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
  23 DEFINE_PER_CPU(int, bpf_prog_active);
  24
  25 int sysctl_unprivileged_bpf_disabled __read_mostly;
  26
  27 static LIST_HEAD(bpf_map_types);
  28
  29 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
  30 {
  31         struct bpf_map_type_list *tl;
  32         struct bpf_map *map;
  33
  34         list_for_each_entry(tl, &bpf_map_types, list_node) {
  35                 if (tl->type == attr->map_type) {
  36                         map = tl->ops->map_alloc(attr);
  37                         if (IS_ERR(map))
  38                                 return map;
  39                         map->ops = tl->ops;
  40                         map->map_type = attr->map_type;
  41                         return map;
  42                 }
  43         }
  44         return ERR_PTR(-EINVAL);
  45 }
  46
  47 /* boot time registration of different map implementations */
  48 void bpf_register_map_type(struct bpf_map_type_list *tl)
  49 {
  50         list_add(&tl->list_node, &bpf_map_types);
  51 }
  52
  53 void *bpf_map_area_alloc(size_t size)
  54 {
  55         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
  56          * trigger under memory pressure as we really just want to
  57          * fail instead.
  58          */
  59         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
  60         void *area;
  61
  62         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
  63                 area = kmalloc(size, GFP_USER | flags);
  64                 if (area != NULL)
  65                         return area;
  66         }
  67
  68         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | flags,
  69                          PAGE_KERNEL);
  70 }
  71
  72 void bpf_map_area_free(void *area)
  73 {
  74         kvfree(area);
  75 }
  76
  77 int bpf_map_precharge_memlock(u32 pages)
  78 {
  79         struct user_struct *user = get_current_user();
  80         unsigned long memlock_limit, cur;
  81
  82         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  83         cur = atomic_long_read(&user->locked_vm);
  84         free_uid(user);
  85         if (cur + pages > memlock_limit)
  86                 return -EPERM;
  87         return 0;
  88 }
  89
  90 static int bpf_map_charge_memlock(struct bpf_map *map)
  91 {
  92         struct user_struct *user = get_current_user();
  93         unsigned long memlock_limit;
  94
  95         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  96
  97         atomic_long_add(map->pages, &user->locked_vm);
  98
  99         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 100                 atomic_long_sub(map->pages, &user->locked_vm);
 101                 free_uid(user);
 102                 return -EPERM;
 103         }
 104         map->user = user;
 105         return 0;
 106 }
 107
 108 static void bpf_map_uncharge_memlock(struct bpf_map *map)
 109 {
 110         struct user_struct *user = map->user;
 111
 112         atomic_long_sub(map->pages, &user->locked_vm);
 113         free_uid(user);
 114 }
 115
 116 /* called from workqueue */
 117 static void bpf_map_free_deferred(struct work_struct *work)
 118 {
 119         struct bpf_map *map = container_of(work, struct bpf_map, work);
 120
 121         bpf_map_uncharge_memlock(map);
 122         /* implementation dependent freeing */
 123         map->ops->map_free(map);
 124 }
 125
 126 static void bpf_map_put_uref(struct bpf_map *map)
 127 {
 128         if (atomic_dec_and_test(&map->usercnt)) {
 129                 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
 130                         bpf_fd_array_map_clear(map);
 131         }
 132 }
 133
 134 /* decrement map refcnt and schedule it for freeing via workqueue
 135  * (unrelying map implementation ops->map_free() might sleep)
 136  */
 137 void bpf_map_put(struct bpf_map *map)
 138 {
 139         if (atomic_dec_and_test(&map->refcnt)) {
 140                 INIT_WORK(&map->work, bpf_map_free_deferred);
 141                 schedule_work(&map->work);
 142         }
 143 }
 144
 145 void bpf_map_put_with_uref(struct bpf_map *map)
 146 {
 147         bpf_map_put_uref(map);
 148         bpf_map_put(map);
 149 }
 150
 151 static int bpf_map_release(struct inode *inode, struct file *filp)
 152 {
 153         struct bpf_map *map = filp->private_data;
 154
 155         if (map->ops->map_release)
 156                 map->ops->map_release(map, filp);
 157
 158         bpf_map_put_with_uref(map);
 159         return 0;
 160 }
 161
 162 #ifdef CONFIG_PROC_FS
