kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26
  27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  28                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  29                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  30                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  33
  34 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  35
  36 DEFINE_PER_CPU(int, bpf_prog_active);
  37 static DEFINE_IDR(prog_idr);
  38 static DEFINE_SPINLOCK(prog_idr_lock);
  39 static DEFINE_IDR(map_idr);
  40 static DEFINE_SPINLOCK(map_idr_lock);
  41
  42 int sysctl_unprivileged_bpf_disabled __read_mostly;
  43
  44 static const struct bpf_map_ops * const bpf_map_types[] = {
  45 #define BPF_PROG_TYPE(_id, _ops)
  46 #define BPF_MAP_TYPE(_id, _ops) \
  47         [_id] = &_ops,
  48 #include <linux/bpf_types.h>
  49 #undef BPF_PROG_TYPE
  50 #undef BPF_MAP_TYPE
  51 };
  52
  53 /*
  54  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  55  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  56  * we don't know about yet.
  57  *
  58  * There is a ToCToU between this function call and the following
  59  * copy_from_user() call. However, this is not a concern since this function is
  60  * meant to be a future-proofing of bits.
  61  */
  62 int bpf_check_uarg_tail_zero(void __user *uaddr,
  63                              size_t expected_size,
  64                              size_t actual_size)
  65 {
  66         unsigned char __user *addr;
  67         unsigned char __user *end;
  68         unsigned char val;
  69         int err;
  70
  71         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  72                 return -E2BIG;
  73
  74         if (unlikely(!access_ok(uaddr, actual_size)))
  75                 return -EFAULT;
  76
  77         if (actual_size <= expected_size)
  78                 return 0;
  79
  80         addr = uaddr + expected_size;
  81         end  = uaddr + actual_size;
  82
  83         for (; addr < end; addr++) {
  84                 err = get_user(val, addr);
  85                 if (err)
  86                         return err;
  87                 if (val)
  88                         return -E2BIG;
  89         }
  90
  91         return 0;
  92 }
  93
  94 const struct bpf_map_ops bpf_map_offload_ops = {
  95         .map_alloc = bpf_map_offload_map_alloc,
  96         .map_free = bpf_map_offload_map_free,
  97         .map_check_btf = map_check_no_btf,
  98 };
  99
 100 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 101 {
 102         const struct bpf_map_ops *ops;
 103         u32 type = attr->map_type;
 104         struct bpf_map *map;
 105         int err;
 106
 107         if (type >= ARRAY_SIZE(bpf_map_types))
 108                 return ERR_PTR(-EINVAL);
 109         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 110         ops = bpf_map_types[type];
 111         if (!ops)
 112                 return ERR_PTR(-EINVAL);
 113
 114         if (ops->map_alloc_check) {
 115                 err = ops->map_alloc_check(attr);
 116                 if (err)
 117                         return ERR_PTR(err);
 118         }
 119         if (attr->map_ifindex)
 120                 ops = &bpf_map_offload_ops;
 121         map = ops->map_alloc(attr);
 122         if (IS_ERR(map))
 123                 return map;
 124         map->ops = ops;
 125         map->map_type = type;
 126         return map;
 127 }
 128
 129 void *bpf_map_area_alloc(size_t size, int numa_node)
 130 {
 131         /* We really just want to fail instead of triggering OOM killer
 132          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 133          * which is used for lower order allocation requests.
 134          *
 135          * It has been observed that higher order allocation requests done by
 136          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 137          * to reclaim memory from the page cache, thus we set
 138          * __GFP_RETRY_MAYFAIL to avoid such situations.
 139          */
 140
 141         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 142         void *area;
 143
 144         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 145                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 146                                     numa_node);
 147                 if (area != NULL)
 148                         return area;
 149         }
 150
 151         return __vmalloc_node_flags_caller(size, numa_node,
 152                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 153                                            flags, __builtin_return_address(0));
 154 }
 155
 156 void bpf_map_area_free(void *area)
 157 {
 158         kvfree(area);
 159 }
 160
 161 static u32 bpf_map_flags_retain_permanent(u32 flags)
 162 {
 163         /* Some map creation flags are not tied to the map object but
 164          * rather to the map fd instead, so they have no meaning upon
 165          * map object inspection since multiple file descriptors with
 166          * different (access) properties can exist here. Thus, given
 167          * this has zero meaning for the map itself, lets clear these
 168          * from here.
 169          */
 170         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 171 }
 172
 173 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 174 {
 175         map->map_type = attr->map_type;
 176         map->key_size = attr->key_size;
 177         map->value_size = attr->value_size;
 178         map->max_entries = attr->max_entries;
 179         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 180         map->numa_node = bpf_map_attr_numa_node(attr);
 181 }
 182
 183 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 184 {
 185         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 186
 187         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 188                 atomic_long_sub(pages, &user->locked_vm);
 189                 return -EPERM;
 190         }
 191         return 0;
 192 }
 193
 194 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 195 {
 196         if (user)
 197                 atomic_long_sub(pages, &user->locked_vm);
 198 }
 199
 200 int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size)
 201 {
 202         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 203         struct user_struct *user;
 204         int ret;
 205
 206         if (size >= U32_MAX - PAGE_SIZE)
 207                 return -E2BIG;
 208
 209         user = get_current_user();
 210         ret = bpf_charge_memlock(user, pages);
 211         if (ret) {
 212                 free_uid(user);
 213                 return ret;
 214         }
 215
 216         mem->pages = pages;
 217         mem->user = user;
 218
 219         return 0;
 220 }
 221
 222 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 223 {
 224         bpf_uncharge_memlock(mem->user, mem->pages);
 225         free_uid(mem->user);
 226 }
 227
 228 void bpf_map_charge_move(struct bpf_map_memory *dst,
 229                          struct bpf_map_memory *src)
 230 {
 231         *dst = *src;
 232
 233         /* Make sure src will not be used for the redundant uncharging. */
 234         memset(src, 0, sizeof(struct bpf_map_memory));
 235 }
 236
 237 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 238 {
 239         int ret;
 240
 241         ret = bpf_charge_memlock(map->memory.user, pages);
 242         if (ret)
 243                 return ret;
 244         map->memory.pages += pages;
 245         return ret;
 246 }
 247
 248 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 249 {
 250         bpf_uncharge_memlock(map->memory.user, pages);
 251         map->memory.pages -= pages;
 252 }
 253
 254 static int bpf_map_alloc_id(struct bpf_map *map)
 255 {
 256         int id;
 257
 258         idr_preload(GFP_KERNEL);
 259         spin_lock_bh(&map_idr_lock);
 260         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 261         if (id > 0)
 262                 map->id = id;
 263         spin_unlock_bh(&map_idr_lock);
 264         idr_preload_end();
 265
 266         if (WARN_ON_ONCE(!id))
 267                 return -ENOSPC;
 268
 269         return id > 0 ? 0 : id;
 270 }
 271
 272 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 273 {
 274         unsigned long flags;
 275
 276         /* Offloaded maps are removed from the IDR store when their device
 277          * disappears - even if someone holds an fd to them they are unusable,
 278          * the memory is gone, all ops will fail; they are simply waiting for
 279          * refcnt to drop to be freed.
 280          */
 281         if (!map->id)
 282                 return;
 283
 284         if (do_idr_lock)
 285                 spin_lock_irqsave(&map_idr_lock, flags);
 286         else
 287                 __acquire(&map_idr_lock);
 288
 289         idr_remove(&map_idr, map->id);
 290         map->id = 0;
 291
 292         if (do_idr_lock)
 293                 spin_unlock_irqrestore(&map_idr_lock, flags);
 294         else
 295                 __release(&map_idr_lock);
 296 }
 297
 298 /* called from workqueue */
 299 static void bpf_map_free_deferred(struct work_struct *work)
 300 {
 301         struct bpf_map *map = container_of(work, struct bpf_map, work);
 302         struct bpf_map_memory mem;
 303
 304         bpf_map_charge_move(&mem, &map->memory);
 305         security_bpf_map_free(map);
 306         /* implementation dependent freeing */
 307         map->ops->map_free(map);
 308         bpf_map_charge_finish(&mem);
 309 }
 310
 311 static void bpf_map_put_uref(struct bpf_map *map)
 312 {
 313         if (atomic_dec_and_test(&map->usercnt)) {
 314                 if (map->ops->map_release_uref)
 315                         map->ops->map_release_uref(map);
 316         }
 317 }
 318
 319 /* decrement map refcnt and schedule it for freeing via workqueue
 320  * (unrelying map implementation ops->map_free() might sleep)
 321  */
 322 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 323 {
 324         if (atomic_dec_and_test(&map->refcnt)) {
 325                 /* bpf_map_free_id() must be called first */
 326                 bpf_map_free_id(map, do_idr_lock);
 327                 btf_put(map->btf);
 328                 INIT_WORK(&map->work, bpf_map_free_deferred);
 329                 schedule_work(&map->work);
 330         }
 331 }
 332
 333 void bpf_map_put(struct bpf_map *map)
 334 {
 335         __bpf_map_put(map, true);
 336 }
 337 EXPORT_SYMBOL_GPL(bpf_map_put);
 338
 339 void bpf_map_put_with_uref(struct bpf_map *map)
 340 {
 341         bpf_map_put_uref(map);
 342         bpf_map_put(map);
 343 }
 344
 345 static int bpf_map_release(struct inode *inode, struct file *filp)
 346 {
 347         struct bpf_map *map = filp->private_data;
 348
 349         if (map->ops->map_release)
 350                 map->ops->map_release(map, filp);
 351
 352         bpf_map_put_with_uref(map);
 353         return 0;
 354 }
 355
 356 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 357 {
 358         fmode_t mode = f.file->f_mode;
 359
 360         /* Our file permissions may have been overridden by global
 361          * map permissions facing syscall side.
 362          */
 363         if (READ_ONCE(map->frozen))
 364                 mode &= ~FMODE_CAN_WRITE;
 365         return mode;
 366 }
 367
 368 #ifdef CONFIG_PROC_FS
 369 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 370 {
 371         const struct bpf_map *map = filp->private_data;
 372         const struct bpf_array *array;
 373         u32 owner_prog_type = 0;
 374         u32 owner_jited = 0;
 375
 376         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 377                 array = container_of(map, struct bpf_array, map);
 378                 owner_prog_type = array->owner_prog_type;
 379                 owner_jited = array->owner_jited;
 380         }
 381
 382         seq_printf(m,
 383                    "map_type:\t%u\n"
 384                    "key_size:\t%u\n"
 385                    "value_size:\t%u\n"
 386                    "max_entries:\t%u\n"
 387                    "map_flags:\t%#x\n"
 388                    "memlock:\t%llu\n"
 389                    "map_id:\t%u\n"
 390                    "frozen:\t%u\n",
 391                    map->map_type,
 392                    map->key_size,
 393                    map->value_size,
 394                    map->max_entries,
 395                    map->map_flags,
 396                    map->memory.pages * 1ULL << PAGE_SHIFT,
 397                    map->id,
 398                    READ_ONCE(map->frozen));
 399
 400         if (owner_prog_type) {
 401                 seq_printf(m, "owner_prog_type:\t%u\n",
 402                            owner_prog_type);
 403                 seq_printf(m, "owner_jited:\t%u\n",
 404                            owner_jited);
 405         }
 406 }
 407 #endif
 408
 409 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 410                               loff_t *ppos)
 411 {
 412         /* We need this handler such that alloc_file() enables
 413          * f_mode with FMODE_CAN_READ.
