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 really just want to fail instead of triggering OOM killer
 140          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 141          * which is used for lower order allocation requests.
 142          *
 143          * It has been observed that higher order allocation requests done by
 144          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 145          * to reclaim memory from the page cache, thus we set
 146          * __GFP_RETRY_MAYFAIL to avoid such situations.
 147          */
 148
 149         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 150         void *area;
 151
 152         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 153                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 154                                     numa_node);
 155                 if (area != NULL)
 156                         return area;
 157         }
 158
 159         return __vmalloc_node_flags_caller(size, numa_node,
 160                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 161                                            flags, __builtin_return_address(0));
 162 }
 163
 164 void bpf_map_area_free(void *area)
 165 {
 166         kvfree(area);
 167 }
 168
 169 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 170 {
 171         map->map_type = attr->map_type;
 172         map->key_size = attr->key_size;
 173         map->value_size = attr->value_size;
 174         map->max_entries = attr->max_entries;
 175         map->map_flags = attr->map_flags;
 176         map->numa_node = bpf_map_attr_numa_node(attr);
 177 }
 178
 179 int bpf_map_precharge_memlock(u32 pages)
 180 {
 181         struct user_struct *user = get_current_user();
 182         unsigned long memlock_limit, cur;
 183
 184         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 185         cur = atomic_long_read(&user->locked_vm);
 186         free_uid(user);
 187         if (cur + pages > memlock_limit)
 188                 return -EPERM;
 189         return 0;
 190 }
 191
 192 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 193 {
 194         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 195
 196         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 197                 atomic_long_sub(pages, &user->locked_vm);
 198                 return -EPERM;
 199         }
 200         return 0;
 201 }
 202
 203 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 204 {
 205         atomic_long_sub(pages, &user->locked_vm);
 206 }
 207
 208 static int bpf_map_init_memlock(struct bpf_map *map)
 209 {
 210         struct user_struct *user = get_current_user();
 211         int ret;
 212
 213         ret = bpf_charge_memlock(user, map->pages);
 214         if (ret) {
 215                 free_uid(user);
 216                 return ret;
 217         }
 218         map->user = user;
 219         return ret;
 220 }
 221
 222 static void bpf_map_release_memlock(struct bpf_map *map)
 223 {
 224         struct user_struct *user = map->user;
 225         bpf_uncharge_memlock(user, map->pages);
 226         free_uid(user);
 227 }
 228
 229 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 230 {
 231         int ret;
 232
 233         ret = bpf_charge_memlock(map->user, pages);
 234         if (ret)
 235                 return ret;
 236         map->pages += pages;
 237         return ret;
 238 }
 239
 240 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 241 {
 242         bpf_uncharge_memlock(map->user, pages);
 243         map->pages -= pages;
 244 }
 245
 246 static int bpf_map_alloc_id(struct bpf_map *map)
 247 {
 248         int id;
 249
 250         idr_preload(GFP_KERNEL);
 251         spin_lock_bh(&map_idr_lock);
 252         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 253         if (id > 0)
 254                 map->id = id;
 255         spin_unlock_bh(&map_idr_lock);
 256         idr_preload_end();
 257
 258         if (WARN_ON_ONCE(!id))
 259                 return -ENOSPC;
 260
 261         return id > 0 ? 0 : id;
 262 }
 263
 264 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 265 {
 266         unsigned long flags;
 267
 268         /* Offloaded maps are removed from the IDR store when their device
 269          * disappears - even if someone holds an fd to them they are unusable,
 270          * the memory is gone, all ops will fail; they are simply waiting for
 271          * refcnt to drop to be freed.
 272          */
 273         if (!map->id)
 274                 return;
 275
 276         if (do_idr_lock)
 277                 spin_lock_irqsave(&map_idr_lock, flags);
 278         else
 279                 __acquire(&map_idr_lock);
 280
 281         idr_remove(&map_idr, map->id);
 282         map->id = 0;
 283
 284         if (do_idr_lock)
 285                 spin_unlock_irqrestore(&map_idr_lock, flags);
 286         else
 287                 __release(&map_idr_lock);
 288 }
 289
 290 /* called from workqueue */
 291 static void bpf_map_free_deferred(struct work_struct *work)
 292 {
 293         struct bpf_map *map = container_of(work, struct bpf_map, work);
 294
 295         bpf_map_release_memlock(map);
 296         security_bpf_map_free(map);
 297         /* implementation dependent freeing */
 298         map->ops->map_free(map);
 299 }
 300
 301 static void bpf_map_put_uref(struct bpf_map *map)
 302 {
 303         if (atomic_dec_and_test(&map->usercnt)) {
 304                 if (map->ops->map_release_uref)
 305                         map->ops->map_release_uref(map);
 306         }
 307 }
 308
 309 /* decrement map refcnt and schedule it for freeing via workqueue
 310  * (unrelying map implementation ops->map_free() might sleep)
 311  */
 312 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 313 {
 314         if (atomic_dec_and_test(&map->refcnt)) {
 315                 /* bpf_map_free_id() must be called first */
 316                 bpf_map_free_id(map, do_idr_lock);
 317                 btf_put(map->btf);
 318                 INIT_WORK(&map->work, bpf_map_free_deferred);
 319                 schedule_work(&map->work);
 320         }
 321 }
 322
 323 void bpf_map_put(struct bpf_map *map)
 324 {
 325         __bpf_map_put(map, true);
 326 }
 327 EXPORT_SYMBOL_GPL(bpf_map_put);
 328
 329 void bpf_map_put_with_uref(struct bpf_map *map)
 330 {
 331         bpf_map_put_uref(map);
 332         bpf_map_put(map);
 333 }
 334
 335 static int bpf_map_release(struct inode *inode, struct file *filp)
 336 {
 337         struct bpf_map *map = filp->private_data;
 338
 339         if (map->ops->map_release)
 340                 map->ops->map_release(map, filp);
 341
 342         bpf_map_put_with_uref(map);
 343         return 0;
 344 }
 345
 346 #ifdef CONFIG_PROC_FS
 347 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 348 {
 349         const struct bpf_map *map = filp->private_data;
 350         const struct bpf_array *array;
 351         u32 owner_prog_type = 0;
 352         u32 owner_jited = 0;
 353
 354         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 355                 array = container_of(map, struct bpf_array, map);
 356                 owner_prog_type = array->owner_prog_type;
 357                 owner_jited = array->owner_jited;
 358         }
 359
 360         seq_printf(m,
 361                    "map_type:\t%u\n"
 362                    "key_size:\t%u\n"
 363                    "value_size:\t%u\n"
 364                    "max_entries:\t%u\n"
 365                    "map_flags:\t%#x\n"
 366                    "memlock:\t%llu\n"
 367                    "map_id:\t%u\n",
 368                    map->map_type,
 369                    map->key_size,
 370                    map->value_size,
 371                    map->max_entries,
 372                    map->map_flags,
 373                    map->pages * 1ULL << PAGE_SHIFT,
 374                    map->id);
 375
 376         if (owner_prog_type) {
 377                 seq_printf(m, "owner_prog_type:\t%u\n",
 378                            owner_prog_type);
 379                 seq_printf(m, "owner_jited:\t%u\n",
 380                            owner_jited);
 381         }
 382 }
 383 #endif
 384
 385 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 386                               loff_t *ppos)
 387 {
 388         /* We need this handler such that alloc_file() enables
 389          * f_mode with FMODE_CAN_READ.
 390          */
 391         return -EINVAL;
 392 }
 393
 394 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 395                                size_t siz, loff_t *ppos)
 396 {
 397         /* We need this handler such that alloc_file() enables
 398          * f_mode with FMODE_CAN_WRITE.
 399          */
 400         return -EINVAL;
 401 }
 402
 403 const struct file_operations bpf_map_fops = {
 404 #ifdef CONFIG_PROC_FS
 405         .show_fdinfo    = bpf_map_show_fdinfo,
 406 #endif
 407         .release        = bpf_map_release,
 408         .read           = bpf_dummy_read,
 409         .write          = bpf_dummy_write,
 410 };
 411
 412 int bpf_map_new_fd(struct bpf_map *map, int flags)
 413 {
 414         int ret;
 415
 416         ret = security_bpf_map(map, OPEN_FMODE(flags));
 417         if (ret < 0)
 418                 return ret;
 419
 420         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 421                                 flags | O_CLOEXEC);
 422 }
 423
 424 int bpf_get_file_flag(int flags)
 425 {
 426         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 427                 return -EINVAL;
 428         if (flags & BPF_F_RDONLY)
 429                 return O_RDONLY;
 430         if (flags & BPF_F_WRONLY)
 431                 return O_WRONLY;
 432         return O_RDWR;
 433 }
 434
 435 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 436 #define CHECK_ATTR(CMD) \
 437         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 438                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 439                    sizeof(*attr) - \
 440                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 441                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 442
 443 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 444  * Return 0 on success and < 0 on error.
