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
1219                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1220         }
1221 }
1222
1223 void bpf_prog_put(struct bpf_prog *prog)
1224 {
1225         __bpf_prog_put(prog, true);
1226 }
1227 EXPORT_SYMBOL_GPL(bpf_prog_put);
1228
1229 static int bpf_prog_release(struct inode *inode, struct file *filp)
1230 {
1231         struct bpf_prog *prog = filp->private_data;
1232
1233         bpf_prog_put(prog);
1234         return 0;
1235 }
1236
1237 #ifdef CONFIG_PROC_FS
1238 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1239 {
1240         const struct bpf_prog *prog = filp->private_data;
1241         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1242
1243         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1244         seq_printf(m,
1245                    "prog_type:\t%u\n"
1246                    "prog_jited:\t%u\n"
1247                    "prog_tag:\t%s\n"
1248                    "memlock:\t%llu\n"
1249                    "prog_id:\t%u\n",
1250                    prog->type,
1251                    prog->jited,
1252                    prog_tag,
1253                    prog->pages * 1ULL << PAGE_SHIFT,
1254                    prog->aux->id);
1255 }
1256 #endif
1257
1258 const struct file_operations bpf_prog_fops = {
1259 #ifdef CONFIG_PROC_FS
1260         .show_fdinfo    = bpf_prog_show_fdinfo,
1261 #endif
1262         .release        = bpf_prog_release,
1263         .read           = bpf_dummy_read,
1264         .write          = bpf_dummy_write,
1265 };
1266
1267 int bpf_prog_new_fd(struct bpf_prog *prog)
1268 {
1269         int ret;
1270
1271         ret = security_bpf_prog(prog);
1272         if (ret < 0)
1273                 return ret;
1274
1275         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1276                                 O_RDWR | O_CLOEXEC);
1277 }
1278
1279 static struct bpf_prog *____bpf_prog_get(struct fd f)
1280 {
1281         if (!f.file)
1282                 return ERR_PTR(-EBADF);
1283         if (f.file->f_op != &bpf_prog_fops) {
1284                 fdput(f);
1285                 return ERR_PTR(-EINVAL);
1286         }
1287
1288         return f.file->private_data;
1289 }
1290
1291 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1292 {
1293         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1294                 atomic_sub(i, &prog->aux->refcnt);
1295                 return ERR_PTR(-EBUSY);
1296         }
1297         return prog;
1298 }
1299 EXPORT_SYMBOL_GPL(bpf_prog_add);
1300
1301 void bpf_prog_sub(struct bpf_prog *prog, int i)
1302 {
1303         /* Only to be used for undoing previous bpf_prog_add() in some
1304          * error path. We still know that another entity in our call
1305          * path holds a reference to the program, thus atomic_sub() can
1306          * be safely used in such cases!
1307          */
1308         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1309 }
1310 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1311
1312 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1313 {
1314         return bpf_prog_add(prog, 1);
1315 }
1316 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1317
1318 /* prog_idr_lock should have been held */
1319 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1320 {
1321         int refold;
1322
1323         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1324
1325         if (refold >= BPF_MAX_REFCNT) {
1326                 __bpf_prog_put(prog, false);
1327                 return ERR_PTR(-EBUSY);
1328         }
1329
1330         if (!refold)
1331                 return ERR_PTR(-ENOENT);
1332
1333         return prog;
1334 }
1335 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1336
1337 bool bpf_prog_get_ok(struct bpf_prog *prog,
1338                             enum bpf_prog_type *attach_type, bool attach_drv)
1339 {
1340         /* not an attachment, just a refcount inc, always allow */
1341         if (!attach_type)
1342                 return true;
1343
1344         if (prog->type != *attach_type)
1345                 return false;
1346         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1347                 return false;
1348
1349         return true;
1350 }
1351
1352 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1353                                        bool attach_drv)
1354 {
1355         struct fd f = fdget(ufd);
1356         struct bpf_prog *prog;
1357
1358         prog = ____bpf_prog_get(f);
1359         if (IS_ERR(prog))
1360                 return prog;
1361         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1362                 prog = ERR_PTR(-EINVAL);
1363                 goto out;
1364         }
1365
1366         prog = bpf_prog_inc(prog);
1367 out:
1368         fdput(f);
1369         return prog;
1370 }
1371
1372 struct bpf_prog *bpf_prog_get(u32 ufd)
1373 {
1374         return __bpf_prog_get(ufd, NULL, false);
1375 }
1376
1377 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1378                                        bool attach_drv)
1379 {
1380         return __bpf_prog_get(ufd, &type, attach_drv);
1381 }
1382 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1383
1384 /* Initially all BPF programs could be loaded w/o specifying
1385  * expected_attach_type. Later for some of them specifying expected_attach_type
1386  * at load time became required so that program could be validated properly.
1387  * Programs of types that are allowed to be loaded both w/ and w/o (for
1388  * backward compatibility) expected_attach_type, should have the default attach
1389  * type assigned to expected_attach_type for the latter case, so that it can be
1390  * validated later at attach time.
1391  *
1392  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1393  * prog type requires it but has some attach types that have to be backward
1394  * compatible.
1395  */
1396 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1397 {
1398         switch (attr->prog_type) {
1399         case BPF_PROG_TYPE_CGROUP_SOCK:
1400                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1401                  * exist so checking for non-zero is the way to go here.