 414          */
 415         return -EINVAL;
 416 }
 417
 418 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 419                                size_t siz, loff_t *ppos)
 420 {
 421         /* We need this handler such that alloc_file() enables
 422          * f_mode with FMODE_CAN_WRITE.
 423          */
 424         return -EINVAL;
 425 }
 426
 427 const struct file_operations bpf_map_fops = {
 428 #ifdef CONFIG_PROC_FS
 429         .show_fdinfo    = bpf_map_show_fdinfo,
 430 #endif
 431         .release        = bpf_map_release,
 432         .read           = bpf_dummy_read,
 433         .write          = bpf_dummy_write,
 434 };
 435
 436 int bpf_map_new_fd(struct bpf_map *map, int flags)
 437 {
 438         int ret;
 439
 440         ret = security_bpf_map(map, OPEN_FMODE(flags));
 441         if (ret < 0)
 442                 return ret;
 443
 444         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 445                                 flags | O_CLOEXEC);
 446 }
 447
 448 int bpf_get_file_flag(int flags)
 449 {
 450         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 451                 return -EINVAL;
 452         if (flags & BPF_F_RDONLY)
 453                 return O_RDONLY;
 454         if (flags & BPF_F_WRONLY)
 455                 return O_WRONLY;
 456         return O_RDWR;
 457 }
 458
 459 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 460 #define CHECK_ATTR(CMD) \
 461         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 462                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 463                    sizeof(*attr) - \
 464                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 465                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 466
 467 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 468  * Return 0 on success and < 0 on error.
 469  */
 470 static int bpf_obj_name_cpy(char *dst, const char *src)
 471 {
 472         const char *end = src + BPF_OBJ_NAME_LEN;
 473
 474         memset(dst, 0, BPF_OBJ_NAME_LEN);
 475         /* Copy all isalnum(), '_' and '.' chars. */
 476         while (src < end && *src) {
 477                 if (!isalnum(*src) &&
 478                     *src != '_' && *src != '.')
 479                         return -EINVAL;
 480                 *dst++ = *src++;
 481         }
 482
 483         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 484         if (src == end)
 485                 return -EINVAL;
 486
 487         return 0;
 488 }
 489
 490 int map_check_no_btf(const struct bpf_map *map,
 491                      const struct btf *btf,
 492                      const struct btf_type *key_type,
 493                      const struct btf_type *value_type)
 494 {
 495         return -ENOTSUPP;
 496 }
 497
 498 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 499                          u32 btf_key_id, u32 btf_value_id)
 500 {
 501         const struct btf_type *key_type, *value_type;
 502         u32 key_size, value_size;
 503         int ret = 0;
 504
 505         /* Some maps allow key to be unspecified. */
 506         if (btf_key_id) {
 507                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 508                 if (!key_type || key_size != map->key_size)
 509                         return -EINVAL;
 510         } else {
 511                 key_type = btf_type_by_id(btf, 0);
 512                 if (!map->ops->map_check_btf)
 513                         return -EINVAL;
 514         }
 515
 516         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 517         if (!value_type || value_size != map->value_size)
 518                 return -EINVAL;
 519
 520         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 521
 522         if (map_value_has_spin_lock(map)) {
 523                 if (map->map_flags & BPF_F_RDONLY_PROG)
 524                         return -EACCES;
 525                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 526                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 527                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 528                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 529                         return -ENOTSUPP;
 530                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 531                     map->value_size) {
 532                         WARN_ONCE(1,
 533                                   "verifier bug spin_lock_off %d value_size %d\n",
 534                                   map->spin_lock_off, map->value_size);
 535                         return -EFAULT;
 536                 }
 537         }
 538
 539         if (map->ops->map_check_btf)
 540                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 541
 542         return ret;
 543 }
 544
 545 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 546 /* called via syscall */
 547 static int map_create(union bpf_attr *attr)
 548 {
 549         int numa_node = bpf_map_attr_numa_node(attr);
 550         struct bpf_map_memory mem;
 551         struct bpf_map *map;
 552         int f_flags;
 553         int err;
 554
 555         err = CHECK_ATTR(BPF_MAP_CREATE);
 556         if (err)
 557                 return -EINVAL;
 558
 559         f_flags = bpf_get_file_flag(attr->map_flags);
 560         if (f_flags < 0)
 561                 return f_flags;
 562
 563         if (numa_node != NUMA_NO_NODE &&
 564             ((unsigned int)numa_node >= nr_node_ids ||
 565              !node_online(numa_node)))
 566                 return -EINVAL;
 567
 568         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 569         map = find_and_alloc_map(attr);
 570         if (IS_ERR(map))
 571                 return PTR_ERR(map);
 572
 573         err = bpf_obj_name_cpy(map->name, attr->map_name);
 574         if (err)
 575                 goto free_map;
 576
 577         atomic_set(&map->refcnt, 1);
 578         atomic_set(&map->usercnt, 1);
 579
 580         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 581                 struct btf *btf;
 582
 583                 if (!attr->btf_value_type_id) {
 584                         err = -EINVAL;
 585                         goto free_map;
 586                 }
 587
 588                 btf = btf_get_by_fd(attr->btf_fd);
 589                 if (IS_ERR(btf)) {
 590                         err = PTR_ERR(btf);
 591                         goto free_map;
 592                 }
 593
 594                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 595                                     attr->btf_value_type_id);
 596                 if (err) {
 597                         btf_put(btf);
 598                         goto free_map;
 599                 }
 600
 601                 map->btf = btf;
 602                 map->btf_key_type_id = attr->btf_key_type_id;
 603                 map->btf_value_type_id = attr->btf_value_type_id;
 604         } else {
 605                 map->spin_lock_off = -EINVAL;
 606         }
 607
 608         err = security_bpf_map_alloc(map);
 609         if (err)
 610                 goto free_map;
 611
 612         err = bpf_map_alloc_id(map);
 613         if (err)
 614                 goto free_map_sec;
 615
 616         err = bpf_map_new_fd(map, f_flags);
 617         if (err < 0) {
 618                 /* failed to allocate fd.
 619                  * bpf_map_put_with_uref() is needed because the above
 620                  * bpf_map_alloc_id() has published the map
 621                  * to the userspace and the userspace may
 622                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 623                  */
 624                 bpf_map_put_with_uref(map);
 625                 return err;
 626         }
 627
 628         return err;
 629
 630 free_map_sec:
 631         security_bpf_map_free(map);
 632 free_map:
 633         btf_put(map->btf);
 634         bpf_map_charge_move(&mem, &map->memory);
 635         map->ops->map_free(map);
 636         bpf_map_charge_finish(&mem);
 637         return err;
 638 }
 639
 640 /* if error is returned, fd is released.
 641  * On success caller should complete fd access with matching fdput()
 642  */
 643 struct bpf_map *__bpf_map_get(struct fd f)
 644 {
 645         if (!f.file)
 646                 return ERR_PTR(-EBADF);
 647         if (f.file->f_op != &bpf_map_fops) {
 648                 fdput(f);
 649                 return ERR_PTR(-EINVAL);
 650         }
 651
 652         return f.file->private_data;
 653 }
 654
 655 /* prog's and map's refcnt limit */
 656 #define BPF_MAX_REFCNT 32768
 657
 658 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 659 {
 660         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 661                 atomic_dec(&map->refcnt);
 662                 return ERR_PTR(-EBUSY);
 663         }
 664         if (uref)
 665                 atomic_inc(&map->usercnt);
 666         return map;
 667 }
 668 EXPORT_SYMBOL_GPL(bpf_map_inc);
 669
 670 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 671 {
 672         struct fd f = fdget(ufd);
 673         struct bpf_map *map;
 674
 675         map = __bpf_map_get(f);
 676         if (IS_ERR(map))
 677                 return map;
 678
 679         map = bpf_map_inc(map, true);
 680         fdput(f);
 681
 682         return map;
 683 }
 684
 685 /* map_idr_lock should have been held */
 686 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 687                                             bool uref)
 688 {
 689         int refold;
 690
 691         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 692
 693         if (refold >= BPF_MAX_REFCNT) {
 694                 __bpf_map_put(map, false);
 695                 return ERR_PTR(-EBUSY);
 696         }
 697
 698         if (!refold)
 699                 return ERR_PTR(-ENOENT);
 700
 701         if (uref)
 702                 atomic_inc(&map->usercnt);
 703
 704         return map;
 705 }
 706
 707 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 708 {
 709         return -ENOTSUPP;
 710 }
 711
 712 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 713 {
 714         if (key_size)
 715                 return memdup_user(ukey, key_size);
 716
 717         if (ukey)
 718                 return ERR_PTR(-EINVAL);
 719
 720         return NULL;
 721 }
 722
 723 /* last field in 'union bpf_attr' used by this command */
 724 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 725
 726 static int map_lookup_elem(union bpf_attr *attr)
 727 {
 728         void __user *ukey = u64_to_user_ptr(attr->key);
 729         void __user *uvalue = u64_to_user_ptr(attr->value);
 730         int ufd = attr->map_fd;
 731         struct bpf_map *map;
 732         void *key, *value, *ptr;
 733         u32 value_size;
 734         struct fd f;
 735         int err;
 736
 737         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 738                 return -EINVAL;
 739
 740         if (attr->flags & ~BPF_F_LOCK)
 741                 return -EINVAL;
 742
 743         f = fdget(ufd);
 744         map = __bpf_map_get(f);
 745         if (IS_ERR(map))
 746                 return PTR_ERR(map);
 747         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 748                 err = -EPERM;
 749                 goto err_put;
 750         }
 751
 752         if ((attr->flags & BPF_F_LOCK) &&
 753             !map_value_has_spin_lock(map)) {
 754                 err = -EINVAL;
 755                 goto err_put;
 756         }
 757
 758         key = __bpf_copy_key(ukey, map->key_size);
 759         if (IS_ERR(key)) {
 760                 err = PTR_ERR(key);
 761                 goto err_put;
 762         }
 763
 764         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 765             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 766             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 767             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 768                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 769         else if (IS_FD_MAP(map))
 770                 value_size = sizeof(u32);
 771         else
 772                 value_size = map->value_size;
 773
 774         err = -ENOMEM;
 775         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 776         if (!value)
 777                 goto free_key;
 778
 779         if (bpf_map_is_dev_bound(map)) {
 780                 err = bpf_map_offload_lookup_elem(map, key, value);
 781                 goto done;
 782         }
 783
 784         preempt_disable();
 785         this_cpu_inc(bpf_prog_active);
 786         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 787             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 788                 err = bpf_percpu_hash_copy(map, key, value);
 789         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 790                 err = bpf_percpu_array_copy(map, key, value);
 791         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 792                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 793         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 794                 err = bpf_stackmap_copy(map, key, value);
 795         } else if (IS_FD_ARRAY(map)) {
 796                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 797         } else if (IS_FD_HASH(map)) {
 798                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 799         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 800                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 801         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 802                    map->map_type == BPF_MAP_TYPE_STACK) {
 803                 err = map->ops->map_peek_elem(map, value);
 804         } else {
 805                 rcu_read_lock();
 806                 if (map->ops->map_lookup_elem_sys_only)
 807                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 808                 else
 809                         ptr = map->ops->map_lookup_elem(map, key);
 810                 if (IS_ERR(ptr)) {
 811                         err = PTR_ERR(ptr);
 812                 } else if (!ptr) {
 813                         err = -ENOENT;
 814                 } else {
 815                         err = 0;
 816                         if (attr->flags & BPF_F_LOCK)
 817                                 /* lock 'ptr' and copy everything but lock */
 818                                 copy_map_value_locked(map, value, ptr, true);
 819                         else
 820                                 copy_map_value(map, value, ptr);
 821                         /* mask lock, since value wasn't zero inited */
 822                         check_and_init_map_lock(map, value);
 823                 }
 824                 rcu_read_unlock();
 825         }
 826         this_cpu_dec(bpf_prog_active);
 827         preempt_enable();
 828
 829 done:
 830         if (err)
 831                 goto free_value;
 832
 833         err = -EFAULT;
 834         if (copy_to_user(uvalue, value, value_size) != 0)
 835                 goto free_value;
 836
 837         err = 0;
 838
 839 free_value:
 840         kfree(value);
 841 free_key:
 842         kfree(key);
 843 err_put:
 844         fdput(f);
 845         return err;
 846 }
 847
 848 static void maybe_wait_bpf_programs(struct bpf_map *map)
 849 {
 850         /* Wait for any running BPF programs to complete so that
 851          * userspace, when we return to it, knows that all programs
 852          * that could be running use the new map value.
