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 #include <uapi/linux/btf.h>
  27
  28 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  29                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  30                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  31                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  32 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  33 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  34
  35 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  36
  37 DEFINE_PER_CPU(int, bpf_prog_active);
  38 static DEFINE_IDR(prog_idr);
  39 static DEFINE_SPINLOCK(prog_idr_lock);
  40 static DEFINE_IDR(map_idr);
  41 static DEFINE_SPINLOCK(map_idr_lock);
  42
  43 int sysctl_unprivileged_bpf_disabled __read_mostly;
  44
  45 static const struct bpf_map_ops * const bpf_map_types[] = {
  46 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  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(size_t 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 <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 146                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 147                                     numa_node);
 148                 if (area != NULL)
 149                         return area;
 150         }
 151
 152         return __vmalloc_node_flags_caller(size, numa_node,
 153                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 154                                            flags, __builtin_return_address(0));
 155 }
 156
 157 void bpf_map_area_free(void *area)
 158 {
 159         kvfree(area);
 160 }
 161
 162 static u32 bpf_map_flags_retain_permanent(u32 flags)
 163 {
 164         /* Some map creation flags are not tied to the map object but
 165          * rather to the map fd instead, so they have no meaning upon
 166          * map object inspection since multiple file descriptors with
 167          * different (access) properties can exist here. Thus, given
 168          * this has zero meaning for the map itself, lets clear these
 169          * from here.
 170          */
 171         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 172 }
 173
 174 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 175 {
 176         map->map_type = attr->map_type;
 177         map->key_size = attr->key_size;
 178         map->value_size = attr->value_size;
 179         map->max_entries = attr->max_entries;
 180         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 181         map->numa_node = bpf_map_attr_numa_node(attr);
 182 }
 183
 184 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 185 {
 186         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 187
 188         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 189                 atomic_long_sub(pages, &user->locked_vm);
 190                 return -EPERM;
 191         }
 192         return 0;
 193 }
 194
 195 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 196 {
 197         if (user)
 198                 atomic_long_sub(pages, &user->locked_vm);
 199 }
 200
 201 int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size)
 202 {
 203         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 204         struct user_struct *user;
 205         int ret;
 206
 207         if (size >= U32_MAX - PAGE_SIZE)
 208                 return -E2BIG;
 209
 210         user = get_current_user();
 211         ret = bpf_charge_memlock(user, pages);
 212         if (ret) {
 213                 free_uid(user);
 214                 return ret;
 215         }
 216
 217         mem->pages = pages;
 218         mem->user = user;
 219
 220         return 0;
 221 }
 222
 223 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 224 {
 225         bpf_uncharge_memlock(mem->user, mem->pages);
 226         free_uid(mem->user);
 227 }
 228
 229 void bpf_map_charge_move(struct bpf_map_memory *dst,
 230                          struct bpf_map_memory *src)
 231 {
 232         *dst = *src;
 233
 234         /* Make sure src will not be used for the redundant uncharging. */
 235         memset(src, 0, sizeof(struct bpf_map_memory));
 236 }
 237
 238 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 239 {
 240         int ret;
 241
 242         ret = bpf_charge_memlock(map->memory.user, pages);
 243         if (ret)
 244                 return ret;
 245         map->memory.pages += pages;
 246         return ret;
 247 }
 248
 249 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 250 {
 251         bpf_uncharge_memlock(map->memory.user, pages);
 252         map->memory.pages -= pages;
 253 }
 254
 255 static int bpf_map_alloc_id(struct bpf_map *map)
 256 {
 257         int id;
 258
 259         idr_preload(GFP_KERNEL);
 260         spin_lock_bh(&map_idr_lock);
 261         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 262         if (id > 0)
 263                 map->id = id;
 264         spin_unlock_bh(&map_idr_lock);
 265         idr_preload_end();
 266
 267         if (WARN_ON_ONCE(!id))
 268                 return -ENOSPC;
 269
 270         return id > 0 ? 0 : id;
 271 }
 272
 273 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 274 {
 275         unsigned long flags;
 276
 277         /* Offloaded maps are removed from the IDR store when their device
 278          * disappears - even if someone holds an fd to them they are unusable,
 279          * the memory is gone, all ops will fail; they are simply waiting for
 280          * refcnt to drop to be freed.
 281          */
 282         if (!map->id)
 283                 return;
 284
 285         if (do_idr_lock)
 286                 spin_lock_irqsave(&map_idr_lock, flags);
 287         else
 288                 __acquire(&map_idr_lock);
 289
 290         idr_remove(&map_idr, map->id);
 291         map->id = 0;
 292
 293         if (do_idr_lock)
 294                 spin_unlock_irqrestore(&map_idr_lock, flags);
 295         else
 296                 __release(&map_idr_lock);
 297 }
 298
 299 /* called from workqueue */
 300 static void bpf_map_free_deferred(struct work_struct *work)
 301 {
 302         struct bpf_map *map = container_of(work, struct bpf_map, work);
 303         struct bpf_map_memory mem;
 304
 305         bpf_map_charge_move(&mem, &map->memory);
 306         security_bpf_map_free(map);
 307         /* implementation dependent freeing */
 308         map->ops->map_free(map);
 309         bpf_map_charge_finish(&mem);
 310 }
 311
 312 static void bpf_map_put_uref(struct bpf_map *map)
 313 {
 314         if (atomic_dec_and_test(&map->usercnt)) {
 315                 if (map->ops->map_release_uref)
 316                         map->ops->map_release_uref(map);
 317         }
 318 }
 319
 320 /* decrement map refcnt and schedule it for freeing via workqueue
 321  * (unrelying map implementation ops->map_free() might sleep)
 322  */
 323 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 324 {
 325         if (atomic_dec_and_test(&map->refcnt)) {
 326                 /* bpf_map_free_id() must be called first */
 327                 bpf_map_free_id(map, do_idr_lock);
 328                 btf_put(map->btf);
 329                 INIT_WORK(&map->work, bpf_map_free_deferred);
 330                 schedule_work(&map->work);
 331         }
 332 }
 333
 334 void bpf_map_put(struct bpf_map *map)
 335 {
 336         __bpf_map_put(map, true);
 337 }
 338 EXPORT_SYMBOL_GPL(bpf_map_put);
 339
 340 void bpf_map_put_with_uref(struct bpf_map *map)
 341 {
 342         bpf_map_put_uref(map);
 343         bpf_map_put(map);
 344 }
 345
 346 static int bpf_map_release(struct inode *inode, struct file *filp)
 347 {
 348         struct bpf_map *map = filp->private_data;
 349
 350         if (map->ops->map_release)
 351                 map->ops->map_release(map, filp);
 352
 353         bpf_map_put_with_uref(map);
 354         return 0;
 355 }
 356
 357 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 358 {
 359         fmode_t mode = f.file->f_mode;
 360
 361         /* Our file permissions may have been overridden by global
 362          * map permissions facing syscall side.
 363          */
 364         if (READ_ONCE(map->frozen))
 365                 mode &= ~FMODE_CAN_WRITE;
 366         return mode;
 367 }
 368
 369 #ifdef CONFIG_PROC_FS
 370 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 371 {
 372         const struct bpf_map *map = filp->private_data;
 373         const struct bpf_array *array;
 374         u32 owner_prog_type = 0;
 375         u32 owner_jited = 0;
 376
 377         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 378                 array = container_of(map, struct bpf_array, map);
 379                 owner_prog_type = array->owner_prog_type;
 380                 owner_jited = array->owner_jited;
 381         }
 382
 383         seq_printf(m,
 384                    "map_type:\t%u\n"
 385                    "key_size:\t%u\n"
 386                    "value_size:\t%u\n"
 387                    "max_entries:\t%u\n"
 388                    "map_flags:\t%#x\n"
 389                    "memlock:\t%llu\n"
 390                    "map_id:\t%u\n"
 391                    "frozen:\t%u\n",
 392                    map->map_type,
 393                    map->key_size,
 394                    map->value_size,
 395                    map->max_entries,
 396                    map->map_flags,
 397                    map->memory.pages * 1ULL << PAGE_SHIFT,
 398                    map->id,
 399                    READ_ONCE(map->frozen));
 400
 401         if (owner_prog_type) {
 402                 seq_printf(m, "owner_prog_type:\t%u\n",
 403                            owner_prog_type);
 404                 seq_printf(m, "owner_jited:\t%u\n",
 405                            owner_jited);
 406         }
 407 }
 408 #endif
 409
 410 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 411                               loff_t *ppos)
 412 {
 413         /* We need this handler such that alloc_file() enables
 414          * f_mode with FMODE_CAN_READ.
 415          */
 416         return -EINVAL;
 417 }
 418
 419 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 420                                size_t siz, loff_t *ppos)
 421 {
 422         /* We need this handler such that alloc_file() enables
 423          * f_mode with FMODE_CAN_WRITE.
 424          */
 425         return -EINVAL;
 426 }
 427
 428 const struct file_operations bpf_map_fops = {
 429 #ifdef CONFIG_PROC_FS
 430         .show_fdinfo    = bpf_map_show_fdinfo,
 431 #endif
 432         .release        = bpf_map_release,
 433         .read           = bpf_dummy_read,
 434         .write          = bpf_dummy_write,
 435 };
 436
 437 int bpf_map_new_fd(struct bpf_map *map, int flags)
 438 {
 439         int ret;
 440
 441         ret = security_bpf_map(map, OPEN_FMODE(flags));
 442         if (ret < 0)
 443                 return ret;
 444
 445         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 446                                 flags | O_CLOEXEC);
 447 }
 448
 449 int bpf_get_file_flag(int flags)
 450 {
 451         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 452                 return -EINVAL;
 453         if (flags & BPF_F_RDONLY)
 454                 return O_RDONLY;
 455         if (flags & BPF_F_WRONLY)
 456                 return O_WRONLY;
 457         return O_RDWR;
 458 }
 459
 460 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 461 #define CHECK_ATTR(CMD) \
 462         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 463                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 464                    sizeof(*attr) - \
 465                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 466                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 467
 468 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 469  * Return 0 on success and < 0 on error.
 470  */
 471 static int bpf_obj_name_cpy(char *dst, const char *src)
 472 {
 473         const char *end = src + BPF_OBJ_NAME_LEN;
 474
 475         memset(dst, 0, BPF_OBJ_NAME_LEN);
 476         /* Copy all isalnum(), '_' and '.' chars. */
 477         while (src < end && *src) {
 478                 if (!isalnum(*src) &&
 479                     *src != '_' && *src != '.')
