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 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 708 {
 709         spin_lock_bh(&map_idr_lock);
 710         map = __bpf_map_inc_not_zero(map, uref);
 711         spin_unlock_bh(&map_idr_lock);
 712
 713         return map;
 714 }
 715 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 716
 717 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 718 {
 719         return -ENOTSUPP;
 720 }
 721
 722 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 723 {
 724         if (key_size)
 725                 return memdup_user(ukey, key_size);
 726
 727         if (ukey)
 728                 return ERR_PTR(-EINVAL);
 729
 730         return NULL;
 731 }
 732
 733 /* last field in 'union bpf_attr' used by this command */
 734 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 735
 736 static int map_lookup_elem(union bpf_attr *attr)
 737 {
 738         void __user *ukey = u64_to_user_ptr(attr->key);
 739         void __user *uvalue = u64_to_user_ptr(attr->value);
 740         int ufd = attr->map_fd;
 741         struct bpf_map *map;
 742         void *key, *value, *ptr;
 743         u32 value_size;
 744         struct fd f;
 745         int err;
 746
 747         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 748                 return -EINVAL;
 749
 750         if (attr->flags & ~BPF_F_LOCK)
 751                 return -EINVAL;
 752
 753         f = fdget(ufd);
 754         map = __bpf_map_get(f);
 755         if (IS_ERR(map))
 756                 return PTR_ERR(map);
 757         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 758                 err = -EPERM;
 759                 goto err_put;
 760         }
 761
 762         if ((attr->flags & BPF_F_LOCK) &&
 763             !map_value_has_spin_lock(map)) {
 764                 err = -EINVAL;
 765                 goto err_put;
 766         }
 767
 768         key = __bpf_copy_key(ukey, map->key_size);
 769         if (IS_ERR(key)) {
 770                 err = PTR_ERR(key);
 771                 goto err_put;
 772         }
 773
 774         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 775             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 776             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 777             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 778                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 779         else if (IS_FD_MAP(map))
 780                 value_size = sizeof(u32);
 781         else
 782                 value_size = map->value_size;
 783
 784         err = -ENOMEM;
 785         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 786         if (!value)
 787                 goto free_key;
 788
 789         if (bpf_map_is_dev_bound(map)) {
 790                 err = bpf_map_offload_lookup_elem(map, key, value);
 791                 goto done;
 792         }
 793
 794         preempt_disable();
 795         this_cpu_inc(bpf_prog_active);
 796         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 797             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 798                 err = bpf_percpu_hash_copy(map, key, value);
 799         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 800                 err = bpf_percpu_array_copy(map, key, value);
 801         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 802                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 803         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 804                 err = bpf_stackmap_copy(map, key, value);
 805         } else if (IS_FD_ARRAY(map)) {
 806                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 807         } else if (IS_FD_HASH(map)) {
 808                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 809         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 810                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 811         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 812                    map->map_type == BPF_MAP_TYPE_STACK) {
 813                 err = map->ops->map_peek_elem(map, value);
 814         } else {
 815                 rcu_read_lock();
 816                 if (map->ops->map_lookup_elem_sys_only)
 817                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 818                 else
 819                         ptr = map->ops->map_lookup_elem(map, key);
 820                 if (IS_ERR(ptr)) {
 821                         err = PTR_ERR(ptr);
 822                 } else if (!ptr) {
 823                         err = -ENOENT;
 824                 } else {
 825                         err = 0;
 826                         if (attr->flags & BPF_F_LOCK)
 827                                 /* lock 'ptr' and copy everything but lock */
 828                                 copy_map_value_locked(map, value, ptr, true);
 829                         else
 830                                 copy_map_value(map, value, ptr);
 831                         /* mask lock, since value wasn't zero inited */
 832                         check_and_init_map_lock(map, value);
 833                 }
 834                 rcu_read_unlock();
 835         }
 836         this_cpu_dec(bpf_prog_active);
 837         preempt_enable();
 838
 839 done:
 840         if (err)
 841                 goto free_value;
 842
 843         err = -EFAULT;
 844         if (copy_to_user(uvalue, value, value_size) != 0)
 845                 goto free_value;
 846
 847         err = 0;
 848
 849 free_value:
 850         kfree(value);
 851 free_key:
 852         kfree(key);
 853 err_put:
 854         fdput(f);
 855         return err;
 856 }
 857
 858 static void maybe_wait_bpf_programs(struct bpf_map *map)
 859 {
 860         /* Wait for any running BPF programs to complete so that
 861          * userspace, when we return to it, knows that all programs
 862          * that could be running use the new map value.
 863          */
 864         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 865             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 866                 synchronize_rcu();
 867 }
 868
 869 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 870
 871 static int map_update_elem(union bpf_attr *attr)
 872 {
 873         void __user *ukey = u64_to_user_ptr(attr->key);
 874         void __user *uvalue = u64_to_user_ptr(attr->value);
 875         int ufd = attr->map_fd;
 876         struct bpf_map *map;
 877         void *key, *value;
 878         u32 value_size;
 879         struct fd f;
 880         int err;
 881
 882         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 883                 return -EINVAL;
 884
 885         f = fdget(ufd);
 886         map = __bpf_map_get(f);
 887         if (IS_ERR(map))
 888                 return PTR_ERR(map);
 889         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 890                 err = -EPERM;
 891                 goto err_put;
 892         }
 893
 894         if ((attr->flags & BPF_F_LOCK) &&
 895             !map_value_has_spin_lock(map)) {
 896                 err = -EINVAL;
 897                 goto err_put;
 898         }
 899
 900         key = __bpf_copy_key(ukey, map->key_size);
 901         if (IS_ERR(key)) {
 902                 err = PTR_ERR(key);
 903                 goto err_put;
 904         }
 905
 906         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 907             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 908             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 909             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 910                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 911         else
 912                 value_size = map->value_size;
 913
 914         err = -ENOMEM;
 915         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 916         if (!value)
 917                 goto free_key;
 918
 919         err = -EFAULT;
 920         if (copy_from_user(value, uvalue, value_size) != 0)
 921                 goto free_value;
 922
 923         /* Need to create a kthread, thus must support schedule */
 924         if (bpf_map_is_dev_bound(map)) {
 925                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 926                 goto out;
 927         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 928                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 929                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 930                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 931                 goto out;
 932         }
 933
 934         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 935          * inside bpf map update or delete otherwise deadlocks are possible
 936          */
 937         preempt_disable();
 938         __this_cpu_inc(bpf_prog_active);
 939         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 940             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 941                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 942         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 943                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 944         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 945                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 946                                                        attr->flags);
 947         } else if (IS_FD_ARRAY(map)) {
 948                 rcu_read_lock();
 949                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 950                                                    attr->flags);
 951                 rcu_read_unlock();
 952         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 953                 rcu_read_lock();
 954                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 955                                                   attr->flags);
 956                 rcu_read_unlock();
 957         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 958                 /* rcu_read_lock() is not needed */
 959                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 960                                                          attr->flags);
 961         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 962                    map->map_type == BPF_MAP_TYPE_STACK) {
 963                 err = map->ops->map_push_elem(map, value, attr->flags);
