git.rot13.org Git - linux/blob - 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/bpf_lirc.h>
  15 #include <linux/btf.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/slab.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/mmzone.h>
  21 #include <linux/anon_inodes.h>
  22 #include <linux/fdtable.h>
  23 #include <linux/file.h>
  24 #include <linux/fs.h>
  25 #include <linux/license.h>
  26 #include <linux/filter.h>
  27 #include <linux/version.h>
  28 #include <linux/kernel.h>
  29 #include <linux/idr.h>
  30 #include <linux/cred.h>
  31 #include <linux/timekeeping.h>
  32 #include <linux/ctype.h>
  33 #include <linux/nospec.h>
  34
  35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  36                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  37                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  38                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  41
  42 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  43
  44 DEFINE_PER_CPU(int, bpf_prog_active);
  45 static DEFINE_IDR(prog_idr);
  46 static DEFINE_SPINLOCK(prog_idr_lock);
  47 static DEFINE_IDR(map_idr);
  48 static DEFINE_SPINLOCK(map_idr_lock);
  49
  50 int sysctl_unprivileged_bpf_disabled __read_mostly;
  51
  52 static const struct bpf_map_ops * const bpf_map_types[] = {
  53 #define BPF_PROG_TYPE(_id, _ops)
  54 #define BPF_MAP_TYPE(_id, _ops) \
  55         [_id] = &_ops,
  56 #include <linux/bpf_types.h>
  57 #undef BPF_PROG_TYPE
  58 #undef BPF_MAP_TYPE
  59 };
  60
  61 /*
  62  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  63  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  64  * we don't know about yet.
  65  *
  66  * There is a ToCToU between this function call and the following
  67  * copy_from_user() call. However, this is not a concern since this function is
  68  * meant to be a future-proofing of bits.
  69  */
  70 int bpf_check_uarg_tail_zero(void __user *uaddr,
  71                              size_t expected_size,
  72                              size_t actual_size)
  73 {
  74         unsigned char __user *addr;
  75         unsigned char __user *end;
  76         unsigned char val;
  77         int err;
  78
  79         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  80                 return -E2BIG;
  81
  82         if (unlikely(!access_ok(uaddr, actual_size)))
  83                 return -EFAULT;
  84
  85         if (actual_size <= expected_size)
  86                 return 0;
  87
  88         addr = uaddr + expected_size;
  89         end  = uaddr + actual_size;
  90
  91         for (; addr < end; addr++) {
  92                 err = get_user(val, addr);
  93                 if (err)
  94                         return err;
  95                 if (val)
  96                         return -E2BIG;
  97         }
  98
  99         return 0;
 100 }
 101
 102 const struct bpf_map_ops bpf_map_offload_ops = {
 103         .map_alloc = bpf_map_offload_map_alloc,
 104         .map_free = bpf_map_offload_map_free,
 105         .map_check_btf = map_check_no_btf,
 106 };
 107
 108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 109 {
 110         const struct bpf_map_ops *ops;
 111         u32 type = attr->map_type;
 112         struct bpf_map *map;
 113         int err;
 114
 115         if (type >= ARRAY_SIZE(bpf_map_types))
 116                 return ERR_PTR(-EINVAL);
 117         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 118         ops = bpf_map_types[type];
 119         if (!ops)
 120                 return ERR_PTR(-EINVAL);
 121
 122         if (ops->map_alloc_check) {
 123                 err = ops->map_alloc_check(attr);
 124                 if (err)
 125                         return ERR_PTR(err);
 126         }
 127         if (attr->map_ifindex)
 128                 ops = &bpf_map_offload_ops;
 129         map = ops->map_alloc(attr);
 130         if (IS_ERR(map))
 131                 return map;
 132         map->ops = ops;
 133         map->map_type = type;
 134         return map;
 135 }
 136
 137 void *bpf_map_area_alloc(size_t size, int numa_node)
 138 {
 139         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
 140          * trigger under memory pressure as we really just want to
 141          * fail instead.
 142          */
 143         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
 144         void *area;
 145
 146         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 147                 area = kmalloc_node(size, GFP_USER | flags, numa_node);
 148                 if (area != NULL)
 149                         return area;
 150         }
 151
 152         return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
 153                                            __builtin_return_address(0));
 154 }
 155
 156 void bpf_map_area_free(void *area)
 157 {
 158         kvfree(area);
 159 }
 160
 161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 162 {
 163         map->map_type = attr->map_type;
 164         map->key_size = attr->key_size;
 165         map->value_size = attr->value_size;
 166         map->max_entries = attr->max_entries;
 167         map->map_flags = attr->map_flags;
 168         map->numa_node = bpf_map_attr_numa_node(attr);
 169 }
 170
 171 int bpf_map_precharge_memlock(u32 pages)
 172 {
 173         struct user_struct *user = get_current_user();
 174         unsigned long memlock_limit, cur;
 175
 176         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 177         cur = atomic_long_read(&user->locked_vm);
 178         free_uid(user);
 179         if (cur + pages > memlock_limit)
 180                 return -EPERM;
 181         return 0;
 182 }
 183
 184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 185 {
 186         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 187
 188         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 189                 atomic_long_sub(pages, &user->locked_vm);
 190                 return -EPERM;
 191         }
 192         return 0;
 193 }
 194
 195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 196 {
 197         atomic_long_sub(pages, &user->locked_vm);
 198 }
 199
 200 static int bpf_map_init_memlock(struct bpf_map *map)
 201 {
 202         struct user_struct *user = get_current_user();
 203         int ret;
 204
 205         ret = bpf_charge_memlock(user, map->pages);
 206         if (ret) {
 207                 free_uid(user);
 208                 return ret;
 209         }
 210         map->user = user;
 211         return ret;
 212 }
 213
 214 static void bpf_map_release_memlock(struct bpf_map *map)
 215 {
 216         struct user_struct *user = map->user;
 217         bpf_uncharge_memlock(user, map->pages);
 218         free_uid(user);
 219 }
 220
 221 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 222 {
 223         int ret;
 224
 225         ret = bpf_charge_memlock(map->user, pages);
 226         if (ret)
 227                 return ret;
 228         map->pages += pages;
 229         return ret;
 230 }
 231
 232 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 233 {
 234         bpf_uncharge_memlock(map->user, pages);
 235         map->pages -= pages;
 236 }
 237
 238 static int bpf_map_alloc_id(struct bpf_map *map)
 239 {
 240         int id;
 241
 242         idr_preload(GFP_KERNEL);
 243         spin_lock_bh(&map_idr_lock);
 244         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 245         if (id > 0)
 246                 map->id = id;
 247         spin_unlock_bh(&map_idr_lock);
 248         idr_preload_end();
 249
 250         if (WARN_ON_ONCE(!id))
 251                 return -ENOSPC;
 252
 253         return id > 0 ? 0 : id;
 254 }
 255
 256 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 257 {
 258         unsigned long flags;
 259
 260         /* Offloaded maps are removed from the IDR store when their device
 261          * disappears - even if someone holds an fd to them they are unusable,
 262          * the memory is gone, all ops will fail; they are simply waiting for
 263          * refcnt to drop to be freed.
 264          */
 265         if (!map->id)
 266                 return;
 267
 268         if (do_idr_lock)
 269                 spin_lock_irqsave(&map_idr_lock, flags);
 270         else
 271                 __acquire(&map_idr_lock);
 272
 273         idr_remove(&map_idr, map->id);
 274         map->id = 0;
 275
 276         if (do_idr_lock)
 277                 spin_unlock_irqrestore(&map_idr_lock, flags);
 278         else
 279                 __release(&map_idr_lock);
 280 }
 281
 282 /* called from workqueue */
 283 static void bpf_map_free_deferred(struct work_struct *work)
 284 {
 285         struct bpf_map *map = container_of(work, struct bpf_map, work);
 286
 287         bpf_map_release_memlock(map);
 288         security_bpf_map_free(map);
 289         /* implementation dependent freeing */
 290         map->ops->map_free(map);
 291 }
 292
 293 static void bpf_map_put_uref(struct bpf_map *map)
 294 {
 295         if (atomic_dec_and_test(&map->usercnt)) {
 296                 if (map->ops->map_release_uref)
 297                         map->ops->map_release_uref(map);
 298         }
 299 }
 300
 301 /* decrement map refcnt and schedule it for freeing via workqueue
 302  * (unrelying map implementation ops->map_free() might sleep)
 303  */
 304 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 305 {
 306         if (atomic_dec_and_test(&map->refcnt)) {
 307                 /* bpf_map_free_id() must be called first */
 308                 bpf_map_free_id(map, do_idr_lock);
 309                 btf_put(map->btf);
 310                 INIT_WORK(&map->work, bpf_map_free_deferred);
 311                 schedule_work(&map->work);
 312         }
 313 }
 314
 315 void bpf_map_put(struct bpf_map *map)
 316 {
 317         __bpf_map_put(map, true);
 318 }
 319 EXPORT_SYMBOL_GPL(bpf_map_put);
 320
 321 void bpf_map_put_with_uref(struct bpf_map *map)
 322 {
 323         bpf_map_put_uref(map);
 324         bpf_map_put(map);
 325 }
 326
 327 static int bpf_map_release(struct inode *inode, struct file *filp)
 328 {
 329         struct bpf_map *map = filp->private_data;
 330
 331         if (map->ops->map_release)
 332                 map->ops->map_release(map, filp);
 333
 334         bpf_map_put_with_uref(map);
 335         return 0;
 336 }
 337
 338 #ifdef CONFIG_PROC_FS
 339 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 340 {
 341         const struct bpf_map *map = filp->private_data;
 342         const struct bpf_array *array;
 343         u32 owner_prog_type = 0;
 344         u32 owner_jited = 0;
 345
 346         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 347                 array = container_of(map, struct bpf_array, map);
 348                 owner_prog_type = array->owner_prog_type;
 349                 owner_jited = array->owner_jited;
 350         }
 351
 352         seq_printf(m,
 353                    "map_type:\t%u\n"
 354                    "key_size:\t%u\n"
 355                    "value_size:\t%u\n"
 356                    "max_entries:\t%u\n"
 357                    "map_flags:\t%#x\n"
 358                    "memlock:\t%llu\n"
 359                    "map_id:\t%u\n",
 360                    map->map_type,
 361                    map->key_size,
 362                    map->value_size,
 363                    map->max_entries,
 364                    map->map_flags,
 365                    map->pages * 1ULL << PAGE_SHIFT,
 366                    map->id);
 367
 368         if (owner_prog_type) {
 369                 seq_printf(m, "owner_prog_type:\t%u\n",
 370                            owner_prog_type);
 371                 seq_printf(m, "owner_jited:\t%u\n",
 372                            owner_jited);
 373         }
 374 }
 375 #endif
 376
 377 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 378                               loff_t *ppos)
 379 {
 380         /* We need this handler such that alloc_file() enables
 381          * f_mode with FMODE_CAN_READ.
