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/vmalloc.h>
  16 #include <linux/mmzone.h>
  17 #include <linux/anon_inodes.h>
  18 #include <linux/file.h>
  19 #include <linux/license.h>
  20 #include <linux/filter.h>
  21 #include <linux/version.h>
  22 #include <linux/kernel.h>
  23
  24 DEFINE_PER_CPU(int, bpf_prog_active);
  25
  26 int sysctl_unprivileged_bpf_disabled __read_mostly;
  27
  28 static LIST_HEAD(bpf_map_types);
  29
  30 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
  31 {
  32         struct bpf_map_type_list *tl;
  33         struct bpf_map *map;
  34
  35         list_for_each_entry(tl, &bpf_map_types, list_node) {
  36                 if (tl->type == attr->map_type) {
  37                         map = tl->ops->map_alloc(attr);
  38                         if (IS_ERR(map))
  39                                 return map;
  40                         map->ops = tl->ops;
  41                         map->map_type = attr->map_type;
  42                         return map;
  43                 }
  44         }
  45         return ERR_PTR(-EINVAL);
  46 }
  47
  48 /* boot time registration of different map implementations */
  49 void bpf_register_map_type(struct bpf_map_type_list *tl)
  50 {
  51         list_add(&tl->list_node, &bpf_map_types);
  52 }
  53
  54 void *bpf_map_area_alloc(size_t size)
  55 {
  56         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
  57          * trigger under memory pressure as we really just want to
  58          * fail instead.
  59          */
  60         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
  61         void *area;
  62
  63         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
  64                 area = kmalloc(size, GFP_USER | flags);
  65                 if (area != NULL)
  66                         return area;
  67         }
  68
  69         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | flags,
  70                          PAGE_KERNEL);
  71 }
  72
  73 void bpf_map_area_free(void *area)
  74 {
  75         kvfree(area);
  76 }
  77
  78 int bpf_map_precharge_memlock(u32 pages)
  79 {
  80         struct user_struct *user = get_current_user();
  81         unsigned long memlock_limit, cur;
  82
  83         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  84         cur = atomic_long_read(&user->locked_vm);
  85         free_uid(user);
  86         if (cur + pages > memlock_limit)
  87                 return -EPERM;
  88         return 0;
  89 }
  90
  91 static int bpf_map_charge_memlock(struct bpf_map *map)
  92 {
  93         struct user_struct *user = get_current_user();
  94         unsigned long memlock_limit;
  95
  96         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  97
  98         atomic_long_add(map->pages, &user->locked_vm);
  99
 100         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 101                 atomic_long_sub(map->pages, &user->locked_vm);
 102                 free_uid(user);
 103                 return -EPERM;
 104         }
 105         map->user = user;
 106         return 0;
 107 }
 108
 109 static void bpf_map_uncharge_memlock(struct bpf_map *map)
 110 {
 111         struct user_struct *user = map->user;
 112
 113         atomic_long_sub(map->pages, &user->locked_vm);
 114         free_uid(user);
 115 }
 116
 117 /* called from workqueue */
 118 static void bpf_map_free_deferred(struct work_struct *work)
 119 {
 120         struct bpf_map *map = container_of(work, struct bpf_map, work);
 121
 122         bpf_map_uncharge_memlock(map);
 123         /* implementation dependent freeing */
 124         map->ops->map_free(map);
 125 }
 126
 127 static void bpf_map_put_uref(struct bpf_map *map)
 128 {
 129         if (atomic_dec_and_test(&map->usercnt)) {
 130                 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
 131                         bpf_fd_array_map_clear(map);
 132         }
 133 }