 964         } else {
 965                 rcu_read_lock();
 966                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 967                 rcu_read_unlock();
 968         }
 969         __this_cpu_dec(bpf_prog_active);
 970         preempt_enable();
 971         maybe_wait_bpf_programs(map);
 972 out:
 973 free_value:
 974         kfree(value);
 975 free_key:
 976         kfree(key);
 977 err_put:
 978         fdput(f);
 979         return err;
 980 }
 981
 982 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 983
 984 static int map_delete_elem(union bpf_attr *attr)
 985 {
 986         void __user *ukey = u64_to_user_ptr(attr->key);
 987         int ufd = attr->map_fd;
 988         struct bpf_map *map;
 989         struct fd f;
 990         void *key;
 991         int err;
 992
 993         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 994                 return -EINVAL;
 995
 996         f = fdget(ufd);
 997         map = __bpf_map_get(f);
 998         if (IS_ERR(map))
 999                 return PTR_ERR(map);
1000         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1001                 err = -EPERM;
1002                 goto err_put;
1003         }
1004
1005         key = __bpf_copy_key(ukey, map->key_size);
1006         if (IS_ERR(key)) {
1007                 err = PTR_ERR(key);
1008                 goto err_put;
1009         }
1010
1011         if (bpf_map_is_dev_bound(map)) {
1012                 err = bpf_map_offload_delete_elem(map, key);
1013                 goto out;
1014         }
1015
1016         preempt_disable();
1017         __this_cpu_inc(bpf_prog_active);
1018         rcu_read_lock();
1019         err = map->ops->map_delete_elem(map, key);
1020         rcu_read_unlock();
1021         __this_cpu_dec(bpf_prog_active);
1022         preempt_enable();
1023         maybe_wait_bpf_programs(map);
1024 out:
1025         kfree(key);
1026 err_put:
1027         fdput(f);
1028         return err;
1029 }
1030
1031 /* last field in 'union bpf_attr' used by this command */
1032 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1033
1034 static int map_get_next_key(union bpf_attr *attr)
1035 {
1036         void __user *ukey = u64_to_user_ptr(attr->key);
1037         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1038         int ufd = attr->map_fd;
1039         struct bpf_map *map;
1040         void *key, *next_key;
1041         struct fd f;
1042         int err;
1043
1044         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1045                 return -EINVAL;
1046
1047         f = fdget(ufd);
1048         map = __bpf_map_get(f);
1049         if (IS_ERR(map))
1050                 return PTR_ERR(map);
1051         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1052                 err = -EPERM;
1053                 goto err_put;
1054         }
1055
1056         if (ukey) {
1057                 key = __bpf_copy_key(ukey, map->key_size);
1058                 if (IS_ERR(key)) {
1059                         err = PTR_ERR(key);
1060                         goto err_put;
1061                 }
1062         } else {
1063                 key = NULL;
1064         }
1065
1066         err = -ENOMEM;
1067         next_key = kmalloc(map->key_size, GFP_USER);
1068         if (!next_key)
1069                 goto free_key;
1070
1071         if (bpf_map_is_dev_bound(map)) {
1072                 err = bpf_map_offload_get_next_key(map, key, next_key);
1073                 goto out;
1074         }
1075
1076         rcu_read_lock();
1077         err = map->ops->map_get_next_key(map, key, next_key);
1078         rcu_read_unlock();
1079 out:
1080         if (err)
1081                 goto free_next_key;
1082
1083         err = -EFAULT;
1084         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1085                 goto free_next_key;
1086
1087         err = 0;
1088
1089 free_next_key:
1090         kfree(next_key);
1091 free_key:
1092         kfree(key);
1093 err_put:
1094         fdput(f);
1095         return err;
1096 }
1097
1098 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1099
1100 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1101 {
1102         void __user *ukey = u64_to_user_ptr(attr->key);
1103         void __user *uvalue = u64_to_user_ptr(attr->value);
1104         int ufd = attr->map_fd;
1105         struct bpf_map *map;
1106         void *key, *value;
1107         u32 value_size;
1108         struct fd f;
1109         int err;
1110
1111         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1112                 return -EINVAL;
1113
1114         f = fdget(ufd);
1115         map = __bpf_map_get(f);
1116         if (IS_ERR(map))
1117                 return PTR_ERR(map);
1118         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1119                 err = -EPERM;
1120                 goto err_put;
1121         }
1122
1123         key = __bpf_copy_key(ukey, map->key_size);
1124         if (IS_ERR(key)) {
1125                 err = PTR_ERR(key);
1126                 goto err_put;
1127         }
1128
1129         value_size = map->value_size;
1130
1131         err = -ENOMEM;
1132         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1133         if (!value)
1134                 goto free_key;
1135
1136         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1137             map->map_type == BPF_MAP_TYPE_STACK) {
1138                 err = map->ops->map_pop_elem(map, value);
1139         } else {
1140                 err = -ENOTSUPP;
1141         }
1142
1143         if (err)
1144                 goto free_value;
1145
1146         if (copy_to_user(uvalue, value, value_size) != 0)
1147                 goto free_value;
1148
1149         err = 0;
1150
1151 free_value:
1152         kfree(value);
1153 free_key:
1154         kfree(key);
1155 err_put:
1156         fdput(f);
1157         return err;
1158 }
1159
1160 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1161
1162 static int map_freeze(const union bpf_attr *attr)
1163 {
1164         int err = 0, ufd = attr->map_fd;
1165         struct bpf_map *map;
1166         struct fd f;
1167
1168         if (CHECK_ATTR(BPF_MAP_FREEZE))
1169                 return -EINVAL;
1170
1171         f = fdget(ufd);
1172         map = __bpf_map_get(f);
1173         if (IS_ERR(map))
1174                 return PTR_ERR(map);
1175         if (READ_ONCE(map->frozen)) {
1176                 err = -EBUSY;
1177                 goto err_put;
1178         }
1179         if (!capable(CAP_SYS_ADMIN)) {
1180                 err = -EPERM;
1181                 goto err_put;
1182         }
1183
1184         WRITE_ONCE(map->frozen, true);
1185 err_put:
1186         fdput(f);
1187         return err;
1188 }
1189
1190 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1191 #define BPF_PROG_TYPE(_id, _name) \
1192         [_id] = & _name ## _prog_ops,
1193 #define BPF_MAP_TYPE(_id, _ops)
1194 #include <linux/bpf_types.h>
1195 #undef BPF_PROG_TYPE
1196 #undef BPF_MAP_TYPE
1197 };
1198
1199 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1200 {
1201         const struct bpf_prog_ops *ops;
1202
1203         if (type >= ARRAY_SIZE(bpf_prog_types))
1204                 return -EINVAL;
1205         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1206         ops = bpf_prog_types[type];
1207         if (!ops)
1208                 return -EINVAL;
1209
1210         if (!bpf_prog_is_dev_bound(prog->aux))
1211                 prog->aux->ops = ops;
1212         else
1213                 prog->aux->ops = &bpf_offload_prog_ops;
1214         prog->type = type;
1215         return 0;
1216 }
1217
1218 /* drop refcnt on maps used by eBPF program and free auxilary data */
1219 static void free_used_maps(struct bpf_prog_aux *aux)
1220 {
1221         enum bpf_cgroup_storage_type stype;
1222         int i;
1223
1224         for_each_cgroup_storage_type(stype) {
1225                 if (!aux->cgroup_storage[stype])
1226                         continue;
1227                 bpf_cgroup_storage_release(aux->prog,
1228                                            aux->cgroup_storage[stype]);
1229         }
1230
1231         for (i = 0; i < aux->used_map_cnt; i++)
1232                 bpf_map_put(aux->used_maps[i]);
1233
1234         kfree(aux->used_maps);
1235 }
1236
1237 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1238 {
1239         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1240         unsigned long user_bufs;
1241
1242         if (user) {
1243                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1244                 if (user_bufs > memlock_limit) {
1245                         atomic_long_sub(pages, &user->locked_vm);
1246                         return -EPERM;
1247                 }
1248         }
1249
1250         return 0;
1251 }
1252
1253 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1254 {
1255         if (user)
1256                 atomic_long_sub(pages, &user->locked_vm);
1257 }
1258
1259 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1260 {
1261         struct user_struct *user = get_current_user();
1262         int ret;
1263
1264         ret = __bpf_prog_charge(user, prog->pages);
1265         if (ret) {
1266                 free_uid(user);
1267                 return ret;
1268         }
1269
1270         prog->aux->user = user;
1271         return 0;
1272 }
1273
1274 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1275 {
1276         struct user_struct *user = prog->aux->user;
1277
1278         __bpf_prog_uncharge(user, prog->pages);
1279         free_uid(user);
1280 }
1281
1282 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1283 {
1284         int id;
1285
1286         idr_preload(GFP_KERNEL);
1287         spin_lock_bh(&prog_idr_lock);
1288         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1289         if (id > 0)
1290                 prog->aux->id = id;
1291         spin_unlock_bh(&prog_idr_lock);
1292         idr_preload_end();
1293
1294         /* id is in [1, INT_MAX) */
1295         if (WARN_ON_ONCE(!id))
1296                 return -ENOSPC;
1297
1298         return id > 0 ? 0 : id;
1299 }
1300
1301 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1302 {
1303         /* cBPF to eBPF migrations are currently not in the idr store.
1304          * Offloaded programs are removed from the store when their device
1305          * disappears - even if someone grabs an fd to them they are unusable,
1306          * simply waiting for refcnt to drop to be freed.
