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/bpf_trace.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/slab.h>
  16 #include <linux/sched/signal.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/mmzone.h>
  19 #include <linux/anon_inodes.h>
  20 #include <linux/file.h>
  21 #include <linux/license.h>
  22 #include <linux/filter.h>
  23 #include <linux/version.h>
  24 #include <linux/kernel.h>
  25 #include <linux/idr.h>
  26 #include <linux/cred.h>
  27 #include <linux/timekeeping.h>
  28 #include <linux/ctype.h>
  29
  30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  31                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  32                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  33                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  36
  37 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  38
  39 DEFINE_PER_CPU(int, bpf_prog_active);
  40 static DEFINE_IDR(prog_idr);
  41 static DEFINE_SPINLOCK(prog_idr_lock);
  42 static DEFINE_IDR(map_idr);
  43 static DEFINE_SPINLOCK(map_idr_lock);
  44
  45 int sysctl_unprivileged_bpf_disabled __read_mostly;
  46
  47 static const struct bpf_map_ops * const bpf_map_types[] = {
  48 #define BPF_PROG_TYPE(_id, _ops)
  49 #define BPF_MAP_TYPE(_id, _ops) \
  50         [_id] = &_ops,
  51 #include <linux/bpf_types.h>
  52 #undef BPF_PROG_TYPE
  53 #undef BPF_MAP_TYPE
  54 };
  55
  56 /*
  57  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  58  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  59  * we don't know about yet.
  60  *
  61  * There is a ToCToU between this function call and the following
  62  * copy_from_user() call. However, this is not a concern since this function is
  63  * meant to be a future-proofing of bits.
  64  */
  65 static int check_uarg_tail_zero(void __user *uaddr,
  66                                 size_t expected_size,
  67                                 size_t actual_size)
  68 {
  69         unsigned char __user *addr;
  70         unsigned char __user *end;
  71         unsigned char val;
  72         int err;
  73
  74         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  75                 return -E2BIG;
  76
  77         if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
  78                 return -EFAULT;
  79
  80         if (actual_size <= expected_size)
  81                 return 0;
  82
  83         addr = uaddr + expected_size;
  84         end  = uaddr + actual_size;
  85
  86         for (; addr < end; addr++) {
  87                 err = get_user(val, addr);
  88                 if (err)
  89                         return err;
  90                 if (val)
  91                         return -E2BIG;
  92         }
  93
  94         return 0;
  95 }
  96
  97 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
  98 {
  99         struct bpf_map *map;
 100
 101         if (attr->map_type >= ARRAY_SIZE(bpf_map_types) ||
 102             !bpf_map_types[attr->map_type])
 103                 return ERR_PTR(-EINVAL);
 104
 105         map = bpf_map_types[attr->map_type]->map_alloc(attr);
 106         if (IS_ERR(map))
 107                 return map;
 108         map->ops = bpf_map_types[attr->map_type];
 109         map->map_type = attr->map_type;
 110         return map;
 111 }
 112
 113 void *bpf_map_area_alloc(size_t size, int numa_node)
 114 {
 115         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
 116          * trigger under memory pressure as we really just want to
 117          * fail instead.
 118          */
 119         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
 120         void *area;
 121
 122         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 123                 area = kmalloc_node(size, GFP_USER | flags, numa_node);
 124                 if (area != NULL)
 125                         return area;
 126         }
 127
 128         return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
 129                                            __builtin_return_address(0));
 130 }
 131
 132 void bpf_map_area_free(void *area)
 133 {
 134         kvfree(area);
 135 }
 136
 137 int bpf_map_precharge_memlock(u32 pages)
 138 {
 139         struct user_struct *user = get_current_user();
 140         unsigned long memlock_limit, cur;
 141
 142         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 143         cur = atomic_long_read(&user->locked_vm);
 144         free_uid(user);
 145         if (cur + pages > memlock_limit)
 146                 return -EPERM;
 147         return 0;
 148 }
 149
 150 static int bpf_map_charge_memlock(struct bpf_map *map)
 151 {
 152         struct user_struct *user = get_current_user();
 153         unsigned long memlock_limit;
 154
 155         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 156
 157         atomic_long_add(map->pages, &user->locked_vm);
 158
 159         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 160                 atomic_long_sub(map->pages, &user->locked_vm);
 161                 free_uid(user);
 162                 return -EPERM;
 163         }
 164         map->user = user;
 165         return 0;
 166 }
 167
 168 static void bpf_map_uncharge_memlock(struct bpf_map *map)
 169 {
 170         struct user_struct *user = map->user;
 171
 172         atomic_long_sub(map->pages, &user->locked_vm);
 173         free_uid(user);
 174 }
 175
 176 static int bpf_map_alloc_id(struct bpf_map *map)
 177 {
 178         int id;
 179
 180         spin_lock_bh(&map_idr_lock);
 181         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 182         if (id > 0)
 183                 map->id = id;
 184         spin_unlock_bh(&map_idr_lock);
 185
 186         if (WARN_ON_ONCE(!id))
 187                 return -ENOSPC;
 188
 189         return id > 0 ? 0 : id;
 190 }
 191
 192 static void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 193 {
 194         unsigned long flags;
 195
 196         if (do_idr_lock)
 197                 spin_lock_irqsave(&map_idr_lock, flags);
 198         else
 199                 __acquire(&map_idr_lock);
 200
 201         idr_remove(&map_idr, map->id);
 202
 203         if (do_idr_lock)
 204                 spin_unlock_irqrestore(&map_idr_lock, flags);
 205         else
 206                 __release(&map_idr_lock);
 207 }
 208
 209 /* called from workqueue */
 210 static void bpf_map_free_deferred(struct work_struct *work)
 211 {
 212         struct bpf_map *map = container_of(work, struct bpf_map, work);
 213
 214         bpf_map_uncharge_memlock(map);
 215         security_bpf_map_free(map);
 216         /* implementation dependent freeing */
 217         map->ops->map_free(map);
 218 }
 219
 220 static void bpf_map_put_uref(struct bpf_map *map)
 221 {
 222         if (atomic_dec_and_test(&map->usercnt)) {
 223                 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
 224                         bpf_fd_array_map_clear(map);
 225         }
 226 }
 227
 228 /* decrement map refcnt and schedule it for freeing via workqueue
 229  * (unrelying map implementation ops->map_free() might sleep)
 230  */
 231 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 232 {
 233         if (atomic_dec_and_test(&map->refcnt)) {
