kernel/bpf/syscall.c

   1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  *
   7  * This program is distributed in the hope that it will be useful, but
   8  * WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  10  * General Public License for more details.
  11  */
  12 #include <linux/bpf.h>
  13 #include <linux/bpf_trace.h>
  14 #include <linux/bpf_lirc.h>
  15 #include <linux/btf.h>
  16 #include <linux/syscalls.h>
  17 #include <linux/slab.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/mmzone.h>
  21 #include <linux/anon_inodes.h>
  22 #include <linux/fdtable.h>
  23 #include <linux/file.h>
  24 #include <linux/fs.h>
  25 #include <linux/license.h>
  26 #include <linux/filter.h>
  27 #include <linux/version.h>
  28 #include <linux/kernel.h>
  29 #include <linux/idr.h>
  30 #include <linux/cred.h>
  31 #include <linux/timekeeping.h>
  32 #include <linux/ctype.h>
  33 #include <linux/nospec.h>
  34
  35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  36                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  37                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  38                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  41
  42 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  43
  44 DEFINE_PER_CPU(int, bpf_prog_active);
  45 static DEFINE_IDR(prog_idr);
  46 static DEFINE_SPINLOCK(prog_idr_lock);
  47 static DEFINE_IDR(map_idr);
  48 static DEFINE_SPINLOCK(map_idr_lock);
  49
  50 int sysctl_unprivileged_bpf_disabled __read_mostly;
  51
  52 static const struct bpf_map_ops * const bpf_map_types[] = {
  53 #define BPF_PROG_TYPE(_id, _ops)
  54 #define BPF_MAP_TYPE(_id, _ops) \
  55         [_id] = &_ops,
  56 #include <linux/bpf_types.h>
  57 #undef BPF_PROG_TYPE
  58 #undef BPF_MAP_TYPE
  59 };
  60
  61 /*
  62  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  63  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  64  * we don't know about yet.
  65  *
  66  * There is a ToCToU between this function call and the following
  67  * copy_from_user() call. However, this is not a concern since this function is
  68  * meant to be a future-proofing of bits.
  69  */
  70 int bpf_check_uarg_tail_zero(void __user *uaddr,
  71                              size_t expected_size,
  72                              size_t actual_size)
  73 {
  74         unsigned char __user *addr;
  75         unsigned char __user *end;
  76         unsigned char val;
  77         int err;
  78
  79         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  80                 return -E2BIG;
  81
  82         if (unlikely(!access_ok(uaddr, actual_size)))
  83                 return -EFAULT;
  84
  85         if (actual_size <= expected_size)
  86                 return 0;
  87
  88         addr = uaddr + expected_size;
  89         end  = uaddr + actual_size;
  90
  91         for (; addr < end; addr++) {
  92                 err = get_user(val, addr);
  93                 if (err)
  94                         return err;
  95                 if (val)
  96                         return -E2BIG;
  97         }
  98
  99         return 0;
 100 }
 101
 102 const struct bpf_map_ops bpf_map_offload_ops = {
 103         .map_alloc = bpf_map_offload_map_alloc,
 104         .map_free = bpf_map_offload_map_free,
 105         .map_check_btf = map_check_no_btf,
 106 };
 107
 108 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 109 {
 110         const struct bpf_map_ops *ops;
 111         u32 type = attr->map_type;
 112         struct bpf_map *map;
 113         int err;
 114
 115         if (type >= ARRAY_SIZE(bpf_map_types))
 116                 return ERR_PTR(-EINVAL);
 117         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 118         ops = bpf_map_types[type];
 119         if (!ops)
 120                 return ERR_PTR(-EINVAL);
 121
 122         if (ops->map_alloc_check) {
 123                 err = ops->map_alloc_check(attr);
 124                 if (err)
 125                         return ERR_PTR(err);
 126         }
 127         if (attr->map_ifindex)
 128                 ops = &bpf_map_offload_ops;
 129         map = ops->map_alloc(attr);
 130         if (IS_ERR(map))
 131                 return map;
 132         map->ops = ops;
 133         map->map_type = type;
 134         return map;
 135 }
 136
 137 void *bpf_map_area_alloc(size_t size, int numa_node)
 138 {
 139         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
 140          * trigger under memory pressure as we really just want to
 141          * fail instead.
 142          */
 143         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
 144         void *area;
 145
 146         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 147                 area = kmalloc_node(size, GFP_USER | flags, numa_node);
 148                 if (area != NULL)
 149                         return area;
 150         }
 151
 152         return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
 153                                            __builtin_return_address(0));
 154 }
 155
 156 void bpf_map_area_free(void *area)
 157 {
 158         kvfree(area);
 159 }
 160
 161 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 162 {
 163         map->map_type = attr->map_type;
 164         map->key_size = attr->key_size;
 165         map->value_size = attr->value_size;
 166         map->max_entries = attr->max_entries;
 167         map->map_flags = attr->map_flags;
 168         map->numa_node = bpf_map_attr_numa_node(attr);
 169 }
 170
 171 int bpf_map_precharge_memlock(u32 pages)
 172 {
 173         struct user_struct *user = get_current_user();
 174         unsigned long memlock_limit, cur;
 175
 176         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 177         cur = atomic_long_read(&user->locked_vm);
 178         free_uid(user);
 179         if (cur + pages > memlock_limit)
 180                 return -EPERM;
 181         return 0;
 182 }
 183
 184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 185 {
 186         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 187
 188         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 189                 atomic_long_sub(pages, &user->locked_vm);
 190                 return -EPERM;
 191         }
 192         return 0;
 193 }
 194
 195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 196 {
 197         atomic_long_sub(pages, &user->locked_vm);
 198 }
 199
 200 static int bpf_map_init_memlock(struct bpf_map *map)
 201 {
 202         struct user_struct *user = get_current_user();
 203         int ret;
 204
 205         ret = bpf_charge_memlock(user, map->pages);
 206         if (ret) {
 207                 free_uid(user);
 208                 return ret;
 209         }
 210         map->user = user;
 211         return ret;
 212 }
 213
 214 static void bpf_map_release_memlock(struct bpf_map *map)
 215 {
 216         struct user_struct *user = map->user;
 217         bpf_uncharge_memlock(user, map->pages);
 218         free_uid(user);
 219 }
 220
 221 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 222 {
 223         int ret;
 224
 225         ret = bpf_charge_memlock(map->user, pages);
 226         if (ret)
 227                 return ret;
 228         map->pages += pages;
 229         return ret;
 230 }
 231
 232 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 233 {
 234         bpf_uncharge_memlock(map->user, pages);
 235         map->pages -= pages;
 236 }
 237
 238 static int bpf_map_alloc_id(struct bpf_map *map)
 239 {
 240         int id;
 241
 242         idr_preload(GFP_KERNEL);
 243         spin_lock_bh(&map_idr_lock);
 244         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 245         if (id > 0)
 246                 map->id = id;
 247         spin_unlock_bh(&map_idr_lock);
 248         idr_preload_end();
 249
 250         if (WARN_ON_ONCE(!id))
 251                 return -ENOSPC;
 252
 253         return id > 0 ? 0 : id;
 254 }
 255
 256 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 257 {
 258         unsigned long flags;
 259
 260         /* Offloaded maps are removed from the IDR store when their device
 261          * disappears - even if someone holds an fd to them they are unusable,
 262          * the memory is gone, all ops will fail; they are simply waiting for
 263          * refcnt to drop to be freed.
 264          */
 265         if (!map->id)
 266                 return;
 267
 268         if (do_idr_lock)
 269                 spin_lock_irqsave(&map_idr_lock, flags);
 270         else
 271                 __acquire(&map_idr_lock);
 272
 273         idr_remove(&map_idr, map->id);
 274         map->id = 0;
 275
 276         if (do_idr_lock)
 277                 spin_unlock_irqrestore(&map_idr_lock, flags);
 278         else
 279                 __release(&map_idr_lock);
 280 }
 281
 282 /* called from workqueue */
 283 static void bpf_map_free_deferred(struct work_struct *work)
 284 {
 285         struct bpf_map *map = container_of(work, struct bpf_map, work);
 286
 287         bpf_map_release_memlock(map);
 288         security_bpf_map_free(map);
 289         /* implementation dependent freeing */
 290         map->ops->map_free(map);
 291 }
 292
 293 static void bpf_map_put_uref(struct bpf_map *map)
 294 {
 295         if (atomic_dec_and_test(&map->usercnt)) {
 296                 if (map->ops->map_release_uref)
 297                         map->ops->map_release_uref(map);
 298         }
 299 }
 300
 301 /* decrement map refcnt and schedule it for freeing via workqueue
 302  * (unrelying map implementation ops->map_free() might sleep)
 303  */
 304 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 305 {
 306         if (atomic_dec_and_test(&map->refcnt)) {
 307                 /* bpf_map_free_id() must be called first */
 308                 bpf_map_free_id(map, do_idr_lock);
 309                 btf_put(map->btf);
 310                 INIT_WORK(&map->work, bpf_map_free_deferred);
 311                 schedule_work(&map->work);
 312         }
 313 }
 314
 315 void bpf_map_put(struct bpf_map *map)
 316 {
 317         __bpf_map_put(map, true);
 318 }
 319 EXPORT_SYMBOL_GPL(bpf_map_put);
 320
 321 void bpf_map_put_with_uref(struct bpf_map *map)
 322 {
 323         bpf_map_put_uref(map);
 324         bpf_map_put(map);
 325 }
 326
 327 static int bpf_map_release(struct inode *inode, struct file *filp)
 328 {
 329         struct bpf_map *map = filp->private_data;
 330
 331         if (map->ops->map_release)
 332                 map->ops->map_release(map, filp);
 333
 334         bpf_map_put_with_uref(map);
 335         return 0;
 336 }
 337
 338 #ifdef CONFIG_PROC_FS
 339 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 340 {
 341         const struct bpf_map *map = filp->private_data;
 342         const struct bpf_array *array;
 343         u32 owner_prog_type = 0;
 344         u32 owner_jited = 0;
 345
 346         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 347                 array = container_of(map, struct bpf_array, map);
 348                 owner_prog_type = array->owner_prog_type;
 349                 owner_jited = array->owner_jited;
 350         }
 351
 352         seq_printf(m,
 353                    "map_type:\t%u\n"
 354                    "key_size:\t%u\n"
 355                    "value_size:\t%u\n"
 356                    "max_entries:\t%u\n"
 357                    "map_flags:\t%#x\n"
 358                    "memlock:\t%llu\n"
 359                    "map_id:\t%u\n",
 360                    map->map_type,
 361                    map->key_size,
 362                    map->value_size,
 363                    map->max_entries,
 364                    map->map_flags,
 365                    map->pages * 1ULL << PAGE_SHIFT,
 366                    map->id);
 367
 368         if (owner_prog_type) {
 369                 seq_printf(m, "owner_prog_type:\t%u\n",
 370                            owner_prog_type);
 371                 seq_printf(m, "owner_jited:\t%u\n",
 372                            owner_jited);
 373         }
 374 }
 375 #endif
 376
 377 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 378                               loff_t *ppos)
 379 {
 380         /* We need this handler such that alloc_file() enables
 381          * f_mode with FMODE_CAN_READ.
