kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26
  27 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  28                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  29                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  30                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  31 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  32 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  33
  34 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  35
  36 DEFINE_PER_CPU(int, bpf_prog_active);
  37 static DEFINE_IDR(prog_idr);
  38 static DEFINE_SPINLOCK(prog_idr_lock);
  39 static DEFINE_IDR(map_idr);
  40 static DEFINE_SPINLOCK(map_idr_lock);
  41
  42 int sysctl_unprivileged_bpf_disabled __read_mostly;
  43
  44 static const struct bpf_map_ops * const bpf_map_types[] = {
  45 #define BPF_PROG_TYPE(_id, _ops)
  46 #define BPF_MAP_TYPE(_id, _ops) \
  47         [_id] = &_ops,
  48 #include <linux/bpf_types.h>
  49 #undef BPF_PROG_TYPE
  50 #undef BPF_MAP_TYPE
  51 };
  52
  53 /*
  54  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  55  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  56  * we don't know about yet.
  57  *
  58  * There is a ToCToU between this function call and the following
  59  * copy_from_user() call. However, this is not a concern since this function is
  60  * meant to be a future-proofing of bits.
  61  */
  62 int bpf_check_uarg_tail_zero(void __user *uaddr,
  63                              size_t expected_size,
  64                              size_t actual_size)
  65 {
  66         unsigned char __user *addr;
  67         unsigned char __user *end;
  68         unsigned char val;
  69         int err;
  70
  71         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  72                 return -E2BIG;
  73
  74         if (unlikely(!access_ok(uaddr, actual_size)))
  75                 return -EFAULT;
  76
  77         if (actual_size <= expected_size)
  78                 return 0;
  79
  80         addr = uaddr + expected_size;
  81         end  = uaddr + actual_size;
  82
  83         for (; addr < end; addr++) {
  84                 err = get_user(val, addr);
  85                 if (err)
  86                         return err;
  87                 if (val)
  88                         return -E2BIG;
  89         }
  90
  91         return 0;
  92 }
  93
  94 const struct bpf_map_ops bpf_map_offload_ops = {
  95         .map_alloc = bpf_map_offload_map_alloc,
  96         .map_free = bpf_map_offload_map_free,
  97         .map_check_btf = map_check_no_btf,
  98 };
  99
 100 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 101 {
 102         const struct bpf_map_ops *ops;
 103         u32 type = attr->map_type;
 104         struct bpf_map *map;
 105         int err;
 106
 107         if (type >= ARRAY_SIZE(bpf_map_types))
 108                 return ERR_PTR(-EINVAL);
 109         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 110         ops = bpf_map_types[type];
 111         if (!ops)
 112                 return ERR_PTR(-EINVAL);
 113
 114         if (ops->map_alloc_check) {
 115                 err = ops->map_alloc_check(attr);
 116                 if (err)
 117                         return ERR_PTR(err);
 118         }
 119         if (attr->map_ifindex)
 120                 ops = &bpf_map_offload_ops;
 121         map = ops->map_alloc(attr);
 122         if (IS_ERR(map))
 123                 return map;
 124         map->ops = ops;
 125         map->map_type = type;
 126         return map;
 127 }
 128
 129 void *bpf_map_area_alloc(size_t size, int numa_node)
 130 {
 131         /* We really just want to fail instead of triggering OOM killer
 132          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 133          * which is used for lower order allocation requests.
 134          *
 135          * It has been observed that higher order allocation requests done by
 136          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 137          * to reclaim memory from the page cache, thus we set
 138          * __GFP_RETRY_MAYFAIL to avoid such situations.
 139          */
 140
 141         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 142         void *area;
 143
 144         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 145                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 146                                     numa_node);
 147                 if (area != NULL)
 148                         return area;
 149         }
 150
 151         return __vmalloc_node_flags_caller(size, numa_node,
 152                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 153                                            flags, __builtin_return_address(0));
 154 }
 155
 156 void bpf_map_area_free(void *area)
 157 {
 158         kvfree(area);
 159 }
 160
 161 static u32 bpf_map_flags_retain_permanent(u32 flags)
 162 {
 163         /* Some map creation flags are not tied to the map object but
 164          * rather to the map fd instead, so they have no meaning upon
 165          * map object inspection since multiple file descriptors with
 166          * different (access) properties can exist here. Thus, given
 167          * this has zero meaning for the map itself, lets clear these
 168          * from here.
 169          */
 170         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 171 }
 172
 173 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 174 {
 175         map->map_type = attr->map_type;
 176         map->key_size = attr->key_size;
 177         map->value_size = attr->value_size;
 178         map->max_entries = attr->max_entries;
 179         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 180         map->numa_node = bpf_map_attr_numa_node(attr);
 181 }
 182
 183 int bpf_map_precharge_memlock(u32 pages)
 184 {
 185         struct user_struct *user = get_current_user();
 186         unsigned long memlock_limit, cur;
 187
 188         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 189         cur = atomic_long_read(&user->locked_vm);
 190         free_uid(user);
 191         if (cur + pages > memlock_limit)
 192                 return -EPERM;
 193         return 0;
 194 }
 195
 196 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 197 {
 198         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 199
 200         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 201                 atomic_long_sub(pages, &user->locked_vm);
 202                 return -EPERM;
 203         }
 204         return 0;
 205 }
 206
 207 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 208 {
 209         atomic_long_sub(pages, &user->locked_vm);
 210 }
 211
 212 static int bpf_map_init_memlock(struct bpf_map *map)
 213 {
 214         struct user_struct *user = get_current_user();
 215         int ret;
 216
 217         ret = bpf_charge_memlock(user, map->pages);
 218         if (ret) {
 219                 free_uid(user);
 220                 return ret;
 221         }
 222         map->user = user;
 223         return ret;
 224 }
 225
 226 static void bpf_map_release_memlock(struct bpf_map *map)
 227 {
 228         struct user_struct *user = map->user;
 229         bpf_uncharge_memlock(user, map->pages);
 230         free_uid(user);
 231 }
 232
 233 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 234 {
 235         int ret;
 236
 237         ret = bpf_charge_memlock(map->user, pages);
 238         if (ret)
 239                 return ret;
 240         map->pages += pages;
 241         return ret;
 242 }
 243
 244 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 245 {
 246         bpf_uncharge_memlock(map->user, pages);
 247         map->pages -= pages;
 248 }
 249
 250 static int bpf_map_alloc_id(struct bpf_map *map)
 251 {
 252         int id;
 253
 254         idr_preload(GFP_KERNEL);
 255         spin_lock_bh(&map_idr_lock);
 256         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 257         if (id > 0)
 258                 map->id = id;
 259         spin_unlock_bh(&map_idr_lock);
 260         idr_preload_end();
 261
 262         if (WARN_ON_ONCE(!id))
 263                 return -ENOSPC;
 264
 265         return id > 0 ? 0 : id;
 266 }
 267
 268 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 269 {
 270         unsigned long flags;
 271
 272         /* Offloaded maps are removed from the IDR store when their device
 273          * disappears - even if someone holds an fd to them they are unusable,
 274          * the memory is gone, all ops will fail; they are simply waiting for
 275          * refcnt to drop to be freed.