 445  */
 446 static int bpf_obj_name_cpy(char *dst, const char *src)
 447 {
 448         const char *end = src + BPF_OBJ_NAME_LEN;
 449
 450         memset(dst, 0, BPF_OBJ_NAME_LEN);
 451
 452         /* Copy all isalnum() and '_' char */
 453         while (src < end && *src) {
 454                 if (!isalnum(*src) && *src != '_')
 455                         return -EINVAL;
 456                 *dst++ = *src++;
 457         }
 458
 459         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 460         if (src == end)
 461                 return -EINVAL;
 462
 463         return 0;
 464 }
 465
 466 int map_check_no_btf(const struct bpf_map *map,
 467                      const struct btf *btf,
 468                      const struct btf_type *key_type,
 469                      const struct btf_type *value_type)
 470 {
 471         return -ENOTSUPP;
 472 }
 473
 474 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 475                          u32 btf_key_id, u32 btf_value_id)
 476 {
 477         const struct btf_type *key_type, *value_type;
 478         u32 key_size, value_size;
 479         int ret = 0;
 480
 481         key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 482         if (!key_type || key_size != map->key_size)
 483                 return -EINVAL;
 484
 485         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 486         if (!value_type || value_size != map->value_size)
 487                 return -EINVAL;
 488
 489         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 490
 491         if (map_value_has_spin_lock(map)) {
 492                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 493                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 494                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE)
 495                         return -ENOTSUPP;
 496                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 497                     map->value_size) {
 498                         WARN_ONCE(1,
 499                                   "verifier bug spin_lock_off %d value_size %d\n",
 500                                   map->spin_lock_off, map->value_size);
 501                         return -EFAULT;
 502                 }
 503         }
 504
 505         if (map->ops->map_check_btf)
 506                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 507
 508         return ret;
 509 }
 510
 511 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 512 /* called via syscall */
 513 static int map_create(union bpf_attr *attr)
 514 {
 515         int numa_node = bpf_map_attr_numa_node(attr);
 516         struct bpf_map *map;
 517         int f_flags;
 518         int err;
 519
 520         err = CHECK_ATTR(BPF_MAP_CREATE);
 521         if (err)
 522                 return -EINVAL;
 523
 524         f_flags = bpf_get_file_flag(attr->map_flags);
 525         if (f_flags < 0)
 526                 return f_flags;
 527
 528         if (numa_node != NUMA_NO_NODE &&
 529             ((unsigned int)numa_node >= nr_node_ids ||
 530              !node_online(numa_node)))
 531                 return -EINVAL;
 532
 533         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 534         map = find_and_alloc_map(attr);
 535         if (IS_ERR(map))
 536                 return PTR_ERR(map);
 537
 538         err = bpf_obj_name_cpy(map->name, attr->map_name);
 539         if (err)
 540                 goto free_map_nouncharge;
 541
 542         atomic_set(&map->refcnt, 1);
 543         atomic_set(&map->usercnt, 1);
 544
 545         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 546                 struct btf *btf;
 547
 548                 if (!attr->btf_key_type_id || !attr->btf_value_type_id) {
 549                         err = -EINVAL;
 550                         goto free_map_nouncharge;
 551                 }
 552
 553                 btf = btf_get_by_fd(attr->btf_fd);
 554                 if (IS_ERR(btf)) {
 555                         err = PTR_ERR(btf);
 556                         goto free_map_nouncharge;
 557                 }
 558
 559                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 560                                     attr->btf_value_type_id);
 561                 if (err) {
 562                         btf_put(btf);
 563                         goto free_map_nouncharge;
 564                 }
 565
 566                 map->btf = btf;
 567                 map->btf_key_type_id = attr->btf_key_type_id;
 568                 map->btf_value_type_id = attr->btf_value_type_id;
 569         } else {
 570                 map->spin_lock_off = -EINVAL;
 571         }
 572
 573         err = security_bpf_map_alloc(map);
 574         if (err)
 575                 goto free_map_nouncharge;
 576
 577         err = bpf_map_init_memlock(map);
 578         if (err)
 579                 goto free_map_sec;
 580
 581         err = bpf_map_alloc_id(map);
 582         if (err)
 583                 goto free_map;
 584
 585         err = bpf_map_new_fd(map, f_flags);
 586         if (err < 0) {
 587                 /* failed to allocate fd.
 588                  * bpf_map_put_with_uref() is needed because the above
 589                  * bpf_map_alloc_id() has published the map
 590                  * to the userspace and the userspace may
 591                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 592                  */
 593                 bpf_map_put_with_uref(map);
 594                 return err;
 595         }
 596
 597         return err;
 598
 599 free_map:
 600         bpf_map_release_memlock(map);
 601 free_map_sec:
 602         security_bpf_map_free(map);
 603 free_map_nouncharge:
 604         btf_put(map->btf);
 605         map->ops->map_free(map);
 606         return err;
 607 }
 608
 609 /* if error is returned, fd is released.
 610  * On success caller should complete fd access with matching fdput()
 611  */
 612 struct bpf_map *__bpf_map_get(struct fd f)
 613 {
 614         if (!f.file)
 615                 return ERR_PTR(-EBADF);
 616         if (f.file->f_op != &bpf_map_fops) {
 617                 fdput(f);
 618                 return ERR_PTR(-EINVAL);
 619         }
 620
 621         return f.file->private_data;
 622 }
 623
 624 /* prog's and map's refcnt limit */
 625 #define BPF_MAX_REFCNT 32768
 626
 627 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 628 {
 629         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 630                 atomic_dec(&map->refcnt);
 631                 return ERR_PTR(-EBUSY);
 632         }
 633         if (uref)
 634                 atomic_inc(&map->usercnt);
 635         return map;
 636 }
 637 EXPORT_SYMBOL_GPL(bpf_map_inc);
 638
 639 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 640 {
 641         struct fd f = fdget(ufd);
 642         struct bpf_map *map;
 643
 644         map = __bpf_map_get(f);
 645         if (IS_ERR(map))
 646                 return map;
 647
 648         map = bpf_map_inc(map, true);
 649         fdput(f);
 650
 651         return map;
 652 }
 653
 654 /* map_idr_lock should have been held */
 655 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 656                                             bool uref)
 657 {
 658         int refold;
 659
 660         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 661
 662         if (refold >= BPF_MAX_REFCNT) {
 663                 __bpf_map_put(map, false);
 664                 return ERR_PTR(-EBUSY);
 665         }
 666
 667         if (!refold)
 668                 return ERR_PTR(-ENOENT);
 669
 670         if (uref)
 671                 atomic_inc(&map->usercnt);
 672
 673         return map;
 674 }
 675
 676 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 677 {
 678         return -ENOTSUPP;
 679 }
 680
 681 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 682 {
 683         if (key_size)
 684                 return memdup_user(ukey, key_size);
 685
 686         if (ukey)
 687                 return ERR_PTR(-EINVAL);
 688
 689         return NULL;
 690 }
 691
 692 /* last field in 'union bpf_attr' used by this command */
 693 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 694
 695 static int map_lookup_elem(union bpf_attr *attr)
 696 {
 697         void __user *ukey = u64_to_user_ptr(attr->key);
 698         void __user *uvalue = u64_to_user_ptr(attr->value);
 699         int ufd = attr->map_fd;
 700         struct bpf_map *map;
 701         void *key, *value, *ptr;
 702         u32 value_size;
 703         struct fd f;
 704         int err;
 705
 706         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 707                 return -EINVAL;
 708
 709         if (attr->flags & ~BPF_F_LOCK)
 710                 return -EINVAL;
 711
 712         f = fdget(ufd);
 713         map = __bpf_map_get(f);
 714         if (IS_ERR(map))
 715                 return PTR_ERR(map);
 716
 717         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 718                 err = -EPERM;
 719                 goto err_put;
 720         }
 721
 722         if ((attr->flags & BPF_F_LOCK) &&
 723             !map_value_has_spin_lock(map)) {
 724                 err = -EINVAL;
 725                 goto err_put;
 726         }
 727
 728         key = __bpf_copy_key(ukey, map->key_size);
 729         if (IS_ERR(key)) {
 730                 err = PTR_ERR(key);
 731                 goto err_put;
 732         }
 733
 734         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 735             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 736             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 737             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 738                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 739         else if (IS_FD_MAP(map))
 740                 value_size = sizeof(u32);
 741         else
 742                 value_size = map->value_size;
 743
 744         err = -ENOMEM;
 745         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 746         if (!value)
 747                 goto free_key;
 748
 749         if (bpf_map_is_dev_bound(map)) {
 750                 err = bpf_map_offload_lookup_elem(map, key, value);
 751                 goto done;
 752         }
 753
 754         preempt_disable();
 755         this_cpu_inc(bpf_prog_active);
 756         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 757             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 758                 err = bpf_percpu_hash_copy(map, key, value);
 759         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 760                 err = bpf_percpu_array_copy(map, key, value);
 761         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 762                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 763         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 764                 err = bpf_stackmap_copy(map, key, value);
 765         } else if (IS_FD_ARRAY(map)) {
 766                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 767         } else if (IS_FD_HASH(map)) {
 768                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 769         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 770                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 771         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 772                    map->map_type == BPF_MAP_TYPE_STACK) {
 773                 err = map->ops->map_peek_elem(map, value);
 774         } else {
 775                 rcu_read_lock();
 776                 ptr = map->ops->map_lookup_elem(map, key);
 777                 if (IS_ERR(ptr)) {
 778                         err = PTR_ERR(ptr);
 779                 } else if (!ptr) {
 780                         err = -ENOENT;
 781                 } else {
 782                         err = 0;
 783                         if (attr->flags & BPF_F_LOCK)
 784                                 /* lock 'ptr' and copy everything but lock */
 785                                 copy_map_value_locked(map, value, ptr, true);
 786                         else
 787                                 copy_map_value(map, value, ptr);
 788                         /* mask lock, since value wasn't zero inited */
 789                         check_and_init_map_lock(map, value);
 790                 }
 791                 rcu_read_unlock();
 792         }
 793         this_cpu_dec(bpf_prog_active);
 794         preempt_enable();
 795
 796 done:
 797         if (err)
 798                 goto free_value;
 799
 800         err = -EFAULT;
 801         if (copy_to_user(uvalue, value, value_size) != 0)
 802                 goto free_value;
 803
 804         err = 0;
 805
 806 free_value:
 807         kfree(value);
 808 free_key:
 809         kfree(key);
 810 err_put:
 811         fdput(f);
 812         return err;
 813 }
 814
 815 static void maybe_wait_bpf_programs(struct bpf_map *map)
 816 {
 817         /* Wait for any running BPF programs to complete so that
 818          * userspace, when we return to it, knows that all programs
 819          * that could be running use the new map value.