 234                 /* bpf_map_free_id() must be called first */
 235                 bpf_map_free_id(map, do_idr_lock);
 236                 INIT_WORK(&map->work, bpf_map_free_deferred);
 237                 schedule_work(&map->work);
 238         }
 239 }
 240
 241 void bpf_map_put(struct bpf_map *map)
 242 {
 243         __bpf_map_put(map, true);
 244 }
 245
 246 void bpf_map_put_with_uref(struct bpf_map *map)
 247 {
 248         bpf_map_put_uref(map);
 249         bpf_map_put(map);
 250 }
 251
 252 static int bpf_map_release(struct inode *inode, struct file *filp)
 253 {
 254         struct bpf_map *map = filp->private_data;
 255
 256         if (map->ops->map_release)
 257                 map->ops->map_release(map, filp);
 258
 259         bpf_map_put_with_uref(map);
 260         return 0;
 261 }
 262
 263 #ifdef CONFIG_PROC_FS
 264 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 265 {
 266         const struct bpf_map *map = filp->private_data;
 267         const struct bpf_array *array;
 268         u32 owner_prog_type = 0;
 269         u32 owner_jited = 0;
 270
 271         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 272                 array = container_of(map, struct bpf_array, map);
 273                 owner_prog_type = array->owner_prog_type;
 274                 owner_jited = array->owner_jited;
 275         }
 276
 277         seq_printf(m,
 278                    "map_type:\t%u\n"
 279                    "key_size:\t%u\n"
 280                    "value_size:\t%u\n"
 281                    "max_entries:\t%u\n"
 282                    "map_flags:\t%#x\n"
 283                    "memlock:\t%llu\n",
 284                    map->map_type,
 285                    map->key_size,
 286                    map->value_size,
 287                    map->max_entries,
 288                    map->map_flags,
 289                    map->pages * 1ULL << PAGE_SHIFT);
 290
 291         if (owner_prog_type) {
 292                 seq_printf(m, "owner_prog_type:\t%u\n",
 293                            owner_prog_type);
 294                 seq_printf(m, "owner_jited:\t%u\n",
 295                            owner_jited);
 296         }
 297 }
 298 #endif
 299
 300 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 301                               loff_t *ppos)
 302 {
 303         /* We need this handler such that alloc_file() enables
 304          * f_mode with FMODE_CAN_READ.
 305          */
 306         return -EINVAL;
 307 }
 308
 309 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 310                                size_t siz, loff_t *ppos)
 311 {
 312         /* We need this handler such that alloc_file() enables
 313          * f_mode with FMODE_CAN_WRITE.
 314          */
 315         return -EINVAL;
 316 }
 317
 318 const struct file_operations bpf_map_fops = {
 319 #ifdef CONFIG_PROC_FS
 320         .show_fdinfo    = bpf_map_show_fdinfo,
 321 #endif
 322         .release        = bpf_map_release,
 323         .read           = bpf_dummy_read,
 324         .write          = bpf_dummy_write,
 325 };
 326
 327 int bpf_map_new_fd(struct bpf_map *map, int flags)
 328 {
 329         int ret;
 330
 331         ret = security_bpf_map(map, OPEN_FMODE(flags));
 332         if (ret < 0)
 333                 return ret;
 334
 335         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 336                                 flags | O_CLOEXEC);
 337 }
 338
 339 int bpf_get_file_flag(int flags)
 340 {
 341         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 342                 return -EINVAL;
 343         if (flags & BPF_F_RDONLY)
 344                 return O_RDONLY;
 345         if (flags & BPF_F_WRONLY)
 346                 return O_WRONLY;
 347         return O_RDWR;
 348 }
 349
 350 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 351 #define CHECK_ATTR(CMD) \
 352         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 353                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 354                    sizeof(*attr) - \
 355                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 356                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 357
 358 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 359  * Return 0 on success and < 0 on error.
 360  */
 361 static int bpf_obj_name_cpy(char *dst, const char *src)
 362 {
 363         const char *end = src + BPF_OBJ_NAME_LEN;
 364
 365         memset(dst, 0, BPF_OBJ_NAME_LEN);
 366
 367         /* Copy all isalnum() and '_' char */
 368         while (src < end && *src) {
 369                 if (!isalnum(*src) && *src != '_')
 370                         return -EINVAL;
 371                 *dst++ = *src++;
 372         }
 373
 374         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 375         if (src == end)
 376                 return -EINVAL;
 377
 378         return 0;
 379 }
 380
 381 #define BPF_MAP_CREATE_LAST_FIELD map_name
 382 /* called via syscall */
 383 static int map_create(union bpf_attr *attr)
 384 {
 385         int numa_node = bpf_map_attr_numa_node(attr);
 386         struct bpf_map *map;
 387         int f_flags;
 388         int err;
 389
 390         err = CHECK_ATTR(BPF_MAP_CREATE);
 391         if (err)
 392                 return -EINVAL;
 393
 394         f_flags = bpf_get_file_flag(attr->map_flags);
 395         if (f_flags < 0)
 396                 return f_flags;
 397
 398         if (numa_node != NUMA_NO_NODE &&
 399             ((unsigned int)numa_node >= nr_node_ids ||
 400              !node_online(numa_node)))
 401                 return -EINVAL;
 402
 403         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 404         map = find_and_alloc_map(attr);
 405         if (IS_ERR(map))
 406                 return PTR_ERR(map);
 407
 408         err = bpf_obj_name_cpy(map->name, attr->map_name);
 409         if (err)
 410                 goto free_map_nouncharge;
 411
 412         atomic_set(&map->refcnt, 1);
 413         atomic_set(&map->usercnt, 1);
 414
 415         err = security_bpf_map_alloc(map);
 416         if (err)
 417                 goto free_map_nouncharge;
 418
 419         err = bpf_map_charge_memlock(map);
 420         if (err)
 421                 goto free_map_sec;
 422
 423         err = bpf_map_alloc_id(map);
 424         if (err)
 425                 goto free_map;
 426
 427         err = bpf_map_new_fd(map, f_flags);
 428         if (err < 0) {
 429                 /* failed to allocate fd.
 430                  * bpf_map_put() is needed because the above
 431                  * bpf_map_alloc_id() has published the map
 432                  * to the userspace and the userspace may
 433                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 434                  */
 435                 bpf_map_put(map);
 436                 return err;
 437         }
 438
 439         trace_bpf_map_create(map, err);
 440         return err;
 441
 442 free_map:
 443         bpf_map_uncharge_memlock(map);
 444 free_map_sec:
 445         security_bpf_map_free(map);
 446 free_map_nouncharge:
 447         map->ops->map_free(map);
 448         return err;
 449 }
 450
 451 /* if error is returned, fd is released.