 276          */
 277         if (!map->id)
 278                 return;
 279
 280         if (do_idr_lock)
 281                 spin_lock_irqsave(&map_idr_lock, flags);
 282         else
 283                 __acquire(&map_idr_lock);
 284
 285         idr_remove(&map_idr, map->id);
 286         map->id = 0;
 287
 288         if (do_idr_lock)
 289                 spin_unlock_irqrestore(&map_idr_lock, flags);
 290         else
 291                 __release(&map_idr_lock);
 292 }
 293
 294 /* called from workqueue */
 295 static void bpf_map_free_deferred(struct work_struct *work)
 296 {
 297         struct bpf_map *map = container_of(work, struct bpf_map, work);
 298
 299         bpf_map_release_memlock(map);
 300         security_bpf_map_free(map);
 301         /* implementation dependent freeing */
 302         map->ops->map_free(map);
 303 }
 304
 305 static void bpf_map_put_uref(struct bpf_map *map)
 306 {
 307         if (atomic_dec_and_test(&map->usercnt)) {
 308                 if (map->ops->map_release_uref)
 309                         map->ops->map_release_uref(map);
 310         }
 311 }
 312
 313 /* decrement map refcnt and schedule it for freeing via workqueue
 314  * (unrelying map implementation ops->map_free() might sleep)
 315  */
 316 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 317 {
 318         if (atomic_dec_and_test(&map->refcnt)) {
 319                 /* bpf_map_free_id() must be called first */
 320                 bpf_map_free_id(map, do_idr_lock);
 321                 btf_put(map->btf);
 322                 INIT_WORK(&map->work, bpf_map_free_deferred);
 323                 schedule_work(&map->work);
 324         }
 325 }
 326
 327 void bpf_map_put(struct bpf_map *map)
 328 {
 329         __bpf_map_put(map, true);
 330 }
 331 EXPORT_SYMBOL_GPL(bpf_map_put);
 332
 333 void bpf_map_put_with_uref(struct bpf_map *map)
 334 {
 335         bpf_map_put_uref(map);
 336         bpf_map_put(map);
 337 }
 338
 339 static int bpf_map_release(struct inode *inode, struct file *filp)
 340 {
 341         struct bpf_map *map = filp->private_data;
 342
 343         if (map->ops->map_release)
 344                 map->ops->map_release(map, filp);
 345
 346         bpf_map_put_with_uref(map);
 347         return 0;
 348 }
 349
 350 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 351 {
 352         fmode_t mode = f.file->f_mode;
 353
 354         /* Our file permissions may have been overridden by global
 355          * map permissions facing syscall side.
 356          */
 357         if (READ_ONCE(map->frozen))
 358                 mode &= ~FMODE_CAN_WRITE;
 359         return mode;
 360 }
 361
 362 #ifdef CONFIG_PROC_FS
 363 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 364 {
 365         const struct bpf_map *map = filp->private_data;
 366         const struct bpf_array *array;
 367         u32 owner_prog_type = 0;
 368         u32 owner_jited = 0;
 369
 370         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 371                 array = container_of(map, struct bpf_array, map);
 372                 owner_prog_type = array->owner_prog_type;
 373                 owner_jited = array->owner_jited;
 374         }
 375
 376         seq_printf(m,
 377                    "map_type:\t%u\n"
 378                    "key_size:\t%u\n"
 379                    "value_size:\t%u\n"
 380                    "max_entries:\t%u\n"
 381                    "map_flags:\t%#x\n"
 382                    "memlock:\t%llu\n"
 383                    "map_id:\t%u\n"
 384                    "frozen:\t%u\n",
 385                    map->map_type,
 386                    map->key_size,
 387                    map->value_size,
 388                    map->max_entries,
 389                    map->map_flags,
 390                    map->pages * 1ULL << PAGE_SHIFT,
 391                    map->id,
 392                    READ_ONCE(map->frozen));
 393
 394         if (owner_prog_type) {
 395                 seq_printf(m, "owner_prog_type:\t%u\n",
 396                            owner_prog_type);
 397                 seq_printf(m, "owner_jited:\t%u\n",
 398                            owner_jited);
 399         }
 400 }
 401 #endif
 402
 403 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 404                               loff_t *ppos)
 405 {
 406         /* We need this handler such that alloc_file() enables
 407          * f_mode with FMODE_CAN_READ.
 408          */
 409         return -EINVAL;
 410 }
 411
 412 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 413                                size_t siz, loff_t *ppos)
 414 {
 415         /* We need this handler such that alloc_file() enables
 416          * f_mode with FMODE_CAN_WRITE.
 417          */
 418         return -EINVAL;
 419 }
 420
 421 const struct file_operations bpf_map_fops = {
 422 #ifdef CONFIG_PROC_FS
 423         .show_fdinfo    = bpf_map_show_fdinfo,
 424 #endif
 425         .release        = bpf_map_release,
 426         .read           = bpf_dummy_read,
 427         .write          = bpf_dummy_write,
 428 };
 429
 430 int bpf_map_new_fd(struct bpf_map *map, int flags)
 431 {
 432         int ret;
 433
 434         ret = security_bpf_map(map, OPEN_FMODE(flags));
 435         if (ret < 0)
 436                 return ret;
 437
 438         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 439                                 flags | O_CLOEXEC);
 440 }
 441
 442 int bpf_get_file_flag(int flags)
 443 {
 444         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 445                 return -EINVAL;
 446         if (flags & BPF_F_RDONLY)
 447                 return O_RDONLY;
 448         if (flags & BPF_F_WRONLY)
 449                 return O_WRONLY;
 450         return O_RDWR;
 451 }
 452
 453 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 454 #define CHECK_ATTR(CMD) \
 455         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 456                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 457                    sizeof(*attr) - \
 458                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 459                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 460
 461 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 462  * Return 0 on success and < 0 on error.
