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         IS_BUILTIN(CONFIG_BPF_UNPRIV_DEFAULT_OFF) ? 2 : 0;
  44
  45 static const struct bpf_map_ops * const bpf_map_types[] = {
  46 #define BPF_PROG_TYPE(_id, _ops)
  47 #define BPF_MAP_TYPE(_id, _ops) \
  48         [_id] = &_ops,
  49 #include <linux/bpf_types.h>
  50 #undef BPF_PROG_TYPE
  51 #undef BPF_MAP_TYPE
  52 };
  53
  54 /*
  55  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  56  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  57  * we don't know about yet.
  58  *
  59  * There is a ToCToU between this function call and the following
  60  * copy_from_user() call. However, this is not a concern since this function is
  61  * meant to be a future-proofing of bits.
  62  */
  63 int bpf_check_uarg_tail_zero(void __user *uaddr,
  64                              size_t expected_size,
  65                              size_t actual_size)
  66 {
  67         unsigned char __user *addr;
  68         unsigned char __user *end;
  69         unsigned char val;
  70         int err;
  71
  72         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  73                 return -E2BIG;
  74
  75         if (unlikely(!access_ok(uaddr, actual_size)))
  76                 return -EFAULT;
  77
  78         if (actual_size <= expected_size)
  79                 return 0;
  80
  81         addr = uaddr + expected_size;
  82         end  = uaddr + actual_size;
  83
  84         for (; addr < end; addr++) {
  85                 err = get_user(val, addr);
  86                 if (err)
  87                         return err;
  88                 if (val)
  89                         return -E2BIG;
  90         }
  91
  92         return 0;
  93 }
  94
  95 const struct bpf_map_ops bpf_map_offload_ops = {
  96         .map_alloc = bpf_map_offload_map_alloc,
  97         .map_free = bpf_map_offload_map_free,
  98         .map_check_btf = map_check_no_btf,
  99 };
 100
 101 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 102 {
 103         const struct bpf_map_ops *ops;
 104         u32 type = attr->map_type;
 105         struct bpf_map *map;
 106         int err;
 107
 108         if (type >= ARRAY_SIZE(bpf_map_types))
 109                 return ERR_PTR(-EINVAL);
 110         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 111         ops = bpf_map_types[type];
 112         if (!ops)
 113                 return ERR_PTR(-EINVAL);
 114
 115         if (ops->map_alloc_check) {
 116                 err = ops->map_alloc_check(attr);
 117                 if (err)
 118                         return ERR_PTR(err);
 119         }
 120         if (attr->map_ifindex)
 121                 ops = &bpf_map_offload_ops;
 122         map = ops->map_alloc(attr);
 123         if (IS_ERR(map))
 124                 return map;
 125         map->ops = ops;
 126         map->map_type = type;
 127         return map;
 128 }
 129
 130 void *bpf_map_area_alloc(u64 size, int numa_node)
 131 {
 132         /* We really just want to fail instead of triggering OOM killer
 133          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 134          * which is used for lower order allocation requests.
 135          *
 136          * It has been observed that higher order allocation requests done by
 137          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 138          * to reclaim memory from the page cache, thus we set
 139          * __GFP_RETRY_MAYFAIL to avoid such situations.
 140          */
 141
 142         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 143         void *area;
 144
 145         if (size >= SIZE_MAX)
 146                 return NULL;
 147
 148         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 149                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 150                                     numa_node);
 151                 if (area != NULL)
 152                         return area;
 153         }
 154
 155         return __vmalloc_node_flags_caller(size, numa_node,
 156                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 157                                            flags, __builtin_return_address(0));
 158 }
 159
 160 void bpf_map_area_free(void *area)
 161 {
 162         kvfree(area);
 163 }
 164
 165 static u32 bpf_map_flags_retain_permanent(u32 flags)
 166 {
 167         /* Some map creation flags are not tied to the map object but
 168          * rather to the map fd instead, so they have no meaning upon
 169          * map object inspection since multiple file descriptors with
 170          * different (access) properties can exist here. Thus, given
 171          * this has zero meaning for the map itself, lets clear these
 172          * from here.
 173          */
 174         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 175 }
 176
 177 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 178 {
 179         map->map_type = attr->map_type;
 180         map->key_size = attr->key_size;
 181         map->value_size = attr->value_size;
 182         map->max_entries = attr->max_entries;
 183         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 184         map->numa_node = bpf_map_attr_numa_node(attr);
 185 }
 186
 187 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 188 {
 189         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 190
 191         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 192                 atomic_long_sub(pages, &user->locked_vm);
 193                 return -EPERM;
 194         }
 195         return 0;
 196 }
 197
 198 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 199 {
 200         if (user)
 201                 atomic_long_sub(pages, &user->locked_vm);
 202 }
 203
 204 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
 205 {
 206         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 207         struct user_struct *user;
 208         int ret;
 209
 210         if (size >= U32_MAX - PAGE_SIZE)
 211                 return -E2BIG;
 212
 213         user = get_current_user();
 214         ret = bpf_charge_memlock(user, pages);
 215         if (ret) {
 216                 free_uid(user);
 217                 return ret;
 218         }
 219
 220         mem->pages = pages;
 221         mem->user = user;
 222
 223         return 0;
 224 }
 225
 226 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 227 {
 228         bpf_uncharge_memlock(mem->user, mem->pages);
 229         free_uid(mem->user);
 230 }
 231
 232 void bpf_map_charge_move(struct bpf_map_memory *dst,
 233                          struct bpf_map_memory *src)
 234 {
 235         *dst = *src;
 236
 237         /* Make sure src will not be used for the redundant uncharging. */
 238         memset(src, 0, sizeof(struct bpf_map_memory));
 239 }
 240
 241 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 242 {
 243         int ret;
 244
 245         ret = bpf_charge_memlock(map->memory.user, pages);
 246         if (ret)
 247                 return ret;
 248         map->memory.pages += pages;
 249         return ret;
 250 }
 251
 252 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 253 {
 254         bpf_uncharge_memlock(map->memory.user, pages);
 255         map->memory.pages -= pages;
 256 }
 257
 258 static int bpf_map_alloc_id(struct bpf_map *map)
 259 {
 260         int id;
 261
 262         idr_preload(GFP_KERNEL);
 263         spin_lock_bh(&map_idr_lock);
 264         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 265         if (id > 0)
 266                 map->id = id;
 267         spin_unlock_bh(&map_idr_lock);
 268         idr_preload_end();
 269
 270         if (WARN_ON_ONCE(!id))
 271                 return -ENOSPC;
 272
 273         return id > 0 ? 0 : id;
 274 }
 275
 276 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 277 {
 278         unsigned long flags;
 279
 280         /* Offloaded maps are removed from the IDR store when their device
 281          * disappears - even if someone holds an fd to them they are unusable,
 282          * the memory is gone, all ops will fail; they are simply waiting for
 283          * refcnt to drop to be freed.
 284          */
 285         if (!map->id)
 286                 return;
 287
 288         if (do_idr_lock)
 289                 spin_lock_irqsave(&map_idr_lock, flags);
 290         else
 291                 __acquire(&map_idr_lock);
 292
 293         idr_remove(&map_idr, map->id);
 294         map->id = 0;
 295
 296         if (do_idr_lock)
 297                 spin_unlock_irqrestore(&map_idr_lock, flags);
 298         else
 299                 __release(&map_idr_lock);
 300 }
 301
 302 /* called from workqueue */
 303 static void bpf_map_free_deferred(struct work_struct *work)
 304 {
 305         struct bpf_map *map = container_of(work, struct bpf_map, work);
 306         struct bpf_map_memory mem;
 307
 308         bpf_map_charge_move(&mem, &map->memory);
 309         security_bpf_map_free(map);
 310         /* implementation dependent freeing */
 311         map->ops->map_free(map);
 312         bpf_map_charge_finish(&mem);
 313 }
 314
 315 static void bpf_map_put_uref(struct bpf_map *map)
 316 {
 317         if (atomic_dec_and_test(&map->usercnt)) {
 318                 if (map->ops->map_release_uref)
 319                         map->ops->map_release_uref(map);
 320         }
 321 }
 322
 323 /* decrement map refcnt and schedule it for freeing via workqueue
 324  * (unrelying map implementation ops->map_free() might sleep)
 325  */
 326 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 327 {
 328         if (atomic_dec_and_test(&map->refcnt)) {
 329                 /* bpf_map_free_id() must be called first */
 330                 bpf_map_free_id(map, do_idr_lock);
 331                 btf_put(map->btf);
 332                 INIT_WORK(&map->work, bpf_map_free_deferred);
 333                 schedule_work(&map->work);
 334         }
 335 }
 336
 337 void bpf_map_put(struct bpf_map *map)
 338 {
 339         __bpf_map_put(map, true);
 340 }
 341 EXPORT_SYMBOL_GPL(bpf_map_put);
 342
 343 void bpf_map_put_with_uref(struct bpf_map *map)
 344 {
 345         bpf_map_put_uref(map);
 346         bpf_map_put(map);
 347 }
 348
 349 static int bpf_map_release(struct inode *inode, struct file *filp)
 350 {
 351         struct bpf_map *map = filp->private_data;
 352
 353         if (map->ops->map_release)
 354                 map->ops->map_release(map, filp);
 355
 356         bpf_map_put_with_uref(map);
 357         return 0;
 358 }
 359
 360 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 361 {
 362         fmode_t mode = f.file->f_mode;
 363
 364         /* Our file permissions may have been overridden by global
 365          * map permissions facing syscall side.
 366          */
 367         if (READ_ONCE(map->frozen))
 368                 mode &= ~FMODE_CAN_WRITE;
 369         return mode;
 370 }
 371
 372 #ifdef CONFIG_PROC_FS
 373 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 374 {
 375         const struct bpf_map *map = filp->private_data;
 376         const struct bpf_array *array;
 377         u32 owner_prog_type = 0;
 378         u32 owner_jited = 0;
 379
 380         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 381                 array = container_of(map, struct bpf_array, map);
 382                 owner_prog_type = array->owner_prog_type;
 383                 owner_jited = array->owner_jited;
 384         }
 385
 386         seq_printf(m,
 387                    "map_type:\t%u\n"
 388                    "key_size:\t%u\n"
 389                    "value_size:\t%u\n"
 390                    "max_entries:\t%u\n"
 391                    "map_flags:\t%#x\n"
 392                    "memlock:\t%llu\n"
 393                    "map_id:\t%u\n"
 394                    "frozen:\t%u\n",
 395                    map->map_type,
 396                    map->key_size,
 397                    map->value_size,
 398                    map->max_entries,
 399                    map->map_flags,
 400                    map->memory.pages * 1ULL << PAGE_SHIFT,
 401                    map->id,
 402                    READ_ONCE(map->frozen));
 403
 404         if (owner_prog_type) {
 405                 seq_printf(m, "owner_prog_type:\t%u\n",
 406                            owner_prog_type);
 407                 seq_printf(m, "owner_jited:\t%u\n",
 408                            owner_jited);
 409         }
 410 }
 411 #endif
 412
 413 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 414                               loff_t *ppos)
 415 {
 416         /* We need this handler such that alloc_file() enables
 417          * f_mode with FMODE_CAN_READ.