 163 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 164 {
 165         const struct bpf_map *map = filp->private_data;
 166
 167         seq_printf(m,
 168                    "map_type:\t%u\n"
 169                    "key_size:\t%u\n"
 170                    "value_size:\t%u\n"
 171                    "max_entries:\t%u\n"
 172                    "map_flags:\t%#x\n",
 173                    map->map_type,
 174                    map->key_size,
 175                    map->value_size,
 176                    map->max_entries,
 177                    map->map_flags);
 178 }
 179 #endif
 180
 181 static const struct file_operations bpf_map_fops = {
 182 #ifdef CONFIG_PROC_FS
 183         .show_fdinfo    = bpf_map_show_fdinfo,
 184 #endif
 185         .release        = bpf_map_release,
 186 };
 187
 188 int bpf_map_new_fd(struct bpf_map *map)
 189 {
 190         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 191                                 O_RDWR | O_CLOEXEC);
 192 }
 193
 194 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 195 #define CHECK_ATTR(CMD) \
 196         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 197                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 198                    sizeof(*attr) - \
 199                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 200                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 201
 202 #define BPF_MAP_CREATE_LAST_FIELD map_flags
 203 /* called via syscall */
 204 static int map_create(union bpf_attr *attr)
 205 {
 206         struct bpf_map *map;
 207         int err;
 208
 209         err = CHECK_ATTR(BPF_MAP_CREATE);
 210         if (err)
 211                 return -EINVAL;
 212
 213         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 214         map = find_and_alloc_map(attr);
 215         if (IS_ERR(map))
 216                 return PTR_ERR(map);
 217
 218         atomic_set(&map->refcnt, 1);
 219         atomic_set(&map->usercnt, 1);
 220
 221         err = bpf_map_charge_memlock(map);
 222         if (err)
 223                 goto free_map_nouncharge;
 224
 225         err = bpf_map_new_fd(map);
 226         if (err < 0)
 227                 /* failed to allocate fd */
 228                 goto free_map;
 229
 230         return err;
 231
 232 free_map:
 233         bpf_map_uncharge_memlock(map);
 234 free_map_nouncharge:
 235         map->ops->map_free(map);
 236         return err;
 237 }
 238
 239 /* if error is returned, fd is released.
 240  * On success caller should complete fd access with matching fdput()
 241  */
 242 struct bpf_map *__bpf_map_get(struct fd f)
 243 {
 244         if (!f.file)
 245                 return ERR_PTR(-EBADF);
 246         if (f.file->f_op != &bpf_map_fops) {
 247                 fdput(f);
 248                 return ERR_PTR(-EINVAL);
 249         }
 250
 251         return f.file->private_data;
 252 }
 253
 254 /* prog's and map's refcnt limit */
 255 #define BPF_MAX_REFCNT 32768
 256
 257 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 258 {
 259         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 260                 atomic_dec(&map->refcnt);
 261                 return ERR_PTR(-EBUSY);
 262         }
 263         if (uref)
 264                 atomic_inc(&map->usercnt);
 265         return map;
 266 }
 267
 268 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 269 {
 270         struct fd f = fdget(ufd);
 271         struct bpf_map *map;
 272
 273         map = __bpf_map_get(f);
 274         if (IS_ERR(map))
 275                 return map;
 276
 277         map = bpf_map_inc(map, true);
 278         fdput(f);
 279
 280         return map;
 281 }
 282
 283 /* helper to convert user pointers passed inside __aligned_u64 fields */
 284 static void __user *u64_to_ptr(__u64 val)
 285 {
 286         return (void __user *) (unsigned long) val;
 287 }
 288
 289 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 290 {
 291         return -ENOTSUPP;
 292 }
 293
 294 /* last field in 'union bpf_attr' used by this command */
 295 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 296
 297 static int map_lookup_elem(union bpf_attr *attr)
 298 {
 299         void __user *ukey = u64_to_ptr(attr->key);
 300         void __user *uvalue = u64_to_ptr(attr->value);
 301         int ufd = attr->map_fd;
 302         struct bpf_map *map;
 303         void *key, *value, *ptr;
 304         u32 value_size;
 305         struct fd f;
 306         int err;
 307
 308         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 309                 return -EINVAL;
 310