 480                         return -EINVAL;
 481                 *dst++ = *src++;
 482         }
 483
 484         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 485         if (src == end)
 486                 return -EINVAL;
 487
 488         return 0;
 489 }
 490
 491 int map_check_no_btf(const struct bpf_map *map,
 492                      const struct btf *btf,
 493                      const struct btf_type *key_type,
 494                      const struct btf_type *value_type)
 495 {
 496         return -ENOTSUPP;
 497 }
 498
 499 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 500                          u32 btf_key_id, u32 btf_value_id)
 501 {
 502         const struct btf_type *key_type, *value_type;
 503         u32 key_size, value_size;
 504         int ret = 0;
 505
 506         /* Some maps allow key to be unspecified. */
 507         if (btf_key_id) {
 508                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 509                 if (!key_type || key_size != map->key_size)
 510                         return -EINVAL;
 511         } else {
 512                 key_type = btf_type_by_id(btf, 0);
 513                 if (!map->ops->map_check_btf)
 514                         return -EINVAL;
 515         }
 516
 517         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 518         if (!value_type || value_size != map->value_size)
 519                 return -EINVAL;
 520
 521         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 522
 523         if (map_value_has_spin_lock(map)) {
 524                 if (map->map_flags & BPF_F_RDONLY_PROG)
 525                         return -EACCES;
 526                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 527                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 528                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 529                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 530                         return -ENOTSUPP;
 531                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 532                     map->value_size) {
 533                         WARN_ONCE(1,
 534                                   "verifier bug spin_lock_off %d value_size %d\n",
 535                                   map->spin_lock_off, map->value_size);
 536                         return -EFAULT;
 537                 }
 538         }
 539
 540         if (map->ops->map_check_btf)
 541                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 542
 543         return ret;
 544 }
 545
 546 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 547 /* called via syscall */
 548 static int map_create(union bpf_attr *attr)
 549 {
 550         int numa_node = bpf_map_attr_numa_node(attr);
 551         struct bpf_map_memory mem;
 552         struct bpf_map *map;
 553         int f_flags;
 554         int err;
 555
 556         err = CHECK_ATTR(BPF_MAP_CREATE);
 557         if (err)
 558                 return -EINVAL;
 559
 560         f_flags = bpf_get_file_flag(attr->map_flags);
 561         if (f_flags < 0)
 562                 return f_flags;
 563
 564         if (numa_node != NUMA_NO_NODE &&
 565             ((unsigned int)numa_node >= nr_node_ids ||
 566              !node_online(numa_node)))
 567                 return -EINVAL;
 568
 569         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 570         map = find_and_alloc_map(attr);
 571         if (IS_ERR(map))
 572                 return PTR_ERR(map);
 573
 574         err = bpf_obj_name_cpy(map->name, attr->map_name);
 575         if (err)
 576                 goto free_map;
 577
 578         atomic_set(&map->refcnt, 1);
 579         atomic_set(&map->usercnt, 1);
 580
 581         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 582                 struct btf *btf;
 583
 584                 if (!attr->btf_value_type_id) {
 585                         err = -EINVAL;
 586                         goto free_map;
 587                 }
 588
 589                 btf = btf_get_by_fd(attr->btf_fd);
 590                 if (IS_ERR(btf)) {
 591                         err = PTR_ERR(btf);
 592                         goto free_map;
 593                 }
 594
 595                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 596                                     attr->btf_value_type_id);
 597                 if (err) {
 598                         btf_put(btf);
 599                         goto free_map;
 600                 }
 601
 602                 map->btf = btf;
 603                 map->btf_key_type_id = attr->btf_key_type_id;
 604                 map->btf_value_type_id = attr->btf_value_type_id;
 605         } else {
 606                 map->spin_lock_off = -EINVAL;
 607         }
 608
 609         err = security_bpf_map_alloc(map);
 610         if (err)
 611                 goto free_map;
 612
 613         err = bpf_map_alloc_id(map);
 614         if (err)
 615                 goto free_map_sec;
 616
 617         err = bpf_map_new_fd(map, f_flags);
 618         if (err < 0) {
 619                 /* failed to allocate fd.
 620                  * bpf_map_put_with_uref() is needed because the above
 621                  * bpf_map_alloc_id() has published the map
 622                  * to the userspace and the userspace may
 623                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 624                  */
 625                 bpf_map_put_with_uref(map);
 626                 return err;
 627         }
 628
 629         return err;
 630
 631 free_map_sec:
 632         security_bpf_map_free(map);
 633 free_map:
 634         btf_put(map->btf);
 635         bpf_map_charge_move(&mem, &map->memory);
 636         map->ops->map_free(map);
 637         bpf_map_charge_finish(&mem);
 638         return err;
 639 }
 640
 641 /* if error is returned, fd is released.
 642  * On success caller should complete fd access with matching fdput()
 643  */
 644 struct bpf_map *__bpf_map_get(struct fd f)
 645 {
 646         if (!f.file)
 647                 return ERR_PTR(-EBADF);
 648         if (f.file->f_op != &bpf_map_fops) {
 649                 fdput(f);
 650                 return ERR_PTR(-EINVAL);
 651         }
 652
 653         return f.file->private_data;
 654 }
 655
 656 /* prog's and map's refcnt limit */
 657 #define BPF_MAX_REFCNT 32768
 658
 659 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 660 {
 661         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 662                 atomic_dec(&map->refcnt);
 663                 return ERR_PTR(-EBUSY);
 664         }
 665         if (uref)
 666                 atomic_inc(&map->usercnt);
 667         return map;
 668 }
 669 EXPORT_SYMBOL_GPL(bpf_map_inc);
 670
 671 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 672 {
 673         struct fd f = fdget(ufd);
 674         struct bpf_map *map;
 675
 676         map = __bpf_map_get(f);
 677         if (IS_ERR(map))
 678                 return map;
 679
 680         map = bpf_map_inc(map, true);
 681         fdput(f);
 682
 683         return map;
 684 }
 685
 686 /* map_idr_lock should have been held */
 687 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map,
 688                                               bool uref)
 689 {
 690         int refold;
 691
 692         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 693
 694         if (refold >= BPF_MAX_REFCNT) {
 695                 __bpf_map_put(map, false);
 696                 return ERR_PTR(-EBUSY);
 697         }
 698
 699         if (!refold)
 700                 return ERR_PTR(-ENOENT);
 701
 702         if (uref)
 703                 atomic_inc(&map->usercnt);
 704
 705         return map;
 706 }
 707
 708 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 709 {
 710         spin_lock_bh(&map_idr_lock);
 711         map = __bpf_map_inc_not_zero(map, uref);
 712         spin_unlock_bh(&map_idr_lock);
 713
 714         return map;
 715 }
 716 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 717
 718 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 719 {
 720         return -ENOTSUPP;
 721 }
 722
 723 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 724 {
 725         if (key_size)
 726                 return memdup_user(ukey, key_size);
 727
 728         if (ukey)
 729                 return ERR_PTR(-EINVAL);
 730
 731         return NULL;
 732 }
 733
 734 /* last field in 'union bpf_attr' used by this command */
 735 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 736
 737 static int map_lookup_elem(union bpf_attr *attr)
 738 {
 739         void __user *ukey = u64_to_user_ptr(attr->key);
 740         void __user *uvalue = u64_to_user_ptr(attr->value);
 741         int ufd = attr->map_fd;
 742         struct bpf_map *map;
 743         void *key, *value, *ptr;
 744         u32 value_size;
 745         struct fd f;
 746         int err;
 747
 748         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 749                 return -EINVAL;
 750
 751         if (attr->flags & ~BPF_F_LOCK)
 752                 return -EINVAL;
 753
 754         f = fdget(ufd);
 755         map = __bpf_map_get(f);
 756         if (IS_ERR(map))
 757                 return PTR_ERR(map);
 758         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 759                 err = -EPERM;
 760                 goto err_put;
 761         }
 762
 763         if ((attr->flags & BPF_F_LOCK) &&
 764             !map_value_has_spin_lock(map)) {
 765                 err = -EINVAL;
 766                 goto err_put;
 767         }
 768
 769         key = __bpf_copy_key(ukey, map->key_size);
 770         if (IS_ERR(key)) {
 771                 err = PTR_ERR(key);
 772                 goto err_put;
 773         }
 774
 775         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 776             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 777             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 778             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 779                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 780         else if (IS_FD_MAP(map))
 781                 value_size = sizeof(u32);
 782         else
 783                 value_size = map->value_size;
 784
 785         err = -ENOMEM;
 786         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 787         if (!value)
 788                 goto free_key;
 789
 790         if (bpf_map_is_dev_bound(map)) {
 791                 err = bpf_map_offload_lookup_elem(map, key, value);
 792                 goto done;
 793         }
 794
 795         preempt_disable();
 796         this_cpu_inc(bpf_prog_active);
 797         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 798             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 799                 err = bpf_percpu_hash_copy(map, key, value);
 800         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 801                 err = bpf_percpu_array_copy(map, key, value);
 802         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 803                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 804         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 805                 err = bpf_stackmap_copy(map, key, value);
 806         } else if (IS_FD_ARRAY(map)) {
 807                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 808         } else if (IS_FD_HASH(map)) {
 809                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 810         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 811                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 812         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 813                    map->map_type == BPF_MAP_TYPE_STACK) {
 814                 err = map->ops->map_peek_elem(map, value);
 815         } else {
 816                 rcu_read_lock();
 817                 if (map->ops->map_lookup_elem_sys_only)
 818                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 819                 else
 820                         ptr = map->ops->map_lookup_elem(map, key);
 821                 if (IS_ERR(ptr)) {
 822                         err = PTR_ERR(ptr);
 823                 } else if (!ptr) {
 824                         err = -ENOENT;
 825                 } else {
 826                         err = 0;
 827                         if (attr->flags & BPF_F_LOCK)
 828                                 /* lock 'ptr' and copy everything but lock */
 829                                 copy_map_value_locked(map, value, ptr, true);
 830                         else
 831                                 copy_map_value(map, value, ptr);
 832                         /* mask lock, since value wasn't zero inited */
 833                         check_and_init_map_lock(map, value);
 834                 }
 835                 rcu_read_unlock();
 836         }
 837         this_cpu_dec(bpf_prog_active);
 838         preempt_enable();
 839
 840 done:
 841         if (err)
 842                 goto free_value;
 843
 844         err = -EFAULT;
 845         if (copy_to_user(uvalue, value, value_size) != 0)
 846                 goto free_value;
 847
 848         err = 0;
 849
 850 free_value:
 851         kfree(value);
 852 free_key:
 853         kfree(key);
 854 err_put:
 855         fdput(f);
 856         return err;
 857 }
 858
 859 static void maybe_wait_bpf_programs(struct bpf_map *map)
 860 {
 861         /* Wait for any running BPF programs to complete so that
 862          * userspace, when we return to it, knows that all programs
 863          * that could be running use the new map value.