 463  */
 464 static int bpf_obj_name_cpy(char *dst, const char *src)
 465 {
 466         const char *end = src + BPF_OBJ_NAME_LEN;
 467
 468         memset(dst, 0, BPF_OBJ_NAME_LEN);
 469         /* Copy all isalnum(), '_' and '.' chars. */
 470         while (src < end && *src) {
 471                 if (!isalnum(*src) &&
 472                     *src != '_' && *src != '.')
 473                         return -EINVAL;
 474                 *dst++ = *src++;
 475         }
 476
 477         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 478         if (src == end)
 479                 return -EINVAL;
 480
 481         return 0;
 482 }
 483
 484 int map_check_no_btf(const struct bpf_map *map,
 485                      const struct btf *btf,
 486                      const struct btf_type *key_type,
 487                      const struct btf_type *value_type)
 488 {
 489         return -ENOTSUPP;
 490 }
 491
 492 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 493                          u32 btf_key_id, u32 btf_value_id)
 494 {
 495         const struct btf_type *key_type, *value_type;
 496         u32 key_size, value_size;
 497         int ret = 0;
 498
 499         /* Some maps allow key to be unspecified. */
 500         if (btf_key_id) {
 501                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 502                 if (!key_type || key_size != map->key_size)
 503                         return -EINVAL;
 504         } else {
 505                 key_type = btf_type_by_id(btf, 0);
 506                 if (!map->ops->map_check_btf)
 507                         return -EINVAL;
 508         }
 509
 510         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 511         if (!value_type || value_size != map->value_size)
 512                 return -EINVAL;
 513
 514         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 515
 516         if (map_value_has_spin_lock(map)) {
 517                 if (map->map_flags & BPF_F_RDONLY_PROG)
 518                         return -EACCES;
 519                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 520                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 521                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 522                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 523                         return -ENOTSUPP;
 524                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 525                     map->value_size) {
 526                         WARN_ONCE(1,
 527                                   "verifier bug spin_lock_off %d value_size %d\n",
 528                                   map->spin_lock_off, map->value_size);
 529                         return -EFAULT;
 530                 }
 531         }
 532
 533         if (map->ops->map_check_btf)
 534                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 535
 536         return ret;
 537 }
 538
 539 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 540 /* called via syscall */
 541 static int map_create(union bpf_attr *attr)
 542 {
 543         int numa_node = bpf_map_attr_numa_node(attr);
 544         struct bpf_map *map;
 545         int f_flags;
 546         int err;
 547
 548         err = CHECK_ATTR(BPF_MAP_CREATE);
 549         if (err)
 550                 return -EINVAL;
 551
 552         f_flags = bpf_get_file_flag(attr->map_flags);
 553         if (f_flags < 0)
 554                 return f_flags;
 555
 556         if (numa_node != NUMA_NO_NODE &&
 557             ((unsigned int)numa_node >= nr_node_ids ||
 558              !node_online(numa_node)))
 559                 return -EINVAL;
 560
 561         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 562         map = find_and_alloc_map(attr);
 563         if (IS_ERR(map))
 564                 return PTR_ERR(map);
 565
 566         err = bpf_obj_name_cpy(map->name, attr->map_name);
 567         if (err)
 568                 goto free_map_nouncharge;
 569
 570         atomic_set(&map->refcnt, 1);
 571         atomic_set(&map->usercnt, 1);
 572
 573         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 574                 struct btf *btf;
 575
 576                 if (!attr->btf_value_type_id) {
 577                         err = -EINVAL;
 578                         goto free_map_nouncharge;
 579                 }
 580
 581                 btf = btf_get_by_fd(attr->btf_fd);
 582                 if (IS_ERR(btf)) {
 583                         err = PTR_ERR(btf);
 584                         goto free_map_nouncharge;
 585                 }
 586
 587                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 588                                     attr->btf_value_type_id);
 589                 if (err) {
 590                         btf_put(btf);
 591                         goto free_map_nouncharge;
 592                 }
 593
 594                 map->btf = btf;
 595                 map->btf_key_type_id = attr->btf_key_type_id;
 596                 map->btf_value_type_id = attr->btf_value_type_id;
 597         } else {
 598                 map->spin_lock_off = -EINVAL;
 599         }
 600
 601         err = security_bpf_map_alloc(map);
 602         if (err)
 603                 goto free_map_nouncharge;
 604
 605         err = bpf_map_init_memlock(map);
 606         if (err)
 607                 goto free_map_sec;
 608
 609         err = bpf_map_alloc_id(map);
 610         if (err)
 611                 goto free_map;
 612
 613         err = bpf_map_new_fd(map, f_flags);
 614         if (err < 0) {
 615                 /* failed to allocate fd.
 616                  * bpf_map_put_with_uref() is needed because the above
 617                  * bpf_map_alloc_id() has published the map
 618                  * to the userspace and the userspace may
 619                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 620                  */
 621                 bpf_map_put_with_uref(map);
 622                 return err;
 623         }
 624
 625         return err;
 626
 627 free_map:
 628         bpf_map_release_memlock(map);
 629 free_map_sec:
 630         security_bpf_map_free(map);
 631 free_map_nouncharge:
 632         btf_put(map->btf);
 633         map->ops->map_free(map);
 634         return err;
 635 }
 636
 637 /* if error is returned, fd is released.