1307          */
1308         if (!prog->aux->id)
1309                 return;
1310
1311         if (do_idr_lock)
1312                 spin_lock_bh(&prog_idr_lock);
1313         else
1314                 __acquire(&prog_idr_lock);
1315
1316         idr_remove(&prog_idr, prog->aux->id);
1317         prog->aux->id = 0;
1318
1319         if (do_idr_lock)
1320                 spin_unlock_bh(&prog_idr_lock);
1321         else
1322                 __release(&prog_idr_lock);
1323 }
1324
1325 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1326 {
1327         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1328
1329         free_used_maps(aux);
1330         bpf_prog_uncharge_memlock(aux->prog);
1331         security_bpf_prog_free(aux);
1332         bpf_prog_free(aux->prog);
1333 }
1334
1335 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1336 {
1337         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1338                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1339                 /* bpf_prog_free_id() must be called first */
1340                 bpf_prog_free_id(prog, do_idr_lock);
1341                 bpf_prog_kallsyms_del_all(prog);
1342                 btf_put(prog->aux->btf);
1343                 kvfree(prog->aux->func_info);
1344                 bpf_prog_free_linfo(prog);
1345
1346                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1347         }
1348 }
1349
1350 void bpf_prog_put(struct bpf_prog *prog)
1351 {
1352         __bpf_prog_put(prog, true);
1353 }
1354 EXPORT_SYMBOL_GPL(bpf_prog_put);
1355
1356 static int bpf_prog_release(struct inode *inode, struct file *filp)
1357 {
1358         struct bpf_prog *prog = filp->private_data;
1359
1360         bpf_prog_put(prog);
1361         return 0;
1362 }
1363
1364 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1365                                struct bpf_prog_stats *stats)
1366 {
1367         u64 nsecs = 0, cnt = 0;
1368         int cpu;
1369
1370         for_each_possible_cpu(cpu) {
1371                 const struct bpf_prog_stats *st;
1372                 unsigned int start;
1373                 u64 tnsecs, tcnt;
1374
1375                 st = per_cpu_ptr(prog->aux->stats, cpu);
1376                 do {
1377                         start = u64_stats_fetch_begin_irq(&st->syncp);
1378                         tnsecs = st->nsecs;
1379                         tcnt = st->cnt;
1380                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1381                 nsecs += tnsecs;
1382                 cnt += tcnt;
1383         }
1384         stats->nsecs = nsecs;
1385         stats->cnt = cnt;
1386 }
1387
1388 #ifdef CONFIG_PROC_FS
1389 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1390 {
1391         const struct bpf_prog *prog = filp->private_data;
1392         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1393         struct bpf_prog_stats stats;
1394
1395         bpf_prog_get_stats(prog, &stats);
1396         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1397         seq_printf(m,
1398                    "prog_type:\t%u\n"
1399                    "prog_jited:\t%u\n"
1400                    "prog_tag:\t%s\n"
1401                    "memlock:\t%llu\n"
1402                    "prog_id:\t%u\n"
1403                    "run_time_ns:\t%llu\n"
1404                    "run_cnt:\t%llu\n",
1405                    prog->type,
1406                    prog->jited,
1407                    prog_tag,
1408                    prog->pages * 1ULL << PAGE_SHIFT,
1409                    prog->aux->id,
1410                    stats.nsecs,
1411                    stats.cnt);
1412 }
1413 #endif
1414
1415 const struct file_operations bpf_prog_fops = {
1416 #ifdef CONFIG_PROC_FS
1417         .show_fdinfo    = bpf_prog_show_fdinfo,
1418 #endif
1419         .release        = bpf_prog_release,
1420         .read           = bpf_dummy_read,
1421         .write          = bpf_dummy_write,
1422 };
1423
1424 int bpf_prog_new_fd(struct bpf_prog *prog)
1425 {
1426         int ret;
1427
1428         ret = security_bpf_prog(prog);
1429         if (ret < 0)
1430                 return ret;
1431
1432         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1433                                 O_RDWR | O_CLOEXEC);
1434 }
1435
1436 static struct bpf_prog *____bpf_prog_get(struct fd f)
1437 {
1438         if (!f.file)
1439                 return ERR_PTR(-EBADF);
1440         if (f.file->f_op != &bpf_prog_fops) {
1441                 fdput(f);
1442                 return ERR_PTR(-EINVAL);
1443         }
1444
1445         return f.file->private_data;
1446 }
1447
1448 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1449 {
1450         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1451                 atomic_sub(i, &prog->aux->refcnt);
1452                 return ERR_PTR(-EBUSY);
1453         }
1454         return prog;
1455 }
1456 EXPORT_SYMBOL_GPL(bpf_prog_add);
1457
1458 void bpf_prog_sub(struct bpf_prog *prog, int i)
1459 {
1460         /* Only to be used for undoing previous bpf_prog_add() in some
1461          * error path. We still know that another entity in our call
1462          * path holds a reference to the program, thus atomic_sub() can
1463          * be safely used in such cases!
1464          */
1465         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1466 }
1467 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1468
1469 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1470 {
1471         return bpf_prog_add(prog, 1);
1472 }
1473 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1474
1475 /* prog_idr_lock should have been held */
1476 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1477 {
1478         int refold;
1479
1480         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1481
1482         if (refold >= BPF_MAX_REFCNT) {
1483                 __bpf_prog_put(prog, false);
1484                 return ERR_PTR(-EBUSY);
1485         }
1486
1487         if (!refold)
1488                 return ERR_PTR(-ENOENT);
1489
1490         return prog;
1491 }
1492 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1493
1494 bool bpf_prog_get_ok(struct bpf_prog *prog,
1495                             enum bpf_prog_type *attach_type, bool attach_drv)
1496 {
1497         /* not an attachment, just a refcount inc, always allow */
1498         if (!attach_type)
1499                 return true;
1500
1501         if (prog->type != *attach_type)
1502                 return false;
1503         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1504                 return false;
1505
1506         return true;
1507 }
1508
1509 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1510                                        bool attach_drv)
1511 {
1512         struct fd f = fdget(ufd);
1513         struct bpf_prog *prog;
1514
1515         prog = ____bpf_prog_get(f);
1516         if (IS_ERR(prog))
1517                 return prog;
1518         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1519                 prog = ERR_PTR(-EINVAL);
1520                 goto out;
1521         }
1522
1523         prog = bpf_prog_inc(prog);
1524 out:
1525         fdput(f);
1526         return prog;
1527 }
1528
1529 struct bpf_prog *bpf_prog_get(u32 ufd)
1530 {
1531         return __bpf_prog_get(ufd, NULL, false);
1532 }
1533
1534 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1535                                        bool attach_drv)
1536 {
1537         return __bpf_prog_get(ufd, &type, attach_drv);
1538 }
1539 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1540
1541 /* Initially all BPF programs could be loaded w/o specifying
1542  * expected_attach_type. Later for some of them specifying expected_attach_type
1543  * at load time became required so that program could be validated properly.
1544  * Programs of types that are allowed to be loaded both w/ and w/o (for
1545  * backward compatibility) expected_attach_type, should have the default attach
1546  * type assigned to expected_attach_type for the latter case, so that it can be
1547  * validated later at attach time.
1548  *
1549  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1550  * prog type requires it but has some attach types that have to be backward
1551  * compatible.
1552  */
1553 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1554 {
1555         switch (attr->prog_type) {
1556         case BPF_PROG_TYPE_CGROUP_SOCK:
1557                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1558                  * exist so checking for non-zero is the way to go here.