 452  * On success caller should complete fd access with matching fdput()
 453  */
 454 struct bpf_map *__bpf_map_get(struct fd f)
 455 {
 456         if (!f.file)
 457                 return ERR_PTR(-EBADF);
 458         if (f.file->f_op != &bpf_map_fops) {
 459                 fdput(f);
 460                 return ERR_PTR(-EINVAL);
 461         }
 462
 463         return f.file->private_data;
 464 }
 465
 466 /* prog's and map's refcnt limit */
 467 #define BPF_MAX_REFCNT 32768
 468
 469 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 470 {
 471         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 472                 atomic_dec(&map->refcnt);
 473                 return ERR_PTR(-EBUSY);
 474         }
 475         if (uref)
 476                 atomic_inc(&map->usercnt);
 477         return map;
 478 }
 479
 480 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 481 {
 482         struct fd f = fdget(ufd);
 483         struct bpf_map *map;
 484
 485         map = __bpf_map_get(f);
 486         if (IS_ERR(map))
 487                 return map;
 488
 489         map = bpf_map_inc(map, true);
 490         fdput(f);
 491
 492         return map;
 493 }
 494
 495 /* map_idr_lock should have been held */
 496 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 497                                             bool uref)
 498 {
 499         int refold;
 500
 501         refold = __atomic_add_unless(&map->refcnt, 1, 0);
 502
 503         if (refold >= BPF_MAX_REFCNT) {
 504                 __bpf_map_put(map, false);
 505                 return ERR_PTR(-EBUSY);
 506         }
 507
 508         if (!refold)
 509                 return ERR_PTR(-ENOENT);
 510
 511         if (uref)
 512                 atomic_inc(&map->usercnt);
 513
 514         return map;
 515 }
 516
 517 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 518 {
 519         return -ENOTSUPP;
 520 }
 521
 522 /* last field in 'union bpf_attr' used by this command */
 523 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 524
 525 static int map_lookup_elem(union bpf_attr *attr)
 526 {
 527         void __user *ukey = u64_to_user_ptr(attr->key);
 528         void __user *uvalue = u64_to_user_ptr(attr->value);
 529         int ufd = attr->map_fd;
 530         struct bpf_map *map;
 531         void *key, *value, *ptr;
 532         u32 value_size;
 533         struct fd f;
 534         int err;
 535
 536         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 537                 return -EINVAL;
 538
 539         f = fdget(ufd);
 540         map = __bpf_map_get(f);
 541         if (IS_ERR(map))
 542                 return PTR_ERR(map);
 543
 544         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 545                 err = -EPERM;
 546                 goto err_put;
 547         }
 548
 549         key = memdup_user(ukey, map->key_size);
 550         if (IS_ERR(key)) {
 551                 err = PTR_ERR(key);
 552                 goto err_put;
 553         }
 554
 555         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 556             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 557             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 558                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 559         else if (IS_FD_MAP(map))
 560                 value_size = sizeof(u32);
 561         else
 562                 value_size = map->value_size;
 563
 564         err = -ENOMEM;
 565         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 566         if (!value)
 567                 goto free_key;
 568
 569         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 570             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 571                 err = bpf_percpu_hash_copy(map, key, value);
 572         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 573                 err = bpf_percpu_array_copy(map, key, value);
 574         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 575                 err = bpf_stackmap_copy(map, key, value);
 576         } else if (IS_FD_ARRAY(map)) {
 577                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 578         } else if (IS_FD_HASH(map)) {
 579                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 580         } else {
 581                 rcu_read_lock();
 582                 ptr = map->ops->map_lookup_elem(map, key);
 583                 if (ptr)
 584                         memcpy(value, ptr, value_size);
 585                 rcu_read_unlock();
 586                 err = ptr ? 0 : -ENOENT;
 587         }
 588
 589         if (err)
 590                 goto free_value;
 591
 592         err = -EFAULT;
 593         if (copy_to_user(uvalue, value, value_size) != 0)
 594                 goto free_value;
 595
 596         trace_bpf_map_lookup_elem(map, ufd, key, value);
 597         err = 0;
 598
 599 free_value:
 600         kfree(value);
 601 free_key:
 602         kfree(key);
 603 err_put:
 604         fdput(f);
 605         return err;
 606 }
 607
 608 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 609
 610 static int map_update_elem(union bpf_attr *attr)
 611 {
 612         void __user *ukey = u64_to_user_ptr(attr->key);
 613         void __user *uvalue = u64_to_user_ptr(attr->value);
 614         int ufd = attr->map_fd;
 615         struct bpf_map *map;
 616         void *key, *value;
 617         u32 value_size;
 618         struct fd f;
 619         int err;
 620
 621         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 622                 return -EINVAL;
 623
 624         f = fdget(ufd);
 625         map = __bpf_map_get(f);
 626         if (IS_ERR(map))
 627                 return PTR_ERR(map);
 628
 629         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 630                 err = -EPERM;
 631                 goto err_put;
 632         }
 633
 634         key = memdup_user(ukey, map->key_size);
 635         if (IS_ERR(key)) {
 636                 err = PTR_ERR(key);
 637                 goto err_put;
 638         }
 639
 640         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 641             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 642             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 643                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 644         else
 645                 value_size = map->value_size;
 646
 647         err = -ENOMEM;
 648         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 649         if (!value)
 650                 goto free_key;
 651
 652         err = -EFAULT;
 653         if (copy_from_user(value, uvalue, value_size) != 0)
 654                 goto free_value;
 655
 656         /* Need to create a kthread, thus must support schedule */
 657         if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
 658                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 659                 goto out;
 660         }
 661
 662         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 663          * inside bpf map update or delete otherwise deadlocks are possible
 664          */
 665         preempt_disable();
 666         __this_cpu_inc(bpf_prog_active);