 864          */
 865         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 866             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 867                 synchronize_rcu();
 868 }
 869
 870 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 871
 872 static int map_update_elem(union bpf_attr *attr)
 873 {
 874         void __user *ukey = u64_to_user_ptr(attr->key);
 875         void __user *uvalue = u64_to_user_ptr(attr->value);
 876         int ufd = attr->map_fd;
 877         struct bpf_map *map;
 878         void *key, *value;
 879         u32 value_size;
 880         struct fd f;
 881         int err;
 882
 883         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 884                 return -EINVAL;
 885
 886         f = fdget(ufd);
 887         map = __bpf_map_get(f);
 888         if (IS_ERR(map))
 889                 return PTR_ERR(map);
 890         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 891                 err = -EPERM;
 892                 goto err_put;
 893         }
 894
 895         if ((attr->flags & BPF_F_LOCK) &&
 896             !map_value_has_spin_lock(map)) {
 897                 err = -EINVAL;
 898                 goto err_put;
 899         }
 900
 901         key = __bpf_copy_key(ukey, map->key_size);
 902         if (IS_ERR(key)) {
 903                 err = PTR_ERR(key);
 904                 goto err_put;
 905         }
 906
 907         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 908             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 909             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 910             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 911                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 912         else
 913                 value_size = map->value_size;
 914
 915         err = -ENOMEM;
 916         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 917         if (!value)
 918                 goto free_key;
 919
 920         err = -EFAULT;
 921         if (copy_from_user(value, uvalue, value_size) != 0)
 922                 goto free_value;
 923
 924         /* Need to create a kthread, thus must support schedule */
 925         if (bpf_map_is_dev_bound(map)) {
 926                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 927                 goto out;
 928         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 929                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 930                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 931                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 932                 goto out;
 933         }
 934
 935         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 936          * inside bpf map update or delete otherwise deadlocks are possible
 937          */
 938         preempt_disable();
 939         __this_cpu_inc(bpf_prog_active);
 940         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 941             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 942                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 943         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 944                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 945         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 946                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 947                                                        attr->flags);
 948         } else if (IS_FD_ARRAY(map)) {
 949                 rcu_read_lock();
 950                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 951                                                    attr->flags);
 952                 rcu_read_unlock();
 953         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 954                 rcu_read_lock();
 955                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 956                                                   attr->flags);
 957                 rcu_read_unlock();
 958         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 959                 /* rcu_read_lock() is not needed */
 960                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 961                                                          attr->flags);
 962         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 963                    map->map_type == BPF_MAP_TYPE_STACK) {
 964                 err = map->ops->map_push_elem(map, value, attr->flags);
 965         } else {
 966                 rcu_read_lock();
 967                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 968                 rcu_read_unlock();
 969         }
 970         __this_cpu_dec(bpf_prog_active);
 971         preempt_enable();
 972         maybe_wait_bpf_programs(map);
 973 out:
 974 free_value:
 975         kfree(value);
 976 free_key:
 977         kfree(key);
 978 err_put:
 979         fdput(f);
 980         return err;
 981 }
 982
 983 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 984
 985 static int map_delete_elem(union bpf_attr *attr)
 986 {
 987         void __user *ukey = u64_to_user_ptr(attr->key);
 988         int ufd = attr->map_fd;
 989         struct bpf_map *map;
 990         struct fd f;
 991         void *key;
 992         int err;
 993
 994         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 995                 return -EINVAL;
 996
 997         f = fdget(ufd);
 998         map = __bpf_map_get(f);
 999         if (IS_ERR(map))
1000                 return PTR_ERR(map);
1001         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1002                 err = -EPERM;
1003                 goto err_put;
1004         }
1005
1006         key = __bpf_copy_key(ukey, map->key_size);
1007         if (IS_ERR(key)) {
1008                 err = PTR_ERR(key);
1009                 goto err_put;
1010         }
1011
1012         if (bpf_map_is_dev_bound(map)) {
1013                 err = bpf_map_offload_delete_elem(map, key);
1014                 goto out;
1015         }
1016
1017         preempt_disable();
1018         __this_cpu_inc(bpf_prog_active);
1019         rcu_read_lock();
1020         err = map->ops->map_delete_elem(map, key);
1021         rcu_read_unlock();
1022         __this_cpu_dec(bpf_prog_active);
1023         preempt_enable();
1024         maybe_wait_bpf_programs(map);
1025 out:
1026         kfree(key);
1027 err_put:
1028         fdput(f);
1029         return err;
1030 }
1031
1032 /* last field in 'union bpf_attr' used by this command */
1033 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1034
1035 static int map_get_next_key(union bpf_attr *attr)
1036 {
1037         void __user *ukey = u64_to_user_ptr(attr->key);
1038         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1039         int ufd = attr->map_fd;
1040         struct bpf_map *map;
1041         void *key, *next_key;
1042         struct fd f;
1043         int err;
1044
1045         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1046                 return -EINVAL;
1047
1048         f = fdget(ufd);
1049         map = __bpf_map_get(f);
1050         if (IS_ERR(map))
1051                 return PTR_ERR(map);
1052         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1053                 err = -EPERM;
1054                 goto err_put;
1055         }
1056
1057         if (ukey) {
1058                 key = __bpf_copy_key(ukey, map->key_size);
1059                 if (IS_ERR(key)) {
1060                         err = PTR_ERR(key);
1061                         goto err_put;
1062                 }
1063         } else {
1064                 key = NULL;
1065         }
1066
1067         err = -ENOMEM;
1068         next_key = kmalloc(map->key_size, GFP_USER);
1069         if (!next_key)
1070                 goto free_key;
1071
1072         if (bpf_map_is_dev_bound(map)) {
1073                 err = bpf_map_offload_get_next_key(map, key, next_key);
1074                 goto out;
1075         }
1076
1077         rcu_read_lock();
1078         err = map->ops->map_get_next_key(map, key, next_key);
1079         rcu_read_unlock();
1080 out:
1081         if (err)
1082                 goto free_next_key;
1083
1084         err = -EFAULT;
1085         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1086                 goto free_next_key;
1087
1088         err = 0;
1089
1090 free_next_key:
1091         kfree(next_key);
1092 free_key:
1093         kfree(key);
1094 err_put:
1095         fdput(f);
1096         return err;
1097 }
1098
1099 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1100
1101 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1102 {
1103         void __user *ukey = u64_to_user_ptr(attr->key);
1104         void __user *uvalue = u64_to_user_ptr(attr->value);
1105         int ufd = attr->map_fd;
1106         struct bpf_map *map;
1107         void *key, *value;
1108         u32 value_size;
1109         struct fd f;
1110         int err;
1111
1112         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1113                 return -EINVAL;
1114
1115         f = fdget(ufd);
1116         map = __bpf_map_get(f);
1117         if (IS_ERR(map))
1118                 return PTR_ERR(map);
1119         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1120                 err = -EPERM;
1121                 goto err_put;
1122         }
1123
1124         key = __bpf_copy_key(ukey, map->key_size);
1125         if (IS_ERR(key)) {
1126                 err = PTR_ERR(key);
1127                 goto err_put;
1128         }
1129
1130         value_size = map->value_size;
1131
1132         err = -ENOMEM;
1133         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1134         if (!value)
1135                 goto free_key;
1136
1137         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1138             map->map_type == BPF_MAP_TYPE_STACK) {
1139                 err = map->ops->map_pop_elem(map, value);
1140         } else {
1141                 err = -ENOTSUPP;
1142         }
1143
1144         if (err)
1145                 goto free_value;
1146
1147         if (copy_to_user(uvalue, value, value_size) != 0)
1148                 goto free_value;
1149
1150         err = 0;
1151
1152 free_value:
1153         kfree(value);
1154 free_key:
1155         kfree(key);
1156 err_put:
1157         fdput(f);
1158         return err;
1159 }
1160
1161 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1162
1163 static int map_freeze(const union bpf_attr *attr)
1164 {
1165         int err = 0, ufd = attr->map_fd;
1166         struct bpf_map *map;
1167         struct fd f;
1168
1169         if (CHECK_ATTR(BPF_MAP_FREEZE))
1170                 return -EINVAL;
1171
1172         f = fdget(ufd);
1173         map = __bpf_map_get(f);
1174         if (IS_ERR(map))
1175                 return PTR_ERR(map);
1176         if (READ_ONCE(map->frozen)) {
1177                 err = -EBUSY;
1178                 goto err_put;
1179         }
1180         if (!capable(CAP_SYS_ADMIN)) {
1181                 err = -EPERM;
1182                 goto err_put;
1183         }
1184
1185         WRITE_ONCE(map->frozen, true);
1186 err_put:
1187         fdput(f);
1188         return err;
1189 }
1190
1191 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1192 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1193         [_id] = & _name ## _prog_ops,
1194 #define BPF_MAP_TYPE(_id, _ops)
1195 #include <linux/bpf_types.h>
1196 #undef BPF_PROG_TYPE
1197 #undef BPF_MAP_TYPE
1198 };
1199
1200 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1201 {
1202         const struct bpf_prog_ops *ops;
1203
1204         if (type >= ARRAY_SIZE(bpf_prog_types))
1205                 return -EINVAL;
1206         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1207         ops = bpf_prog_types[type];
1208         if (!ops)
1209                 return -EINVAL;
1210
1211         if (!bpf_prog_is_dev_bound(prog->aux))
1212                 prog->aux->ops = ops;
1213         else
1214                 prog->aux->ops = &bpf_offload_prog_ops;
1215         prog->type = type;
1216         return 0;
1217 }
1218
1219 /* drop refcnt on maps used by eBPF program and free auxilary data */
1220 static void free_used_maps(struct bpf_prog_aux *aux)
1221 {
1222         enum bpf_cgroup_storage_type stype;
1223         int i;
1224
1225         for_each_cgroup_storage_type(stype) {
1226                 if (!aux->cgroup_storage[stype])
1227                         continue;
1228                 bpf_cgroup_storage_release(aux->prog,
1229                                            aux->cgroup_storage[stype]);
1230         }
1231
1232         for (i = 0; i < aux->used_map_cnt; i++)
1233                 bpf_map_put(aux->used_maps[i]);
1234
1235         kfree(aux->used_maps);
1236 }
1237
1238 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1239 {
1240         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1241         unsigned long user_bufs;
1242
1243         if (user) {
1244                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1245                 if (user_bufs > memlock_limit) {
1246                         atomic_long_sub(pages, &user->locked_vm);
1247                         return -EPERM;
1248                 }
1249         }
1250
1251         return 0;
1252 }
1253
1254 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1255 {
1256         if (user)
1257                 atomic_long_sub(pages, &user->locked_vm);
1258 }
1259
1260 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1261 {
1262         struct user_struct *user = get_current_user();
1263         int ret;
1264
1265         ret = __bpf_prog_charge(user, prog->pages);
1266         if (ret) {
1267                 free_uid(user);
1268                 return ret;
1269         }
1270
1271         prog->aux->user = user;
1272         return 0;
1273 }
1274
1275 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1276 {
1277         struct user_struct *user = prog->aux->user;
1278
1279         __bpf_prog_uncharge(user, prog->pages);
1280         free_uid(user);
1281 }
1282
1283 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1284 {
1285         int id;
1286
1287         idr_preload(GFP_KERNEL);
1288         spin_lock_bh(&prog_idr_lock);
1289         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1290         if (id > 0)
1291                 prog->aux->id = id;
1292         spin_unlock_bh(&prog_idr_lock);
1293         idr_preload_end();
1294
1295         /* id is in [1, INT_MAX) */
1296         if (WARN_ON_ONCE(!id))
1297                 return -ENOSPC;
1298
1299         return id > 0 ? 0 : id;
1300 }
1301
1302 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1303 {
1304         /* cBPF to eBPF migrations are currently not in the idr store.