1402                  */
1403                 if (!attr->expected_attach_type)
1404                         attr->expected_attach_type =
1405                                 BPF_CGROUP_INET_SOCK_CREATE;
1406                 break;
1407         }
1408 }
1409
1410 static int
1411 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1412                                 enum bpf_attach_type expected_attach_type)
1413 {
1414         switch (prog_type) {
1415         case BPF_PROG_TYPE_CGROUP_SOCK:
1416                 switch (expected_attach_type) {
1417                 case BPF_CGROUP_INET_SOCK_CREATE:
1418                 case BPF_CGROUP_INET4_POST_BIND:
1419                 case BPF_CGROUP_INET6_POST_BIND:
1420                         return 0;
1421                 default:
1422                         return -EINVAL;
1423                 }
1424         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1425                 switch (expected_attach_type) {
1426                 case BPF_CGROUP_INET4_BIND:
1427                 case BPF_CGROUP_INET6_BIND:
1428                 case BPF_CGROUP_INET4_CONNECT:
1429                 case BPF_CGROUP_INET6_CONNECT:
1430                 case BPF_CGROUP_UDP4_SENDMSG:
1431                 case BPF_CGROUP_UDP6_SENDMSG:
1432                         return 0;
1433                 default:
1434                         return -EINVAL;
1435                 }
1436         default:
1437                 return 0;
1438         }
1439 }
1440
1441 /* last field in 'union bpf_attr' used by this command */
1442 #define BPF_PROG_LOAD_LAST_FIELD func_info_cnt
1443
1444 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1445 {
1446         enum bpf_prog_type type = attr->prog_type;
1447         struct bpf_prog *prog;
1448         int err;
1449         char license[128];
1450         bool is_gpl;
1451
1452         if (CHECK_ATTR(BPF_PROG_LOAD))
1453                 return -EINVAL;
1454
1455         if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1456                 return -EINVAL;
1457
1458         /* copy eBPF program license from user space */
1459         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1460                               sizeof(license) - 1) < 0)
1461                 return -EFAULT;
1462         license[sizeof(license) - 1] = 0;
1463
1464         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1465         is_gpl = license_is_gpl_compatible(license);
1466
1467         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1468                 return -E2BIG;
1469
1470         if (type == BPF_PROG_TYPE_KPROBE &&
1471             attr->kern_version != LINUX_VERSION_CODE)
1472                 return -EINVAL;
1473
1474         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1475             type != BPF_PROG_TYPE_CGROUP_SKB &&
1476             !capable(CAP_SYS_ADMIN))
1477                 return -EPERM;
1478
1479         bpf_prog_load_fixup_attach_type(attr);
1480         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1481                 return -EINVAL;
1482
1483         /* plain bpf_prog allocation */
1484         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1485         if (!prog)
1486                 return -ENOMEM;
1487
1488         prog->expected_attach_type = attr->expected_attach_type;
1489
1490         prog->aux->offload_requested = !!attr->prog_ifindex;
1491
1492         err = security_bpf_prog_alloc(prog->aux);
1493         if (err)
1494                 goto free_prog_nouncharge;
1495
1496         err = bpf_prog_charge_memlock(prog);
1497         if (err)
1498                 goto free_prog_sec;
1499
1500         prog->len = attr->insn_cnt;
1501
1502         err = -EFAULT;
1503         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1504                            bpf_prog_insn_size(prog)) != 0)
1505                 goto free_prog;
1506
1507         prog->orig_prog = NULL;
1508         prog->jited = 0;
1509
1510         atomic_set(&prog->aux->refcnt, 1);
1511         prog->gpl_compatible = is_gpl ? 1 : 0;
1512
1513         if (bpf_prog_is_dev_bound(prog->aux)) {
1514                 err = bpf_prog_offload_init(prog, attr);
1515                 if (err)
1516                         goto free_prog;
1517         }
1518
1519         /* find program type: socket_filter vs tracing_filter */
1520         err = find_prog_type(type, prog);
1521         if (err < 0)
1522                 goto free_prog;
1523
1524         prog->aux->load_time = ktime_get_boot_ns();
1525         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1526         if (err)
1527                 goto free_prog;
1528
1529         /* run eBPF verifier */
1530         err = bpf_check(&prog, attr, uattr);
1531         if (err < 0)
1532                 goto free_used_maps;
1533
1534         prog = bpf_prog_select_runtime(prog, &err);
1535         if (err < 0)
1536                 goto free_used_maps;
1537
1538         err = bpf_prog_alloc_id(prog);
1539         if (err)
1540                 goto free_used_maps;
1541
1542         err = bpf_prog_new_fd(prog);
1543         if (err < 0) {
1544                 /* failed to allocate fd.