 853          */
 854         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 855             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 856                 synchronize_rcu();
 857 }
 858
 859 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 860
 861 static int map_update_elem(union bpf_attr *attr)
 862 {
 863         void __user *ukey = u64_to_user_ptr(attr->key);
 864         void __user *uvalue = u64_to_user_ptr(attr->value);
 865         int ufd = attr->map_fd;
 866         struct bpf_map *map;
 867         void *key, *value;
 868         u32 value_size;
 869         struct fd f;
 870         int err;
 871
 872         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 873                 return -EINVAL;
 874
 875         f = fdget(ufd);
 876         map = __bpf_map_get(f);
 877         if (IS_ERR(map))
 878                 return PTR_ERR(map);
 879         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 880                 err = -EPERM;
 881                 goto err_put;
 882         }
 883
 884         if ((attr->flags & BPF_F_LOCK) &&
 885             !map_value_has_spin_lock(map)) {
 886                 err = -EINVAL;
 887                 goto err_put;
 888         }
 889
 890         key = __bpf_copy_key(ukey, map->key_size);
 891         if (IS_ERR(key)) {
 892                 err = PTR_ERR(key);
 893                 goto err_put;
 894         }
 895
 896         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 897             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 898             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 899             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 900                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 901         else
 902                 value_size = map->value_size;
 903
 904         err = -ENOMEM;
 905         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 906         if (!value)
 907                 goto free_key;
 908
 909         err = -EFAULT;
 910         if (copy_from_user(value, uvalue, value_size) != 0)
 911                 goto free_value;
 912
 913         /* Need to create a kthread, thus must support schedule */
 914         if (bpf_map_is_dev_bound(map)) {
 915                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 916                 goto out;
 917         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 918                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 919                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 920                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 921                 goto out;
 922         }
 923
 924         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 925          * inside bpf map update or delete otherwise deadlocks are possible
 926          */
 927         preempt_disable();
 928         __this_cpu_inc(bpf_prog_active);
 929         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 930             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 931                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 932         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 933                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 934         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 935                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 936                                                        attr->flags);
 937         } else if (IS_FD_ARRAY(map)) {
 938                 rcu_read_lock();
 939                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 940                                                    attr->flags);
 941                 rcu_read_unlock();
 942         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 943                 rcu_read_lock();
 944                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 945                                                   attr->flags);
 946                 rcu_read_unlock();
 947         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 948                 /* rcu_read_lock() is not needed */
 949                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 950                                                          attr->flags);
 951         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 952                    map->map_type == BPF_MAP_TYPE_STACK) {
 953                 err = map->ops->map_push_elem(map, value, attr->flags);
 954         } else {
 955                 rcu_read_lock();
 956                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 957                 rcu_read_unlock();
 958         }
 959         __this_cpu_dec(bpf_prog_active);
 960         preempt_enable();
 961         maybe_wait_bpf_programs(map);
 962 out:
 963 free_value:
 964         kfree(value);
 965 free_key:
 966         kfree(key);
 967 err_put:
 968         fdput(f);
 969         return err;
 970 }
 971
 972 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 973
 974 static int map_delete_elem(union bpf_attr *attr)
 975 {
 976         void __user *ukey = u64_to_user_ptr(attr->key);
 977         int ufd = attr->map_fd;
 978         struct bpf_map *map;
 979         struct fd f;
 980         void *key;
 981         int err;
 982
 983         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 984                 return -EINVAL;
 985
 986         f = fdget(ufd);
 987         map = __bpf_map_get(f);
 988         if (IS_ERR(map))
 989                 return PTR_ERR(map);
 990         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 991                 err = -EPERM;
 992                 goto err_put;
 993         }
 994
 995         key = __bpf_copy_key(ukey, map->key_size);
 996         if (IS_ERR(key)) {
 997                 err = PTR_ERR(key);
 998                 goto err_put;
 999         }
1000
1001         if (bpf_map_is_dev_bound(map)) {
1002                 err = bpf_map_offload_delete_elem(map, key);
1003                 goto out;
1004         }
1005
1006         preempt_disable();
1007         __this_cpu_inc(bpf_prog_active);
1008         rcu_read_lock();
1009         err = map->ops->map_delete_elem(map, key);
1010         rcu_read_unlock();
1011         __this_cpu_dec(bpf_prog_active);
1012         preempt_enable();
1013         maybe_wait_bpf_programs(map);
1014 out:
1015         kfree(key);
1016 err_put:
1017         fdput(f);
1018         return err;
1019 }
1020
1021 /* last field in 'union bpf_attr' used by this command */
1022 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1023
1024 static int map_get_next_key(union bpf_attr *attr)
1025 {
1026         void __user *ukey = u64_to_user_ptr(attr->key);
1027         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1028         int ufd = attr->map_fd;
1029         struct bpf_map *map;
1030         void *key, *next_key;
1031         struct fd f;
1032         int err;
1033
1034         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1035                 return -EINVAL;
1036
1037         f = fdget(ufd);
1038         map = __bpf_map_get(f);
1039         if (IS_ERR(map))
1040                 return PTR_ERR(map);
1041         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1042                 err = -EPERM;
1043                 goto err_put;
1044         }
1045
1046         if (ukey) {
1047                 key = __bpf_copy_key(ukey, map->key_size);
1048                 if (IS_ERR(key)) {
1049                         err = PTR_ERR(key);
1050                         goto err_put;
1051                 }
1052         } else {
1053                 key = NULL;
1054         }
1055
1056         err = -ENOMEM;
1057         next_key = kmalloc(map->key_size, GFP_USER);
1058         if (!next_key)
1059                 goto free_key;
1060
1061         if (bpf_map_is_dev_bound(map)) {
1062                 err = bpf_map_offload_get_next_key(map, key, next_key);
1063                 goto out;
1064         }
1065
1066         rcu_read_lock();
1067         err = map->ops->map_get_next_key(map, key, next_key);
1068         rcu_read_unlock();
1069 out:
1070         if (err)
1071                 goto free_next_key;
1072
1073         err = -EFAULT;
1074         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1075                 goto free_next_key;
1076
1077         err = 0;
1078
1079 free_next_key:
1080         kfree(next_key);
1081 free_key:
1082         kfree(key);
1083 err_put:
1084         fdput(f);
1085         return err;
1086 }
1087
1088 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1089
1090 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1091 {
1092         void __user *ukey = u64_to_user_ptr(attr->key);
1093         void __user *uvalue = u64_to_user_ptr(attr->value);
1094         int ufd = attr->map_fd;
1095         struct bpf_map *map;
1096         void *key, *value;
1097         u32 value_size;
1098         struct fd f;
1099         int err;
1100
1101         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1102                 return -EINVAL;
1103
1104         f = fdget(ufd);
1105         map = __bpf_map_get(f);
1106         if (IS_ERR(map))
1107                 return PTR_ERR(map);
1108         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1109                 err = -EPERM;
1110                 goto err_put;
1111         }
1112
1113         key = __bpf_copy_key(ukey, map->key_size);
1114         if (IS_ERR(key)) {
1115                 err = PTR_ERR(key);
1116                 goto err_put;
1117         }
1118
1119         value_size = map->value_size;
1120
1121         err = -ENOMEM;
1122         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1123         if (!value)
1124                 goto free_key;
1125
1126         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1127             map->map_type == BPF_MAP_TYPE_STACK) {
1128                 err = map->ops->map_pop_elem(map, value);
1129         } else {
1130                 err = -ENOTSUPP;
1131         }
1132
1133         if (err)
1134                 goto free_value;
1135
1136         if (copy_to_user(uvalue, value, value_size) != 0)
1137                 goto free_value;
1138
1139         err = 0;
1140
1141 free_value:
1142         kfree(value);
1143 free_key:
1144         kfree(key);
1145 err_put:
1146         fdput(f);
1147         return err;
1148 }
1149
1150 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1151
1152 static int map_freeze(const union bpf_attr *attr)
1153 {
1154         int err = 0, ufd = attr->map_fd;
1155         struct bpf_map *map;
1156         struct fd f;
1157
1158         if (CHECK_ATTR(BPF_MAP_FREEZE))
1159                 return -EINVAL;
1160
1161         f = fdget(ufd);
1162         map = __bpf_map_get(f);
1163         if (IS_ERR(map))
1164                 return PTR_ERR(map);
1165         if (READ_ONCE(map->frozen)) {
1166                 err = -EBUSY;
1167                 goto err_put;
1168         }
1169         if (!capable(CAP_SYS_ADMIN)) {
1170                 err = -EPERM;
1171                 goto err_put;
1172         }
1173
1174         WRITE_ONCE(map->frozen, true);
1175 err_put:
1176         fdput(f);
1177         return err;
1178 }
1179
1180 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1181 #define BPF_PROG_TYPE(_id, _name) \
1182         [_id] = & _name ## _prog_ops,
1183 #define BPF_MAP_TYPE(_id, _ops)
1184 #include <linux/bpf_types.h>
1185 #undef BPF_PROG_TYPE
1186 #undef BPF_MAP_TYPE
1187 };
1188
1189 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1190 {
1191         const struct bpf_prog_ops *ops;
1192
1193         if (type >= ARRAY_SIZE(bpf_prog_types))