 820          */
 821         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 822             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 823                 synchronize_rcu();
 824 }
 825
 826 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 827
 828 static int map_update_elem(union bpf_attr *attr)
 829 {
 830         void __user *ukey = u64_to_user_ptr(attr->key);
 831         void __user *uvalue = u64_to_user_ptr(attr->value);
 832         int ufd = attr->map_fd;
 833         struct bpf_map *map;
 834         void *key, *value;
 835         u32 value_size;
 836         struct fd f;
 837         int err;
 838
 839         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 840                 return -EINVAL;
 841
 842         f = fdget(ufd);
 843         map = __bpf_map_get(f);
 844         if (IS_ERR(map))
 845                 return PTR_ERR(map);
 846
 847         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 848                 err = -EPERM;
 849                 goto err_put;
 850         }
 851
 852         if ((attr->flags & BPF_F_LOCK) &&
 853             !map_value_has_spin_lock(map)) {
 854                 err = -EINVAL;
 855                 goto err_put;
 856         }
 857
 858         key = __bpf_copy_key(ukey, map->key_size);
 859         if (IS_ERR(key)) {
 860                 err = PTR_ERR(key);
 861                 goto err_put;
 862         }
 863
 864         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 865             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 866             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 867             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 868                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 869         else
 870                 value_size = map->value_size;
 871
 872         err = -ENOMEM;
 873         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 874         if (!value)
 875                 goto free_key;
 876
 877         err = -EFAULT;
 878         if (copy_from_user(value, uvalue, value_size) != 0)
 879                 goto free_value;
 880
 881         /* Need to create a kthread, thus must support schedule */
 882         if (bpf_map_is_dev_bound(map)) {
 883                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 884                 goto out;
 885         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 886                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 887                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 888                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 889                 goto out;
 890         }
 891
 892         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 893          * inside bpf map update or delete otherwise deadlocks are possible
 894          */
 895         preempt_disable();
 896         __this_cpu_inc(bpf_prog_active);
 897         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 898             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 899                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 900         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 901                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 902         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 903                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 904                                                        attr->flags);
 905         } else if (IS_FD_ARRAY(map)) {
 906                 rcu_read_lock();
 907                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 908                                                    attr->flags);
 909                 rcu_read_unlock();
 910         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 911                 rcu_read_lock();
 912                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 913                                                   attr->flags);
 914                 rcu_read_unlock();
 915         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 916                 /* rcu_read_lock() is not needed */
 917                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 918                                                          attr->flags);
 919         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 920                    map->map_type == BPF_MAP_TYPE_STACK) {
 921                 err = map->ops->map_push_elem(map, value, attr->flags);
 922         } else {
 923                 rcu_read_lock();
 924                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 925                 rcu_read_unlock();
 926         }
 927         __this_cpu_dec(bpf_prog_active);
 928         preempt_enable();
 929         maybe_wait_bpf_programs(map);
 930 out:
 931 free_value:
 932         kfree(value);
 933 free_key:
 934         kfree(key);
 935 err_put:
 936         fdput(f);
 937         return err;
 938 }
 939
 940 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 941
 942 static int map_delete_elem(union bpf_attr *attr)
 943 {
 944         void __user *ukey = u64_to_user_ptr(attr->key);
 945         int ufd = attr->map_fd;
 946         struct bpf_map *map;
 947         struct fd f;
 948         void *key;
 949         int err;
 950
 951         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 952                 return -EINVAL;
 953
 954         f = fdget(ufd);
 955         map = __bpf_map_get(f);
 956         if (IS_ERR(map))
 957                 return PTR_ERR(map);
 958
 959         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 960                 err = -EPERM;
 961                 goto err_put;
 962         }
 963
 964         key = __bpf_copy_key(ukey, map->key_size);
 965         if (IS_ERR(key)) {
 966                 err = PTR_ERR(key);
 967                 goto err_put;
 968         }
 969
 970         if (bpf_map_is_dev_bound(map)) {
 971                 err = bpf_map_offload_delete_elem(map, key);
 972                 goto out;
 973         }
 974
 975         preempt_disable();
 976         __this_cpu_inc(bpf_prog_active);
 977         rcu_read_lock();
 978         err = map->ops->map_delete_elem(map, key);
 979         rcu_read_unlock();
 980         __this_cpu_dec(bpf_prog_active);
 981         preempt_enable();
 982         maybe_wait_bpf_programs(map);
 983 out:
 984         kfree(key);
 985 err_put:
 986         fdput(f);
 987         return err;
 988 }
 989
 990 /* last field in 'union bpf_attr' used by this command */
 991 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 992
 993 static int map_get_next_key(union bpf_attr *attr)
 994 {
 995         void __user *ukey = u64_to_user_ptr(attr->key);
 996         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 997         int ufd = attr->map_fd;
 998         struct bpf_map *map;
 999         void *key, *next_key;
1000         struct fd f;
1001         int err;
1002
1003         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1004                 return -EINVAL;
1005
1006         f = fdget(ufd);
1007         map = __bpf_map_get(f);
1008         if (IS_ERR(map))
1009                 return PTR_ERR(map);
1010
1011         if (!(f.file->f_mode & FMODE_CAN_READ)) {
1012                 err = -EPERM;
1013                 goto err_put;
1014         }
1015
1016         if (ukey) {
1017                 key = __bpf_copy_key(ukey, map->key_size);
1018                 if (IS_ERR(key)) {
1019                         err = PTR_ERR(key);
1020                         goto err_put;
1021                 }
1022         } else {
1023                 key = NULL;
1024         }
1025
1026         err = -ENOMEM;
1027         next_key = kmalloc(map->key_size, GFP_USER);
1028         if (!next_key)
1029                 goto free_key;
1030
1031         if (bpf_map_is_dev_bound(map)) {
1032                 err = bpf_map_offload_get_next_key(map, key, next_key);
1033                 goto out;
1034         }
1035
1036         rcu_read_lock();
1037         err = map->ops->map_get_next_key(map, key, next_key);
1038         rcu_read_unlock();
1039 out:
1040         if (err)
1041                 goto free_next_key;
1042
1043         err = -EFAULT;
1044         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1045                 goto free_next_key;
1046
1047         err = 0;
1048
1049 free_next_key:
1050         kfree(next_key);
1051 free_key:
1052         kfree(key);
1053 err_put:
1054         fdput(f);
1055         return err;
1056 }
1057
1058 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1059
1060 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1061 {
1062         void __user *ukey = u64_to_user_ptr(attr->key);
1063         void __user *uvalue = u64_to_user_ptr(attr->value);
1064         int ufd = attr->map_fd;
1065         struct bpf_map *map;
1066         void *key, *value;
1067         u32 value_size;
1068         struct fd f;
1069         int err;
1070
1071         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1072                 return -EINVAL;
1073
1074         f = fdget(ufd);
1075         map = __bpf_map_get(f);
1076         if (IS_ERR(map))
1077                 return PTR_ERR(map);
1078
1079         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
1080                 err = -EPERM;
1081                 goto err_put;
1082         }
1083
1084         key = __bpf_copy_key(ukey, map->key_size);
1085         if (IS_ERR(key)) {
1086                 err = PTR_ERR(key);
1087                 goto err_put;
1088         }
1089
1090         value_size = map->value_size;
1091
1092         err = -ENOMEM;
1093         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1094         if (!value)
1095                 goto free_key;
1096
1097         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1098             map->map_type == BPF_MAP_TYPE_STACK) {
1099                 err = map->ops->map_pop_elem(map, value);
1100         } else {
1101                 err = -ENOTSUPP;
1102         }
1103
1104         if (err)
1105                 goto free_value;
1106
1107         if (copy_to_user(uvalue, value, value_size) != 0)
1108                 goto free_value;
1109
1110         err = 0;
1111
1112 free_value:
1113         kfree(value);
1114 free_key:
1115         kfree(key);
1116 err_put:
1117         fdput(f);
1118         return err;
1119 }
1120
1121 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1122 #define BPF_PROG_TYPE(_id, _name) \
1123         [_id] = & _name ## _prog_ops,
1124 #define BPF_MAP_TYPE(_id, _ops)
1125 #include <linux/bpf_types.h>
1126 #undef BPF_PROG_TYPE
1127 #undef BPF_MAP_TYPE
1128 };
1129
1130 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1131 {
1132         const struct bpf_prog_ops *ops;
1133
1134         if (type >= ARRAY_SIZE(bpf_prog_types))
1135                 return -EINVAL;
1136         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1137         ops = bpf_prog_types[type];
1138         if (!ops)
1139                 return -EINVAL;
1140
1141         if (!bpf_prog_is_dev_bound(prog->aux))
1142                 prog->aux->ops = ops;
1143         else
1144                 prog->aux->ops = &bpf_offload_prog_ops;
1145         prog->type = type;
1146         return 0;
1147 }
1148
1149 /* drop refcnt on maps used by eBPF program and free auxilary data */
1150 static void free_used_maps(struct bpf_prog_aux *aux)
1151 {
1152         enum bpf_cgroup_storage_type stype;
1153         int i;
1154
1155         for_each_cgroup_storage_type(stype) {
1156                 if (!aux->cgroup_storage[stype])
1157                         continue;
1158                 bpf_cgroup_storage_release(aux->prog,
1159                                            aux->cgroup_storage[stype]);
1160         }
1161
1162         for (i = 0; i < aux->used_map_cnt; i++)
1163                 bpf_map_put(aux->used_maps[i]);
1164
1165         kfree(aux->used_maps);
1166 }
1167
1168 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1169 {
1170         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1171         unsigned long user_bufs;
1172
1173         if (user) {
1174                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1175                 if (user_bufs > memlock_limit) {
1176                         atomic_long_sub(pages, &user->locked_vm);
1177                         return -EPERM;
1178                 }
1179         }
1180
1181         return 0;
1182 }
1183
1184 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1185 {
1186         if (user)
1187                 atomic_long_sub(pages, &user->locked_vm);
1188 }
1189
1190 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1191 {
1192         struct user_struct *user = get_current_user();
1193         int ret;
1194
1195         ret = __bpf_prog_charge(user, prog->pages);
1196         if (ret) {
1197                 free_uid(user);
1198                 return ret;
1199         }
1200
1201         prog->aux->user = user;
1202         return 0;
1203 }
1204
1205 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1206 {
1207         struct user_struct *user = prog->aux->user;
1208
1209         __bpf_prog_uncharge(user, prog->pages);
1210         free_uid(user);
1211 }
1212
1213 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1214 {
1215         int id;
1216
1217         idr_preload(GFP_KERNEL);
1218         spin_lock_bh(&prog_idr_lock);
1219         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1220         if (id > 0)
1221                 prog->aux->id = id;
1222         spin_unlock_bh(&prog_idr_lock);
1223         idr_preload_end();
1224
1225         /* id is in [1, INT_MAX) */
1226         if (WARN_ON_ONCE(!id))
1227                 return -ENOSPC;
1228
1229         return id > 0 ? 0 : id;
1230 }
1231
1232 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1233 {
1234         /* cBPF to eBPF migrations are currently not in the idr store.