1545                  * bpf_prog_put() is needed because the above
1546                  * bpf_prog_alloc_id() has published the prog
1547                  * to the userspace and the userspace may
1548                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1549                  */
1550                 bpf_prog_put(prog);
1551                 return err;
1552         }
1553
1554         bpf_prog_kallsyms_add(prog);
1555         return err;
1556
1557 free_used_maps:
1558         bpf_prog_kallsyms_del_subprogs(prog);
1559         free_used_maps(prog->aux);
1560 free_prog:
1561         bpf_prog_uncharge_memlock(prog);
1562 free_prog_sec:
1563         security_bpf_prog_free(prog->aux);
1564 free_prog_nouncharge:
1565         bpf_prog_free(prog);
1566         return err;
1567 }
1568
1569 #define BPF_OBJ_LAST_FIELD file_flags
1570
1571 static int bpf_obj_pin(const union bpf_attr *attr)
1572 {
1573         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1574                 return -EINVAL;
1575
1576         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1577 }
1578
1579 static int bpf_obj_get(const union bpf_attr *attr)
1580 {
1581         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1582             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1583                 return -EINVAL;
1584
1585         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1586                                 attr->file_flags);
1587 }
1588
1589 struct bpf_raw_tracepoint {
1590         struct bpf_raw_event_map *btp;
1591         struct bpf_prog *prog;
1592 };
1593
1594 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1595 {
1596         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1597
1598         if (raw_tp->prog) {
1599                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1600                 bpf_prog_put(raw_tp->prog);
1601         }
1602         kfree(raw_tp);
1603         return 0;
1604 }
1605
1606 static const struct file_operations bpf_raw_tp_fops = {
1607         .release        = bpf_raw_tracepoint_release,
1608         .read           = bpf_dummy_read,
1609         .write          = bpf_dummy_write,
1610 };
1611
1612 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1613
1614 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1615 {
1616         struct bpf_raw_tracepoint *raw_tp;
1617         struct bpf_raw_event_map *btp;
1618         struct bpf_prog *prog;
1619         char tp_name[128];
1620         int tp_fd, err;
1621
1622         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1623                               sizeof(tp_name) - 1) < 0)
1624                 return -EFAULT;
1625         tp_name[sizeof(tp_name) - 1] = 0;
1626
1627         btp = bpf_find_raw_tracepoint(tp_name);
1628         if (!btp)
1629                 return -ENOENT;
1630
1631         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1632         if (!raw_tp)
1633                 return -ENOMEM;
1634         raw_tp->btp = btp;
1635
1636         prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1637                                  BPF_PROG_TYPE_RAW_TRACEPOINT);
1638         if (IS_ERR(prog)) {
1639                 err = PTR_ERR(prog);
1640                 goto out_free_tp;
1641         }
1642
1643         err = bpf_probe_register(raw_tp->btp, prog);
1644         if (err)
1645                 goto out_put_prog;
1646
1647         raw_tp->prog = prog;
1648         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1649                                  O_CLOEXEC);
1650         if (tp_fd < 0) {
1651                 bpf_probe_unregister(raw_tp->btp, prog);
1652                 err = tp_fd;
1653                 goto out_put_prog;
1654         }
1655         return tp_fd;
1656
1657 out_put_prog:
1658         bpf_prog_put(prog);
1659 out_free_tp:
1660         kfree(raw_tp);
1661         return err;
1662 }
1663
1664 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1665                                              enum bpf_attach_type attach_type)
1666 {
1667         switch (prog->type) {
1668         case BPF_PROG_TYPE_CGROUP_SOCK:
1669         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1670                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1671         default:
1672                 return 0;
1673         }
1674 }
1675
1676 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1677
1678 #define BPF_F_ATTACH_MASK \
1679         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1680
1681 static int bpf_prog_attach(const union bpf_attr *attr)
1682 {
1683         enum bpf_prog_type ptype;
1684         struct bpf_prog *prog;
1685         int ret;
1686
1687         if (!capable(CAP_NET_ADMIN))
1688                 return -EPERM;
1689
1690         if (CHECK_ATTR(BPF_PROG_ATTACH))
1691                 return -EINVAL;
1692
1693         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1694                 return -EINVAL;
1695
1696         switch (attr->attach_type) {
1697         case BPF_CGROUP_INET_INGRESS:
1698         case BPF_CGROUP_INET_EGRESS:
1699                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1700                 break;
1701         case BPF_CGROUP_INET_SOCK_CREATE:
1702         case BPF_CGROUP_INET4_POST_BIND:
1703         case BPF_CGROUP_INET6_POST_BIND:
1704                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1705                 break;
1706         case BPF_CGROUP_INET4_BIND:
1707         case BPF_CGROUP_INET6_BIND:
1708         case BPF_CGROUP_INET4_CONNECT:
1709         case BPF_CGROUP_INET6_CONNECT:
1710         case BPF_CGROUP_UDP4_SENDMSG:
1711         case BPF_CGROUP_UDP6_SENDMSG:
1712                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1713                 break;
1714         case BPF_CGROUP_SOCK_OPS:
1715                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1716                 break;
1717         case BPF_CGROUP_DEVICE:
1718                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1719                 break;
1720         case BPF_SK_MSG_VERDICT:
1721                 ptype = BPF_PROG_TYPE_SK_MSG;
1722                 break;
1723         case BPF_SK_SKB_STREAM_PARSER:
1724         case BPF_SK_SKB_STREAM_VERDICT:
1725                 ptype = BPF_PROG_TYPE_SK_SKB;
1726                 break;
1727         case BPF_LIRC_MODE2:
1728                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1729                 break;
1730         case BPF_FLOW_DISSECTOR:
1731                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1732                 break;
1733         default:
1734                 return -EINVAL;
1735         }
1736
1737         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1738         if (IS_ERR(prog))
1739                 return PTR_ERR(prog);