 134
 135 /* decrement map refcnt and schedule it for freeing via workqueue
 136  * (unrelying map implementation ops->map_free() might sleep)
 137  */
 138 void bpf_map_put(struct bpf_map *map)
 139 {
 140         if (atomic_dec_and_test(&map->refcnt)) {
 141                 INIT_WORK(&map->work, bpf_map_free_deferred);
 142                 schedule_work(&map->work);
 143         }
 144 }
 145
 146 void bpf_map_put_with_uref(struct bpf_map *map)
 147 {
 148         bpf_map_put_uref(map);
 149         bpf_map_put(map);
 150 }
 151
 152 static int bpf_map_release(struct inode *inode, struct file *filp)
 153 {
 154         struct bpf_map *map = filp->private_data;
 155
 156         if (map->ops->map_release)
 157                 map->ops->map_release(map, filp);
 158
 159         bpf_map_put_with_uref(map);
 160         return 0;
 161 }
 162
 163 #ifdef CONFIG_PROC_FS
 164 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 165 {
 166         const struct bpf_map *map = filp->private_data;
 167         const struct bpf_array *array;
 168         u32 owner_prog_type = 0;
 169
 170         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 171                 array = container_of(map, struct bpf_array, map);
 172                 owner_prog_type = array->owner_prog_type;
 173         }
 174
 175         seq_printf(m,
 176                    "map_type:\t%u\n"
 177                    "key_size:\t%u\n"
 178                    "value_size:\t%u\n"
 179                    "max_entries:\t%u\n"
 180                    "map_flags:\t%#x\n"
 181                    "memlock:\t%llu\n",
 182                    map->map_type,
 183                    map->key_size,
 184                    map->value_size,
 185                    map->max_entries,
 186                    map->map_flags,
 187                    map->pages * 1ULL << PAGE_SHIFT);
 188
 189         if (owner_prog_type)
 190                 seq_printf(m, "owner_prog_type:\t%u\n",
 191                            owner_prog_type);
 192 }
 193 #endif
 194
 195 static const struct file_operations bpf_map_fops = {
 196 #ifdef CONFIG_PROC_FS
 197         .show_fdinfo    = bpf_map_show_fdinfo,
 198 #endif
 199         .release        = bpf_map_release,
 200 };
 201
 202 int bpf_map_new_fd(struct bpf_map *map)
 203 {
 204         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 205                                 O_RDWR | O_CLOEXEC);
 206 }
 207
 208 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 209 #define CHECK_ATTR(CMD) \
 210         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 211                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 212                    sizeof(*attr) - \
 213                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 214                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 215
 216 #define BPF_MAP_CREATE_LAST_FIELD map_flags
 217 /* called via syscall */
 218 static int map_create(union bpf_attr *attr)
 219 {
 220         struct bpf_map *map;
 221         int err;
 222
 223         err = CHECK_ATTR(BPF_MAP_CREATE);
 224         if (err)
 225                 return -EINVAL;
 226
 227         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 228         map = find_and_alloc_map(attr);
 229         if (IS_ERR(map))
 230                 return PTR_ERR(map);
 231
 232         atomic_set(&map->refcnt, 1);
 233         atomic_set(&map->usercnt, 1);
 234
 235         err = bpf_map_charge_memlock(map);
 236         if (err)
 237                 goto free_map_nouncharge;
 238
 239         err = bpf_map_new_fd(map);
 240         if (err < 0)
 241                 /* failed to allocate fd */
 242                 goto free_map;
 243
 244         return err;
 245
 246 free_map:
 247         bpf_map_uncharge_memlock(map);
 248 free_map_nouncharge:
 249         map->ops->map_free(map);
 250         return err;
 251 }
 252
 253 /* if error is returned, fd is released.