1235          * Offloaded programs are removed from the store when their device
1236          * disappears - even if someone grabs an fd to them they are unusable,
1237          * simply waiting for refcnt to drop to be freed.
1238          */
1239         if (!prog->aux->id)
1240                 return;
1241
1242         if (do_idr_lock)
1243                 spin_lock_bh(&prog_idr_lock);
1244         else
1245                 __acquire(&prog_idr_lock);
1246
1247         idr_remove(&prog_idr, prog->aux->id);
1248         prog->aux->id = 0;
1249
1250         if (do_idr_lock)
1251                 spin_unlock_bh(&prog_idr_lock);
1252         else
1253                 __release(&prog_idr_lock);
1254 }
1255
1256 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1257 {
1258         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1259
1260         free_used_maps(aux);
1261         bpf_prog_uncharge_memlock(aux->prog);
1262         security_bpf_prog_free(aux);
1263         bpf_prog_free(aux->prog);
1264 }
1265
1266 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1267 {
1268         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1269                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1270                 /* bpf_prog_free_id() must be called first */
1271                 bpf_prog_free_id(prog, do_idr_lock);
1272                 bpf_prog_kallsyms_del_all(prog);
1273                 btf_put(prog->aux->btf);
1274                 kvfree(prog->aux->func_info);
1275                 bpf_prog_free_linfo(prog);
1276
1277                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1278         }
1279 }
1280
1281 void bpf_prog_put(struct bpf_prog *prog)
1282 {
1283         __bpf_prog_put(prog, true);
1284 }
1285 EXPORT_SYMBOL_GPL(bpf_prog_put);
1286
1287 static int bpf_prog_release(struct inode *inode, struct file *filp)
1288 {
1289         struct bpf_prog *prog = filp->private_data;
1290
1291         bpf_prog_put(prog);
1292         return 0;
1293 }
1294
1295 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1296                                struct bpf_prog_stats *stats)
1297 {
1298         u64 nsecs = 0, cnt = 0;
1299         int cpu;
1300
1301         for_each_possible_cpu(cpu) {
1302                 const struct bpf_prog_stats *st;
1303                 unsigned int start;
1304                 u64 tnsecs, tcnt;
1305
1306                 st = per_cpu_ptr(prog->aux->stats, cpu);
1307                 do {
1308                         start = u64_stats_fetch_begin_irq(&st->syncp);
1309                         tnsecs = st->nsecs;
1310                         tcnt = st->cnt;
1311                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1312                 nsecs += tnsecs;
1313                 cnt += tcnt;
1314         }
1315         stats->nsecs = nsecs;
1316         stats->cnt = cnt;
1317 }
1318
1319 #ifdef CONFIG_PROC_FS
1320 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1321 {
1322         const struct bpf_prog *prog = filp->private_data;
1323         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1324         struct bpf_prog_stats stats;
1325
1326         bpf_prog_get_stats(prog, &stats);
1327         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1328         seq_printf(m,
1329                    "prog_type:\t%u\n"
1330                    "prog_jited:\t%u\n"
1331                    "prog_tag:\t%s\n"
1332                    "memlock:\t%llu\n"
1333                    "prog_id:\t%u\n"
1334                    "run_time_ns:\t%llu\n"
1335                    "run_cnt:\t%llu\n",
1336                    prog->type,
1337                    prog->jited,
1338                    prog_tag,
1339                    prog->pages * 1ULL << PAGE_SHIFT,
1340                    prog->aux->id,
1341                    stats.nsecs,
1342                    stats.cnt);
1343 }
1344 #endif
1345
1346 const struct file_operations bpf_prog_fops = {
1347 #ifdef CONFIG_PROC_FS
1348         .show_fdinfo    = bpf_prog_show_fdinfo,
1349 #endif
1350         .release        = bpf_prog_release,
1351         .read           = bpf_dummy_read,
1352         .write          = bpf_dummy_write,
1353 };
1354
1355 int bpf_prog_new_fd(struct bpf_prog *prog)
1356 {
1357         int ret;
1358
1359         ret = security_bpf_prog(prog);
1360         if (ret < 0)
1361                 return ret;
1362
1363         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1364                                 O_RDWR | O_CLOEXEC);
1365 }
1366
1367 static struct bpf_prog *____bpf_prog_get(struct fd f)
1368 {
1369         if (!f.file)
1370                 return ERR_PTR(-EBADF);
1371         if (f.file->f_op != &bpf_prog_fops) {
1372                 fdput(f);
1373                 return ERR_PTR(-EINVAL);
1374         }
1375
1376         return f.file->private_data;
1377 }
1378
1379 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1380 {
1381         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1382                 atomic_sub(i, &prog->aux->refcnt);
1383                 return ERR_PTR(-EBUSY);
1384         }
1385         return prog;
1386 }
1387 EXPORT_SYMBOL_GPL(bpf_prog_add);
1388
1389 void bpf_prog_sub(struct bpf_prog *prog, int i)
1390 {
1391         /* Only to be used for undoing previous bpf_prog_add() in some
1392          * error path. We still know that another entity in our call
1393          * path holds a reference to the program, thus atomic_sub() can
1394          * be safely used in such cases!
1395          */
1396         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1397 }
1398 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1399
1400 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1401 {
1402         return bpf_prog_add(prog, 1);
1403 }
1404 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1405
1406 /* prog_idr_lock should have been held */
1407 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1408 {
1409         int refold;
1410
1411         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1412
1413         if (refold >= BPF_MAX_REFCNT) {
1414                 __bpf_prog_put(prog, false);
1415                 return ERR_PTR(-EBUSY);
1416         }
1417
1418         if (!refold)
1419                 return ERR_PTR(-ENOENT);
1420
1421         return prog;
1422 }
1423 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1424
1425 bool bpf_prog_get_ok(struct bpf_prog *prog,
1426                             enum bpf_prog_type *attach_type, bool attach_drv)
1427 {
1428         /* not an attachment, just a refcount inc, always allow */
1429         if (!attach_type)
1430                 return true;
1431
1432         if (prog->type != *attach_type)
1433                 return false;
1434         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1435                 return false;
1436
1437         return true;
1438 }
1439
1440 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1441                                        bool attach_drv)
1442 {
1443         struct fd f = fdget(ufd);
1444         struct bpf_prog *prog;
1445
1446         prog = ____bpf_prog_get(f);
1447         if (IS_ERR(prog))
1448                 return prog;
1449         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1450                 prog = ERR_PTR(-EINVAL);
1451                 goto out;
1452         }
1453
1454         prog = bpf_prog_inc(prog);
1455 out:
1456         fdput(f);
1457         return prog;
1458 }
1459
1460 struct bpf_prog *bpf_prog_get(u32 ufd)
1461 {
1462         return __bpf_prog_get(ufd, NULL, false);
1463 }
1464
1465 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1466                                        bool attach_drv)
1467 {
1468         return __bpf_prog_get(ufd, &type, attach_drv);
1469 }
1470 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1471
1472 /* Initially all BPF programs could be loaded w/o specifying
1473  * expected_attach_type. Later for some of them specifying expected_attach_type
1474  * at load time became required so that program could be validated properly.
1475  * Programs of types that are allowed to be loaded both w/ and w/o (for
1476  * backward compatibility) expected_attach_type, should have the default attach
1477  * type assigned to expected_attach_type for the latter case, so that it can be
1478  * validated later at attach time.
1479  *
1480  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1481  * prog type requires it but has some attach types that have to be backward
1482  * compatible.
1483  */
1484 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1485 {
1486         switch (attr->prog_type) {
1487         case BPF_PROG_TYPE_CGROUP_SOCK:
1488                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1489                  * exist so checking for non-zero is the way to go here.