1305          * Offloaded programs are removed from the store when their device
1306          * disappears - even if someone grabs an fd to them they are unusable,
1307          * simply waiting for refcnt to drop to be freed.
1308          */
1309         if (!prog->aux->id)
1310                 return;
1311
1312         if (do_idr_lock)
1313                 spin_lock_bh(&prog_idr_lock);
1314         else
1315                 __acquire(&prog_idr_lock);
1316
1317         idr_remove(&prog_idr, prog->aux->id);
1318         prog->aux->id = 0;
1319
1320         if (do_idr_lock)
1321                 spin_unlock_bh(&prog_idr_lock);
1322         else
1323                 __release(&prog_idr_lock);
1324 }
1325
1326 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1327 {
1328         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1329
1330         kvfree(aux->func_info);
1331         free_used_maps(aux);
1332         bpf_prog_uncharge_memlock(aux->prog);
1333         security_bpf_prog_free(aux);
1334         bpf_prog_free(aux->prog);
1335 }
1336
1337 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1338 {
1339         bpf_prog_kallsyms_del_all(prog);
1340         btf_put(prog->aux->btf);
1341         bpf_prog_free_linfo(prog);
1342
1343         if (deferred)
1344                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1345         else
1346                 __bpf_prog_put_rcu(&prog->aux->rcu);
1347 }
1348
1349 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1350 {
1351         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1352                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1353                 /* bpf_prog_free_id() must be called first */
1354                 bpf_prog_free_id(prog, do_idr_lock);
1355                 __bpf_prog_put_noref(prog, true);
1356         }
1357 }
1358
1359 void bpf_prog_put(struct bpf_prog *prog)
1360 {
1361         __bpf_prog_put(prog, true);
1362 }
1363 EXPORT_SYMBOL_GPL(bpf_prog_put);
1364
1365 static int bpf_prog_release(struct inode *inode, struct file *filp)
1366 {
1367         struct bpf_prog *prog = filp->private_data;
1368
1369         bpf_prog_put(prog);
1370         return 0;
1371 }
1372
1373 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1374                                struct bpf_prog_stats *stats)
1375 {
1376         u64 nsecs = 0, cnt = 0;
1377         int cpu;
1378
1379         for_each_possible_cpu(cpu) {
1380                 const struct bpf_prog_stats *st;
1381                 unsigned int start;
1382                 u64 tnsecs, tcnt;
1383
1384                 st = per_cpu_ptr(prog->aux->stats, cpu);
1385                 do {
1386                         start = u64_stats_fetch_begin_irq(&st->syncp);
1387                         tnsecs = st->nsecs;
1388                         tcnt = st->cnt;
1389                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1390                 nsecs += tnsecs;
1391                 cnt += tcnt;
1392         }
1393         stats->nsecs = nsecs;
1394         stats->cnt = cnt;
1395 }
1396
1397 #ifdef CONFIG_PROC_FS
1398 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1399 {
1400         const struct bpf_prog *prog = filp->private_data;
1401         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1402         struct bpf_prog_stats stats;
1403
1404         bpf_prog_get_stats(prog, &stats);
1405         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1406         seq_printf(m,
1407                    "prog_type:\t%u\n"
1408                    "prog_jited:\t%u\n"
1409                    "prog_tag:\t%s\n"
1410                    "memlock:\t%llu\n"
1411                    "prog_id:\t%u\n"
1412                    "run_time_ns:\t%llu\n"
1413                    "run_cnt:\t%llu\n",
1414                    prog->type,
1415                    prog->jited,
1416                    prog_tag,
1417                    prog->pages * 1ULL << PAGE_SHIFT,
1418                    prog->aux->id,
1419                    stats.nsecs,
1420                    stats.cnt);
1421 }
1422 #endif
1423
1424 const struct file_operations bpf_prog_fops = {
1425 #ifdef CONFIG_PROC_FS
1426         .show_fdinfo    = bpf_prog_show_fdinfo,
1427 #endif
1428         .release        = bpf_prog_release,
1429         .read           = bpf_dummy_read,
1430         .write          = bpf_dummy_write,
1431 };
1432
1433 int bpf_prog_new_fd(struct bpf_prog *prog)
1434 {
1435         int ret;
1436
1437         ret = security_bpf_prog(prog);
1438         if (ret < 0)
1439                 return ret;
1440
1441         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1442                                 O_RDWR | O_CLOEXEC);
1443 }
1444
1445 static struct bpf_prog *____bpf_prog_get(struct fd f)
1446 {
1447         if (!f.file)
1448                 return ERR_PTR(-EBADF);
1449         if (f.file->f_op != &bpf_prog_fops) {
1450                 fdput(f);
1451                 return ERR_PTR(-EINVAL);
1452         }
1453
1454         return f.file->private_data;
1455 }
1456
1457 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1458 {
1459         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1460                 atomic_sub(i, &prog->aux->refcnt);
1461                 return ERR_PTR(-EBUSY);
1462         }
1463         return prog;
1464 }
1465 EXPORT_SYMBOL_GPL(bpf_prog_add);
1466
1467 void bpf_prog_sub(struct bpf_prog *prog, int i)
1468 {
1469         /* Only to be used for undoing previous bpf_prog_add() in some
1470          * error path. We still know that another entity in our call
1471          * path holds a reference to the program, thus atomic_sub() can
1472          * be safely used in such cases!
1473          */
1474         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1475 }
1476 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1477
1478 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1479 {
1480         return bpf_prog_add(prog, 1);
1481 }
1482 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1483
1484 /* prog_idr_lock should have been held */
1485 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1486 {
1487         int refold;
1488
1489         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1490
1491         if (refold >= BPF_MAX_REFCNT) {
1492                 __bpf_prog_put(prog, false);
1493                 return ERR_PTR(-EBUSY);
1494         }
1495
1496         if (!refold)
1497                 return ERR_PTR(-ENOENT);
1498
1499         return prog;
1500 }
1501 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1502
1503 bool bpf_prog_get_ok(struct bpf_prog *prog,
1504                             enum bpf_prog_type *attach_type, bool attach_drv)
1505 {
1506         /* not an attachment, just a refcount inc, always allow */
1507         if (!attach_type)
1508                 return true;
1509
1510         if (prog->type != *attach_type)
1511                 return false;
1512         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1513                 return false;
1514
1515         return true;
1516 }
1517
1518 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1519                                        bool attach_drv)
1520 {
1521         struct fd f = fdget(ufd);
1522         struct bpf_prog *prog;
1523
1524         prog = ____bpf_prog_get(f);
1525         if (IS_ERR(prog))
1526                 return prog;
1527         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1528                 prog = ERR_PTR(-EINVAL);
1529                 goto out;
1530         }
1531
1532         prog = bpf_prog_inc(prog);
1533 out:
1534         fdput(f);
1535         return prog;
1536 }
1537
1538 struct bpf_prog *bpf_prog_get(u32 ufd)
1539 {
1540         return __bpf_prog_get(ufd, NULL, false);
1541 }
1542
1543 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1544                                        bool attach_drv)
1545 {
1546         return __bpf_prog_get(ufd, &type, attach_drv);
1547 }
1548 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1549
1550 /* Initially all BPF programs could be loaded w/o specifying
1551  * expected_attach_type. Later for some of them specifying expected_attach_type
1552  * at load time became required so that program could be validated properly.
1553  * Programs of types that are allowed to be loaded both w/ and w/o (for
1554  * backward compatibility) expected_attach_type, should have the default attach
1555  * type assigned to expected_attach_type for the latter case, so that it can be
1556  * validated later at attach time.
1557  *
1558  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1559  * prog type requires it but has some attach types that have to be backward
1560  * compatible.
1561  */
1562 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1563 {
1564         switch (attr->prog_type) {
1565         case BPF_PROG_TYPE_CGROUP_SOCK:
1566                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1567                  * exist so checking for non-zero is the way to go here.