 382          */
 383         return -EINVAL;
 384 }
 385
 386 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 387                                size_t siz, loff_t *ppos)
 388 {
 389         /* We need this handler such that alloc_file() enables
 390          * f_mode with FMODE_CAN_WRITE.
 391          */
 392         return -EINVAL;
 393 }
 394
 395 const struct file_operations bpf_map_fops = {
 396 #ifdef CONFIG_PROC_FS
 397         .show_fdinfo    = bpf_map_show_fdinfo,
 398 #endif
 399         .release        = bpf_map_release,
 400         .read           = bpf_dummy_read,
 401         .write          = bpf_dummy_write,
 402 };
 403
 404 int bpf_map_new_fd(struct bpf_map *map, int flags)
 405 {
 406         int ret;
 407
 408         ret = security_bpf_map(map, OPEN_FMODE(flags));
 409         if (ret < 0)
 410                 return ret;
 411
 412         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 413                                 flags | O_CLOEXEC);
 414 }
 415
 416 int bpf_get_file_flag(int flags)
 417 {
 418         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 419                 return -EINVAL;
 420         if (flags & BPF_F_RDONLY)
 421                 return O_RDONLY;
 422         if (flags & BPF_F_WRONLY)
 423                 return O_WRONLY;
 424         return O_RDWR;
 425 }
 426
 427 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 428 #define CHECK_ATTR(CMD) \
 429         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 430                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 431                    sizeof(*attr) - \
 432                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 433                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 434
 435 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 436  * Return 0 on success and < 0 on error.
 437  */
 438 static int bpf_obj_name_cpy(char *dst, const char *src)
 439 {
 440         const char *end = src + BPF_OBJ_NAME_LEN;
 441
 442         memset(dst, 0, BPF_OBJ_NAME_LEN);
 443
 444         /* Copy all isalnum() and '_' char */
 445         while (src < end && *src) {
 446                 if (!isalnum(*src) && *src != '_')
 447                         return -EINVAL;
 448                 *dst++ = *src++;
 449         }
 450
 451         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 452         if (src == end)
 453                 return -EINVAL;
 454
 455         return 0;
 456 }
 457
 458 int map_check_no_btf(const struct bpf_map *map,
 459                      const struct btf *btf,
 460                      const struct btf_type *key_type,
 461                      const struct btf_type *value_type)
 462 {
 463         return -ENOTSUPP;
 464 }
 465
 466 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 467                          u32 btf_key_id, u32 btf_value_id)
 468 {
 469         const struct btf_type *key_type, *value_type;
 470         u32 key_size, value_size;
 471         int ret = 0;
 472
 473         key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 474         if (!key_type || key_size != map->key_size)
 475                 return -EINVAL;
 476
 477         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 478         if (!value_type || value_size != map->value_size)
 479                 return -EINVAL;
 480
 481         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 482
 483         if (map_value_has_spin_lock(map)) {
 484                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 485                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 486                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
 487                         return -ENOTSUPP;
 488                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 489                     map->value_size) {
 490                         WARN_ONCE(1,
 491                                   "verifier bug spin_lock_off %d value_size %d\n",
 492                                   map->spin_lock_off, map->value_size);
 493                         return -EFAULT;
 494                 }
 495         }
 496
 497         if (map->ops->map_check_btf)
 498                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 499
 500         return ret;
 501 }
 502
 503 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 504 /* called via syscall */
 505 static int map_create(union bpf_attr *attr)
 506 {
 507         int numa_node = bpf_map_attr_numa_node(attr);
 508         struct bpf_map *map;
 509         int f_flags;
 510         int err;
 511
 512         err = CHECK_ATTR(BPF_MAP_CREATE);
 513         if (err)
 514                 return -EINVAL;
 515
 516         f_flags = bpf_get_file_flag(attr->map_flags);
 517         if (f_flags < 0)
 518                 return f_flags;
 519
 520         if (numa_node != NUMA_NO_NODE &&
 521             ((unsigned int)numa_node >= nr_node_ids ||
 522              !node_online(numa_node)))
 523                 return -EINVAL;
 524
 525         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 526         map = find_and_alloc_map(attr);
 527         if (IS_ERR(map))
 528                 return PTR_ERR(map);
 529
 530         err = bpf_obj_name_cpy(map->name, attr->map_name);
 531         if (err)
 532                 goto free_map_nouncharge;
 533
 534         atomic_set(&map->refcnt, 1);
 535         atomic_set(&map->usercnt, 1);
 536
 537         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 538                 struct btf *btf;
 539
 540                 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
 541                         err = -EINVAL;
 542                         goto free_map_nouncharge;
 543                 }
 544
 545                 btf = btf_get_by_fd(attr->btf_fd);
 546                 if (IS_ERR(btf)) {
 547                         err = PTR_ERR(btf);
 548                         goto free_map_nouncharge;
 549                 }
 550
 551                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 552                                     attr->btf_value_type_id);
 553                 if (err) {
 554                         btf_put(btf);
 555                         goto free_map_nouncharge;
 556                 }
 557
 558                 map->btf = btf;
 559                 map->btf_key_type_id = attr->btf_key_type_id;
 560                 map->btf_value_type_id = attr->btf_value_type_id;
 561         } else {
 562                 map->spin_lock_off = -EINVAL;
 563         }
 564
 565         err = security_bpf_map_alloc(map);
 566         if (err)
 567                 goto free_map_nouncharge;
 568
 569         err = bpf_map_init_memlock(map);
 570         if (err)
 571                 goto free_map_sec;
 572
 573         err = bpf_map_alloc_id(map);
 574         if (err)
 575                 goto free_map;
 576
 577         err = bpf_map_new_fd(map, f_flags);
 578         if (err < 0) {
 579                 /* failed to allocate fd.
 580                  * bpf_map_put() is needed because the above
 581                  * bpf_map_alloc_id() has published the map
 582                  * to the userspace and the userspace may
 583                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 584                  */
 585                 bpf_map_put(map);
 586                 return err;
 587         }
 588
 589         return err;
 590
 591 free_map:
 592         bpf_map_release_memlock(map);
 593 free_map_sec:
 594         security_bpf_map_free(map);
 595 free_map_nouncharge:
 596         btf_put(map->btf);
 597         map->ops->map_free(map);
 598         return err;
 599 }
 600
 601 /* if error is returned, fd is released.
 602  * On success caller should complete fd access with matching fdput()
 603  */
 604 struct bpf_map *__bpf_map_get(struct fd f)
 605 {
 606         if (!f.file)
 607                 return ERR_PTR(-EBADF);
 608         if (f.file->f_op != &bpf_map_fops) {
 609                 fdput(f);
 610                 return ERR_PTR(-EINVAL);
 611         }
 612
 613         return f.file->private_data;
 614 }
 615
 616 /* prog's and map's refcnt limit */
 617 #define BPF_MAX_REFCNT 32768
 618
 619 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 620 {
 621         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 622                 atomic_dec(&map->refcnt);
 623                 return ERR_PTR(-EBUSY);
 624         }
 625         if (uref)
 626                 atomic_inc(&map->usercnt);
 627         return map;
 628 }
 629 EXPORT_SYMBOL_GPL(bpf_map_inc);
 630
 631 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 632 {
 633         struct fd f = fdget(ufd);
 634         struct bpf_map *map;
 635
 636         map = __bpf_map_get(f);
 637         if (IS_ERR(map))
 638                 return map;
 639
 640         map = bpf_map_inc(map, true);
 641         fdput(f);
 642
 643         return map;
 644 }
 645
 646 /* map_idr_lock should have been held */
 647 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 648                                             bool uref)
 649 {
 650         int refold;
 651
 652         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 653
 654         if (refold >= BPF_MAX_REFCNT) {
 655                 __bpf_map_put(map, false);
 656                 return ERR_PTR(-EBUSY);
 657         }
 658
 659         if (!refold)
 660                 return ERR_PTR(-ENOENT);
 661
 662         if (uref)
 663                 atomic_inc(&map->usercnt);
 664
 665         return map;
 666 }
 667
 668 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 669 {
 670         return -ENOTSUPP;
 671 }
 672
 673 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 674 {
 675         if (key_size)
 676                 return memdup_user(ukey, key_size);
 677
 678         if (ukey)
 679                 return ERR_PTR(-EINVAL);
 680
 681         return NULL;
 682 }
 683
 684 /* last field in 'union bpf_attr' used by this command */
 685 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 686
 687 static int map_lookup_elem(union bpf_attr *attr)
 688 {
 689         void __user *ukey = u64_to_user_ptr(attr->key);
 690         void __user *uvalue = u64_to_user_ptr(attr->value);
 691         int ufd = attr->map_fd;
 692         struct bpf_map *map;
 693         void *key, *value, *ptr;
 694         u32 value_size;
 695         struct fd f;
 696         int err;
 697
 698         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 699                 return -EINVAL;
 700
 701         if (attr->flags & ~BPF_F_LOCK)
 702                 return -EINVAL;
 703
 704         f = fdget(ufd);
 705         map = __bpf_map_get(f);
 706         if (IS_ERR(map))
 707                 return PTR_ERR(map);
 708
 709         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 710                 err = -EPERM;
 711                 goto err_put;
 712         }
 713
 714         if ((attr->flags & BPF_F_LOCK) &&
 715             !map_value_has_spin_lock(map)) {
 716                 err = -EINVAL;
 717                 goto err_put;
 718         }
 719
 720         key = __bpf_copy_key(ukey, map->key_size);
 721         if (IS_ERR(key)) {
 722                 err = PTR_ERR(key);
 723                 goto err_put;
 724         }
 725
 726         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 727             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 728             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 729             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 730                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 731         else if (IS_FD_MAP(map))
 732                 value_size = sizeof(u32);
 733         else
 734                 value_size = map->value_size;
 735
 736         err = -ENOMEM;
 737         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 738         if (!value)
 739                 goto free_key;
 740
 741         if (bpf_map_is_dev_bound(map)) {
 742                 err = bpf_map_offload_lookup_elem(map, key, value);
 743         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 744                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 745                 err = bpf_percpu_hash_copy(map, key, value);
 746         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 747                 err = bpf_percpu_array_copy(map, key, value);
 748         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 749                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 750         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 751                 err = bpf_stackmap_copy(map, key, value);
 752         } else if (IS_FD_ARRAY(map)) {
 753                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 754         } else if (IS_FD_HASH(map)) {
 755                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 756         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 757                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 758         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 759                    map->map_type == BPF_MAP_TYPE_STACK) {
 760                 err = map->ops->map_peek_elem(map, value);
 761         } else {
 762                 rcu_read_lock();
 763                 ptr = map->ops->map_lookup_elem(map, key);
 764                 if (IS_ERR(ptr)) {
 765                         err = PTR_ERR(ptr);
 766                 } else if (!ptr) {
 767                         err = -ENOENT;
 768                 } else {
 769                         err = 0;
 770                         if (attr->flags & BPF_F_LOCK)
 771                                 /* lock 'ptr' and copy everything but lock */
 772                                 copy_map_value_locked(map, value, ptr, true);
 773                         else
 774                                 copy_map_value(map, value, ptr);
 775                         /* mask lock, since value wasn't zero inited */
 776                         check_and_init_map_lock(map, value);
 777                 }
 778                 rcu_read_unlock();
 779         }
 780
 781         if (err)
 782                 goto free_value;
 783
 784         err = -EFAULT;
 785         if (copy_to_user(uvalue, value, value_size) != 0)
 786                 goto free_value;
 787
 788         err = 0;
 789
 790 free_value:
 791         kfree(value);
 792 free_key:
 793         kfree(key);
 794 err_put:
 795         fdput(f);
 796         return err;
 797 }
 798
 799 static void maybe_wait_bpf_programs(struct bpf_map *map)
 800 {
 801         /* Wait for any running BPF programs to complete so that
 802          * userspace, when we return to it, knows that all programs
 803          * that could be running use the new map value.