 382          */
 383         return -EINVAL;
 384 }
 385
 386 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 387                                size_t siz, loff_t *ppos)
 388 {
 389         /* We need this handler such that alloc_file() enables
 390          * f_mode with FMODE_CAN_WRITE.
 391          */
 392         return -EINVAL;
 393 }
 394
 395 const struct file_operations bpf_map_fops = {
 396 #ifdef CONFIG_PROC_FS
 397         .show_fdinfo    = bpf_map_show_fdinfo,
 398 #endif
 399         .release        = bpf_map_release,
 400         .read           = bpf_dummy_read,
 401         .write          = bpf_dummy_write,
 402 };
 403
 404 int bpf_map_new_fd(struct bpf_map *map, int flags)
 405 {
 406         int ret;
 407
 408         ret = security_bpf_map(map, OPEN_FMODE(flags));
 409         if (ret < 0)
 410                 return ret;
 411
 412         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 413                                 flags | O_CLOEXEC);
 414 }
 415
 416 int bpf_get_file_flag(int flags)
 417 {
 418         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 419                 return -EINVAL;
 420         if (flags & BPF_F_RDONLY)
 421                 return O_RDONLY;
 422         if (flags & BPF_F_WRONLY)
 423                 return O_WRONLY;
 424         return O_RDWR;
 425 }
 426
 427 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 428 #define CHECK_ATTR(CMD) \
 429         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 430                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 431                    sizeof(*attr) - \
 432                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 433                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 434
 435 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 436  * Return 0 on success and < 0 on error.
 437  */
 438 static int bpf_obj_name_cpy(char *dst, const char *src)
 439 {
 440         const char *end = src + BPF_OBJ_NAME_LEN;
 441
 442         memset(dst, 0, BPF_OBJ_NAME_LEN);
 443
 444         /* Copy all isalnum() and '_' char */
 445         while (src < end && *src) {
 446                 if (!isalnum(*src) && *src != '_')
 447                         return -EINVAL;
 448                 *dst++ = *src++;
 449         }
 450
 451         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 452         if (src == end)
 453                 return -EINVAL;
 454
 455         return 0;
 456 }
 457
 458 int map_check_no_btf(const struct bpf_map *map,
 459                      const struct btf *btf,
 460                      const struct btf_type *key_type,
 461                      const struct btf_type *value_type)
 462 {
 463         return -ENOTSUPP;
 464 }
 465
 466 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 467                          u32 btf_key_id, u32 btf_value_id)
 468 {
 469         const struct btf_type *key_type, *value_type;
 470         u32 key_size, value_size;
 471         int ret = 0;
 472
 473         key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 474         if (!key_type || key_size != map->key_size)
 475                 return -EINVAL;
 476
 477         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 478         if (!value_type || value_size != map->value_size)
 479                 return -EINVAL;
 480
 481         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 482
 483         if (map_value_has_spin_lock(map)) {
 484                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 485                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 486                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
 487                         return -ENOTSUPP;
 488                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 489                     map->value_size) {
 490                         WARN_ONCE(1,
 491                                   "verifier bug spin_lock_off %d value_size %d\n",
 492                                   map->spin_lock_off, map->value_size);
 493                         return -EFAULT;
 494                 }
 495         }
 496
 497         if (map->ops->map_check_btf)
 498                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 499
 500         return ret;
 501 }
 502
 503 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 504 /* called via syscall */
 505 static int map_create(union bpf_attr *attr)
 506 {
 507         int numa_node = bpf_map_attr_numa_node(attr);
 508         struct bpf_map *map;
 509         int f_flags;
 510         int err;
 511
 512         err = CHECK_ATTR(BPF_MAP_CREATE);
 513         if (err)
 514                 return -EINVAL;
 515
 516         f_flags = bpf_get_file_flag(attr->map_flags);
 517         if (f_flags < 0)
 518                 return f_flags;
 519
 520         if (numa_node != NUMA_NO_NODE &&
 521             ((unsigned int)numa_node >= nr_node_ids ||
 522              !node_online(numa_node)))
 523                 return -EINVAL;
 524
 525         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 526         map = find_and_alloc_map(attr);
 527         if (IS_ERR(map))
 528                 return PTR_ERR(map);
 529
 530         err = bpf_obj_name_cpy(map->name, attr->map_name);
 531         if (err)
 532                 goto free_map_nouncharge;
 533
 534         atomic_set(&map->refcnt, 1);
 535         atomic_set(&map->usercnt, 1);
 536
 537         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 538                 struct btf *btf;
 539
 540                 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
 541                         err = -EINVAL;
 542                         goto free_map_nouncharge;
 543                 }
 544
 545                 btf = btf_get_by_fd(attr->btf_fd);
 546                 if (IS_ERR(btf)) {
 547                         err = PTR_ERR(btf);
 548                         goto free_map_nouncharge;
 549                 }
 550
 551                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 552                                     attr->btf_value_type_id);
 553                 if (err) {
 554                         btf_put(btf);
 555                         goto free_map_nouncharge;
 556                 }
 557
 558                 map->btf = btf;
 559                 map->btf_key_type_id = attr->btf_key_type_id;
 560                 map->btf_value_type_id = attr->btf_value_type_id;
 561         } else {
 562                 map->spin_lock_off = -EINVAL;
 563         }
 564
 565         err = security_bpf_map_alloc(map);
 566         if (err)
 567                 goto free_map_nouncharge;
 568
 569         err = bpf_map_init_memlock(map);
 570         if (err)
 571                 goto free_map_sec;
 572
 573         err = bpf_map_alloc_id(map);
 574         if (err)
 575                 goto free_map;
 576
 577         err = bpf_map_new_fd(map, f_flags);
 578         if (err < 0) {
 579                 /* failed to allocate fd.
 580                  * bpf_map_put_with_uref() is needed because the above
 581                  * bpf_map_alloc_id() has published the map
 582                  * to the userspace and the userspace may
 583                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 584                  */
 585                 bpf_map_put_with_uref(map);
 586                 return err;
 587         }
 588
 589         return err;
 590
 591 free_map:
 592         bpf_map_release_memlock(map);
 593 free_map_sec:
 594         security_bpf_map_free(map);
 595 free_map_nouncharge:
 596         btf_put(map->btf);
 597         map->ops->map_free(map);
 598         return err;
 599 }
 600
 601 /* if error is returned, fd is released.
 602  * On success caller should complete fd access with matching fdput()
 603  */
 604 struct bpf_map *__bpf_map_get(struct fd f)
 605 {
 606         if (!f.file)
 607                 return ERR_PTR(-EBADF);
 608         if (f.file->f_op != &bpf_map_fops) {
 609                 fdput(f);
 610                 return ERR_PTR(-EINVAL);
 611         }
 612
 613         return f.file->private_data;
 614 }
 615
 616 /* prog's and map's refcnt limit */
 617 #define BPF_MAX_REFCNT 32768
 618
 619 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 620 {
 621         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 622                 atomic_dec(&map->refcnt);
 623                 return ERR_PTR(-EBUSY);
 624         }
 625         if (uref)
 626                 atomic_inc(&map->usercnt);
 627         return map;
 628 }
 629 EXPORT_SYMBOL_GPL(bpf_map_inc);
 630
 631 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 632 {
 633         struct fd f = fdget(ufd);
 634         struct bpf_map *map;
 635
 636         map = __bpf_map_get(f);
 637         if (IS_ERR(map))
 638                 return map;
 639
 640         map = bpf_map_inc(map, true);
 641         fdput(f);
 642
 643         return map;
 644 }
 645
 646 /* map_idr_lock should have been held */
 647 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 648                                             bool uref)
 649 {
 650         int refold;
 651
 652         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 653
 654         if (refold >= BPF_MAX_REFCNT) {
 655                 __bpf_map_put(map, false);
 656                 return ERR_PTR(-EBUSY);
 657         }
 658
 659         if (!refold)
 660                 return ERR_PTR(-ENOENT);
 661
 662         if (uref)
 663                 atomic_inc(&map->usercnt);
 664
 665         return map;
 666 }
 667
 668 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 669 {
 670         return -ENOTSUPP;
 671 }
 672
 673 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 674 {
 675         if (key_size)
 676                 return memdup_user(ukey, key_size);
 677
 678         if (ukey)
 679                 return ERR_PTR(-EINVAL);
 680
 681         return NULL;
 682 }
 683
 684 /* last field in 'union bpf_attr' used by this command */
 685 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 686
 687 static int map_lookup_elem(union bpf_attr *attr)
 688 {
 689         void __user *ukey = u64_to_user_ptr(attr->key);
 690         void __user *uvalue = u64_to_user_ptr(attr->value);
 691         int ufd = attr->map_fd;
 692         struct bpf_map *map;
 693         void *key, *value, *ptr;
 694         u32 value_size;
 695         struct fd f;
 696         int err;
 697
 698         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 699                 return -EINVAL;
 700
 701         if (attr->flags & ~BPF_F_LOCK)
 702                 return -EINVAL;
 703
 704         f = fdget(ufd);
 705         map = __bpf_map_get(f);
 706         if (IS_ERR(map))
 707                 return PTR_ERR(map);
 708
 709         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 710                 err = -EPERM;
 711                 goto err_put;
 712         }
 713
 714         if ((attr->flags & BPF_F_LOCK) &&
 715             !map_value_has_spin_lock(map)) {
 716                 err = -EINVAL;
 717                 goto err_put;
 718         }
 719
 720         key = __bpf_copy_key(ukey, map->key_size);
 721         if (IS_ERR(key)) {
 722                 err = PTR_ERR(key);
 723                 goto err_put;
 724         }
 725
 726         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 727             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 728             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 729             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 730                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 731         else if (IS_FD_MAP(map))
 732                 value_size = sizeof(u32);
 733         else
 734                 value_size = map->value_size;
 735
 736         err = -ENOMEM;
 737         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 738         if (!value)
 739                 goto free_key;
 740
 741         if (bpf_map_is_dev_bound(map)) {
 742                 err = bpf_map_offload_lookup_elem(map, key, value);
 743                 goto done;
 744         }
 745
 746         preempt_disable();
 747         this_cpu_inc(bpf_prog_active);
 748         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 749             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 750                 err = bpf_percpu_hash_copy(map, key, value);
 751         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 752                 err = bpf_percpu_array_copy(map, key, value);
 753         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 754                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 755         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 756                 err = bpf_stackmap_copy(map, key, value);
 757         } else if (IS_FD_ARRAY(map)) {
 758                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 759         } else if (IS_FD_HASH(map)) {
 760                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 761         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 762                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 763         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 764                    map->map_type == BPF_MAP_TYPE_STACK) {
 765                 err = map->ops->map_peek_elem(map, value);
 766         } else {
 767                 rcu_read_lock();
 768                 ptr = map->ops->map_lookup_elem(map, key);
 769                 if (IS_ERR(ptr)) {
 770                         err = PTR_ERR(ptr);
 771                 } else if (!ptr) {
 772                         err = -ENOENT;
 773                 } else {
 774                         err = 0;
 775                         if (attr->flags & BPF_F_LOCK)
 776                                 /* lock 'ptr' and copy everything but lock */
 777                                 copy_map_value_locked(map, value, ptr, true);
 778                         else
 779                                 copy_map_value(map, value, ptr);
 780                         /* mask lock, since value wasn't zero inited */
 781                         check_and_init_map_lock(map, value);
 782                 }
 783                 rcu_read_unlock();
 784         }
 785         this_cpu_dec(bpf_prog_active);
 786         preempt_enable();
 787
 788 done:
 789         if (err)
 790                 goto free_value;
 791
 792         err = -EFAULT;
 793         if (copy_to_user(uvalue, value, value_size) != 0)
 794                 goto free_value;
 795
 796         err = 0;
 797
 798 free_value:
 799         kfree(value);
 800 free_key:
 801         kfree(key);
 802 err_put:
 803         fdput(f);
 804         return err;
 805 }
 806
 807 static void maybe_wait_bpf_programs(struct bpf_map *map)
 808 {
 809         /* Wait for any running BPF programs to complete so that
 810          * userspace, when we return to it, knows that all programs
 811          * that could be running use the new map value.