 638  * On success caller should complete fd access with matching fdput()
 639  */
 640 struct bpf_map *__bpf_map_get(struct fd f)
 641 {
 642         if (!f.file)
 643                 return ERR_PTR(-EBADF);
 644         if (f.file->f_op != &bpf_map_fops) {
 645                 fdput(f);
 646                 return ERR_PTR(-EINVAL);
 647         }
 648
 649         return f.file->private_data;
 650 }
 651
 652 /* prog's and map's refcnt limit */
 653 #define BPF_MAX_REFCNT 32768
 654
 655 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 656 {
 657         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 658                 atomic_dec(&map->refcnt);
 659                 return ERR_PTR(-EBUSY);
 660         }
 661         if (uref)
 662                 atomic_inc(&map->usercnt);
 663         return map;
 664 }
 665 EXPORT_SYMBOL_GPL(bpf_map_inc);
 666
 667 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 668 {
 669         struct fd f = fdget(ufd);
 670         struct bpf_map *map;
 671
 672         map = __bpf_map_get(f);
 673         if (IS_ERR(map))
 674                 return map;
 675
 676         map = bpf_map_inc(map, true);
 677         fdput(f);
 678
 679         return map;
 680 }
 681
 682 /* map_idr_lock should have been held */
 683 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 684                                             bool uref)
 685 {
 686         int refold;
 687
 688         refold = atomic_fetch_add_unless(&map->refcnt, 1, 0);
 689
 690         if (refold >= BPF_MAX_REFCNT) {
 691                 __bpf_map_put(map, false);
 692                 return ERR_PTR(-EBUSY);
 693         }
 694
 695         if (!refold)
 696                 return ERR_PTR(-ENOENT);
 697
 698         if (uref)
 699                 atomic_inc(&map->usercnt);
 700
 701         return map;
 702 }
 703
 704 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 705 {
 706         return -ENOTSUPP;
 707 }
 708
 709 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 710 {
 711         if (key_size)
 712                 return memdup_user(ukey, key_size);
 713
 714         if (ukey)
 715                 return ERR_PTR(-EINVAL);
 716
 717         return NULL;
 718 }
 719
 720 /* last field in 'union bpf_attr' used by this command */
 721 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 722
 723 static int map_lookup_elem(union bpf_attr *attr)
 724 {
 725         void __user *ukey = u64_to_user_ptr(attr->key);
 726         void __user *uvalue = u64_to_user_ptr(attr->value);
 727         int ufd = attr->map_fd;
 728         struct bpf_map *map;
 729         void *key, *value, *ptr;
 730         u32 value_size;
 731         struct fd f;
 732         int err;
 733
 734         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 735                 return -EINVAL;
 736
 737         if (attr->flags & ~BPF_F_LOCK)
 738                 return -EINVAL;
 739
 740         f = fdget(ufd);
 741         map = __bpf_map_get(f);
 742         if (IS_ERR(map))
 743                 return PTR_ERR(map);
 744         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 745                 err = -EPERM;
 746                 goto err_put;
 747         }
 748
 749         if ((attr->flags & BPF_F_LOCK) &&
 750             !map_value_has_spin_lock(map)) {
 751                 err = -EINVAL;
 752                 goto err_put;
 753         }
 754
 755         key = __bpf_copy_key(ukey, map->key_size);
 756         if (IS_ERR(key)) {
 757                 err = PTR_ERR(key);
 758                 goto err_put;
 759         }
 760
 761         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 762             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 763             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 764             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 765                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 766         else if (IS_FD_MAP(map))
 767                 value_size = sizeof(u32);
 768         else
 769                 value_size = map->value_size;
 770
 771         err = -ENOMEM;
 772         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 773         if (!value)
 774                 goto free_key;
 775
 776         if (bpf_map_is_dev_bound(map)) {
 777                 err = bpf_map_offload_lookup_elem(map, key, value);
 778                 goto done;
 779         }
 780
 781         preempt_disable();
 782         this_cpu_inc(bpf_prog_active);
 783         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 784             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 785                 err = bpf_percpu_hash_copy(map, key, value);
 786         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 787                 err = bpf_percpu_array_copy(map, key, value);
 788         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 789                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 790         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 791                 err = bpf_stackmap_copy(map, key, value);
 792         } else if (IS_FD_ARRAY(map)) {
 793                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 794         } else if (IS_FD_HASH(map)) {
 795                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 796         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 797                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 798         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 799                    map->map_type == BPF_MAP_TYPE_STACK) {
 800                 err = map->ops->map_peek_elem(map, value);
 801         } else {
 802                 rcu_read_lock();
 803                 if (map->ops->map_lookup_elem_sys_only)
 804                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 805                 else
 806                         ptr = map->ops->map_lookup_elem(map, key);
 807                 if (IS_ERR(ptr)) {
 808                         err = PTR_ERR(ptr);
 809                 } else if (!ptr) {
 810                         err = -ENOENT;
 811                 } else {
 812                         err = 0;
 813                         if (attr->flags & BPF_F_LOCK)
 814                                 /* lock 'ptr' and copy everything but lock */
 815                                 copy_map_value_locked(map, value, ptr, true);
 816                         else
 817                                 copy_map_value(map, value, ptr);
 818                         /* mask lock, since value wasn't zero inited */
 819                         check_and_init_map_lock(map, value);
 820                 }
 821                 rcu_read_unlock();
 822         }
 823         this_cpu_dec(bpf_prog_active);
 824         preempt_enable();
 825
 826 done:
 827         if (err)
 828                 goto free_value;
 829
 830         err = -EFAULT;
 831         if (copy_to_user(uvalue, value, value_size) != 0)
 832                 goto free_value;
 833
 834         err = 0;
 835
 836 free_value:
 837         kfree(value);
 838 free_key:
 839         kfree(key);
 840 err_put:
 841         fdput(f);
 842         return err;
 843 }
 844
 845 static void maybe_wait_bpf_programs(struct bpf_map *map)
 846 {
 847         /* Wait for any running BPF programs to complete so that
 848          * userspace, when we return to it, knows that all programs
 849          * that could be running use the new map value.