 804          */
 805         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 806             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 807                 synchronize_rcu();
 808 }
 809
 810 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 811
 812 static int map_update_elem(union bpf_attr *attr)
 813 {
 814         void __user *ukey = u64_to_user_ptr(attr->key);
 815         void __user *uvalue = u64_to_user_ptr(attr->value);
 816         int ufd = attr->map_fd;
 817         struct bpf_map *map;
 818         void *key, *value;
 819         u32 value_size;
 820         struct fd f;
 821         int err;
 822
 823         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 824                 return -EINVAL;
 825
 826         f = fdget(ufd);
 827         map = __bpf_map_get(f);
 828         if (IS_ERR(map))
 829                 return PTR_ERR(map);
 830
 831         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 832                 err = -EPERM;
 833                 goto err_put;
 834         }
 835
 836         if ((attr->flags & BPF_F_LOCK) &&
 837             !map_value_has_spin_lock(map)) {
 838                 err = -EINVAL;
 839                 goto err_put;
 840         }
 841
 842         key = __bpf_copy_key(ukey, map->key_size);
 843         if (IS_ERR(key)) {
 844                 err = PTR_ERR(key);
 845                 goto err_put;
 846         }
 847
 848         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 849             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 850             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 851             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 852                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 853         else
 854                 value_size = map->value_size;
 855
 856         err = -ENOMEM;
 857         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 858         if (!value)
 859                 goto free_key;
 860
 861         err = -EFAULT;
 862         if (copy_from_user(value, uvalue, value_size) != 0)
 863                 goto free_value;
 864
 865         /* Need to create a kthread, thus must support schedule */
 866         if (bpf_map_is_dev_bound(map)) {
 867                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 868                 goto out;
 869         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 870                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 871                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 872                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 873                 goto out;
 874         }
 875
 876         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 877          * inside bpf map update or delete otherwise deadlocks are possible
 878          */
 879         preempt_disable();
 880         __this_cpu_inc(bpf_prog_active);
 881         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 882             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 883                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 884         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 885                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 886         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 887                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 888                                                        attr->flags);
 889         } else if (IS_FD_ARRAY(map)) {
 890                 rcu_read_lock();
 891                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 892                                                    attr->flags);
 893                 rcu_read_unlock();
 894         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 895                 rcu_read_lock();
 896                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 897                                                   attr->flags);
 898                 rcu_read_unlock();
 899         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 900                 /* rcu_read_lock() is not needed */
 901                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 902                                                          attr->flags);
 903         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 904                    map->map_type == BPF_MAP_TYPE_STACK) {
 905                 err = map->ops->map_push_elem(map, value, attr->flags);
 906         } else {
 907                 rcu_read_lock();
 908                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 909                 rcu_read_unlock();
 910         }
 911         __this_cpu_dec(bpf_prog_active);
 912         preempt_enable();
 913         maybe_wait_bpf_programs(map);
 914 out:
 915 free_value:
 916         kfree(value);
 917 free_key:
 918         kfree(key);
 919 err_put:
 920         fdput(f);
 921         return err;
 922 }
 923
 924 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 925
 926 static int map_delete_elem(union bpf_attr *attr)
 927 {
 928         void __user *ukey = u64_to_user_ptr(attr->key);
 929         int ufd = attr->map_fd;
 930         struct bpf_map *map;
 931         struct fd f;
 932         void *key;
 933         int err;
 934
 935         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 936                 return -EINVAL;
 937
 938         f = fdget(ufd);
 939         map = __bpf_map_get(f);
 940         if (IS_ERR(map))
 941                 return PTR_ERR(map);
 942
 943         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 944                 err = -EPERM;
 945                 goto err_put;
 946         }
 947
 948         key = __bpf_copy_key(ukey, map->key_size);
 949         if (IS_ERR(key)) {
 950                 err = PTR_ERR(key);
 951                 goto err_put;
 952         }
 953
 954         if (bpf_map_is_dev_bound(map)) {
 955                 err = bpf_map_offload_delete_elem(map, key);
 956                 goto out;
 957         }
 958
 959         preempt_disable();
 960         __this_cpu_inc(bpf_prog_active);
 961         rcu_read_lock();
 962         err = map->ops->map_delete_elem(map, key);
 963         rcu_read_unlock();
 964         __this_cpu_dec(bpf_prog_active);
 965         preempt_enable();
 966         maybe_wait_bpf_programs(map);
 967 out:
 968         kfree(key);
 969 err_put:
 970         fdput(f);
 971         return err;
 972 }
 973
 974 /* last field in 'union bpf_attr' used by this command */
 975 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 976
 977 static int map_get_next_key(union bpf_attr *attr)
 978 {
 979         void __user *ukey = u64_to_user_ptr(attr->key);
 980         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 981         int ufd = attr->map_fd;
 982         struct bpf_map *map;
 983         void *key, *next_key;
 984         struct fd f;
 985         int err;
 986
 987         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 988                 return -EINVAL;
 989
 990         f = fdget(ufd);
 991         map = __bpf_map_get(f);
 992         if (IS_ERR(map))
 993                 return PTR_ERR(map);
 994
 995         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 996                 err = -EPERM;
 997                 goto err_put;
 998         }
 999
1000         if (ukey) {
1001                 key = __bpf_copy_key(ukey, map->key_size);
1002                 if (IS_ERR(key)) {
1003                         err = PTR_ERR(key);
1004                         goto err_put;
1005                 }
1006         } else {
1007                 key = NULL;
1008         }
1009
1010         err = -ENOMEM;
1011         next_key = kmalloc(map->key_size, GFP_USER);
1012         if (!next_key)
1013                 goto free_key;
1014
1015         if (bpf_map_is_dev_bound(map)) {
1016                 err = bpf_map_offload_get_next_key(map, key, next_key);
1017                 goto out;
1018         }
1019
1020         rcu_read_lock();
1021         err = map->ops->map_get_next_key(map, key, next_key);
1022         rcu_read_unlock();
1023 out:
1024         if (err)
1025                 goto free_next_key;
1026
1027         err = -EFAULT;
1028         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1029                 goto free_next_key;
1030
1031         err = 0;
1032
1033 free_next_key:
1034         kfree(next_key);
1035 free_key:
1036         kfree(key);
1037 err_put:
1038         fdput(f);
1039         return err;
1040 }
1041
1042 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1043
1044 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1045 {
1046         void __user *ukey = u64_to_user_ptr(attr->key);
1047         void __user *uvalue = u64_to_user_ptr(attr->value);
1048         int ufd = attr->map_fd;
1049         struct bpf_map *map;
1050         void *key, *value;
1051         u32 value_size;
1052         struct fd f;
1053         int err;
1054
1055         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1056                 return -EINVAL;
1057
1058         f = fdget(ufd);
1059         map = __bpf_map_get(f);
1060         if (IS_ERR(map))
1061                 return PTR_ERR(map);
1062
1063         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1064                 err = -EPERM;
1065                 goto err_put;
1066         }
1067
1068         key = __bpf_copy_key(ukey, map->key_size);
1069         if (IS_ERR(key)) {
1070                 err = PTR_ERR(key);
1071                 goto err_put;
1072         }
1073
1074         value_size = map->value_size;
1075
1076         err = -ENOMEM;
1077         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1078         if (!value)
1079                 goto free_key;
1080
1081         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1082             map->map_type == BPF_MAP_TYPE_STACK) {
1083                 err = map->ops->map_pop_elem(map, value);
1084         } else {
1085                 err = -ENOTSUPP;
1086         }
1087
1088         if (err)
1089                 goto free_value;
1090
1091         if (copy_to_user(uvalue, value, value_size) != 0)
1092                 goto free_value;
1093
1094         err = 0;
1095
1096 free_value:
1097         kfree(value);
1098 free_key:
1099         kfree(key);
1100 err_put:
1101         fdput(f);
1102         return err;
1103 }
1104
1105 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1106 #define BPF_PROG_TYPE(_id, _name) \
1107         [_id] = & _name ## _prog_ops,