 254  * On success caller should complete fd access with matching fdput()
 255  */
 256 struct bpf_map *__bpf_map_get(struct fd f)
 257 {
 258         if (!f.file)
 259                 return ERR_PTR(-EBADF);
 260         if (f.file->f_op != &bpf_map_fops) {
 261                 fdput(f);
 262                 return ERR_PTR(-EINVAL);
 263         }
 264
 265         return f.file->private_data;
 266 }
 267
 268 /* prog's and map's refcnt limit */
 269 #define BPF_MAX_REFCNT 32768
 270
 271 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 272 {
 273         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 274                 atomic_dec(&map->refcnt);
 275                 return ERR_PTR(-EBUSY);
 276         }
 277         if (uref)
 278                 atomic_inc(&map->usercnt);
 279         return map;
 280 }
 281
 282 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 283 {
 284         struct fd f = fdget(ufd);
 285         struct bpf_map *map;
 286
 287         map = __bpf_map_get(f);
 288         if (IS_ERR(map))
 289                 return map;
 290
 291         map = bpf_map_inc(map, true);
 292         fdput(f);
 293
 294         return map;
 295 }
 296
 297 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 298 {
 299         return -ENOTSUPP;
 300 }
 301
 302 /* last field in 'union bpf_attr' used by this command */
 303 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 304
 305 static int map_lookup_elem(union bpf_attr *attr)
 306 {
 307         void __user *ukey = u64_to_user_ptr(attr->key);
 308         void __user *uvalue = u64_to_user_ptr(attr->value);
 309         int ufd = attr->map_fd;
 310         struct bpf_map *map;
 311         void *key, *value, *ptr;
 312         u32 value_size;
 313         struct fd f;
 314         int err;
 315
 316         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 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         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 334             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 335             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 336                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 337         else
 338                 value_size = map->value_size;
 339
 340         err = -ENOMEM;
 341         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 342         if (!value)
 343                 goto free_key;
 344
 345         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 346             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 347                 err = bpf_percpu_hash_copy(map, key, value);
 348         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 349                 err = bpf_percpu_array_copy(map, key, value);
 350         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 351                 err = bpf_stackmap_copy(map, key, value);
 352         } else {
 353                 rcu_read_lock();
 354                 ptr = map->ops->map_lookup_elem(map, key);
 355                 if (ptr)
 356                         memcpy(value, ptr, value_size);
 357                 rcu_read_unlock();
 358                 err = ptr ? 0 : -ENOENT;
 359         }
 360
 361         if (err)
 362                 goto free_value;
 363
 364         err = -EFAULT;
 365         if (copy_to_user(uvalue, value, value_size) != 0)
 366                 goto free_value;
 367
 368         err = 0;
 369
 370 free_value:
 371         kfree(value);
 372 free_key:
 373         kfree(key);
 374 err_put:
 375         fdput(f);
 376         return err;
 377 }
 378
 379 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 380
 381 static int map_update_elem(union bpf_attr *attr)
 382 {
 383         void __user *ukey = u64_to_user_ptr(attr->key);
 384         void __user *uvalue = u64_to_user_ptr(attr->value);
 385         int ufd = attr->map_fd;
 386         struct bpf_map *map;
 387         void *key, *value;
 388         u32 value_size;
 389         struct fd f;
 390         int err;
 391
 392         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 393                 return -EINVAL;
 394
 395         f = fdget(ufd);
 396         map = __bpf_map_get(f);
 397         if (IS_ERR(map))
 398                 return PTR_ERR(map);
 399
 400         err = -ENOMEM;
 401         key = kmalloc(map->key_size, GFP_USER);
 402         if (!key)
 403                 goto err_put;
 404
 405         err = -EFAULT;
 406         if (copy_from_user(key, ukey, map->key_size) != 0)
 407                 goto free_key;
 408
 409         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 410             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 411             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 412                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 413         else
 414                 value_size = map->value_size;
 415
 416         err = -ENOMEM;
 417         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 418         if (!value)
 419                 goto free_key;
 420
 421         err = -EFAULT;
 422         if (copy_from_user(value, uvalue, value_size) != 0)
 423                 goto free_value;
 424