 812          */
 813         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 814             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 815                 synchronize_rcu();
 816 }
 817
 818 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 819
 820 static int map_update_elem(union bpf_attr *attr)
 821 {
 822         void __user *ukey = u64_to_user_ptr(attr->key);
 823         void __user *uvalue = u64_to_user_ptr(attr->value);
 824         int ufd = attr->map_fd;
 825         struct bpf_map *map;
 826         void *key, *value;
 827         u32 value_size;
 828         struct fd f;
 829         int err;
 830
 831         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 832                 return -EINVAL;
 833
 834         f = fdget(ufd);
 835         map = __bpf_map_get(f);
 836         if (IS_ERR(map))
 837                 return PTR_ERR(map);
 838
 839         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 840                 err = -EPERM;
 841                 goto err_put;
 842         }
 843
 844         if ((attr->flags & BPF_F_LOCK) &&
 845             !map_value_has_spin_lock(map)) {
 846                 err = -EINVAL;
 847                 goto err_put;
 848         }
 849
 850         key = __bpf_copy_key(ukey, map->key_size);
 851         if (IS_ERR(key)) {
 852                 err = PTR_ERR(key);
 853                 goto err_put;
 854         }
 855
 856         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 857             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 858             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 859             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 860                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 861         else
 862                 value_size = map->value_size;
 863
 864         err = -ENOMEM;
 865         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 866         if (!value)
 867                 goto free_key;
 868
 869         err = -EFAULT;
 870         if (copy_from_user(value, uvalue, value_size) != 0)
 871                 goto free_value;
 872
 873         /* Need to create a kthread, thus must support schedule */
 874         if (bpf_map_is_dev_bound(map)) {
 875                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 876                 goto out;
 877         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 878                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 879                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 880                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 881                 goto out;
 882         }
 883
 884         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 885          * inside bpf map update or delete otherwise deadlocks are possible
 886          */
 887         preempt_disable();
 888         __this_cpu_inc(bpf_prog_active);
 889         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 890             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 891                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 892         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 893                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 894         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 895                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 896                                                        attr->flags);
 897         } else if (IS_FD_ARRAY(map)) {
 898                 rcu_read_lock();
 899                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 900                                                    attr->flags);
 901                 rcu_read_unlock();
 902         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 903                 rcu_read_lock();
 904                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 905                                                   attr->flags);
 906                 rcu_read_unlock();
 907         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 908                 /* rcu_read_lock() is not needed */
 909                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 910                                                          attr->flags);
 911         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 912                    map->map_type == BPF_MAP_TYPE_STACK) {
 913                 err = map->ops->map_push_elem(map, value, attr->flags);
 914         } else {
 915                 rcu_read_lock();
 916                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 917                 rcu_read_unlock();
 918         }
 919         __this_cpu_dec(bpf_prog_active);
 920         preempt_enable();
 921         maybe_wait_bpf_programs(map);
 922 out:
 923 free_value:
 924         kfree(value);
 925 free_key:
 926         kfree(key);
 927 err_put:
 928         fdput(f);
 929         return err;
 930 }
 931
 932 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 933
 934 static int map_delete_elem(union bpf_attr *attr)
 935 {
 936         void __user *ukey = u64_to_user_ptr(attr->key);
 937         int ufd = attr->map_fd;
 938         struct bpf_map *map;
 939         struct fd f;
 940         void *key;
 941         int err;
 942
 943         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 944                 return -EINVAL;
 945
 946         f = fdget(ufd);
 947         map = __bpf_map_get(f);
 948         if (IS_ERR(map))
 949                 return PTR_ERR(map);
 950
 951         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 952                 err = -EPERM;
 953                 goto err_put;
 954         }
 955
 956         key = __bpf_copy_key(ukey, map->key_size);
 957         if (IS_ERR(key)) {
 958                 err = PTR_ERR(key);
 959                 goto err_put;
 960         }
 961
 962         if (bpf_map_is_dev_bound(map)) {
 963                 err = bpf_map_offload_delete_elem(map, key);
 964                 goto out;
 965         }
 966
 967         preempt_disable();
 968         __this_cpu_inc(bpf_prog_active);
 969         rcu_read_lock();
 970         err = map->ops->map_delete_elem(map, key);
 971         rcu_read_unlock();
 972         __this_cpu_dec(bpf_prog_active);
 973         preempt_enable();
 974         maybe_wait_bpf_programs(map);
 975 out:
 976         kfree(key);
 977 err_put:
 978         fdput(f);
 979         return err;
 980 }
 981
 982 /* last field in 'union bpf_attr' used by this command */
 983 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 984
 985 static int map_get_next_key(union bpf_attr *attr)
 986 {
 987         void __user *ukey = u64_to_user_ptr(attr->key);
 988         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 989         int ufd = attr->map_fd;
 990         struct bpf_map *map;
 991         void *key, *next_key;
 992         struct fd f;
 993         int err;
 994
 995         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 996                 return -EINVAL;
 997
 998         f = fdget(ufd);
 999         map = __bpf_map_get(f);
1000         if (IS_ERR(map))
1001                 return PTR_ERR(map);
1002
1003         if (!(f.file->f_mode & FMODE_CAN_READ)) {
1004                 err = -EPERM;
1005                 goto err_put;
1006         }
1007
1008         if (ukey) {
1009                 key = __bpf_copy_key(ukey, map->key_size);
1010                 if (IS_ERR(key)) {
1011                         err = PTR_ERR(key);
1012                         goto err_put;
1013                 }
1014         } else {
1015                 key = NULL;
1016         }
1017
1018         err = -ENOMEM;
1019         next_key = kmalloc(map->key_size, GFP_USER);
1020         if (!next_key)
1021                 goto free_key;
1022
1023         if (bpf_map_is_dev_bound(map)) {
1024                 err = bpf_map_offload_get_next_key(map, key, next_key);
1025                 goto out;
1026         }
1027
1028         rcu_read_lock();
1029         err = map->ops->map_get_next_key(map, key, next_key);
1030         rcu_read_unlock();
1031 out:
1032         if (err)
1033                 goto free_next_key;
1034
1035         err = -EFAULT;
1036         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1037                 goto free_next_key;
1038
1039         err = 0;
1040
1041 free_next_key:
1042         kfree(next_key);
1043 free_key:
1044         kfree(key);
1045 err_put:
1046         fdput(f);
1047         return err;
1048 }
1049
1050 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1051
1052 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1053 {
1054         void __user *ukey = u64_to_user_ptr(attr->key);
1055         void __user *uvalue = u64_to_user_ptr(attr->value);
1056         int ufd = attr->map_fd;
1057         struct bpf_map *map;
1058         void *key, *value;
1059         u32 value_size;
1060         struct fd f;
1061         int err;
1062
1063         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1064                 return -EINVAL;
1065
1066         f = fdget(ufd);
1067         map = __bpf_map_get(f);
1068         if (IS_ERR(map))
1069                 return PTR_ERR(map);
1070
1071         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1072                 err = -EPERM;
1073                 goto err_put;
1074         }
1075
1076         key = __bpf_copy_key(ukey, map->key_size);
1077         if (IS_ERR(key)) {
1078                 err = PTR_ERR(key);
1079                 goto err_put;
1080         }
1081
1082         value_size = map->value_size;
1083
1084         err = -ENOMEM;
1085         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1086         if (!value)
1087                 goto free_key;
1088
1089         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1090             map->map_type == BPF_MAP_TYPE_STACK) {
1091                 err = map->ops->map_pop_elem(map, value);
1092         } else {
1093                 err = -ENOTSUPP;
1094         }
1095
1096         if (err)
1097                 goto free_value;
1098
1099         if (copy_to_user(uvalue, value, value_size) != 0)
1100                 goto free_value;
1101
1102         err = 0;
1103
1104 free_value:
1105         kfree(value);
1106 free_key:
1107         kfree(key);
1108 err_put:
1109         fdput(f);
1110         return err;
1111 }
1112
1113 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1114 #define BPF_PROG_TYPE(_id, _name) \
1115         [_id] = & _name ## _prog_ops,
1116 #define BPF_MAP_TYPE(_id, _ops)
1117 #include <linux/bpf_types.h>
1118 #undef BPF_PROG_TYPE
1119 #undef BPF_MAP_TYPE
1120 };
1121
1122 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1123 {
1124         const struct bpf_prog_ops *ops;
1125
1126         if (type >= ARRAY_SIZE(bpf_prog_types))
1127                 return -EINVAL;
1128         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1129         ops = bpf_prog_types[type];
1130         if (!ops)