1490                  */
1491                 if (!attr->expected_attach_type)
1492                         attr->expected_attach_type =
1493                                 BPF_CGROUP_INET_SOCK_CREATE;
1494                 break;
1495         }
1496 }
1497
1498 static int
1499 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1500                                 enum bpf_attach_type expected_attach_type)
1501 {
1502         switch (prog_type) {
1503         case BPF_PROG_TYPE_CGROUP_SOCK:
1504                 switch (expected_attach_type) {
1505                 case BPF_CGROUP_INET_SOCK_CREATE:
1506                 case BPF_CGROUP_INET4_POST_BIND:
1507                 case BPF_CGROUP_INET6_POST_BIND:
1508                         return 0;
1509                 default:
1510                         return -EINVAL;
1511                 }
1512         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1513                 switch (expected_attach_type) {
1514                 case BPF_CGROUP_INET4_BIND:
1515                 case BPF_CGROUP_INET6_BIND:
1516                 case BPF_CGROUP_INET4_CONNECT:
1517                 case BPF_CGROUP_INET6_CONNECT:
1518                 case BPF_CGROUP_UDP4_SENDMSG:
1519                 case BPF_CGROUP_UDP6_SENDMSG:
1520                         return 0;
1521                 default:
1522                         return -EINVAL;
1523                 }
1524         default:
1525                 return 0;
1526         }
1527 }
1528
1529 /* last field in 'union bpf_attr' used by this command */
1530 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1531
1532 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1533 {
1534         enum bpf_prog_type type = attr->prog_type;
1535         struct bpf_prog *prog;
1536         int err;
1537         char license[128];
1538         bool is_gpl;
1539
1540         if (CHECK_ATTR(BPF_PROG_LOAD))
1541                 return -EINVAL;
1542
1543         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1544                 return -EINVAL;
1545
1546         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1547             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1548             !capable(CAP_SYS_ADMIN))
1549                 return -EPERM;
1550
1551         /* copy eBPF program license from user space */
1552         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1553                               sizeof(license) - 1) < 0)
1554                 return -EFAULT;
1555         license[sizeof(license) - 1] = 0;
1556
1557         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1558         is_gpl = license_is_gpl_compatible(license);
1559
1560         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1561                 return -E2BIG;
1562         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1563             type != BPF_PROG_TYPE_CGROUP_SKB &&
1564             !capable(CAP_SYS_ADMIN))
1565                 return -EPERM;
1566
1567         bpf_prog_load_fixup_attach_type(attr);
1568         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1569                 return -EINVAL;
1570
1571         /* plain bpf_prog allocation */
1572         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1573         if (!prog)
1574                 return -ENOMEM;
1575
1576         prog->expected_attach_type = attr->expected_attach_type;
1577
1578         prog->aux->offload_requested = !!attr->prog_ifindex;
1579
1580         err = security_bpf_prog_alloc(prog->aux);
1581         if (err)
1582                 goto free_prog_nouncharge;
1583
1584         err = bpf_prog_charge_memlock(prog);
1585         if (err)
1586                 goto free_prog_sec;
1587
1588         prog->len = attr->insn_cnt;
1589
1590         err = -EFAULT;
1591         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1592                            bpf_prog_insn_size(prog)) != 0)
1593                 goto free_prog;
1594
1595         prog->orig_prog = NULL;
1596         prog->jited = 0;
1597
1598         atomic_set(&prog->aux->refcnt, 1);
1599         prog->gpl_compatible = is_gpl ? 1 : 0;
1600
1601         if (bpf_prog_is_dev_bound(prog->aux)) {
1602                 err = bpf_prog_offload_init(prog, attr);
1603                 if (err)
1604                         goto free_prog;
1605         }
1606
1607         /* find program type: socket_filter vs tracing_filter */
1608         err = find_prog_type(type, prog);
1609         if (err < 0)
1610                 goto free_prog;
1611
1612         prog->aux->load_time = ktime_get_boot_ns();
1613         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1614         if (err)
1615                 goto free_prog;
1616
1617         /* run eBPF verifier */
1618         err = bpf_check(&prog, attr, uattr);
1619         if (err < 0)
1620                 goto free_used_maps;
1621
1622         prog = bpf_prog_select_runtime(prog, &err);
1623         if (err < 0)
1624                 goto free_used_maps;
1625
1626         err = bpf_prog_alloc_id(prog);
1627         if (err)
1628                 goto free_used_maps;
1629
1630         err = bpf_prog_new_fd(prog);
1631         if (err < 0) {
1632                 /* failed to allocate fd.
1633                  * bpf_prog_put() is needed because the above
1634                  * bpf_prog_alloc_id() has published the prog
1635                  * to the userspace and the userspace may
1636                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1637                  */
1638                 bpf_prog_put(prog);
1639                 return err;
1640         }
1641
1642         bpf_prog_kallsyms_add(prog);
1643         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1644         return err;
1645
1646 free_used_maps:
1647         bpf_prog_free_linfo(prog);
1648         kvfree(prog->aux->func_info);
1649         btf_put(prog->aux->btf);
1650         bpf_prog_kallsyms_del_subprogs(prog);
1651         free_used_maps(prog->aux);
1652 free_prog:
1653         bpf_prog_uncharge_memlock(prog);
1654 free_prog_sec:
1655         security_bpf_prog_free(prog->aux);
1656 free_prog_nouncharge:
1657         bpf_prog_free(prog);
1658         return err;
1659 }
1660
1661 #define BPF_OBJ_LAST_FIELD file_flags
1662
1663 static int bpf_obj_pin(const union bpf_attr *attr)
1664 {
1665         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1666                 return -EINVAL;
1667
1668         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1669 }
1670
1671 static int bpf_obj_get(const union bpf_attr *attr)
1672 {
1673         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1674             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1675                 return -EINVAL;
1676
1677         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1678                                 attr->file_flags);
1679 }
1680
1681 struct bpf_raw_tracepoint {
1682         struct bpf_raw_event_map *btp;
1683         struct bpf_prog *prog;
1684 };
1685
1686 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1687 {
1688         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1689
1690         if (raw_tp->prog) {
1691                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1692                 bpf_prog_put(raw_tp->prog);
1693         }
1694         bpf_put_raw_tracepoint(raw_tp->btp);
1695         kfree(raw_tp);
1696         return 0;
1697 }
1698
1699 static const struct file_operations bpf_raw_tp_fops = {
1700         .release        = bpf_raw_tracepoint_release,
1701         .read           = bpf_dummy_read,
1702         .write          = bpf_dummy_write,
1703 };
1704
1705 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1706
1707 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1708 {
1709         struct bpf_raw_tracepoint *raw_tp;
1710         struct bpf_raw_event_map *btp;
1711         struct bpf_prog *prog;
1712         char tp_name[128];
1713         int tp_fd, err;
1714
1715         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1716                               sizeof(tp_name) - 1) < 0)
1717                 return -EFAULT;
1718         tp_name[sizeof(tp_name) - 1] = 0;
1719
1720         btp = bpf_get_raw_tracepoint(tp_name);
1721         if (!btp)
1722                 return -ENOENT;
1723
1724         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1725         if (!raw_tp) {
1726                 err = -ENOMEM;
1727                 goto out_put_btp;
1728         }
1729         raw_tp->btp = btp;
1730
1731         prog = bpf_prog_get_type(attr->raw_tracepoint.prog_fd,
1732                                  BPF_PROG_TYPE_RAW_TRACEPOINT);
1733         if (IS_ERR(prog)) {
1734                 err = PTR_ERR(prog);
1735                 goto out_free_tp;
1736         }
1737
1738         err = bpf_probe_register(raw_tp->btp, prog);
1739         if (err)
1740                 goto out_put_prog;
1741
1742         raw_tp->prog = prog;
1743         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1744                                  O_CLOEXEC);
1745         if (tp_fd < 0) {
1746                 bpf_probe_unregister(raw_tp->btp, prog);
1747                 err = tp_fd;
1748                 goto out_put_prog;
1749         }
1750         return tp_fd;
1751
1752 out_put_prog:
1753         bpf_prog_put(prog);
1754 out_free_tp:
1755         kfree(raw_tp);
1756 out_put_btp:
1757         bpf_put_raw_tracepoint(btp);
1758         return err;
1759 }
1760
1761 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1762                                              enum bpf_attach_type attach_type)
1763 {
1764         switch (prog->type) {
1765         case BPF_PROG_TYPE_CGROUP_SOCK:
1766         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1767                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1768         default:
1769                 return 0;
1770         }
1771 }
1772
1773 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1774
1775 #define BPF_F_ATTACH_MASK \
1776         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1777
1778 static int bpf_prog_attach(const union bpf_attr *attr)
1779 {
1780         enum bpf_prog_type ptype;
1781         struct bpf_prog *prog;
1782         int ret;
1783
1784         if (!capable(CAP_NET_ADMIN))
1785                 return -EPERM;
1786
1787         if (CHECK_ATTR(BPF_PROG_ATTACH))
1788                 return -EINVAL;
1789
1790         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1791                 return -EINVAL;
1792
1793         switch (attr->attach_type) {
1794         case BPF_CGROUP_INET_INGRESS:
1795         case BPF_CGROUP_INET_EGRESS:
1796                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1797                 break;
1798         case BPF_CGROUP_INET_SOCK_CREATE:
1799         case BPF_CGROUP_INET4_POST_BIND:
1800         case BPF_CGROUP_INET6_POST_BIND:
1801                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1802                 break;
1803         case BPF_CGROUP_INET4_BIND:
1804         case BPF_CGROUP_INET6_BIND:
1805         case BPF_CGROUP_INET4_CONNECT:
1806         case BPF_CGROUP_INET6_CONNECT:
1807         case BPF_CGROUP_UDP4_SENDMSG:
1808         case BPF_CGROUP_UDP6_SENDMSG:
1809                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1810                 break;
1811         case BPF_CGROUP_SOCK_OPS:
1812                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1813                 break;
1814         case BPF_CGROUP_DEVICE:
1815                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1816                 break;
1817         case BPF_SK_MSG_VERDICT:
1818                 ptype = BPF_PROG_TYPE_SK_MSG;
1819                 break;
1820         case BPF_SK_SKB_STREAM_PARSER:
1821         case BPF_SK_SKB_STREAM_VERDICT:
1822                 ptype = BPF_PROG_TYPE_SK_SKB;
1823                 break;
1824         case BPF_LIRC_MODE2:
1825                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1826                 break;
1827         case BPF_FLOW_DISSECTOR:
1828                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1829                 break;
1830         default:
1831                 return -EINVAL;
1832         }
1833
1834         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1835         if (IS_ERR(prog))
1836                 return PTR_ERR(prog);
1837
1838         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1839                 bpf_prog_put(prog);
1840                 return -EINVAL;
1841         }
1842
1843         switch (ptype) {
1844         case BPF_PROG_TYPE_SK_SKB:
1845         case BPF_PROG_TYPE_SK_MSG:
1846                 ret = sock_map_get_from_fd(attr, prog);
1847                 break;
1848         case BPF_PROG_TYPE_LIRC_MODE2:
1849                 ret = lirc_prog_attach(attr, prog);
1850                 break;
1851         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1852                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1853                 break;
1854         default:
1855                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1856         }
1857
1858         if (ret)
1859                 bpf_prog_put(prog);
1860         return ret;
1861 }
1862
1863 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1864
1865 static int bpf_prog_detach(const union bpf_attr *attr)
1866 {
1867         enum bpf_prog_type ptype;
1868
1869         if (!capable(CAP_NET_ADMIN))
1870                 return -EPERM;
1871
1872         if (CHECK_ATTR(BPF_PROG_DETACH))
1873                 return -EINVAL;
1874
1875         switch (attr->attach_type) {
1876         case BPF_CGROUP_INET_INGRESS:
1877         case BPF_CGROUP_INET_EGRESS:
1878                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1879                 break;
1880         case BPF_CGROUP_INET_SOCK_CREATE:
1881         case BPF_CGROUP_INET4_POST_BIND:
1882         case BPF_CGROUP_INET6_POST_BIND:
1883                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1884                 break;
1885         case BPF_CGROUP_INET4_BIND:
1886         case BPF_CGROUP_INET6_BIND:
1887         case BPF_CGROUP_INET4_CONNECT:
1888         case BPF_CGROUP_INET6_CONNECT:
1889         case BPF_CGROUP_UDP4_SENDMSG:
1890         case BPF_CGROUP_UDP6_SENDMSG:
1891                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1892                 break;
1893         case BPF_CGROUP_SOCK_OPS:
1894                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1895                 break;
1896         case BPF_CGROUP_DEVICE:
1897                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1898                 break;
1899         case BPF_SK_MSG_VERDICT:
1900                 return sock_map_get_from_fd(attr, NULL);
1901         case BPF_SK_SKB_STREAM_PARSER:
1902         case BPF_SK_SKB_STREAM_VERDICT:
1903                 return sock_map_get_from_fd(attr, NULL);
1904         case BPF_LIRC_MODE2:
1905                 return lirc_prog_detach(attr);
1906         case BPF_FLOW_DISSECTOR:
1907                 return skb_flow_dissector_bpf_prog_detach(attr);
1908         default:
1909                 return -EINVAL;
1910         }
1911
1912         return cgroup_bpf_prog_detach(attr, ptype);
1913 }
1914
1915 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1916
1917 static int bpf_prog_query(const union bpf_attr *attr,
1918                           union bpf_attr __user *uattr)
1919 {
1920         if (!capable(CAP_NET_ADMIN))
1921                 return -EPERM;
1922         if (CHECK_ATTR(BPF_PROG_QUERY))
1923                 return -EINVAL;
1924         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1925                 return -EINVAL;
1926
1927         switch (attr->query.attach_type) {
1928         case BPF_CGROUP_INET_INGRESS:
1929         case BPF_CGROUP_INET_EGRESS:
1930         case BPF_CGROUP_INET_SOCK_CREATE:
1931         case BPF_CGROUP_INET4_BIND:
1932         case BPF_CGROUP_INET6_BIND:
1933         case BPF_CGROUP_INET4_POST_BIND:
1934         case BPF_CGROUP_INET6_POST_BIND:
1935         case BPF_CGROUP_INET4_CONNECT:
1936         case BPF_CGROUP_INET6_CONNECT:
1937         case BPF_CGROUP_UDP4_SENDMSG:
1938         case BPF_CGROUP_UDP6_SENDMSG:
1939         case BPF_CGROUP_SOCK_OPS:
1940         case BPF_CGROUP_DEVICE:
1941                 break;
1942         case BPF_LIRC_MODE2:
1943                 return lirc_prog_query(attr, uattr);
1944         default:
1945                 return -EINVAL;
1946         }
1947
1948         return cgroup_bpf_prog_query(attr, uattr);
1949 }
1950
1951 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1952
1953 static int bpf_prog_test_run(const union bpf_attr *attr,
1954                              union bpf_attr __user *uattr)
1955 {
1956         struct bpf_prog *prog;
1957         int ret = -ENOTSUPP;
1958
1959         if (!capable(CAP_SYS_ADMIN))
1960                 return -EPERM;
1961         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1962                 return -EINVAL;
1963
1964         prog = bpf_prog_get(attr->test.prog_fd);
1965         if (IS_ERR(prog))
1966                 return PTR_ERR(prog);
1967
1968         if (prog->aux->ops->test_run)
1969                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1970
1971         bpf_prog_put(prog);
1972         return ret;
1973 }
1974
1975 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1976
1977 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1978                                union bpf_attr __user *uattr,
1979                                struct idr *idr,
1980                                spinlock_t *lock)
1981 {
1982         u32 next_id = attr->start_id;
1983         int err = 0;
1984
1985         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1986                 return -EINVAL;
1987
1988         if (!capable(CAP_SYS_ADMIN))
1989                 return -EPERM;
1990
1991         next_id++;
1992         spin_lock_bh(lock);
1993         if (!idr_get_next(idr, &next_id))
1994                 err = -ENOENT;
1995         spin_unlock_bh(lock);
1996
1997         if (!err)
1998                 err = put_user(next_id, &uattr->next_id);
1999
2000         return err;
2001 }
2002
2003 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2004
2005 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2006 {
2007         struct bpf_prog *prog;
2008         u32 id = attr->prog_id;
2009         int fd;
2010
2011         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2012                 return -EINVAL;
2013
2014         if (!capable(CAP_SYS_ADMIN))
2015                 return -EPERM;
2016
2017         spin_lock_bh(&prog_idr_lock);
2018         prog = idr_find(&prog_idr, id);
2019         if (prog)
2020                 prog = bpf_prog_inc_not_zero(prog);
2021         else
2022                 prog = ERR_PTR(-ENOENT);
2023         spin_unlock_bh(&prog_idr_lock);
2024
2025         if (IS_ERR(prog))
2026                 return PTR_ERR(prog);
2027
2028         fd = bpf_prog_new_fd(prog);
2029         if (fd < 0)
2030                 bpf_prog_put(prog);
2031
2032         return fd;
2033 }
2034
2035 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2036
2037 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2038 {
2039         struct bpf_map *map;
2040         u32 id = attr->map_id;
2041         int f_flags;
2042         int fd;
2043
2044         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2045             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2046                 return -EINVAL;
2047
2048         if (!capable(CAP_SYS_ADMIN))
2049                 return -EPERM;
2050
2051         f_flags = bpf_get_file_flag(attr->open_flags);
2052         if (f_flags < 0)
2053                 return f_flags;
2054
2055         spin_lock_bh(&map_idr_lock);
2056         map = idr_find(&map_idr, id);
2057         if (map)
2058                 map = bpf_map_inc_not_zero(map, true);
2059         else
2060                 map = ERR_PTR(-ENOENT);
2061         spin_unlock_bh(&map_idr_lock);
2062
2063         if (IS_ERR(map))
2064                 return PTR_ERR(map);
2065
2066         fd = bpf_map_new_fd(map, f_flags);
2067         if (fd < 0)
2068                 bpf_map_put_with_uref(map);
2069
2070         return fd;
2071 }
2072
2073 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2074                                               unsigned long addr)
2075 {
2076         int i;
2077
2078         for (i = 0; i < prog->aux->used_map_cnt; i++)
2079                 if (prog->aux->used_maps[i] == (void *)addr)
2080                         return prog->aux->used_maps[i];
2081         return NULL;
2082 }
2083
2084 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2085 {
2086         const struct bpf_map *map;
2087         struct bpf_insn *insns;
2088         u64 imm;
2089         int i;
2090
2091         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2092                         GFP_USER);
2093         if (!insns)
2094                 return insns;
2095
2096         for (i = 0; i < prog->len; i++) {
2097                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2098                         insns[i].code = BPF_JMP | BPF_CALL;
2099                         insns[i].imm = BPF_FUNC_tail_call;
2100                         /* fall-through */
2101                 }
2102                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2103                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2104                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2105                                 insns[i].code = BPF_JMP | BPF_CALL;
2106                         if (!bpf_dump_raw_ok())
2107                                 insns[i].imm = 0;
2108                         continue;
2109                 }
2110
2111                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2112                         continue;
2113
2114                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2115                 map = bpf_map_from_imm(prog, imm);
2116                 if (map) {
2117                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
2118                         insns[i].imm = map->id;
2119                         insns[i + 1].imm = 0;
2120                         continue;
2121                 }
2122         }
2123
2124         return insns;
2125 }
2126
2127 static int set_info_rec_size(struct bpf_prog_info *info)
2128 {
2129         /*
2130          * Ensure info.*_rec_size is the same as kernel expected size
2131          *
2132          * or
2133          *
2134          * Only allow zero *_rec_size if both _rec_size and _cnt are
2135          * zero.  In this case, the kernel will set the expected
2136          * _rec_size back to the info.