 425         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 426          * inside bpf map update or delete otherwise deadlocks are possible
 427          */
 428         preempt_disable();
 429         __this_cpu_inc(bpf_prog_active);
 430         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 431             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 432                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 433         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 434                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 435         } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
 436                    map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
 437                    map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) {
 438                 rcu_read_lock();
 439                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 440                                                    attr->flags);
 441                 rcu_read_unlock();
 442         } else {
 443                 rcu_read_lock();
 444                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 445                 rcu_read_unlock();
 446         }
 447         __this_cpu_dec(bpf_prog_active);
 448         preempt_enable();
 449
 450 free_value:
 451         kfree(value);
 452 free_key:
 453         kfree(key);
 454 err_put:
 455         fdput(f);
 456         return err;
 457 }
 458
 459 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 460
 461 static int map_delete_elem(union bpf_attr *attr)
 462 {
 463         void __user *ukey = u64_to_user_ptr(attr->key);
 464         int ufd = attr->map_fd;
 465         struct bpf_map *map;
 466         struct fd f;
 467         void *key;
 468         int err;
 469
 470         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 471                 return -EINVAL;
 472
 473         f = fdget(ufd);
 474         map = __bpf_map_get(f);
 475         if (IS_ERR(map))
 476                 return PTR_ERR(map);
 477
 478         err = -ENOMEM;
 479         key = kmalloc(map->key_size, GFP_USER);
 480         if (!key)
 481                 goto err_put;
 482
 483         err = -EFAULT;
 484         if (copy_from_user(key, ukey, map->key_size) != 0)
 485                 goto free_key;
 486
 487         preempt_disable();
 488         __this_cpu_inc(bpf_prog_active);
 489         rcu_read_lock();
 490         err = map->ops->map_delete_elem(map, key);
 491         rcu_read_unlock();
 492         __this_cpu_dec(bpf_prog_active);
 493         preempt_enable();
 494
 495 free_key:
 496         kfree(key);
 497 err_put:
 498         fdput(f);
 499         return err;
 500 }
 501
 502 /* last field in 'union bpf_attr' used by this command */
 503 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 504
 505 static int map_get_next_key(union bpf_attr *attr)
 506 {
 507         void __user *ukey = u64_to_user_ptr(attr->key);
 508         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 509         int ufd = attr->map_fd;
 510         struct bpf_map *map;
 511         void *key, *next_key;
 512         struct fd f;
 513         int err;
 514
 515         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 516                 return -EINVAL;
 517
 518         f = fdget(ufd);
 519         map = __bpf_map_get(f);
 520         if (IS_ERR(map))
 521                 return PTR_ERR(map);
 522
 523         err = -ENOMEM;
 524         key = kmalloc(map->key_size, GFP_USER);
 525         if (!key)
 526                 goto err_put;
 527
 528         err = -EFAULT;
 529         if (copy_from_user(key, ukey, map->key_size) != 0)
 530                 goto free_key;
 531
 532         err = -ENOMEM;
 533         next_key = kmalloc(map->key_size, GFP_USER);
 534         if (!next_key)
 535                 goto free_key;
 536
 537         rcu_read_lock();
 538         err = map->ops->map_get_next_key(map, key, next_key);
 539         rcu_read_unlock();
 540         if (err)
 541                 goto free_next_key;
 542
 543         err = -EFAULT;
 544         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 545                 goto free_next_key;
 546
 547         err = 0;
 548
 549 free_next_key:
 550         kfree(next_key);
 551 free_key:
 552         kfree(key);
 553 err_put:
 554         fdput(f);
 555         return err;
 556 }
 557
 558 static LIST_HEAD(bpf_prog_types);
 559
 560 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 561 {
 562         struct bpf_prog_type_list *tl;
 563
 564         list_for_each_entry(tl, &bpf_prog_types, list_node) {
 565                 if (tl->type == type) {
 566                         prog->aux->ops = tl->ops;
 567                         prog->type = type;
 568                         return 0;
 569                 }
 570         }
 571
 572         return -EINVAL;
 573 }
 574
 575 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
 576 {
 577         list_add(&tl->list_node, &bpf_prog_types);
 578 }
 579
 580 /* fixup insn->imm field of bpf_call instructions:
 581  * if (insn->imm == BPF_FUNC_map_lookup_elem)
 582  *      insn->imm = bpf_map_lookup_elem - __bpf_call_base;
 583  * else if (insn->imm == BPF_FUNC_map_update_elem)
 584  *      insn->imm = bpf_map_update_elem - __bpf_call_base;
 585  * else ...