1108 #define BPF_MAP_TYPE(_id, _ops)
1109 #include <linux/bpf_types.h>
1110 #undef BPF_PROG_TYPE
1111 #undef BPF_MAP_TYPE
1112 };
1113
1114 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1115 {
1116         const struct bpf_prog_ops *ops;
1117
1118         if (type >= ARRAY_SIZE(bpf_prog_types))
1119                 return -EINVAL;
1120         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1121         ops = bpf_prog_types[type];
1122         if (!ops)
1123                 return -EINVAL;
1124
1125         if (!bpf_prog_is_dev_bound(prog->aux))
1126                 prog->aux->ops = ops;
1127         else
1128                 prog->aux->ops = &bpf_offload_prog_ops;
1129         prog->type = type;
1130         return 0;
1131 }
1132
1133 /* drop refcnt on maps used by eBPF program and free auxilary data */
1134 static void free_used_maps(struct bpf_prog_aux *aux)
1135 {
1136         enum bpf_cgroup_storage_type stype;
1137         int i;
1138
1139         for_each_cgroup_storage_type(stype) {
1140                 if (!aux->cgroup_storage[stype])
1141                         continue;
1142                 bpf_cgroup_storage_release(aux->prog,
1143                                            aux->cgroup_storage[stype]);
1144         }
1145
1146         for (i = 0; i < aux->used_map_cnt; i++)
1147                 bpf_map_put(aux->used_maps[i]);
1148
1149         kfree(aux->used_maps);
1150 }
1151
1152 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1153 {
1154         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1155         unsigned long user_bufs;
1156
1157         if (user) {
1158                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1159                 if (user_bufs > memlock_limit) {
1160                         atomic_long_sub(pages, &user->locked_vm);
1161                         return -EPERM;
1162                 }
1163         }
1164
1165         return 0;
1166 }
1167
1168 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1169 {
1170         if (user)
1171                 atomic_long_sub(pages, &user->locked_vm);
1172 }
1173
1174 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1175 {
1176         struct user_struct *user = get_current_user();
1177         int ret;
1178
1179         ret = __bpf_prog_charge(user, prog->pages);
1180         if (ret) {
1181                 free_uid(user);
1182                 return ret;
1183         }
1184
1185         prog->aux->user = user;
1186         return 0;
1187 }
1188
1189 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1190 {
1191         struct user_struct *user = prog->aux->user;
1192
1193         __bpf_prog_uncharge(user, prog->pages);
1194         free_uid(user);
1195 }
1196
1197 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1198 {
1199         int id;
1200
1201         idr_preload(GFP_KERNEL);
1202         spin_lock_bh(&prog_idr_lock);
1203         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1204         if (id > 0)
1205                 prog->aux->id = id;
1206         spin_unlock_bh(&prog_idr_lock);
1207         idr_preload_end();
1208
1209         /* id is in [1, INT_MAX) */
1210         if (WARN_ON_ONCE(!id))
1211                 return -ENOSPC;
1212
1213         return id > 0 ? 0 : id;
1214 }
1215
1216 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1217 {
1218         /* cBPF to eBPF migrations are currently not in the idr store.
1219          * Offloaded programs are removed from the store when their device
1220          * disappears - even if someone grabs an fd to them they are unusable,
1221          * simply waiting for refcnt to drop to be freed.
1222          */
1223         if (!prog->aux->id)
1224                 return;
1225
1226         if (do_idr_lock)
1227                 spin_lock_bh(&prog_idr_lock);
1228         else
1229                 __acquire(&prog_idr_lock);
1230
1231         idr_remove(&prog_idr, prog->aux->id);
1232         prog->aux->id = 0;
1233
1234         if (do_idr_lock)
1235                 spin_unlock_bh(&prog_idr_lock);
1236         else
1237                 __release(&prog_idr_lock);
1238 }
1239
1240 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1241 {
1242         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1243
1244         free_used_maps(aux);
1245         bpf_prog_uncharge_memlock(aux->prog);
1246         security_bpf_prog_free(aux);
1247         bpf_prog_free(aux->prog);
1248 }
1249
1250 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1251 {
1252         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1253                 /* bpf_prog_free_id() must be called first */
1254                 bpf_prog_free_id(prog, do_idr_lock);
1255                 bpf_prog_kallsyms_del_all(prog);
1256                 btf_put(prog->aux->btf);
1257                 kvfree(prog->aux->func_info);
1258                 bpf_prog_free_linfo(prog);
1259
1260                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1261         }
1262 }
1263
1264 void bpf_prog_put(struct bpf_prog *prog)
1265 {
1266         __bpf_prog_put(prog, true);
1267 }
1268 EXPORT_SYMBOL_GPL(bpf_prog_put);
1269
1270 static int bpf_prog_release(struct inode *inode, struct file *filp)
1271 {
1272         struct bpf_prog *prog = filp->private_data;
1273
1274         bpf_prog_put(prog);
1275         return 0;
1276 }
1277
1278 #ifdef CONFIG_PROC_FS
1279 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1280 {
1281         const struct bpf_prog *prog = filp->private_data;
1282         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1283
1284         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1285         seq_printf(m,
1286                    "prog_type:\t%u\n"
1287                    "prog_jited:\t%u\n"
1288                    "prog_tag:\t%s\n"
1289                    "memlock:\t%llu\n"
1290                    "prog_id:\t%u\n",
1291                    prog->type,
1292                    prog->jited,
1293                    prog_tag,
1294                    prog->pages * 1ULL << PAGE_SHIFT,
1295                    prog->aux->id);
1296 }
1297 #endif
1298
1299 const struct file_operations bpf_prog_fops = {
1300 #ifdef CONFIG_PROC_FS
1301         .show_fdinfo    = bpf_prog_show_fdinfo,
1302 #endif
1303         .release        = bpf_prog_release,
1304         .read           = bpf_dummy_read,
1305         .write          = bpf_dummy_write,
1306 };
1307
1308 int bpf_prog_new_fd(struct bpf_prog *prog)
1309 {
1310         int ret;
1311
1312         ret = security_bpf_prog(prog);
1313         if (ret < 0)
1314                 return ret;
1315
1316         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1317                                 O_RDWR | O_CLOEXEC);
1318 }
1319
1320 static struct bpf_prog *____bpf_prog_get(struct fd f)
1321 {
1322         if (!f.file)
1323                 return ERR_PTR(-EBADF);
1324         if (f.file->f_op != &bpf_prog_fops) {
1325                 fdput(f);
1326                 return ERR_PTR(-EINVAL);
1327         }
1328
1329         return f.file->private_data;
1330 }
1331
1332 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1333 {
1334         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1335                 atomic_sub(i, &prog->aux->refcnt);
1336                 return ERR_PTR(-EBUSY);
1337         }
1338         return prog;
1339 }
1340 EXPORT_SYMBOL_GPL(bpf_prog_add);
1341
1342 void bpf_prog_sub(struct bpf_prog *prog, int i)
1343 {
1344         /* Only to be used for undoing previous bpf_prog_add() in some
1345          * error path. We still know that another entity in our call
1346          * path holds a reference to the program, thus atomic_sub() can
1347          * be safely used in such cases!
1348          */
1349         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1350 }
1351 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1352
1353 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1354 {
1355         return bpf_prog_add(prog, 1);
1356 }
1357 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1358
1359 /* prog_idr_lock should have been held */
1360 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1361 {
1362         int refold;
1363
1364         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1365
1366         if (refold >= BPF_MAX_REFCNT) {
1367                 __bpf_prog_put(prog, false);
1368                 return ERR_PTR(-EBUSY);
1369         }
1370
1371         if (!refold)
1372                 return ERR_PTR(-ENOENT);
1373
1374         return prog;
1375 }
1376 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1377
1378 bool bpf_prog_get_ok(struct bpf_prog *prog,
1379                             enum bpf_prog_type *attach_type, bool attach_drv)
1380 {
1381         /* not an attachment, just a refcount inc, always allow */
1382         if (!attach_type)
1383                 return true;
1384
1385         if (prog->type != *attach_type)
1386                 return false;
1387         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1388                 return false;
1389
1390         return true;
1391 }
1392
1393 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1394                                        bool attach_drv)
1395 {
1396         struct fd f = fdget(ufd);
1397         struct bpf_prog *prog;
1398
1399         prog = ____bpf_prog_get(f);
1400         if (IS_ERR(prog))
1401                 return prog;
1402         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1403                 prog = ERR_PTR(-EINVAL);
1404                 goto out;
1405         }
1406
1407         prog = bpf_prog_inc(prog);
1408 out:
1409         fdput(f);
1410         return prog;
1411 }
1412
1413 struct bpf_prog *bpf_prog_get(u32 ufd)
1414 {
1415         return __bpf_prog_get(ufd, NULL, false);
1416 }
1417
1418 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1419                                        bool attach_drv)
1420 {
1421         return __bpf_prog_get(ufd, &type, attach_drv);
1422 }
1423 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1424
1425 /* Initially all BPF programs could be loaded w/o specifying
1426  * expected_attach_type. Later for some of them specifying expected_attach_type
1427  * at load time became required so that program could be validated properly.
1428  * Programs of types that are allowed to be loaded both w/ and w/o (for
1429  * backward compatibility) expected_attach_type, should have the default attach
1430  * type assigned to expected_attach_type for the latter case, so that it can be
1431  * validated later at attach time.