2137          */
2138
2139         if ((info->nr_func_info || info->func_info_rec_size) &&
2140             info->func_info_rec_size != sizeof(struct bpf_func_info))
2141                 return -EINVAL;
2142
2143         if ((info->nr_line_info || info->line_info_rec_size) &&
2144             info->line_info_rec_size != sizeof(struct bpf_line_info))
2145                 return -EINVAL;
2146
2147         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2148             info->jited_line_info_rec_size != sizeof(__u64))
2149                 return -EINVAL;
2150
2151         info->func_info_rec_size = sizeof(struct bpf_func_info);
2152         info->line_info_rec_size = sizeof(struct bpf_line_info);
2153         info->jited_line_info_rec_size = sizeof(__u64);
2154
2155         return 0;
2156 }
2157
2158 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2159                                    const union bpf_attr *attr,
2160                                    union bpf_attr __user *uattr)
2161 {
2162         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2163         struct bpf_prog_info info = {};
2164         u32 info_len = attr->info.info_len;
2165         struct bpf_prog_stats stats;
2166         char __user *uinsns;
2167         u32 ulen;
2168         int err;
2169
2170         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2171         if (err)
2172                 return err;
2173         info_len = min_t(u32, sizeof(info), info_len);
2174
2175         if (copy_from_user(&info, uinfo, info_len))
2176                 return -EFAULT;
2177
2178         info.type = prog->type;
2179         info.id = prog->aux->id;
2180         info.load_time = prog->aux->load_time;
2181         info.created_by_uid = from_kuid_munged(current_user_ns(),
2182                                                prog->aux->user->uid);
2183         info.gpl_compatible = prog->gpl_compatible;
2184
2185         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2186         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2187
2188         ulen = info.nr_map_ids;
2189         info.nr_map_ids = prog->aux->used_map_cnt;
2190         ulen = min_t(u32, info.nr_map_ids, ulen);
2191         if (ulen) {
2192                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2193                 u32 i;
2194
2195                 for (i = 0; i < ulen; i++)
2196                         if (put_user(prog->aux->used_maps[i]->id,
2197                                      &user_map_ids[i]))
2198                                 return -EFAULT;
2199         }
2200
2201         err = set_info_rec_size(&info);
2202         if (err)
2203                 return err;
2204
2205         bpf_prog_get_stats(prog, &stats);
2206         info.run_time_ns = stats.nsecs;
2207         info.run_cnt = stats.cnt;
2208
2209         if (!capable(CAP_SYS_ADMIN)) {
2210                 info.jited_prog_len = 0;
2211                 info.xlated_prog_len = 0;
2212                 info.nr_jited_ksyms = 0;
2213                 info.nr_jited_func_lens = 0;
2214                 info.nr_func_info = 0;
2215                 info.nr_line_info = 0;
2216                 info.nr_jited_line_info = 0;
2217                 goto done;
2218         }
2219
2220         ulen = info.xlated_prog_len;
2221         info.xlated_prog_len = bpf_prog_insn_size(prog);
2222         if (info.xlated_prog_len && ulen) {
2223                 struct bpf_insn *insns_sanitized;
2224                 bool fault;
2225
2226                 if (prog->blinded && !bpf_dump_raw_ok()) {
2227                         info.xlated_prog_insns = 0;
2228                         goto done;
2229                 }
2230                 insns_sanitized = bpf_insn_prepare_dump(prog);
2231                 if (!insns_sanitized)
2232                         return -ENOMEM;
2233                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2234                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2235                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2236                 kfree(insns_sanitized);
2237                 if (fault)
2238                         return -EFAULT;
2239         }
2240
2241         if (bpf_prog_is_dev_bound(prog->aux)) {
2242                 err = bpf_prog_offload_info_fill(&info, prog);
2243                 if (err)
2244                         return err;
2245                 goto done;
2246         }
2247
2248         /* NOTE: the following code is supposed to be skipped for offload.
2249          * bpf_prog_offload_info_fill() is the place to fill similar fields
2250          * for offload.
2251          */
2252         ulen = info.jited_prog_len;
2253         if (prog->aux->func_cnt) {
2254                 u32 i;
2255
2256                 info.jited_prog_len = 0;
2257                 for (i = 0; i < prog->aux->func_cnt; i++)
2258                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2259         } else {
2260                 info.jited_prog_len = prog->jited_len;
2261         }
2262
2263         if (info.jited_prog_len && ulen) {
2264                 if (bpf_dump_raw_ok()) {
2265                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2266                         ulen = min_t(u32, info.jited_prog_len, ulen);
2267
2268                         /* for multi-function programs, copy the JITed
2269                          * instructions for all the functions
2270                          */
2271                         if (prog->aux->func_cnt) {
2272                                 u32 len, free, i;
2273                                 u8 *img;
2274
2275                                 free = ulen;
2276                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2277                                         len = prog->aux->func[i]->jited_len;
2278                                         len = min_t(u32, len, free);
2279                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2280                                         if (copy_to_user(uinsns, img, len))
2281                                                 return -EFAULT;
2282                                         uinsns += len;
2283                                         free -= len;
2284                                         if (!free)
2285                                                 break;
2286                                 }
2287                         } else {
2288                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2289                                         return -EFAULT;
2290                         }
2291                 } else {
2292                         info.jited_prog_insns = 0;
2293                 }
2294         }
2295
2296         ulen = info.nr_jited_ksyms;
2297         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2298         if (ulen) {
2299                 if (bpf_dump_raw_ok()) {
2300                         unsigned long ksym_addr;
2301                         u64 __user *user_ksyms;
2302                         u32 i;
2303
2304                         /* copy the address of the kernel symbol
2305                          * corresponding to each function
2306                          */
2307                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2308                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2309                         if (prog->aux->func_cnt) {
2310                                 for (i = 0; i < ulen; i++) {
2311                                         ksym_addr = (unsigned long)
2312                                                 prog->aux->func[i]->bpf_func;
2313                                         if (put_user((u64) ksym_addr,
2314                                                      &user_ksyms[i]))
2315                                                 return -EFAULT;
2316                                 }
2317                         } else {
2318                                 ksym_addr = (unsigned long) prog->bpf_func;
2319                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2320                                         return -EFAULT;
2321                         }
2322                 } else {
2323                         info.jited_ksyms = 0;
2324                 }
2325         }
2326
2327         ulen = info.nr_jited_func_lens;
2328         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2329         if (ulen) {
2330                 if (bpf_dump_raw_ok()) {
2331                         u32 __user *user_lens;
2332                         u32 func_len, i;
2333
2334                         /* copy the JITed image lengths for each function */
2335                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2336                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2337                         if (prog->aux->func_cnt) {
2338                                 for (i = 0; i < ulen; i++) {
2339                                         func_len =
2340                                                 prog->aux->func[i]->jited_len;
2341                                         if (put_user(func_len, &user_lens[i]))
2342                                                 return -EFAULT;
2343                                 }
2344                         } else {
2345                                 func_len = prog->jited_len;
2346                                 if (put_user(func_len, &user_lens[0]))
2347                                         return -EFAULT;
2348                         }
2349                 } else {
2350                         info.jited_func_lens = 0;
2351                 }
2352         }
2353
2354         if (prog->aux->btf)
2355                 info.btf_id = btf_id(prog->aux->btf);
2356
2357         ulen = info.nr_func_info;
2358         info.nr_func_info = prog->aux->func_info_cnt;
2359         if (info.nr_func_info && ulen) {
2360                 char __user *user_finfo;
2361
2362                 user_finfo = u64_to_user_ptr(info.func_info);
2363                 ulen = min_t(u32, info.nr_func_info, ulen);
2364                 if (copy_to_user(user_finfo, prog->aux->func_info,
2365                                  info.func_info_rec_size * ulen))
2366                         return -EFAULT;
2367         }
2368
2369         ulen = info.nr_line_info;
2370         info.nr_line_info = prog->aux->nr_linfo;
2371         if (info.nr_line_info && ulen) {
2372                 __u8 __user *user_linfo;
2373
2374                 user_linfo = u64_to_user_ptr(info.line_info);
2375                 ulen = min_t(u32, info.nr_line_info, ulen);
2376                 if (copy_to_user(user_linfo, prog->aux->linfo,
2377                                  info.line_info_rec_size * ulen))
2378                         return -EFAULT;
2379         }
2380
2381         ulen = info.nr_jited_line_info;
2382         if (prog->aux->jited_linfo)
2383                 info.nr_jited_line_info = prog->aux->nr_linfo;
2384         else
2385                 info.nr_jited_line_info = 0;
2386         if (info.nr_jited_line_info && ulen) {
2387                 if (bpf_dump_raw_ok()) {
2388                         __u64 __user *user_linfo;
2389                         u32 i;
2390
2391                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2392                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2393                         for (i = 0; i < ulen; i++) {
2394                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2395                                              &user_linfo[i]))
2396                                         return -EFAULT;
2397                         }
2398                 } else {
2399                         info.jited_line_info = 0;
2400                 }
2401         }
2402
2403         ulen = info.nr_prog_tags;
2404         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2405         if (ulen) {
2406                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2407                 u32 i;
2408
2409                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2410                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2411                 if (prog->aux->func_cnt) {
2412                         for (i = 0; i < ulen; i++) {
2413                                 if (copy_to_user(user_prog_tags[i],
2414                                                  prog->aux->func[i]->tag,
2415                                                  BPF_TAG_SIZE))
2416                                         return -EFAULT;
2417                         }
2418                 } else {
2419                         if (copy_to_user(user_prog_tags[0],
2420                                          prog->tag, BPF_TAG_SIZE))
2421                                 return -EFAULT;
2422                 }