 586  *
 587  * this function is called after eBPF program passed verification
 588  */
 589 static void fixup_bpf_calls(struct bpf_prog *prog)
 590 {
 591         const struct bpf_func_proto *fn;
 592         int i;
 593
 594         for (i = 0; i < prog->len; i++) {
 595                 struct bpf_insn *insn = &prog->insnsi[i];
 596
 597                 if (insn->code == (BPF_JMP | BPF_CALL)) {
 598                         /* we reach here when program has bpf_call instructions
 599                          * and it passed bpf_check(), means that
 600                          * ops->get_func_proto must have been supplied, check it
 601                          */
 602                         BUG_ON(!prog->aux->ops->get_func_proto);
 603
 604                         if (insn->imm == BPF_FUNC_get_route_realm)
 605                                 prog->dst_needed = 1;
 606                         if (insn->imm == BPF_FUNC_get_prandom_u32)
 607                                 bpf_user_rnd_init_once();
 608                         if (insn->imm == BPF_FUNC_xdp_adjust_head)
 609                                 prog->xdp_adjust_head = 1;
 610                         if (insn->imm == BPF_FUNC_tail_call) {
 611                                 /* mark bpf_tail_call as different opcode
 612                                  * to avoid conditional branch in
 613                                  * interpeter for every normal call
 614                                  * and to prevent accidental JITing by
 615                                  * JIT compiler that doesn't support
 616                                  * bpf_tail_call yet
 617                                  */
 618                                 insn->imm = 0;
 619                                 insn->code |= BPF_X;
 620                                 continue;
 621                         }
 622
 623                         fn = prog->aux->ops->get_func_proto(insn->imm);
 624                         /* all functions that have prototype and verifier allowed
 625                          * programs to call them, must be real in-kernel functions
 626                          */
 627                         BUG_ON(!fn->func);
 628                         insn->imm = fn->func - __bpf_call_base;
 629                 }
 630         }
 631 }
 632
 633 /* drop refcnt on maps used by eBPF program and free auxilary data */
 634 static void free_used_maps(struct bpf_prog_aux *aux)
 635 {
 636         int i;
 637
 638         for (i = 0; i < aux->used_map_cnt; i++)
 639                 bpf_map_put(aux->used_maps[i]);
 640
 641         kfree(aux->used_maps);
 642 }
 643
 644 int __bpf_prog_charge(struct user_struct *user, u32 pages)
 645 {
 646         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 647         unsigned long user_bufs;
 648
 649         if (user) {
 650                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
 651                 if (user_bufs > memlock_limit) {
 652                         atomic_long_sub(pages, &user->locked_vm);
 653                         return -EPERM;
 654                 }
 655         }
 656
 657         return 0;
 658 }
 659
 660 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
 661 {
 662         if (user)
 663                 atomic_long_sub(pages, &user->locked_vm);
 664 }
 665
 666 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
 667 {
 668         struct user_struct *user = get_current_user();
 669         int ret;
 670
 671         ret = __bpf_prog_charge(user, prog->pages);
 672         if (ret) {
 673                 free_uid(user);
 674                 return ret;
 675         }
 676
 677         prog->aux->user = user;
 678         return 0;
 679 }
 680
 681 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
 682 {
 683         struct user_struct *user = prog->aux->user;
 684
 685         __bpf_prog_uncharge(user, prog->pages);
 686         free_uid(user);
 687 }
 688
 689 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
 690 {
 691         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 692
 693         free_used_maps(aux);