1568                  */
1569                 if (!attr->expected_attach_type)
1570                         attr->expected_attach_type =
1571                                 BPF_CGROUP_INET_SOCK_CREATE;
1572                 break;
1573         }
1574 }
1575
1576 static int
1577 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1578                            enum bpf_attach_type expected_attach_type,
1579                            u32 btf_id)
1580 {
1581         switch (prog_type) {
1582         case BPF_PROG_TYPE_TRACING:
1583                 if (btf_id > BTF_MAX_TYPE)
1584                         return -EINVAL;
1585                 break;
1586         default:
1587                 if (btf_id)
1588                         return -EINVAL;
1589                 break;
1590         }
1591
1592         switch (prog_type) {
1593         case BPF_PROG_TYPE_CGROUP_SOCK:
1594                 switch (expected_attach_type) {
1595                 case BPF_CGROUP_INET_SOCK_CREATE:
1596                 case BPF_CGROUP_INET4_POST_BIND:
1597                 case BPF_CGROUP_INET6_POST_BIND:
1598                         return 0;
1599                 default:
1600                         return -EINVAL;
1601                 }
1602         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1603                 switch (expected_attach_type) {
1604                 case BPF_CGROUP_INET4_BIND:
1605                 case BPF_CGROUP_INET6_BIND:
1606                 case BPF_CGROUP_INET4_CONNECT:
1607                 case BPF_CGROUP_INET6_CONNECT:
1608                 case BPF_CGROUP_UDP4_SENDMSG:
1609                 case BPF_CGROUP_UDP6_SENDMSG:
1610                 case BPF_CGROUP_UDP4_RECVMSG:
1611                 case BPF_CGROUP_UDP6_RECVMSG:
1612                         return 0;
1613                 default:
1614                         return -EINVAL;
1615                 }
1616         case BPF_PROG_TYPE_CGROUP_SKB:
1617                 switch (expected_attach_type) {
1618                 case BPF_CGROUP_INET_INGRESS:
1619                 case BPF_CGROUP_INET_EGRESS:
1620                         return 0;
1621                 default:
1622                         return -EINVAL;
1623                 }
1624         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1625                 switch (expected_attach_type) {
1626                 case BPF_CGROUP_SETSOCKOPT:
1627                 case BPF_CGROUP_GETSOCKOPT:
1628                         return 0;
1629                 default:
1630                         return -EINVAL;
1631                 }
1632         default:
1633                 return 0;
1634         }
1635 }
1636
1637 /* last field in 'union bpf_attr' used by this command */
1638 #define BPF_PROG_LOAD_LAST_FIELD attach_btf_id
1639
1640 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1641 {
1642         enum bpf_prog_type type = attr->prog_type;
1643         struct bpf_prog *prog;
1644         int err;
1645         char license[128];
1646         bool is_gpl;
1647
1648         if (CHECK_ATTR(BPF_PROG_LOAD))
1649                 return -EINVAL;
1650
1651         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1652                                  BPF_F_ANY_ALIGNMENT |
1653                                  BPF_F_TEST_STATE_FREQ |
1654                                  BPF_F_TEST_RND_HI32))
1655                 return -EINVAL;
1656
1657         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1658             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1659             !capable(CAP_SYS_ADMIN))
1660                 return -EPERM;
1661
1662         /* copy eBPF program license from user space */
1663         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1664                               sizeof(license) - 1) < 0)
1665                 return -EFAULT;
1666         license[sizeof(license) - 1] = 0;
1667
1668         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1669         is_gpl = license_is_gpl_compatible(license);
1670
1671         if (attr->insn_cnt == 0 ||
1672             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1673                 return -E2BIG;
1674         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1675             type != BPF_PROG_TYPE_CGROUP_SKB &&
1676             !capable(CAP_SYS_ADMIN))
1677                 return -EPERM;
1678
1679         bpf_prog_load_fixup_attach_type(attr);
1680         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1681                                        attr->attach_btf_id))
1682                 return -EINVAL;
1683
1684         /* plain bpf_prog allocation */
1685         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1686         if (!prog)
1687                 return -ENOMEM;
1688
1689         prog->expected_attach_type = attr->expected_attach_type;
1690         prog->aux->attach_btf_id = attr->attach_btf_id;
1691
1692         prog->aux->offload_requested = !!attr->prog_ifindex;
1693
1694         err = security_bpf_prog_alloc(prog->aux);
1695         if (err)
1696                 goto free_prog_nouncharge;
1697
1698         err = bpf_prog_charge_memlock(prog);
1699         if (err)
1700                 goto free_prog_sec;
1701
1702         prog->len = attr->insn_cnt;
1703
1704         err = -EFAULT;
1705         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1706                            bpf_prog_insn_size(prog)) != 0)
1707                 goto free_prog;
1708
1709         prog->orig_prog = NULL;
1710         prog->jited = 0;
1711
1712         atomic_set(&prog->aux->refcnt, 1);
1713         prog->gpl_compatible = is_gpl ? 1 : 0;
1714
1715         if (bpf_prog_is_dev_bound(prog->aux)) {
1716                 err = bpf_prog_offload_init(prog, attr);
1717                 if (err)
1718                         goto free_prog;
1719         }
1720
1721         /* find program type: socket_filter vs tracing_filter */
1722         err = find_prog_type(type, prog);
1723         if (err < 0)
1724                 goto free_prog;
1725
1726         prog->aux->load_time = ktime_get_boottime_ns();
1727         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1728         if (err)
1729                 goto free_prog;
1730
1731         /* run eBPF verifier */
1732         err = bpf_check(&prog, attr, uattr);
1733         if (err < 0)
1734                 goto free_used_maps;
1735
1736         prog = bpf_prog_select_runtime(prog, &err);
1737         if (err < 0)
1738                 goto free_used_maps;
1739
1740         err = bpf_prog_alloc_id(prog);
1741         if (err)
1742                 goto free_used_maps;
1743
1744         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1745          * effectively publicly exposed. However, retrieving via
1746          * bpf_prog_get_fd_by_id() will take another reference,
1747          * therefore it cannot be gone underneath us.
1748          *
1749          * Only for the time /after/ successful bpf_prog_new_fd()
1750          * and before returning to userspace, we might just hold
1751          * one reference and any parallel close on that fd could
1752          * rip everything out. Hence, below notifications must
1753          * happen before bpf_prog_new_fd().
1754          *
1755          * Also, any failure handling from this point onwards must
1756          * be using bpf_prog_put() given the program is exposed.
1757          */
1758         bpf_prog_kallsyms_add(prog);
1759         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1760
1761         err = bpf_prog_new_fd(prog);
1762         if (err < 0)
1763                 bpf_prog_put(prog);
1764         return err;
1765
1766 free_used_maps:
1767         /* In case we have subprogs, we need to wait for a grace
1768          * period before we can tear down JIT memory since symbols
1769          * are already exposed under kallsyms.
1770          */
1771         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1772         return err;
1773 free_prog:
1774         bpf_prog_uncharge_memlock(prog);
1775 free_prog_sec:
1776         security_bpf_prog_free(prog->aux);
1777 free_prog_nouncharge:
1778         bpf_prog_free(prog);
1779         return err;
1780 }
1781
1782 #define BPF_OBJ_LAST_FIELD file_flags
1783
1784 static int bpf_obj_pin(const union bpf_attr *attr)
1785 {
1786         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1787                 return -EINVAL;
1788
1789         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1790 }
1791
1792 static int bpf_obj_get(const union bpf_attr *attr)
1793 {
1794         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1795             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1796                 return -EINVAL;
1797
1798         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1799                                 attr->file_flags);
1800 }
1801
1802 static int bpf_tracing_prog_release(struct inode *inode, struct file *filp)
1803 {
1804         struct bpf_prog *prog = filp->private_data;
1805
1806         WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1807         bpf_prog_put(prog);
1808         return 0;
1809 }
1810
1811 static const struct file_operations bpf_tracing_prog_fops = {
1812         .release        = bpf_tracing_prog_release,
1813         .read           = bpf_dummy_read,
1814         .write          = bpf_dummy_write,
1815 };
1816
1817 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
1818 {
1819         int tr_fd, err;
1820
1821         if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
1822             prog->expected_attach_type != BPF_TRACE_FEXIT) {
1823                 err = -EINVAL;
1824                 goto out_put_prog;
1825         }
1826
1827         err = bpf_trampoline_link_prog(prog);
1828         if (err)
1829                 goto out_put_prog;
1830
1831         tr_fd = anon_inode_getfd("bpf-tracing-prog", &bpf_tracing_prog_fops,
1832                                  prog, O_CLOEXEC);
1833         if (tr_fd < 0) {
1834                 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1835                 err = tr_fd;
1836                 goto out_put_prog;
1837         }
1838         return tr_fd;
1839
1840 out_put_prog:
1841         bpf_prog_put(prog);
1842         return err;
1843 }
1844
1845 struct bpf_raw_tracepoint {
1846         struct bpf_raw_event_map *btp;
1847         struct bpf_prog *prog;
1848 };
1849
1850 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1851 {
1852         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1853
1854         if (raw_tp->prog) {
1855                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1856                 bpf_prog_put(raw_tp->prog);
1857         }
1858         bpf_put_raw_tracepoint(raw_tp->btp);
1859         kfree(raw_tp);
1860         return 0;
1861 }
1862
1863 static const struct file_operations bpf_raw_tp_fops = {
1864         .release        = bpf_raw_tracepoint_release,
1865         .read           = bpf_dummy_read,
1866         .write          = bpf_dummy_write,
1867 };
1868
1869 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1870
1871 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1872 {
1873         struct bpf_raw_tracepoint *raw_tp;
1874         struct bpf_raw_event_map *btp;
1875         struct bpf_prog *prog;
1876         const char *tp_name;
1877         char buf[128];
1878         int tp_fd, err;
1879
1880         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1881                 return -EINVAL;
1882
1883         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1884         if (IS_ERR(prog))
1885                 return PTR_ERR(prog);
1886
1887         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1888             prog->type != BPF_PROG_TYPE_TRACING &&
1889             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1890                 err = -EINVAL;
1891                 goto out_put_prog;
1892         }
1893
1894         if (prog->type == BPF_PROG_TYPE_TRACING) {
1895                 if (attr->raw_tracepoint.name) {
1896                         /* The attach point for this category of programs
1897                          * should be specified via btf_id during program load.
1898                          */
1899                         err = -EINVAL;
1900                         goto out_put_prog;
1901                 }
1902                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
1903                         tp_name = prog->aux->attach_func_name;
1904                 else
1905                         return bpf_tracing_prog_attach(prog);
1906         } else {
1907                 if (strncpy_from_user(buf,
1908                                       u64_to_user_ptr(attr->raw_tracepoint.name),
1909                                       sizeof(buf) - 1) < 0) {
1910                         err = -EFAULT;
1911                         goto out_put_prog;
1912                 }
1913                 buf[sizeof(buf) - 1] = 0;
1914                 tp_name = buf;
1915         }
1916
1917         btp = bpf_get_raw_tracepoint(tp_name);
1918         if (!btp) {
1919                 err = -ENOENT;
1920                 goto out_put_prog;
1921         }
1922
1923         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1924         if (!raw_tp) {
1925                 err = -ENOMEM;
1926                 goto out_put_btp;
1927         }
1928         raw_tp->btp = btp;
1929         raw_tp->prog = prog;
1930
1931         err = bpf_probe_register(raw_tp->btp, prog);
1932         if (err)
1933                 goto out_free_tp;
1934
1935         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1936                                  O_CLOEXEC);
1937         if (tp_fd < 0) {
1938                 bpf_probe_unregister(raw_tp->btp, prog);
1939                 err = tp_fd;
1940                 goto out_free_tp;
1941         }
1942         return tp_fd;
1943
1944 out_free_tp:
1945         kfree(raw_tp);
1946 out_put_btp:
1947         bpf_put_raw_tracepoint(btp);
1948 out_put_prog:
1949         bpf_prog_put(prog);
1950         return err;
1951 }
1952
1953 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1954                                              enum bpf_attach_type attach_type)
1955 {
1956         switch (prog->type) {
1957         case BPF_PROG_TYPE_CGROUP_SOCK:
1958         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1959         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1960                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1961         case BPF_PROG_TYPE_CGROUP_SKB:
1962                 return prog->enforce_expected_attach_type &&
1963                         prog->expected_attach_type != attach_type ?