1740
1741         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1742                 bpf_prog_put(prog);
1743                 return -EINVAL;
1744         }
1745
1746         switch (ptype) {
1747         case BPF_PROG_TYPE_SK_SKB:
1748         case BPF_PROG_TYPE_SK_MSG:
1749                 ret = sock_map_get_from_fd(attr, prog);
1750                 break;
1751         case BPF_PROG_TYPE_LIRC_MODE2:
1752                 ret = lirc_prog_attach(attr, prog);
1753                 break;
1754         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1755                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1756                 break;
1757         default:
1758                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1759         }
1760
1761         if (ret)
1762                 bpf_prog_put(prog);
1763         return ret;
1764 }
1765
1766 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1767
1768 static int bpf_prog_detach(const union bpf_attr *attr)
1769 {
1770         enum bpf_prog_type ptype;
1771
1772         if (!capable(CAP_NET_ADMIN))
1773                 return -EPERM;
1774
1775         if (CHECK_ATTR(BPF_PROG_DETACH))
1776                 return -EINVAL;
1777
1778         switch (attr->attach_type) {
1779         case BPF_CGROUP_INET_INGRESS:
1780         case BPF_CGROUP_INET_EGRESS:
1781                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1782                 break;
1783         case BPF_CGROUP_INET_SOCK_CREATE:
1784         case BPF_CGROUP_INET4_POST_BIND:
1785         case BPF_CGROUP_INET6_POST_BIND:
1786                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1787                 break;
1788         case BPF_CGROUP_INET4_BIND:
1789         case BPF_CGROUP_INET6_BIND:
1790         case BPF_CGROUP_INET4_CONNECT:
1791         case BPF_CGROUP_INET6_CONNECT:
1792         case BPF_CGROUP_UDP4_SENDMSG:
1793         case BPF_CGROUP_UDP6_SENDMSG:
1794                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1795                 break;
1796         case BPF_CGROUP_SOCK_OPS:
1797                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1798                 break;
1799         case BPF_CGROUP_DEVICE:
1800                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1801                 break;
1802         case BPF_SK_MSG_VERDICT:
1803                 return sock_map_get_from_fd(attr, NULL);
1804         case BPF_SK_SKB_STREAM_PARSER:
1805         case BPF_SK_SKB_STREAM_VERDICT:
1806                 return sock_map_get_from_fd(attr, NULL);
1807         case BPF_LIRC_MODE2:
1808                 return lirc_prog_detach(attr);
1809         case BPF_FLOW_DISSECTOR:
1810                 return skb_flow_dissector_bpf_prog_detach(attr);
1811         default:
1812                 return -EINVAL;
1813         }
1814
1815         return cgroup_bpf_prog_detach(attr, ptype);
1816 }
1817
1818 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1819
1820 static int bpf_prog_query(const union bpf_attr *attr,
1821                           union bpf_attr __user *uattr)
1822 {
1823         if (!capable(CAP_NET_ADMIN))
1824                 return -EPERM;
1825         if (CHECK_ATTR(BPF_PROG_QUERY))
1826                 return -EINVAL;
1827         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1828                 return -EINVAL;
1829
1830         switch (attr->query.attach_type) {
1831         case BPF_CGROUP_INET_INGRESS:
1832         case BPF_CGROUP_INET_EGRESS:
1833         case BPF_CGROUP_INET_SOCK_CREATE:
1834         case BPF_CGROUP_INET4_BIND:
1835         case BPF_CGROUP_INET6_BIND:
1836         case BPF_CGROUP_INET4_POST_BIND:
1837         case BPF_CGROUP_INET6_POST_BIND:
1838         case BPF_CGROUP_INET4_CONNECT:
1839         case BPF_CGROUP_INET6_CONNECT:
1840         case BPF_CGROUP_UDP4_SENDMSG:
1841         case BPF_CGROUP_UDP6_SENDMSG:
1842         case BPF_CGROUP_SOCK_OPS:
1843         case BPF_CGROUP_DEVICE:
1844                 break;
1845         case BPF_LIRC_MODE2:
1846                 return lirc_prog_query(attr, uattr);
1847         default:
1848                 return -EINVAL;
1849         }
1850
1851         return cgroup_bpf_prog_query(attr, uattr);
1852 }
1853
1854 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1855
1856 static int bpf_prog_test_run(const union bpf_attr *attr,
1857                              union bpf_attr __user *uattr)
1858 {
1859         struct bpf_prog *prog;
1860         int ret = -ENOTSUPP;
1861
1862         if (!capable(CAP_SYS_ADMIN))
1863                 return -EPERM;
1864         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1865                 return -EINVAL;
1866
1867         prog = bpf_prog_get(attr->test.prog_fd);
1868         if (IS_ERR(prog))
1869                 return PTR_ERR(prog);
1870
1871         if (prog->aux->ops->test_run)
1872                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1873
1874         bpf_prog_put(prog);
1875         return ret;
1876 }
1877
1878 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1879
1880 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1881                                union bpf_attr __user *uattr,
1882                                struct idr *idr,
1883                                spinlock_t *lock)
1884 {
1885         u32 next_id = attr->start_id;
1886         int err = 0;
1887
1888         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1889                 return -EINVAL;
1890
1891         if (!capable(CAP_SYS_ADMIN))
1892                 return -EPERM;
1893
1894         next_id++;
1895         spin_lock_bh(lock);
1896         if (!idr_get_next(idr, &next_id))
1897                 err = -ENOENT;
1898         spin_unlock_bh(lock);
1899
1900         if (!err)
1901                 err = put_user(next_id, &uattr->next_id);
1902
1903         return err;
1904 }
1905
1906 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1907
1908 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1909 {
1910         struct bpf_prog *prog;
1911         u32 id = attr->prog_id;
1912         int fd;
1913
1914         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1915                 return -EINVAL;
1916
1917         if (!capable(CAP_SYS_ADMIN))
1918                 return -EPERM;
1919
1920         spin_lock_bh(&prog_idr_lock);
1921         prog = idr_find(&prog_idr, id);
1922         if (prog)
1923                 prog = bpf_prog_inc_not_zero(prog);
1924         else
1925                 prog = ERR_PTR(-ENOENT);
1926         spin_unlock_bh(&prog_idr_lock);
1927
1928         if (IS_ERR(prog))
1929                 return PTR_ERR(prog);
1930
1931         fd = bpf_prog_new_fd(prog);
1932         if (fd < 0)
1933                 bpf_prog_put(prog);
1934
1935         return fd;
1936 }
1937