 694         bpf_prog_uncharge_memlock(aux->prog);
 695         bpf_prog_free(aux->prog);
 696 }
 697
 698 void bpf_prog_put(struct bpf_prog *prog)
 699 {
 700         if (atomic_dec_and_test(&prog->aux->refcnt))
 701                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
 702 }
 703 EXPORT_SYMBOL_GPL(bpf_prog_put);
 704
 705 static int bpf_prog_release(struct inode *inode, struct file *filp)
 706 {
 707         struct bpf_prog *prog = filp->private_data;
 708
 709         bpf_prog_put(prog);
 710         return 0;
 711 }
 712
 713 #ifdef CONFIG_PROC_FS
 714 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
 715 {
 716         const struct bpf_prog *prog = filp->private_data;
 717         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
 718
 719         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
 720         seq_printf(m,
 721                    "prog_type:\t%u\n"
 722                    "prog_jited:\t%u\n"
 723                    "prog_tag:\t%s\n"
 724                    "memlock:\t%llu\n",
 725                    prog->type,
 726                    prog->jited,
 727                    prog_tag,
 728                    prog->pages * 1ULL << PAGE_SHIFT);
 729 }
 730 #endif
 731
 732 static const struct file_operations bpf_prog_fops = {
 733 #ifdef CONFIG_PROC_FS
 734         .show_fdinfo    = bpf_prog_show_fdinfo,
 735 #endif
 736         .release        = bpf_prog_release,
 737 };
 738
 739 int bpf_prog_new_fd(struct bpf_prog *prog)
 740 {
 741         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
 742                                 O_RDWR | O_CLOEXEC);
 743 }
 744
 745 static struct bpf_prog *____bpf_prog_get(struct fd f)
 746 {
 747         if (!f.file)
 748                 return ERR_PTR(-EBADF);
 749         if (f.file->f_op != &bpf_prog_fops) {
 750                 fdput(f);
 751                 return ERR_PTR(-EINVAL);
 752         }
 753
 754         return f.file->private_data;
 755 }
 756
 757 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
 758 {
 759         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
 760                 atomic_sub(i, &prog->aux->refcnt);
 761                 return ERR_PTR(-EBUSY);
 762         }
 763         return prog;
 764 }
 765 EXPORT_SYMBOL_GPL(bpf_prog_add);
 766
 767 void bpf_prog_sub(struct bpf_prog *prog, int i)
 768 {
 769         /* Only to be used for undoing previous bpf_prog_add() in some
 770          * error path. We still know that another entity in our call
 771          * path holds a reference to the program, thus atomic_sub() can
 772          * be safely used in such cases!
 773          */
 774         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
 775 }
 776 EXPORT_SYMBOL_GPL(bpf_prog_sub);
 777
 778 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
 779 {
 780         return bpf_prog_add(prog, 1);
 781 }
 782 EXPORT_SYMBOL_GPL(bpf_prog_inc);
 783
 784 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
 785 {
 786         struct fd f = fdget(ufd);
 787         struct bpf_prog *prog;
 788
 789         prog = ____bpf_prog_get(f);
 790         if (IS_ERR(prog))
 791                 return prog;
 792         if (type && prog->type != *type) {
 793                 prog = ERR_PTR(-EINVAL);
 794                 goto out;
 795         }
 796
 797         prog = bpf_prog_inc(prog);
 798 out:
 799         fdput(f);
 800         return prog;
 801 }
 802
 803 struct bpf_prog *bpf_prog_get(u32 ufd)
 804 {
 805         return __bpf_prog_get(ufd, NULL);
 806 }
 807
 808 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
 809 {
 810         return __bpf_prog_get(ufd, &type);
 811 }
 812 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
 813
 814 /* last field in 'union bpf_attr' used by this command */
 815 #define BPF_PROG_LOAD_LAST_FIELD kern_version
 816
 817 static int bpf_prog_load(union bpf_attr *attr)
 818 {
 819         enum bpf_prog_type type = attr->prog_type;
 820         struct bpf_prog *prog;
 821         int err;
 822         char license[128];
 823         bool is_gpl;
 824
 825         if (CHECK_ATTR(BPF_PROG_LOAD))