1131                 return -EINVAL;
1132
1133         if (!bpf_prog_is_dev_bound(prog->aux))
1134                 prog->aux->ops = ops;
1135         else
1136                 prog->aux->ops = &bpf_offload_prog_ops;
1137         prog->type = type;
1138         return 0;
1139 }
1140
1141 /* drop refcnt on maps used by eBPF program and free auxilary data */
1142 static void free_used_maps(struct bpf_prog_aux *aux)
1143 {
1144         enum bpf_cgroup_storage_type stype;
1145         int i;
1146
1147         for_each_cgroup_storage_type(stype) {
1148                 if (!aux->cgroup_storage[stype])
1149                         continue;
1150                 bpf_cgroup_storage_release(aux->prog,
1151                                            aux->cgroup_storage[stype]);
1152         }
1153
1154         for (i = 0; i < aux->used_map_cnt; i++)
1155                 bpf_map_put(aux->used_maps[i]);
1156
1157         kfree(aux->used_maps);
1158 }
1159
1160 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1161 {
1162         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1163         unsigned long user_bufs;
1164
1165         if (user) {
1166                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1167                 if (user_bufs > memlock_limit) {
1168                         atomic_long_sub(pages, &user->locked_vm);
1169                         return -EPERM;
1170                 }
1171         }
1172
1173         return 0;
1174 }
1175
1176 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1177 {
1178         if (user)
1179                 atomic_long_sub(pages, &user->locked_vm);
1180 }
1181
1182 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1183 {
1184         struct user_struct *user = get_current_user();
1185         int ret;
1186
1187         ret = __bpf_prog_charge(user, prog->pages);
1188         if (ret) {
1189                 free_uid(user);
1190                 return ret;
1191         }
1192
1193         prog->aux->user = user;
1194         return 0;
1195 }
1196
1197 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1198 {
1199         struct user_struct *user = prog->aux->user;
1200
1201         __bpf_prog_uncharge(user, prog->pages);
1202         free_uid(user);
1203 }
1204
1205 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1206 {
1207         int id;
1208
1209         idr_preload(GFP_KERNEL);
1210         spin_lock_bh(&prog_idr_lock);
1211         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1212         if (id > 0)
1213                 prog->aux->id = id;
1214         spin_unlock_bh(&prog_idr_lock);
1215         idr_preload_end();
1216
1217         /* id is in [1, INT_MAX) */
1218         if (WARN_ON_ONCE(!id))
1219                 return -ENOSPC;
1220
1221         return id > 0 ? 0 : id;
1222 }
1223
1224 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1225 {
1226         /* cBPF to eBPF migrations are currently not in the idr store.
1227          * Offloaded programs are removed from the store when their device
1228          * disappears - even if someone grabs an fd to them they are unusable,
1229          * simply waiting for refcnt to drop to be freed.
1230          */
1231         if (!prog->aux->id)
1232                 return;
1233
1234         if (do_idr_lock)
1235                 spin_lock_bh(&prog_idr_lock);
1236         else
1237                 __acquire(&prog_idr_lock);
1238
1239         idr_remove(&prog_idr, prog->aux->id);
1240         prog->aux->id = 0;
1241
1242         if (do_idr_lock)
1243                 spin_unlock_bh(&prog_idr_lock);
1244         else
1245                 __release(&prog_idr_lock);
1246 }
1247
1248 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1249 {
1250         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1251
1252         free_used_maps(aux);
1253         bpf_prog_uncharge_memlock(aux->prog);
1254         security_bpf_prog_free(aux);
1255         bpf_prog_free(aux->prog);
1256 }
1257
1258 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1259 {
1260         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1261                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1262                 /* bpf_prog_free_id() must be called first */
1263                 bpf_prog_free_id(prog, do_idr_lock);
1264                 bpf_prog_kallsyms_del_all(prog);
1265                 btf_put(prog->aux->btf);
1266                 kvfree(prog->aux->func_info);
1267                 bpf_prog_free_linfo(prog);
1268
1269                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1270         }
1271 }
1272
1273 void bpf_prog_put(struct bpf_prog *prog)
1274 {
1275         __bpf_prog_put(prog, true);
1276 }
1277 EXPORT_SYMBOL_GPL(bpf_prog_put);
1278
1279 static int bpf_prog_release(struct inode *inode, struct file *filp)
1280 {
1281         struct bpf_prog *prog = filp->private_data;
1282
1283         bpf_prog_put(prog);
1284         return 0;
1285 }
1286
1287 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1288                                struct bpf_prog_stats *stats)
1289 {
1290         u64 nsecs = 0, cnt = 0;
1291         int cpu;
1292
1293         for_each_possible_cpu(cpu) {
1294                 const struct bpf_prog_stats *st;
1295                 unsigned int start;
1296                 u64 tnsecs, tcnt;
1297
1298                 st = per_cpu_ptr(prog->aux->stats, cpu);
1299                 do {
1300                         start = u64_stats_fetch_begin_irq(&st->syncp);
1301                         tnsecs = st->nsecs;
1302                         tcnt = st->cnt;
1303                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1304                 nsecs += tnsecs;
1305                 cnt += tcnt;
1306         }
1307         stats->nsecs = nsecs;
1308         stats->cnt = cnt;
1309 }
1310
1311 #ifdef CONFIG_PROC_FS
1312 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1313 {
1314         const struct bpf_prog *prog = filp->private_data;
1315         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1316         struct bpf_prog_stats stats;
1317
1318         bpf_prog_get_stats(prog, &stats);
1319         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1320         seq_printf(m,
1321                    "prog_type:\t%u\n"
1322                    "prog_jited:\t%u\n"
1323                    "prog_tag:\t%s\n"
1324                    "memlock:\t%llu\n"
1325                    "prog_id:\t%u\n"
1326                    "run_time_ns:\t%llu\n"
1327                    "run_cnt:\t%llu\n",
1328                    prog->type,
1329                    prog->jited,
1330                    prog_tag,
1331                    prog->pages * 1ULL << PAGE_SHIFT,
1332                    prog->aux->id,
1333                    stats.nsecs,
1334                    stats.cnt);
1335 }
1336 #endif
1337
1338 const struct file_operations bpf_prog_fops = {
1339 #ifdef CONFIG_PROC_FS
1340         .show_fdinfo    = bpf_prog_show_fdinfo,
1341 #endif
1342         .release        = bpf_prog_release,
1343         .read           = bpf_dummy_read,
1344         .write          = bpf_dummy_write,
1345 };
1346
1347 int bpf_prog_new_fd(struct bpf_prog *prog)
1348 {
1349         int ret;
1350
1351         ret = security_bpf_prog(prog);
1352         if (ret < 0)
1353                 return ret;
1354
1355         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1356                                 O_RDWR | O_CLOEXEC);
1357 }
1358
1359 static struct bpf_prog *____bpf_prog_get(struct fd f)
1360 {
1361         if (!f.file)
1362                 return ERR_PTR(-EBADF);
1363         if (f.file->f_op != &bpf_prog_fops) {
1364                 fdput(f);
1365                 return ERR_PTR(-EINVAL);
1366         }
1367
1368         return f.file->private_data;
1369 }
1370
1371 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1372 {
1373         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1374                 atomic_sub(i, &prog->aux->refcnt);
1375                 return ERR_PTR(-EBUSY);
1376         }
1377         return prog;
1378 }
1379 EXPORT_SYMBOL_GPL(bpf_prog_add);
1380
1381 void bpf_prog_sub(struct bpf_prog *prog, int i)
1382 {
1383         /* Only to be used for undoing previous bpf_prog_add() in some
1384          * error path. We still know that another entity in our call
1385          * path holds a reference to the program, thus atomic_sub() can
1386          * be safely used in such cases!
1387          */
1388         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1389 }
1390 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1391
1392 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1393 {
1394         return bpf_prog_add(prog, 1);
1395 }
1396 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1397
1398 /* prog_idr_lock should have been held */
1399 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1400 {
1401         int refold;
1402
1403         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1404
1405         if (refold >= BPF_MAX_REFCNT) {
1406                 __bpf_prog_put(prog, false);
1407                 return ERR_PTR(-EBUSY);
1408         }
1409
1410         if (!refold)
1411                 return ERR_PTR(-ENOENT);
1412
1413         return prog;
1414 }
1415 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1416
1417 bool bpf_prog_get_ok(struct bpf_prog *prog,
1418                             enum bpf_prog_type *attach_type, bool attach_drv)
1419 {
1420         /* not an attachment, just a refcount inc, always allow */
1421         if (!attach_type)
1422                 return true;
1423
1424         if (prog->type != *attach_type)
1425                 return false;
1426         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1427                 return false;
1428
1429         return true;
1430 }
1431
1432 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1433                                        bool attach_drv)
1434 {
1435         struct fd f = fdget(ufd);
1436         struct bpf_prog *prog;
1437
1438         prog = ____bpf_prog_get(f);
1439         if (IS_ERR(prog))
1440                 return prog;
1441         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1442                 prog = ERR_PTR(-EINVAL);
1443                 goto out;
1444         }
1445
1446         prog = bpf_prog_inc(prog);
1447 out:
1448         fdput(f);
1449         return prog;
1450 }
1451
1452 struct bpf_prog *bpf_prog_get(u32 ufd)
1453 {
1454         return __bpf_prog_get(ufd, NULL, false);
1455 }
1456
1457 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1458                                        bool attach_drv)
1459 {
1460         return __bpf_prog_get(ufd, &type, attach_drv);
1461 }
1462 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1463
1464 /* Initially all BPF programs could be loaded w/o specifying
1465  * expected_attach_type. Later for some of them specifying expected_attach_type
1466  * at load time became required so that program could be validated properly.