 418          */
 419         return -EINVAL;
 420 }
 421
 422 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 423                                size_t siz, loff_t *ppos)
 424 {
 425         /* We need this handler such that alloc_file() enables
 426          * f_mode with FMODE_CAN_WRITE.
 427          */
 428         return -EINVAL;
 429 }
 430
 431 const struct file_operations bpf_map_fops = {
 432 #ifdef CONFIG_PROC_FS
 433         .show_fdinfo    = bpf_map_show_fdinfo,
 434 #endif
 435         .release        = bpf_map_release,
 436         .read           = bpf_dummy_read,
 437         .write          = bpf_dummy_write,
 438 };
 439
 440 int bpf_map_new_fd(struct bpf_map *map, int flags)
 441 {
 442         int ret;
 443
 444         ret = security_bpf_map(map, OPEN_FMODE(flags));
 445         if (ret < 0)
 446                 return ret;
 447
 448         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 449                                 flags | O_CLOEXEC);
 450 }
 451
 452 int bpf_get_file_flag(int flags)
 453 {
 454         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 455                 return -EINVAL;
 456         if (flags & BPF_F_RDONLY)
 457                 return O_RDONLY;
 458         if (flags & BPF_F_WRONLY)
 459                 return O_WRONLY;
 460         return O_RDWR;
 461 }
 462
 463 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 464 #define CHECK_ATTR(CMD) \
 465         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 466                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 467                    sizeof(*attr) - \
 468                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 469                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 470
 471 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 472  * Return 0 on success and < 0 on error.
 473  */
 474 static int bpf_obj_name_cpy(char *dst, const char *src)
 475 {
 476         const char *end = src + BPF_OBJ_NAME_LEN;
 477
 478         memset(dst, 0, BPF_OBJ_NAME_LEN);
 479         /* Copy all isalnum(), '_' and '.' chars. */
 480         while (src < end && *src) {
 481                 if (!isalnum(*src) &&
 482                     *src != '_' && *src != '.')
 483                         return -EINVAL;
 484                 *dst++ = *src++;
 485         }
 486
 487         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 488         if (src == end)
 489                 return -EINVAL;
 490
 491         return 0;
 492 }
 493
 494 int map_check_no_btf(const struct bpf_map *map,
 495                      const struct btf *btf,
 496                      const struct btf_type *key_type,
 497                      const struct btf_type *value_type)
 498 {
 499         return -ENOTSUPP;
 500 }
 501
 502 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 503                          u32 btf_key_id, u32 btf_value_id)
 504 {
 505         const struct btf_type *key_type, *value_type;
 506         u32 key_size, value_size;
 507         int ret = 0;
 508
 509         /* Some maps allow key to be unspecified. */
 510         if (btf_key_id) {
 511                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 512                 if (!key_type || key_size != map->key_size)
 513                         return -EINVAL;
 514         } else {
 515                 key_type = btf_type_by_id(btf, 0);
 516                 if (!map->ops->map_check_btf)
 517                         return -EINVAL;
 518         }
 519
 520         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 521         if (!value_type || value_size != map->value_size)
 522                 return -EINVAL;
 523
 524         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 525
 526         if (map_value_has_spin_lock(map)) {
 527                 if (map->map_flags & BPF_F_RDONLY_PROG)
 528                         return -EACCES;
 529                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 530                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 531                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 532                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 533                         return -ENOTSUPP;
 534                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 535                     map->value_size) {
 536                         WARN_ONCE(1,
 537                                   "verifier bug spin_lock_off %d value_size %d\n",
 538                                   map->spin_lock_off, map->value_size);
 539                         return -EFAULT;
 540                 }
 541         }
 542
 543         if (map->ops->map_check_btf)
 544                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 545
 546         return ret;
 547 }
 548
 549 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 550 /* called via syscall */
 551 static int map_create(union bpf_attr *attr)
 552 {
 553         int numa_node = bpf_map_attr_numa_node(attr);
 554         struct bpf_map_memory mem;
 555         struct bpf_map *map;
 556         int f_flags;
 557         int err;
 558
 559         err = CHECK_ATTR(BPF_MAP_CREATE);
 560         if (err)
 561                 return -EINVAL;
 562
 563         f_flags = bpf_get_file_flag(attr->map_flags);
 564         if (f_flags < 0)
 565                 return f_flags;
 566
 567         if (numa_node != NUMA_NO_NODE &&
 568             ((unsigned int)numa_node >= nr_node_ids ||
 569              !node_online(numa_node)))
 570                 return -EINVAL;
 571
 572         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 573         map = find_and_alloc_map(attr);
 574         if (IS_ERR(map))
 575                 return PTR_ERR(map);
 576
 577         err = bpf_obj_name_cpy(map->name, attr->map_name);
 578         if (err)
 579                 goto free_map;
 580
 581         atomic_set(&map->refcnt, 1);
 582         atomic_set(&map->usercnt, 1);
 583
 584         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 585                 struct btf *btf;
 586
 587                 if (!attr->btf_value_type_id) {
 588                         err = -EINVAL;
 589                         goto free_map;
 590                 }
 591
 592                 btf = btf_get_by_fd(attr->btf_fd);
 593                 if (IS_ERR(btf)) {
 594                         err = PTR_ERR(btf);
 595                         goto free_map;
 596                 }
 597
 598                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 599                                     attr->btf_value_type_id);
 600                 if (err) {
 601                         btf_put(btf);
 602                         goto free_map;
 603                 }
 604
 605                 map->btf = btf;
 606                 map->btf_key_type_id = attr->btf_key_type_id;
 607                 map->btf_value_type_id = attr->btf_value_type_id;
 608         } else {
 609                 map->spin_lock_off = -EINVAL;
 610         }
 611
 612         err = security_bpf_map_alloc(map);
 613         if (err)
 614                 goto free_map;
 615
 616         err = bpf_map_alloc_id(map);
 617         if (err)
 618                 goto free_map_sec;
 619
 620         err = bpf_map_new_fd(map, f_flags);
 621         if (err < 0) {
 622                 /* failed to allocate fd.
 623                  * bpf_map_put_with_uref() is needed because the above
 624                  * bpf_map_alloc_id() has published the map
 625                  * to the userspace and the userspace may
 626                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 627                  */
 628                 bpf_map_put_with_uref(map);
 629                 return err;
 630         }
 631
 632         return err;
 633
 634 free_map_sec:
 635         security_bpf_map_free(map);
 636 free_map:
 637         btf_put(map->btf);
 638         bpf_map_charge_move(&mem, &map->memory);
 639         map->ops->map_free(map);
 640         bpf_map_charge_finish(&mem);
 641         return err;
 642 }
 643
 644 /* if error is returned, fd is released.
 645  * On success caller should complete fd access with matching fdput()
 646  */
 647 struct bpf_map *__bpf_map_get(struct fd f)
 648 {
 649         if (!f.file)
 650                 return ERR_PTR(-EBADF);
 651         if (f.file->f_op != &bpf_map_fops) {
 652                 fdput(f);
 653                 return ERR_PTR(-EINVAL);
 654         }
 655
 656         return f.file->private_data;
 657 }
 658
 659 /* prog's and map's refcnt limit */
 660 #define BPF_MAX_REFCNT 32768
 661
 662 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 663 {
 664         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 665                 atomic_dec(&map->refcnt);
 666                 return ERR_PTR(-EBUSY);
 667         }
 668         if (uref)
 669                 atomic_inc(&map->usercnt);
 670         return map;
 671 }
 672 EXPORT_SYMBOL_GPL(bpf_map_inc);
 673
 674 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 675 {
 676         struct fd f = fdget(ufd);
 677         struct bpf_map *map;
 678
 679         map = __bpf_map_get(f);
 680         if (IS_ERR(map))
 681                 return map;
 682
 683         map = bpf_map_inc(map, true);
 684         fdput(f);
 685
 686         return map;
 687 }
 688
 689 /* map_idr_lock should have been held */
 690 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map,
 691                                               bool uref)
 692 {
 693         int refold;
 694
 695         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 696
 697         if (refold >= BPF_MAX_REFCNT) {
 698                 __bpf_map_put(map, false);
 699                 return ERR_PTR(-EBUSY);
 700         }
 701
 702         if (!refold)
 703                 return ERR_PTR(-ENOENT);
 704
 705         if (uref)
 706                 atomic_inc(&map->usercnt);
 707
 708         return map;
 709 }
 710
 711 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 712 {
 713         spin_lock_bh(&map_idr_lock);
 714         map = __bpf_map_inc_not_zero(map, uref);
 715         spin_unlock_bh(&map_idr_lock);
 716
 717         return map;
 718 }
 719 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 720
 721 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 722 {
 723         return -ENOTSUPP;
 724 }
 725
 726 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 727 {
 728         if (key_size)
 729                 return memdup_user(ukey, key_size);
 730
 731         if (ukey)
 732                 return ERR_PTR(-EINVAL);
 733
 734         return NULL;
 735 }
 736
 737 /* last field in 'union bpf_attr' used by this command */
 738 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 739
 740 static int map_lookup_elem(union bpf_attr *attr)
 741 {
 742         void __user *ukey = u64_to_user_ptr(attr->key);
 743         void __user *uvalue = u64_to_user_ptr(attr->value);
 744         int ufd = attr->map_fd;
 745         struct bpf_map *map;
 746         void *key, *value, *ptr;
 747         u32 value_size;
 748         struct fd f;
 749         int err;
 750
 751         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 752                 return -EINVAL;
 753
 754         if (attr->flags & ~BPF_F_LOCK)
 755                 return -EINVAL;
 756
 757         f = fdget(ufd);
 758         map = __bpf_map_get(f);
 759         if (IS_ERR(map))
 760                 return PTR_ERR(map);
 761         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 762                 err = -EPERM;
 763                 goto err_put;
 764         }
 765
 766         if ((attr->flags & BPF_F_LOCK) &&
 767             !map_value_has_spin_lock(map)) {
 768                 err = -EINVAL;
 769                 goto err_put;
 770         }
 771
 772         key = __bpf_copy_key(ukey, map->key_size);
 773         if (IS_ERR(key)) {
 774                 err = PTR_ERR(key);
 775                 goto err_put;
 776         }
 777
 778         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 779             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 780             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 781             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 782                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 783         else if (IS_FD_MAP(map))
 784                 value_size = sizeof(u32);
 785         else
 786                 value_size = map->value_size;
 787
 788         err = -ENOMEM;
 789         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 790         if (!value)
 791                 goto free_key;
 792
 793         if (bpf_map_is_dev_bound(map)) {
 794                 err = bpf_map_offload_lookup_elem(map, key, value);
 795                 goto done;
 796         }
 797
 798         preempt_disable();
 799         this_cpu_inc(bpf_prog_active);
 800         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 801             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 802                 err = bpf_percpu_hash_copy(map, key, value);
 803         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 804                 err = bpf_percpu_array_copy(map, key, value);
 805         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 806                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 807         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 808                 err = bpf_stackmap_copy(map, key, value);
 809         } else if (IS_FD_ARRAY(map)) {
 810                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 811         } else if (IS_FD_HASH(map)) {
 812                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 813         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 814                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 815         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 816                    map->map_type == BPF_MAP_TYPE_STACK) {
 817                 err = map->ops->map_peek_elem(map, value);
 818         } else {
 819                 rcu_read_lock();
 820                 if (map->ops->map_lookup_elem_sys_only)
 821                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 822                 else
 823                         ptr = map->ops->map_lookup_elem(map, key);
 824                 if (IS_ERR(ptr)) {
 825                         err = PTR_ERR(ptr);
 826                 } else if (!ptr) {
 827                         err = -ENOENT;
 828                 } else {
 829                         err = 0;
 830                         if (attr->flags & BPF_F_LOCK)
 831                                 /* lock 'ptr' and copy everything but lock */
 832                                 copy_map_value_locked(map, value, ptr, true);
 833                         else
 834                                 copy_map_value(map, value, ptr);
 835                         /* mask lock, since value wasn't zero inited */
 836                         check_and_init_map_lock(map, value);
 837                 }
 838                 rcu_read_unlock();
 839         }
 840         this_cpu_dec(bpf_prog_active);
 841         preempt_enable();
 842
 843 done:
 844         if (err)
 845                 goto free_value;
 846
 847         err = -EFAULT;
 848         if (copy_to_user(uvalue, value, value_size) != 0)
 849                 goto free_value;
 850
 851         err = 0;
 852
 853 free_value:
 854         kfree(value);
 855 free_key:
 856         kfree(key);
 857 err_put:
 858         fdput(f);
 859         return err;
 860 }
 861
 862 static void maybe_wait_bpf_programs(struct bpf_map *map)
 863 {
 864         /* Wait for any running BPF programs to complete so that
 865          * userspace, when we return to it, knows that all programs
 866          * that could be running use the new map value.