 311         f = fdget(ufd);
 312         map = __bpf_map_get(f);
 313         if (IS_ERR(map))
 314                 return PTR_ERR(map);
 315
 316         err = -ENOMEM;
 317         key = kmalloc(map->key_size, GFP_USER);
 318         if (!key)
 319                 goto err_put;
 320
 321         err = -EFAULT;
 322         if (copy_from_user(key, ukey, map->key_size) != 0)
 323                 goto free_key;
 324
 325         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 326             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 327                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 328         else
 329                 value_size = map->value_size;
 330
 331         err = -ENOMEM;
 332         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 333         if (!value)
 334                 goto free_key;
 335
 336         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
 337                 err = bpf_percpu_hash_copy(map, key, value);
 338         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 339                 err = bpf_percpu_array_copy(map, key, value);
 340         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 341                 err = bpf_stackmap_copy(map, key, value);
 342         } else {
 343                 rcu_read_lock();
 344                 ptr = map->ops->map_lookup_elem(map, key);
 345                 if (ptr)
 346                         memcpy(value, ptr, value_size);
 347                 rcu_read_unlock();
 348                 err = ptr ? 0 : -ENOENT;
 349         }
 350
 351         if (err)
 352                 goto free_value;
 353
 354         err = -EFAULT;
 355         if (copy_to_user(uvalue, value, value_size) != 0)
 356                 goto free_value;
 357
 358         err = 0;
 359
 360 free_value:
 361         kfree(value);
 362 free_key:
 363         kfree(key);
 364 err_put:
 365         fdput(f);
 366         return err;
 367 }
 368
 369 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 370
 371 static int map_update_elem(union bpf_attr *attr)
 372 {
 373         void __user *ukey = u64_to_ptr(attr->key);
 374         void __user *uvalue = u64_to_ptr(attr->value);
 375         int ufd = attr->map_fd;
 376         struct bpf_map *map;
 377         void *key, *value;
 378         u32 value_size;
 379         struct fd f;
 380         int err;
 381
 382         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 383                 return -EINVAL;
 384
 385         f = fdget(ufd);
 386         map = __bpf_map_get(f);
 387         if (IS_ERR(map))
 388                 return PTR_ERR(map);
 389
 390         err = -ENOMEM;
 391         key = kmalloc(map->key_size, GFP_USER);
 392         if (!key)
 393                 goto err_put;
 394
 395         err = -EFAULT;
 396         if (copy_from_user(key, ukey, map->key_size) != 0)
 397                 goto free_key;
 398
 399         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 400             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 401                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 402         else
 403                 value_size = map->value_size;
 404
 405         err = -ENOMEM;
 406         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 407         if (!value)
 408                 goto free_key;
 409
 410         err = -EFAULT;
 411         if (copy_from_user(value, uvalue, value_size) != 0)
 412                 goto free_value;
 413
 414         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 415          * inside bpf map update or delete otherwise deadlocks are possible
 416          */
 417         preempt_disable();
 418         __this_cpu_inc(bpf_prog_active);
 419         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH) {
 420                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 421         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 422                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 423         } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
 424                    map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
 425                    map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY) {
 426                 rcu_read_lock();
 427                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 428                                                    attr->flags);
 429                 rcu_read_unlock();
 430         } else {
 431                 rcu_read_lock();
 432                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 433                 rcu_read_unlock();