1467  * Programs of types that are allowed to be loaded both w/ and w/o (for
1468  * backward compatibility) expected_attach_type, should have the default attach
1469  * type assigned to expected_attach_type for the latter case, so that it can be
1470  * validated later at attach time.
1471  *
1472  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1473  * prog type requires it but has some attach types that have to be backward
1474  * compatible.
1475  */
1476 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1477 {
1478         switch (attr->prog_type) {
1479         case BPF_PROG_TYPE_CGROUP_SOCK:
1480                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1481                  * exist so checking for non-zero is the way to go here.
1482                  */
1483                 if (!attr->expected_attach_type)
1484                         attr->expected_attach_type =
1485                                 BPF_CGROUP_INET_SOCK_CREATE;
1486                 break;
1487         }
1488 }
1489
1490 static int
1491 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1492                                 enum bpf_attach_type expected_attach_type)
1493 {
1494         switch (prog_type) {
1495         case BPF_PROG_TYPE_CGROUP_SOCK:
1496                 switch (expected_attach_type) {
1497                 case BPF_CGROUP_INET_SOCK_CREATE:
1498                 case BPF_CGROUP_INET4_POST_BIND:
1499                 case BPF_CGROUP_INET6_POST_BIND:
1500                         return 0;
1501                 default:
1502                         return -EINVAL;
1503                 }
1504         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1505                 switch (expected_attach_type) {
1506                 case BPF_CGROUP_INET4_BIND:
1507                 case BPF_CGROUP_INET6_BIND:
1508                 case BPF_CGROUP_INET4_CONNECT:
1509                 case BPF_CGROUP_INET6_CONNECT:
1510                 case BPF_CGROUP_UDP4_SENDMSG:
1511                 case BPF_CGROUP_UDP6_SENDMSG:
1512                         return 0;
1513                 default:
1514                         return -EINVAL;
1515                 }
1516         default:
1517                 return 0;
1518         }
1519 }
1520
1521 /* last field in 'union bpf_attr' used by this command */
1522 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1523
1524 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1525 {
1526         enum bpf_prog_type type = attr->prog_type;
1527         struct bpf_prog *prog;
1528         int err;
1529         char license[128];
1530         bool is_gpl;
1531
1532         if (CHECK_ATTR(BPF_PROG_LOAD))
1533                 return -EINVAL;
1534
1535         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1536                 return -EINVAL;
1537
1538         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1539             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1540             !capable(CAP_SYS_ADMIN))
1541                 return -EPERM;
1542
1543         /* copy eBPF program license from user space */
1544         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1545                               sizeof(license) - 1) < 0)
1546                 return -EFAULT;
1547         license[sizeof(license) - 1] = 0;
1548
1549         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1550         is_gpl = license_is_gpl_compatible(license);
1551
1552         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1553                 return -E2BIG;
1554         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1555             type != BPF_PROG_TYPE_CGROUP_SKB &&
1556             !capable(CAP_SYS_ADMIN))
1557                 return -EPERM;
1558
1559         bpf_prog_load_fixup_attach_type(attr);
1560         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1561                 return -EINVAL;
1562
1563         /* plain bpf_prog allocation */
1564         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1565         if (!prog)
1566                 return -ENOMEM;
1567
1568         prog->expected_attach_type = attr->expected_attach_type;
1569
1570         prog->aux->offload_requested = !!attr->prog_ifindex;
1571
1572         err = security_bpf_prog_alloc(prog->aux);
1573         if (err)
1574                 goto free_prog_nouncharge;
1575
1576         err = bpf_prog_charge_memlock(prog);
1577         if (err)
1578                 goto free_prog_sec;
1579
1580         prog->len = attr->insn_cnt;
1581
1582         err = -EFAULT;
1583         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1584                            bpf_prog_insn_size(prog)) != 0)
1585                 goto free_prog;
1586
1587         prog->orig_prog = NULL;
1588         prog->jited = 0;
1589
1590         atomic_set(&prog->aux->refcnt, 1);
1591         prog->gpl_compatible = is_gpl ? 1 : 0;
1592
1593         if (bpf_prog_is_dev_bound(prog->aux)) {
1594                 err = bpf_prog_offload_init(prog, attr);
1595                 if (err)
1596                         goto free_prog;
1597         }
1598
1599         /* find program type: socket_filter vs tracing_filter */
1600         err = find_prog_type(type, prog);
1601         if (err < 0)
1602                 goto free_prog;
1603
1604         prog->aux->load_time = ktime_get_boot_ns();
1605         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1606         if (err)
1607                 goto free_prog;
1608
1609         /* run eBPF verifier */
1610         err = bpf_check(&prog, attr, uattr);
1611         if (err < 0)
1612                 goto free_used_maps;
1613
1614         prog = bpf_prog_select_runtime(prog, &err);
1615         if (err < 0)
1616                 goto free_used_maps;
1617
1618         err = bpf_prog_alloc_id(prog);
1619         if (err)
1620                 goto free_used_maps;
1621
1622         err = bpf_prog_new_fd(prog);
1623         if (err < 0) {
1624                 /* failed to allocate fd.
1625                  * bpf_prog_put() is needed because the above
1626                  * bpf_prog_alloc_id() has published the prog
1627                  * to the userspace and the userspace may
1628                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1629                  */
1630                 bpf_prog_put(prog);
1631                 return err;
1632         }
1633
1634         bpf_prog_kallsyms_add(prog);
1635         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1636         return err;
1637
1638 free_used_maps:
1639         bpf_prog_free_linfo(prog);
1640         kvfree(prog->aux->func_info);
1641         btf_put(prog->aux->btf);
1642         bpf_prog_kallsyms_del_subprogs(prog);
1643         free_used_maps(prog->aux);
1644 free_prog:
1645         bpf_prog_uncharge_memlock(prog);
1646 free_prog_sec:
1647         security_bpf_prog_free(prog->aux);
1648 free_prog_nouncharge:
1649         bpf_prog_free(prog);
1650         return err;
1651 }
1652
1653 #define BPF_OBJ_LAST_FIELD file_flags
1654
1655 static int bpf_obj_pin(const union bpf_attr *attr)
1656 {
1657         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1658                 return -EINVAL;
1659
1660         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1661 }
1662
1663 static int bpf_obj_get(const union bpf_attr *attr)
1664 {
1665         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1666             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1667                 return -EINVAL;
1668
1669         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1670                                 attr->file_flags);
1671 }
1672
1673 struct bpf_raw_tracepoint {
1674         struct bpf_raw_event_map *btp;
1675         struct bpf_prog *prog;
1676 };
1677
1678 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1679 {
1680         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1681
1682         if (raw_tp->prog) {
1683                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1684                 bpf_prog_put(raw_tp->prog);
1685         }
1686         bpf_put_raw_tracepoint(raw_tp->btp);
1687         kfree(raw_tp);
1688         return 0;
1689 }
1690
1691 static const struct file_operations bpf_raw_tp_fops = {
1692         .release        = bpf_raw_tracepoint_release,
1693         .read           = bpf_dummy_read,
1694         .write          = bpf_dummy_write,
1695 };
1696
1697 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1698
1699 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1700 {
1701         struct bpf_raw_tracepoint *raw_tp;
1702         struct bpf_raw_event_map *btp;
1703         struct bpf_prog *prog;
1704         char tp_name[128];
1705         int tp_fd, err;
1706
1707         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1708                               sizeof(tp_name) - 1) < 0)
1709                 return -EFAULT;
1710         tp_name[sizeof(tp_name) - 1] = 0;
1711
1712         btp = bpf_get_raw_tracepoint(tp_name);
1713         if (!btp)
1714                 return -ENOENT;
1715
1716         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1717         if (!raw_tp) {
1718                 err = -ENOMEM;
1719                 goto out_put_btp;
1720         }
1721         raw_tp->btp = btp;
1722
1723         prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1724                                  BPF_PROG_TYPE_RAW_TRACEPOINT);
1725         if (IS_ERR(prog)) {
1726                 err = PTR_ERR(prog);
1727                 goto out_free_tp;
1728         }
1729
1730         err = bpf_probe_register(raw_tp->btp, prog);
1731         if (err)
1732                 goto out_put_prog;
1733
1734         raw_tp->prog = prog;
1735         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1736                                  O_CLOEXEC);
1737         if (tp_fd < 0) {
1738                 bpf_probe_unregister(raw_tp->btp, prog);
1739                 err = tp_fd;
1740                 goto out_put_prog;
1741         }
1742         return tp_fd;
1743
1744 out_put_prog:
1745         bpf_prog_put(prog);
1746 out_free_tp:
1747         kfree(raw_tp);
1748 out_put_btp:
1749         bpf_put_raw_tracepoint(btp);
1750         return err;
1751 }
1752
1753 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1754                                              enum bpf_attach_type attach_type)
1755 {
1756         switch (prog->type) {
1757         case BPF_PROG_TYPE_CGROUP_SOCK:
1758         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1759                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1760         default:
1761                 return 0;
1762         }
1763 }
1764
1765 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1766
1767 #define BPF_F_ATTACH_MASK \
1768         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1769
1770 static int bpf_prog_attach(const union bpf_attr *attr)
1771 {
1772         enum bpf_prog_type ptype;
1773         struct bpf_prog *prog;
1774         int ret;
1775
1776         if (!capable(CAP_NET_ADMIN))
1777                 return -EPERM;
1778
1779         if (CHECK_ATTR(BPF_PROG_ATTACH))
1780                 return -EINVAL;
1781
1782         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1783                 return -EINVAL;
1784
1785         switch (attr->attach_type) {
1786         case BPF_CGROUP_INET_INGRESS:
1787         case BPF_CGROUP_INET_EGRESS:
1788                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1789                 break;
1790         case BPF_CGROUP_INET_SOCK_CREATE:
1791         case BPF_CGROUP_INET4_POST_BIND:
1792         case BPF_CGROUP_INET6_POST_BIND:
1793                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1794                 break;
1795         case BPF_CGROUP_INET4_BIND:
1796         case BPF_CGROUP_INET6_BIND:
1797         case BPF_CGROUP_INET4_CONNECT:
1798         case BPF_CGROUP_INET6_CONNECT:
1799         case BPF_CGROUP_UDP4_SENDMSG:
1800         case BPF_CGROUP_UDP6_SENDMSG:
1801                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1802                 break;
1803         case BPF_CGROUP_SOCK_OPS:
1804                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1805                 break;
1806         case BPF_CGROUP_DEVICE:
1807                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1808                 break;
1809         case BPF_SK_MSG_VERDICT:
1810                 ptype = BPF_PROG_TYPE_SK_MSG;
1811                 break;
1812         case BPF_SK_SKB_STREAM_PARSER:
1813         case BPF_SK_SKB_STREAM_VERDICT:
1814                 ptype = BPF_PROG_TYPE_SK_SKB;
1815                 break;
1816         case BPF_LIRC_MODE2:
1817                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1818                 break;
1819         case BPF_FLOW_DISSECTOR:
1820                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1821                 break;
1822         default:
1823                 return -EINVAL;
1824         }
1825
1826         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1827         if (IS_ERR(prog))
1828                 return PTR_ERR(prog);
1829
1830         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1831                 bpf_prog_put(prog);
1832                 return -EINVAL;
1833         }
1834
1835         switch (ptype) {
1836         case BPF_PROG_TYPE_SK_SKB:
1837         case BPF_PROG_TYPE_SK_MSG:
1838                 ret = sock_map_get_from_fd(attr, prog);
1839                 break;
1840         case BPF_PROG_TYPE_LIRC_MODE2:
1841                 ret = lirc_prog_attach(attr, prog);
1842                 break;
1843         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1844                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1845                 break;
1846         default:
1847                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1848         }
1849
1850         if (ret)
1851                 bpf_prog_put(prog);
1852         return ret;
1853 }
1854
1855 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1856
1857 static int bpf_prog_detach(const union bpf_attr *attr)
1858 {
1859         enum bpf_prog_type ptype;
1860
1861         if (!capable(CAP_NET_ADMIN))
1862                 return -EPERM;
1863
1864         if (CHECK_ATTR(BPF_PROG_DETACH))
1865                 return -EINVAL;
1866
1867         switch (attr->attach_type) {
1868         case BPF_CGROUP_INET_INGRESS:
1869         case BPF_CGROUP_INET_EGRESS:
1870                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1871                 break;
1872         case BPF_CGROUP_INET_SOCK_CREATE:
1873         case BPF_CGROUP_INET4_POST_BIND:
1874         case BPF_CGROUP_INET6_POST_BIND:
1875                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1876                 break;
1877         case BPF_CGROUP_INET4_BIND:
1878         case BPF_CGROUP_INET6_BIND:
1879         case BPF_CGROUP_INET4_CONNECT:
1880         case BPF_CGROUP_INET6_CONNECT:
1881         case BPF_CGROUP_UDP4_SENDMSG:
1882         case BPF_CGROUP_UDP6_SENDMSG:
1883                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1884                 break;
1885         case BPF_CGROUP_SOCK_OPS:
1886                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1887                 break;
1888         case BPF_CGROUP_DEVICE:
1889                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1890                 break;
1891         case BPF_SK_MSG_VERDICT:
1892                 return sock_map_get_from_fd(attr, NULL);
1893         case BPF_SK_SKB_STREAM_PARSER:
1894         case BPF_SK_SKB_STREAM_VERDICT:
1895                 return sock_map_get_from_fd(attr, NULL);
1896         case BPF_LIRC_MODE2:
1897                 return lirc_prog_detach(attr);
1898         case BPF_FLOW_DISSECTOR:
1899                 return skb_flow_dissector_bpf_prog_detach(attr);
1900         default:
1901                 return -EINVAL;
1902         }
1903
1904         return cgroup_bpf_prog_detach(attr, ptype);
1905 }
1906
1907 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1908
1909 static int bpf_prog_query(const union bpf_attr *attr,
1910                           union bpf_attr __user *uattr)
1911 {
1912         if (!capable(CAP_NET_ADMIN))
1913                 return -EPERM;
1914         if (CHECK_ATTR(BPF_PROG_QUERY))
1915                 return -EINVAL;
1916         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1917                 return -EINVAL;
1918
1919         switch (attr->query.attach_type) {
1920         case BPF_CGROUP_INET_INGRESS:
1921         case BPF_CGROUP_INET_EGRESS:
1922         case BPF_CGROUP_INET_SOCK_CREATE:
1923         case BPF_CGROUP_INET4_BIND:
1924         case BPF_CGROUP_INET6_BIND:
1925         case BPF_CGROUP_INET4_POST_BIND:
1926         case BPF_CGROUP_INET6_POST_BIND:
1927         case BPF_CGROUP_INET4_CONNECT:
1928         case BPF_CGROUP_INET6_CONNECT:
1929         case BPF_CGROUP_UDP4_SENDMSG:
1930         case BPF_CGROUP_UDP6_SENDMSG:
1931         case BPF_CGROUP_SOCK_OPS:
1932         case BPF_CGROUP_DEVICE:
1933                 break;
1934         case BPF_LIRC_MODE2:
1935                 return lirc_prog_query(attr, uattr);
1936         default:
1937                 return -EINVAL;
1938         }
1939
1940         return cgroup_bpf_prog_query(attr, uattr);
1941 }
1942
1943 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1944
1945 static int bpf_prog_test_run(const union bpf_attr *attr,
1946                              union bpf_attr __user *uattr)
1947 {
1948         struct bpf_prog *prog;
1949         int ret = -ENOTSUPP;
1950
1951         if (!capable(CAP_SYS_ADMIN))
1952                 return -EPERM;
1953         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1954                 return -EINVAL;
1955
1956         prog = bpf_prog_get(attr->test.prog_fd);
1957         if (IS_ERR(prog))
1958                 return PTR_ERR(prog);
1959
1960         if (prog->aux->ops->test_run)
1961                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1962
1963         bpf_prog_put(prog);
1964         return ret;
1965 }
1966
1967 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1968
1969 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1970                                union bpf_attr __user *uattr,
1971                                struct idr *idr,
1972                                spinlock_t *lock)
1973 {
1974         u32 next_id = attr->start_id;
1975         int err = 0;
1976
1977         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1978                 return -EINVAL;
1979
1980         if (!capable(CAP_SYS_ADMIN))
1981                 return -EPERM;
1982
1983         next_id++;
1984         spin_lock_bh(lock);
1985         if (!idr_get_next(idr, &next_id))
1986                 err = -ENOENT;
1987         spin_unlock_bh(lock);
1988
1989         if (!err)
1990                 err = put_user(next_id, &uattr->next_id);
1991
1992         return err;
1993 }
1994
1995 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1996
1997 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1998 {
1999         struct bpf_prog *prog;
2000         u32 id = attr->prog_id;
2001         int fd;
2002
2003         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2004                 return -EINVAL;
2005
2006         if (!capable(CAP_SYS_ADMIN))
2007                 return -EPERM;
2008
2009         spin_lock_bh(&prog_idr_lock);
2010         prog = idr_find(&prog_idr, id);
2011         if (prog)
2012                 prog = bpf_prog_inc_not_zero(prog);
2013         else
2014                 prog = ERR_PTR(-ENOENT);
2015         spin_unlock_bh(&prog_idr_lock);
2016
2017         if (IS_ERR(prog))
2018                 return PTR_ERR(prog);
2019
2020         fd = bpf_prog_new_fd(prog);
2021         if (fd < 0)
2022                 bpf_prog_put(prog);
2023
2024         return fd;
2025 }
2026
2027 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2028
2029 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2030 {
2031         struct bpf_map *map;
2032         u32 id = attr->map_id;
2033         int f_flags;
2034         int fd;
2035
2036         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2037             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2038                 return -EINVAL;
2039
2040         if (!capable(CAP_SYS_ADMIN))
2041                 return -EPERM;
2042
2043         f_flags = bpf_get_file_flag(attr->open_flags);
2044         if (f_flags < 0)
2045                 return f_flags;
2046
2047         spin_lock_bh(&map_idr_lock);
2048         map = idr_find(&map_idr, id);
2049         if (map)
2050                 map = bpf_map_inc_not_zero(map, true);
2051         else
2052                 map = ERR_PTR(-ENOENT);
2053         spin_unlock_bh(&map_idr_lock);
2054
2055         if (IS_ERR(map))
2056                 return PTR_ERR(map);
2057
2058         fd = bpf_map_new_fd(map, f_flags);
2059         if (fd < 0)
2060                 bpf_map_put_with_uref(map);
2061
2062         return fd;
2063 }
2064
2065 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2066                                               unsigned long addr)
2067 {
2068         int i;
2069
2070         for (i = 0; i < prog->aux->used_map_cnt; i++)
2071                 if (prog->aux->used_maps[i] == (void *)addr)
2072                         return prog->aux->used_maps[i];
2073         return NULL;
2074 }
2075
2076 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2077 {
2078         const struct bpf_map *map;
2079         struct bpf_insn *insns;
2080         u64 imm;
2081         int i;
2082
2083         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2084                         GFP_USER);
2085         if (!insns)
2086                 return insns;
2087
2088         for (i = 0; i < prog->len; i++) {
2089                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2090                         insns[i].code = BPF_JMP | BPF_CALL;
2091                         insns[i].imm = BPF_FUNC_tail_call;
2092                         /* fall-through */
2093                 }
2094                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2095                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2096                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2097                                 insns[i].code = BPF_JMP | BPF_CALL;
2098                         if (!bpf_dump_raw_ok())
2099                                 insns[i].imm = 0;
2100                         continue;
2101                 }
2102
2103                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2104                         continue;
2105
2106                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2107                 map = bpf_map_from_imm(prog, imm);
2108                 if (map) {
2109                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2110                         insns[i].imm = map->id;
2111                         insns[i + 1].imm = 0;
2112                         continue;
2113                 }
2114         }
2115
2116         return insns;
2117 }
2118
2119 static int set_info_rec_size(struct bpf_prog_info *info)
2120 {
2121         /*
2122          * Ensure info.*_rec_size is the same as kernel expected size
2123          *
2124          * or
2125          *
2126          * Only allow zero *_rec_size if both _rec_size and _cnt are
2127          * zero.  In this case, the kernel will set the expected
2128          * _rec_size back to the info.