1194                 return -EINVAL;
1195         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1196         ops = bpf_prog_types[type];
1197         if (!ops)
1198                 return -EINVAL;
1199
1200         if (!bpf_prog_is_dev_bound(prog->aux))
1201                 prog->aux->ops = ops;
1202         else
1203                 prog->aux->ops = &bpf_offload_prog_ops;
1204         prog->type = type;
1205         return 0;
1206 }
1207
1208 /* drop refcnt on maps used by eBPF program and free auxilary data */
1209 static void free_used_maps(struct bpf_prog_aux *aux)
1210 {
1211         enum bpf_cgroup_storage_type stype;
1212         int i;
1213
1214         for_each_cgroup_storage_type(stype) {
1215                 if (!aux->cgroup_storage[stype])
1216                         continue;
1217                 bpf_cgroup_storage_release(aux->prog,
1218                                            aux->cgroup_storage[stype]);
1219         }
1220
1221         for (i = 0; i < aux->used_map_cnt; i++)
1222                 bpf_map_put(aux->used_maps[i]);
1223
1224         kfree(aux->used_maps);
1225 }
1226
1227 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1228 {
1229         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1230         unsigned long user_bufs;
1231
1232         if (user) {
1233                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1234                 if (user_bufs > memlock_limit) {
1235                         atomic_long_sub(pages, &user->locked_vm);
1236                         return -EPERM;
1237                 }
1238         }
1239
1240         return 0;
1241 }
1242
1243 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1244 {
1245         if (user)
1246                 atomic_long_sub(pages, &user->locked_vm);
1247 }
1248
1249 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1250 {
1251         struct user_struct *user = get_current_user();
1252         int ret;
1253
1254         ret = __bpf_prog_charge(user, prog->pages);
1255         if (ret) {
1256                 free_uid(user);
1257                 return ret;
1258         }
1259
1260         prog->aux->user = user;
1261         return 0;
1262 }
1263
1264 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1265 {
1266         struct user_struct *user = prog->aux->user;
1267
1268         __bpf_prog_uncharge(user, prog->pages);
1269         free_uid(user);
1270 }
1271
1272 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1273 {
1274         int id;
1275
1276         idr_preload(GFP_KERNEL);
1277         spin_lock_bh(&prog_idr_lock);
1278         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1279         if (id > 0)
1280                 prog->aux->id = id;
1281         spin_unlock_bh(&prog_idr_lock);
1282         idr_preload_end();
1283
1284         /* id is in [1, INT_MAX) */
1285         if (WARN_ON_ONCE(!id))
1286                 return -ENOSPC;
1287
1288         return id > 0 ? 0 : id;
1289 }
1290
1291 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1292 {
1293         /* cBPF to eBPF migrations are currently not in the idr store.
1294          * Offloaded programs are removed from the store when their device
1295          * disappears - even if someone grabs an fd to them they are unusable,
1296          * simply waiting for refcnt to drop to be freed.
1297          */
1298         if (!prog->aux->id)
1299                 return;
1300
1301         if (do_idr_lock)
1302                 spin_lock_bh(&prog_idr_lock);
1303         else
1304                 __acquire(&prog_idr_lock);
1305
1306         idr_remove(&prog_idr, prog->aux->id);
1307         prog->aux->id = 0;
1308
1309         if (do_idr_lock)
1310                 spin_unlock_bh(&prog_idr_lock);
1311         else
1312                 __release(&prog_idr_lock);
1313 }
1314
1315 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1316 {
1317         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1318
1319         free_used_maps(aux);
1320         bpf_prog_uncharge_memlock(aux->prog);
1321         security_bpf_prog_free(aux);
1322         bpf_prog_free(aux->prog);
1323 }
1324
1325 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1326 {
1327         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1328                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1329                 /* bpf_prog_free_id() must be called first */
1330                 bpf_prog_free_id(prog, do_idr_lock);
1331                 bpf_prog_kallsyms_del_all(prog);
1332                 btf_put(prog->aux->btf);
1333                 kvfree(prog->aux->func_info);
1334                 bpf_prog_free_linfo(prog);
1335
1336                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1337         }
1338 }
1339
1340 void bpf_prog_put(struct bpf_prog *prog)
1341 {
1342         __bpf_prog_put(prog, true);
1343 }
1344 EXPORT_SYMBOL_GPL(bpf_prog_put);
1345
1346 static int bpf_prog_release(struct inode *inode, struct file *filp)
1347 {
1348         struct bpf_prog *prog = filp->private_data;
1349
1350         bpf_prog_put(prog);
1351         return 0;
1352 }
1353
1354 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1355                                struct bpf_prog_stats *stats)
1356 {
1357         u64 nsecs = 0, cnt = 0;
1358         int cpu;
1359
1360         for_each_possible_cpu(cpu) {
1361                 const struct bpf_prog_stats *st;
1362                 unsigned int start;
1363                 u64 tnsecs, tcnt;
1364
1365                 st = per_cpu_ptr(prog->aux->stats, cpu);
1366                 do {
1367                         start = u64_stats_fetch_begin_irq(&st->syncp);
1368                         tnsecs = st->nsecs;
1369                         tcnt = st->cnt;
1370                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1371                 nsecs += tnsecs;
1372                 cnt += tcnt;
1373         }
1374         stats->nsecs = nsecs;
1375         stats->cnt = cnt;
1376 }
1377
1378 #ifdef CONFIG_PROC_FS
1379 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1380 {
1381         const struct bpf_prog *prog = filp->private_data;
1382         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1383         struct bpf_prog_stats stats;
1384
1385         bpf_prog_get_stats(prog, &stats);
1386         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1387         seq_printf(m,
1388                    "prog_type:\t%u\n"
1389                    "prog_jited:\t%u\n"
1390                    "prog_tag:\t%s\n"
1391                    "memlock:\t%llu\n"
1392                    "prog_id:\t%u\n"
1393                    "run_time_ns:\t%llu\n"
1394                    "run_cnt:\t%llu\n",
1395                    prog->type,
1396                    prog->jited,
1397                    prog_tag,
1398                    prog->pages * 1ULL << PAGE_SHIFT,
1399                    prog->aux->id,
1400                    stats.nsecs,
1401                    stats.cnt);
1402 }
1403 #endif
1404
1405 const struct file_operations bpf_prog_fops = {
1406 #ifdef CONFIG_PROC_FS
1407         .show_fdinfo    = bpf_prog_show_fdinfo,
1408 #endif
1409         .release        = bpf_prog_release,
1410         .read           = bpf_dummy_read,
1411         .write          = bpf_dummy_write,
1412 };
1413
1414 int bpf_prog_new_fd(struct bpf_prog *prog)
1415 {
1416         int ret;
1417
1418         ret = security_bpf_prog(prog);
1419         if (ret < 0)
1420                 return ret;
1421
1422         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1423                                 O_RDWR | O_CLOEXEC);
1424 }
1425
1426 static struct bpf_prog *____bpf_prog_get(struct fd f)
1427 {
1428         if (!f.file)
1429                 return ERR_PTR(-EBADF);
1430         if (f.file->f_op != &bpf_prog_fops) {
1431                 fdput(f);
1432                 return ERR_PTR(-EINVAL);
1433         }
1434
1435         return f.file->private_data;
1436 }
1437
1438 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1439 {
1440         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1441                 atomic_sub(i, &prog->aux->refcnt);
1442                 return ERR_PTR(-EBUSY);
1443         }
1444         return prog;
1445 }
1446 EXPORT_SYMBOL_GPL(bpf_prog_add);
1447
1448 void bpf_prog_sub(struct bpf_prog *prog, int i)
1449 {
1450         /* Only to be used for undoing previous bpf_prog_add() in some
1451          * error path. We still know that another entity in our call
1452          * path holds a reference to the program, thus atomic_sub() can
1453          * be safely used in such cases!
1454          */
1455         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1456 }
1457 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1458
1459 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1460 {
1461         return bpf_prog_add(prog, 1);
1462 }
1463 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1464
1465 /* prog_idr_lock should have been held */
1466 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1467 {
1468         int refold;
1469
1470         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1471
1472         if (refold >= BPF_MAX_REFCNT) {
1473                 __bpf_prog_put(prog, false);
1474                 return ERR_PTR(-EBUSY);
1475         }
1476
1477         if (!refold)
1478                 return ERR_PTR(-ENOENT);
1479
1480         return prog;
1481 }
1482 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1483
1484 bool bpf_prog_get_ok(struct bpf_prog *prog,
1485                             enum bpf_prog_type *attach_type, bool attach_drv)
1486 {
1487         /* not an attachment, just a refcount inc, always allow */
1488         if (!attach_type)
1489                 return true;
1490
1491         if (prog->type != *attach_type)
1492                 return false;
1493         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1494                 return false;
1495
1496         return true;
1497 }
1498
1499 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1500                                        bool attach_drv)
1501 {
1502         struct fd f = fdget(ufd);
1503         struct bpf_prog *prog;
1504
1505         prog = ____bpf_prog_get(f);
1506         if (IS_ERR(prog))
1507                 return prog;
1508         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1509                 prog = ERR_PTR(-EINVAL);
1510                 goto out;
1511         }
1512
1513         prog = bpf_prog_inc(prog);
1514 out:
1515         fdput(f);
1516         return prog;
1517 }
1518
1519 struct bpf_prog *bpf_prog_get(u32 ufd)
1520 {
1521         return __bpf_prog_get(ufd, NULL, false);
1522 }
1523
1524 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1525                                        bool attach_drv)
1526 {
1527         return __bpf_prog_get(ufd, &type, attach_drv);
1528 }
1529 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1530
1531 /* Initially all BPF programs could be loaded w/o specifying
1532  * expected_attach_type. Later for some of them specifying expected_attach_type
1533  * at load time became required so that program could be validated properly.
1534  * Programs of types that are allowed to be loaded both w/ and w/o (for
1535  * backward compatibility) expected_attach_type, should have the default attach
1536  * type assigned to expected_attach_type for the latter case, so that it can be
1537  * validated later at attach time.
1538  *
1539  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1540  * prog type requires it but has some attach types that have to be backward
1541  * compatible.