 850          */
 851         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 852             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 853                 synchronize_rcu();
 854 }
 855
 856 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 857
 858 static int map_update_elem(union bpf_attr *attr)
 859 {
 860         void __user *ukey = u64_to_user_ptr(attr->key);
 861         void __user *uvalue = u64_to_user_ptr(attr->value);
 862         int ufd = attr->map_fd;
 863         struct bpf_map *map;
 864         void *key, *value;
 865         u32 value_size;
 866         struct fd f;
 867         int err;
 868
 869         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 870                 return -EINVAL;
 871
 872         f = fdget(ufd);
 873         map = __bpf_map_get(f);
 874         if (IS_ERR(map))
 875                 return PTR_ERR(map);
 876         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 877                 err = -EPERM;
 878                 goto err_put;
 879         }
 880
 881         if ((attr->flags & BPF_F_LOCK) &&
 882             !map_value_has_spin_lock(map)) {
 883                 err = -EINVAL;
 884                 goto err_put;
 885         }
 886
 887         key = __bpf_copy_key(ukey, map->key_size);
 888         if (IS_ERR(key)) {
 889                 err = PTR_ERR(key);
 890                 goto err_put;
 891         }
 892
 893         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 894             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 895             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 896             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 897                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 898         else
 899                 value_size = map->value_size;
 900
 901         err = -ENOMEM;
 902         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 903         if (!value)
 904                 goto free_key;
 905
 906         err = -EFAULT;
 907         if (copy_from_user(value, uvalue, value_size) != 0)
 908                 goto free_value;
 909
 910         /* Need to create a kthread, thus must support schedule */
 911         if (bpf_map_is_dev_bound(map)) {
 912                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 913                 goto out;
 914         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 915                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 916                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 917                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 918                 goto out;
 919         }
 920
 921         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 922          * inside bpf map update or delete otherwise deadlocks are possible
 923          */
 924         preempt_disable();
 925         __this_cpu_inc(bpf_prog_active);
 926         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 927             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 928                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 929         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 930                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 931         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 932                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 933                                                        attr->flags);
 934         } else if (IS_FD_ARRAY(map)) {
 935                 rcu_read_lock();
 936                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 937                                                    attr->flags);
 938                 rcu_read_unlock();
 939         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 940                 rcu_read_lock();
 941                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 942                                                   attr->flags);
 943                 rcu_read_unlock();
 944         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 945                 /* rcu_read_lock() is not needed */
 946                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 947                                                          attr->flags);
 948         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 949                    map->map_type == BPF_MAP_TYPE_STACK) {
 950                 err = map->ops->map_push_elem(map, value, attr->flags);
 951         } else {
 952                 rcu_read_lock();
 953                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 954                 rcu_read_unlock();
 955         }
 956         __this_cpu_dec(bpf_prog_active);
 957         preempt_enable();
 958         maybe_wait_bpf_programs(map);
 959 out:
 960 free_value:
 961         kfree(value);
 962 free_key:
 963         kfree(key);
 964 err_put:
 965         fdput(f);
 966         return err;
 967 }
 968
 969 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 970
 971 static int map_delete_elem(union bpf_attr *attr)
 972 {
 973         void __user *ukey = u64_to_user_ptr(attr->key);
 974         int ufd = attr->map_fd;
 975         struct bpf_map *map;
 976         struct fd f;
 977         void *key;
 978         int err;
 979
 980         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 981                 return -EINVAL;
 982
 983         f = fdget(ufd);
 984         map = __bpf_map_get(f);
 985         if (IS_ERR(map))
 986                 return PTR_ERR(map);
 987         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 988                 err = -EPERM;
 989                 goto err_put;
 990         }
 991
 992         key = __bpf_copy_key(ukey, map->key_size);
 993         if (IS_ERR(key)) {
 994                 err = PTR_ERR(key);
 995                 goto err_put;
 996         }
 997
 998         if (bpf_map_is_dev_bound(map)) {
 999                 err = bpf_map_offload_delete_elem(map, key);
1000                 goto out;
1001         }
1002
1003         preempt_disable();
1004         __this_cpu_inc(bpf_prog_active);
1005         rcu_read_lock();
1006         err = map->ops->map_delete_elem(map, key);
1007         rcu_read_unlock();
1008         __this_cpu_dec(bpf_prog_active);
1009         preempt_enable();
1010         maybe_wait_bpf_programs(map);
1011 out:
1012         kfree(key);
1013 err_put:
1014         fdput(f);
1015         return err;
1016 }
1017
1018 /* last field in 'union bpf_attr' used by this command */
1019 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1020
1021 static int map_get_next_key(union bpf_attr *attr)
1022 {
1023         void __user *ukey = u64_to_user_ptr(attr->key);
1024         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1025         int ufd = attr->map_fd;
1026         struct bpf_map *map;
1027         void *key, *next_key;
1028         struct fd f;
1029         int err;
1030
1031         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1032                 return -EINVAL;
1033
1034         f = fdget(ufd);
1035         map = __bpf_map_get(f);
1036         if (IS_ERR(map))
1037                 return PTR_ERR(map);
1038         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1039                 err = -EPERM;
1040                 goto err_put;
1041         }
1042
1043         if (ukey) {
1044                 key = __bpf_copy_key(ukey, map->key_size);
1045                 if (IS_ERR(key)) {