2129          */
2130
2131         if ((info->nr_func_info || info->func_info_rec_size) &&
2132             info->func_info_rec_size != sizeof(struct bpf_func_info))
2133                 return -EINVAL;
2134
2135         if ((info->nr_line_info || info->line_info_rec_size) &&
2136             info->line_info_rec_size != sizeof(struct bpf_line_info))
2137                 return -EINVAL;
2138
2139         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2140             info->jited_line_info_rec_size != sizeof(__u64))
2141                 return -EINVAL;
2142
2143         info->func_info_rec_size = sizeof(struct bpf_func_info);
2144         info->line_info_rec_size = sizeof(struct bpf_line_info);
2145         info->jited_line_info_rec_size = sizeof(__u64);
2146
2147         return 0;
2148 }
2149
2150 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2151                                    const union bpf_attr *attr,
2152                                    union bpf_attr __user *uattr)
2153 {
2154         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2155         struct bpf_prog_info info = {};
2156         u32 info_len = attr->info.info_len;
2157         struct bpf_prog_stats stats;
2158         char __user *uinsns;
2159         u32 ulen;
2160         int err;
2161
2162         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2163         if (err)
2164                 return err;
2165         info_len = min_t(u32, sizeof(info), info_len);
2166
2167         if (copy_from_user(&info, uinfo, info_len))
2168                 return -EFAULT;
2169
2170         info.type = prog->type;
2171         info.id = prog->aux->id;
2172         info.load_time = prog->aux->load_time;
2173         info.created_by_uid = from_kuid_munged(current_user_ns(),
2174                                                prog->aux->user->uid);
2175         info.gpl_compatible = prog->gpl_compatible;
2176
2177         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2178         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2179
2180         ulen = info.nr_map_ids;
2181         info.nr_map_ids = prog->aux->used_map_cnt;
2182         ulen = min_t(u32, info.nr_map_ids, ulen);
2183         if (ulen) {
2184                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2185                 u32 i;
2186
2187                 for (i = 0; i < ulen; i++)
2188                         if (put_user(prog->aux->used_maps[i]->id,
2189                                      &user_map_ids[i]))
2190                                 return -EFAULT;
2191         }
2192
2193         err = set_info_rec_size(&info);
2194         if (err)
2195                 return err;
2196
2197         bpf_prog_get_stats(prog, &stats);
2198         info.run_time_ns = stats.nsecs;
2199         info.run_cnt = stats.cnt;
2200
2201         if (!capable(CAP_SYS_ADMIN)) {
2202                 info.jited_prog_len = 0;
2203                 info.xlated_prog_len = 0;
2204                 info.nr_jited_ksyms = 0;
2205                 info.nr_jited_func_lens = 0;
2206                 info.nr_func_info = 0;
2207                 info.nr_line_info = 0;
2208                 info.nr_jited_line_info = 0;
2209                 goto done;
2210         }
2211
2212         ulen = info.xlated_prog_len;
2213         info.xlated_prog_len = bpf_prog_insn_size(prog);
2214         if (info.xlated_prog_len && ulen) {
2215                 struct bpf_insn *insns_sanitized;
2216                 bool fault;
2217
2218                 if (prog->blinded && !bpf_dump_raw_ok()) {
2219                         info.xlated_prog_insns = 0;
2220                         goto done;
2221                 }
2222                 insns_sanitized = bpf_insn_prepare_dump(prog);
2223                 if (!insns_sanitized)
2224                         return -ENOMEM;
2225                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2226                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2227                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2228                 kfree(insns_sanitized);
2229                 if (fault)
2230                         return -EFAULT;
2231         }
2232
2233         if (bpf_prog_is_dev_bound(prog->aux)) {
2234                 err = bpf_prog_offload_info_fill(&info, prog);
2235                 if (err)
2236                         return err;
2237                 goto done;
2238         }
2239
2240         /* NOTE: the following code is supposed to be skipped for offload.
2241          * bpf_prog_offload_info_fill() is the place to fill similar fields
2242          * for offload.
2243          */
2244         ulen = info.jited_prog_len;
2245         if (prog->aux->func_cnt) {
2246                 u32 i;
2247
2248                 info.jited_prog_len = 0;
2249                 for (i = 0; i < prog->aux->func_cnt; i++)
2250                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2251         } else {
2252                 info.jited_prog_len = prog->jited_len;
2253         }
2254
2255         if (info.jited_prog_len && ulen) {
2256                 if (bpf_dump_raw_ok()) {
2257                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2258                         ulen = min_t(u32, info.jited_prog_len, ulen);
2259
2260                         /* for multi-function programs, copy the JITed
2261                          * instructions for all the functions
2262                          */
2263                         if (prog->aux->func_cnt) {
2264                                 u32 len, free, i;
2265                                 u8 *img;
2266
2267                                 free = ulen;
2268                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2269                                         len = prog->aux->func[i]->jited_len;
2270                                         len = min_t(u32, len, free);
2271                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2272                                         if (copy_to_user(uinsns, img, len))
2273                                                 return -EFAULT;
2274                                         uinsns += len;
2275                                         free -= len;
2276                                         if (!free)
2277                                                 break;
2278                                 }
2279                         } else {
2280                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2281                                         return -EFAULT;
2282                         }
2283                 } else {
2284                         info.jited_prog_insns = 0;
2285                 }
2286         }
2287
2288         ulen = info.nr_jited_ksyms;
2289         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2290         if (ulen) {
2291                 if (bpf_dump_raw_ok()) {
2292                         unsigned long ksym_addr;
2293                         u64 __user *user_ksyms;
2294                         u32 i;
2295
2296                         /* copy the address of the kernel symbol
2297                          * corresponding to each function
2298                          */
2299                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2300                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2301                         if (prog->aux->func_cnt) {
2302                                 for (i = 0; i < ulen; i++) {
2303                                         ksym_addr = (unsigned long)
2304                                                 prog->aux->func[i]->bpf_func;
2305                                         if (put_user((u64) ksym_addr,
2306                                                      &user_ksyms[i]))
2307                                                 return -EFAULT;
2308                                 }
2309                         } else {
2310                                 ksym_addr = (unsigned long) prog->bpf_func;
2311                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2312                                         return -EFAULT;
2313                         }
2314                 } else {
2315                         info.jited_ksyms = 0;
2316                 }
2317         }
2318
2319         ulen = info.nr_jited_func_lens;
2320         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2321         if (ulen) {
2322                 if (bpf_dump_raw_ok()) {
2323                         u32 __user *user_lens;
2324                         u32 func_len, i;
2325
2326                         /* copy the JITed image lengths for each function */
2327                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2328                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2329                         if (prog->aux->func_cnt) {
2330                                 for (i = 0; i < ulen; i++) {
2331                                         func_len =
2332                                                 prog->aux->func[i]->jited_len;
2333                                         if (put_user(func_len, &user_lens[i]))
2334                                                 return -EFAULT;
2335                                 }
2336                         } else {
2337                                 func_len = prog->jited_len;
2338                                 if (put_user(func_len, &user_lens[0]))
2339                                         return -EFAULT;
2340                         }
2341                 } else {
2342                         info.jited_func_lens = 0;
2343                 }
2344         }
2345
2346         if (prog->aux->btf)
2347                 info.btf_id = btf_id(prog->aux->btf);
2348
2349         ulen = info.nr_func_info;
2350         info.nr_func_info = prog->aux->func_info_cnt;
2351         if (info.nr_func_info && ulen) {
2352                 char __user *user_finfo;
2353
2354                 user_finfo = u64_to_user_ptr(info.func_info);
2355                 ulen = min_t(u32, info.nr_func_info, ulen);
2356                 if (copy_to_user(user_finfo, prog->aux->func_info,
2357                                  info.func_info_rec_size * ulen))
2358                         return -EFAULT;
2359         }
2360
2361         ulen = info.nr_line_info;
2362         info.nr_line_info = prog->aux->nr_linfo;
2363         if (info.nr_line_info && ulen) {
2364                 __u8 __user *user_linfo;
2365
2366                 user_linfo = u64_to_user_ptr(info.line_info);
2367                 ulen = min_t(u32, info.nr_line_info, ulen);
2368                 if (copy_to_user(user_linfo, prog->aux->linfo,
2369                                  info.line_info_rec_size * ulen))
2370                         return -EFAULT;
2371         }
2372
2373         ulen = info.nr_jited_line_info;
2374         if (prog->aux->jited_linfo)
2375                 info.nr_jited_line_info = prog->aux->nr_linfo;
2376         else
2377                 info.nr_jited_line_info = 0;
2378         if (info.nr_jited_line_info && ulen) {
2379                 if (bpf_dump_raw_ok()) {
2380                         __u64 __user *user_linfo;
2381                         u32 i;
2382
2383                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2384                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2385                         for (i = 0; i < ulen; i++) {
2386                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2387                                              &user_linfo[i]))
2388                                         return -EFAULT;
2389                         }
2390                 } else {
2391                         info.jited_line_info = 0;
2392                 }
2393         }
2394
2395         ulen = info.nr_prog_tags;
2396         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2397         if (ulen) {
2398                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2399                 u32 i;
2400
2401                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2402                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2403                 if (prog->aux->func_cnt) {
2404                         for (i = 0; i < ulen; i++) {
2405                                 if (copy_to_user(user_prog_tags[i],
2406                                                  prog->aux->func[i]->tag,
2407                                                  BPF_TAG_SIZE))
2408                                         return -EFAULT;
2409                         }
2410                 } else {
2411                         if (copy_to_user(user_prog_tags[0],
2412                                          prog->tag, BPF_TAG_SIZE))
2413                                 return -EFAULT;
2414                 }