 667         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 668             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 669                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 670         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 671                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 672         } else if (map->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY ||
 673                    map->map_type == BPF_MAP_TYPE_PROG_ARRAY ||
 674                    map->map_type == BPF_MAP_TYPE_CGROUP_ARRAY ||
 675                    map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS) {
 676                 rcu_read_lock();
 677                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 678                                                    attr->flags);
 679                 rcu_read_unlock();
 680         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 681                 rcu_read_lock();
 682                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 683                                                   attr->flags);
 684                 rcu_read_unlock();
 685         } else {
 686                 rcu_read_lock();
 687                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 688                 rcu_read_unlock();
 689         }
 690         __this_cpu_dec(bpf_prog_active);
 691         preempt_enable();
 692 out:
 693         if (!err)
 694                 trace_bpf_map_update_elem(map, ufd, key, value);
 695 free_value:
 696         kfree(value);
 697 free_key:
 698         kfree(key);
 699 err_put:
 700         fdput(f);
 701         return err;
 702 }
 703
 704 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 705
 706 static int map_delete_elem(union bpf_attr *attr)
 707 {
 708         void __user *ukey = u64_to_user_ptr(attr->key);
 709         int ufd = attr->map_fd;
 710         struct bpf_map *map;
 711         struct fd f;
 712         void *key;
 713         int err;
 714
 715         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 716                 return -EINVAL;
 717
 718         f = fdget(ufd);
 719         map = __bpf_map_get(f);
 720         if (IS_ERR(map))
 721                 return PTR_ERR(map);
 722
 723         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 724                 err = -EPERM;
 725                 goto err_put;
 726         }
 727
 728         key = memdup_user(ukey, map->key_size);
 729         if (IS_ERR(key)) {
 730                 err = PTR_ERR(key);
 731                 goto err_put;
 732         }
 733
 734         preempt_disable();
 735         __this_cpu_inc(bpf_prog_active);
 736         rcu_read_lock();
 737         err = map->ops->map_delete_elem(map, key);
 738         rcu_read_unlock();
 739         __this_cpu_dec(bpf_prog_active);
 740         preempt_enable();
 741
 742         if (!err)
 743                 trace_bpf_map_delete_elem(map, ufd, key);
 744         kfree(key);
 745 err_put:
 746         fdput(f);
 747         return err;
 748 }
 749
 750 /* last field in 'union bpf_attr' used by this command */
 751 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 752
 753 static int map_get_next_key(union bpf_attr *attr)
 754 {
 755         void __user *ukey = u64_to_user_ptr(attr->key);
 756         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 757         int ufd = attr->map_fd;
 758         struct bpf_map *map;
 759         void *key, *next_key;
 760         struct fd f;
 761         int err;
 762
 763         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 764                 return -EINVAL;
 765
 766         f = fdget(ufd);
 767         map = __bpf_map_get(f);
 768         if (IS_ERR(map))
 769                 return PTR_ERR(map);
 770
 771         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 772                 err = -EPERM;
 773                 goto err_put;
 774         }
 775
 776         if (ukey) {
 777                 key = memdup_user(ukey, map->key_size);
 778                 if (IS_ERR(key)) {
 779                         err = PTR_ERR(key);
 780                         goto err_put;
 781                 }
 782         } else {
 783                 key = NULL;
 784         }
 785
 786         err = -ENOMEM;
 787         next_key = kmalloc(map->key_size, GFP_USER);
 788         if (!next_key)
 789                 goto free_key;
 790
 791         rcu_read_lock();
 792         err = map->ops->map_get_next_key(map, key, next_key);
 793         rcu_read_unlock();
 794         if (err)
 795                 goto free_next_key;
 796
 797         err = -EFAULT;
 798         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 799                 goto free_next_key;
 800
 801         trace_bpf_map_next_key(map, ufd, key, next_key);
 802         err = 0;
 803
 804 free_next_key:
 805         kfree(next_key);
 806 free_key:
 807         kfree(key);
 808 err_put:
 809         fdput(f);
 810         return err;
 811 }
 812
 813 static const struct bpf_prog_ops * const bpf_prog_types[] = {
 814 #define BPF_PROG_TYPE(_id, _name) \
 815         [_id] = & _name ## _prog_ops,
 816 #define BPF_MAP_TYPE(_id, _ops)
 817 #include <linux/bpf_types.h>
 818 #undef BPF_PROG_TYPE
 819 #undef BPF_MAP_TYPE
 820 };
 821
 822 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 823 {
 824         if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
 825                 return -EINVAL;
 826
 827         if (!bpf_prog_is_dev_bound(prog->aux))
 828                 prog->aux->ops = bpf_prog_types[type];
 829         else
 830                 prog->aux->ops = &bpf_offload_prog_ops;
 831         prog->type = type;
 832         return 0;
 833 }
 834
 835 /* drop refcnt on maps used by eBPF program and free auxilary data */
 836 static void free_used_maps(struct bpf_prog_aux *aux)
 837 {
 838         int i;
 839
 840         for (i = 0; i < aux->used_map_cnt; i++)
 841                 bpf_map_put(aux->used_maps[i]);
 842
 843         kfree(aux->used_maps);
 844 }
 845
 846 int __bpf_prog_charge(struct user_struct *user, u32 pages)
 847 {
 848         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 849         unsigned long user_bufs;
 850
 851         if (user) {
 852                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
 853                 if (user_bufs > memlock_limit) {
 854                         atomic_long_sub(pages, &user->locked_vm);
 855                         return -EPERM;
 856                 }
 857         }
 858
 859         return 0;
 860 }
 861
 862 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
 863 {
 864         if (user)
 865                 atomic_long_sub(pages, &user->locked_vm);
 866 }
 867
 868 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
 869 {
 870         struct user_struct *user = get_current_user();
 871         int ret;
 872
 873         ret = __bpf_prog_charge(user, prog->pages);
 874         if (ret) {
 875                 free_uid(user);
 876                 return ret;
 877         }
 878
 879         prog->aux->user = user;
 880         return 0;
 881 }
 882
 883 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
 884 {
 885         struct user_struct *user = prog->aux->user;
 886
 887         __bpf_prog_uncharge(user, prog->pages);
 888         free_uid(user);
 889 }
 890
 891 static int bpf_prog_alloc_id(struct bpf_prog *prog)
 892 {
 893         int id;
 894
 895         spin_lock_bh(&prog_idr_lock);
 896         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