 867          */
 868         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 869             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 870                 synchronize_rcu();
 871 }
 872
 873 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 874
 875 static int map_update_elem(union bpf_attr *attr)
 876 {
 877         void __user *ukey = u64_to_user_ptr(attr->key);
 878         void __user *uvalue = u64_to_user_ptr(attr->value);
 879         int ufd = attr->map_fd;
 880         struct bpf_map *map;
 881         void *key, *value;
 882         u32 value_size;
 883         struct fd f;
 884         int err;
 885
 886         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 887                 return -EINVAL;
 888
 889         f = fdget(ufd);
 890         map = __bpf_map_get(f);
 891         if (IS_ERR(map))
 892                 return PTR_ERR(map);
 893         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 894                 err = -EPERM;
 895                 goto err_put;
 896         }
 897
 898         if ((attr->flags & BPF_F_LOCK) &&
 899             !map_value_has_spin_lock(map)) {
 900                 err = -EINVAL;
 901                 goto err_put;
 902         }
 903
 904         key = __bpf_copy_key(ukey, map->key_size);
 905         if (IS_ERR(key)) {
 906                 err = PTR_ERR(key);
 907                 goto err_put;
 908         }
 909
 910         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 911             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 912             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 913             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 914                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 915         else
 916                 value_size = map->value_size;
 917
 918         err = -ENOMEM;
 919         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 920         if (!value)
 921                 goto free_key;
 922
 923         err = -EFAULT;
 924         if (copy_from_user(value, uvalue, value_size) != 0)
 925                 goto free_value;
 926
 927         /* Need to create a kthread, thus must support schedule */
 928         if (bpf_map_is_dev_bound(map)) {
 929                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 930                 goto out;
 931         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 932                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 933                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 934                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 935                 goto out;
 936         }
 937
 938         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 939          * inside bpf map update or delete otherwise deadlocks are possible
 940          */
 941         preempt_disable();
 942         __this_cpu_inc(bpf_prog_active);
 943         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 944             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 945                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 946         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 947                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 948         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 949                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 950                                                        attr->flags);
 951         } else if (IS_FD_ARRAY(map)) {
 952                 rcu_read_lock();
 953                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 954                                                    attr->flags);
 955                 rcu_read_unlock();
 956         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 957                 rcu_read_lock();
 958                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 959                                                   attr->flags);
 960                 rcu_read_unlock();
 961         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 962                 /* rcu_read_lock() is not needed */
 963                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 964                                                          attr->flags);
 965         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 966                    map->map_type == BPF_MAP_TYPE_STACK) {
 967                 err = map->ops->map_push_elem(map, value, attr->flags);
 968         } else {
 969                 rcu_read_lock();
 970                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 971                 rcu_read_unlock();
 972         }
 973         __this_cpu_dec(bpf_prog_active);
 974         preempt_enable();
 975         maybe_wait_bpf_programs(map);
 976 out:
 977 free_value:
 978         kfree(value);
 979 free_key:
 980         kfree(key);
 981 err_put:
 982         fdput(f);
 983         return err;
 984 }
 985
 986 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 987
 988 static int map_delete_elem(union bpf_attr *attr)
 989 {
 990         void __user *ukey = u64_to_user_ptr(attr->key);
 991         int ufd = attr->map_fd;
 992         struct bpf_map *map;
 993         struct fd f;
 994         void *key;
 995         int err;
 996
 997         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 998                 return -EINVAL;
 999
1000         f = fdget(ufd);
1001         map = __bpf_map_get(f);
1002         if (IS_ERR(map))
1003                 return PTR_ERR(map);
1004         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1005                 err = -EPERM;
1006                 goto err_put;
1007         }
1008
1009         key = __bpf_copy_key(ukey, map->key_size);
1010         if (IS_ERR(key)) {
1011                 err = PTR_ERR(key);
1012                 goto err_put;
1013         }
1014
1015         if (bpf_map_is_dev_bound(map)) {
1016                 err = bpf_map_offload_delete_elem(map, key);
1017                 goto out;
1018         }
1019
1020         preempt_disable();
1021         __this_cpu_inc(bpf_prog_active);
1022         rcu_read_lock();
1023         err = map->ops->map_delete_elem(map, key);
1024         rcu_read_unlock();
1025         __this_cpu_dec(bpf_prog_active);
1026         preempt_enable();
1027         maybe_wait_bpf_programs(map);
1028 out:
1029         kfree(key);
1030 err_put:
1031         fdput(f);
1032         return err;
1033 }
1034
1035 /* last field in 'union bpf_attr' used by this command */
1036 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1037
1038 static int map_get_next_key(union bpf_attr *attr)
1039 {
1040         void __user *ukey = u64_to_user_ptr(attr->key);
1041         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1042         int ufd = attr->map_fd;
1043         struct bpf_map *map;
1044         void *key, *next_key;
1045         struct fd f;
1046         int err;
1047
1048         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1049                 return -EINVAL;
1050
1051         f = fdget(ufd);
1052         map = __bpf_map_get(f);
1053         if (IS_ERR(map))
1054                 return PTR_ERR(map);
1055         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1056                 err = -EPERM;
1057                 goto err_put;
1058         }
1059
1060         if (ukey) {
1061                 key = __bpf_copy_key(ukey, map->key_size);
1062                 if (IS_ERR(key)) {
1063                         err = PTR_ERR(key);
1064                         goto err_put;
1065                 }
1066         } else {
1067                 key = NULL;
1068         }
1069
1070         err = -ENOMEM;
1071         next_key = kmalloc(map->key_size, GFP_USER);
1072         if (!next_key)
1073                 goto free_key;
1074
1075         if (bpf_map_is_dev_bound(map)) {
1076                 err = bpf_map_offload_get_next_key(map, key, next_key);
1077                 goto out;
1078         }
1079
1080         rcu_read_lock();
1081         err = map->ops->map_get_next_key(map, key, next_key);
1082         rcu_read_unlock();
1083 out:
1084         if (err)
1085                 goto free_next_key;
1086
1087         err = -EFAULT;
1088         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1089                 goto free_next_key;
1090
1091         err = 0;
1092
1093 free_next_key:
1094         kfree(next_key);
1095 free_key:
1096         kfree(key);
1097 err_put:
1098         fdput(f);
1099         return err;
1100 }
1101
1102 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1103
1104 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1105 {
1106         void __user *ukey = u64_to_user_ptr(attr->key);
1107         void __user *uvalue = u64_to_user_ptr(attr->value);
1108         int ufd = attr->map_fd;
1109         struct bpf_map *map;
1110         void *key, *value;
1111         u32 value_size;
1112         struct fd f;
1113         int err;
1114
1115         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1116                 return -EINVAL;
1117
1118         f = fdget(ufd);
1119         map = __bpf_map_get(f);
1120         if (IS_ERR(map))
1121                 return PTR_ERR(map);
1122         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1123             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1124                 err = -EPERM;
1125                 goto err_put;
1126         }
1127
1128         key = __bpf_copy_key(ukey, map->key_size);
1129         if (IS_ERR(key)) {
1130                 err = PTR_ERR(key);
1131                 goto err_put;
1132         }
1133
1134         value_size = map->value_size;
1135
1136         err = -ENOMEM;
1137         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1138         if (!value)
1139                 goto free_key;
1140
1141         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1142             map->map_type == BPF_MAP_TYPE_STACK) {
1143                 err = map->ops->map_pop_elem(map, value);
1144         } else {
1145                 err = -ENOTSUPP;
1146         }
1147
1148         if (err)
1149                 goto free_value;
1150
1151         if (copy_to_user(uvalue, value, value_size) != 0) {
1152                 err = -EFAULT;
1153                 goto free_value;
1154         }
1155
1156         err = 0;
1157
1158 free_value:
1159         kfree(value);
1160 free_key:
1161         kfree(key);
1162 err_put:
1163         fdput(f);
1164         return err;
1165 }
1166
1167 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1168
1169 static int map_freeze(const union bpf_attr *attr)
1170 {
1171         int err = 0, ufd = attr->map_fd;
1172         struct bpf_map *map;
1173         struct fd f;
1174
1175         if (CHECK_ATTR(BPF_MAP_FREEZE))
1176                 return -EINVAL;
1177
1178         f = fdget(ufd);
1179         map = __bpf_map_get(f);
1180         if (IS_ERR(map))
1181                 return PTR_ERR(map);
1182         if (READ_ONCE(map->frozen)) {
1183                 err = -EBUSY;
1184                 goto err_put;
1185         }
1186         if (!capable(CAP_SYS_ADMIN)) {
1187                 err = -EPERM;
1188                 goto err_put;
1189         }
1190
1191         WRITE_ONCE(map->frozen, true);
1192 err_put:
1193         fdput(f);
1194         return err;
1195 }
1196
1197 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1198 #define BPF_PROG_TYPE(_id, _name) \
1199         [_id] = & _name ## _prog_ops,
1200 #define BPF_MAP_TYPE(_id, _ops)
1201 #include <linux/bpf_types.h>
1202 #undef BPF_PROG_TYPE
1203 #undef BPF_MAP_TYPE
1204 };
1205
1206 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1207 {
1208         const struct bpf_prog_ops *ops;
1209
1210         if (type >= ARRAY_SIZE(bpf_prog_types))
1211                 return -EINVAL;
1212         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1213         ops = bpf_prog_types[type];
1214         if (!ops)
1215                 return -EINVAL;
1216
1217         if (!bpf_prog_is_dev_bound(prog->aux))
1218                 prog->aux->ops = ops;
1219         else
1220                 prog->aux->ops = &bpf_offload_prog_ops;
1221         prog->type = type;
1222         return 0;
1223 }
1224
1225 /* drop refcnt on maps used by eBPF program and free auxilary data */
1226 static void free_used_maps(struct bpf_prog_aux *aux)
1227 {
1228         enum bpf_cgroup_storage_type stype;
1229         int i;
1230
1231         for_each_cgroup_storage_type(stype) {
1232                 if (!aux->cgroup_storage[stype])
1233                         continue;
1234                 bpf_cgroup_storage_release(aux->prog,
1235                                            aux->cgroup_storage[stype]);
1236         }
1237
1238         for (i = 0; i < aux->used_map_cnt; i++)
1239                 bpf_map_put(aux->used_maps[i]);
1240
1241         kfree(aux->used_maps);
1242 }
1243
1244 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1245 {
1246         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1247         unsigned long user_bufs;
1248
1249         if (user) {
1250                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1251                 if (user_bufs > memlock_limit) {
1252                         atomic_long_sub(pages, &user->locked_vm);
1253                         return -EPERM;
1254                 }
1255         }
1256
1257         return 0;
1258 }
1259
1260 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1261 {
1262         if (user)
1263                 atomic_long_sub(pages, &user->locked_vm);
1264 }
1265
1266 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1267 {
1268         struct user_struct *user = get_current_user();
1269         int ret;
1270
1271         ret = __bpf_prog_charge(user, prog->pages);
1272         if (ret) {
1273                 free_uid(user);
1274                 return ret;
1275         }
1276
1277         prog->aux->user = user;
1278         return 0;
1279 }
1280
1281 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1282 {
1283         struct user_struct *user = prog->aux->user;
1284
1285         __bpf_prog_uncharge(user, prog->pages);
1286         free_uid(user);
1287 }
1288
1289 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1290 {
1291         int id;
1292
1293         idr_preload(GFP_KERNEL);
1294         spin_lock_bh(&prog_idr_lock);
1295         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1296         if (id > 0)
1297                 prog->aux->id = id;
1298         spin_unlock_bh(&prog_idr_lock);
1299         idr_preload_end();
1300
1301         /* id is in [1, INT_MAX) */
1302         if (WARN_ON_ONCE(!id))
1303                 return -ENOSPC;
1304
1305         return id > 0 ? 0 : id;
1306 }
1307
1308 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1309 {
1310         /* cBPF to eBPF migrations are currently not in the idr store.