2415         }
2416
2417 done:
2418         if (copy_to_user(uinfo, &info, info_len) ||
2419             put_user(info_len, &uattr->info.info_len))
2420                 return -EFAULT;
2421
2422         return 0;
2423 }
2424
2425 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2426                                   const union bpf_attr *attr,
2427                                   union bpf_attr __user *uattr)
2428 {
2429         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2430         struct bpf_map_info info = {};
2431         u32 info_len = attr->info.info_len;
2432         int err;
2433
2434         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2435         if (err)
2436                 return err;
2437         info_len = min_t(u32, sizeof(info), info_len);
2438
2439         info.type = map->map_type;
2440         info.id = map->id;
2441         info.key_size = map->key_size;
2442         info.value_size = map->value_size;
2443         info.max_entries = map->max_entries;
2444         info.map_flags = map->map_flags;
2445         memcpy(info.name, map->name, sizeof(map->name));
2446
2447         if (map->btf) {
2448                 info.btf_id = btf_id(map->btf);
2449                 info.btf_key_type_id = map->btf_key_type_id;
2450                 info.btf_value_type_id = map->btf_value_type_id;
2451         }
2452
2453         if (bpf_map_is_dev_bound(map)) {
2454                 err = bpf_map_offload_info_fill(&info, map);
2455                 if (err)
2456                         return err;
2457         }
2458
2459         if (copy_to_user(uinfo, &info, info_len) ||
2460             put_user(info_len, &uattr->info.info_len))
2461                 return -EFAULT;
2462
2463         return 0;
2464 }
2465
2466 static int bpf_btf_get_info_by_fd(struct btf *btf,
2467                                   const union bpf_attr *attr,
2468                                   union bpf_attr __user *uattr)
2469 {
2470         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2471         u32 info_len = attr->info.info_len;
2472         int err;
2473
2474         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2475         if (err)
2476                 return err;
2477
2478         return btf_get_info_by_fd(btf, attr, uattr);
2479 }
2480
2481 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2482
2483 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2484                                   union bpf_attr __user *uattr)
2485 {
2486         int ufd = attr->info.bpf_fd;
2487         struct fd f;
2488         int err;
2489
2490         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2491                 return -EINVAL;
2492
2493         f = fdget(ufd);
2494         if (!f.file)
2495                 return -EBADFD;
2496
2497         if (f.file->f_op == &bpf_prog_fops)
2498                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2499                                               uattr);
2500         else if (f.file->f_op == &bpf_map_fops)
2501                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2502                                              uattr);
2503         else if (f.file->f_op == &btf_fops)
2504                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2505         else
2506                 err = -EINVAL;
2507
2508         fdput(f);
2509         return err;
2510 }
2511
2512 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2513
2514 static int bpf_btf_load(const union bpf_attr *attr)
2515 {
2516         if (CHECK_ATTR(BPF_BTF_LOAD))
2517                 return -EINVAL;
2518
2519         if (!capable(CAP_SYS_ADMIN))
2520                 return -EPERM;
2521
2522         return btf_new_fd(attr);
2523 }
2524
2525 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2526
2527 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2528 {
2529         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2530                 return -EINVAL;
2531
2532         if (!capable(CAP_SYS_ADMIN))
2533                 return -EPERM;
2534
2535         return btf_get_fd_by_id(attr->btf_id);
2536 }
2537
2538 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2539                                     union bpf_attr __user *uattr,
2540                                     u32 prog_id, u32 fd_type,
2541                                     const char *buf, u64 probe_offset,
2542                                     u64 probe_addr)
2543 {
2544         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2545         u32 len = buf ? strlen(buf) : 0, input_len;
2546         int err = 0;
2547
2548         if (put_user(len, &uattr->task_fd_query.buf_len))
2549                 return -EFAULT;
2550         input_len = attr->task_fd_query.buf_len;
2551         if (input_len && ubuf) {
2552                 if (!len) {
2553                         /* nothing to copy, just make ubuf NULL terminated */
2554                         char zero = '\0';
2555
2556                         if (put_user(zero, ubuf))
2557                                 return -EFAULT;
2558                 } else if (input_len >= len + 1) {
2559                         /* ubuf can hold the string with NULL terminator */
2560                         if (copy_to_user(ubuf, buf, len + 1))
2561                                 return -EFAULT;
2562                 } else {
2563                         /* ubuf cannot hold the string with NULL terminator,
2564                          * do a partial copy with NULL terminator.
2565                          */
2566                         char zero = '\0';
2567
2568                         err = -ENOSPC;
2569                         if (copy_to_user(ubuf, buf, input_len - 1))
2570                                 return -EFAULT;
2571                         if (put_user(zero, ubuf + input_len - 1))
2572                                 return -EFAULT;
2573                 }
2574         }
2575
2576         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2577             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2578             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2579             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2580                 return -EFAULT;
2581
2582         return err;
2583 }
2584
2585 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2586
2587 static int bpf_task_fd_query(const union bpf_attr *attr,
2588                              union bpf_attr __user *uattr)
2589 {
2590         pid_t pid = attr->task_fd_query.pid;
2591         u32 fd = attr->task_fd_query.fd;
2592         const struct perf_event *event;
2593         struct files_struct *files;
2594         struct task_struct *task;
2595         struct file *file;
2596         int err;
2597
2598         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2599                 return -EINVAL;
2600
2601         if (!capable(CAP_SYS_ADMIN))
2602                 return -EPERM;
2603
2604         if (attr->task_fd_query.flags != 0)
2605                 return -EINVAL;
2606
2607         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2608         if (!task)
2609                 return -ENOENT;
2610
2611         files = get_files_struct(task);
2612         put_task_struct(task);
2613         if (!files)
2614                 return -ENOENT;
2615
2616         err = 0;
2617         spin_lock(&files->file_lock);
2618         file = fcheck_files(files, fd);
2619         if (!file)
2620                 err = -EBADF;
2621         else
2622                 get_file(file);
2623         spin_unlock(&files->file_lock);
2624         put_files_struct(files);
2625
2626         if (err)
2627                 goto out;
2628
2629         if (file->f_op == &bpf_raw_tp_fops) {
2630                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2631                 struct bpf_raw_event_map *btp = raw_tp->btp;
2632
2633                 err = bpf_task_fd_query_copy(attr, uattr,
2634                                              raw_tp->prog->aux->id,
2635                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2636                                              btp->tp->name, 0, 0);
2637                 goto put_file;
2638         }
2639
2640         event = perf_get_event(file);
2641         if (!IS_ERR(event)) {
2642                 u64 probe_offset, probe_addr;
2643                 u32 prog_id, fd_type;
2644                 const char *buf;
2645
2646                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2647                                               &buf, &probe_offset,
2648                                               &probe_addr);
2649                 if (!err)
2650                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2651                                                      fd_type, buf,
2652                                                      probe_offset,
2653                                                      probe_addr);
2654                 goto put_file;
2655         }
2656
2657         err = -ENOTSUPP;
2658 put_file:
2659         fput(file);
2660 out:
2661         return err;
2662 }
2663
2664 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2665 {
2666         union bpf_attr attr = {};
2667         int err;
2668
2669         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2670                 return -EPERM;
2671
2672         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2673         if (err)
2674                 return err;
2675         size = min_t(u32, size, sizeof(attr));
2676
2677         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2678         if (copy_from_user(&attr, uattr, size) != 0)
2679                 return -EFAULT;
2680
2681         err = security_bpf(cmd, &attr, size);
2682         if (err < 0)
2683                 return err;
2684
2685         switch (cmd) {
2686         case BPF_MAP_CREATE:
2687                 err = map_create(&attr);
2688                 break;
2689         case BPF_MAP_LOOKUP_ELEM:
2690                 err = map_lookup_elem(&attr);
2691                 break;
2692         case BPF_MAP_UPDATE_ELEM:
2693                 err = map_update_elem(&attr);
2694                 break;
2695         case BPF_MAP_DELETE_ELEM:
2696                 err = map_delete_elem(&attr);
2697                 break;
2698         case BPF_MAP_GET_NEXT_KEY:
2699                 err = map_get_next_key(&attr);
2700                 break;
2701         case BPF_PROG_LOAD:
2702                 err = bpf_prog_load(&attr, uattr);
2703                 break;
2704         case BPF_OBJ_PIN:
2705                 err = bpf_obj_pin(&attr);
2706                 break;
2707         case BPF_OBJ_GET:
2708                 err = bpf_obj_get(&attr);
2709                 break;
2710         case BPF_PROG_ATTACH:
2711                 err = bpf_prog_attach(&attr);
2712                 break;
2713         case BPF_PROG_DETACH:
2714                 err = bpf_prog_detach(&attr);
2715                 break;
2716         case BPF_PROG_QUERY:
2717                 err = bpf_prog_query(&attr, uattr);
2718                 break;
2719         case BPF_PROG_TEST_RUN:
2720                 err = bpf_prog_test_run(&attr, uattr);
2721                 break;
2722         case BPF_PROG_GET_NEXT_ID:
2723                 err = bpf_obj_get_next_id(&attr, uattr,
2724                                           &prog_idr, &prog_idr_lock);
2725                 break;
2726         case BPF_MAP_GET_NEXT_ID:
2727                 err = bpf_obj_get_next_id(&attr, uattr,
2728                                           &map_idr, &map_idr_lock);
2729                 break;
2730         case BPF_PROG_GET_FD_BY_ID:
2731                 err = bpf_prog_get_fd_by_id(&attr);
2732                 break;
2733         case BPF_MAP_GET_FD_BY_ID:
2734                 err = bpf_map_get_fd_by_id(&attr);
2735                 break;
2736         case BPF_OBJ_GET_INFO_BY_FD:
2737                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2738                 break;
2739         case BPF_RAW_TRACEPOINT_OPEN:
2740                 err = bpf_raw_tracepoint_open(&attr);
2741                 break;
2742         case BPF_BTF_LOAD:
2743                 err = bpf_btf_load(&attr);
2744                 break;
2745         case BPF_BTF_GET_FD_BY_ID:
2746                 err = bpf_btf_get_fd_by_id(&attr);
2747                 break;
2748         case BPF_TASK_FD_QUERY:
2749                 err = bpf_task_fd_query(&attr, uattr);
2750                 break;
2751         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2752                 err = map_lookup_and_delete_elem(&attr);
2753                 break;
2754         default:
2755                 err = -EINVAL;
2756                 break;
2757         }
2758
2759         return err;
2760 }