 897         if (id > 0)
 898                 prog->aux->id = id;
 899         spin_unlock_bh(&prog_idr_lock);
 900
 901         /* id is in [1, INT_MAX) */
 902         if (WARN_ON_ONCE(!id))
 903                 return -ENOSPC;
 904
 905         return id > 0 ? 0 : id;
 906 }
 907
 908 static void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
 909 {
 910         /* cBPF to eBPF migrations are currently not in the idr store. */
 911         if (!prog->aux->id)
 912                 return;
 913
 914         if (do_idr_lock)
 915                 spin_lock_bh(&prog_idr_lock);
 916         else
 917                 __acquire(&prog_idr_lock);
 918
 919         idr_remove(&prog_idr, prog->aux->id);
 920
 921         if (do_idr_lock)
 922                 spin_unlock_bh(&prog_idr_lock);
 923         else
 924                 __release(&prog_idr_lock);
 925 }
 926
 927 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
 928 {
 929         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 930
 931         free_used_maps(aux);
 932         bpf_prog_uncharge_memlock(aux->prog);
 933         security_bpf_prog_free(aux);
 934         bpf_prog_free(aux->prog);
 935 }
 936
 937 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
 938 {
 939         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 940                 trace_bpf_prog_put_rcu(prog);
 941                 /* bpf_prog_free_id() must be called first */
 942                 bpf_prog_free_id(prog, do_idr_lock);
 943                 bpf_prog_kallsyms_del(prog);
 944                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
 945         }
 946 }
 947
 948 void bpf_prog_put(struct bpf_prog *prog)
 949 {
 950         __bpf_prog_put(prog, true);
 951 }
 952 EXPORT_SYMBOL_GPL(bpf_prog_put);
 953
 954 static int bpf_prog_release(struct inode *inode, struct file *filp)
 955 {
 956         struct bpf_prog *prog = filp->private_data;
 957
 958         bpf_prog_put(prog);
 959         return 0;
 960 }
 961
 962 #ifdef CONFIG_PROC_FS
 963 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
 964 {
 965         const struct bpf_prog *prog = filp->private_data;
 966         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
 967
 968         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
 969         seq_printf(m,
 970                    "prog_type:\t%u\n"
 971                    "prog_jited:\t%u\n"
 972                    "prog_tag:\t%s\n"
 973                    "memlock:\t%llu\n",
 974                    prog->type,
 975                    prog->jited,
 976                    prog_tag,
 977                    prog->pages * 1ULL << PAGE_SHIFT);
 978 }
 979 #endif
 980
 981 const struct file_operations bpf_prog_fops = {
 982 #ifdef CONFIG_PROC_FS
 983         .show_fdinfo    = bpf_prog_show_fdinfo,
 984 #endif
 985         .release        = bpf_prog_release,
 986         .read           = bpf_dummy_read,
 987         .write          = bpf_dummy_write,
 988 };
 989
 990 int bpf_prog_new_fd(struct bpf_prog *prog)
 991 {
 992         int ret;
 993
 994         ret = security_bpf_prog(prog);
 995         if (ret < 0)
 996                 return ret;
 997
 998         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
 999                                 O_RDWR | O_CLOEXEC);
1000 }
1001
1002 static struct bpf_prog *____bpf_prog_get(struct fd f)
1003 {
1004         if (!f.file)
1005                 return ERR_PTR(-EBADF);
1006         if (f.file->f_op != &bpf_prog_fops) {
1007                 fdput(f);
1008                 return ERR_PTR(-EINVAL);
1009         }
1010
1011         return f.file->private_data;
1012 }
1013
1014 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1015 {
1016         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1017                 atomic_sub(i, &prog->aux->refcnt);
1018                 return ERR_PTR(-EBUSY);
1019         }
1020         return prog;
1021 }
1022 EXPORT_SYMBOL_GPL(bpf_prog_add);
1023
1024 void bpf_prog_sub(struct bpf_prog *prog, int i)
1025 {
1026         /* Only to be used for undoing previous bpf_prog_add() in some
1027          * error path. We still know that another entity in our call
1028          * path holds a reference to the program, thus atomic_sub() can
1029          * be safely used in such cases!
1030          */
1031         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1032 }
1033 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1034
1035 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1036 {
1037         return bpf_prog_add(prog, 1);
1038 }
1039 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1040
1041 /* prog_idr_lock should have been held */
1042 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1043 {
1044         int refold;
1045
1046         refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1047
1048         if (refold >= BPF_MAX_REFCNT) {
1049                 __bpf_prog_put(prog, false);
1050                 return ERR_PTR(-EBUSY);
1051         }
1052
1053         if (!refold)
1054                 return ERR_PTR(-ENOENT);
1055
1056         return prog;
1057 }
1058 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1059
1060 static bool bpf_prog_get_ok(struct bpf_prog *prog,
1061                             enum bpf_prog_type *attach_type, bool attach_drv)
1062 {
1063         /* not an attachment, just a refcount inc, always allow */
1064         if (!attach_type)
1065                 return true;
1066
1067         if (prog->type != *attach_type)
1068                 return false;
1069         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1070                 return false;
1071
1072         return true;
1073 }
1074
1075 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1076                                        bool attach_drv)
1077 {
1078         struct fd f = fdget(ufd);
1079         struct bpf_prog *prog;
1080
1081         prog = ____bpf_prog_get(f);
1082         if (IS_ERR(prog))
1083                 return prog;
1084         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1085                 prog = ERR_PTR(-EINVAL);
1086                 goto out;
1087         }
1088
1089         prog = bpf_prog_inc(prog);
1090 out:
1091         fdput(f);
1092         return prog;
1093 }
1094
1095 struct bpf_prog *bpf_prog_get(u32 ufd)
1096 {
1097         return __bpf_prog_get(ufd, NULL, false);
1098 }
1099
1100 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1101                                        bool attach_drv)
1102 {
1103         struct bpf_prog *prog = __bpf_prog_get(ufd, &type, attach_drv);
1104
1105         if (!IS_ERR(prog))
1106                 trace_bpf_prog_get_type(prog);
1107         return prog;
1108 }
1109 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1110
1111 /* last field in 'union bpf_attr' used by this command */
1112 #define BPF_PROG_LOAD_LAST_FIELD prog_ifindex
1113
1114 static int bpf_prog_load(union bpf_attr *attr)
1115 {
1116         enum bpf_prog_type type = attr->prog_type;
1117         struct bpf_prog *prog;
1118         int err;
1119         char license[128];
1120         bool is_gpl;
1121
1122         if (CHECK_ATTR(BPF_PROG_LOAD))
1123                 return -EINVAL;
1124
1125         if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1126                 return -EINVAL;
1127