 826                 return -EINVAL;
 827
 828         /* copy eBPF program license from user space */
 829         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
 830                               sizeof(license) - 1) < 0)
 831                 return -EFAULT;
 832         license[sizeof(license) - 1] = 0;
 833
 834         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 835         is_gpl = license_is_gpl_compatible(license);
 836
 837         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
 838                 return -E2BIG;
 839
 840         if (type == BPF_PROG_TYPE_KPROBE &&
 841             attr->kern_version != LINUX_VERSION_CODE)
 842                 return -EINVAL;
 843
 844         if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
 845                 return -EPERM;
 846
 847         /* plain bpf_prog allocation */
 848         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
 849         if (!prog)
 850                 return -ENOMEM;
 851
 852         err = bpf_prog_charge_memlock(prog);
 853         if (err)
 854                 goto free_prog_nouncharge;
 855
 856         prog->len = attr->insn_cnt;
 857
 858         err = -EFAULT;
 859         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
 860                            bpf_prog_insn_size(prog)) != 0)
 861                 goto free_prog;
 862
 863         prog->orig_prog = NULL;
 864         prog->jited = 0;
 865
 866         atomic_set(&prog->aux->refcnt, 1);
 867         prog->gpl_compatible = is_gpl ? 1 : 0;
 868
 869         /* find program type: socket_filter vs tracing_filter */
 870         err = find_prog_type(type, prog);
 871         if (err < 0)
 872                 goto free_prog;
 873
 874         /* run eBPF verifier */
 875         err = bpf_check(&prog, attr);
 876         if (err < 0)
 877                 goto free_used_maps;
 878
 879         /* fixup BPF_CALL->imm field */
 880         fixup_bpf_calls(prog);
 881
 882         /* eBPF program is ready to be JITed */
 883         prog = bpf_prog_select_runtime(prog, &err);
 884         if (err < 0)
 885                 goto free_used_maps;
 886
 887         err = bpf_prog_new_fd(prog);
 888         if (err < 0)
 889                 /* failed to allocate fd */
 890                 goto free_used_maps;
 891
 892         return err;
 893
 894 free_used_maps:
 895         free_used_maps(prog->aux);
 896 free_prog:
 897         bpf_prog_uncharge_memlock(prog);
 898 free_prog_nouncharge:
 899         bpf_prog_free(prog);
 900         return err;
 901 }
 902
 903 #define BPF_OBJ_LAST_FIELD bpf_fd
 904
 905 static int bpf_obj_pin(const union bpf_attr *attr)
 906 {
 907         if (CHECK_ATTR(BPF_OBJ))
 908                 return -EINVAL;
 909
 910         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
 911 }
 912
 913 static int bpf_obj_get(const union bpf_attr *attr)
 914 {
 915         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
 916                 return -EINVAL;
 917
 918         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname));
 919 }
 920
 921 #ifdef CONFIG_CGROUP_BPF
 922
 923 #define BPF_PROG_ATTACH_LAST_FIELD attach_type
 924
 925 static int bpf_prog_attach(const union bpf_attr *attr)
 926 {
 927         struct bpf_prog *prog;
 928         struct cgroup *cgrp;
 929         enum bpf_prog_type ptype;
 930
 931         if (!capable(CAP_NET_ADMIN))
 932                 return -EPERM;
 933
 934         if (CHECK_ATTR(BPF_PROG_ATTACH))
 935                 return -EINVAL;
 936
 937         switch (attr->attach_type) {
 938         case BPF_CGROUP_INET_INGRESS:
 939         case BPF_CGROUP_INET_EGRESS:
 940                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
 941                 break;
 942         case BPF_CGROUP_INET_SOCK_CREATE:
 943                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
 944                 break;
 945         default:
 946                 return -EINVAL;
 947         }
 948
 949         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
 950         if (IS_ERR(prog))
 951                 return PTR_ERR(prog);
 952