1432  *
1433  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1434  * prog type requires it but has some attach types that have to be backward
1435  * compatible.
1436  */
1437 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1438 {
1439         switch (attr->prog_type) {
1440         case BPF_PROG_TYPE_CGROUP_SOCK:
1441                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1442                  * exist so checking for non-zero is the way to go here.
1443                  */
1444                 if (!attr->expected_attach_type)
1445                         attr->expected_attach_type =
1446                                 BPF_CGROUP_INET_SOCK_CREATE;
1447                 break;
1448         }
1449 }
1450
1451 static int
1452 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1453                                 enum bpf_attach_type expected_attach_type)
1454 {
1455         switch (prog_type) {
1456         case BPF_PROG_TYPE_CGROUP_SOCK:
1457                 switch (expected_attach_type) {
1458                 case BPF_CGROUP_INET_SOCK_CREATE:
1459                 case BPF_CGROUP_INET4_POST_BIND:
1460                 case BPF_CGROUP_INET6_POST_BIND:
1461                         return 0;
1462                 default:
1463                         return -EINVAL;
1464                 }
1465         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1466                 switch (expected_attach_type) {
1467                 case BPF_CGROUP_INET4_BIND:
1468                 case BPF_CGROUP_INET6_BIND:
1469                 case BPF_CGROUP_INET4_CONNECT:
1470                 case BPF_CGROUP_INET6_CONNECT:
1471                 case BPF_CGROUP_UDP4_SENDMSG:
1472                 case BPF_CGROUP_UDP6_SENDMSG:
1473                         return 0;
1474                 default:
1475                         return -EINVAL;
1476                 }
1477         default:
1478                 return 0;
1479         }
1480 }
1481
1482 /* last field in 'union bpf_attr' used by this command */
1483 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1484
1485 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1486 {
1487         enum bpf_prog_type type = attr->prog_type;
1488         struct bpf_prog *prog;
1489         int err;
1490         char license[128];
1491         bool is_gpl;
1492
1493         if (CHECK_ATTR(BPF_PROG_LOAD))
1494                 return -EINVAL;
1495
1496         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1497                 return -EINVAL;
1498
1499         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1500             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1501             !capable(CAP_SYS_ADMIN))
1502                 return -EPERM;
1503
1504         /* copy eBPF program license from user space */
1505         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1506                               sizeof(license) - 1) < 0)
1507                 return -EFAULT;
1508         license[sizeof(license) - 1] = 0;
1509
1510         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1511         is_gpl = license_is_gpl_compatible(license);
1512
1513         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1514                 return -E2BIG;
1515         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1516             type != BPF_PROG_TYPE_CGROUP_SKB &&
1517             !capable(CAP_SYS_ADMIN))
1518                 return -EPERM;
1519
1520         bpf_prog_load_fixup_attach_type(attr);
1521         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1522                 return -EINVAL;
1523
1524         /* plain bpf_prog allocation */
1525         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1526         if (!prog)
1527                 return -ENOMEM;
1528
1529         prog->expected_attach_type = attr->expected_attach_type;
1530
1531         prog->aux->offload_requested = !!attr->prog_ifindex;
1532
1533         err = security_bpf_prog_alloc(prog->aux);
1534         if (err)
1535                 goto free_prog_nouncharge;
1536
1537         err = bpf_prog_charge_memlock(prog);
1538         if (err)
1539                 goto free_prog_sec;
1540
1541         prog->len = attr->insn_cnt;
1542
1543         err = -EFAULT;
1544         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1545                            bpf_prog_insn_size(prog)) != 0)
1546                 goto free_prog;
1547
1548         prog->orig_prog = NULL;
1549         prog->jited = 0;
1550
1551         atomic_set(&prog->aux->refcnt, 1);
1552         prog->gpl_compatible = is_gpl ? 1 : 0;
1553
1554         if (bpf_prog_is_dev_bound(prog->aux)) {
1555                 err = bpf_prog_offload_init(prog, attr);
1556                 if (err)
1557                         goto free_prog;
1558         }
1559
1560         /* find program type: socket_filter vs tracing_filter */
1561         err = find_prog_type(type, prog);
1562         if (err < 0)
1563                 goto free_prog;
1564
1565         prog->aux->load_time = ktime_get_boot_ns();
1566         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1567         if (err)
1568                 goto free_prog;
1569
1570         /* run eBPF verifier */
1571         err = bpf_check(&prog, attr, uattr);
1572         if (err < 0)
1573                 goto free_used_maps;
1574
1575         prog = bpf_prog_select_runtime(prog, &err);
1576         if (err < 0)
1577                 goto free_used_maps;
1578
1579         err = bpf_prog_alloc_id(prog);
1580         if (err)
1581                 goto free_used_maps;
1582
1583         err = bpf_prog_new_fd(prog);
1584         if (err < 0) {
1585                 /* failed to allocate fd.
1586                  * bpf_prog_put() is needed because the above
1587                  * bpf_prog_alloc_id() has published the prog
1588                  * to the userspace and the userspace may
1589                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1590                  */
1591                 bpf_prog_put(prog);
1592                 return err;
1593         }
1594
1595         bpf_prog_kallsyms_add(prog);
1596         return err;
1597
1598 free_used_maps:
1599         bpf_prog_free_linfo(prog);
1600         kvfree(prog->aux->func_info);
1601         btf_put(prog->aux->btf);
1602         bpf_prog_kallsyms_del_subprogs(prog);
1603         free_used_maps(prog->aux);
1604 free_prog:
1605         bpf_prog_uncharge_memlock(prog);
1606 free_prog_sec:
1607         security_bpf_prog_free(prog->aux);
1608 free_prog_nouncharge:
1609         bpf_prog_free(prog);
1610         return err;
1611 }
1612
1613 #define BPF_OBJ_LAST_FIELD file_flags
1614
1615 static int bpf_obj_pin(const union bpf_attr *attr)
1616 {
1617         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1618                 return -EINVAL;
1619
1620         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1621 }
1622
1623 static int bpf_obj_get(const union bpf_attr *attr)
1624 {
1625         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1626             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1627                 return -EINVAL;
1628
1629         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1630                                 attr->file_flags);
1631 }
1632
1633 struct bpf_raw_tracepoint {
1634         struct bpf_raw_event_map *btp;
1635         struct bpf_prog *prog;
1636 };
1637
1638 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1639 {
1640         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1641
1642         if (raw_tp->prog) {
1643                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1644                 bpf_prog_put(raw_tp->prog);
1645         }
1646         bpf_put_raw_tracepoint(raw_tp->btp);
1647         kfree(raw_tp);
1648         return 0;
1649 }
1650
1651 static const struct file_operations bpf_raw_tp_fops = {
1652         .release        = bpf_raw_tracepoint_release,
1653         .read           = bpf_dummy_read,
1654         .write          = bpf_dummy_write,
1655 };
1656
1657 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1658
1659 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1660 {
1661         struct bpf_raw_tracepoint *raw_tp;
1662         struct bpf_raw_event_map *btp;
1663         struct bpf_prog *prog;
1664         char tp_name[128];
1665         int tp_fd, err;
1666
1667         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1668                               sizeof(tp_name) - 1) < 0)
1669                 return -EFAULT;
1670         tp_name[sizeof(tp_name) - 1] = 0;
1671
1672         btp = bpf_get_raw_tracepoint(tp_name);
1673         if (!btp)
1674                 return -ENOENT;
1675
1676         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1677         if (!raw_tp) {
1678                 err = -ENOMEM;
1679                 goto out_put_btp;
1680         }
1681         raw_tp->btp = btp;
1682
1683         prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1684                                  BPF_PROG_TYPE_RAW_TRACEPOINT);
1685         if (IS_ERR(prog)) {
1686                 err = PTR_ERR(prog);
1687                 goto out_free_tp;
1688         }
1689
1690         err = bpf_probe_register(raw_tp->btp, prog);
1691         if (err)
1692                 goto out_put_prog;
1693
1694         raw_tp->prog = prog;
1695         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1696                                  O_CLOEXEC);
1697         if (tp_fd < 0) {
1698                 bpf_probe_unregister(raw_tp->btp, prog);
1699                 err = tp_fd;
1700                 goto out_put_prog;
1701         }
1702         return tp_fd;
1703
1704 out_put_prog:
1705         bpf_prog_put(prog);
1706 out_free_tp:
1707         kfree(raw_tp);
1708 out_put_btp:
1709         bpf_put_raw_tracepoint(btp);
1710         return err;
1711 }
1712
1713 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1714                                              enum bpf_attach_type attach_type)
1715 {
1716         switch (prog->type) {
1717         case BPF_PROG_TYPE_CGROUP_SOCK:
1718         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1719                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1720         default:
1721                 return 0;
1722         }
1723 }
1724
1725 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1726
1727 #define BPF_F_ATTACH_MASK \
1728         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1729
1730 static int bpf_prog_attach(const union bpf_attr *attr)
1731 {
1732         enum bpf_prog_type ptype;
1733         struct bpf_prog *prog;
1734         int ret;
1735
1736         if (!capable(CAP_NET_ADMIN))
1737                 return -EPERM;
1738
1739         if (CHECK_ATTR(BPF_PROG_ATTACH))
1740                 return -EINVAL;
1741
1742         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1743                 return -EINVAL;
1744
1745         switch (attr->attach_type) {
1746         case BPF_CGROUP_INET_INGRESS:
1747         case BPF_CGROUP_INET_EGRESS:
1748                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1749                 break;
1750         case BPF_CGROUP_INET_SOCK_CREATE:
1751         case BPF_CGROUP_INET4_POST_BIND:
1752         case BPF_CGROUP_INET6_POST_BIND:
1753                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1754                 break;
1755         case BPF_CGROUP_INET4_BIND:
1756         case BPF_CGROUP_INET6_BIND:
1757         case BPF_CGROUP_INET4_CONNECT:
1758         case BPF_CGROUP_INET6_CONNECT:
1759         case BPF_CGROUP_UDP4_SENDMSG:
1760         case BPF_CGROUP_UDP6_SENDMSG:
1761                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1762                 break;
1763         case BPF_CGROUP_SOCK_OPS:
1764                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1765                 break;
1766         case BPF_CGROUP_DEVICE:
1767                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1768                 break;
1769         case BPF_SK_MSG_VERDICT:
1770                 ptype = BPF_PROG_TYPE_SK_MSG;
1771                 break;
1772         case BPF_SK_SKB_STREAM_PARSER:
1773         case BPF_SK_SKB_STREAM_VERDICT:
1774                 ptype = BPF_PROG_TYPE_SK_SKB;
1775                 break;
1776         case BPF_LIRC_MODE2:
1777                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1778                 break;
1779         case BPF_FLOW_DISSECTOR:
1780                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1781                 break;
1782         default:
1783                 return -EINVAL;
1784         }
1785
1786         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1787         if (IS_ERR(prog))
1788                 return PTR_ERR(prog);
1789
1790         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1791                 bpf_prog_put(prog);
1792                 return -EINVAL;
1793         }
1794
1795         switch (ptype) {
1796         case BPF_PROG_TYPE_SK_SKB:
1797         case BPF_PROG_TYPE_SK_MSG:
1798                 ret = sock_map_get_from_fd(attr, prog);
1799                 break;
1800         case BPF_PROG_TYPE_LIRC_MODE2:
1801                 ret = lirc_prog_attach(attr, prog);
1802                 break;
1803         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1804                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1805                 break;
1806         default:
1807                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1808         }
1809
1810         if (ret)
1811                 bpf_prog_put(prog);
1812         return ret;
1813 }
1814
1815 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1816
1817 static int bpf_prog_detach(const union bpf_attr *attr)
1818 {
1819         enum bpf_prog_type ptype;
1820
1821         if (!capable(CAP_NET_ADMIN))
1822                 return -EPERM;
1823
1824         if (CHECK_ATTR(BPF_PROG_DETACH))
1825                 return -EINVAL;
1826
1827         switch (attr->attach_type) {
1828         case BPF_CGROUP_INET_INGRESS:
1829         case BPF_CGROUP_INET_EGRESS:
1830                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1831                 break;
1832         case BPF_CGROUP_INET_SOCK_CREATE:
1833         case BPF_CGROUP_INET4_POST_BIND:
1834         case BPF_CGROUP_INET6_POST_BIND:
1835                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1836                 break;
1837         case BPF_CGROUP_INET4_BIND:
1838         case BPF_CGROUP_INET6_BIND:
1839         case BPF_CGROUP_INET4_CONNECT:
1840         case BPF_CGROUP_INET6_CONNECT:
1841         case BPF_CGROUP_UDP4_SENDMSG:
1842         case BPF_CGROUP_UDP6_SENDMSG:
1843                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1844                 break;
1845         case BPF_CGROUP_SOCK_OPS:
1846                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1847                 break;
1848         case BPF_CGROUP_DEVICE:
1849                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1850                 break;
1851         case BPF_SK_MSG_VERDICT:
1852                 return sock_map_get_from_fd(attr, NULL);
1853         case BPF_SK_SKB_STREAM_PARSER:
1854         case BPF_SK_SKB_STREAM_VERDICT:
1855                 return sock_map_get_from_fd(attr, NULL);
1856         case BPF_LIRC_MODE2:
1857                 return lirc_prog_detach(attr);
1858         case BPF_FLOW_DISSECTOR:
1859                 return skb_flow_dissector_bpf_prog_detach(attr);
1860         default:
1861                 return -EINVAL;
1862         }
1863
1864         return cgroup_bpf_prog_detach(attr, ptype);
1865 }
1866
1867 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1868
1869 static int bpf_prog_query(const union bpf_attr *attr,
1870                           union bpf_attr __user *uattr)
1871 {
1872         if (!capable(CAP_NET_ADMIN))
1873                 return -EPERM;
1874         if (CHECK_ATTR(BPF_PROG_QUERY))
1875                 return -EINVAL;
1876         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1877                 return -EINVAL;
1878
1879         switch (attr->query.attach_type) {
1880         case BPF_CGROUP_INET_INGRESS:
1881         case BPF_CGROUP_INET_EGRESS:
1882         case BPF_CGROUP_INET_SOCK_CREATE:
1883         case BPF_CGROUP_INET4_BIND:
1884         case BPF_CGROUP_INET6_BIND:
1885         case BPF_CGROUP_INET4_POST_BIND:
1886         case BPF_CGROUP_INET6_POST_BIND:
1887         case BPF_CGROUP_INET4_CONNECT:
1888         case BPF_CGROUP_INET6_CONNECT:
1889         case BPF_CGROUP_UDP4_SENDMSG:
1890         case BPF_CGROUP_UDP6_SENDMSG:
1891         case BPF_CGROUP_SOCK_OPS:
1892         case BPF_CGROUP_DEVICE:
1893                 break;
1894         case BPF_LIRC_MODE2:
1895                 return lirc_prog_query(attr, uattr);
1896         default:
1897                 return -EINVAL;
1898         }
1899
1900         return cgroup_bpf_prog_query(attr, uattr);
1901 }
1902
1903 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1904
1905 static int bpf_prog_test_run(const union bpf_attr *attr,
1906                              union bpf_attr __user *uattr)
1907 {
1908         struct bpf_prog *prog;
1909         int ret = -ENOTSUPP;
1910
1911         if (!capable(CAP_SYS_ADMIN))
1912                 return -EPERM;
1913         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1914                 return -EINVAL;
1915
1916         prog = bpf_prog_get(attr->test.prog_fd);
1917         if (IS_ERR(prog))
1918                 return PTR_ERR(prog);
1919
1920         if (prog->aux->ops->test_run)
1921                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1922
1923         bpf_prog_put(prog);
1924         return ret;
1925 }
1926
1927 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1928
1929 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1930                                union bpf_attr __user *uattr,
1931                                struct idr *idr,
1932                                spinlock_t *lock)
1933 {
1934         u32 next_id = attr->start_id;
1935         int err = 0;
1936
1937         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1938                 return -EINVAL;
1939
1940         if (!capable(CAP_SYS_ADMIN))
1941                 return -EPERM;
1942
1943         next_id++;
1944         spin_lock_bh(lock);
1945         if (!idr_get_next(idr, &next_id))
1946                 err = -ENOENT;
1947         spin_unlock_bh(lock);
1948
1949         if (!err)
1950                 err = put_user(next_id, &uattr->next_id);
1951
1952         return err;
1953 }
1954
1955 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1956
1957 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1958 {
1959         struct bpf_prog *prog;
1960         u32 id = attr->prog_id;
1961         int fd;
1962
1963         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1964                 return -EINVAL;
1965
1966         if (!capable(CAP_SYS_ADMIN))
1967                 return -EPERM;
1968
1969         spin_lock_bh(&prog_idr_lock);
1970         prog = idr_find(&prog_idr, id);
1971         if (prog)
1972                 prog = bpf_prog_inc_not_zero(prog);
1973         else
1974                 prog = ERR_PTR(-ENOENT);
1975         spin_unlock_bh(&prog_idr_lock);
1976
1977         if (IS_ERR(prog))
1978                 return PTR_ERR(prog);
1979
1980         fd = bpf_prog_new_fd(prog);
1981         if (fd < 0)
1982                 bpf_prog_put(prog);
1983
1984         return fd;
1985 }
1986
1987 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1988
1989 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1990 {
1991         struct bpf_map *map;
1992         u32 id = attr->map_id;
1993         int f_flags;
1994         int fd;
1995
1996         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1997             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1998                 return -EINVAL;
1999
2000         if (!capable(CAP_SYS_ADMIN))
2001                 return -EPERM;
2002
2003         f_flags = bpf_get_file_flag(attr->open_flags);
2004         if (f_flags < 0)
2005                 return f_flags;
2006
2007         spin_lock_bh(&map_idr_lock);
2008         map = idr_find(&map_idr, id);
2009         if (map)
2010                 map = bpf_map_inc_not_zero(map, true);
2011         else
2012                 map = ERR_PTR(-ENOENT);
2013         spin_unlock_bh(&map_idr_lock);
2014
2015         if (IS_ERR(map))
2016                 return PTR_ERR(map);
2017
2018         fd = bpf_map_new_fd(map, f_flags);
2019         if (fd < 0)
2020                 bpf_map_put(map);
2021
2022         return fd;
2023 }
2024
2025 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2026                                               unsigned long addr)
2027 {
2028         int i;
2029
2030         for (i = 0; i < prog->aux->used_map_cnt; i++)
2031                 if (prog->aux->used_maps[i] == (void *)addr)
2032                         return prog->aux->used_maps[i];
2033         return NULL;
2034 }
2035
2036 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2037 {
2038         const struct bpf_map *map;
2039         struct bpf_insn *insns;
2040         u64 imm;
2041         int i;
2042
2043         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2044                         GFP_USER);
2045         if (!insns)
2046                 return insns;
2047
2048         for (i = 0; i < prog->len; i++) {
2049                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2050                         insns[i].code = BPF_JMP | BPF_CALL;
2051                         insns[i].imm = BPF_FUNC_tail_call;
2052                         /* fall-through */
2053                 }
2054                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2055                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2056                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2057                                 insns[i].code = BPF_JMP | BPF_CALL;
2058                         if (!bpf_dump_raw_ok())
2059                                 insns[i].imm = 0;
2060                         continue;
2061                 }
2062
2063                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2064                         continue;
2065
2066                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2067                 map = bpf_map_from_imm(prog, imm);
2068                 if (map) {
2069                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2070                         insns[i].imm = map->id;
2071                         insns[i + 1].imm = 0;
2072                         continue;
2073                 }
2074         }
2075
2076         return insns;
2077 }
2078
2079 static int set_info_rec_size(struct bpf_prog_info *info)
2080 {
2081         /*
2082          * Ensure info.*_rec_size is the same as kernel expected size
2083          *
2084          * or
2085          *
2086          * Only allow zero *_rec_size if both _rec_size and _cnt are
2087          * zero.  In this case, the kernel will set the expected
2088          * _rec_size back to the info.