1128         /* copy eBPF program license from user space */
1129         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1130                               sizeof(license) - 1) < 0)
1131                 return -EFAULT;
1132         license[sizeof(license) - 1] = 0;
1133
1134         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1135         is_gpl = license_is_gpl_compatible(license);
1136
1137         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1138                 return -E2BIG;
1139
1140         if (type == BPF_PROG_TYPE_KPROBE &&
1141             attr->kern_version != LINUX_VERSION_CODE)
1142                 return -EINVAL;
1143
1144         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1145             type != BPF_PROG_TYPE_CGROUP_SKB &&
1146             !capable(CAP_SYS_ADMIN))
1147                 return -EPERM;
1148
1149         /* plain bpf_prog allocation */
1150         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1151         if (!prog)
1152                 return -ENOMEM;
1153
1154         err = security_bpf_prog_alloc(prog->aux);
1155         if (err)
1156                 goto free_prog_nouncharge;
1157
1158         err = bpf_prog_charge_memlock(prog);
1159         if (err)
1160                 goto free_prog_sec;
1161
1162         prog->len = attr->insn_cnt;
1163
1164         err = -EFAULT;
1165         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1166                            bpf_prog_insn_size(prog)) != 0)
1167                 goto free_prog;
1168
1169         prog->orig_prog = NULL;
1170         prog->jited = 0;
1171
1172         atomic_set(&prog->aux->refcnt, 1);
1173         prog->gpl_compatible = is_gpl ? 1 : 0;
1174
1175         if (attr->prog_ifindex) {
1176                 err = bpf_prog_offload_init(prog, attr);
1177                 if (err)
1178                         goto free_prog;
1179         }
1180
1181         /* find program type: socket_filter vs tracing_filter */
1182         err = find_prog_type(type, prog);
1183         if (err < 0)
1184                 goto free_prog;
1185
1186         prog->aux->load_time = ktime_get_boot_ns();
1187         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1188         if (err)
1189                 goto free_prog;
1190
1191         /* run eBPF verifier */
1192         err = bpf_check(&prog, attr);
1193         if (err < 0)
1194                 goto free_used_maps;
1195
1196         /* eBPF program is ready to be JITed */
1197         prog = bpf_prog_select_runtime(prog, &err);
1198         if (err < 0)
1199                 goto free_used_maps;
1200
1201         err = bpf_prog_alloc_id(prog);
1202         if (err)
1203                 goto free_used_maps;
1204
1205         err = bpf_prog_new_fd(prog);
1206         if (err < 0) {
1207                 /* failed to allocate fd.
1208                  * bpf_prog_put() is needed because the above
1209                  * bpf_prog_alloc_id() has published the prog
1210                  * to the userspace and the userspace may
1211                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1212                  */
1213                 bpf_prog_put(prog);
1214                 return err;
1215         }
1216
1217         bpf_prog_kallsyms_add(prog);
1218         trace_bpf_prog_load(prog, err);
1219         return err;
1220
1221 free_used_maps:
1222         free_used_maps(prog->aux);
1223 free_prog:
1224         bpf_prog_uncharge_memlock(prog);
1225 free_prog_sec:
1226         security_bpf_prog_free(prog->aux);
1227 free_prog_nouncharge:
1228         bpf_prog_free(prog);
1229         return err;
1230 }
1231
1232 #define BPF_OBJ_LAST_FIELD file_flags
1233
1234 static int bpf_obj_pin(const union bpf_attr *attr)
1235 {
1236         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1237                 return -EINVAL;
1238
1239         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1240 }
1241
1242 static int bpf_obj_get(const union bpf_attr *attr)
1243 {
1244         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1245             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1246                 return -EINVAL;
1247
1248         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1249                                 attr->file_flags);
1250 }
1251
1252 #ifdef CONFIG_CGROUP_BPF
1253
1254 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1255
1256 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1257 {
1258         struct bpf_prog *prog = NULL;
1259         int ufd = attr->target_fd;
1260         struct bpf_map *map;
1261         struct fd f;
1262         int err;
1263
1264         f = fdget(ufd);
1265         map = __bpf_map_get(f);
1266         if (IS_ERR(map))
1267                 return PTR_ERR(map);
1268
1269         if (attach) {
1270                 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1271                                          BPF_PROG_TYPE_SK_SKB);
1272                 if (IS_ERR(prog)) {
1273                         fdput(f);
1274                         return PTR_ERR(prog);
1275                 }
1276         }
1277
1278         err = sock_map_prog(map, prog, attr->attach_type);
1279         if (err) {
1280                 fdput(f);
1281                 if (prog)
1282                         bpf_prog_put(prog);
1283                 return err;
1284         }
1285
1286         fdput(f);
1287         return 0;
1288 }
1289
1290 #define BPF_F_ATTACH_MASK \
1291         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1292
1293 static int bpf_prog_attach(const union bpf_attr *attr)
1294 {
1295         enum bpf_prog_type ptype;
1296         struct bpf_prog *prog;
1297         struct cgroup *cgrp;
1298         int ret;
1299
1300         if (!capable(CAP_NET_ADMIN))
1301                 return -EPERM;
1302
1303         if (CHECK_ATTR(BPF_PROG_ATTACH))
1304                 return -EINVAL;
1305
1306         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1307                 return -EINVAL;
1308
1309         switch (attr->attach_type) {
1310         case BPF_CGROUP_INET_INGRESS:
1311         case BPF_CGROUP_INET_EGRESS:
1312                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1313                 break;
1314         case BPF_CGROUP_INET_SOCK_CREATE:
1315                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1316                 break;
1317         case BPF_CGROUP_SOCK_OPS:
1318                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1319                 break;
1320         case BPF_CGROUP_DEVICE:
1321                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1322                 break;
1323         case BPF_SK_SKB_STREAM_PARSER:
1324         case BPF_SK_SKB_STREAM_VERDICT:
1325                 return sockmap_get_from_fd(attr, true);
1326         default:
1327                 return -EINVAL;
1328         }
1329
1330         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1331         if (IS_ERR(prog))
1332                 return PTR_ERR(prog);
1333
1334         cgrp = cgroup_get_from_fd(attr->target_fd);
1335         if (IS_ERR(cgrp)) {
1336                 bpf_prog_put(prog);
1337                 return PTR_ERR(cgrp);
1338         }
1339
1340         ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1341                                 attr->attach_flags);
1342         if (ret)