1542  */
1543 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1544 {
1545         switch (attr->prog_type) {
1546         case BPF_PROG_TYPE_CGROUP_SOCK:
1547                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1548                  * exist so checking for non-zero is the way to go here.
1549                  */
1550                 if (!attr->expected_attach_type)
1551                         attr->expected_attach_type =
1552                                 BPF_CGROUP_INET_SOCK_CREATE;
1553                 break;
1554         }
1555 }
1556
1557 static int
1558 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1559                                 enum bpf_attach_type expected_attach_type)
1560 {
1561         switch (prog_type) {
1562         case BPF_PROG_TYPE_CGROUP_SOCK:
1563                 switch (expected_attach_type) {
1564                 case BPF_CGROUP_INET_SOCK_CREATE:
1565                 case BPF_CGROUP_INET4_POST_BIND:
1566                 case BPF_CGROUP_INET6_POST_BIND:
1567                         return 0;
1568                 default:
1569                         return -EINVAL;
1570                 }
1571         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1572                 switch (expected_attach_type) {
1573                 case BPF_CGROUP_INET4_BIND:
1574                 case BPF_CGROUP_INET6_BIND:
1575                 case BPF_CGROUP_INET4_CONNECT:
1576                 case BPF_CGROUP_INET6_CONNECT:
1577                 case BPF_CGROUP_UDP4_SENDMSG:
1578                 case BPF_CGROUP_UDP6_SENDMSG:
1579                 case BPF_CGROUP_UDP4_RECVMSG:
1580                 case BPF_CGROUP_UDP6_RECVMSG:
1581                         return 0;
1582                 default:
1583                         return -EINVAL;
1584                 }
1585         case BPF_PROG_TYPE_CGROUP_SKB:
1586                 switch (expected_attach_type) {
1587                 case BPF_CGROUP_INET_INGRESS:
1588                 case BPF_CGROUP_INET_EGRESS:
1589                         return 0;
1590                 default:
1591                         return -EINVAL;
1592                 }
1593         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1594                 switch (expected_attach_type) {
1595                 case BPF_CGROUP_SETSOCKOPT:
1596                 case BPF_CGROUP_GETSOCKOPT:
1597                         return 0;
1598                 default:
1599                         return -EINVAL;
1600                 }
1601         default:
1602                 return 0;
1603         }
1604 }
1605
1606 /* last field in 'union bpf_attr' used by this command */
1607 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1608
1609 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1610 {
1611         enum bpf_prog_type type = attr->prog_type;
1612         struct bpf_prog *prog;
1613         int err;
1614         char license[128];
1615         bool is_gpl;
1616
1617         if (CHECK_ATTR(BPF_PROG_LOAD))
1618                 return -EINVAL;
1619
1620         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1621                                  BPF_F_ANY_ALIGNMENT |
1622                                  BPF_F_TEST_RND_HI32))
1623                 return -EINVAL;
1624
1625         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1626             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1627             !capable(CAP_SYS_ADMIN))
1628                 return -EPERM;
1629
1630         /* copy eBPF program license from user space */
1631         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1632                               sizeof(license) - 1) < 0)
1633                 return -EFAULT;
1634         license[sizeof(license) - 1] = 0;
1635
1636         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1637         is_gpl = license_is_gpl_compatible(license);
1638
1639         if (attr->insn_cnt == 0 ||
1640             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1641                 return -E2BIG;
1642         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1643             type != BPF_PROG_TYPE_CGROUP_SKB &&
1644             !capable(CAP_SYS_ADMIN))
1645                 return -EPERM;
1646
1647         bpf_prog_load_fixup_attach_type(attr);
1648         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1649                 return -EINVAL;
1650
1651         /* plain bpf_prog allocation */
1652         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1653         if (!prog)
1654                 return -ENOMEM;
1655
1656         prog->expected_attach_type = attr->expected_attach_type;
1657
1658         prog->aux->offload_requested = !!attr->prog_ifindex;
1659
1660         err = security_bpf_prog_alloc(prog->aux);
1661         if (err)
1662                 goto free_prog_nouncharge;
1663
1664         err = bpf_prog_charge_memlock(prog);
1665         if (err)
1666                 goto free_prog_sec;
1667
1668         prog->len = attr->insn_cnt;
1669
1670         err = -EFAULT;
1671         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1672                            bpf_prog_insn_size(prog)) != 0)
1673                 goto free_prog;
1674
1675         prog->orig_prog = NULL;
1676         prog->jited = 0;
1677
1678         atomic_set(&prog->aux->refcnt, 1);
1679         prog->gpl_compatible = is_gpl ? 1 : 0;
1680
1681         if (bpf_prog_is_dev_bound(prog->aux)) {
1682                 err = bpf_prog_offload_init(prog, attr);
1683                 if (err)
1684                         goto free_prog;
1685         }
1686
1687         /* find program type: socket_filter vs tracing_filter */
1688         err = find_prog_type(type, prog);
1689         if (err < 0)
1690                 goto free_prog;
1691
1692         prog->aux->load_time = ktime_get_boottime_ns();
1693         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1694         if (err)
1695                 goto free_prog;
1696
1697         /* run eBPF verifier */
1698         err = bpf_check(&prog, attr, uattr);
1699         if (err < 0)
1700                 goto free_used_maps;
1701
1702         prog = bpf_prog_select_runtime(prog, &err);
1703         if (err < 0)
1704                 goto free_used_maps;
1705
1706         err = bpf_prog_alloc_id(prog);
1707         if (err)
1708                 goto free_used_maps;
1709
1710         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1711          * effectively publicly exposed. However, retrieving via
1712          * bpf_prog_get_fd_by_id() will take another reference,
1713          * therefore it cannot be gone underneath us.
1714          *
1715          * Only for the time /after/ successful bpf_prog_new_fd()
1716          * and before returning to userspace, we might just hold
1717          * one reference and any parallel close on that fd could
1718          * rip everything out. Hence, below notifications must
1719          * happen before bpf_prog_new_fd().
1720          *
1721          * Also, any failure handling from this point onwards must
1722          * be using bpf_prog_put() given the program is exposed.
1723          */
1724         bpf_prog_kallsyms_add(prog);
1725         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1726
1727         err = bpf_prog_new_fd(prog);
1728         if (err < 0)
1729                 bpf_prog_put(prog);
1730         return err;
1731
1732 free_used_maps:
1733         bpf_prog_free_linfo(prog);
1734         kvfree(prog->aux->func_info);
1735         btf_put(prog->aux->btf);
1736         bpf_prog_kallsyms_del_subprogs(prog);
1737         free_used_maps(prog->aux);
1738 free_prog:
1739         bpf_prog_uncharge_memlock(prog);
1740 free_prog_sec:
1741         security_bpf_prog_free(prog->aux);
1742 free_prog_nouncharge:
1743         bpf_prog_free(prog);
1744         return err;
1745 }
1746
1747 #define BPF_OBJ_LAST_FIELD file_flags
1748
1749 static int bpf_obj_pin(const union bpf_attr *attr)
1750 {
1751         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1752                 return -EINVAL;
1753
1754         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1755 }
1756
1757 static int bpf_obj_get(const union bpf_attr *attr)
1758 {
1759         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1760             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1761                 return -EINVAL;
1762
1763         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1764                                 attr->file_flags);
1765 }
1766
1767 struct bpf_raw_tracepoint {
1768         struct bpf_raw_event_map *btp;
1769         struct bpf_prog *prog;
1770 };
1771
1772 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1773 {
1774         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1775
1776         if (raw_tp->prog) {
1777                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1778                 bpf_prog_put(raw_tp->prog);
1779         }
1780         bpf_put_raw_tracepoint(raw_tp->btp);
1781         kfree(raw_tp);
1782         return 0;
1783 }
1784
1785 static const struct file_operations bpf_raw_tp_fops = {
1786         .release        = bpf_raw_tracepoint_release,
1787         .read           = bpf_dummy_read,
1788         .write          = bpf_dummy_write,
1789 };
1790
1791 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1792
1793 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1794 {
1795         struct bpf_raw_tracepoint *raw_tp;
1796         struct bpf_raw_event_map *btp;
1797         struct bpf_prog *prog;
1798         char tp_name[128];
1799         int tp_fd, err;
1800
1801         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1802                               sizeof(tp_name) - 1) < 0)
1803                 return -EFAULT;
1804         tp_name[sizeof(tp_name) - 1] = 0;
1805
1806         btp = bpf_get_raw_tracepoint(tp_name);
1807         if (!btp)
1808                 return -ENOENT;
1809
1810         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1811         if (!raw_tp) {
1812                 err = -ENOMEM;
1813                 goto out_put_btp;
1814         }
1815         raw_tp->btp = btp;
1816
1817         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1818         if (IS_ERR(prog)) {
1819                 err = PTR_ERR(prog);
1820                 goto out_free_tp;
1821         }
1822         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1823             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1824                 err = -EINVAL;
1825                 goto out_put_prog;
1826         }
1827
1828         err = bpf_probe_register(raw_tp->btp, prog);
1829         if (err)
1830                 goto out_put_prog;
1831
1832         raw_tp->prog = prog;
1833         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1834                                  O_CLOEXEC);
1835         if (tp_fd < 0) {
1836                 bpf_probe_unregister(raw_tp->btp, prog);
1837                 err = tp_fd;
1838                 goto out_put_prog;
1839         }
1840         return tp_fd;
1841
1842 out_put_prog:
1843         bpf_prog_put(prog);
1844 out_free_tp:
1845         kfree(raw_tp);
1846 out_put_btp:
1847         bpf_put_raw_tracepoint(btp);
1848         return err;
1849 }
1850
1851 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1852                                              enum bpf_attach_type attach_type)
1853 {
1854         switch (prog->type) {
1855         case BPF_PROG_TYPE_CGROUP_SOCK:
1856         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1857         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1858                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1859         case BPF_PROG_TYPE_CGROUP_SKB:
1860                 return prog->enforce_expected_attach_type &&
1861                         prog->expected_attach_type != attach_type ?