 434         }
 435         __this_cpu_dec(bpf_prog_active);
 436         preempt_enable();
 437
 438 free_value:
 439         kfree(value);
 440 free_key:
 441         kfree(key);
 442 err_put:
 443         fdput(f);
 444         return err;
 445 }
 446
 447 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 448
 449 static int map_delete_elem(union bpf_attr *attr)
 450 {
 451         void __user *ukey = u64_to_ptr(attr->key);
 452         int ufd = attr->map_fd;
 453         struct bpf_map *map;
 454         struct fd f;
 455         void *key;
 456         int err;
 457
 458         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 459                 return -EINVAL;
 460
 461         f = fdget(ufd);
 462         map = __bpf_map_get(f);
 463         if (IS_ERR(map))
 464                 return PTR_ERR(map);
 465
 466         err = -ENOMEM;
 467         key = kmalloc(map->key_size, GFP_USER);
 468         if (!key)
 469                 goto err_put;
 470
 471         err = -EFAULT;
 472         if (copy_from_user(key, ukey, map->key_size) != 0)
 473                 goto free_key;
 474
 475         preempt_disable();
 476         __this_cpu_inc(bpf_prog_active);
 477         rcu_read_lock();
 478         err = map->ops->map_delete_elem(map, key);
 479         rcu_read_unlock();
 480         __this_cpu_dec(bpf_prog_active);
 481         preempt_enable();
 482
 483 free_key:
 484         kfree(key);
 485 err_put:
 486         fdput(f);
 487         return err;
 488 }
 489
 490 /* last field in 'union bpf_attr' used by this command */
 491 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 492
 493 static int map_get_next_key(union bpf_attr *attr)
 494 {
 495         void __user *ukey = u64_to_ptr(attr->key);
 496         void __user *unext_key = u64_to_ptr(attr->next_key);
 497         int ufd = attr->map_fd;
 498         struct bpf_map *map;
 499         void *key, *next_key;
 500         struct fd f;
 501         int err;
 502
 503         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 504                 return -EINVAL;
 505
 506         f = fdget(ufd);
 507         map = __bpf_map_get(f);
 508         if (IS_ERR(map))
 509                 return PTR_ERR(map);
 510
 511         if (ukey) {
 512                 err = -ENOMEM;
 513                 key = kmalloc(map->key_size, GFP_USER);
 514                 if (!key)
 515                         goto err_put;
 516
 517                 err = -EFAULT;
 518                 if (copy_from_user(key, ukey, map->key_size) != 0)
 519                         goto free_key;
 520         } else {
 521                 key = NULL;
 522         }
 523
 524         err = -ENOMEM;
 525         next_key = kmalloc(map->key_size, GFP_USER);
 526         if (!next_key)
 527                 goto free_key;
 528
 529         rcu_read_lock();
 530         err = map->ops->map_get_next_key(map, key, next_key);
 531         rcu_read_unlock();
 532         if (err)
 533                 goto free_next_key;
 534
 535         err = -EFAULT;
 536         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 537                 goto free_next_key;
 538
 539         err = 0;
 540
 541 free_next_key:
 542         kfree(next_key);
 543 free_key:
 544         kfree(key);
 545 err_put:
 546         fdput(f);
 547         return err;
 548 }
 549
 550 static LIST_HEAD(bpf_prog_types);
 551
 552 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 553 {
 554         struct bpf_prog_type_list *tl;
 555
 556         list_for_each_entry(tl, &bpf_prog_types, list_node) {
 557                 if (tl->type == type) {
 558                         prog->aux->ops = tl->ops;
 559                         prog->type = type;
 560                         return 0;
 561                 }
 562         }
 563
 564         return -EINVAL;
 565 }
 566
 567 void bpf_register_prog_type(struct bpf_prog_type_list *tl)
 568 {
 569         list_add(&tl->list_node, &bpf_prog_types);
 570 }
 571
 572 /* drop refcnt on maps used by eBPF program and free auxilary data */
 573 static void free_used_maps(struct bpf_prog_aux *aux)
 574 {
 575         int i;
 576
 577         for (i = 0; i < aux->used_map_cnt; i++)
 578                 bpf_map_put(aux->used_maps[i]);
 579
 580         kfree(aux->used_maps);
 581 }
 582
 583 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
 584 {
 585         struct user_struct *user = get_current_user();
 586         unsigned long memlock_limit;
 587
 588         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 589
 590         atomic_long_add(prog->pages, &user->locked_vm);