1559                  */
1560                 if (!attr->expected_attach_type)
1561                         attr->expected_attach_type =
1562                                 BPF_CGROUP_INET_SOCK_CREATE;
1563                 break;
1564         }
1565 }
1566
1567 static int
1568 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1569                                 enum bpf_attach_type expected_attach_type)
1570 {
1571         switch (prog_type) {
1572         case BPF_PROG_TYPE_CGROUP_SOCK:
1573                 switch (expected_attach_type) {
1574                 case BPF_CGROUP_INET_SOCK_CREATE:
1575                 case BPF_CGROUP_INET4_POST_BIND:
1576                 case BPF_CGROUP_INET6_POST_BIND:
1577                         return 0;
1578                 default:
1579                         return -EINVAL;
1580                 }
1581         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1582                 switch (expected_attach_type) {
1583                 case BPF_CGROUP_INET4_BIND:
1584                 case BPF_CGROUP_INET6_BIND:
1585                 case BPF_CGROUP_INET4_CONNECT:
1586                 case BPF_CGROUP_INET6_CONNECT:
1587                 case BPF_CGROUP_UDP4_SENDMSG:
1588                 case BPF_CGROUP_UDP6_SENDMSG:
1589                 case BPF_CGROUP_UDP4_RECVMSG:
1590                 case BPF_CGROUP_UDP6_RECVMSG:
1591                         return 0;
1592                 default:
1593                         return -EINVAL;
1594                 }
1595         case BPF_PROG_TYPE_CGROUP_SKB:
1596                 switch (expected_attach_type) {
1597                 case BPF_CGROUP_INET_INGRESS:
1598                 case BPF_CGROUP_INET_EGRESS:
1599                         return 0;
1600                 default:
1601                         return -EINVAL;
1602                 }
1603         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1604                 switch (expected_attach_type) {
1605                 case BPF_CGROUP_SETSOCKOPT:
1606                 case BPF_CGROUP_GETSOCKOPT:
1607                         return 0;
1608                 default:
1609                         return -EINVAL;
1610                 }
1611         default:
1612                 return 0;
1613         }
1614 }
1615
1616 /* last field in 'union bpf_attr' used by this command */
1617 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1618
1619 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1620 {
1621         enum bpf_prog_type type = attr->prog_type;
1622         struct bpf_prog *prog;
1623         int err;
1624         char license[128];
1625         bool is_gpl;
1626
1627         if (CHECK_ATTR(BPF_PROG_LOAD))
1628                 return -EINVAL;
1629
1630         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1631                                  BPF_F_ANY_ALIGNMENT |
1632                                  BPF_F_TEST_STATE_FREQ |
1633                                  BPF_F_TEST_RND_HI32))
1634                 return -EINVAL;
1635
1636         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1637             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1638             !capable(CAP_SYS_ADMIN))
1639                 return -EPERM;
1640
1641         /* copy eBPF program license from user space */
1642         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1643                               sizeof(license) - 1) < 0)
1644                 return -EFAULT;
1645         license[sizeof(license) - 1] = 0;
1646
1647         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1648         is_gpl = license_is_gpl_compatible(license);
1649
1650         if (attr->insn_cnt == 0 ||
1651             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1652                 return -E2BIG;
1653         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1654             type != BPF_PROG_TYPE_CGROUP_SKB &&
1655             !capable(CAP_SYS_ADMIN))
1656                 return -EPERM;
1657
1658         bpf_prog_load_fixup_attach_type(attr);
1659         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1660                 return -EINVAL;
1661
1662         /* plain bpf_prog allocation */
1663         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1664         if (!prog)
1665                 return -ENOMEM;
1666
1667         prog->expected_attach_type = attr->expected_attach_type;
1668
1669         prog->aux->offload_requested = !!attr->prog_ifindex;
1670
1671         err = security_bpf_prog_alloc(prog->aux);
1672         if (err)
1673                 goto free_prog_nouncharge;
1674
1675         err = bpf_prog_charge_memlock(prog);
1676         if (err)
1677                 goto free_prog_sec;
1678
1679         prog->len = attr->insn_cnt;
1680
1681         err = -EFAULT;
1682         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1683                            bpf_prog_insn_size(prog)) != 0)
1684                 goto free_prog;
1685
1686         prog->orig_prog = NULL;
1687         prog->jited = 0;
1688
1689         atomic_set(&prog->aux->refcnt, 1);
1690         prog->gpl_compatible = is_gpl ? 1 : 0;
1691
1692         if (bpf_prog_is_dev_bound(prog->aux)) {
1693                 err = bpf_prog_offload_init(prog, attr);
1694                 if (err)
1695                         goto free_prog;
1696         }
1697
1698         /* find program type: socket_filter vs tracing_filter */
1699         err = find_prog_type(type, prog);
1700         if (err < 0)
1701                 goto free_prog;
1702
1703         prog->aux->load_time = ktime_get_boottime_ns();
1704         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1705         if (err)
1706                 goto free_prog;
1707
1708         /* run eBPF verifier */
1709         err = bpf_check(&prog, attr, uattr);
1710         if (err < 0)
1711                 goto free_used_maps;
1712
1713         prog = bpf_prog_select_runtime(prog, &err);
1714         if (err < 0)
1715                 goto free_used_maps;
1716
1717         err = bpf_prog_alloc_id(prog);
1718         if (err)
1719                 goto free_used_maps;
1720
1721         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1722          * effectively publicly exposed. However, retrieving via
1723          * bpf_prog_get_fd_by_id() will take another reference,
1724          * therefore it cannot be gone underneath us.
1725          *
1726          * Only for the time /after/ successful bpf_prog_new_fd()
1727          * and before returning to userspace, we might just hold
1728          * one reference and any parallel close on that fd could
1729          * rip everything out. Hence, below notifications must
1730          * happen before bpf_prog_new_fd().
1731          *
1732          * Also, any failure handling from this point onwards must
1733          * be using bpf_prog_put() given the program is exposed.
1734          */
1735         bpf_prog_kallsyms_add(prog);
1736         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1737
1738         err = bpf_prog_new_fd(prog);
1739         if (err < 0)
1740                 bpf_prog_put(prog);
1741         return err;
1742
1743 free_used_maps:
1744         bpf_prog_free_linfo(prog);
1745         kvfree(prog->aux->func_info);
1746         btf_put(prog->aux->btf);
1747         bpf_prog_kallsyms_del_subprogs(prog);
1748         free_used_maps(prog->aux);
1749 free_prog:
1750         bpf_prog_uncharge_memlock(prog);
1751 free_prog_sec:
1752         security_bpf_prog_free(prog->aux);
1753 free_prog_nouncharge:
1754         bpf_prog_free(prog);
1755         return err;
1756 }
1757
1758 #define BPF_OBJ_LAST_FIELD file_flags
1759
1760 static int bpf_obj_pin(const union bpf_attr *attr)
1761 {
1762         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1763                 return -EINVAL;
1764
1765         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1766 }
1767
1768 static int bpf_obj_get(const union bpf_attr *attr)
1769 {
1770         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1771             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1772                 return -EINVAL;
1773
1774         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1775                                 attr->file_flags);
1776 }
1777
1778 struct bpf_raw_tracepoint {
1779         struct bpf_raw_event_map *btp;
1780         struct bpf_prog *prog;
1781 };
1782
1783 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1784 {
1785         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1786
1787         if (raw_tp->prog) {
1788                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1789                 bpf_prog_put(raw_tp->prog);
1790         }
1791         bpf_put_raw_tracepoint(raw_tp->btp);
1792         kfree(raw_tp);
1793         return 0;
1794 }
1795
1796 static const struct file_operations bpf_raw_tp_fops = {
1797         .release        = bpf_raw_tracepoint_release,
1798         .read           = bpf_dummy_read,
1799         .write          = bpf_dummy_write,
1800 };
1801
1802 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1803
1804 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1805 {
1806         struct bpf_raw_tracepoint *raw_tp;
1807         struct bpf_raw_event_map *btp;
1808         struct bpf_prog *prog;
1809         char tp_name[128];
1810         int tp_fd, err;
1811
1812         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1813                               sizeof(tp_name) - 1) < 0)
1814                 return -EFAULT;
1815         tp_name[sizeof(tp_name) - 1] = 0;
1816
1817         btp = bpf_get_raw_tracepoint(tp_name);
1818         if (!btp)
1819                 return -ENOENT;
1820
1821         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1822         if (!raw_tp) {
1823                 err = -ENOMEM;
1824                 goto out_put_btp;
1825         }
1826         raw_tp->btp = btp;
1827
1828         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1829         if (IS_ERR(prog)) {
1830                 err = PTR_ERR(prog);
1831                 goto out_free_tp;
1832         }
1833         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1834             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1835                 err = -EINVAL;
1836                 goto out_put_prog;
1837         }
1838
1839         err = bpf_probe_register(raw_tp->btp, prog);
1840         if (err)
1841                 goto out_put_prog;
1842
1843         raw_tp->prog = prog;
1844         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1845                                  O_CLOEXEC);
1846         if (tp_fd < 0) {
1847                 bpf_probe_unregister(raw_tp->btp, prog);
1848                 err = tp_fd;
1849                 goto out_put_prog;
1850         }
1851         return tp_fd;
1852
1853 out_put_prog:
1854         bpf_prog_put(prog);
1855 out_free_tp:
1856         kfree(raw_tp);
1857 out_put_btp:
1858         bpf_put_raw_tracepoint(btp);
1859         return err;
1860 }
1861
1862 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1863                                              enum bpf_attach_type attach_type)
1864 {
1865         switch (prog->type) {
1866         case BPF_PROG_TYPE_CGROUP_SOCK:
1867         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1868         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1869                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1870         case BPF_PROG_TYPE_CGROUP_SKB:
1871                 return prog->enforce_expected_attach_type &&
1872                         prog->expected_attach_type != attach_type ?
