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