1964                         -EINVAL : 0;
1965         default:
1966                 return 0;
1967         }
1968 }
1969
1970 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1971
1972 #define BPF_F_ATTACH_MASK \
1973         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1974
1975 static int bpf_prog_attach(const union bpf_attr *attr)
1976 {
1977         enum bpf_prog_type ptype;
1978         struct bpf_prog *prog;
1979         int ret;
1980
1981         if (!capable(CAP_NET_ADMIN))
1982                 return -EPERM;
1983
1984         if (CHECK_ATTR(BPF_PROG_ATTACH))
1985                 return -EINVAL;
1986
1987         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1988                 return -EINVAL;
1989
1990         switch (attr->attach_type) {
1991         case BPF_CGROUP_INET_INGRESS:
1992         case BPF_CGROUP_INET_EGRESS:
1993                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1994                 break;
1995         case BPF_CGROUP_INET_SOCK_CREATE:
1996         case BPF_CGROUP_INET4_POST_BIND:
1997         case BPF_CGROUP_INET6_POST_BIND:
1998                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1999                 break;
2000         case BPF_CGROUP_INET4_BIND:
2001         case BPF_CGROUP_INET6_BIND:
2002         case BPF_CGROUP_INET4_CONNECT:
2003         case BPF_CGROUP_INET6_CONNECT:
2004         case BPF_CGROUP_UDP4_SENDMSG:
2005         case BPF_CGROUP_UDP6_SENDMSG:
2006         case BPF_CGROUP_UDP4_RECVMSG:
2007         case BPF_CGROUP_UDP6_RECVMSG:
2008                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2009                 break;
2010         case BPF_CGROUP_SOCK_OPS:
2011                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2012                 break;
2013         case BPF_CGROUP_DEVICE:
2014                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2015                 break;
2016         case BPF_SK_MSG_VERDICT:
2017                 ptype = BPF_PROG_TYPE_SK_MSG;
2018                 break;
2019         case BPF_SK_SKB_STREAM_PARSER:
2020         case BPF_SK_SKB_STREAM_VERDICT:
2021                 ptype = BPF_PROG_TYPE_SK_SKB;
2022                 break;
2023         case BPF_LIRC_MODE2:
2024                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
2025                 break;
2026         case BPF_FLOW_DISSECTOR:
2027                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2028                 break;
2029         case BPF_CGROUP_SYSCTL:
2030                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2031                 break;
2032         case BPF_CGROUP_GETSOCKOPT:
2033         case BPF_CGROUP_SETSOCKOPT:
2034                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2035                 break;
2036         default:
2037                 return -EINVAL;
2038         }
2039
2040         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2041         if (IS_ERR(prog))
2042                 return PTR_ERR(prog);
2043
2044         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2045                 bpf_prog_put(prog);
2046                 return -EINVAL;
2047         }
2048
2049         switch (ptype) {
2050         case BPF_PROG_TYPE_SK_SKB:
2051         case BPF_PROG_TYPE_SK_MSG:
2052                 ret = sock_map_get_from_fd(attr, prog);
2053                 break;
2054         case BPF_PROG_TYPE_LIRC_MODE2:
2055                 ret = lirc_prog_attach(attr, prog);
2056                 break;
2057         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2058                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2059                 break;
2060         default:
2061                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2062         }
2063
2064         if (ret)
2065                 bpf_prog_put(prog);
2066         return ret;
2067 }
2068
2069 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2070
2071 static int bpf_prog_detach(const union bpf_attr *attr)
2072 {
2073         enum bpf_prog_type ptype;
2074
2075         if (!capable(CAP_NET_ADMIN))
2076                 return -EPERM;
2077
2078         if (CHECK_ATTR(BPF_PROG_DETACH))
2079                 return -EINVAL;
2080
2081         switch (attr->attach_type) {
2082         case BPF_CGROUP_INET_INGRESS:
2083         case BPF_CGROUP_INET_EGRESS:
2084                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2085                 break;
2086         case BPF_CGROUP_INET_SOCK_CREATE:
2087         case BPF_CGROUP_INET4_POST_BIND:
2088         case BPF_CGROUP_INET6_POST_BIND:
2089                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2090                 break;
2091         case BPF_CGROUP_INET4_BIND:
2092         case BPF_CGROUP_INET6_BIND:
2093         case BPF_CGROUP_INET4_CONNECT:
2094         case BPF_CGROUP_INET6_CONNECT:
2095         case BPF_CGROUP_UDP4_SENDMSG:
2096         case BPF_CGROUP_UDP6_SENDMSG:
2097         case BPF_CGROUP_UDP4_RECVMSG:
2098         case BPF_CGROUP_UDP6_RECVMSG:
2099                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2100                 break;
2101         case BPF_CGROUP_SOCK_OPS:
2102                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2103                 break;
2104         case BPF_CGROUP_DEVICE:
2105                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2106                 break;
2107         case BPF_SK_MSG_VERDICT:
2108                 return sock_map_get_from_fd(attr, NULL);
2109         case BPF_SK_SKB_STREAM_PARSER:
2110         case BPF_SK_SKB_STREAM_VERDICT:
2111                 return sock_map_get_from_fd(attr, NULL);
2112         case BPF_LIRC_MODE2:
2113                 return lirc_prog_detach(attr);
2114         case BPF_FLOW_DISSECTOR:
2115                 return skb_flow_dissector_bpf_prog_detach(attr);
2116         case BPF_CGROUP_SYSCTL:
2117                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2118                 break;
2119         case BPF_CGROUP_GETSOCKOPT:
2120         case BPF_CGROUP_SETSOCKOPT:
2121                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2122                 break;
2123         default:
2124                 return -EINVAL;
2125         }
2126
2127         return cgroup_bpf_prog_detach(attr, ptype);
2128 }
2129
2130 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2131
2132 static int bpf_prog_query(const union bpf_attr *attr,
2133                           union bpf_attr __user *uattr)
2134 {
2135         if (!capable(CAP_NET_ADMIN))
2136                 return -EPERM;
2137         if (CHECK_ATTR(BPF_PROG_QUERY))
2138                 return -EINVAL;
2139         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2140                 return -EINVAL;
2141
2142         switch (attr->query.attach_type) {
2143         case BPF_CGROUP_INET_INGRESS:
2144         case BPF_CGROUP_INET_EGRESS:
2145         case BPF_CGROUP_INET_SOCK_CREATE:
2146         case BPF_CGROUP_INET4_BIND:
2147         case BPF_CGROUP_INET6_BIND:
2148         case BPF_CGROUP_INET4_POST_BIND:
2149         case BPF_CGROUP_INET6_POST_BIND:
2150         case BPF_CGROUP_INET4_CONNECT:
2151         case BPF_CGROUP_INET6_CONNECT:
2152         case BPF_CGROUP_UDP4_SENDMSG:
2153         case BPF_CGROUP_UDP6_SENDMSG:
2154         case BPF_CGROUP_UDP4_RECVMSG:
2155         case BPF_CGROUP_UDP6_RECVMSG:
2156         case BPF_CGROUP_SOCK_OPS:
2157         case BPF_CGROUP_DEVICE:
2158         case BPF_CGROUP_SYSCTL:
2159         case BPF_CGROUP_GETSOCKOPT:
2160         case BPF_CGROUP_SETSOCKOPT:
2161                 break;
2162         case BPF_LIRC_MODE2:
2163                 return lirc_prog_query(attr, uattr);
2164         case BPF_FLOW_DISSECTOR:
2165                 return skb_flow_dissector_prog_query(attr, uattr);
2166         default:
2167                 return -EINVAL;
2168         }
2169
2170         return cgroup_bpf_prog_query(attr, uattr);
2171 }
2172
2173 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2174
2175 static int bpf_prog_test_run(const union bpf_attr *attr,
2176                              union bpf_attr __user *uattr)
2177 {
2178         struct bpf_prog *prog;
2179         int ret = -ENOTSUPP;
2180
2181         if (!capable(CAP_SYS_ADMIN))
2182                 return -EPERM;
2183         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2184                 return -EINVAL;
2185
2186         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2187             (!attr->test.ctx_size_in && attr->test.ctx_in))
2188                 return -EINVAL;
2189
2190         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2191             (!attr->test.ctx_size_out && attr->test.ctx_out))
2192                 return -EINVAL;
2193
2194         prog = bpf_prog_get(attr->test.prog_fd);
2195         if (IS_ERR(prog))
2196                 return PTR_ERR(prog);
2197
2198         if (prog->aux->ops->test_run)
2199                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2200
2201         bpf_prog_put(prog);
2202         return ret;
2203 }
2204
2205 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2206
2207 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2208                                union bpf_attr __user *uattr,
2209                                struct idr *idr,
2210                                spinlock_t *lock)
2211 {
2212         u32 next_id = attr->start_id;
2213         int err = 0;
2214
2215         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2216                 return -EINVAL;
2217
2218         if (!capable(CAP_SYS_ADMIN))
2219                 return -EPERM;
2220
2221         next_id++;
2222         spin_lock_bh(lock);
2223         if (!idr_get_next(idr, &next_id))
2224                 err = -ENOENT;
2225         spin_unlock_bh(lock);
2226
2227         if (!err)
2228                 err = put_user(next_id, &uattr->next_id);
2229
2230         return err;
2231 }
2232
2233 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2234
2235 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2236 {
2237         struct bpf_prog *prog;
2238         u32 id = attr->prog_id;
2239         int fd;
2240
2241         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2242                 return -EINVAL;
2243
2244         if (!capable(CAP_SYS_ADMIN))
2245                 return -EPERM;
2246
2247         spin_lock_bh(&prog_idr_lock);
2248         prog = idr_find(&prog_idr, id);
2249         if (prog)
2250                 prog = bpf_prog_inc_not_zero(prog);
2251         else
2252                 prog = ERR_PTR(-ENOENT);
2253         spin_unlock_bh(&prog_idr_lock);
2254
2255         if (IS_ERR(prog))
2256                 return PTR_ERR(prog);
2257
2258         fd = bpf_prog_new_fd(prog);
2259         if (fd < 0)
2260                 bpf_prog_put(prog);
2261
2262         return fd;
2263 }
2264
2265 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2266
2267 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2268 {
2269         struct bpf_map *map;
2270         u32 id = attr->map_id;
2271         int f_flags;
2272         int fd;
2273
2274         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2275             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2276                 return -EINVAL;
2277
2278         if (!capable(CAP_SYS_ADMIN))
2279                 return -EPERM;
2280
2281         f_flags = bpf_get_file_flag(attr->open_flags);
2282         if (f_flags < 0)
2283                 return f_flags;
2284
2285         spin_lock_bh(&map_idr_lock);
2286         map = idr_find(&map_idr, id);
2287         if (map)
2288                 map = __bpf_map_inc_not_zero(map, true);
2289         else
2290                 map = ERR_PTR(-ENOENT);
2291         spin_unlock_bh(&map_idr_lock);
2292
2293         if (IS_ERR(map))
2294                 return PTR_ERR(map);
2295
2296         fd = bpf_map_new_fd(map, f_flags);
2297         if (fd < 0)
2298                 bpf_map_put_with_uref(map);
2299
2300         return fd;
2301 }
2302
2303 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2304                                               unsigned long addr, u32 *off,
2305                                               u32 *type)
2306 {
2307         const struct bpf_map *map;
2308         int i;
2309
2310         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2311                 map = prog->aux->used_maps[i];
2312                 if (map == (void *)addr) {
2313                         *type = BPF_PSEUDO_MAP_FD;
2314                         return map;
2315                 }
2316                 if (!map->ops->map_direct_value_meta)
2317                         continue;
2318                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2319                         *type = BPF_PSEUDO_MAP_VALUE;
2320                         return map;
2321                 }
2322         }
2323
2324         return NULL;
2325 }
2326
2327 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2328 {
2329         const struct bpf_map *map;
2330         struct bpf_insn *insns;
2331         u32 off, type;
2332         u64 imm;
2333         int i;
2334
2335         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2336                         GFP_USER);
2337         if (!insns)
2338                 return insns;
2339
2340         for (i = 0; i < prog->len; i++) {
2341                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2342                         insns[i].code = BPF_JMP | BPF_CALL;
2343                         insns[i].imm = BPF_FUNC_tail_call;
2344                         /* fall-through */
2345                 }
2346                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2347                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2348                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2349                                 insns[i].code = BPF_JMP | BPF_CALL;
2350                         if (!bpf_dump_raw_ok())
2351                                 insns[i].imm = 0;
2352                         continue;
2353                 }
2354
2355                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2356                         continue;
2357
2358                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2359                 map = bpf_map_from_imm(prog, imm, &off, &type);
2360                 if (map) {
2361                         insns[i].src_reg = type;
2362                         insns[i].imm = map->id;
2363                         insns[i + 1].imm = off;
2364                         continue;
2365                 }
2366         }
2367
2368         return insns;
2369 }
2370
2371 static int set_info_rec_size(struct bpf_prog_info *info)
2372 {
2373         /*
2374          * Ensure info.*_rec_size is the same as kernel expected size
2375          *
2376          * or
2377          *
2378          * Only allow zero *_rec_size if both _rec_size and _cnt are
2379          * zero.  In this case, the kernel will set the expected
2380          * _rec_size back to the info.