1343                 bpf_prog_put(prog);
1344         cgroup_put(cgrp);
1345
1346         return ret;
1347 }
1348
1349 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1350
1351 static int bpf_prog_detach(const union bpf_attr *attr)
1352 {
1353         enum bpf_prog_type ptype;
1354         struct bpf_prog *prog;
1355         struct cgroup *cgrp;
1356         int ret;
1357
1358         if (!capable(CAP_NET_ADMIN))
1359                 return -EPERM;
1360
1361         if (CHECK_ATTR(BPF_PROG_DETACH))
1362                 return -EINVAL;
1363
1364         switch (attr->attach_type) {
1365         case BPF_CGROUP_INET_INGRESS:
1366         case BPF_CGROUP_INET_EGRESS:
1367                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1368                 break;
1369         case BPF_CGROUP_INET_SOCK_CREATE:
1370                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1371                 break;
1372         case BPF_CGROUP_SOCK_OPS:
1373                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1374                 break;
1375         case BPF_CGROUP_DEVICE:
1376                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1377                 break;
1378         case BPF_SK_SKB_STREAM_PARSER:
1379         case BPF_SK_SKB_STREAM_VERDICT:
1380                 return sockmap_get_from_fd(attr, false);
1381         default:
1382                 return -EINVAL;
1383         }
1384
1385         cgrp = cgroup_get_from_fd(attr->target_fd);
1386         if (IS_ERR(cgrp))
1387                 return PTR_ERR(cgrp);
1388
1389         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1390         if (IS_ERR(prog))
1391                 prog = NULL;
1392
1393         ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1394         if (prog)
1395                 bpf_prog_put(prog);
1396         cgroup_put(cgrp);
1397         return ret;
1398 }
1399
1400 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1401
1402 static int bpf_prog_query(const union bpf_attr *attr,
1403                           union bpf_attr __user *uattr)
1404 {
1405         struct cgroup *cgrp;
1406         int ret;
1407
1408         if (!capable(CAP_NET_ADMIN))
1409                 return -EPERM;
1410         if (CHECK_ATTR(BPF_PROG_QUERY))
1411                 return -EINVAL;
1412         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1413                 return -EINVAL;
1414
1415         switch (attr->query.attach_type) {
1416         case BPF_CGROUP_INET_INGRESS:
1417         case BPF_CGROUP_INET_EGRESS:
1418         case BPF_CGROUP_INET_SOCK_CREATE:
1419         case BPF_CGROUP_SOCK_OPS:
1420         case BPF_CGROUP_DEVICE:
1421                 break;
1422         default:
1423                 return -EINVAL;
1424         }
1425         cgrp = cgroup_get_from_fd(attr->query.target_fd);
1426         if (IS_ERR(cgrp))
1427                 return PTR_ERR(cgrp);
1428         ret = cgroup_bpf_query(cgrp, attr, uattr);
1429         cgroup_put(cgrp);
1430         return ret;
1431 }
1432 #endif /* CONFIG_CGROUP_BPF */
1433
1434 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1435
1436 static int bpf_prog_test_run(const union bpf_attr *attr,
1437                              union bpf_attr __user *uattr)
1438 {
1439         struct bpf_prog *prog;
1440         int ret = -ENOTSUPP;
1441
1442         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1443                 return -EINVAL;
1444
1445         prog = bpf_prog_get(attr->test.prog_fd);
1446         if (IS_ERR(prog))
1447                 return PTR_ERR(prog);
1448
1449         if (prog->aux->ops->test_run)
1450                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1451
1452         bpf_prog_put(prog);
1453         return ret;
1454 }
1455
1456 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1457
1458 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1459                                union bpf_attr __user *uattr,
1460                                struct idr *idr,
1461                                spinlock_t *lock)
1462 {
1463         u32 next_id = attr->start_id;
1464         int err = 0;
1465
1466         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1467                 return -EINVAL;
1468
1469         if (!capable(CAP_SYS_ADMIN))
1470                 return -EPERM;
1471
1472         next_id++;
1473         spin_lock_bh(lock);
1474         if (!idr_get_next(idr, &next_id))
1475                 err = -ENOENT;
1476         spin_unlock_bh(lock);
1477
1478         if (!err)
1479                 err = put_user(next_id, &uattr->next_id);
1480
1481         return err;
1482 }
1483
1484 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1485
1486 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1487 {
1488         struct bpf_prog *prog;
1489         u32 id = attr->prog_id;
1490         int fd;
1491
1492         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1493                 return -EINVAL;
1494
1495         if (!capable(CAP_SYS_ADMIN))
1496                 return -EPERM;
1497
1498         spin_lock_bh(&prog_idr_lock);
1499         prog = idr_find(&prog_idr, id);
1500         if (prog)
1501                 prog = bpf_prog_inc_not_zero(prog);
1502         else
1503                 prog = ERR_PTR(-ENOENT);
1504         spin_unlock_bh(&prog_idr_lock);
1505
1506         if (IS_ERR(prog))
1507                 return PTR_ERR(prog);
1508
1509         fd = bpf_prog_new_fd(prog);
1510         if (fd < 0)
1511                 bpf_prog_put(prog);
1512
1513         return fd;
1514 }
1515
1516 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1517
1518 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1519 {
1520         struct bpf_map *map;
1521         u32 id = attr->map_id;
1522         int f_flags;
1523         int fd;
1524
1525         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1526             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1527                 return -EINVAL;
1528
1529         if (!capable(CAP_SYS_ADMIN))
1530                 return -EPERM;
1531
1532         f_flags = bpf_get_file_flag(attr->open_flags);
1533         if (f_flags < 0)
1534                 return f_flags;
1535
1536         spin_lock_bh(&map_idr_lock);
1537         map = idr_find(&map_idr, id);
1538         if (map)
1539                 map = bpf_map_inc_not_zero(map, true);
1540         else
1541                 map = ERR_PTR(-ENOENT);
1542         spin_unlock_bh(&map_idr_lock);
1543
1544         if (IS_ERR(map))
1545                 return PTR_ERR(map);
1546
1547         fd = bpf_map_new_fd(map, f_flags);
1548         if (fd < 0)
1549                 bpf_map_put(map);
1550
1551         return fd;
1552 }
1553
1554 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1555                                    const union bpf_attr *attr,
1556                                    union bpf_attr __user *uattr)
1557 {
1558         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1559         struct bpf_prog_info info = {};
1560         u32 info_len = attr->info.info_len;
1561         char __user *uinsns;
1562         u32 ulen;
1563         int err;
1564
1565         err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1566         if (err)
1567                 return err;
1568         info_len = min_t(u32, sizeof(info), info_len);
1569
1570         if (copy_from_user(&info, uinfo, info_len))