1873                         -EINVAL : 0;
1874         default:
1875                 return 0;
1876         }
1877 }
1878
1879 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1880
1881 #define BPF_F_ATTACH_MASK \
1882         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1883
1884 static int bpf_prog_attach(const union bpf_attr *attr)
1885 {
1886         enum bpf_prog_type ptype;
1887         struct bpf_prog *prog;
1888         int ret;
1889
1890         if (!capable(CAP_NET_ADMIN))
1891                 return -EPERM;
1892
1893         if (CHECK_ATTR(BPF_PROG_ATTACH))
1894                 return -EINVAL;
1895
1896         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1897                 return -EINVAL;
1898
1899         switch (attr->attach_type) {
1900         case BPF_CGROUP_INET_INGRESS:
1901         case BPF_CGROUP_INET_EGRESS:
1902                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1903                 break;
1904         case BPF_CGROUP_INET_SOCK_CREATE:
1905         case BPF_CGROUP_INET4_POST_BIND:
1906         case BPF_CGROUP_INET6_POST_BIND:
1907                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1908                 break;
1909         case BPF_CGROUP_INET4_BIND:
1910         case BPF_CGROUP_INET6_BIND:
1911         case BPF_CGROUP_INET4_CONNECT:
1912         case BPF_CGROUP_INET6_CONNECT:
1913         case BPF_CGROUP_UDP4_SENDMSG:
1914         case BPF_CGROUP_UDP6_SENDMSG:
1915         case BPF_CGROUP_UDP4_RECVMSG:
1916         case BPF_CGROUP_UDP6_RECVMSG:
1917                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1918                 break;
1919         case BPF_CGROUP_SOCK_OPS:
1920                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1921                 break;
1922         case BPF_CGROUP_DEVICE:
1923                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1924                 break;
1925         case BPF_SK_MSG_VERDICT:
1926                 ptype = BPF_PROG_TYPE_SK_MSG;
1927                 break;
1928         case BPF_SK_SKB_STREAM_PARSER:
1929         case BPF_SK_SKB_STREAM_VERDICT:
1930                 ptype = BPF_PROG_TYPE_SK_SKB;
1931                 break;
1932         case BPF_LIRC_MODE2:
1933                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1934                 break;
1935         case BPF_FLOW_DISSECTOR:
1936                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1937                 break;
1938         case BPF_CGROUP_SYSCTL:
1939                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1940                 break;
1941         case BPF_CGROUP_GETSOCKOPT:
1942         case BPF_CGROUP_SETSOCKOPT:
1943                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1944                 break;
1945         default:
1946                 return -EINVAL;
1947         }
1948
1949         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1950         if (IS_ERR(prog))
1951                 return PTR_ERR(prog);
1952
1953         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1954                 bpf_prog_put(prog);
1955                 return -EINVAL;
1956         }
1957
1958         switch (ptype) {
1959         case BPF_PROG_TYPE_SK_SKB:
1960         case BPF_PROG_TYPE_SK_MSG:
1961                 ret = sock_map_get_from_fd(attr, prog);
1962                 break;
1963         case BPF_PROG_TYPE_LIRC_MODE2:
1964                 ret = lirc_prog_attach(attr, prog);
1965                 break;
1966         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1967                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1968                 break;
1969         default:
1970                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1971         }
1972
1973         if (ret)
1974                 bpf_prog_put(prog);
1975         return ret;
1976 }
1977
1978 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1979
1980 static int bpf_prog_detach(const union bpf_attr *attr)
1981 {
1982         enum bpf_prog_type ptype;
1983
1984         if (!capable(CAP_NET_ADMIN))
1985                 return -EPERM;
1986
1987         if (CHECK_ATTR(BPF_PROG_DETACH))
1988                 return -EINVAL;
1989
1990         switch (attr->attach_type) {
1991         case BPF_CGROUP_INET_INGRESS:
1992         case BPF_CGROUP_INET_EGRESS:
1993                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1994                 break;
1995         case BPF_CGROUP_INET_SOCK_CREATE:
1996         case BPF_CGROUP_INET4_POST_BIND:
1997         case BPF_CGROUP_INET6_POST_BIND:
1998                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1999                 break;
2000         case BPF_CGROUP_INET4_BIND:
2001         case BPF_CGROUP_INET6_BIND:
2002         case BPF_CGROUP_INET4_CONNECT:
2003         case BPF_CGROUP_INET6_CONNECT:
2004         case BPF_CGROUP_UDP4_SENDMSG:
2005         case BPF_CGROUP_UDP6_SENDMSG:
2006         case BPF_CGROUP_UDP4_RECVMSG:
2007         case BPF_CGROUP_UDP6_RECVMSG:
2008                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2009                 break;
2010         case BPF_CGROUP_SOCK_OPS:
2011                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2012                 break;
2013         case BPF_CGROUP_DEVICE:
2014                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2015                 break;
2016         case BPF_SK_MSG_VERDICT:
2017                 return sock_map_get_from_fd(attr, NULL);
2018         case BPF_SK_SKB_STREAM_PARSER:
2019         case BPF_SK_SKB_STREAM_VERDICT:
2020                 return sock_map_get_from_fd(attr, NULL);
2021         case BPF_LIRC_MODE2:
2022                 return lirc_prog_detach(attr);
2023         case BPF_FLOW_DISSECTOR:
2024                 return skb_flow_dissector_bpf_prog_detach(attr);
2025         case BPF_CGROUP_SYSCTL:
2026                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2027                 break;
2028         case BPF_CGROUP_GETSOCKOPT:
2029         case BPF_CGROUP_SETSOCKOPT:
2030                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2031                 break;
2032         default:
2033                 return -EINVAL;
2034         }
2035
2036         return cgroup_bpf_prog_detach(attr, ptype);
2037 }
2038
2039 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2040
2041 static int bpf_prog_query(const union bpf_attr *attr,
2042                           union bpf_attr __user *uattr)
2043 {
2044         if (!capable(CAP_NET_ADMIN))
2045                 return -EPERM;
2046         if (CHECK_ATTR(BPF_PROG_QUERY))
2047                 return -EINVAL;
2048         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2049                 return -EINVAL;
2050
2051         switch (attr->query.attach_type) {
2052         case BPF_CGROUP_INET_INGRESS:
2053         case BPF_CGROUP_INET_EGRESS:
2054         case BPF_CGROUP_INET_SOCK_CREATE:
2055         case BPF_CGROUP_INET4_BIND:
2056         case BPF_CGROUP_INET6_BIND:
2057         case BPF_CGROUP_INET4_POST_BIND:
2058         case BPF_CGROUP_INET6_POST_BIND:
2059         case BPF_CGROUP_INET4_CONNECT:
2060         case BPF_CGROUP_INET6_CONNECT:
2061         case BPF_CGROUP_UDP4_SENDMSG:
2062         case BPF_CGROUP_UDP6_SENDMSG:
2063         case BPF_CGROUP_UDP4_RECVMSG:
2064         case BPF_CGROUP_UDP6_RECVMSG:
2065         case BPF_CGROUP_SOCK_OPS:
2066         case BPF_CGROUP_DEVICE:
2067         case BPF_CGROUP_SYSCTL:
2068         case BPF_CGROUP_GETSOCKOPT:
2069         case BPF_CGROUP_SETSOCKOPT:
2070                 break;
2071         case BPF_LIRC_MODE2:
2072                 return lirc_prog_query(attr, uattr);
2073         case BPF_FLOW_DISSECTOR:
2074                 return skb_flow_dissector_prog_query(attr, uattr);
2075         default:
2076                 return -EINVAL;
2077         }
2078
2079         return cgroup_bpf_prog_query(attr, uattr);
2080 }
2081
2082 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2083
2084 static int bpf_prog_test_run(const union bpf_attr *attr,
2085                              union bpf_attr __user *uattr)
2086 {
2087         struct bpf_prog *prog;
2088         int ret = -ENOTSUPP;
2089
2090         if (!capable(CAP_SYS_ADMIN))
2091                 return -EPERM;
2092         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2093                 return -EINVAL;
2094
2095         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2096             (!attr->test.ctx_size_in && attr->test.ctx_in))
2097                 return -EINVAL;
2098
2099         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2100             (!attr->test.ctx_size_out && attr->test.ctx_out))
2101                 return -EINVAL;
2102
2103         prog = bpf_prog_get(attr->test.prog_fd);
2104         if (IS_ERR(prog))
2105                 return PTR_ERR(prog);
2106
2107         if (prog->aux->ops->test_run)
2108                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2109
2110         bpf_prog_put(prog);
2111         return ret;
2112 }
2113
2114 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2115
2116 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2117                                union bpf_attr __user *uattr,
2118                                struct idr *idr,
2119                                spinlock_t *lock)
2120 {
2121         u32 next_id = attr->start_id;
2122         int err = 0;
2123
2124         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2125                 return -EINVAL;
2126
2127         if (!capable(CAP_SYS_ADMIN))
2128                 return -EPERM;
2129
2130         next_id++;
2131         spin_lock_bh(lock);
2132         if (!idr_get_next(idr, &next_id))
2133                 err = -ENOENT;