2381          */
2382
2383         if ((info->nr_func_info || info->func_info_rec_size) &&
2384             info->func_info_rec_size != sizeof(struct bpf_func_info))
2385                 return -EINVAL;
2386
2387         if ((info->nr_line_info || info->line_info_rec_size) &&
2388             info->line_info_rec_size != sizeof(struct bpf_line_info))
2389                 return -EINVAL;
2390
2391         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2392             info->jited_line_info_rec_size != sizeof(__u64))
2393                 return -EINVAL;
2394
2395         info->func_info_rec_size = sizeof(struct bpf_func_info);
2396         info->line_info_rec_size = sizeof(struct bpf_line_info);
2397         info->jited_line_info_rec_size = sizeof(__u64);
2398
2399         return 0;
2400 }
2401
2402 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2403                                    const union bpf_attr *attr,
2404                                    union bpf_attr __user *uattr)
2405 {
2406         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2407         struct bpf_prog_info info = {};
2408         u32 info_len = attr->info.info_len;
2409         struct bpf_prog_stats stats;
2410         char __user *uinsns;
2411         u32 ulen;
2412         int err;
2413
2414         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2415         if (err)
2416                 return err;
2417         info_len = min_t(u32, sizeof(info), info_len);
2418
2419         if (copy_from_user(&info, uinfo, info_len))
2420                 return -EFAULT;
2421
2422         info.type = prog->type;
2423         info.id = prog->aux->id;
2424         info.load_time = prog->aux->load_time;
2425         info.created_by_uid = from_kuid_munged(current_user_ns(),
2426                                                prog->aux->user->uid);
2427         info.gpl_compatible = prog->gpl_compatible;
2428
2429         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2430         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2431
2432         ulen = info.nr_map_ids;
2433         info.nr_map_ids = prog->aux->used_map_cnt;
2434         ulen = min_t(u32, info.nr_map_ids, ulen);
2435         if (ulen) {
2436                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2437                 u32 i;
2438
2439                 for (i = 0; i < ulen; i++)
2440                         if (put_user(prog->aux->used_maps[i]->id,
2441                                      &user_map_ids[i]))
2442                                 return -EFAULT;
2443         }
2444
2445         err = set_info_rec_size(&info);
2446         if (err)
2447                 return err;
2448
2449         bpf_prog_get_stats(prog, &stats);
2450         info.run_time_ns = stats.nsecs;
2451         info.run_cnt = stats.cnt;
2452
2453         if (!capable(CAP_SYS_ADMIN)) {
2454                 info.jited_prog_len = 0;
2455                 info.xlated_prog_len = 0;
2456                 info.nr_jited_ksyms = 0;
2457                 info.nr_jited_func_lens = 0;
2458                 info.nr_func_info = 0;
2459                 info.nr_line_info = 0;
2460                 info.nr_jited_line_info = 0;
2461                 goto done;
2462         }
2463
2464         ulen = info.xlated_prog_len;
2465         info.xlated_prog_len = bpf_prog_insn_size(prog);
2466         if (info.xlated_prog_len && ulen) {
2467                 struct bpf_insn *insns_sanitized;
2468                 bool fault;
2469
2470                 if (prog->blinded && !bpf_dump_raw_ok()) {
2471                         info.xlated_prog_insns = 0;
2472                         goto done;
2473                 }
2474                 insns_sanitized = bpf_insn_prepare_dump(prog);
2475                 if (!insns_sanitized)
2476                         return -ENOMEM;
2477                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2478                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2479                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2480                 kfree(insns_sanitized);
2481                 if (fault)
2482                         return -EFAULT;
2483         }
2484
2485         if (bpf_prog_is_dev_bound(prog->aux)) {
2486                 err = bpf_prog_offload_info_fill(&info, prog);
2487                 if (err)
2488                         return err;
2489                 goto done;
2490         }
2491
2492         /* NOTE: the following code is supposed to be skipped for offload.
2493          * bpf_prog_offload_info_fill() is the place to fill similar fields
2494          * for offload.
2495          */
2496         ulen = info.jited_prog_len;
2497         if (prog->aux->func_cnt) {
2498                 u32 i;
2499
2500                 info.jited_prog_len = 0;
2501                 for (i = 0; i < prog->aux->func_cnt; i++)
2502                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2503         } else {
2504                 info.jited_prog_len = prog->jited_len;
2505         }
2506
2507         if (info.jited_prog_len && ulen) {
2508                 if (bpf_dump_raw_ok()) {
2509                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2510                         ulen = min_t(u32, info.jited_prog_len, ulen);
2511
2512                         /* for multi-function programs, copy the JITed
2513                          * instructions for all the functions
2514                          */
2515                         if (prog->aux->func_cnt) {
2516                                 u32 len, free, i;
2517                                 u8 *img;
2518
2519                                 free = ulen;
2520                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2521                                         len = prog->aux->func[i]->jited_len;
2522                                         len = min_t(u32, len, free);
2523                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2524                                         if (copy_to_user(uinsns, img, len))
2525                                                 return -EFAULT;
2526                                         uinsns += len;
2527                                         free -= len;
2528                                         if (!free)
2529                                                 break;
2530                                 }
2531                         } else {
2532                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2533                                         return -EFAULT;
2534                         }
2535                 } else {
2536                         info.jited_prog_insns = 0;
2537                 }
2538         }
2539
2540         ulen = info.nr_jited_ksyms;
2541         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2542         if (ulen) {
2543                 if (bpf_dump_raw_ok()) {
2544                         unsigned long ksym_addr;
2545                         u64 __user *user_ksyms;
2546                         u32 i;
2547
2548                         /* copy the address of the kernel symbol
2549                          * corresponding to each function
2550                          */
2551                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2552                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2553                         if (prog->aux->func_cnt) {
2554                                 for (i = 0; i < ulen; i++) {
2555                                         ksym_addr = (unsigned long)
2556                                                 prog->aux->func[i]->bpf_func;
2557                                         if (put_user((u64) ksym_addr,
2558                                                      &user_ksyms[i]))
2559                                                 return -EFAULT;
2560                                 }
2561                         } else {
2562                                 ksym_addr = (unsigned long) prog->bpf_func;
2563                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2564                                         return -EFAULT;
2565                         }
2566                 } else {
2567                         info.jited_ksyms = 0;
2568                 }
2569         }
2570
2571         ulen = info.nr_jited_func_lens;
2572         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2573         if (ulen) {
2574                 if (bpf_dump_raw_ok()) {
2575                         u32 __user *user_lens;
2576                         u32 func_len, i;
2577
2578                         /* copy the JITed image lengths for each function */
2579                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2580                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2581                         if (prog->aux->func_cnt) {
2582                                 for (i = 0; i < ulen; i++) {
2583                                         func_len =
2584                                                 prog->aux->func[i]->jited_len;
2585                                         if (put_user(func_len, &user_lens[i]))
2586                                                 return -EFAULT;
2587                                 }
2588                         } else {
2589                                 func_len = prog->jited_len;
2590                                 if (put_user(func_len, &user_lens[0]))
2591                                         return -EFAULT;
2592                         }
2593                 } else {
2594                         info.jited_func_lens = 0;
2595                 }
2596         }
2597
2598         if (prog->aux->btf)
2599                 info.btf_id = btf_id(prog->aux->btf);
2600
2601         ulen = info.nr_func_info;
2602         info.nr_func_info = prog->aux->func_info_cnt;
2603         if (info.nr_func_info && ulen) {
2604                 char __user *user_finfo;
2605
2606                 user_finfo = u64_to_user_ptr(info.func_info);
2607                 ulen = min_t(u32, info.nr_func_info, ulen);
2608                 if (copy_to_user(user_finfo, prog->aux->func_info,
2609                                  info.func_info_rec_size * ulen))
2610                         return -EFAULT;
2611         }
2612
2613         ulen = info.nr_line_info;
2614         info.nr_line_info = prog->aux->nr_linfo;
2615         if (info.nr_line_info && ulen) {
2616                 __u8 __user *user_linfo;
2617
2618                 user_linfo = u64_to_user_ptr(info.line_info);
2619                 ulen = min_t(u32, info.nr_line_info, ulen);
2620                 if (copy_to_user(user_linfo, prog->aux->linfo,
2621                                  info.line_info_rec_size * ulen))
2622                         return -EFAULT;
2623         }
2624
2625         ulen = info.nr_jited_line_info;
2626         if (prog->aux->jited_linfo)
2627                 info.nr_jited_line_info = prog->aux->nr_linfo;
2628         else
2629                 info.nr_jited_line_info = 0;
2630         if (info.nr_jited_line_info && ulen) {
2631                 if (bpf_dump_raw_ok()) {
2632                         __u64 __user *user_linfo;
2633                         u32 i;
2634
2635                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2636                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2637                         for (i = 0; i < ulen; i++) {
2638                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2639                                              &user_linfo[i]))
2640                                         return -EFAULT;
2641                         }
2642                 } else {
2643                         info.jited_line_info = 0;
2644                 }
2645         }
2646
2647         ulen = info.nr_prog_tags;
2648         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2649         if (ulen) {
2650                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2651                 u32 i;
2652
2653                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2654                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2655                 if (prog->aux->func_cnt) {
2656                         for (i = 0; i < ulen; i++) {
2657                                 if (copy_to_user(user_prog_tags[i],
2658                                                  prog->aux->func[i]->tag,
2659                                                  BPF_TAG_SIZE))
2660                                         return -EFAULT;
2661                         }
2662                 } else {
2663                         if (copy_to_user(user_prog_tags[0],
2664                                          prog->tag, BPF_TAG_SIZE))
2665                                 return -EFAULT;
2666                 }
2667         }
2668
2669 done:
2670         if (copy_to_user(uinfo, &info, info_len) ||
2671             put_user(info_len, &uattr->info.info_len))