1571                 return -EFAULT;
1572
1573         info.type = prog->type;
1574         info.id = prog->aux->id;
1575         info.load_time = prog->aux->load_time;
1576         info.created_by_uid = from_kuid_munged(current_user_ns(),
1577                                                prog->aux->user->uid);
1578
1579         memcpy(info.tag, prog->tag, sizeof(prog->tag));
1580         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1581
1582         ulen = info.nr_map_ids;
1583         info.nr_map_ids = prog->aux->used_map_cnt;
1584         ulen = min_t(u32, info.nr_map_ids, ulen);
1585         if (ulen) {
1586                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1587                 u32 i;
1588
1589                 for (i = 0; i < ulen; i++)
1590                         if (put_user(prog->aux->used_maps[i]->id,
1591                                      &user_map_ids[i]))
1592                                 return -EFAULT;
1593         }
1594
1595         if (!capable(CAP_SYS_ADMIN)) {
1596                 info.jited_prog_len = 0;
1597                 info.xlated_prog_len = 0;
1598                 goto done;
1599         }
1600
1601         ulen = info.jited_prog_len;
1602         info.jited_prog_len = prog->jited_len;
1603         if (info.jited_prog_len && ulen) {
1604                 uinsns = u64_to_user_ptr(info.jited_prog_insns);
1605                 ulen = min_t(u32, info.jited_prog_len, ulen);
1606                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
1607                         return -EFAULT;
1608         }
1609
1610         ulen = info.xlated_prog_len;
1611         info.xlated_prog_len = bpf_prog_insn_size(prog);
1612         if (info.xlated_prog_len && ulen) {
1613                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1614                 ulen = min_t(u32, info.xlated_prog_len, ulen);
1615                 if (copy_to_user(uinsns, prog->insnsi, ulen))
1616                         return -EFAULT;
1617         }
1618
1619 done:
1620         if (copy_to_user(uinfo, &info, info_len) ||
1621             put_user(info_len, &uattr->info.info_len))
1622                 return -EFAULT;
1623
1624         return 0;
1625 }
1626
1627 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1628                                   const union bpf_attr *attr,
1629                                   union bpf_attr __user *uattr)
1630 {
1631         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1632         struct bpf_map_info info = {};
1633         u32 info_len = attr->info.info_len;
1634         int err;
1635
1636         err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1637         if (err)
1638                 return err;
1639         info_len = min_t(u32, sizeof(info), info_len);
1640
1641         info.type = map->map_type;
1642         info.id = map->id;
1643         info.key_size = map->key_size;
1644         info.value_size = map->value_size;
1645         info.max_entries = map->max_entries;
1646         info.map_flags = map->map_flags;
1647         memcpy(info.name, map->name, sizeof(map->name));
1648
1649         if (copy_to_user(uinfo, &info, info_len) ||
1650             put_user(info_len, &uattr->info.info_len))
1651                 return -EFAULT;
1652
1653         return 0;
1654 }
1655
1656 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1657
1658 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1659                                   union bpf_attr __user *uattr)
1660 {
1661         int ufd = attr->info.bpf_fd;
1662         struct fd f;
1663         int err;
1664
1665         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1666                 return -EINVAL;
1667
1668         f = fdget(ufd);
1669         if (!f.file)
1670                 return -EBADFD;
1671
1672         if (f.file->f_op == &bpf_prog_fops)
1673                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1674                                               uattr);
1675         else if (f.file->f_op == &bpf_map_fops)
1676                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1677                                              uattr);
1678         else
1679                 err = -EINVAL;
1680
1681         fdput(f);
1682         return err;
1683 }
1684
1685 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1686 {
1687         union bpf_attr attr = {};
1688         int err;
1689
1690         if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1691                 return -EPERM;
1692
1693         err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1694         if (err)
1695                 return err;
1696         size = min_t(u32, size, sizeof(attr));
1697
1698         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1699         if (copy_from_user(&attr, uattr, size) != 0)
1700                 return -EFAULT;
1701
1702         err = security_bpf(cmd, &attr, size);
1703         if (err < 0)
1704                 return err;
1705
1706         switch (cmd) {
1707         case BPF_MAP_CREATE:
1708                 err = map_create(&attr);
1709                 break;
1710         case BPF_MAP_LOOKUP_ELEM:
1711                 err = map_lookup_elem(&attr);
1712                 break;
1713         case BPF_MAP_UPDATE_ELEM:
1714                 err = map_update_elem(&attr);
1715                 break;
1716         case BPF_MAP_DELETE_ELEM:
1717                 err = map_delete_elem(&attr);
1718                 break;
1719         case BPF_MAP_GET_NEXT_KEY:
1720                 err = map_get_next_key(&attr);
1721                 break;
1722         case BPF_PROG_LOAD:
1723                 err = bpf_prog_load(&attr);
1724                 break;
1725         case BPF_OBJ_PIN:
1726                 err = bpf_obj_pin(&attr);
1727                 break;
1728         case BPF_OBJ_GET:
1729                 err = bpf_obj_get(&attr);
1730                 break;
1731 #ifdef CONFIG_CGROUP_BPF
1732         case BPF_PROG_ATTACH:
1733                 err = bpf_prog_attach(&attr);
1734                 break;
1735         case BPF_PROG_DETACH:
1736                 err = bpf_prog_detach(&attr);
1737                 break;
1738         case BPF_PROG_QUERY:
1739                 err = bpf_prog_query(&attr, uattr);
1740                 break;
1741 #endif
1742         case BPF_PROG_TEST_RUN:
1743                 err = bpf_prog_test_run(&attr, uattr);
1744                 break;
1745         case BPF_PROG_GET_NEXT_ID:
1746                 err = bpf_obj_get_next_id(&attr, uattr,
1747                                           &prog_idr, &prog_idr_lock);
1748                 break;
1749         case BPF_MAP_GET_NEXT_ID:
1750                 err = bpf_obj_get_next_id(&attr, uattr,
1751                                           &map_idr, &map_idr_lock);
1752                 break;
1753         case BPF_PROG_GET_FD_BY_ID:
1754                 err = bpf_prog_get_fd_by_id(&attr);
1755                 break;
1756         case BPF_MAP_GET_FD_BY_ID:
1757                 err = bpf_map_get_fd_by_id(&attr);
1758                 break;
1759         case BPF_OBJ_GET_INFO_BY_FD:
1760                 err = bpf_obj_get_info_by_fd(&attr, uattr);
1761                 break;
1762         default:
1763                 err = -EINVAL;
1764                 break;
1765         }
1766
1767         return err;
1768 }