2134         spin_unlock_bh(lock);
2135
2136         if (!err)
2137                 err = put_user(next_id, &uattr->next_id);
2138
2139         return err;
2140 }
2141
2142 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2143
2144 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2145 {
2146         struct bpf_prog *prog;
2147         u32 id = attr->prog_id;
2148         int fd;
2149
2150         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2151                 return -EINVAL;
2152
2153         if (!capable(CAP_SYS_ADMIN))
2154                 return -EPERM;
2155
2156         spin_lock_bh(&prog_idr_lock);
2157         prog = idr_find(&prog_idr, id);
2158         if (prog)
2159                 prog = bpf_prog_inc_not_zero(prog);
2160         else
2161                 prog = ERR_PTR(-ENOENT);
2162         spin_unlock_bh(&prog_idr_lock);
2163
2164         if (IS_ERR(prog))
2165                 return PTR_ERR(prog);
2166
2167         fd = bpf_prog_new_fd(prog);
2168         if (fd < 0)
2169                 bpf_prog_put(prog);
2170
2171         return fd;
2172 }
2173
2174 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2175
2176 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2177 {
2178         struct bpf_map *map;
2179         u32 id = attr->map_id;
2180         int f_flags;
2181         int fd;
2182
2183         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2184             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2185                 return -EINVAL;
2186
2187         if (!capable(CAP_SYS_ADMIN))
2188                 return -EPERM;
2189
2190         f_flags = bpf_get_file_flag(attr->open_flags);
2191         if (f_flags < 0)
2192                 return f_flags;
2193
2194         spin_lock_bh(&map_idr_lock);
2195         map = idr_find(&map_idr, id);
2196         if (map)
2197                 map = __bpf_map_inc_not_zero(map, true);
2198         else
2199                 map = ERR_PTR(-ENOENT);
2200         spin_unlock_bh(&map_idr_lock);
2201
2202         if (IS_ERR(map))
2203                 return PTR_ERR(map);
2204
2205         fd = bpf_map_new_fd(map, f_flags);
2206         if (fd < 0)
2207                 bpf_map_put_with_uref(map);
2208
2209         return fd;
2210 }
2211
2212 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2213                                               unsigned long addr, u32 *off,
2214                                               u32 *type)
2215 {
2216         const struct bpf_map *map;
2217         int i;
2218
2219         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2220                 map = prog->aux->used_maps[i];
2221                 if (map == (void *)addr) {
2222                         *type = BPF_PSEUDO_MAP_FD;
2223                         return map;
2224                 }
2225                 if (!map->ops->map_direct_value_meta)
2226                         continue;
2227                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2228                         *type = BPF_PSEUDO_MAP_VALUE;
2229                         return map;
2230                 }
2231         }
2232
2233         return NULL;
2234 }
2235
2236 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2237 {
2238         const struct bpf_map *map;
2239         struct bpf_insn *insns;
2240         u32 off, type;
2241         u64 imm;
2242         int i;
2243
2244         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2245                         GFP_USER);
2246         if (!insns)
2247                 return insns;
2248
2249         for (i = 0; i < prog->len; i++) {
2250                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2251                         insns[i].code = BPF_JMP | BPF_CALL;
2252                         insns[i].imm = BPF_FUNC_tail_call;
2253                         /* fall-through */
2254                 }
2255                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2256                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2257                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2258                                 insns[i].code = BPF_JMP | BPF_CALL;
2259                         if (!bpf_dump_raw_ok())
2260                                 insns[i].imm = 0;
2261                         continue;
2262                 }
2263
2264                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2265                         continue;
2266
2267                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2268                 map = bpf_map_from_imm(prog, imm, &off, &type);
2269                 if (map) {
2270                         insns[i].src_reg = type;
2271                         insns[i].imm = map->id;
2272                         insns[i + 1].imm = off;
2273                         continue;
2274                 }
2275         }
2276
2277         return insns;
2278 }
2279
2280 static int set_info_rec_size(struct bpf_prog_info *info)
2281 {
2282         /*
2283          * Ensure info.*_rec_size is the same as kernel expected size
2284          *
2285          * or
2286          *
2287          * Only allow zero *_rec_size if both _rec_size and _cnt are
2288          * zero.  In this case, the kernel will set the expected
2289          * _rec_size back to the info.
2290          */
2291
2292         if ((info->nr_func_info || info->func_info_rec_size) &&
2293             info->func_info_rec_size != sizeof(struct bpf_func_info))
2294                 return -EINVAL;
2295
2296         if ((info->nr_line_info || info->line_info_rec_size) &&
2297             info->line_info_rec_size != sizeof(struct bpf_line_info))
2298                 return -EINVAL;
2299
2300         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2301             info->jited_line_info_rec_size != sizeof(__u64))
2302                 return -EINVAL;
2303
2304         info->func_info_rec_size = sizeof(struct bpf_func_info);
2305         info->line_info_rec_size = sizeof(struct bpf_line_info);
2306         info->jited_line_info_rec_size = sizeof(__u64);
2307
2308         return 0;
2309 }
2310
2311 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2312                                    const union bpf_attr *attr,
2313                                    union bpf_attr __user *uattr)
2314 {
2315         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2316         struct bpf_prog_info info = {};
2317         u32 info_len = attr->info.info_len;
2318         struct bpf_prog_stats stats;
2319         char __user *uinsns;
2320         u32 ulen;
2321         int err;
2322
2323         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2324         if (err)
2325                 return err;
2326         info_len = min_t(u32, sizeof(info), info_len);
2327
2328         if (copy_from_user(&info, uinfo, info_len))
2329                 return -EFAULT;
2330
2331         info.type = prog->type;
2332         info.id = prog->aux->id;
2333         info.load_time = prog->aux->load_time;
2334         info.created_by_uid = from_kuid_munged(current_user_ns(),
2335                                                prog->aux->user->uid);
2336         info.gpl_compatible = prog->gpl_compatible;
2337
2338         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2339         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2340
2341         ulen = info.nr_map_ids;
2342         info.nr_map_ids = prog->aux->used_map_cnt;
2343         ulen = min_t(u32, info.nr_map_ids, ulen);
2344         if (ulen) {
2345                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2346                 u32 i;
2347
2348                 for (i = 0; i < ulen; i++)
2349                         if (put_user(prog->aux->used_maps[i]->id,
2350                                      &user_map_ids[i]))
2351                                 return -EFAULT;
2352         }
2353
2354         err = set_info_rec_size(&info);
2355         if (err)
2356                 return err;
2357
2358         bpf_prog_get_stats(prog, &stats);
2359         info.run_time_ns = stats.nsecs;
2360         info.run_cnt = stats.cnt;
2361
2362         if (!capable(CAP_SYS_ADMIN)) {
2363                 info.jited_prog_len = 0;
2364                 info.xlated_prog_len = 0;
2365                 info.nr_jited_ksyms = 0;
2366                 info.nr_jited_func_lens = 0;
2367                 info.nr_func_info = 0;
2368                 info.nr_line_info = 0;
2369                 info.nr_jited_line_info = 0;
2370                 goto done;
2371         }
2372
2373         ulen = info.xlated_prog_len;
2374         info.xlated_prog_len = bpf_prog_insn_size(prog);
2375         if (info.xlated_prog_len && ulen) {
2376                 struct bpf_insn *insns_sanitized;
2377                 bool fault;
2378
2379                 if (prog->blinded && !bpf_dump_raw_ok()) {
2380                         info.xlated_prog_insns = 0;
2381                         goto done;
2382                 }
2383                 insns_sanitized = bpf_insn_prepare_dump(prog);
2384                 if (!insns_sanitized)
2385                         return -ENOMEM;
2386                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2387                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2388                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2389                 kfree(insns_sanitized);
2390                 if (fault)
2391                         return -EFAULT;
2392         }
2393
2394         if (bpf_prog_is_dev_bound(prog->aux)) {
2395                 err = bpf_prog_offload_info_fill(&info, prog);
2396                 if (err)
2397                         return err;
2398                 goto done;
2399         }
2400
2401         /* NOTE: the following code is supposed to be skipped for offload.