1862                         -EINVAL : 0;
1863         default:
1864                 return 0;
1865         }
1866 }
1867
1868 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1869
1870 #define BPF_F_ATTACH_MASK \
1871         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1872
1873 static int bpf_prog_attach(const union bpf_attr *attr)
1874 {
1875         enum bpf_prog_type ptype;
1876         struct bpf_prog *prog;
1877         int ret;
1878
1879         if (!capable(CAP_NET_ADMIN))
1880                 return -EPERM;
1881
1882         if (CHECK_ATTR(BPF_PROG_ATTACH))
1883                 return -EINVAL;
1884
1885         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1886                 return -EINVAL;
1887
1888         switch (attr->attach_type) {
1889         case BPF_CGROUP_INET_INGRESS:
1890         case BPF_CGROUP_INET_EGRESS:
1891                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1892                 break;
1893         case BPF_CGROUP_INET_SOCK_CREATE:
1894         case BPF_CGROUP_INET4_POST_BIND:
1895         case BPF_CGROUP_INET6_POST_BIND:
1896                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1897                 break;
1898         case BPF_CGROUP_INET4_BIND:
1899         case BPF_CGROUP_INET6_BIND:
1900         case BPF_CGROUP_INET4_CONNECT:
1901         case BPF_CGROUP_INET6_CONNECT:
1902         case BPF_CGROUP_UDP4_SENDMSG:
1903         case BPF_CGROUP_UDP6_SENDMSG:
1904         case BPF_CGROUP_UDP4_RECVMSG:
1905         case BPF_CGROUP_UDP6_RECVMSG:
1906                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1907                 break;
1908         case BPF_CGROUP_SOCK_OPS:
1909                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1910                 break;
1911         case BPF_CGROUP_DEVICE:
1912                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1913                 break;
1914         case BPF_SK_MSG_VERDICT:
1915                 ptype = BPF_PROG_TYPE_SK_MSG;
1916                 break;
1917         case BPF_SK_SKB_STREAM_PARSER:
1918         case BPF_SK_SKB_STREAM_VERDICT:
1919                 ptype = BPF_PROG_TYPE_SK_SKB;
1920                 break;
1921         case BPF_LIRC_MODE2:
1922                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1923                 break;
1924         case BPF_FLOW_DISSECTOR:
1925                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1926                 break;
1927         case BPF_CGROUP_SYSCTL:
1928                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1929                 break;
1930         case BPF_CGROUP_GETSOCKOPT:
1931         case BPF_CGROUP_SETSOCKOPT:
1932                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1933                 break;
1934         default:
1935                 return -EINVAL;
1936         }
1937
1938         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1939         if (IS_ERR(prog))
1940                 return PTR_ERR(prog);
1941
1942         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1943                 bpf_prog_put(prog);
1944                 return -EINVAL;
1945         }
1946
1947         switch (ptype) {
1948         case BPF_PROG_TYPE_SK_SKB:
1949         case BPF_PROG_TYPE_SK_MSG:
1950                 ret = sock_map_get_from_fd(attr, prog);
1951                 break;
1952         case BPF_PROG_TYPE_LIRC_MODE2:
1953                 ret = lirc_prog_attach(attr, prog);
1954                 break;
1955         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1956                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1957                 break;
1958         default:
1959                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1960         }
1961
1962         if (ret)
1963                 bpf_prog_put(prog);
1964         return ret;
1965 }
1966
1967 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1968
1969 static int bpf_prog_detach(const union bpf_attr *attr)
1970 {
1971         enum bpf_prog_type ptype;
1972
1973         if (!capable(CAP_NET_ADMIN))
1974                 return -EPERM;
1975
1976         if (CHECK_ATTR(BPF_PROG_DETACH))
1977                 return -EINVAL;
1978
1979         switch (attr->attach_type) {
1980         case BPF_CGROUP_INET_INGRESS:
1981         case BPF_CGROUP_INET_EGRESS:
1982                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1983                 break;
1984         case BPF_CGROUP_INET_SOCK_CREATE:
1985         case BPF_CGROUP_INET4_POST_BIND:
1986         case BPF_CGROUP_INET6_POST_BIND:
1987                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1988                 break;
1989         case BPF_CGROUP_INET4_BIND:
1990         case BPF_CGROUP_INET6_BIND:
1991         case BPF_CGROUP_INET4_CONNECT:
1992         case BPF_CGROUP_INET6_CONNECT:
1993         case BPF_CGROUP_UDP4_SENDMSG:
1994         case BPF_CGROUP_UDP6_SENDMSG:
1995         case BPF_CGROUP_UDP4_RECVMSG:
1996         case BPF_CGROUP_UDP6_RECVMSG:
1997                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1998                 break;
1999         case BPF_CGROUP_SOCK_OPS:
2000                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2001                 break;
2002         case BPF_CGROUP_DEVICE:
2003                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2004                 break;
2005         case BPF_SK_MSG_VERDICT:
2006                 return sock_map_get_from_fd(attr, NULL);
2007         case BPF_SK_SKB_STREAM_PARSER:
2008         case BPF_SK_SKB_STREAM_VERDICT:
2009                 return sock_map_get_from_fd(attr, NULL);
2010         case BPF_LIRC_MODE2:
2011                 return lirc_prog_detach(attr);
2012         case BPF_FLOW_DISSECTOR:
2013                 return skb_flow_dissector_bpf_prog_detach(attr);
2014         case BPF_CGROUP_SYSCTL:
2015                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2016                 break;
2017         case BPF_CGROUP_GETSOCKOPT:
2018         case BPF_CGROUP_SETSOCKOPT:
2019                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2020                 break;
2021         default:
2022                 return -EINVAL;
2023         }
2024
2025         return cgroup_bpf_prog_detach(attr, ptype);
2026 }
2027
2028 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2029
2030 static int bpf_prog_query(const union bpf_attr *attr,
2031                           union bpf_attr __user *uattr)
2032 {
2033         if (!capable(CAP_NET_ADMIN))
2034                 return -EPERM;
2035         if (CHECK_ATTR(BPF_PROG_QUERY))
2036                 return -EINVAL;
2037         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2038                 return -EINVAL;
2039
2040         switch (attr->query.attach_type) {
2041         case BPF_CGROUP_INET_INGRESS:
2042         case BPF_CGROUP_INET_EGRESS:
2043         case BPF_CGROUP_INET_SOCK_CREATE:
2044         case BPF_CGROUP_INET4_BIND:
2045         case BPF_CGROUP_INET6_BIND:
2046         case BPF_CGROUP_INET4_POST_BIND:
2047         case BPF_CGROUP_INET6_POST_BIND:
2048         case BPF_CGROUP_INET4_CONNECT:
2049         case BPF_CGROUP_INET6_CONNECT:
2050         case BPF_CGROUP_UDP4_SENDMSG:
2051         case BPF_CGROUP_UDP6_SENDMSG:
2052         case BPF_CGROUP_UDP4_RECVMSG:
2053         case BPF_CGROUP_UDP6_RECVMSG:
2054         case BPF_CGROUP_SOCK_OPS:
2055         case BPF_CGROUP_DEVICE:
2056         case BPF_CGROUP_SYSCTL:
2057         case BPF_CGROUP_GETSOCKOPT:
2058         case BPF_CGROUP_SETSOCKOPT:
2059                 break;
2060         case BPF_LIRC_MODE2:
2061                 return lirc_prog_query(attr, uattr);
2062         case BPF_FLOW_DISSECTOR:
2063                 return skb_flow_dissector_prog_query(attr, uattr);
2064         default:
2065                 return -EINVAL;
2066         }
2067
2068         return cgroup_bpf_prog_query(attr, uattr);
2069 }
2070
2071 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2072
2073 static int bpf_prog_test_run(const union bpf_attr *attr,
2074                              union bpf_attr __user *uattr)
2075 {
2076         struct bpf_prog *prog;
2077         int ret = -ENOTSUPP;
2078
2079         if (!capable(CAP_SYS_ADMIN))
2080                 return -EPERM;
2081         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2082                 return -EINVAL;
2083
2084         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2085             (!attr->test.ctx_size_in && attr->test.ctx_in))
2086                 return -EINVAL;
2087
2088         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2089             (!attr->test.ctx_size_out && attr->test.ctx_out))
2090                 return -EINVAL;
2091
2092         prog = bpf_prog_get(attr->test.prog_fd);
2093         if (IS_ERR(prog))
2094                 return PTR_ERR(prog);
2095
2096         if (prog->aux->ops->test_run)
2097                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2098
2099         bpf_prog_put(prog);
2100         return ret;
2101 }
2102
2103 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2104
2105 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2106                                union bpf_attr __user *uattr,
2107                                struct idr *idr,
2108                                spinlock_t *lock)
2109 {
2110         u32 next_id = attr->start_id;
2111         int err = 0;
2112
2113         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2114                 return -EINVAL;
2115
2116         if (!capable(CAP_SYS_ADMIN))
2117                 return -EPERM;
2118
2119         next_id++;
2120         spin_lock_bh(lock);
2121         if (!idr_get_next(idr, &next_id))
2122                 err = -ENOENT;
2123         spin_unlock_bh(lock);
2124
2125         if (!err)
2126                 err = put_user(next_id, &uattr->next_id);
2127
2128         return err;
2129 }
2130
2131 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2132
2133 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2134 {
2135         struct bpf_prog *prog;
2136         u32 id = attr->prog_id;
2137         int fd;
2138
2139         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2140                 return -EINVAL;
2141
2142         if (!capable(CAP_SYS_ADMIN))
2143                 return -EPERM;
2144
2145         spin_lock_bh(&prog_idr_lock);
2146         prog = idr_find(&prog_idr, id);
2147         if (prog)
2148                 prog = bpf_prog_inc_not_zero(prog);
2149         else
2150                 prog = ERR_PTR(-ENOENT);
2151         spin_unlock_bh(&prog_idr_lock);
2152
2153         if (IS_ERR(prog))
2154                 return PTR_ERR(prog);
2155
2156         fd = bpf_prog_new_fd(prog);
2157         if (fd < 0)
2158                 bpf_prog_put(prog);
2159
2160         return fd;
2161 }
2162
2163 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2164
2165 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2166 {
2167         struct bpf_map *map;
2168         u32 id = attr->map_id;
2169         int f_flags;
2170         int fd;
2171
2172         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2173             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2174                 return -EINVAL;
2175
2176         if (!capable(CAP_SYS_ADMIN))
2177                 return -EPERM;
2178
2179         f_flags = bpf_get_file_flag(attr->open_flags);
2180         if (f_flags < 0)
2181                 return f_flags;
2182
2183         spin_lock_bh(&map_idr_lock);
2184         map = idr_find(&map_idr, id);
2185         if (map)
2186                 map = bpf_map_inc_not_zero(map, true);
2187         else
2188                 map = ERR_PTR(-ENOENT);
2189         spin_unlock_bh(&map_idr_lock);
2190
2191         if (IS_ERR(map))
2192                 return PTR_ERR(map);
2193
2194         fd = bpf_map_new_fd(map, f_flags);
2195         if (fd < 0)
2196                 bpf_map_put_with_uref(map);
2197
2198         return fd;
2199 }
2200
2201 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2202                                               unsigned long addr, u32 *off,
2203                                               u32 *type)
2204 {
2205         const struct bpf_map *map;
2206         int i;
2207
2208         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2209                 map = prog->aux->used_maps[i];
2210                 if (map == (void *)addr) {
2211                         *type = BPF_PSEUDO_MAP_FD;
2212                         return map;
2213                 }
2214                 if (!map->ops->map_direct_value_meta)
2215                         continue;
2216                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2217                         *type = BPF_PSEUDO_MAP_VALUE;
2218                         return map;
2219                 }
2220         }
2221
2222         return NULL;
2223 }
2224
2225 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2226 {
2227         const struct bpf_map *map;
2228         struct bpf_insn *insns;
2229         u32 off, type;
2230         u64 imm;
2231         int i;
2232
2233         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2234                         GFP_USER);
2235         if (!insns)
2236                 return insns;
2237
2238         for (i = 0; i < prog->len; i++) {
2239                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2240                         insns[i].code = BPF_JMP | BPF_CALL;
2241                         insns[i].imm = BPF_FUNC_tail_call;
2242                         /* fall-through */
2243                 }
2244                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2245                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2246                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2247                                 insns[i].code = BPF_JMP | BPF_CALL;
2248                         if (!bpf_dump_raw_ok())
2249                                 insns[i].imm = 0;
2250                         continue;
2251                 }
2252
2253                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2254                         continue;
2255
2256                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2257                 map = bpf_map_from_imm(prog, imm, &off, &type);
2258                 if (map) {
2259                         insns[i].src_reg = type;
2260                         insns[i].imm = map->id;
2261                         insns[i + 1].imm = off;
2262                         continue;
2263                 }
2264         }
2265
2266         return insns;
2267 }
2268
2269 static int set_info_rec_size(struct bpf_prog_info *info)
2270 {
2271         /*
2272          * Ensure info.*_rec_size is the same as kernel expected size
2273          *
2274          * or
2275          *
2276          * Only allow zero *_rec_size if both _rec_size and _cnt are
2277          * zero.  In this case, the kernel will set the expected
2278          * _rec_size back to the info.