 953         cgrp = cgroup_get_from_fd(attr->target_fd);
 954         if (IS_ERR(cgrp)) {
 955                 bpf_prog_put(prog);
 956                 return PTR_ERR(cgrp);
 957         }
 958
 959         cgroup_bpf_update(cgrp, prog, attr->attach_type);
 960         cgroup_put(cgrp);
 961
 962         return 0;
 963 }
 964
 965 #define BPF_PROG_DETACH_LAST_FIELD attach_type
 966
 967 static int bpf_prog_detach(const union bpf_attr *attr)
 968 {
 969         struct cgroup *cgrp;
 970
 971         if (!capable(CAP_NET_ADMIN))
 972                 return -EPERM;
 973
 974         if (CHECK_ATTR(BPF_PROG_DETACH))
 975                 return -EINVAL;
 976
 977         switch (attr->attach_type) {
 978         case BPF_CGROUP_INET_INGRESS:
 979         case BPF_CGROUP_INET_EGRESS:
 980         case BPF_CGROUP_INET_SOCK_CREATE:
 981                 cgrp = cgroup_get_from_fd(attr->target_fd);
 982                 if (IS_ERR(cgrp))
 983                         return PTR_ERR(cgrp);
 984
 985                 cgroup_bpf_update(cgrp, NULL, attr->attach_type);
 986                 cgroup_put(cgrp);
 987                 break;
 988
 989         default:
 990                 return -EINVAL;
 991         }
 992
 993         return 0;
 994 }
 995 #endif /* CONFIG_CGROUP_BPF */
 996
 997 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 998 {
 999         union bpf_attr attr = {};
1000         int err;
1001
1002         if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1003                 return -EPERM;
1004
1005         if (!access_ok(VERIFY_READ, uattr, 1))
1006                 return -EFAULT;
1007
1008         if (size > PAGE_SIZE)   /* silly large */
1009                 return -E2BIG;
1010
1011         /* If we're handed a bigger struct than we know of,
1012          * ensure all the unknown bits are 0 - i.e. new
1013          * user-space does not rely on any kernel feature
1014          * extensions we dont know about yet.
1015          */
1016         if (size > sizeof(attr)) {
1017                 unsigned char __user *addr;
1018                 unsigned char __user *end;
1019                 unsigned char val;
1020
1021                 addr = (void __user *)uattr + sizeof(attr);
1022                 end  = (void __user *)uattr + size;
1023
1024                 for (; addr < end; addr++) {
1025                         err = get_user(val, addr);
1026                         if (err)
1027                                 return err;
1028                         if (val)
1029                                 return -E2BIG;
1030                 }
1031                 size = sizeof(attr);
1032         }
1033
1034         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1035         if (copy_from_user(&attr, uattr, size) != 0)
1036                 return -EFAULT;
1037
1038         switch (cmd) {
1039         case BPF_MAP_CREATE:
1040                 err = map_create(&attr);
1041                 break;
1042         case BPF_MAP_LOOKUP_ELEM:
1043                 err = map_lookup_elem(&attr);
1044                 break;
1045         case BPF_MAP_UPDATE_ELEM:
1046                 err = map_update_elem(&attr);
1047                 break;
1048         case BPF_MAP_DELETE_ELEM:
1049                 err = map_delete_elem(&attr);
1050                 break;
1051         case BPF_MAP_GET_NEXT_KEY:
1052                 err = map_get_next_key(&attr);
1053                 break;
1054         case BPF_PROG_LOAD:
1055                 err = bpf_prog_load(&attr);
1056                 break;
1057         case BPF_OBJ_PIN:
1058                 err = bpf_obj_pin(&attr);
1059                 break;
1060         case BPF_OBJ_GET:
1061                 err = bpf_obj_get(&attr);
1062                 break;
1063
1064 #ifdef CONFIG_CGROUP_BPF
1065         case BPF_PROG_ATTACH:
1066                 err = bpf_prog_attach(&attr);
1067                 break;
1068         case BPF_PROG_DETACH:
1069                 err = bpf_prog_detach(&attr);
1070                 break;
1071 #endif
1072
1073         default:
1074                 err = -EINVAL;
1075                 break;
1076         }
1077
1078         return err;
1079 }