2402          * bpf_prog_offload_info_fill() is the place to fill similar fields
2403          * for offload.
2404          */
2405         ulen = info.jited_prog_len;
2406         if (prog->aux->func_cnt) {
2407                 u32 i;
2408
2409                 info.jited_prog_len = 0;
2410                 for (i = 0; i < prog->aux->func_cnt; i++)
2411                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2412         } else {
2413                 info.jited_prog_len = prog->jited_len;
2414         }
2415
2416         if (info.jited_prog_len && ulen) {
2417                 if (bpf_dump_raw_ok()) {
2418                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2419                         ulen = min_t(u32, info.jited_prog_len, ulen);
2420
2421                         /* for multi-function programs, copy the JITed
2422                          * instructions for all the functions
2423                          */
2424                         if (prog->aux->func_cnt) {
2425                                 u32 len, free, i;
2426                                 u8 *img;
2427
2428                                 free = ulen;
2429                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2430                                         len = prog->aux->func[i]->jited_len;
2431                                         len = min_t(u32, len, free);
2432                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2433                                         if (copy_to_user(uinsns, img, len))
2434                                                 return -EFAULT;
2435                                         uinsns += len;
2436                                         free -= len;
2437                                         if (!free)
2438                                                 break;
2439                                 }
2440                         } else {
2441                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2442                                         return -EFAULT;
2443                         }
2444                 } else {
2445                         info.jited_prog_insns = 0;
2446                 }
2447         }
2448
2449         ulen = info.nr_jited_ksyms;
2450         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2451         if (ulen) {
2452                 if (bpf_dump_raw_ok()) {
2453                         unsigned long ksym_addr;
2454                         u64 __user *user_ksyms;
2455                         u32 i;
2456
2457                         /* copy the address of the kernel symbol
2458                          * corresponding to each function
2459                          */
2460                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2461                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2462                         if (prog->aux->func_cnt) {
2463                                 for (i = 0; i < ulen; i++) {
2464                                         ksym_addr = (unsigned long)
2465                                                 prog->aux->func[i]->bpf_func;
2466                                         if (put_user((u64) ksym_addr,
2467                                                      &user_ksyms[i]))
2468                                                 return -EFAULT;
2469                                 }
2470                         } else {
2471                                 ksym_addr = (unsigned long) prog->bpf_func;
2472                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2473                                         return -EFAULT;
2474                         }
2475                 } else {
2476                         info.jited_ksyms = 0;
2477                 }
2478         }
2479
2480         ulen = info.nr_jited_func_lens;
2481         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2482         if (ulen) {
2483                 if (bpf_dump_raw_ok()) {
2484                         u32 __user *user_lens;
2485                         u32 func_len, i;
2486
2487                         /* copy the JITed image lengths for each function */
2488                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2489                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2490                         if (prog->aux->func_cnt) {
2491                                 for (i = 0; i < ulen; i++) {
2492                                         func_len =
2493                                                 prog->aux->func[i]->jited_len;
2494                                         if (put_user(func_len, &user_lens[i]))
2495                                                 return -EFAULT;
2496                                 }
2497                         } else {
2498                                 func_len = prog->jited_len;
2499                                 if (put_user(func_len, &user_lens[0]))
2500                                         return -EFAULT;
2501                         }
2502                 } else {
2503                         info.jited_func_lens = 0;
2504                 }
2505         }
2506
2507         if (prog->aux->btf)
2508                 info.btf_id = btf_id(prog->aux->btf);
2509
2510         ulen = info.nr_func_info;
2511         info.nr_func_info = prog->aux->func_info_cnt;
2512         if (info.nr_func_info && ulen) {
2513                 char __user *user_finfo;
2514
2515                 user_finfo = u64_to_user_ptr(info.func_info);
2516                 ulen = min_t(u32, info.nr_func_info, ulen);
2517                 if (copy_to_user(user_finfo, prog->aux->func_info,
2518                                  info.func_info_rec_size * ulen))
2519                         return -EFAULT;
2520         }
2521
2522         ulen = info.nr_line_info;
2523         info.nr_line_info = prog->aux->nr_linfo;
2524         if (info.nr_line_info && ulen) {
2525                 __u8 __user *user_linfo;
2526
2527                 user_linfo = u64_to_user_ptr(info.line_info);
2528                 ulen = min_t(u32, info.nr_line_info, ulen);
2529                 if (copy_to_user(user_linfo, prog->aux->linfo,
2530                                  info.line_info_rec_size * ulen))
2531                         return -EFAULT;
2532         }
2533
2534         ulen = info.nr_jited_line_info;
2535         if (prog->aux->jited_linfo)
2536                 info.nr_jited_line_info = prog->aux->nr_linfo;
2537         else
2538                 info.nr_jited_line_info = 0;
2539         if (info.nr_jited_line_info && ulen) {
2540                 if (bpf_dump_raw_ok()) {
2541                         __u64 __user *user_linfo;
2542                         u32 i;
2543
2544                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2545                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2546                         for (i = 0; i < ulen; i++) {
2547                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2548                                              &user_linfo[i]))
2549                                         return -EFAULT;
2550                         }
2551                 } else {
2552                         info.jited_line_info = 0;
2553                 }
2554         }
2555
2556         ulen = info.nr_prog_tags;
2557         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2558         if (ulen) {
2559                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2560                 u32 i;
2561
2562                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2563                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2564                 if (prog->aux->func_cnt) {
2565                         for (i = 0; i < ulen; i++) {
2566                                 if (copy_to_user(user_prog_tags[i],
2567                                                  prog->aux->func[i]->tag,
2568                                                  BPF_TAG_SIZE))
2569                                         return -EFAULT;
2570                         }
2571                 } else {
2572                         if (copy_to_user(user_prog_tags[0],
2573                                          prog->tag, BPF_TAG_SIZE))
2574                                 return -EFAULT;
2575                 }
2576         }
2577
2578 done:
2579         if (copy_to_user(uinfo, &info, info_len) ||
2580             put_user(info_len, &uattr->info.info_len))
2581                 return -EFAULT;
2582
2583         return 0;
2584 }
2585
2586 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2587                                   const union bpf_attr *attr,
2588                                   union bpf_attr __user *uattr)
2589 {
2590         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2591         struct bpf_map_info info = {};
2592         u32 info_len = attr->info.info_len;
2593         int err;
2594
2595         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2596         if (err)
2597                 return err;
2598         info_len = min_t(u32, sizeof(info), info_len);
2599
2600         info.type = map->map_type;
2601         info.id = map->id;
2602         info.key_size = map->key_size;
2603         info.value_size = map->value_size;
2604         info.max_entries = map->max_entries;
2605         info.map_flags = map->map_flags;
2606         memcpy(info.name, map->name, sizeof(map->name));
2607
2608         if (map->btf) {
2609                 info.btf_id = btf_id(map->btf);
2610                 info.btf_key_type_id = map->btf_key_type_id;
2611                 info.btf_value_type_id = map->btf_value_type_id;
2612         }
2613
2614         if (bpf_map_is_dev_bound(map)) {
2615                 err = bpf_map_offload_info_fill(&info, map);
2616                 if (err)
2617                         return err;
2618         }
2619
2620         if (copy_to_user(uinfo, &info, info_len) ||
2621             put_user(info_len, &uattr->info.info_len))
2622                 return -EFAULT;
2623
2624         return 0;
2625 }
2626
2627 static int bpf_btf_get_info_by_fd(struct btf *btf,
2628                                   const union bpf_attr *attr,
2629                                   union bpf_attr __user *uattr)
2630 {
2631         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2632         u32 info_len = attr->info.info_len;
2633         int err;
2634
2635         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2636         if (err)
2637                 return err;
2638
2639         return btf_get_info_by_fd(btf, attr, uattr);
2640 }
2641
2642 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2643
2644 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2645                                   union bpf_attr __user *uattr)
2646 {
2647         int ufd = attr->info.bpf_fd;
2648         struct fd f;
2649         int err;
2650
2651         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2652                 return -EINVAL;
2653
2654         f = fdget(ufd);
2655         if (!f.file)
2656                 return -EBADFD;
2657
2658         if (f.file->f_op == &bpf_prog_fops)
2659                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2660                                               uattr);
2661         else if (f.file->f_op == &bpf_map_fops)
2662                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2663                                              uattr);
2664         else if (f.file->f_op == &btf_fops)
2665                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2666         else
2667                 err = -EINVAL;
2668
2669         fdput(f);
2670         return err;
2671 }
2672
2673 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2674
2675 static int bpf_btf_load(const union bpf_attr *attr)
2676 {
2677         if (CHECK_ATTR(BPF_BTF_LOAD))
2678                 return -EINVAL;
2679
2680         if (!capable(CAP_SYS_ADMIN))
2681                 return -EPERM;
2682
2683         return btf_new_fd(attr);
2684 }
2685
2686 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2687
2688 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2689 {
2690         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2691                 return -EINVAL;
2692
2693         if (!capable(CAP_SYS_ADMIN))
2694                 return -EPERM;
2695
2696         return btf_get_fd_by_id(attr->btf_id);
2697 }
2698
2699 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2700                                     union bpf_attr __user *uattr,
2701                                     u32 prog_id, u32 fd_type,
2702                                     const char *buf, u64 probe_offset,
2703                                     u64 probe_addr)
2704 {
2705         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2706         u32 len = buf ? strlen(buf) : 0, input_len;
2707         int err = 0;
2708
2709         if (put_user(len, &uattr->task_fd_query.buf_len))
2710                 return -EFAULT;
2711         input_len = attr->task_fd_query.buf_len;
2712         if (input_len && ubuf) {
2713                 if (!len) {
2714                         /* nothing to copy, just make ubuf NULL terminated */
2715                         char zero = '\0';
2716
2717                         if (put_user(zero, ubuf))
2718                                 return -EFAULT;
2719                 } else if (input_len >= len + 1) {
2720                         /* ubuf can hold the string with NULL terminator */
2721                         if (copy_to_user(ubuf, buf, len + 1))
2722                                 return -EFAULT;
2723                 } else {
2724                         /* ubuf cannot hold the string with NULL terminator,
2725                          * do a partial copy with NULL terminator.