2672                 return -EFAULT;
2673
2674         return 0;
2675 }
2676
2677 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2678                                   const union bpf_attr *attr,
2679                                   union bpf_attr __user *uattr)
2680 {
2681         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2682         struct bpf_map_info info = {};
2683         u32 info_len = attr->info.info_len;
2684         int err;
2685
2686         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2687         if (err)
2688                 return err;
2689         info_len = min_t(u32, sizeof(info), info_len);
2690
2691         info.type = map->map_type;
2692         info.id = map->id;
2693         info.key_size = map->key_size;
2694         info.value_size = map->value_size;
2695         info.max_entries = map->max_entries;
2696         info.map_flags = map->map_flags;
2697         memcpy(info.name, map->name, sizeof(map->name));
2698
2699         if (map->btf) {
2700                 info.btf_id = btf_id(map->btf);
2701                 info.btf_key_type_id = map->btf_key_type_id;
2702                 info.btf_value_type_id = map->btf_value_type_id;
2703         }
2704
2705         if (bpf_map_is_dev_bound(map)) {
2706                 err = bpf_map_offload_info_fill(&info, map);
2707                 if (err)
2708                         return err;
2709         }
2710
2711         if (copy_to_user(uinfo, &info, info_len) ||
2712             put_user(info_len, &uattr->info.info_len))
2713                 return -EFAULT;
2714
2715         return 0;
2716 }
2717
2718 static int bpf_btf_get_info_by_fd(struct btf *btf,
2719                                   const union bpf_attr *attr,
2720                                   union bpf_attr __user *uattr)
2721 {
2722         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2723         u32 info_len = attr->info.info_len;
2724         int err;
2725
2726         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2727         if (err)
2728                 return err;
2729
2730         return btf_get_info_by_fd(btf, attr, uattr);
2731 }
2732
2733 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2734
2735 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2736                                   union bpf_attr __user *uattr)
2737 {
2738         int ufd = attr->info.bpf_fd;
2739         struct fd f;
2740         int err;
2741
2742         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2743                 return -EINVAL;
2744
2745         f = fdget(ufd);
2746         if (!f.file)
2747                 return -EBADFD;
2748
2749         if (f.file->f_op == &bpf_prog_fops)
2750                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2751                                               uattr);
2752         else if (f.file->f_op == &bpf_map_fops)
2753                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2754                                              uattr);
2755         else if (f.file->f_op == &btf_fops)
2756                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2757         else
2758                 err = -EINVAL;
2759
2760         fdput(f);
2761         return err;
2762 }
2763
2764 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2765
2766 static int bpf_btf_load(const union bpf_attr *attr)
2767 {
2768         if (CHECK_ATTR(BPF_BTF_LOAD))
2769                 return -EINVAL;
2770
2771         if (!capable(CAP_SYS_ADMIN))
2772                 return -EPERM;
2773
2774         return btf_new_fd(attr);
2775 }
2776
2777 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2778
2779 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2780 {
2781         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2782                 return -EINVAL;
2783
2784         if (!capable(CAP_SYS_ADMIN))
2785                 return -EPERM;
2786
2787         return btf_get_fd_by_id(attr->btf_id);
2788 }
2789
2790 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2791                                     union bpf_attr __user *uattr,
2792                                     u32 prog_id, u32 fd_type,
2793                                     const char *buf, u64 probe_offset,
2794                                     u64 probe_addr)
2795 {
2796         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2797         u32 len = buf ? strlen(buf) : 0, input_len;
2798         int err = 0;
2799
2800         if (put_user(len, &uattr->task_fd_query.buf_len))
2801                 return -EFAULT;
2802         input_len = attr->task_fd_query.buf_len;
2803         if (input_len && ubuf) {
2804                 if (!len) {
2805                         /* nothing to copy, just make ubuf NULL terminated */
2806                         char zero = '\0';
2807
2808                         if (put_user(zero, ubuf))
2809                                 return -EFAULT;
2810                 } else if (input_len >= len + 1) {
2811                         /* ubuf can hold the string with NULL terminator */
2812                         if (copy_to_user(ubuf, buf, len + 1))
2813                                 return -EFAULT;
2814                 } else {
2815                         /* ubuf cannot hold the string with NULL terminator,
2816                          * do a partial copy with NULL terminator.
2817                          */
2818                         char zero = '\0';
2819
2820                         err = -ENOSPC;
2821                         if (copy_to_user(ubuf, buf, input_len - 1))
2822                                 return -EFAULT;
2823                         if (put_user(zero, ubuf + input_len - 1))
2824                                 return -EFAULT;
2825                 }
2826         }
2827
2828         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2829             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2830             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2831             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2832                 return -EFAULT;
2833
2834         return err;
2835 }
2836
2837 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2838
2839 static int bpf_task_fd_query(const union bpf_attr *attr,
2840                              union bpf_attr __user *uattr)
2841 {
2842         pid_t pid = attr->task_fd_query.pid;
2843         u32 fd = attr->task_fd_query.fd;
2844         const struct perf_event *event;
2845         struct files_struct *files;
2846         struct task_struct *task;
2847         struct file *file;
2848         int err;
2849
2850         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2851                 return -EINVAL;
2852
2853         if (!capable(CAP_SYS_ADMIN))
2854                 return -EPERM;
2855
2856         if (attr->task_fd_query.flags != 0)
2857                 return -EINVAL;
2858
2859         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2860         if (!task)
2861                 return -ENOENT;
2862
2863         files = get_files_struct(task);
2864         put_task_struct(task);
2865         if (!files)
2866                 return -ENOENT;
2867
2868         err = 0;
2869         spin_lock(&files->file_lock);
2870         file = fcheck_files(files, fd);
2871         if (!file)
2872                 err = -EBADF;
2873         else
2874                 get_file(file);
2875         spin_unlock(&files->file_lock);
2876         put_files_struct(files);
2877
2878         if (err)
2879                 goto out;
2880
2881         if (file->f_op == &bpf_raw_tp_fops) {
2882                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2883                 struct bpf_raw_event_map *btp = raw_tp->btp;
2884
2885                 err = bpf_task_fd_query_copy(attr, uattr,
2886                                              raw_tp->prog->aux->id,
2887                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2888                                              btp->tp->name, 0, 0);
2889                 goto put_file;
2890         }
2891
2892         event = perf_get_event(file);
2893         if (!IS_ERR(event)) {
2894                 u64 probe_offset, probe_addr;
2895                 u32 prog_id, fd_type;
2896                 const char *buf;
2897
2898                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2899                                               &buf, &probe_offset,
2900                                               &probe_addr);
2901                 if (!err)
2902                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2903                                                      fd_type, buf,
2904                                                      probe_offset,
2905                                                      probe_addr);
2906                 goto put_file;
2907         }
2908
2909         err = -ENOTSUPP;
2910 put_file:
2911         fput(file);
2912 out:
2913         return err;
2914 }
2915
2916 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2917 {
2918         union bpf_attr attr = {};
2919         int err;
2920
2921         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2922                 return -EPERM;
2923
2924         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2925         if (err)
2926                 return err;
2927         size = min_t(u32, size, sizeof(attr));
2928
2929         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2930         if (copy_from_user(&attr, uattr, size) != 0)
2931                 return -EFAULT;
2932
2933         err = security_bpf(cmd, &attr, size);
2934         if (err < 0)
2935                 return err;
2936
2937         switch (cmd) {
2938         case BPF_MAP_CREATE:
2939                 err = map_create(&attr);
2940                 break;
2941         case BPF_MAP_LOOKUP_ELEM:
2942                 err = map_lookup_elem(&attr);
2943                 break;
2944         case BPF_MAP_UPDATE_ELEM:
2945                 err = map_update_elem(&attr);
2946                 break;
2947         case BPF_MAP_DELETE_ELEM:
2948                 err = map_delete_elem(&attr);
2949                 break;
2950         case BPF_MAP_GET_NEXT_KEY:
2951                 err = map_get_next_key(&attr);
2952                 break;
2953         case BPF_MAP_FREEZE:
2954                 err = map_freeze(&attr);
2955                 break;
2956         case BPF_PROG_LOAD:
2957                 err = bpf_prog_load(&attr, uattr);
2958                 break;
2959         case BPF_OBJ_PIN:
2960                 err = bpf_obj_pin(&attr);
2961                 break;
2962         case BPF_OBJ_GET:
2963                 err = bpf_obj_get(&attr);
2964                 break;
2965         case BPF_PROG_ATTACH:
2966                 err = bpf_prog_attach(&attr);
2967                 break;
2968         case BPF_PROG_DETACH:
2969                 err = bpf_prog_detach(&attr);
2970                 break;
2971         case BPF_PROG_QUERY:
2972                 err = bpf_prog_query(&attr, uattr);
2973                 break;
2974         case BPF_PROG_TEST_RUN:
2975                 err = bpf_prog_test_run(&attr, uattr);
2976                 break;
2977         case BPF_PROG_GET_NEXT_ID:
2978                 err = bpf_obj_get_next_id(&attr, uattr,
2979                                           &prog_idr, &prog_idr_lock);
2980                 break;
2981         case BPF_MAP_GET_NEXT_ID:
2982                 err = bpf_obj_get_next_id(&attr, uattr,
2983                                           &map_idr, &map_idr_lock);
2984                 break;
2985         case BPF_BTF_GET_NEXT_ID:
2986                 err = bpf_obj_get_next_id(&attr, uattr,
2987                                           &btf_idr, &btf_idr_lock);
2988                 break;
2989         case BPF_PROG_GET_FD_BY_ID:
2990                 err = bpf_prog_get_fd_by_id(&attr);
2991                 break;
2992         case BPF_MAP_GET_FD_BY_ID:
2993                 err = bpf_map_get_fd_by_id(&attr);
2994                 break;
2995         case BPF_OBJ_GET_INFO_BY_FD:
2996                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2997                 break;
2998         case BPF_RAW_TRACEPOINT_OPEN:
2999                 err = bpf_raw_tracepoint_open(&attr);
3000                 break;
3001         case BPF_BTF_LOAD:
3002                 err = bpf_btf_load(&attr);
3003                 break;
3004         case BPF_BTF_GET_FD_BY_ID:
3005                 err = bpf_btf_get_fd_by_id(&attr);
3006                 break;
3007         case BPF_TASK_FD_QUERY:
3008                 err = bpf_task_fd_query(&attr, uattr);
3009                 break;
3010         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3011                 err = map_lookup_and_delete_elem(&attr);
3012                 break;
3013         default:
3014                 err = -EINVAL;
3015                 break;
3016         }
3017
3018         return err;
3019 }