1311          * Offloaded programs are removed from the store when their device
1312          * disappears - even if someone grabs an fd to them they are unusable,
1313          * simply waiting for refcnt to drop to be freed.
1314          */
1315         if (!prog->aux->id)
1316                 return;
1317
1318         if (do_idr_lock)
1319                 spin_lock_bh(&prog_idr_lock);
1320         else
1321                 __acquire(&prog_idr_lock);
1322
1323         idr_remove(&prog_idr, prog->aux->id);
1324         prog->aux->id = 0;
1325
1326         if (do_idr_lock)
1327                 spin_unlock_bh(&prog_idr_lock);
1328         else
1329                 __release(&prog_idr_lock);
1330 }
1331
1332 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1333 {
1334         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1335
1336         kvfree(aux->func_info);
1337         free_used_maps(aux);
1338         bpf_prog_uncharge_memlock(aux->prog);
1339         security_bpf_prog_free(aux);
1340         bpf_prog_free(aux->prog);
1341 }
1342
1343 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1344 {
1345         bpf_prog_kallsyms_del_all(prog);
1346         btf_put(prog->aux->btf);
1347         bpf_prog_free_linfo(prog);
1348
1349         if (deferred)
1350                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1351         else
1352                 __bpf_prog_put_rcu(&prog->aux->rcu);
1353 }
1354
1355 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1356 {
1357         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1358                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1359                 /* bpf_prog_free_id() must be called first */
1360                 bpf_prog_free_id(prog, do_idr_lock);
1361                 __bpf_prog_put_noref(prog, true);
1362         }
1363 }
1364
1365 void bpf_prog_put(struct bpf_prog *prog)
1366 {
1367         __bpf_prog_put(prog, true);
1368 }
1369 EXPORT_SYMBOL_GPL(bpf_prog_put);
1370
1371 static int bpf_prog_release(struct inode *inode, struct file *filp)
1372 {
1373         struct bpf_prog *prog = filp->private_data;
1374
1375         bpf_prog_put(prog);
1376         return 0;
1377 }
1378
1379 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1380                                struct bpf_prog_stats *stats)
1381 {
1382         u64 nsecs = 0, cnt = 0;
1383         int cpu;
1384
1385         for_each_possible_cpu(cpu) {
1386                 const struct bpf_prog_stats *st;
1387                 unsigned int start;
1388                 u64 tnsecs, tcnt;
1389
1390                 st = per_cpu_ptr(prog->aux->stats, cpu);
1391                 do {
1392                         start = u64_stats_fetch_begin_irq(&st->syncp);
1393                         tnsecs = st->nsecs;
1394                         tcnt = st->cnt;
1395                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1396                 nsecs += tnsecs;
1397                 cnt += tcnt;
1398         }
1399         stats->nsecs = nsecs;
1400         stats->cnt = cnt;
1401 }
1402
1403 #ifdef CONFIG_PROC_FS
1404 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1405 {
1406         const struct bpf_prog *prog = filp->private_data;
1407         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1408         struct bpf_prog_stats stats;
1409
1410         bpf_prog_get_stats(prog, &stats);
1411         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1412         seq_printf(m,
1413                    "prog_type:\t%u\n"
1414                    "prog_jited:\t%u\n"
1415                    "prog_tag:\t%s\n"
1416                    "memlock:\t%llu\n"
1417                    "prog_id:\t%u\n"
1418                    "run_time_ns:\t%llu\n"
1419                    "run_cnt:\t%llu\n",
1420                    prog->type,
1421                    prog->jited,
1422                    prog_tag,
1423                    prog->pages * 1ULL << PAGE_SHIFT,
1424                    prog->aux->id,
1425                    stats.nsecs,
1426                    stats.cnt);
1427 }
1428 #endif
1429
1430 const struct file_operations bpf_prog_fops = {
1431 #ifdef CONFIG_PROC_FS
1432         .show_fdinfo    = bpf_prog_show_fdinfo,
1433 #endif
1434         .release        = bpf_prog_release,
1435         .read           = bpf_dummy_read,
1436         .write          = bpf_dummy_write,
1437 };
1438
1439 int bpf_prog_new_fd(struct bpf_prog *prog)
1440 {
1441         int ret;
1442
1443         ret = security_bpf_prog(prog);
1444         if (ret < 0)
1445                 return ret;
1446
1447         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1448                                 O_RDWR | O_CLOEXEC);
1449 }
1450
1451 static struct bpf_prog *____bpf_prog_get(struct fd f)
1452 {
1453         if (!f.file)
1454                 return ERR_PTR(-EBADF);
1455         if (f.file->f_op != &bpf_prog_fops) {
1456                 fdput(f);
1457                 return ERR_PTR(-EINVAL);
1458         }
1459
1460         return f.file->private_data;
1461 }
1462
1463 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1464 {
1465         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1466                 atomic_sub(i, &prog->aux->refcnt);
1467                 return ERR_PTR(-EBUSY);
1468         }
1469         return prog;
1470 }
1471 EXPORT_SYMBOL_GPL(bpf_prog_add);
1472
1473 void bpf_prog_sub(struct bpf_prog *prog, int i)
1474 {
1475         /* Only to be used for undoing previous bpf_prog_add() in some
1476          * error path. We still know that another entity in our call
1477          * path holds a reference to the program, thus atomic_sub() can
1478          * be safely used in such cases!
1479          */
1480         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1481 }
1482 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1483
1484 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1485 {
1486         return bpf_prog_add(prog, 1);
1487 }
1488 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1489
1490 /* prog_idr_lock should have been held */
1491 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1492 {
1493         int refold;
1494
1495         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1496
1497         if (refold >= BPF_MAX_REFCNT) {
1498                 __bpf_prog_put(prog, false);
1499                 return ERR_PTR(-EBUSY);
1500         }
1501
1502         if (!refold)
1503                 return ERR_PTR(-ENOENT);
1504
1505         return prog;
1506 }
1507 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1508
1509 bool bpf_prog_get_ok(struct bpf_prog *prog,
1510                             enum bpf_prog_type *attach_type, bool attach_drv)
1511 {
1512         /* not an attachment, just a refcount inc, always allow */
1513         if (!attach_type)
1514                 return true;
1515
1516         if (prog->type != *attach_type)
1517                 return false;
1518         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1519                 return false;
1520
1521         return true;
1522 }
1523
1524 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1525                                        bool attach_drv)
1526 {
1527         struct fd f = fdget(ufd);
1528         struct bpf_prog *prog;
1529
1530         prog = ____bpf_prog_get(f);
1531         if (IS_ERR(prog))
1532                 return prog;
1533         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1534                 prog = ERR_PTR(-EINVAL);
1535                 goto out;
1536         }
1537
1538         prog = bpf_prog_inc(prog);
1539 out:
1540         fdput(f);
1541         return prog;
1542 }
1543
1544 struct bpf_prog *bpf_prog_get(u32 ufd)
1545 {
1546         return __bpf_prog_get(ufd, NULL, false);
1547 }
1548
1549 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1550                                        bool attach_drv)
1551 {
1552         return __bpf_prog_get(ufd, &type, attach_drv);
1553 }
1554 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1555
1556 /* Initially all BPF programs could be loaded w/o specifying
1557  * expected_attach_type. Later for some of them specifying expected_attach_type
1558  * at load time became required so that program could be validated properly.
1559  * Programs of types that are allowed to be loaded both w/ and w/o (for
1560  * backward compatibility) expected_attach_type, should have the default attach
1561  * type assigned to expected_attach_type for the latter case, so that it can be
1562  * validated later at attach time.
1563  *
1564  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1565  * prog type requires it but has some attach types that have to be backward
1566  * compatible.
1567  */
1568 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1569 {
1570         switch (attr->prog_type) {
1571         case BPF_PROG_TYPE_CGROUP_SOCK:
1572                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1573                  * exist so checking for non-zero is the way to go here.
