_elem(&attr);
 644                 break;
 645         case BPF_MAP_UPDATE_ELEM:
 646                 err = map_update_elem(&attr);
 647                 break;
 648         case BPF_MAP_DELETE_ELEM:
 649                 err = map_delete_elem(&attr);
 650                 break;
 651         case BPF_MAP_GET_NEXT_KEY:
 652                 err = map_get_next_key(&attr);
 653                 break;
 654         case BPF_PROG_LOAD:
 655                 err = bpf_prog_load(&attr);
 656                 break;
 657         default:
 658                 err = -EINVAL;
 659                 break;
 660         }
 661
 662         return err;
 663 }