1938 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1939
1940 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1941 {
1942         struct bpf_map *map;
1943         u32 id = attr->map_id;
1944         int f_flags;
1945         int fd;
1946
1947         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1948             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1949                 return -EINVAL;
1950
1951         if (!capable(CAP_SYS_ADMIN))
1952                 return -EPERM;
1953
1954         f_flags = bpf_get_file_flag(attr->open_flags);
1955         if (f_flags < 0)
1956                 return f_flags;
1957
1958         spin_lock_bh(&map_idr_lock);
1959         map = idr_find(&map_idr, id);
1960         if (map)
1961                 map = bpf_map_inc_not_zero(map, true);
1962         else
1963                 map = ERR_PTR(-ENOENT);
1964         spin_unlock_bh(&map_idr_lock);
1965
1966         if (IS_ERR(map))
1967                 return PTR_ERR(map);
1968
1969         fd = bpf_map_new_fd(map, f_flags);
1970         if (fd < 0)
1971                 bpf_map_put(map);
1972
1973         return fd;
1974 }
1975
1976 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1977                                               unsigned long addr)
1978 {
1979         int i;
1980
1981         for (i = 0; i < prog->aux->used_map_cnt; i++)
1982                 if (prog->aux->used_maps[i] == (void *)addr)
1983                         return prog->aux->used_maps[i];
1984         return NULL;
1985 }
1986
1987 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1988 {
1989         const struct bpf_map *map;
1990         struct bpf_insn *insns;
1991         u64 imm;
1992         int i;
1993
1994         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1995                         GFP_USER);
1996         if (!insns)
1997                 return insns;
1998
1999         for (i = 0; i < prog->len; i++) {
2000                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2001                         insns[i].code = BPF_JMP | BPF_CALL;
2002                         insns[i].imm = BPF_FUNC_tail_call;
2003                         /* fall-through */
2004                 }
2005                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2006                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2007                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2008                                 insns[i].code = BPF_JMP | BPF_CALL;
2009                         if (!bpf_dump_raw_ok())
2010                                 insns[i].imm = 0;
2011                         continue;
2012                 }
2013
2014                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2015                         continue;
2016
2017                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2018                 map = bpf_map_from_imm(prog, imm);
2019                 if (map) {
2020                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2021                         insns[i].imm = map->id;
2022                         insns[i + 1].imm = 0;
2023                         continue;
2024                 }
2025
2026                 if (!bpf_dump_raw_ok() &&
2027                     imm == (unsigned long)prog->aux) {
2028                         insns[i].imm = 0;
2029                         insns[i + 1].imm = 0;
2030                         continue;
2031                 }
2032         }
2033
2034         return insns;
2035 }
2036
2037 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2038                                    const union bpf_attr *attr,
2039                                    union bpf_attr __user *uattr)
2040 {
2041         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2042         struct bpf_prog_info info = {};
2043         u32 info_len = attr->info.info_len;
2044         char __user *uinsns;
2045         u32 ulen;
2046         int err;
2047
2048         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2049         if (err)
2050                 return err;
2051         info_len = min_t(u32, sizeof(info), info_len);
2052
2053         if (copy_from_user(&info, uinfo, info_len))
2054                 return -EFAULT;
2055
2056         info.type = prog->type;
2057         info.id = prog->aux->id;
2058         info.load_time = prog->aux->load_time;
2059         info.created_by_uid = from_kuid_munged(current_user_ns(),
2060                                                prog->aux->user->uid);
2061         info.gpl_compatible = prog->gpl_compatible;
2062
2063         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2064         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2065
2066         ulen = info.nr_map_ids;
2067         info.nr_map_ids = prog->aux->used_map_cnt;
2068         ulen = min_t(u32, info.nr_map_ids, ulen);
2069         if (ulen) {
2070                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2071                 u32 i;
2072
2073                 for (i = 0; i < ulen; i++)
2074                         if (put_user(prog->aux->used_maps[i]->id,
2075                                      &user_map_ids[i]))
2076                                 return -EFAULT;
2077         }
2078
2079         if (!capable(CAP_SYS_ADMIN)) {
2080                 info.jited_prog_len = 0;
2081                 info.xlated_prog_len = 0;
2082                 info.nr_jited_ksyms = 0;
2083                 info.nr_jited_func_lens = 0;
2084                 info.func_info_cnt = 0;
2085                 goto done;
2086         }
2087
2088         ulen = info.xlated_prog_len;
2089         info.xlated_prog_len = bpf_prog_insn_size(prog);
2090         if (info.xlated_prog_len && ulen) {
2091                 struct bpf_insn *insns_sanitized;
2092                 bool fault;
2093
2094                 if (prog->blinded && !bpf_dump_raw_ok()) {
2095                         info.xlated_prog_insns = 0;
2096                         goto done;
2097                 }
2098                 insns_sanitized = bpf_insn_prepare_dump(prog);
2099                 if (!insns_sanitized)
2100                         return -ENOMEM;
2101                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2102                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2103                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2104                 kfree(insns_sanitized);
2105                 if (fault)
2106                         return -EFAULT;
2107         }
2108
2109         if (bpf_prog_is_dev_bound(prog->aux)) {
2110                 err = bpf_prog_offload_info_fill(&info, prog);
2111                 if (err)
2112                         return err;
2113                 goto done;
2114         }
2115
2116         /* NOTE: the following code is supposed to be skipped for offload.