1574                  */
1575                 if (!attr->expected_attach_type)
1576                         attr->expected_attach_type =
1577                                 BPF_CGROUP_INET_SOCK_CREATE;
1578                 break;
1579         }
1580 }
1581
1582 static int
1583 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1584                                 enum bpf_attach_type expected_attach_type)
1585 {
1586         switch (prog_type) {
1587         case BPF_PROG_TYPE_CGROUP_SOCK:
1588                 switch (expected_attach_type) {
1589                 case BPF_CGROUP_INET_SOCK_CREATE:
1590                 case BPF_CGROUP_INET4_POST_BIND:
1591                 case BPF_CGROUP_INET6_POST_BIND:
1592                         return 0;
1593                 default:
1594                         return -EINVAL;
1595                 }
1596         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1597                 switch (expected_attach_type) {
1598                 case BPF_CGROUP_INET4_BIND:
1599                 case BPF_CGROUP_INET6_BIND:
1600                 case BPF_CGROUP_INET4_CONNECT:
1601                 case BPF_CGROUP_INET6_CONNECT:
1602                 case BPF_CGROUP_UDP4_SENDMSG:
1603                 case BPF_CGROUP_UDP6_SENDMSG:
1604                 case BPF_CGROUP_UDP4_RECVMSG:
1605                 case BPF_CGROUP_UDP6_RECVMSG:
1606                         return 0;
1607                 default:
1608                         return -EINVAL;
1609                 }
1610         case BPF_PROG_TYPE_CGROUP_SKB:
1611                 switch (expected_attach_type) {
1612                 case BPF_CGROUP_INET_INGRESS:
1613                 case BPF_CGROUP_INET_EGRESS:
1614                         return 0;
1615                 default:
1616                         return -EINVAL;
1617                 }
1618         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1619                 switch (expected_attach_type) {
1620                 case BPF_CGROUP_SETSOCKOPT:
1621                 case BPF_CGROUP_GETSOCKOPT:
1622                         return 0;
1623                 default:
1624                         return -EINVAL;
1625                 }
1626         default:
1627                 return 0;
1628         }
1629 }
1630
1631 /* last field in 'union bpf_attr' used by this command */
1632 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1633
1634 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1635 {
1636         enum bpf_prog_type type = attr->prog_type;
1637         struct bpf_prog *prog;
1638         int err;
1639         char license[128];
1640         bool is_gpl;
1641
1642         if (CHECK_ATTR(BPF_PROG_LOAD))
1643                 return -EINVAL;
1644
1645         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1646                                  BPF_F_ANY_ALIGNMENT |
1647                                  BPF_F_TEST_STATE_FREQ |
1648                                  BPF_F_TEST_RND_HI32))
1649                 return -EINVAL;
1650
1651         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1652             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1653             !capable(CAP_SYS_ADMIN))
1654                 return -EPERM;
1655
1656         /* copy eBPF program license from user space */
1657         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1658                               sizeof(license) - 1) < 0)
1659                 return -EFAULT;
1660         license[sizeof(license) - 1] = 0;
1661
1662         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1663         is_gpl = license_is_gpl_compatible(license);
1664
1665         if (attr->insn_cnt == 0 ||
1666             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1667                 return -E2BIG;
1668         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1669             type != BPF_PROG_TYPE_CGROUP_SKB &&
1670             !capable(CAP_SYS_ADMIN))
1671                 return -EPERM;
1672
1673         bpf_prog_load_fixup_attach_type(attr);
1674         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1675                 return -EINVAL;
1676
1677         /* plain bpf_prog allocation */
1678         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1679         if (!prog)
1680                 return -ENOMEM;
1681
1682         prog->expected_attach_type = attr->expected_attach_type;
1683
1684         prog->aux->offload_requested = !!attr->prog_ifindex;
1685
1686         err = security_bpf_prog_alloc(prog->aux);
1687         if (err)
1688                 goto free_prog_nouncharge;
1689
1690         err = bpf_prog_charge_memlock(prog);
1691         if (err)
1692                 goto free_prog_sec;
1693
1694         prog->len = attr->insn_cnt;
1695
1696         err = -EFAULT;
1697         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1698                            bpf_prog_insn_size(prog)) != 0)
1699                 goto free_prog;
1700
1701         prog->orig_prog = NULL;
1702         prog->jited = 0;
1703
1704         atomic_set(&prog->aux->refcnt, 1);
1705         prog->gpl_compatible = is_gpl ? 1 : 0;
1706
1707         if (bpf_prog_is_dev_bound(prog->aux)) {
1708                 err = bpf_prog_offload_init(prog, attr);
1709                 if (err)
1710                         goto free_prog;
1711         }
1712
1713         /* find program type: socket_filter vs tracing_filter */
1714         err = find_prog_type(type, prog);
1715         if (err < 0)
1716                 goto free_prog;
1717
1718         prog->aux->load_time = ktime_get_boottime_ns();
1719         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1720         if (err)
1721                 goto free_prog;
1722
1723         /* run eBPF verifier */
1724         err = bpf_check(&prog, attr, uattr);
1725         if (err < 0)
1726                 goto free_used_maps;
1727
1728         prog = bpf_prog_select_runtime(prog, &err);
1729         if (err < 0)
1730                 goto free_used_maps;
1731
1732         err = bpf_prog_alloc_id(prog);
1733         if (err)
1734                 goto free_used_maps;
1735
1736         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1737          * effectively publicly exposed. However, retrieving via
1738          * bpf_prog_get_fd_by_id() will take another reference,
1739          * therefore it cannot be gone underneath us.
1740          *
1741          * Only for the time /after/ successful bpf_prog_new_fd()
1742          * and before returning to userspace, we might just hold
1743          * one reference and any parallel close on that fd could
1744          * rip everything out. Hence, below notifications must
1745          * happen before bpf_prog_new_fd().
1746          *
1747          * Also, any failure handling from this point onwards must
1748          * be using bpf_prog_put() given the program is exposed.
1749          */
1750         bpf_prog_kallsyms_add(prog);
1751         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1752
1753         err = bpf_prog_new_fd(prog);
1754         if (err < 0)
1755                 bpf_prog_put(prog);
1756         return err;
1757
1758 free_used_maps:
1759         /* In case we have subprogs, we need to wait for a grace
1760          * period before we can tear down JIT memory since symbols
1761          * are already exposed under kallsyms.
1762          */
1763         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1764         return err;
1765 free_prog:
1766         bpf_prog_uncharge_memlock(prog);
1767 free_prog_sec:
1768         security_bpf_prog_free(prog->aux);
1769 free_prog_nouncharge:
1770         bpf_prog_free(prog);
1771         return err;
1772 }
1773
1774 #define BPF_OBJ_LAST_FIELD file_flags
1775
1776 static int bpf_obj_pin(const union bpf_attr *attr)
1777 {
1778         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1779                 return -EINVAL;
1780
1781         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1782 }
1783
1784 static int bpf_obj_get(const union bpf_attr *attr)
1785 {
1786         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1787             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1788                 return -EINVAL;
1789
1790         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1791                                 attr->file_flags);
1792 }
1793
1794 struct bpf_raw_tracepoint {
1795         struct bpf_raw_event_map *btp;
1796         struct bpf_prog *prog;
1797 };
1798
1799 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1800 {
1801         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1802
1803         if (raw_tp->prog) {
1804                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1805                 bpf_prog_put(raw_tp->prog);
1806         }
1807         bpf_put_raw_tracepoint(raw_tp->btp);
1808         kfree(raw_tp);
1809         return 0;
1810 }
1811
1812 static const struct file_operations bpf_raw_tp_fops = {
1813         .release        = bpf_raw_tracepoint_release,
1814         .read           = bpf_dummy_read,
1815         .write          = bpf_dummy_write,
1816 };
1817
1818 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1819
1820 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1821 {
1822         struct bpf_raw_tracepoint *raw_tp;
1823         struct bpf_raw_event_map *btp;
1824         struct bpf_prog *prog;
1825         char tp_name[128];
1826         int tp_fd, err;
1827
1828         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1829                               sizeof(tp_name) - 1) < 0)
1830                 return -EFAULT;
1831         tp_name[sizeof(tp_name) - 1] = 0;
1832
1833         btp = bpf_get_raw_tracepoint(tp_name);
1834         if (!btp)
1835                 return -ENOENT;
1836
1837         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1838         if (!raw_tp) {
1839                 err = -ENOMEM;
1840                 goto out_put_btp;
1841         }
1842         raw_tp->btp = btp;
1843
1844         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1845         if (IS_ERR(prog)) {
1846                 err = PTR_ERR(prog);
1847                 goto out_free_tp;
1848         }
1849         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1850             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1851                 err = -EINVAL;
1852                 goto out_put_prog;
1853         }
1854
1855         err = bpf_probe_register(raw_tp->btp, prog);
1856         if (err)
1857                 goto out_put_prog;
1858
1859         raw_tp->prog = prog;
1860         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1861                                  O_CLOEXEC);
1862         if (tp_fd < 0) {
1863                 bpf_probe_unregister(raw_tp->btp, prog);
1864                 err = tp_fd;
1865                 goto out_put_prog;
1866         }
1867         return tp_fd;
1868
1869 out_put_prog:
1870         bpf_prog_put(prog);
1871 out_free_tp:
1872         kfree(raw_tp);
1873 out_put_btp:
1874         bpf_put_raw_tracepoint(btp);
1875         return err;
1876 }
1877
1878 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1879                                              enum bpf_attach_type attach_type)
1880 {
1881         switch (prog->type) {
1882         case BPF_PROG_TYPE_CGROUP_SOCK:
1883         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1884         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1885                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1886         case BPF_PROG_TYPE_CGROUP_SKB:
1887                 return prog->enforce_expected_attach_type &&
1888                         prog->expected_attach_type != attach_type ?