2423         }
2424
2425 done:
2426         if (copy_to_user(uinfo, &info, info_len) ||
2427             put_user(info_len, &uattr->info.info_len))
2428                 return -EFAULT;
2429
2430         return 0;
2431 }
2432
2433 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2434                                   const union bpf_attr *attr,
2435                                   union bpf_attr __user *uattr)
2436 {
2437         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2438         struct bpf_map_info info = {};
2439         u32 info_len = attr->info.info_len;
2440         int err;
2441
2442         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2443         if (err)
2444                 return err;
2445         info_len = min_t(u32, sizeof(info), info_len);
2446
2447         info.type = map->map_type;
2448         info.id = map->id;
2449         info.key_size = map->key_size;
2450         info.value_size = map->value_size;
2451         info.max_entries = map->max_entries;
2452         info.map_flags = map->map_flags;
2453         memcpy(info.name, map->name, sizeof(map->name));
2454
2455         if (map->btf) {
2456                 info.btf_id = btf_id(map->btf);
2457                 info.btf_key_type_id = map->btf_key_type_id;
2458                 info.btf_value_type_id = map->btf_value_type_id;
2459         }
2460
2461         if (bpf_map_is_dev_bound(map)) {
2462                 err = bpf_map_offload_info_fill(&info, map);
2463                 if (err)
2464                         return err;
2465         }
2466
2467         if (copy_to_user(uinfo, &info, info_len) ||
2468             put_user(info_len, &uattr->info.info_len))
2469                 return -EFAULT;
2470
2471         return 0;
2472 }
2473
2474 static int bpf_btf_get_info_by_fd(struct btf *btf,
2475                                   const union bpf_attr *attr,
2476                                   union bpf_attr __user *uattr)
2477 {
2478         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2479         u32 info_len = attr->info.info_len;
2480         int err;
2481
2482         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2483         if (err)
2484                 return err;
2485
2486         return btf_get_info_by_fd(btf, attr, uattr);
2487 }
2488
2489 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2490
2491 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2492                                   union bpf_attr __user *uattr)
2493 {
2494         int ufd = attr->info.bpf_fd;
2495         struct fd f;
2496         int err;
2497
2498         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2499                 return -EINVAL;
2500
2501         f = fdget(ufd);
2502         if (!f.file)
2503                 return -EBADFD;
2504
2505         if (f.file->f_op == &bpf_prog_fops)
2506                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2507                                               uattr);
2508         else if (f.file->f_op == &bpf_map_fops)
2509                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2510                                              uattr);
2511         else if (f.file->f_op == &btf_fops)
2512                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2513         else
2514                 err = -EINVAL;
2515
2516         fdput(f);
2517         return err;
2518 }
2519
2520 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2521
2522 static int bpf_btf_load(const union bpf_attr *attr)
2523 {
2524         if (CHECK_ATTR(BPF_BTF_LOAD))
2525                 return -EINVAL;
2526
2527         if (!capable(CAP_SYS_ADMIN))
2528                 return -EPERM;
2529
2530         return btf_new_fd(attr);
2531 }
2532
2533 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2534
2535 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2536 {
2537         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2538                 return -EINVAL;
2539
2540         if (!capable(CAP_SYS_ADMIN))
2541                 return -EPERM;
2542
2543         return btf_get_fd_by_id(attr->btf_id);
2544 }
2545
2546 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2547                                     union bpf_attr __user *uattr,
2548                                     u32 prog_id, u32 fd_type,
2549                                     const char *buf, u64 probe_offset,
2550                                     u64 probe_addr)
2551 {
2552         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2553         u32 len = buf ? strlen(buf) : 0, input_len;
2554         int err = 0;
2555
2556         if (put_user(len, &uattr->task_fd_query.buf_len))
2557                 return -EFAULT;
2558         input_len = attr->task_fd_query.buf_len;
2559         if (input_len && ubuf) {
2560                 if (!len) {
2561                         /* nothing to copy, just make ubuf NULL terminated */
2562                         char zero = '\0';
2563
2564                         if (put_user(zero, ubuf))
2565                                 return -EFAULT;
2566                 } else if (input_len >= len + 1) {
2567                         /* ubuf can hold the string with NULL terminator */
2568                         if (copy_to_user(ubuf, buf, len + 1))
2569                                 return -EFAULT;
2570                 } else {
2571                         /* ubuf cannot hold the string with NULL terminator,
2572                          * do a partial copy with NULL terminator.
2573                          */
2574                         char zero = '\0';
2575
2576                         err = -ENOSPC;
2577                         if (copy_to_user(ubuf, buf, input_len - 1))
2578                                 return -EFAULT;
2579                         if (put_user(zero, ubuf + input_len - 1))
2580                                 return -EFAULT;
2581                 }
2582         }
2583
2584         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2585             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2586             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2587             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2588                 return -EFAULT;
2589
2590         return err;
2591 }
2592
2593 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2594
2595 static int bpf_task_fd_query(const union bpf_attr *attr,
2596                              union bpf_attr __user *uattr)
2597 {
2598         pid_t pid = attr->task_fd_query.pid;
2599         u32 fd = attr->task_fd_query.fd;
2600         const struct perf_event *event;
2601         struct files_struct *files;
2602         struct task_struct *task;
2603         struct file *file;
2604         int err;
2605
2606         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2607                 return -EINVAL;
2608
2609         if (!capable(CAP_SYS_ADMIN))
2610                 return -EPERM;
2611
2612         if (attr->task_fd_query.flags != 0)
2613                 return -EINVAL;
2614
2615         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2616         if (!task)
2617                 return -ENOENT;
2618
2619         files = get_files_struct(task);
2620         put_task_struct(task);
2621         if (!files)
2622                 return -ENOENT;
2623
2624         err = 0;
2625         spin_lock(&files->file_lock);
2626         file = fcheck_files(files, fd);
2627         if (!file)
2628                 err = -EBADF;
2629         else
2630                 get_file(file);
2631         spin_unlock(&files->file_lock);
2632         put_files_struct(files);
2633
2634         if (err)
2635                 goto out;
2636
2637         if (file->f_op == &bpf_raw_tp_fops) {
2638                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2639                 struct bpf_raw_event_map *btp = raw_tp->btp;
2640
2641                 err = bpf_task_fd_query_copy(attr, uattr,
2642                                              raw_tp->prog->aux->id,
2643                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2644                                              btp->tp->name, 0, 0);
2645                 goto put_file;
2646         }
2647
2648         event = perf_get_event(file);
2649         if (!IS_ERR(event)) {
2650                 u64 probe_offset, probe_addr;
2651                 u32 prog_id, fd_type;
2652                 const char *buf;
2653
2654                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2655                                               &buf, &probe_offset,
2656                                               &probe_addr);
2657                 if (!err)
2658                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2659                                                      fd_type, buf,
2660                                                      probe_offset,
2661                                                      probe_addr);
2662                 goto put_file;
2663         }
2664
2665         err = -ENOTSUPP;
2666 put_file:
2667         fput(file);
2668 out:
2669         return err;
2670 }
2671
2672 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2673 {
2674         union bpf_attr attr = {};
2675         int err;
2676
2677         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2678                 return -EPERM;
2679
2680         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2681         if (err)
2682                 return err;
2683         size = min_t(u32, size, sizeof(attr));
2684
2685         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2686         if (copy_from_user(&attr, uattr, size) != 0)
2687                 return -EFAULT;
2688
2689         err = security_bpf(cmd, &attr, size);
2690         if (err < 0)
2691                 return err;
2692
2693         switch (cmd) {
2694         case BPF_MAP_CREATE:
2695                 err = map_create(&attr);
2696                 break;
2697         case BPF_MAP_LOOKUP_ELEM:
2698                 err = map_lookup_elem(&attr);
2699                 break;
2700         case BPF_MAP_UPDATE_ELEM:
2701                 err = map_update_elem(&attr);
2702                 break;
2703         case BPF_MAP_DELETE_ELEM:
2704                 err = map_delete_elem(&attr);
2705                 break;
2706         case BPF_MAP_GET_NEXT_KEY:
2707                 err = map_get_next_key(&attr);
2708                 break;
2709         case BPF_PROG_LOAD:
2710                 err = bpf_prog_load(&attr, uattr);
2711                 break;
2712         case BPF_OBJ_PIN:
2713                 err = bpf_obj_pin(&attr);
2714                 break;
2715         case BPF_OBJ_GET:
2716                 err = bpf_obj_get(&attr);
2717                 break;
2718         case BPF_PROG_ATTACH:
2719                 err = bpf_prog_attach(&attr);
2720                 break;
2721         case BPF_PROG_DETACH:
2722                 err = bpf_prog_detach(&attr);
2723                 break;
2724         case BPF_PROG_QUERY:
2725                 err = bpf_prog_query(&attr, uattr);
2726                 break;
2727         case BPF_PROG_TEST_RUN:
2728                 err = bpf_prog_test_run(&attr, uattr);
2729                 break;
2730         case BPF_PROG_GET_NEXT_ID:
2731                 err = bpf_obj_get_next_id(&attr, uattr,
2732                                           &prog_idr, &prog_idr_lock);
2733                 break;
2734         case BPF_MAP_GET_NEXT_ID:
2735                 err = bpf_obj_get_next_id(&attr, uattr,
2736                                           &map_idr, &map_idr_lock);
2737                 break;
2738         case BPF_PROG_GET_FD_BY_ID:
2739                 err = bpf_prog_get_fd_by_id(&attr);
2740                 break;
2741         case BPF_MAP_GET_FD_BY_ID:
2742                 err = bpf_map_get_fd_by_id(&attr);
2743                 break;
2744         case BPF_OBJ_GET_INFO_BY_FD:
2745                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2746                 break;
2747         case BPF_RAW_TRACEPOINT_OPEN:
2748                 err = bpf_raw_tracepoint_open(&attr);
2749                 break;
2750         case BPF_BTF_LOAD:
2751                 err = bpf_btf_load(&attr);
2752                 break;
2753         case BPF_BTF_GET_FD_BY_ID:
2754                 err = bpf_btf_get_fd_by_id(&attr);
2755                 break;
2756         case BPF_TASK_FD_QUERY:
2757                 err = bpf_task_fd_query(&attr, uattr);
2758                 break;
2759         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2760                 err = map_lookup_and_delete_elem(&attr);
2761                 break;
2762         default:
2763                 err = -EINVAL;
2764                 break;
2765         }
2766
2767         return err;
2768 }