 591         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 592                 atomic_long_sub(prog->pages, &user->locked_vm);
 593                 free_uid(user);
 594                 return -EPERM;
 595         }
 596         prog->aux->user = user;
 597         return 0;
 598 }
 599
 600 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
 601 {
 602         struct user_struct *user = prog->aux->user;
 603
 604         atomic_long_sub(prog->pages, &user->locked_vm);
 605         free_uid(user);
 606 }
 607
 608 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
 609 {
 610         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 611
 612         free_used_maps(aux);
 613         bpf_prog_uncharge_memlock(aux->prog);
 614         bpf_prog_free(aux->prog);
 615 }
 616
 617 void bpf_prog_put(struct bpf_prog *prog)
 618 {
 619         if (atomic_dec_and_test(&prog->aux->refcnt))
 620                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
 621 }
 622 EXPORT_SYMBOL_GPL(bpf_prog_put);
 623
 624 static int bpf_prog_release(struct inode *inode, struct file *filp)
 625 {
 626         struct bpf_prog *prog = filp->private_data;
 627
 628         bpf_prog_put(prog);
 629         return 0;
 630 }
 631
 632 static const struct file_operations bpf_prog_fops = {
 633         .release = bpf_prog_release,
 634 };
 635
 636 int bpf_prog_new_fd(struct bpf_prog *prog)
 637 {
 638         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
 639                                 O_RDWR | O_CLOEXEC);
 640 }
 641
 642 static struct bpf_prog *____bpf_prog_get(struct fd f)
 643 {
 644         if (!f.file)
 645                 return ERR_PTR(-EBADF);
 646         if (f.file->f_op != &bpf_prog_fops) {
 647                 fdput(f);
 648                 return ERR_PTR(-EINVAL);
 649         }
 650
 651         return f.file->private_data;
 652 }
 653
 654 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
 655 {
 656         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
 657                 atomic_sub(i, &prog->aux->refcnt);
 658                 return ERR_PTR(-EBUSY);
 659         }
 660         return prog;
 661 }
 662 EXPORT_SYMBOL_GPL(bpf_prog_add);
 663
 664 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
 665 {
 666         return bpf_prog_add(prog, 1);
 667 }
 668
 669 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *type)
 670 {
 671         struct fd f = fdget(ufd);
 672         struct bpf_prog *prog;
 673
 674         prog = ____bpf_prog_get(f);
 675         if (IS_ERR(prog))
 676                 return prog;
 677         if (type && prog->type != *type) {
 678                 prog = ERR_PTR(-EINVAL);
 679                 goto out;
 680         }
 681
 682         prog = bpf_prog_inc(prog);
 683 out:
 684         fdput(f);
 685         return prog;
 686 }
 687
 688 struct bpf_prog *bpf_prog_get(u32 ufd)
 689 {
 690         return __bpf_prog_get(ufd, NULL);
 691 }
 692
 693 struct bpf_prog *bpf_prog_get_type(u32 ufd, enum bpf_prog_type type)
 694 {
 695         return __bpf_prog_get(ufd, &type);
 696 }
 697 EXPORT_SYMBOL_GPL(bpf_prog_get_type);
 698
 699 /* last field in 'union bpf_attr' used by this command */
 700 #define BPF_PROG_LOAD_LAST_FIELD kern_version
 701
 702 static int bpf_prog_load(union bpf_attr *attr)
 703 {
 704         enum bpf_prog_type type = attr->prog_type;
 705         struct bpf_prog *prog;
 706         int err;
 707         char license[128];
 708         bool is_gpl;
 709
 710         if (CHECK_ATTR(BPF_PROG_LOAD))
 711                 return -EINVAL;
 712
 713         /* copy eBPF program license from user space */
 714         if (strncpy_from_user(license, u64_to_ptr(attr->license),
 715                               sizeof(license) - 1) < 0)
 716                 return -EFAULT;
 717         license[sizeof(license) - 1] = 0;
 718
 719         /* eBPF programs must be GPL compatible to use GPL-ed functions */
 720         is_gpl = license_is_gpl_compatible(license);
 721
 722         if (attr->insn_cnt >= BPF_MAXINSNS)
 723                 return -EINVAL;
 724
 725         if (type == BPF_PROG_TYPE_KPROBE &&
 726             attr->kern_version != LINUX_VERSION_CODE)
 727                 return -EINVAL;
 728
 729         if (type != BPF_PROG_TYPE_SOCKET_FILTER && !capable(CAP_SYS_ADMIN))
 730                 return -EPERM;
 731
 732         /* plain bpf_prog allocation */
 733         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
 734         if (!prog)
 735                 return -ENOMEM;
 736
 737         err = bpf_prog_charge_memlock(prog);