1046                         err = PTR_ERR(key);
1047                         goto err_put;
1048                 }
1049         } else {
1050                 key = NULL;
1051         }
1052
1053         err = -ENOMEM;
1054         next_key = kmalloc(map->key_size, GFP_USER);
1055         if (!next_key)
1056                 goto free_key;
1057
1058         if (bpf_map_is_dev_bound(map)) {
1059                 err = bpf_map_offload_get_next_key(map, key, next_key);
1060                 goto out;
1061         }
1062
1063         rcu_read_lock();
1064         err = map->ops->map_get_next_key(map, key, next_key);
1065         rcu_read_unlock();
1066 out:
1067         if (err)
1068                 goto free_next_key;
1069
1070         err = -EFAULT;
1071         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1072                 goto free_next_key;
1073
1074         err = 0;
1075
1076 free_next_key:
1077         kfree(next_key);
1078 free_key:
1079         kfree(key);
1080 err_put:
1081         fdput(f);
1082         return err;
1083 }
1084
1085 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1086
1087 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1088 {
1089         void __user *ukey = u64_to_user_ptr(attr->key);
1090         void __user *uvalue = u64_to_user_ptr(attr->value);
1091         int ufd = attr->map_fd;
1092         struct bpf_map *map;
1093         void *key, *value;
1094         u32 value_size;
1095         struct fd f;
1096         int err;
1097
1098         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1099                 return -EINVAL;
1100
1101         f = fdget(ufd);
1102         map = __bpf_map_get(f);
1103         if (IS_ERR(map))
1104                 return PTR_ERR(map);
1105         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1106                 err = -EPERM;
1107                 goto err_put;
1108         }
1109
1110         key = __bpf_copy_key(ukey, map->key_size);
1111         if (IS_ERR(key)) {
1112                 err = PTR_ERR(key);
1113                 goto err_put;
1114         }
1115
1116         value_size = map->value_size;
1117
1118         err = -ENOMEM;
1119         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1120         if (!value)
1121                 goto free_key;
1122
1123         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1124             map->map_type == BPF_MAP_TYPE_STACK) {
1125                 err = map->ops->map_pop_elem(map, value);
1126         } else {
1127                 err = -ENOTSUPP;
1128         }
1129
1130         if (err)
1131                 goto free_value;
1132
1133         if (copy_to_user(uvalue, value, value_size) != 0)
1134                 goto free_value;
1135
1136         err = 0;
1137
1138 free_value:
1139         kfree(value);
1140 free_key:
1141         kfree(key);
1142 err_put:
1143         fdput(f);
1144         return err;
1145 }
1146
1147 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1148
1149 static int map_freeze(const union bpf_attr *attr)
1150 {
1151         int err = 0, ufd = attr->map_fd;
1152         struct bpf_map *map;
1153         struct fd f;
1154
1155         if (CHECK_ATTR(BPF_MAP_FREEZE))
1156                 return -EINVAL;
1157
1158         f = fdget(ufd);
1159         map = __bpf_map_get(f);
1160         if (IS_ERR(map))
1161                 return PTR_ERR(map);
1162         if (READ_ONCE(map->frozen)) {
1163                 err = -EBUSY;
1164                 goto err_put;
1165         }
1166         if (!capable(CAP_SYS_ADMIN)) {
1167                 err = -EPERM;
1168                 goto err_put;
1169         }
1170
1171         WRITE_ONCE(map->frozen, true);
1172 err_put:
1173         fdput(f);
1174         return err;
1175 }
1176
1177 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1178 #define BPF_PROG_TYPE(_id, _name) \
1179         [_id] = & _name ## _prog_ops,
1180 #define BPF_MAP_TYPE(_id, _ops)
1181 #include <linux/bpf_types.h>
1182 #undef BPF_PROG_TYPE
1183 #undef BPF_MAP_TYPE
1184 };
1185
1186 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1187 {
1188         const struct bpf_prog_ops *ops;
1189
1190         if (type >= ARRAY_SIZE(bpf_prog_types))
1191                 return -EINVAL;
1192         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1193         ops = bpf_prog_types[type];
1194         if (!ops)
1195                 return -EINVAL;
1196
1197         if (!bpf_prog_is_dev_bound(prog->aux))
1198                 prog->aux->ops = ops;
1199         else
1200                 prog->aux->ops = &bpf_offload_prog_ops;
1201         prog->type = type;
1202         return 0;
1203 }
1204
1205 /* drop refcnt on maps used by eBPF program and free auxilary data */
1206 static void free_used_maps(struct bpf_prog_aux *aux)
1207 {
1208         enum bpf_cgroup_storage_type stype;
1209         int i;
1210
1211         for_each_cgroup_storage_type(stype) {
1212                 if (!aux->cgroup_storage[stype])
1213                         continue;
1214                 bpf_cgroup_storage_release(aux->prog,
1215                                            aux->cgroup_storage[stype]);
1216         }
1217
1218         for (i = 0; i < aux->used_map_cnt; i++)
1219                 bpf_map_put(aux->used_maps[i]);
1220
1221         kfree(aux->used_maps);
1222 }
1223
1224 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1225 {
1226         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1227         unsigned long user_bufs;
1228
1229         if (user) {
1230                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1231                 if (user_bufs > memlock_limit) {
1232                         atomic_long_sub(pages, &user->locked_vm);
1233                         return -EPERM;
1234                 }
1235         }
1236
1237         return 0;
1238 }
1239
1240 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1241 {
1242         if (user)
1243                 atomic_long_sub(pages, &user->locked_vm);
1244 }
1245
1246 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1247 {
1248         struct user_struct *user = get_current_user();
1249         int ret;
1250
1251         ret = __bpf_prog_charge(user, prog->pages);
1252         if (ret) {
1253                 free_uid(user);
1254                 return ret;
1255         }
1256
1257         prog->aux->user = user;
1258         return 0;
1259 }
1260
1261 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1262 {
1263         struct user_struct *user = prog->aux->user;
1264
1265         __bpf_prog_uncharge(user, prog->pages);
1266         free_uid(user);
1267 }
1268
1269 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1270 {
1271         int id;
1272
1273         idr_preload(GFP_KERNEL);
1274         spin_lock_bh(&prog_idr_lock);
1275         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1276         if (id > 0)
1277                 prog->aux->id = id;
1278         spin_unlock_bh(&prog_idr_lock);
1279         idr_preload_end();
1280
1281         /* id is in [1, INT_MAX) */
1282         if (WARN_ON_ONCE(!id))
1283                 return -ENOSPC;
1284
1285         return id > 0 ? 0 : id;
1286 }
1287
1288 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1289 {
1290         /* cBPF to eBPF migrations are currently not in the idr store.
1291          * Offloaded programs are removed from the store when their device
1292          * disappears - even if someone grabs an fd to them they are unusable,
1293          * simply waiting for refcnt to drop to be freed.