2279          */
2280
2281         if ((info->nr_func_info || info->func_info_rec_size) &&
2282             info->func_info_rec_size != sizeof(struct bpf_func_info))
2283                 return -EINVAL;
2284
2285         if ((info->nr_line_info || info->line_info_rec_size) &&
2286             info->line_info_rec_size != sizeof(struct bpf_line_info))
2287                 return -EINVAL;
2288
2289         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2290             info->jited_line_info_rec_size != sizeof(__u64))
2291                 return -EINVAL;
2292
2293         info->func_info_rec_size = sizeof(struct bpf_func_info);
2294         info->line_info_rec_size = sizeof(struct bpf_line_info);
2295         info->jited_line_info_rec_size = sizeof(__u64);
2296
2297         return 0;
2298 }
2299
2300 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2301                                    const union bpf_attr *attr,
2302                                    union bpf_attr __user *uattr)
2303 {
2304         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2305         struct bpf_prog_info info = {};
2306         u32 info_len = attr->info.info_len;
2307         struct bpf_prog_stats stats;
2308         char __user *uinsns;
2309         u32 ulen;
2310         int err;
2311
2312         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2313         if (err)
2314                 return err;
2315         info_len = min_t(u32, sizeof(info), info_len);
2316
2317         if (copy_from_user(&info, uinfo, info_len))
2318                 return -EFAULT;
2319
2320         info.type = prog->type;
2321         info.id = prog->aux->id;
2322         info.load_time = prog->aux->load_time;
2323         info.created_by_uid = from_kuid_munged(current_user_ns(),
2324                                                prog->aux->user->uid);
2325         info.gpl_compatible = prog->gpl_compatible;
2326
2327         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2328         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2329
2330         ulen = info.nr_map_ids;
2331         info.nr_map_ids = prog->aux->used_map_cnt;
2332         ulen = min_t(u32, info.nr_map_ids, ulen);
2333         if (ulen) {
2334                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2335                 u32 i;
2336
2337                 for (i = 0; i < ulen; i++)
2338                         if (put_user(prog->aux->used_maps[i]->id,
2339                                      &user_map_ids[i]))
2340                                 return -EFAULT;
2341         }
2342
2343         err = set_info_rec_size(&info);
2344         if (err)
2345                 return err;
2346
2347         bpf_prog_get_stats(prog, &stats);
2348         info.run_time_ns = stats.nsecs;
2349         info.run_cnt = stats.cnt;
2350
2351         if (!capable(CAP_SYS_ADMIN)) {
2352                 info.jited_prog_len = 0;
2353                 info.xlated_prog_len = 0;
2354                 info.nr_jited_ksyms = 0;
2355                 info.nr_jited_func_lens = 0;
2356                 info.nr_func_info = 0;
2357                 info.nr_line_info = 0;
2358                 info.nr_jited_line_info = 0;
2359                 goto done;
2360         }
2361
2362         ulen = info.xlated_prog_len;
2363         info.xlated_prog_len = bpf_prog_insn_size(prog);
2364         if (info.xlated_prog_len && ulen) {
2365                 struct bpf_insn *insns_sanitized;
2366                 bool fault;
2367
2368                 if (prog->blinded && !bpf_dump_raw_ok()) {
2369                         info.xlated_prog_insns = 0;
2370                         goto done;
2371                 }
2372                 insns_sanitized = bpf_insn_prepare_dump(prog);
2373                 if (!insns_sanitized)
2374                         return -ENOMEM;
2375                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2376                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2377                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2378                 kfree(insns_sanitized);
2379                 if (fault)
2380                         return -EFAULT;
2381         }
2382
2383         if (bpf_prog_is_dev_bound(prog->aux)) {
2384                 err = bpf_prog_offload_info_fill(&info, prog);
2385                 if (err)
2386                         return err;
2387                 goto done;
2388         }
2389
2390         /* NOTE: the following code is supposed to be skipped for offload.
2391          * bpf_prog_offload_info_fill() is the place to fill similar fields
2392          * for offload.
2393          */
2394         ulen = info.jited_prog_len;
2395         if (prog->aux->func_cnt) {
2396                 u32 i;
2397
2398                 info.jited_prog_len = 0;
2399                 for (i = 0; i < prog->aux->func_cnt; i++)
2400                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2401         } else {
2402                 info.jited_prog_len = prog->jited_len;
2403         }
2404
2405         if (info.jited_prog_len && ulen) {
2406                 if (bpf_dump_raw_ok()) {
2407                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2408                         ulen = min_t(u32, info.jited_prog_len, ulen);
2409
2410                         /* for multi-function programs, copy the JITed
2411                          * instructions for all the functions
2412                          */
2413                         if (prog->aux->func_cnt) {
2414                                 u32 len, free, i;
2415                                 u8 *img;
2416
2417                                 free = ulen;
2418                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2419                                         len = prog->aux->func[i]->jited_len;
2420                                         len = min_t(u32, len, free);
2421                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2422                                         if (copy_to_user(uinsns, img, len))
2423                                                 return -EFAULT;
2424                                         uinsns += len;
2425                                         free -= len;
2426                                         if (!free)
2427                                                 break;
2428                                 }
2429                         } else {
2430                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2431                                         return -EFAULT;
2432                         }
2433                 } else {
2434                         info.jited_prog_insns = 0;
2435                 }
2436         }
2437
2438         ulen = info.nr_jited_ksyms;
2439         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2440         if (ulen) {
2441                 if (bpf_dump_raw_ok()) {
2442                         unsigned long ksym_addr;
2443                         u64 __user *user_ksyms;
2444                         u32 i;
2445
2446                         /* copy the address of the kernel symbol
2447                          * corresponding to each function
2448                          */
2449                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2450                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2451                         if (prog->aux->func_cnt) {
2452                                 for (i = 0; i < ulen; i++) {
2453                                         ksym_addr = (unsigned long)
2454                                                 prog->aux->func[i]->bpf_func;
2455                                         if (put_user((u64) ksym_addr,
2456                                                      &user_ksyms[i]))
2457                                                 return -EFAULT;
2458                                 }
2459                         } else {
2460                                 ksym_addr = (unsigned long) prog->bpf_func;
2461                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2462                                         return -EFAULT;
2463                         }
2464                 } else {
2465                         info.jited_ksyms = 0;
2466                 }
2467         }
2468
2469         ulen = info.nr_jited_func_lens;
2470         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2471         if (ulen) {
2472                 if (bpf_dump_raw_ok()) {
2473                         u32 __user *user_lens;
2474                         u32 func_len, i;
2475
2476                         /* copy the JITed image lengths for each function */
2477                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2478                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2479                         if (prog->aux->func_cnt) {
2480                                 for (i = 0; i < ulen; i++) {
2481                                         func_len =
2482                                                 prog->aux->func[i]->jited_len;
2483                                         if (put_user(func_len, &user_lens[i]))
2484                                                 return -EFAULT;
2485                                 }
2486                         } else {
2487                                 func_len = prog->jited_len;
2488                                 if (put_user(func_len, &user_lens[0]))
2489                                         return -EFAULT;
2490                         }
2491                 } else {
2492                         info.jited_func_lens = 0;
2493                 }
2494         }
2495
2496         if (prog->aux->btf)
2497                 info.btf_id = btf_id(prog->aux->btf);
2498
2499         ulen = info.nr_func_info;
2500         info.nr_func_info = prog->aux->func_info_cnt;
2501         if (info.nr_func_info && ulen) {
2502                 char __user *user_finfo;
2503
2504                 user_finfo = u64_to_user_ptr(info.func_info);
2505                 ulen = min_t(u32, info.nr_func_info, ulen);
2506                 if (copy_to_user(user_finfo, prog->aux->func_info,
2507                                  info.func_info_rec_size * ulen))
2508                         return -EFAULT;
2509         }
2510
2511         ulen = info.nr_line_info;
2512         info.nr_line_info = prog->aux->nr_linfo;
2513         if (info.nr_line_info && ulen) {
2514                 __u8 __user *user_linfo;
2515
2516                 user_linfo = u64_to_user_ptr(info.line_info);
2517                 ulen = min_t(u32, info.nr_line_info, ulen);
2518                 if (copy_to_user(user_linfo, prog->aux->linfo,
2519                                  info.line_info_rec_size * ulen))
2520                         return -EFAULT;
2521         }
2522
2523         ulen = info.nr_jited_line_info;
2524         if (prog->aux->jited_linfo)
2525                 info.nr_jited_line_info = prog->aux->nr_linfo;
2526         else
2527                 info.nr_jited_line_info = 0;
2528         if (info.nr_jited_line_info && ulen) {
2529                 if (bpf_dump_raw_ok()) {
2530                         __u64 __user *user_linfo;
2531                         u32 i;
2532
2533                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2534                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2535                         for (i = 0; i < ulen; i++) {
2536                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2537                                              &user_linfo[i]))
2538                                         return -EFAULT;
2539                         }
2540                 } else {
2541                         info.jited_line_info = 0;
2542                 }
2543         }
2544
2545         ulen = info.nr_prog_tags;
2546         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2547         if (ulen) {
2548                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2549                 u32 i;
2550
2551                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2552                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2553                 if (prog->aux->func_cnt) {
2554                         for (i = 0; i < ulen; i++) {
2555                                 if (copy_to_user(user_prog_tags[i],
2556                                                  prog->aux->func[i]->tag,
2557                                                  BPF_TAG_SIZE))
2558                                         return -EFAULT;
2559                         }
2560                 } else {
2561                         if (copy_to_user(user_prog_tags[0],
2562                                          prog->tag, BPF_TAG_SIZE))
2563                                 return -EFAULT;
2564                 }
2565         }
2566
2567 done:
2568         if (copy_to_user(uinfo, &info, info_len) ||
2569             put_user(info_len, &uattr->info.info_len))
2570                 return -EFAULT;
2571
2572         return 0;
2573 }
2574
2575 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2576                                   const union bpf_attr *attr,
2577                                   union bpf_attr __user *uattr)
2578 {
2579         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2580         struct bpf_map_info info = {};
2581         u32 info_len = attr->info.info_len;
2582         int err;
2583
2584         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2585         if (err)
2586                 return err;
2587         info_len = min_t(u32, sizeof(info), info_len);
2588
2589         info.type = map->map_type;
2590         info.id = map->id;
2591         info.key_size = map->key_size;
2592         info.value_size = map->value_size;
2593         info.max_entries = map->max_entries;
2594         info.map_flags = map->map_flags;
2595         memcpy(info.name, map->name, sizeof(map->name));
2596
2597         if (map->btf) {
2598                 info.btf_id = btf_id(map->btf);
2599                 info.btf_key_type_id = map->btf_key_type_id;
2600                 info.btf_value_type_id = map->btf_value_type_id;
2601         }
2602
2603         if (bpf_map_is_dev_bound(map)) {
2604                 err = bpf_map_offload_info_fill(&info, map);
2605                 if (err)
2606                         return err;
2607         }
2608
2609         if (copy_to_user(uinfo, &info, info_len) ||
2610             put_user(info_len, &uattr->info.info_len))
2611                 return -EFAULT;
2612
2613         return 0;
2614 }
2615
2616 static int bpf_btf_get_info_by_fd(struct btf *btf,
2617                                   const union bpf_attr *attr,
2618                                   union bpf_attr __user *uattr)
2619 {
2620         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2621         u32 info_len = attr->info.info_len;
2622         int err;
2623
2624         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2625         if (err)
2626                 return err;
2627
2628         return btf_get_info_by_fd(btf, attr, uattr);
2629 }
2630
2631 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2632
2633 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2634                                   union bpf_attr __user *uattr)
2635 {
2636         int ufd = attr->info.bpf_fd;
2637         struct fd f;
2638         int err;
2639
2640         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2641                 return -EINVAL;
2642
2643         f = fdget(ufd);
2644         if (!f.file)
2645                 return -EBADFD;
2646
2647         if (f.file->f_op == &bpf_prog_fops)
2648                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2649                                               uattr);
2650         else if (f.file->f_op == &bpf_map_fops)
2651                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2652                                              uattr);
2653         else if (f.file->f_op == &btf_fops)
2654                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2655         else
2656                 err = -EINVAL;
2657
2658         fdput(f);
2659         return err;
2660 }
2661
2662 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2663
2664 static int bpf_btf_load(const union bpf_attr *attr)
2665 {
2666         if (CHECK_ATTR(BPF_BTF_LOAD))
2667                 return -EINVAL;
2668
2669         if (!capable(CAP_SYS_ADMIN))
2670                 return -EPERM;
2671
2672         return btf_new_fd(attr);
2673 }
2674
2675 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2676
2677 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2678 {
2679         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2680                 return -EINVAL;
2681
2682         if (!capable(CAP_SYS_ADMIN))
2683                 return -EPERM;
2684
2685         return btf_get_fd_by_id(attr->btf_id);
2686 }
2687
2688 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2689                                     union bpf_attr __user *uattr,
2690                                     u32 prog_id, u32 fd_type,
2691                                     const char *buf, u64 probe_offset,
2692                                     u64 probe_addr)
2693 {
2694         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2695         u32 len = buf ? strlen(buf) : 0, input_len;
2696         int err = 0;
2697
2698         if (put_user(len, &uattr->task_fd_query.buf_len))
2699                 return -EFAULT;
2700         input_len = attr->task_fd_query.buf_len;
2701         if (input_len && ubuf) {
2702                 if (!len) {
2703                         /* nothing to copy, just make ubuf NULL terminated */
2704                         char zero = '\0';
2705
2706                         if (put_user(zero, ubuf))
2707                                 return -EFAULT;
2708                 } else if (input_len >= len + 1) {
2709                         /* ubuf can hold the string with NULL terminator */
2710                         if (copy_to_user(ubuf, buf, len + 1))
2711                                 return -EFAULT;
2712                 } else {
2713                         /* ubuf cannot hold the string with NULL terminator,
2714                          * do a partial copy with NULL terminator.
2715                          */
2716                         char zero = '\0';
2717
2718                         err = -ENOSPC;
2719                         if (copy_to_user(ubuf, buf, input_len - 1))
2720                                 return -EFAULT;
2721                         if (put_user(zero, ubuf + input_len - 1))
2722                                 return -EFAULT;
2723                 }
2724         }
2725
2726         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2727             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2728             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2729             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2730                 return -EFAULT;
2731
2732         return err;
2733 }
2734
2735 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2736
2737 static int bpf_task_fd_query(const union bpf_attr *attr,
2738                              union bpf_attr __user *uattr)
2739 {
2740         pid_t pid = attr->task_fd_query.pid;
2741         u32 fd = attr->task_fd_query.fd;
2742         const struct perf_event *event;
2743         struct files_struct *files;
2744         struct task_struct *task;
2745         struct file *file;
2746         int err;
2747
2748         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2749                 return -EINVAL;
2750
2751         if (!capable(CAP_SYS_ADMIN))
2752                 return -EPERM;
2753
2754         if (attr->task_fd_query.flags != 0)
2755                 return -EINVAL;
2756
2757         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2758         if (!task)
2759                 return -ENOENT;
2760
2761         files = get_files_struct(task);
2762         put_task_struct(task);
2763         if (!files)
2764                 return -ENOENT;
2765
2766         err = 0;
2767         spin_lock(&files->file_lock);
2768         file = fcheck_files(files, fd);
2769         if (!file)
2770                 err = -EBADF;
2771         else
2772                 get_file(file);
2773         spin_unlock(&files->file_lock);
2774         put_files_struct(files);
2775
2776         if (err)
2777                 goto out;
2778
2779         if (file->f_op == &bpf_raw_tp_fops) {
2780                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2781                 struct bpf_raw_event_map *btp = raw_tp->btp;
2782
2783                 err = bpf_task_fd_query_copy(attr, uattr,
2784                                              raw_tp->prog->aux->id,
2785                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2786                                              btp->tp->name, 0, 0);
2787                 goto put_file;
2788         }
2789
2790         event = perf_get_event(file);
2791         if (!IS_ERR(event)) {
2792                 u64 probe_offset, probe_addr;
2793                 u32 prog_id, fd_type;
2794                 const char *buf;
2795
2796                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2797                                               &buf, &probe_offset,
2798                                               &probe_addr);
2799                 if (!err)
2800                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2801                                                      fd_type, buf,
2802                                                      probe_offset,
2803                                                      probe_addr);
2804                 goto put_file;
2805         }
2806
2807         err = -ENOTSUPP;
2808 put_file:
2809         fput(file);
2810 out:
2811         return err;
2812 }
2813
2814 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2815 {
2816         union bpf_attr attr = {};
2817         int err;
2818
2819         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2820                 return -EPERM;
2821
2822         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2823         if (err)
2824                 return err;
2825         size = min_t(u32, size, sizeof(attr));
2826
2827         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2828         if (copy_from_user(&attr, uattr, size) != 0)
2829                 return -EFAULT;
2830
2831         err = security_bpf(cmd, &attr, size);
2832         if (err < 0)
2833                 return err;
2834
2835         switch (cmd) {
2836         case BPF_MAP_CREATE:
2837                 err = map_create(&attr);
2838                 break;
2839         case BPF_MAP_LOOKUP_ELEM:
2840                 err = map_lookup_elem(&attr);
2841                 break;
2842         case BPF_MAP_UPDATE_ELEM:
2843                 err = map_update_elem(&attr);
2844                 break;
2845         case BPF_MAP_DELETE_ELEM:
2846                 err = map_delete_elem(&attr);
2847                 break;
2848         case BPF_MAP_GET_NEXT_KEY:
2849                 err = map_get_next_key(&attr);
2850                 break;
2851         case BPF_MAP_FREEZE:
2852                 err = map_freeze(&attr);
2853                 break;
2854         case BPF_PROG_LOAD:
2855                 err = bpf_prog_load(&attr, uattr);
2856                 break;
2857         case BPF_OBJ_PIN:
2858                 err = bpf_obj_pin(&attr);
2859                 break;
2860         case BPF_OBJ_GET:
2861                 err = bpf_obj_get(&attr);
2862                 break;
2863         case BPF_PROG_ATTACH:
2864                 err = bpf_prog_attach(&attr);
2865                 break;
2866         case BPF_PROG_DETACH:
2867                 err = bpf_prog_detach(&attr);
2868                 break;
2869         case BPF_PROG_QUERY:
2870                 err = bpf_prog_query(&attr, uattr);
2871                 break;
2872         case BPF_PROG_TEST_RUN:
2873                 err = bpf_prog_test_run(&attr, uattr);
2874                 break;
2875         case BPF_PROG_GET_NEXT_ID:
2876                 err = bpf_obj_get_next_id(&attr, uattr,
2877                                           &prog_idr, &prog_idr_lock);
2878                 break;
2879         case BPF_MAP_GET_NEXT_ID:
2880                 err = bpf_obj_get_next_id(&attr, uattr,
2881                                           &map_idr, &map_idr_lock);
2882                 break;
2883         case BPF_PROG_GET_FD_BY_ID:
2884                 err = bpf_prog_get_fd_by_id(&attr);
2885                 break;
2886         case BPF_MAP_GET_FD_BY_ID:
2887                 err = bpf_map_get_fd_by_id(&attr);
2888                 break;
2889         case BPF_OBJ_GET_INFO_BY_FD:
2890                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2891                 break;
2892         case BPF_RAW_TRACEPOINT_OPEN:
2893                 err = bpf_raw_tracepoint_open(&attr);
2894                 break;
2895         case BPF_BTF_LOAD:
2896                 err = bpf_btf_load(&attr);
2897                 break;
2898         case BPF_BTF_GET_FD_BY_ID:
2899                 err = bpf_btf_get_fd_by_id(&attr);
2900                 break;
2901         case BPF_TASK_FD_QUERY:
2902                 err = bpf_task_fd_query(&attr, uattr);
2903                 break;
2904         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2905                 err = map_lookup_and_delete_elem(&attr);
2906                 break;
2907         default:
2908                 err = -EINVAL;
2909                 break;
2910         }
2911
2912         return err;
2913 }