 738         if (err)
 739                 goto free_prog_nouncharge;
 740
 741         prog->len = attr->insn_cnt;
 742
 743         err = -EFAULT;
 744         if (copy_from_user(prog->insns, u64_to_ptr(attr->insns),
 745                            prog->len * sizeof(struct bpf_insn)) != 0)
 746                 goto free_prog;
 747
 748         prog->orig_prog = NULL;
 749         prog->jited = 0;
 750
 751         atomic_set(&prog->aux->refcnt, 1);
 752         prog->gpl_compatible = is_gpl ? 1 : 0;
 753
 754         /* find program type: socket_filter vs tracing_filter */
 755         err = find_prog_type(type, prog);
 756         if (err < 0)
 757                 goto free_prog;
 758
 759         /* run eBPF verifier */
 760         err = bpf_check(&prog, attr);
 761         if (err < 0)
 762                 goto free_used_maps;
 763
 764         /* eBPF program is ready to be JITed */
 765         prog = bpf_prog_select_runtime(prog, &err);
 766         if (err < 0)
 767                 goto free_used_maps;
 768
 769         err = bpf_prog_new_fd(prog);
 770         if (err < 0)
 771                 /* failed to allocate fd */
 772                 goto free_used_maps;
 773
 774         return err;
 775
 776 free_used_maps:
 777         free_used_maps(prog->aux);
 778 free_prog:
 779         bpf_prog_uncharge_memlock(prog);
 780 free_prog_nouncharge:
 781         bpf_prog_free(prog);
 782         return err;
 783 }
 784
 785 #define BPF_OBJ_LAST_FIELD bpf_fd
 786
 787 static int bpf_obj_pin(const union bpf_attr *attr)
 788 {
 789         if (CHECK_ATTR(BPF_OBJ))
 790                 return -EINVAL;
 791
 792         return bpf_obj_pin_user(attr->bpf_fd, u64_to_ptr(attr->pathname));
 793 }
 794
 795 static int bpf_obj_get(const union bpf_attr *attr)
 796 {
 797         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0)
 798                 return -EINVAL;
 799
 800         return bpf_obj_get_user(u64_to_ptr(attr->pathname));
 801 }
 802
 803 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
 804 {
 805         union bpf_attr attr;
 806         int err;
 807
 808         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
 809                 return -EPERM;
 810
 811         if (!access_ok(VERIFY_READ, uattr, 1))
 812                 return -EFAULT;
 813
 814         if (size > PAGE_SIZE)   /* silly large */
 815                 return -E2BIG;
 816
 817         /* If we're handed a bigger struct than we know of,
 818          * ensure all the unknown bits are 0 - i.e. new
 819          * user-space does not rely on any kernel feature
 820          * extensions we dont know about yet.
 821          */
 822         if (size > sizeof(attr)) {
 823                 unsigned char __user *addr;
 824                 unsigned char __user *end;
 825                 unsigned char val;
 826
 827                 addr = (void __user *)uattr + sizeof(attr);
 828                 end  = (void __user *)uattr + size;
 829
 830                 for (; addr < end; addr++) {
 831                         err = get_user(val, addr);
 832                         if (err)
 833                                 return err;
 834                         if (val)
 835                                 return -E2BIG;
 836                 }
 837                 size = sizeof(attr);
 838         }
 839
 840         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
 841         memset(&attr, 0, sizeof(attr));
 842         if (copy_from_user(&attr, uattr, size) != 0)
 843                 return -EFAULT;
 844
 845         switch (cmd) {
 846         case BPF_MAP_CREATE:
 847                 err = map_create(&attr);
 848                 break;
 849         case BPF_MAP_LOOKUP_ELEM:
 850                 err = map_lookup_elem(&attr);
 851                 break;
 852         case BPF_MAP_UPDATE_ELEM:
 853                 err = map_update_elem(&attr);
 854                 break;
 855         case BPF_MAP_DELETE_ELEM:
 856                 err = map_delete_elem(&attr);
 857                 break;
 858         case BPF_MAP_GET_NEXT_KEY:
 859                 err = map_get_next_key(&attr);
 860                 break;
 861         case BPF_PROG_LOAD:
 862                 err = bpf_prog_load(&attr);
 863                 break;
 864         case BPF_OBJ_PIN:
 865                 err = bpf_obj_pin(&attr);
 866                 break;
 867         case BPF_OBJ_GET:
 868                 err = bpf_obj_get(&attr);
 869                 break;
 870         default:
 871                 err = -EINVAL;
 872                 break;
 873         }
 874
 875         return err;
 876 }