1294          */
1295         if (!prog->aux->id)
1296                 return;
1297
1298         if (do_idr_lock)
1299                 spin_lock_bh(&prog_idr_lock);
1300         else
1301                 __acquire(&prog_idr_lock);
1302
1303         idr_remove(&prog_idr, prog->aux->id);
1304         prog->aux->id = 0;
1305
1306         if (do_idr_lock)
1307                 spin_unlock_bh(&prog_idr_lock);
1308         else
1309                 __release(&prog_idr_lock);
1310 }
1311
1312 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1313 {
1314         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1315
1316         free_used_maps(aux);
1317         bpf_prog_uncharge_memlock(aux->prog);
1318         security_bpf_prog_free(aux);
1319         bpf_prog_free(aux->prog);
1320 }
1321
1322 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1323 {
1324         if (atomic_dec_and_test(&prog->aux->refcnt)) {
1325                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1326                 /* bpf_prog_free_id() must be called first */
1327                 bpf_prog_free_id(prog, do_idr_lock);
1328                 bpf_prog_kallsyms_del_all(prog);
1329                 btf_put(prog->aux->btf);
1330                 kvfree(prog->aux->func_info);
1331                 bpf_prog_free_linfo(prog);
1332
1333                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1334         }
1335 }
1336
1337 void bpf_prog_put(struct bpf_prog *prog)
1338 {
1339         __bpf_prog_put(prog, true);
1340 }
1341 EXPORT_SYMBOL_GPL(bpf_prog_put);
1342
1343 static int bpf_prog_release(struct inode *inode, struct file *filp)
1344 {
1345         struct bpf_prog *prog = filp->private_data;
1346
1347         bpf_prog_put(prog);
1348         return 0;
1349 }
1350
1351 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1352                                struct bpf_prog_stats *stats)
1353 {
1354         u64 nsecs = 0, cnt = 0;
1355         int cpu;
1356
1357         for_each_possible_cpu(cpu) {
1358                 const struct bpf_prog_stats *st;
1359                 unsigned int start;
1360                 u64 tnsecs, tcnt;
1361
1362                 st = per_cpu_ptr(prog->aux->stats, cpu);
1363                 do {
1364                         start = u64_stats_fetch_begin_irq(&st->syncp);
1365                         tnsecs = st->nsecs;
1366                         tcnt = st->cnt;
1367                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1368                 nsecs += tnsecs;
1369                 cnt += tcnt;
1370         }
1371         stats->nsecs = nsecs;
1372         stats->cnt = cnt;
1373 }
1374
1375 #ifdef CONFIG_PROC_FS
1376 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1377 {
1378         const struct bpf_prog *prog = filp->private_data;
1379         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1380         struct bpf_prog_stats stats;
1381
1382         bpf_prog_get_stats(prog, &stats);
1383         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1384         seq_printf(m,
1385                    "prog_type:\t%u\n"
1386                    "prog_jited:\t%u\n"
1387                    "prog_tag:\t%s\n"
1388                    "memlock:\t%llu\n"
1389                    "prog_id:\t%u\n"
1390                    "run_time_ns:\t%llu\n"
1391                    "run_cnt:\t%llu\n",
1392                    prog->type,
1393                    prog->jited,
1394                    prog_tag,
1395                    prog->pages * 1ULL << PAGE_SHIFT,
1396                    prog->aux->id,
1397                    stats.nsecs,
1398                    stats.cnt);
1399 }
1400 #endif
1401
1402 const struct file_operations bpf_prog_fops = {
1403 #ifdef CONFIG_PROC_FS
1404         .show_fdinfo    = bpf_prog_show_fdinfo,
1405 #endif
1406         .release        = bpf_prog_release,
1407         .read           = bpf_dummy_read,
1408         .write          = bpf_dummy_write,
1409 };
1410
1411 int bpf_prog_new_fd(struct bpf_prog *prog)
1412 {
1413         int ret;
1414
1415         ret = security_bpf_prog(prog);
1416         if (ret < 0)
1417                 return ret;
1418
1419         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1420                                 O_RDWR | O_CLOEXEC);
1421 }
1422
1423 static struct bpf_prog *____bpf_prog_get(struct fd f)
1424 {
1425         if (!f.file)
1426                 return ERR_PTR(-EBADF);
1427         if (f.file->f_op != &bpf_prog_fops) {
1428                 fdput(f);
1429                 return ERR_PTR(-EINVAL);
1430         }
1431
1432         return f.file->private_data;
1433 }
1434
1435 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1436 {
1437         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1438                 atomic_sub(i, &prog->aux->refcnt);
1439                 return ERR_PTR(-EBUSY);
1440         }
1441         return prog;
1442 }
1443 EXPORT_SYMBOL_GPL(bpf_prog_add);
1444
1445 void bpf_prog_sub(struct bpf_prog *prog, int i)
1446 {
1447         /* Only to be used for undoing previous bpf_prog_add() in some
1448          * error path. We still know that another entity in our call
1449          * path holds a reference to the program, thus atomic_sub() can
1450          * be safely used in such cases!
1451          */
1452         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1453 }
1454 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1455
1456 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1457 {
1458         return bpf_prog_add(prog, 1);
1459 }
1460 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1461
1462 /* prog_idr_lock should have been held */
1463 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1464 {
1465         int refold;
1466
1467         refold = atomic_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1468
1469         if (refold >= BPF_MAX_REFCNT) {
1470                 __bpf_prog_put(prog, false);
1471                 return ERR_PTR(-EBUSY);
1472         }
1473
1474         if (!refold)
1475                 return ERR_PTR(-ENOENT);
1476
1477         return prog;
1478 }
1479 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1480
1481 bool bpf_prog_get_ok(struct bpf_prog *prog,
1482                             enum bpf_prog_type *attach_type, bool attach_drv)
1483 {
1484         /* not an attachment, just a refcount inc, always allow */
1485         if (!attach_type)
1486                 return true;
1487
1488         if (prog->type != *attach_type)
1489                 return false;
1490         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1491                 return false;
1492
1493         return true;
1494 }
1495
1496 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1497                                        bool attach_drv)
1498 {
1499         struct fd f = fdget(ufd);
1500         struct bpf_prog *prog;
1501
1502         prog = ____bpf_prog_get(f);
1503         if (IS_ERR(prog))
1504                 return prog;
1505         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1506                 prog = ERR_PTR(-EINVAL);
1507                 goto out;
1508         }
1509
1510         prog = bpf_prog_inc(prog);
1511 out:
1512         fdput(f);
1513         return prog;
1514 }
1515
1516 struct bpf_prog *bpf_prog_get(u32 ufd)
1517 {
1518         return __bpf_prog_get(ufd, NULL, false);
1519 }
1520
1521 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1522                                        bool attach_drv)
1523 {
1524         return __bpf_prog_get(ufd, &type, attach_drv);
1525 }
1526 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1527
1528 /* Initially all BPF programs could be loaded w/o specifying
1529  * expected_attach_type. Later for some of them specifying expected_attach_type
1530  * at load time became required so that program could be validated properly.
1531  * Programs of types that are allowed to be loaded both w/ and w/o (for
1532  * backward compatibility) expected_attach_type, should have the default attach
1533  * type assigned to expected_attach_type for the latter case, so that it can be
1534  * validated later at attach time.
1535  *
1536  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1537  * prog type requires it but has some attach types that have to be backward
1538  * compatible.
1539  */
1540 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1541 {
1542         switch (attr->prog_type) {
1543         case BPF_PROG_TYPE_CGROUP_SOCK:
1544                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1545                  * exist so checking for non-zero is the way to go here.