2117          * bpf_prog_offload_info_fill() is the place to fill similar fields
2118          * for offload.
2119          */
2120         ulen = info.jited_prog_len;
2121         if (prog->aux->func_cnt) {
2122                 u32 i;
2123
2124                 info.jited_prog_len = 0;
2125                 for (i = 0; i < prog->aux->func_cnt; i++)
2126                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2127         } else {
2128                 info.jited_prog_len = prog->jited_len;
2129         }
2130
2131         if (info.jited_prog_len && ulen) {
2132                 if (bpf_dump_raw_ok()) {
2133                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2134                         ulen = min_t(u32, info.jited_prog_len, ulen);
2135
2136                         /* for multi-function programs, copy the JITed
2137                          * instructions for all the functions
2138                          */
2139                         if (prog->aux->func_cnt) {
2140                                 u32 len, free, i;
2141                                 u8 *img;
2142
2143                                 free = ulen;
2144                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2145                                         len = prog->aux->func[i]->jited_len;
2146                                         len = min_t(u32, len, free);
2147                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2148                                         if (copy_to_user(uinsns, img, len))
2149                                                 return -EFAULT;
2150                                         uinsns += len;
2151                                         free -= len;
2152                                         if (!free)
2153                                                 break;
2154                                 }
2155                         } else {
2156                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2157                                         return -EFAULT;
2158                         }
2159                 } else {
2160                         info.jited_prog_insns = 0;
2161                 }
2162         }
2163
2164         ulen = info.nr_jited_ksyms;
2165         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2166         if (info.nr_jited_ksyms && ulen) {
2167                 if (bpf_dump_raw_ok()) {
2168                         unsigned long ksym_addr;
2169                         u64 __user *user_ksyms;
2170                         u32 i;
2171
2172                         /* copy the address of the kernel symbol
2173                          * corresponding to each function
2174                          */
2175                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2176                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2177                         if (prog->aux->func_cnt) {
2178                                 for (i = 0; i < ulen; i++) {
2179                                         ksym_addr = (unsigned long)
2180                                                 prog->aux->func[i]->bpf_func;
2181                                         if (put_user((u64) ksym_addr,
2182                                                      &user_ksyms[i]))
2183                                                 return -EFAULT;
2184                                 }
2185                         } else {
2186                                 ksym_addr = (unsigned long) prog->bpf_func;
2187                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2188                                         return -EFAULT;
2189                         }
2190                 } else {
2191                         info.jited_ksyms = 0;
2192                 }
2193         }
2194
2195         ulen = info.nr_jited_func_lens;
2196         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2197         if (info.nr_jited_func_lens && ulen) {
2198                 if (bpf_dump_raw_ok()) {
2199                         u32 __user *user_lens;
2200                         u32 func_len, i;
2201
2202                         /* copy the JITed image lengths for each function */
2203                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2204                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2205                         if (prog->aux->func_cnt) {
2206                                 for (i = 0; i < ulen; i++) {
2207                                         func_len =
2208                                                 prog->aux->func[i]->jited_len;
2209                                         if (put_user(func_len, &user_lens[i]))
2210                                                 return -EFAULT;
2211                                 }
2212                         } else {
2213                                 func_len = prog->jited_len;
2214                                 if (put_user(func_len, &user_lens[0]))
2215                                         return -EFAULT;
2216                         }
2217                 } else {
2218                         info.jited_func_lens = 0;
2219                 }
2220         }
2221
2222         if (prog->aux->btf) {
2223                 u32 krec_size = sizeof(struct bpf_func_info);
2224                 u32 ucnt, urec_size;
2225
2226                 info.btf_id = btf_id(prog->aux->btf);
2227
2228                 ucnt = info.func_info_cnt;
2229                 info.func_info_cnt = prog->aux->func_info_cnt;
2230                 urec_size = info.func_info_rec_size;
2231                 info.func_info_rec_size = krec_size;
2232                 if (ucnt) {
2233                         /* expect passed-in urec_size is what the kernel expects */
2234                         if (urec_size != info.func_info_rec_size)
2235                                 return -EINVAL;
2236
2237                         if (bpf_dump_raw_ok()) {
2238                                 char __user *user_finfo;
2239
2240                                 user_finfo = u64_to_user_ptr(info.func_info);