2089          */
2090
2091         if ((info->nr_func_info || info->func_info_rec_size) &&
2092             info->func_info_rec_size != sizeof(struct bpf_func_info))
2093                 return -EINVAL;
2094
2095         if ((info->nr_line_info || info->line_info_rec_size) &&
2096             info->line_info_rec_size != sizeof(struct bpf_line_info))
2097                 return -EINVAL;
2098
2099         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2100             info->jited_line_info_rec_size != sizeof(__u64))
2101                 return -EINVAL;
2102
2103         info->func_info_rec_size = sizeof(struct bpf_func_info);
2104         info->line_info_rec_size = sizeof(struct bpf_line_info);
2105         info->jited_line_info_rec_size = sizeof(__u64);
2106
2107         return 0;
2108 }
2109
2110 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2111                                    const union bpf_attr *attr,
2112                                    union bpf_attr __user *uattr)
2113 {
2114         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2115         struct bpf_prog_info info = {};
2116         u32 info_len = attr->info.info_len;
2117         char __user *uinsns;
2118         u32 ulen;
2119         int err;
2120
2121         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2122         if (err)
2123                 return err;
2124         info_len = min_t(u32, sizeof(info), info_len);
2125
2126         if (copy_from_user(&info, uinfo, info_len))
2127                 return -EFAULT;
2128
2129         info.type = prog->type;
2130         info.id = prog->aux->id;
2131         info.load_time = prog->aux->load_time;
2132         info.created_by_uid = from_kuid_munged(current_user_ns(),
2133                                                prog->aux->user->uid);
2134         info.gpl_compatible = prog->gpl_compatible;
2135
2136         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2137         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2138
2139         ulen = info.nr_map_ids;
2140         info.nr_map_ids = prog->aux->used_map_cnt;
2141         ulen = min_t(u32, info.nr_map_ids, ulen);
2142         if (ulen) {
2143                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2144                 u32 i;
2145
2146                 for (i = 0; i < ulen; i++)
2147                         if (put_user(prog->aux->used_maps[i]->id,
2148                                      &user_map_ids[i]))
2149                                 return -EFAULT;
2150         }
2151
2152         err = set_info_rec_size(&info);
2153         if (err)
2154                 return err;
2155
2156         if (!capable(CAP_SYS_ADMIN)) {
2157                 info.jited_prog_len = 0;
2158                 info.xlated_prog_len = 0;
2159                 info.nr_jited_ksyms = 0;
2160                 info.nr_jited_func_lens = 0;
2161                 info.nr_func_info = 0;
2162                 info.nr_line_info = 0;
2163                 info.nr_jited_line_info = 0;
2164                 goto done;
2165         }
2166
2167         ulen = info.xlated_prog_len;
2168         info.xlated_prog_len = bpf_prog_insn_size(prog);
2169         if (info.xlated_prog_len && ulen) {
2170                 struct bpf_insn *insns_sanitized;
2171                 bool fault;
2172
2173                 if (prog->blinded && !bpf_dump_raw_ok()) {
2174                         info.xlated_prog_insns = 0;
2175                         goto done;
2176                 }
2177                 insns_sanitized = bpf_insn_prepare_dump(prog);
2178                 if (!insns_sanitized)
2179                         return -ENOMEM;
2180                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2181                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2182                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2183                 kfree(insns_sanitized);
2184                 if (fault)
2185                         return -EFAULT;
2186         }
2187
2188         if (bpf_prog_is_dev_bound(prog->aux)) {
2189                 err = bpf_prog_offload_info_fill(&info, prog);
2190                 if (err)
2191                         return err;
2192                 goto done;
2193         }
2194
2195         /* NOTE: the following code is supposed to be skipped for offload.
2196          * bpf_prog_offload_info_fill() is the place to fill similar fields
2197          * for offload.
2198          */
2199         ulen = info.jited_prog_len;
2200         if (prog->aux->func_cnt) {
2201                 u32 i;
2202
2203                 info.jited_prog_len = 0;
2204                 for (i = 0; i < prog->aux->func_cnt; i++)
2205                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2206         } else {
2207                 info.jited_prog_len = prog->jited_len;
2208         }
2209
2210         if (info.jited_prog_len && ulen) {
2211                 if (bpf_dump_raw_ok()) {
2212                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2213                         ulen = min_t(u32, info.jited_prog_len, ulen);
2214
2215                         /* for multi-function programs, copy the JITed
2216                          * instructions for all the functions
2217                          */
2218                         if (prog->aux->func_cnt) {
2219                                 u32 len, free, i;
2220                                 u8 *img;
2221
2222                                 free = ulen;
2223                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2224                                         len = prog->aux->func[i]->jited_len;
2225                                         len = min_t(u32, len, free);
2226                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2227                                         if (copy_to_user(uinsns, img, len))
2228                                                 return -EFAULT;
2229                                         uinsns += len;
2230                                         free -= len;
2231                                         if (!free)
2232                                                 break;
2233                                 }
2234                         } else {
2235                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2236                                         return -EFAULT;
2237                         }
2238                 } else {
2239                         info.jited_prog_insns = 0;
2240                 }
2241         }
2242
2243         ulen = info.nr_jited_ksyms;
2244         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2245         if (ulen) {
2246                 if (bpf_dump_raw_ok()) {
2247                         unsigned long ksym_addr;
2248                         u64 __user *user_ksyms;
2249                         u32 i;
2250
2251                         /* copy the address of the kernel symbol
2252                          * corresponding to each function
2253                          */
2254                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2255                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2256                         if (prog->aux->func_cnt) {
2257                                 for (i = 0; i < ulen; i++) {
2258                                         ksym_addr = (unsigned long)
2259                                                 prog->aux->func[i]->bpf_func;
2260                                         if (put_user((u64) ksym_addr,
2261                                                      &user_ksyms[i]))
2262                                                 return -EFAULT;
2263                                 }
2264                         } else {
2265                                 ksym_addr = (unsigned long) prog->bpf_func;
2266                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2267                                         return -EFAULT;
2268                         }
2269                 } else {
2270                         info.jited_ksyms = 0;
2271                 }
2272         }
2273
2274         ulen = info.nr_jited_func_lens;
2275         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2276         if (ulen) {
2277                 if (bpf_dump_raw_ok()) {
2278                         u32 __user *user_lens;
2279                         u32 func_len, i;
2280
2281                         /* copy the JITed image lengths for each function */
2282                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2283                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2284                         if (prog->aux->func_cnt) {
2285                                 for (i = 0; i < ulen; i++) {
2286                                         func_len =
2287                                                 prog->aux->func[i]->jited_len;
2288                                         if (put_user(func_len, &user_lens[i]))
2289                                                 return -EFAULT;
2290                                 }
2291                         } else {
2292                                 func_len = prog->jited_len;
2293                                 if (put_user(func_len, &user_lens[0]))
2294                                         return -EFAULT;
2295                         }
2296                 } else {
2297                         info.jited_func_lens = 0;
2298                 }
2299         }
2300
2301         if (prog->aux->btf)
2302                 info.btf_id = btf_id(prog->aux->btf);
2303
2304         ulen = info.nr_func_info;
2305         info.nr_func_info = prog->aux->func_info_cnt;
2306         if (info.nr_func_info && ulen) {
2307                 char __user *user_finfo;
2308
2309                 user_finfo = u64_to_user_ptr(info.func_info);
2310                 ulen = min_t(u32, info.nr_func_info, ulen);
2311                 if (copy_to_user(user_finfo, prog->aux->func_info,
2312                                  info.func_info_rec_size * ulen))
2313                         return -EFAULT;
2314         }
2315
2316         ulen = info.nr_line_info;
2317         info.nr_line_info = prog->aux->nr_linfo;
2318         if (info.nr_line_info && ulen) {
2319                 __u8 __user *user_linfo;
2320
2321                 user_linfo = u64_to_user_ptr(info.line_info);
2322                 ulen = min_t(u32, info.nr_line_info, ulen);
2323                 if (copy_to_user(user_linfo, prog->aux->linfo,
2324                                  info.line_info_rec_size * ulen))
2325                         return -EFAULT;
2326         }
2327
2328         ulen = info.nr_jited_line_info;
2329         if (prog->aux->jited_linfo)
2330                 info.nr_jited_line_info = prog->aux->nr_linfo;
2331         else
2332                 info.nr_jited_line_info = 0;
2333         if (info.nr_jited_line_info && ulen) {
2334                 if (bpf_dump_raw_ok()) {
2335                         __u64 __user *user_linfo;
2336                         u32 i;
2337
2338                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2339                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2340                         for (i = 0; i < ulen; i++) {
2341                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2342                                              &user_linfo[i]))
2343                                         return -EFAULT;
2344                         }
2345                 } else {
2346                         info.jited_line_info = 0;
2347                 }
2348         }
2349
2350         ulen = info.nr_prog_tags;
2351         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2352         if (ulen) {
2353                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2354                 u32 i;
2355
2356                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2357                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2358                 if (prog->aux->func_cnt) {
2359                         for (i = 0; i < ulen; i++) {
2360                                 if (copy_to_user(user_prog_tags[i],
2361                                                  prog->aux->func[i]->tag,
2362                                                  BPF_TAG_SIZE))
2363                                         return -EFAULT;
2364                         }
2365                 } else {
2366                         if (copy_to_user(user_prog_tags[0],
2367                                          prog->tag, BPF_TAG_SIZE))
2368                                 return -EFAULT;
2369                 }
2370         }
2371
2372 done:
2373         if (copy_to_user(uinfo, &info, info_len) ||