1889                         -EINVAL : 0;
1890         default:
1891                 return 0;
1892         }
1893 }
1894
1895 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1896
1897 #define BPF_F_ATTACH_MASK \
1898         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1899
1900 static int bpf_prog_attach(const union bpf_attr *attr)
1901 {
1902         enum bpf_prog_type ptype;
1903         struct bpf_prog *prog;
1904         int ret;
1905
1906         if (!capable(CAP_NET_ADMIN))
1907                 return -EPERM;
1908
1909         if (CHECK_ATTR(BPF_PROG_ATTACH))
1910                 return -EINVAL;
1911
1912         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1913                 return -EINVAL;
1914
1915         switch (attr->attach_type) {
1916         case BPF_CGROUP_INET_INGRESS:
1917         case BPF_CGROUP_INET_EGRESS:
1918                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1919                 break;
1920         case BPF_CGROUP_INET_SOCK_CREATE:
1921         case BPF_CGROUP_INET4_POST_BIND:
1922         case BPF_CGROUP_INET6_POST_BIND:
1923                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1924                 break;
1925         case BPF_CGROUP_INET4_BIND:
1926         case BPF_CGROUP_INET6_BIND:
1927         case BPF_CGROUP_INET4_CONNECT:
1928         case BPF_CGROUP_INET6_CONNECT:
1929         case BPF_CGROUP_UDP4_SENDMSG:
1930         case BPF_CGROUP_UDP6_SENDMSG:
1931         case BPF_CGROUP_UDP4_RECVMSG:
1932         case BPF_CGROUP_UDP6_RECVMSG:
1933                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1934                 break;
1935         case BPF_CGROUP_SOCK_OPS:
1936                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1937                 break;
1938         case BPF_CGROUP_DEVICE:
1939                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1940                 break;
1941         case BPF_SK_MSG_VERDICT:
1942                 ptype = BPF_PROG_TYPE_SK_MSG;
1943                 break;
1944         case BPF_SK_SKB_STREAM_PARSER:
1945         case BPF_SK_SKB_STREAM_VERDICT:
1946                 ptype = BPF_PROG_TYPE_SK_SKB;
1947                 break;
1948         case BPF_LIRC_MODE2:
1949                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1950                 break;
1951         case BPF_FLOW_DISSECTOR:
1952                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1953                 break;
1954         case BPF_CGROUP_SYSCTL:
1955                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1956                 break;
1957         case BPF_CGROUP_GETSOCKOPT:
1958         case BPF_CGROUP_SETSOCKOPT:
1959                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
1960                 break;
1961         default:
1962                 return -EINVAL;
1963         }
1964
1965         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1966         if (IS_ERR(prog))
1967                 return PTR_ERR(prog);
1968
1969         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1970                 bpf_prog_put(prog);
1971                 return -EINVAL;
1972         }
1973
1974         switch (ptype) {
1975         case BPF_PROG_TYPE_SK_SKB:
1976         case BPF_PROG_TYPE_SK_MSG:
1977                 ret = sock_map_get_from_fd(attr, prog);
1978                 break;
1979         case BPF_PROG_TYPE_LIRC_MODE2:
1980                 ret = lirc_prog_attach(attr, prog);
1981                 break;
1982         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1983                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1984                 break;
1985         default:
1986                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1987         }
1988
1989         if (ret)
1990                 bpf_prog_put(prog);
1991         return ret;
1992 }
1993
1994 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1995
1996 static int bpf_prog_detach(const union bpf_attr *attr)
1997 {
1998         enum bpf_prog_type ptype;
1999
2000         if (!capable(CAP_NET_ADMIN))
2001                 return -EPERM;
2002
2003         if (CHECK_ATTR(BPF_PROG_DETACH))
2004                 return -EINVAL;
2005
2006         switch (attr->attach_type) {
2007         case BPF_CGROUP_INET_INGRESS:
2008         case BPF_CGROUP_INET_EGRESS:
2009                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2010                 break;
2011         case BPF_CGROUP_INET_SOCK_CREATE:
2012         case BPF_CGROUP_INET4_POST_BIND:
2013         case BPF_CGROUP_INET6_POST_BIND:
2014                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2015                 break;
2016         case BPF_CGROUP_INET4_BIND:
2017         case BPF_CGROUP_INET6_BIND:
2018         case BPF_CGROUP_INET4_CONNECT:
2019         case BPF_CGROUP_INET6_CONNECT:
2020         case BPF_CGROUP_UDP4_SENDMSG:
2021         case BPF_CGROUP_UDP6_SENDMSG:
2022         case BPF_CGROUP_UDP4_RECVMSG:
2023         case BPF_CGROUP_UDP6_RECVMSG:
2024                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2025                 break;
2026         case BPF_CGROUP_SOCK_OPS:
2027                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2028                 break;
2029         case BPF_CGROUP_DEVICE:
2030                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2031                 break;
2032         case BPF_SK_MSG_VERDICT:
2033                 return sock_map_prog_detach(attr, BPF_PROG_TYPE_SK_MSG);
2034         case BPF_SK_SKB_STREAM_PARSER:
2035         case BPF_SK_SKB_STREAM_VERDICT:
2036                 return sock_map_prog_detach(attr, BPF_PROG_TYPE_SK_SKB);
2037         case BPF_LIRC_MODE2:
2038                 return lirc_prog_detach(attr);
2039         case BPF_FLOW_DISSECTOR:
2040                 return skb_flow_dissector_bpf_prog_detach(attr);
2041         case BPF_CGROUP_SYSCTL:
2042                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2043                 break;
2044         case BPF_CGROUP_GETSOCKOPT:
2045         case BPF_CGROUP_SETSOCKOPT:
2046                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2047                 break;
2048         default:
2049                 return -EINVAL;
2050         }
2051
2052         return cgroup_bpf_prog_detach(attr, ptype);
2053 }
2054
2055 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2056
2057 static int bpf_prog_query(const union bpf_attr *attr,
2058                           union bpf_attr __user *uattr)
2059 {
2060         if (!capable(CAP_NET_ADMIN))
2061                 return -EPERM;
2062         if (CHECK_ATTR(BPF_PROG_QUERY))
2063                 return -EINVAL;
2064         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2065                 return -EINVAL;
2066
2067         switch (attr->query.attach_type) {
2068         case BPF_CGROUP_INET_INGRESS:
2069         case BPF_CGROUP_INET_EGRESS:
2070         case BPF_CGROUP_INET_SOCK_CREATE:
2071         case BPF_CGROUP_INET4_BIND:
2072         case BPF_CGROUP_INET6_BIND:
2073         case BPF_CGROUP_INET4_POST_BIND:
2074         case BPF_CGROUP_INET6_POST_BIND:
2075         case BPF_CGROUP_INET4_CONNECT:
2076         case BPF_CGROUP_INET6_CONNECT:
2077         case BPF_CGROUP_UDP4_SENDMSG:
2078         case BPF_CGROUP_UDP6_SENDMSG:
2079         case BPF_CGROUP_UDP4_RECVMSG:
2080         case BPF_CGROUP_UDP6_RECVMSG:
2081         case BPF_CGROUP_SOCK_OPS:
2082         case BPF_CGROUP_DEVICE:
2083         case BPF_CGROUP_SYSCTL:
2084         case BPF_CGROUP_GETSOCKOPT:
2085         case BPF_CGROUP_SETSOCKOPT:
2086                 break;
2087         case BPF_LIRC_MODE2:
2088                 return lirc_prog_query(attr, uattr);
2089         case BPF_FLOW_DISSECTOR:
2090                 return skb_flow_dissector_prog_query(attr, uattr);
2091         default:
2092                 return -EINVAL;
2093         }
2094
2095         return cgroup_bpf_prog_query(attr, uattr);
2096 }
2097
2098 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2099
2100 static int bpf_prog_test_run(const union bpf_attr *attr,
2101                              union bpf_attr __user *uattr)
2102 {
2103         struct bpf_prog *prog;
2104         int ret = -ENOTSUPP;
2105
2106         if (!capable(CAP_SYS_ADMIN))
2107                 return -EPERM;
2108         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2109                 return -EINVAL;
2110
2111         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2112             (!attr->test.ctx_size_in && attr->test.ctx_in))
2113                 return -EINVAL;
2114
2115         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2116             (!attr->test.ctx_size_out && attr->test.ctx_out))
2117                 return -EINVAL;
2118
2119         prog = bpf_prog_get(attr->test.prog_fd);
2120         if (IS_ERR(prog))
2121                 return PTR_ERR(prog);
2122
2123         if (prog->aux->ops->test_run)
2124                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2125
2126         bpf_prog_put(prog);
2127         return ret;
2128 }
2129
2130 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2131
2132 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2133                                union bpf_attr __user *uattr,
2134                                struct idr *idr,
2135                                spinlock_t *lock)
2136 {
2137         u32 next_id = attr->start_id;
2138         int err = 0;
2139
2140         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2141                 return -EINVAL;
2142
2143         if (!capable(CAP_SYS_ADMIN))
2144                 return -EPERM;
2145
2146         next_id++;
2147         spin_lock_bh(lock);
2148         if (!idr_get_next(idr, &next_id))
2149                 err = -ENOENT;
2150         spin_unlock_bh(lock);
2151
2152         if (!err)
2153                 err = put_user(next_id, &uattr->next_id);
2154
2155         return err;
2156 }
2157
2158 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2159
2160 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2161 {
2162         struct bpf_prog *prog;
2163         u32 id = attr->prog_id;
2164         int fd;
2165
2166         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2167                 return -EINVAL;
2168
2169         if (!capable(CAP_SYS_ADMIN))
2170                 return -EPERM;
2171
2172         spin_lock_bh(&prog_idr_lock);
2173         prog = idr_find(&prog_idr, id);
2174         if (prog)
2175                 prog = bpf_prog_inc_not_zero(prog);
2176         else
2177                 prog = ERR_PTR(-ENOENT);
2178         spin_unlock_bh(&prog_idr_lock);
2179
2180         if (IS_ERR(prog))
2181                 return PTR_ERR(prog);
2182
2183         fd = bpf_prog_new_fd(prog);
2184         if (fd < 0)
2185                 bpf_prog_put(prog);
2186
2187         return fd;
2188 }
2189
2190 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2191
2192 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2193 {
2194         struct bpf_map *map;
2195         u32 id = attr->map_id;
2196         int f_flags;
2197         int fd;
2198
2199         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2200             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2201                 return -EINVAL;
2202
2203         if (!capable(CAP_SYS_ADMIN))
2204                 return -EPERM;
2205
2206         f_flags = bpf_get_file_flag(attr->open_flags);
2207         if (f_flags < 0)
2208                 return f_flags;
2209
2210         spin_lock_bh(&map_idr_lock);
2211         map = idr_find(&map_idr, id);
2212         if (map)
2213                 map = __bpf_map_inc_not_zero(map, true);
2214         else
2215                 map = ERR_PTR(-ENOENT);
2216         spin_unlock_bh(&map_idr_lock);
2217
2218         if (IS_ERR(map))
2219                 return PTR_ERR(map);
2220
2221         fd = bpf_map_new_fd(map, f_flags);
2222         if (fd < 0)
2223                 bpf_map_put_with_uref(map);
2224
2225         return fd;
2226 }
2227
2228 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2229                                               unsigned long addr, u32 *off,
2230                                               u32 *type)
2231 {
2232         const struct bpf_map *map;
2233         int i;
2234
2235         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2236                 map = prog->aux->used_maps[i];
2237                 if (map == (void *)addr) {
2238                         *type = BPF_PSEUDO_MAP_FD;
2239                         return map;
2240                 }
2241                 if (!map->ops->map_direct_value_meta)
2242                         continue;
2243                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2244                         *type = BPF_PSEUDO_MAP_VALUE;
2245                         return map;
2246                 }
2247         }
2248
2249         return NULL;
2250 }
2251
2252 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
2253                                               const struct cred *f_cred)
2254 {
2255         const struct bpf_map *map;
2256         struct bpf_insn *insns;
2257         u32 off, type;
2258         u64 imm;
2259         int i;
2260
2261         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2262                         GFP_USER);
2263         if (!insns)
2264                 return insns;
2265
2266         for (i = 0; i < prog->len; i++) {
2267                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2268                         insns[i].code = BPF_JMP | BPF_CALL;
2269                         insns[i].imm = BPF_FUNC_tail_call;
2270                         /* fall-through */
2271                 }
2272                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2273                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2274                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2275                                 insns[i].code = BPF_JMP | BPF_CALL;
2276                         if (!bpf_dump_raw_ok(f_cred))
2277                                 insns[i].imm = 0;
2278                         continue;
2279                 }
2280
2281                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2282                         continue;
2283
2284                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2285                 map = bpf_map_from_imm(prog, imm, &off, &type);
2286                 if (map) {
2287                         insns[i].src_reg = type;
2288                         insns[i].imm = map->id;
2289                         insns[i + 1].imm = off;
2290                         continue;
2291                 }
2292         }
2293
2294         return insns;
2295 }
2296
2297 static int set_info_rec_size(struct bpf_prog_info *info)
2298 {
2299         /*
2300          * Ensure info.*_rec_size is the same as kernel expected size
2301          *
2302          * or
2303          *
2304          * Only allow zero *_rec_size if both _rec_size and _cnt are
2305          * zero.  In this case, the kernel will set the expected
2306          * _rec_size back to the info.
