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