2241                                 ucnt = min_t(u32, info.func_info_cnt, ucnt);
2242                                 if (copy_to_user(user_finfo, prog->aux->func_info,
2243                                                  krec_size * ucnt))
2244                                         return -EFAULT;
2245                         } else {
2246                                 info.func_info_cnt = 0;
2247                         }
2248                 }
2249         } else {
2250                 info.func_info_cnt = 0;
2251         }
2252
2253 done:
2254         if (copy_to_user(uinfo, &info, info_len) ||
2255             put_user(info_len, &uattr->info.info_len))
2256                 return -EFAULT;
2257
2258         return 0;
2259 }
2260
2261 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2262                                   const union bpf_attr *attr,
2263                                   union bpf_attr __user *uattr)
2264 {
2265         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2266         struct bpf_map_info info = {};
2267         u32 info_len = attr->info.info_len;
2268         int err;
2269
2270         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2271         if (err)
2272                 return err;
2273         info_len = min_t(u32, sizeof(info), info_len);
2274
2275         info.type = map->map_type;
2276         info.id = map->id;
2277         info.key_size = map->key_size;
2278         info.value_size = map->value_size;
2279         info.max_entries = map->max_entries;
2280         info.map_flags = map->map_flags;
2281         memcpy(info.name, map->name, sizeof(map->name));
2282
2283         if (map->btf) {
2284                 info.btf_id = btf_id(map->btf);
2285                 info.btf_key_type_id = map->btf_key_type_id;
2286                 info.btf_value_type_id = map->btf_value_type_id;
2287         }
2288
2289         if (bpf_map_is_dev_bound(map)) {
2290                 err = bpf_map_offload_info_fill(&info, map);
2291                 if (err)
2292                         return err;
2293         }
2294
2295         if (copy_to_user(uinfo, &info, info_len) ||
2296             put_user(info_len, &uattr->info.info_len))
2297                 return -EFAULT;
2298
2299         return 0;
2300 }
2301
2302 static int bpf_btf_get_info_by_fd(struct btf *btf,
2303                                   const union bpf_attr *attr,
2304                                   union bpf_attr __user *uattr)
2305 {
2306         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2307         u32 info_len = attr->info.info_len;
2308         int err;
2309
2310         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2311         if (err)
2312                 return err;
2313
2314         return btf_get_info_by_fd(btf, attr, uattr);
2315 }
2316
2317 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2318
2319 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2320                                   union bpf_attr __user *uattr)
2321 {
2322         int ufd = attr->info.bpf_fd;
2323         struct fd f;
2324         int err;
2325
2326         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2327                 return -EINVAL;
2328
2329         f = fdget(ufd);
2330         if (!f.file)
2331                 return -EBADFD;
2332
2333         if (f.file->f_op == &bpf_prog_fops)
2334                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2335                                               uattr);
2336         else if (f.file->f_op == &bpf_map_fops)
2337                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2338                                              uattr);
2339         else if (f.file->f_op == &btf_fops)
2340                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2341         else
2342                 err = -EINVAL;
2343
2344         fdput(f);
2345         return err;
2346 }
2347
2348 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2349
2350 static int bpf_btf_load(const union bpf_attr *attr)
2351 {
2352         if (CHECK_ATTR(BPF_BTF_LOAD))
2353                 return -EINVAL;
2354
2355         if (!capable(CAP_SYS_ADMIN))
2356                 return -EPERM;
2357
2358         return btf_new_fd(attr);
2359 }
2360
2361 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2362
2363 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2364 {
2365         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2366                 return -EINVAL;
2367
2368         if (!capable(CAP_SYS_ADMIN))
2369                 return -EPERM;
2370
2371         return btf_get_fd_by_id(attr->btf_id);
2372 }
2373
2374 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2375                                     union bpf_attr __user *uattr,
2376                                     u32 prog_id, u32 fd_type,
2377                                     const char *buf, u64 probe_offset,
2378                                     u64 probe_addr)
2379 {
2380         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2381         u32 len = buf ? strlen(buf) : 0, input_len;
2382         int err = 0;
2383
2384         if (put_user(len, &uattr->task_fd_query.buf_len))
2385                 return -EFAULT;
2386         input_len = attr->task_fd_query.buf_len;
2387         if (input_len && ubuf) {
2388                 if (!len) {
2389                         /* nothing to copy, just make ubuf NULL terminated */
2390                         char zero = '\0';
2391
2392                         if (put_user(zero, ubuf))
2393                                 return -EFAULT;
2394                 } else if (input_len >= len + 1) {
2395                         /* ubuf can hold the string with NULL terminator */
2396                         if (copy_to_user(ubuf, buf, len + 1))
2397                                 return -EFAULT;
2398                 } else {
2399                         /* ubuf cannot hold the string with NULL terminator,
2400                          * do a partial copy with NULL terminator.