2307          */
2308
2309         if ((info->nr_func_info || info->func_info_rec_size) &&
2310             info->func_info_rec_size != sizeof(struct bpf_func_info))
2311                 return -EINVAL;
2312
2313         if ((info->nr_line_info || info->line_info_rec_size) &&
2314             info->line_info_rec_size != sizeof(struct bpf_line_info))
2315                 return -EINVAL;
2316
2317         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2318             info->jited_line_info_rec_size != sizeof(__u64))
2319                 return -EINVAL;
2320
2321         info->func_info_rec_size = sizeof(struct bpf_func_info);
2322         info->line_info_rec_size = sizeof(struct bpf_line_info);
2323         info->jited_line_info_rec_size = sizeof(__u64);
2324
2325         return 0;
2326 }
2327
2328 static int bpf_prog_get_info_by_fd(struct file *file,
2329                                    struct bpf_prog *prog,
2330                                    const union bpf_attr *attr,
2331                                    union bpf_attr __user *uattr)
2332 {
2333         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2334         struct bpf_prog_info info;
2335         u32 info_len = attr->info.info_len;
2336         struct bpf_prog_stats stats;
2337         char __user *uinsns;
2338         u32 ulen;
2339         int err;
2340
2341         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2342         if (err)
2343                 return err;
2344         info_len = min_t(u32, sizeof(info), info_len);
2345
2346         memset(&info, 0, sizeof(info));
2347         if (copy_from_user(&info, uinfo, info_len))
2348                 return -EFAULT;
2349
2350         info.type = prog->type;
2351         info.id = prog->aux->id;
2352         info.load_time = prog->aux->load_time;
2353         info.created_by_uid = from_kuid_munged(current_user_ns(),
2354                                                prog->aux->user->uid);
2355         info.gpl_compatible = prog->gpl_compatible;
2356
2357         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2358         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2359
2360         ulen = info.nr_map_ids;
2361         info.nr_map_ids = prog->aux->used_map_cnt;
2362         ulen = min_t(u32, info.nr_map_ids, ulen);
2363         if (ulen) {
2364                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2365                 u32 i;
2366
2367                 for (i = 0; i < ulen; i++)
2368                         if (put_user(prog->aux->used_maps[i]->id,
2369                                      &user_map_ids[i]))
2370                                 return -EFAULT;
2371         }
2372
2373         err = set_info_rec_size(&info);
2374         if (err)
2375                 return err;
2376
2377         bpf_prog_get_stats(prog, &stats);
2378         info.run_time_ns = stats.nsecs;
2379         info.run_cnt = stats.cnt;
2380
2381         if (!capable(CAP_SYS_ADMIN)) {
2382                 info.jited_prog_len = 0;
2383                 info.xlated_prog_len = 0;
2384                 info.nr_jited_ksyms = 0;
2385                 info.nr_jited_func_lens = 0;
2386                 info.nr_func_info = 0;
2387                 info.nr_line_info = 0;
2388                 info.nr_jited_line_info = 0;
2389                 goto done;
2390         }
2391
2392         ulen = info.xlated_prog_len;
2393         info.xlated_prog_len = bpf_prog_insn_size(prog);
2394         if (info.xlated_prog_len && ulen) {
2395                 struct bpf_insn *insns_sanitized;
2396                 bool fault;
2397
2398                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
2399                         info.xlated_prog_insns = 0;
2400                         goto done;
2401                 }
2402                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
2403                 if (!insns_sanitized)
2404                         return -ENOMEM;
2405                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2406                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2407                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2408                 kfree(insns_sanitized);
2409                 if (fault)
2410                         return -EFAULT;
2411         }
2412
2413         if (bpf_prog_is_dev_bound(prog->aux)) {
2414                 err = bpf_prog_offload_info_fill(&info, prog);
2415                 if (err)
2416                         return err;
2417                 goto done;
2418         }
2419
2420         /* NOTE: the following code is supposed to be skipped for offload.
2421          * bpf_prog_offload_info_fill() is the place to fill similar fields
2422          * for offload.
2423          */
2424         ulen = info.jited_prog_len;
2425         if (prog->aux->func_cnt) {
2426                 u32 i;
2427
2428                 info.jited_prog_len = 0;
2429                 for (i = 0; i < prog->aux->func_cnt; i++)
2430                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2431         } else {
2432                 info.jited_prog_len = prog->jited_len;
2433         }
2434
2435         if (info.jited_prog_len && ulen) {
2436                 if (bpf_dump_raw_ok(file->f_cred)) {
2437                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2438                         ulen = min_t(u32, info.jited_prog_len, ulen);
2439
2440                         /* for multi-function programs, copy the JITed
2441                          * instructions for all the functions
2442                          */
2443                         if (prog->aux->func_cnt) {
2444                                 u32 len, free, i;
2445                                 u8 *img;
2446
2447                                 free = ulen;
2448                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2449                                         len = prog->aux->func[i]->jited_len;
2450                                         len = min_t(u32, len, free);
2451                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2452                                         if (copy_to_user(uinsns, img, len))
2453                                                 return -EFAULT;
2454                                         uinsns += len;
2455                                         free -= len;
2456                                         if (!free)
2457                                                 break;
2458                                 }
2459                         } else {
2460                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2461                                         return -EFAULT;
2462                         }
2463                 } else {
2464                         info.jited_prog_insns = 0;
2465                 }
2466         }
2467
2468         ulen = info.nr_jited_ksyms;
2469         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2470         if (ulen) {
2471                 if (bpf_dump_raw_ok(file->f_cred)) {
2472                         unsigned long ksym_addr;
2473                         u64 __user *user_ksyms;
2474                         u32 i;
2475
2476                         /* copy the address of the kernel symbol
2477                          * corresponding to each function
2478                          */
2479                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2480                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2481                         if (prog->aux->func_cnt) {
2482                                 for (i = 0; i < ulen; i++) {
2483                                         ksym_addr = (unsigned long)
2484                                                 prog->aux->func[i]->bpf_func;
2485                                         if (put_user((u64) ksym_addr,
2486                                                      &user_ksyms[i]))
2487                                                 return -EFAULT;
2488                                 }
2489                         } else {
2490                                 ksym_addr = (unsigned long) prog->bpf_func;
2491                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2492                                         return -EFAULT;
2493                         }
2494                 } else {
2495                         info.jited_ksyms = 0;
2496                 }
2497         }
2498
2499         ulen = info.nr_jited_func_lens;
2500         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2501         if (ulen) {
2502                 if (bpf_dump_raw_ok(file->f_cred)) {
2503                         u32 __user *user_lens;
2504                         u32 func_len, i;
2505
2506                         /* copy the JITed image lengths for each function */
2507                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2508                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2509                         if (prog->aux->func_cnt) {
2510                                 for (i = 0; i < ulen; i++) {
2511                                         func_len =
2512                                                 prog->aux->func[i]->jited_len;
2513                                         if (put_user(func_len, &user_lens[i]))
2514                                                 return -EFAULT;
2515                                 }
2516                         } else {
2517                                 func_len = prog->jited_len;
2518                                 if (put_user(func_len, &user_lens[0]))
2519                                         return -EFAULT;
2520                         }
2521                 } else {
2522                         info.jited_func_lens = 0;
2523                 }
2524         }
2525
2526         if (prog->aux->btf)
2527                 info.btf_id = btf_id(prog->aux->btf);
2528
2529         ulen = info.nr_func_info;
2530         info.nr_func_info = prog->aux->func_info_cnt;
2531         if (info.nr_func_info && ulen) {
2532                 char __user *user_finfo;
2533
2534                 user_finfo = u64_to_user_ptr(info.func_info);
2535                 ulen = min_t(u32, info.nr_func_info, ulen);
2536                 if (copy_to_user(user_finfo, prog->aux->func_info,
2537                                  info.func_info_rec_size * ulen))
2538                         return -EFAULT;
2539         }
2540
2541         ulen = info.nr_line_info;
2542         info.nr_line_info = prog->aux->nr_linfo;
2543         if (info.nr_line_info && ulen) {
2544                 __u8 __user *user_linfo;
2545
2546                 user_linfo = u64_to_user_ptr(info.line_info);
2547                 ulen = min_t(u32, info.nr_line_info, ulen);
2548                 if (copy_to_user(user_linfo, prog->aux->linfo,
2549                                  info.line_info_rec_size * ulen))
2550                         return -EFAULT;
2551         }
2552
2553         ulen = info.nr_jited_line_info;
2554         if (prog->aux->jited_linfo)
2555                 info.nr_jited_line_info = prog->aux->nr_linfo;
2556         else
2557                 info.nr_jited_line_info = 0;
2558         if (info.nr_jited_line_info && ulen) {
2559                 if (bpf_dump_raw_ok(file->f_cred)) {
2560                         __u64 __user *user_linfo;
2561                         u32 i;
2562
2563                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2564                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2565                         for (i = 0; i < ulen; i++) {
2566                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2567                                              &user_linfo[i]))
2568                                         return -EFAULT;
2569                         }
2570                 } else {
2571                         info.jited_line_info = 0;
2572                 }
2573         }
2574
2575         ulen = info.nr_prog_tags;
2576         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2577         if (ulen) {
2578                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2579                 u32 i;
2580
2581                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2582                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2583                 if (prog->aux->func_cnt) {
2584                         for (i = 0; i < ulen; i++) {
2585                                 if (copy_to_user(user_prog_tags[i],
2586                                                  prog->aux->func[i]->tag,
2587                                                  BPF_TAG_SIZE))
2588                                         return -EFAULT;
2589                         }
2590                 } else {
2591                         if (copy_to_user(user_prog_tags[0],
2592                                          prog->tag, BPF_TAG_SIZE))
2593                                 return -EFAULT;
2594                 }
2595         }
2596
2597 done:
2598         if (copy_to_user(uinfo, &info, info_len) ||
2599             put_user(info_len, &uattr->info.info_len))
2600                 return -EFAULT;
2601