2374             put_user(info_len, &uattr->info.info_len))
2375                 return -EFAULT;
2376
2377         return 0;
2378 }
2379
2380 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2381                                   const union bpf_attr *attr,
2382                                   union bpf_attr __user *uattr)
2383 {
2384         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2385         struct bpf_map_info info = {};
2386         u32 info_len = attr->info.info_len;
2387         int err;
2388
2389         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2390         if (err)
2391                 return err;
2392         info_len = min_t(u32, sizeof(info), info_len);
2393
2394         info.type = map->map_type;
2395         info.id = map->id;
2396         info.key_size = map->key_size;
2397         info.value_size = map->value_size;
2398         info.max_entries = map->max_entries;
2399         info.map_flags = map->map_flags;
2400         memcpy(info.name, map->name, sizeof(map->name));
2401
2402         if (map->btf) {
2403                 info.btf_id = btf_id(map->btf);
2404                 info.btf_key_type_id = map->btf_key_type_id;
2405                 info.btf_value_type_id = map->btf_value_type_id;
2406         }
2407
2408         if (bpf_map_is_dev_bound(map)) {
2409                 err = bpf_map_offload_info_fill(&info, map);
2410                 if (err)
2411                         return err;
2412         }
2413
2414         if (copy_to_user(uinfo, &info, info_len) ||
2415             put_user(info_len, &uattr->info.info_len))
2416                 return -EFAULT;
2417
2418         return 0;
2419 }
2420
2421 static int bpf_btf_get_info_by_fd(struct btf *btf,
2422                                   const union bpf_attr *attr,
2423                                   union bpf_attr __user *uattr)
2424 {
2425         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2426         u32 info_len = attr->info.info_len;
2427         int err;
2428
2429         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2430         if (err)
2431                 return err;
2432
2433         return btf_get_info_by_fd(btf, attr, uattr);
2434 }
2435
2436 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2437
2438 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2439                                   union bpf_attr __user *uattr)
2440 {
2441         int ufd = attr->info.bpf_fd;
2442         struct fd f;
2443         int err;
2444
2445         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2446                 return -EINVAL;
2447
2448         f = fdget(ufd);
2449         if (!f.file)
2450                 return -EBADFD;
2451
2452         if (f.file->f_op == &bpf_prog_fops)
2453                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2454                                               uattr);
2455         else if (f.file->f_op == &bpf_map_fops)
2456                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2457                                              uattr);
2458         else if (f.file->f_op == &btf_fops)
2459                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2460         else
2461                 err = -EINVAL;
2462
2463         fdput(f);
2464         return err;
2465 }
2466
2467 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2468
2469 static int bpf_btf_load(const union bpf_attr *attr)
2470 {
2471         if (CHECK_ATTR(BPF_BTF_LOAD))
2472                 return -EINVAL;
2473
2474         if (!capable(CAP_SYS_ADMIN))
2475                 return -EPERM;
2476
2477         return btf_new_fd(attr);
2478 }
2479
2480 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2481
2482 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2483 {
2484         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2485                 return -EINVAL;
2486
2487         if (!capable(CAP_SYS_ADMIN))
2488                 return -EPERM;
2489
2490         return btf_get_fd_by_id(attr->btf_id);
2491 }
2492
2493 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2494                                     union bpf_attr __user *uattr,
2495                                     u32 prog_id, u32 fd_type,
2496                                     const char *buf, u64 probe_offset,
2497                                     u64 probe_addr)
2498 {
2499         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2500         u32 len = buf ? strlen(buf) : 0, input_len;
2501         int err = 0;
2502
2503         if (put_user(len, &uattr->task_fd_query.buf_len))
2504                 return -EFAULT;
2505         input_len = attr->task_fd_query.buf_len;
2506         if (input_len && ubuf) {
2507                 if (!len) {
2508                         /* nothing to copy, just make ubuf NULL terminated */
2509                         char zero = '\0';
2510
2511                         if (put_user(zero, ubuf))
2512                                 return -EFAULT;
2513                 } else if (input_len >= len + 1) {
2514                         /* ubuf can hold the string with NULL terminator */
2515                         if (copy_to_user(ubuf, buf, len + 1))
2516                                 return -EFAULT;
2517                 } else {
2518                         /* ubuf cannot hold the string with NULL terminator,
2519                          * do a partial copy with NULL terminator.
2520                          */
2521                         char zero = '\0';
2522
2523                         err = -ENOSPC;
2524                         if (copy_to_user(ubuf, buf, input_len - 1))
2525                                 return -EFAULT;
2526                         if (put_user(zero, ubuf + input_len - 1))
2527                                 return -EFAULT;
2528                 }
2529         }
2530
2531         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2532             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2533             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2534             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2535                 return -EFAULT;
2536
2537         return err;
2538 }
2539
2540 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2541
2542 static int bpf_task_fd_query(const union bpf_attr *attr,
2543                              union bpf_attr __user *uattr)
2544 {
2545         pid_t pid = attr->task_fd_query.pid;
2546         u32 fd = attr->task_fd_query.fd;
2547         const struct perf_event *event;
2548         struct files_struct *files;
2549         struct task_struct *task;
2550         struct file *file;
2551         int err;
2552
2553         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2554                 return -EINVAL;
2555
2556         if (!capable(CAP_SYS_ADMIN))
2557                 return -EPERM;
2558
2559         if (attr->task_fd_query.flags != 0)
2560                 return -EINVAL;
2561
2562         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2563         if (!task)
2564                 return -ENOENT;
2565
2566         files = get_files_struct(task);
2567         put_task_struct(task);
2568         if (!files)
2569                 return -ENOENT;
2570
2571         err = 0;
2572         spin_lock(&files->file_lock);
2573         file = fcheck_files(files, fd);
2574         if (!file)
2575                 err = -EBADF;
2576         else
2577                 get_file(file);
2578         spin_unlock(&files->file_lock);
2579         put_files_struct(files);
2580
2581         if (err)
2582                 goto out;
2583
2584         if (file->f_op == &bpf_raw_tp_fops) {
2585                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2586                 struct bpf_raw_event_map *btp = raw_tp->btp;
2587
2588                 err = bpf_task_fd_query_copy(attr, uattr,
2589                                              raw_tp->prog->aux->id,
2590                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2591                                              btp->tp->name, 0, 0);
2592                 goto put_file;
2593         }
2594
2595         event = perf_get_event(file);
2596         if (!IS_ERR(event)) {
2597                 u64 probe_offset, probe_addr;
2598                 u32 prog_id, fd_type;
2599                 const char *buf;
2600
2601                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2602                                               &buf, &probe_offset,
2603                                               &probe_addr);
2604                 if (!err)
2605                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2606                                                      fd_type, buf,
2607                                                      probe_offset,
2608                                                      probe_addr);
2609                 goto put_file;
2610         }
2611
2612         err = -ENOTSUPP;
2613 put_file:
2614         fput(file);
2615 out:
2616         return err;
2617 }
2618
2619 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2620 {
2621         union bpf_attr attr = {};
2622         int err;
2623
2624         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2625                 return -EPERM;
2626
2627         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2628         if (err)
2629                 return err;
2630         size = min_t(u32, size, sizeof(attr));
2631
2632         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2633         if (copy_from_user(&attr, uattr, size) != 0)
2634                 return -EFAULT;
2635
2636         err = security_bpf(cmd, &attr, size);
2637         if (err < 0)
2638                 return err;
2639
2640         switch (cmd) {
2641         case BPF_MAP_CREATE:
2642                 err = map_create(&attr);
2643                 break;
2644         case BPF_MAP_LOOKUP_ELEM:
2645                 err = map_lookup_elem(&attr);
2646                 break;
2647         case BPF_MAP_UPDATE_ELEM:
2648                 err = map_update_elem(&attr);
2649                 break;
2650         case BPF_MAP_DELETE_ELEM:
2651                 err = map_delete_elem(&attr);
2652                 break;
2653         case BPF_MAP_GET_NEXT_KEY:
2654                 err = map_get_next_key(&attr);
2655                 break;
2656         case BPF_PROG_LOAD:
2657                 err = bpf_prog_load(&attr, uattr);
2658                 break;
2659         case BPF_OBJ_PIN:
2660                 err = bpf_obj_pin(&attr);
2661                 break;
2662         case BPF_OBJ_GET:
2663                 err = bpf_obj_get(&attr);
2664                 break;
2665         case BPF_PROG_ATTACH:
2666                 err = bpf_prog_attach(&attr);
2667                 break;
2668         case BPF_PROG_DETACH:
2669                 err = bpf_prog_detach(&attr);
2670                 break;
2671         case BPF_PROG_QUERY:
2672                 err = bpf_prog_query(&attr, uattr);
2673                 break;
2674         case BPF_PROG_TEST_RUN:
2675                 err = bpf_prog_test_run(&attr, uattr);
2676                 break;
2677         case BPF_PROG_GET_NEXT_ID:
2678                 err = bpf_obj_get_next_id(&attr, uattr,
2679                                           &prog_idr, &prog_idr_lock);
2680                 break;
2681         case BPF_MAP_GET_NEXT_ID:
2682                 err = bpf_obj_get_next_id(&attr, uattr,
2683                                           &map_idr, &map_idr_lock);
2684                 break;
2685         case BPF_PROG_GET_FD_BY_ID:
2686                 err = bpf_prog_get_fd_by_id(&attr);
2687                 break;
2688         case BPF_MAP_GET_FD_BY_ID:
2689                 err = bpf_map_get_fd_by_id(&attr);
2690                 break;
2691         case BPF_OBJ_GET_INFO_BY_FD:
2692                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2693                 break;
2694         case BPF_RAW_TRACEPOINT_OPEN:
2695                 err = bpf_raw_tracepoint_open(&attr);
2696                 break;
2697         case BPF_BTF_LOAD:
2698                 err = bpf_btf_load(&attr);
2699                 break;
2700         case BPF_BTF_GET_FD_BY_ID:
2701                 err = bpf_btf_get_fd_by_id(&attr);
2702                 break;
2703         case BPF_TASK_FD_QUERY:
2704                 err = bpf_task_fd_query(&attr, uattr);
2705                 break;
2706         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2707                 err = map_lookup_and_delete_elem(&attr);
2708                 break;
2709         default:
2710                 err = -EINVAL;
2711                 break;
2712         }
2713
2714         return err;
2715 }