2401                          */
2402                         char zero = '\0';
2403
2404                         err = -ENOSPC;
2405                         if (copy_to_user(ubuf, buf, input_len - 1))
2406                                 return -EFAULT;
2407                         if (put_user(zero, ubuf + input_len - 1))
2408                                 return -EFAULT;
2409                 }
2410         }
2411
2412         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2413             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2414             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2415             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2416                 return -EFAULT;
2417
2418         return err;
2419 }
2420
2421 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2422
2423 static int bpf_task_fd_query(const union bpf_attr *attr,
2424                              union bpf_attr __user *uattr)
2425 {
2426         pid_t pid = attr->task_fd_query.pid;
2427         u32 fd = attr->task_fd_query.fd;
2428         const struct perf_event *event;
2429         struct files_struct *files;
2430         struct task_struct *task;
2431         struct file *file;
2432         int err;
2433
2434         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2435                 return -EINVAL;
2436
2437         if (!capable(CAP_SYS_ADMIN))
2438                 return -EPERM;
2439
2440         if (attr->task_fd_query.flags != 0)
2441                 return -EINVAL;
2442
2443         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2444         if (!task)
2445                 return -ENOENT;
2446
2447         files = get_files_struct(task);
2448         put_task_struct(task);
2449         if (!files)
2450                 return -ENOENT;
2451
2452         err = 0;
2453         spin_lock(&files->file_lock);
2454         file = fcheck_files(files, fd);
2455         if (!file)
2456                 err = -EBADF;
2457         else
2458                 get_file(file);
2459         spin_unlock(&files->file_lock);
2460         put_files_struct(files);
2461
2462         if (err)
2463                 goto out;
2464
2465         if (file->f_op == &bpf_raw_tp_fops) {
2466                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2467                 struct bpf_raw_event_map *btp = raw_tp->btp;
2468
2469                 err = bpf_task_fd_query_copy(attr, uattr,
2470                                              raw_tp->prog->aux->id,
2471                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2472                                              btp->tp->name, 0, 0);
2473                 goto put_file;
2474         }
2475
2476         event = perf_get_event(file);
2477         if (!IS_ERR(event)) {
2478                 u64 probe_offset, probe_addr;
2479                 u32 prog_id, fd_type;
2480                 const char *buf;
2481
2482                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2483                                               &buf, &probe_offset,
2484                                               &probe_addr);
2485                 if (!err)
2486                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2487                                                      fd_type, buf,
2488                                                      probe_offset,
2489                                                      probe_addr);
2490                 goto put_file;
2491         }
2492
2493         err = -ENOTSUPP;
2494 put_file:
2495         fput(file);
2496 out:
2497         return err;
2498 }
2499
2500 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2501 {
2502         union bpf_attr attr = {};
2503         int err;
2504
2505         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2506                 return -EPERM;
2507
2508         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2509         if (err)
2510                 return err;
2511         size = min_t(u32, size, sizeof(attr));
2512
2513         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2514         if (copy_from_user(&attr, uattr, size) != 0)
2515                 return -EFAULT;
2516
2517         err = security_bpf(cmd, &attr, size);
2518         if (err < 0)
2519                 return err;
2520
2521         switch (cmd) {
2522         case BPF_MAP_CREATE:
2523                 err = map_create(&attr);
2524                 break;
2525         case BPF_MAP_LOOKUP_ELEM:
2526                 err = map_lookup_elem(&attr);
2527                 break;
2528         case BPF_MAP_UPDATE_ELEM:
2529                 err = map_update_elem(&attr);
2530                 break;
2531         case BPF_MAP_DELETE_ELEM:
2532                 err = map_delete_elem(&attr);
2533                 break;
2534         case BPF_MAP_GET_NEXT_KEY:
2535                 err = map_get_next_key(&attr);
2536                 break;
2537         case BPF_PROG_LOAD:
2538                 err = bpf_prog_load(&attr, uattr);
2539                 break;
2540         case BPF_OBJ_PIN:
2541                 err = bpf_obj_pin(&attr);
2542                 break;
2543         case BPF_OBJ_GET:
2544                 err = bpf_obj_get(&attr);
2545                 break;
2546         case BPF_PROG_ATTACH:
2547                 err = bpf_prog_attach(&attr);
2548                 break;
2549         case BPF_PROG_DETACH:
2550                 err = bpf_prog_detach(&attr);
2551                 break;
2552         case BPF_PROG_QUERY:
2553                 err = bpf_prog_query(&attr, uattr);
2554                 break;
2555         case BPF_PROG_TEST_RUN:
2556                 err = bpf_prog_test_run(&attr, uattr);
2557                 break;
2558         case BPF_PROG_GET_NEXT_ID:
2559                 err = bpf_obj_get_next_id(&attr, uattr,
2560                                           &prog_idr, &prog_idr_lock);
2561                 break;
2562         case BPF_MAP_GET_NEXT_ID:
2563                 err = bpf_obj_get_next_id(&attr, uattr,
2564                                           &map_idr, &map_idr_lock);
2565                 break;
2566         case BPF_PROG_GET_FD_BY_ID:
2567                 err = bpf_prog_get_fd_by_id(&attr);
2568                 break;
2569         case BPF_MAP_GET_FD_BY_ID:
2570                 err = bpf_map_get_fd_by_id(&attr);
2571                 break;
2572         case BPF_OBJ_GET_INFO_BY_FD:
2573                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2574                 break;
2575         case BPF_RAW_TRACEPOINT_OPEN:
2576                 err = bpf_raw_tracepoint_open(&attr);
2577                 break;
2578         case BPF_BTF_LOAD:
2579                 err = bpf_btf_load(&attr);
2580                 break;
2581         case BPF_BTF_GET_FD_BY_ID:
2582                 err = bpf_btf_get_fd_by_id(&attr);
2583                 break;
2584         case BPF_TASK_FD_QUERY:
2585                 err = bpf_task_fd_query(&attr, uattr);
2586                 break;
2587         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2588                 err = map_lookup_and_delete_elem(&attr);
2589                 break;
2590         default:
2591                 err = -EINVAL;
2592                 break;
2593         }
2594
2595         return err;
2596 }