2602         return 0;
2603 }
2604
2605 static int bpf_map_get_info_by_fd(struct file *file,
2606                                   struct bpf_map *map,
2607                                   const union bpf_attr *attr,
2608                                   union bpf_attr __user *uattr)
2609 {
2610         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2611         struct bpf_map_info info;
2612         u32 info_len = attr->info.info_len;
2613         int err;
2614
2615         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2616         if (err)
2617                 return err;
2618         info_len = min_t(u32, sizeof(info), info_len);
2619
2620         memset(&info, 0, sizeof(info));
2621         info.type = map->map_type;
2622         info.id = map->id;
2623         info.key_size = map->key_size;
2624         info.value_size = map->value_size;
2625         info.max_entries = map->max_entries;
2626         info.map_flags = map->map_flags;
2627         memcpy(info.name, map->name, sizeof(map->name));
2628
2629         if (map->btf) {
2630                 info.btf_id = btf_id(map->btf);
2631                 info.btf_key_type_id = map->btf_key_type_id;
2632                 info.btf_value_type_id = map->btf_value_type_id;
2633         }
2634
2635         if (bpf_map_is_dev_bound(map)) {
2636                 err = bpf_map_offload_info_fill(&info, map);
2637                 if (err)
2638                         return err;
2639         }
2640
2641         if (copy_to_user(uinfo, &info, info_len) ||
2642             put_user(info_len, &uattr->info.info_len))
2643                 return -EFAULT;
2644
2645         return 0;
2646 }
2647
2648 static int bpf_btf_get_info_by_fd(struct file *file,
2649                                   struct btf *btf,
2650                                   const union bpf_attr *attr,
2651                                   union bpf_attr __user *uattr)
2652 {
2653         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2654         u32 info_len = attr->info.info_len;
2655         int err;
2656
2657         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2658         if (err)
2659                 return err;
2660
2661         return btf_get_info_by_fd(btf, attr, uattr);
2662 }
2663
2664 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2665
2666 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2667                                   union bpf_attr __user *uattr)
2668 {
2669         int ufd = attr->info.bpf_fd;
2670         struct fd f;
2671         int err;
2672
2673         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2674                 return -EINVAL;
2675
2676         f = fdget(ufd);
2677         if (!f.file)
2678                 return -EBADFD;
2679
2680         if (f.file->f_op == &bpf_prog_fops)
2681                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
2682                                               uattr);
2683         else if (f.file->f_op == &bpf_map_fops)
2684                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
2685                                              uattr);
2686         else if (f.file->f_op == &btf_fops)
2687                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
2688         else
2689                 err = -EINVAL;
2690
2691         fdput(f);
2692         return err;
2693 }
2694
2695 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2696
2697 static int bpf_btf_load(const union bpf_attr *attr)
2698 {
2699         if (CHECK_ATTR(BPF_BTF_LOAD))
2700                 return -EINVAL;
2701
2702         if (!capable(CAP_SYS_ADMIN))
2703                 return -EPERM;
2704
2705         return btf_new_fd(attr);
2706 }
2707
2708 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2709
2710 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2711 {
2712         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2713                 return -EINVAL;
2714
2715         if (!capable(CAP_SYS_ADMIN))
2716                 return -EPERM;
2717
2718         return btf_get_fd_by_id(attr->btf_id);
2719 }
2720
2721 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2722                                     union bpf_attr __user *uattr,
2723                                     u32 prog_id, u32 fd_type,
2724                                     const char *buf, u64 probe_offset,
2725                                     u64 probe_addr)
2726 {
2727         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2728         u32 len = buf ? strlen(buf) : 0, input_len;
2729         int err = 0;
2730
2731         if (put_user(len, &uattr->task_fd_query.buf_len))
2732                 return -EFAULT;
2733         input_len = attr->task_fd_query.buf_len;
2734         if (input_len && ubuf) {
2735                 if (!len) {
2736                         /* nothing to copy, just make ubuf NULL terminated */
2737                         char zero = '\0';
2738
2739                         if (put_user(zero, ubuf))
2740                                 return -EFAULT;
2741                 } else if (input_len >= len + 1) {
2742                         /* ubuf can hold the string with NULL terminator */
2743                         if (copy_to_user(ubuf, buf, len + 1))
2744                                 return -EFAULT;
2745                 } else {
2746                         /* ubuf cannot hold the string with NULL terminator,
2747                          * do a partial copy with NULL terminator.
2748                          */
2749                         char zero = '\0';
2750
2751                         err = -ENOSPC;
2752                         if (copy_to_user(ubuf, buf, input_len - 1))
2753                                 return -EFAULT;
2754                         if (put_user(zero, ubuf + input_len - 1))
2755                                 return -EFAULT;
2756                 }
2757         }
2758
2759         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2760             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2761             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2762             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2763                 return -EFAULT;
2764
2765         return err;
2766 }
2767
2768 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2769
2770 static int bpf_task_fd_query(const union bpf_attr *attr,
2771                              union bpf_attr __user *uattr)
2772 {
2773         pid_t pid = attr->task_fd_query.pid;
2774         u32 fd = attr->task_fd_query.fd;
2775         const struct perf_event *event;
2776         struct files_struct *files;
2777         struct task_struct *task;
2778         struct file *file;
2779         int err;
2780
2781         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2782                 return -EINVAL;
2783
2784         if (!capable(CAP_SYS_ADMIN))
2785                 return -EPERM;
2786
2787         if (attr->task_fd_query.flags != 0)
2788                 return -EINVAL;
2789
2790         rcu_read_lock();
2791         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2792         rcu_read_unlock();
2793         if (!task)
2794                 return -ENOENT;
2795
2796         files = get_files_struct(task);
2797         put_task_struct(task);
2798         if (!files)
2799                 return -ENOENT;
2800
2801         err = 0;
2802         spin_lock(&files->file_lock);
2803         file = fcheck_files(files, fd);
2804         if (!file)
2805                 err = -EBADF;
2806         else
2807                 get_file(file);
2808         spin_unlock(&files->file_lock);
2809         put_files_struct(files);
2810
2811         if (err)
2812                 goto out;
2813
2814         if (file->f_op == &bpf_raw_tp_fops) {
2815                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2816                 struct bpf_raw_event_map *btp = raw_tp->btp;
2817
2818                 err = bpf_task_fd_query_copy(attr, uattr,
2819                                              raw_tp->prog->aux->id,
2820                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2821                                              btp->tp->name, 0, 0);
2822                 goto put_file;
2823         }
2824
2825         event = perf_get_event(file);
2826         if (!IS_ERR(event)) {
2827                 u64 probe_offset, probe_addr;
2828                 u32 prog_id, fd_type;
2829                 const char *buf;
2830
2831                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2832                                               &buf, &probe_offset,
2833                                               &probe_addr);
2834                 if (!err)
2835                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2836                                                      fd_type, buf,
2837                                                      probe_offset,
2838                                                      probe_addr);
2839                 goto put_file;
2840         }
2841
2842         err = -ENOTSUPP;
2843 put_file:
2844         fput(file);
2845 out:
2846         return err;
2847 }
2848
2849 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2850 {
2851         union bpf_attr attr;
2852         int err;
2853
2854         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2855                 return -EPERM;
2856
2857         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2858         if (err)
2859                 return err;
2860         size = min_t(u32, size, sizeof(attr));
2861
2862         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2863         memset(&attr, 0, sizeof(attr));
2864         if (copy_from_user(&attr, uattr, size) != 0)
2865                 return -EFAULT;
2866
2867         err = security_bpf(cmd, &attr, size);
2868         if (err < 0)
2869                 return err;
2870
2871         switch (cmd) {
2872         case BPF_MAP_CREATE:
2873                 err = map_create(&attr);
2874                 break;
2875         case BPF_MAP_LOOKUP_ELEM:
2876                 err = map_lookup_elem(&attr);
2877                 break;
2878         case BPF_MAP_UPDATE_ELEM:
2879                 err = map_update_elem(&attr);
2880                 break;
2881         case BPF_MAP_DELETE_ELEM:
2882                 err = map_delete_elem(&attr);
2883                 break;
2884         case BPF_MAP_GET_NEXT_KEY:
2885                 err = map_get_next_key(&attr);
2886                 break;
2887         case BPF_MAP_FREEZE:
2888                 err = map_freeze(&attr);
2889                 break;
2890         case BPF_PROG_LOAD:
2891                 err = bpf_prog_load(&attr, uattr);
2892                 break;
2893         case BPF_OBJ_PIN:
2894                 err = bpf_obj_pin(&attr);
2895                 break;
2896         case BPF_OBJ_GET:
2897                 err = bpf_obj_get(&attr);
2898                 break;
2899         case BPF_PROG_ATTACH:
2900                 err = bpf_prog_attach(&attr);
2901                 break;
2902         case BPF_PROG_DETACH:
2903                 err = bpf_prog_detach(&attr);
2904                 break;
2905         case BPF_PROG_QUERY:
2906                 err = bpf_prog_query(&attr, uattr);
2907                 break;
2908         case BPF_PROG_TEST_RUN:
2909                 err = bpf_prog_test_run(&attr, uattr);
2910                 break;
2911         case BPF_PROG_GET_NEXT_ID:
2912                 err = bpf_obj_get_next_id(&attr, uattr,
2913                                           &prog_idr, &prog_idr_lock);
2914                 break;
2915         case BPF_MAP_GET_NEXT_ID:
2916                 err = bpf_obj_get_next_id(&attr, uattr,
2917                                           &map_idr, &map_idr_lock);
2918                 break;
2919         case BPF_BTF_GET_NEXT_ID:
2920                 err = bpf_obj_get_next_id(&attr, uattr,
2921                                           &btf_idr, &btf_idr_lock);
2922                 break;
2923         case BPF_PROG_GET_FD_BY_ID:
2924                 err = bpf_prog_get_fd_by_id(&attr);
2925                 break;
2926         case BPF_MAP_GET_FD_BY_ID:
2927                 err = bpf_map_get_fd_by_id(&attr);
2928                 break;
2929         case BPF_OBJ_GET_INFO_BY_FD:
2930                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2931                 break;
2932         case BPF_RAW_TRACEPOINT_OPEN:
2933                 err = bpf_raw_tracepoint_open(&attr);
2934                 break;
2935         case BPF_BTF_LOAD:
2936                 err = bpf_btf_load(&attr);
2937                 break;
2938         case BPF_BTF_GET_FD_BY_ID:
2939                 err = bpf_btf_get_fd_by_id(&attr);
2940                 break;
2941         case BPF_TASK_FD_QUERY:
2942                 err = bpf_task_fd_query(&attr, uattr);
2943                 break;
2944         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2945                 err = map_lookup_and_delete_elem(&attr);
2946                 break;
2947         default:
2948                 err = -EINVAL;
2949                 break;
2950         }
2951
2952         return err;
2953 }