2726                          */
2727                         char zero = '\0';
2728
2729                         err = -ENOSPC;
2730                         if (copy_to_user(ubuf, buf, input_len - 1))
2731                                 return -EFAULT;
2732                         if (put_user(zero, ubuf + input_len - 1))
2733                                 return -EFAULT;
2734                 }
2735         }
2736
2737         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2738             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2739             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2740             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2741                 return -EFAULT;
2742
2743         return err;
2744 }
2745
2746 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2747
2748 static int bpf_task_fd_query(const union bpf_attr *attr,
2749                              union bpf_attr __user *uattr)
2750 {
2751         pid_t pid = attr->task_fd_query.pid;
2752         u32 fd = attr->task_fd_query.fd;
2753         const struct perf_event *event;
2754         struct files_struct *files;
2755         struct task_struct *task;
2756         struct file *file;
2757         int err;
2758
2759         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2760                 return -EINVAL;
2761
2762         if (!capable(CAP_SYS_ADMIN))
2763                 return -EPERM;
2764
2765         if (attr->task_fd_query.flags != 0)
2766                 return -EINVAL;
2767
2768         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2769         if (!task)
2770                 return -ENOENT;
2771
2772         files = get_files_struct(task);
2773         put_task_struct(task);
2774         if (!files)
2775                 return -ENOENT;
2776
2777         err = 0;
2778         spin_lock(&files->file_lock);
2779         file = fcheck_files(files, fd);
2780         if (!file)
2781                 err = -EBADF;
2782         else
2783                 get_file(file);
2784         spin_unlock(&files->file_lock);
2785         put_files_struct(files);
2786
2787         if (err)
2788                 goto out;
2789
2790         if (file->f_op == &bpf_raw_tp_fops) {
2791                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2792                 struct bpf_raw_event_map *btp = raw_tp->btp;
2793
2794                 err = bpf_task_fd_query_copy(attr, uattr,
2795                                              raw_tp->prog->aux->id,
2796                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2797                                              btp->tp->name, 0, 0);
2798                 goto put_file;
2799         }
2800
2801         event = perf_get_event(file);
2802         if (!IS_ERR(event)) {
2803                 u64 probe_offset, probe_addr;
2804                 u32 prog_id, fd_type;
2805                 const char *buf;
2806
2807                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2808                                               &buf, &probe_offset,
2809                                               &probe_addr);
2810                 if (!err)
2811                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2812                                                      fd_type, buf,
2813                                                      probe_offset,
2814                                                      probe_addr);
2815                 goto put_file;
2816         }
2817
2818         err = -ENOTSUPP;
2819 put_file:
2820         fput(file);
2821 out:
2822         return err;
2823 }
2824
2825 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2826 {
2827         union bpf_attr attr = {};
2828         int err;
2829
2830         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2831                 return -EPERM;
2832
2833         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2834         if (err)
2835                 return err;
2836         size = min_t(u32, size, sizeof(attr));
2837
2838         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2839         if (copy_from_user(&attr, uattr, size) != 0)
2840                 return -EFAULT;
2841
2842         err = security_bpf(cmd, &attr, size);
2843         if (err < 0)
2844                 return err;
2845
2846         switch (cmd) {
2847         case BPF_MAP_CREATE:
2848                 err = map_create(&attr);
2849                 break;
2850         case BPF_MAP_LOOKUP_ELEM:
2851                 err = map_lookup_elem(&attr);
2852                 break;
2853         case BPF_MAP_UPDATE_ELEM:
2854                 err = map_update_elem(&attr);
2855                 break;
2856         case BPF_MAP_DELETE_ELEM:
2857                 err = map_delete_elem(&attr);
2858                 break;
2859         case BPF_MAP_GET_NEXT_KEY:
2860                 err = map_get_next_key(&attr);
2861                 break;
2862         case BPF_MAP_FREEZE:
2863                 err = map_freeze(&attr);
2864                 break;
2865         case BPF_PROG_LOAD:
2866                 err = bpf_prog_load(&attr, uattr);
2867                 break;
2868         case BPF_OBJ_PIN:
2869                 err = bpf_obj_pin(&attr);
2870                 break;
2871         case BPF_OBJ_GET:
2872                 err = bpf_obj_get(&attr);
2873                 break;
2874         case BPF_PROG_ATTACH:
2875                 err = bpf_prog_attach(&attr);
2876                 break;
2877         case BPF_PROG_DETACH:
2878                 err = bpf_prog_detach(&attr);
2879                 break;
2880         case BPF_PROG_QUERY:
2881                 err = bpf_prog_query(&attr, uattr);
2882                 break;
2883         case BPF_PROG_TEST_RUN:
2884                 err = bpf_prog_test_run(&attr, uattr);
2885                 break;
2886         case BPF_PROG_GET_NEXT_ID:
2887                 err = bpf_obj_get_next_id(&attr, uattr,
2888                                           &prog_idr, &prog_idr_lock);
2889                 break;
2890         case BPF_MAP_GET_NEXT_ID:
2891                 err = bpf_obj_get_next_id(&attr, uattr,
2892                                           &map_idr, &map_idr_lock);
2893                 break;
2894         case BPF_BTF_GET_NEXT_ID:
2895                 err = bpf_obj_get_next_id(&attr, uattr,
2896                                           &btf_idr, &btf_idr_lock);
2897                 break;
2898         case BPF_PROG_GET_FD_BY_ID:
2899                 err = bpf_prog_get_fd_by_id(&attr);
2900                 break;
2901         case BPF_MAP_GET_FD_BY_ID:
2902                 err = bpf_map_get_fd_by_id(&attr);
2903                 break;
2904         case BPF_OBJ_GET_INFO_BY_FD:
2905                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2906                 break;
2907         case BPF_RAW_TRACEPOINT_OPEN:
2908                 err = bpf_raw_tracepoint_open(&attr);
2909                 break;
2910         case BPF_BTF_LOAD:
2911                 err = bpf_btf_load(&attr);
2912                 break;
2913         case BPF_BTF_GET_FD_BY_ID:
2914                 err = bpf_btf_get_fd_by_id(&attr);
2915                 break;
2916         case BPF_TASK_FD_QUERY:
2917                 err = bpf_task_fd_query(&attr, uattr);
2918                 break;
2919         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2920                 err = map_lookup_and_delete_elem(&attr);
2921                 break;
2922         default:
2923                 err = -EINVAL;
2924                 break;
2925         }
2926
2927         return err;
2928 }