1546                  */
1547                 if (!attr->expected_attach_type)
1548                         attr->expected_attach_type =
1549                                 BPF_CGROUP_INET_SOCK_CREATE;
1550                 break;
1551         }
1552 }
1553
1554 static int
1555 bpf_prog_load_check_attach_type(enum bpf_prog_type prog_type,
1556                                 enum bpf_attach_type expected_attach_type)
1557 {
1558         switch (prog_type) {
1559         case BPF_PROG_TYPE_CGROUP_SOCK:
1560                 switch (expected_attach_type) {
1561                 case BPF_CGROUP_INET_SOCK_CREATE:
1562                 case BPF_CGROUP_INET4_POST_BIND:
1563                 case BPF_CGROUP_INET6_POST_BIND:
1564                         return 0;
1565                 default:
1566                         return -EINVAL;
1567                 }
1568         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1569                 switch (expected_attach_type) {
1570                 case BPF_CGROUP_INET4_BIND:
1571                 case BPF_CGROUP_INET6_BIND:
1572                 case BPF_CGROUP_INET4_CONNECT:
1573                 case BPF_CGROUP_INET6_CONNECT:
1574                 case BPF_CGROUP_UDP4_SENDMSG:
1575                 case BPF_CGROUP_UDP6_SENDMSG:
1576                         return 0;
1577                 default:
1578                         return -EINVAL;
1579                 }
1580         default:
1581                 return 0;
1582         }
1583 }
1584
1585 /* last field in 'union bpf_attr' used by this command */
1586 #define BPF_PROG_LOAD_LAST_FIELD line_info_cnt
1587
1588 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1589 {
1590         enum bpf_prog_type type = attr->prog_type;
1591         struct bpf_prog *prog;
1592         int err;
1593         char license[128];
1594         bool is_gpl;
1595
1596         if (CHECK_ATTR(BPF_PROG_LOAD))
1597                 return -EINVAL;
1598
1599         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT | BPF_F_ANY_ALIGNMENT))
1600                 return -EINVAL;
1601
1602         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1603             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1604             !capable(CAP_SYS_ADMIN))
1605                 return -EPERM;
1606
1607         /* copy eBPF program license from user space */
1608         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1609                               sizeof(license) - 1) < 0)
1610                 return -EFAULT;
1611         license[sizeof(license) - 1] = 0;
1612
1613         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1614         is_gpl = license_is_gpl_compatible(license);
1615
1616         if (attr->insn_cnt == 0 ||
1617             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1618                 return -E2BIG;
1619         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1620             type != BPF_PROG_TYPE_CGROUP_SKB &&
1621             !capable(CAP_SYS_ADMIN))
1622                 return -EPERM;
1623
1624         bpf_prog_load_fixup_attach_type(attr);
1625         if (bpf_prog_load_check_attach_type(type, attr->expected_attach_type))
1626                 return -EINVAL;
1627
1628         /* plain bpf_prog allocation */
1629         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1630         if (!prog)
1631                 return -ENOMEM;
1632
1633         prog->expected_attach_type = attr->expected_attach_type;
1634
1635         prog->aux->offload_requested = !!attr->prog_ifindex;
1636
1637         err = security_bpf_prog_alloc(prog->aux);
1638         if (err)
1639                 goto free_prog_nouncharge;
1640
1641         err = bpf_prog_charge_memlock(prog);
1642         if (err)
1643                 goto free_prog_sec;
1644
1645         prog->len = attr->insn_cnt;
1646
1647         err = -EFAULT;
1648         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1649                            bpf_prog_insn_size(prog)) != 0)
1650                 goto free_prog;
1651
1652         prog->orig_prog = NULL;
1653         prog->jited = 0;
1654
1655         atomic_set(&prog->aux->refcnt, 1);
1656         prog->gpl_compatible = is_gpl ? 1 : 0;
1657
1658         if (bpf_prog_is_dev_bound(prog->aux)) {
1659                 err = bpf_prog_offload_init(prog, attr);
1660                 if (err)
1661                         goto free_prog;
1662         }
1663
1664         /* find program type: socket_filter vs tracing_filter */
1665         err = find_prog_type(type, prog);
1666         if (err < 0)
1667                 goto free_prog;
1668
1669         prog->aux->load_time = ktime_get_boot_ns();
1670         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1671         if (err)
1672                 goto free_prog;
1673
1674         /* run eBPF verifier */
1675         err = bpf_check(&prog, attr, uattr);
1676         if (err < 0)
1677                 goto free_used_maps;
1678
1679         prog = bpf_prog_select_runtime(prog, &err);
1680         if (err < 0)
1681                 goto free_used_maps;
1682
1683         err = bpf_prog_alloc_id(prog);
1684         if (err)
1685                 goto free_used_maps;
1686
1687         err = bpf_prog_new_fd(prog);
1688         if (err < 0) {
1689                 /* failed to allocate fd.
1690                  * bpf_prog_put() is needed because the above
1691                  * bpf_prog_alloc_id() has published the prog
1692                  * to the userspace and the userspace may
1693                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1694                  */
1695                 bpf_prog_put(prog);
1696                 return err;
1697         }
1698
1699         bpf_prog_kallsyms_add(prog);
1700         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1701         return err;
1702
1703 free_used_maps:
1704         bpf_prog_free_linfo(prog);
1705         kvfree(prog->aux->func_info);
1706         btf_put(prog->aux->btf);
1707         bpf_prog_kallsyms_del_subprogs(prog);
1708         free_used_maps(prog->aux);
1709 free_prog:
1710         bpf_prog_uncharge_memlock(prog);
1711 free_prog_sec:
1712         security_bpf_prog_free(prog->aux);
1713 free_prog_nouncharge:
1714         bpf_prog_free(prog);
1715         return err;
1716 }
1717
1718 #define BPF_OBJ_LAST_FIELD file_flags
1719
1720 static int bpf_obj_pin(const union bpf_attr *attr)
1721 {
1722         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1723                 return -EINVAL;
1724
1725         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1726 }
1727
1728 static int bpf_obj_get(const union bpf_attr *attr)
1729 {
1730         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1731             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1732                 return -EINVAL;
1733
1734         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1735                                 attr->file_flags);
1736 }
1737
1738 struct bpf_raw_tracepoint {
1739         struct bpf_raw_event_map *btp;
1740         struct bpf_prog *prog;
1741 };
1742
1743 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1744 {
1745         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1746
1747         if (raw_tp->prog) {
1748                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1749                 bpf_prog_put(raw_tp->prog);
1750         }
1751         bpf_put_raw_tracepoint(raw_tp->btp);
1752         kfree(raw_tp);
1753         return 0;
1754 }
1755
1756 static const struct file_operations bpf_raw_tp_fops = {
1757         .release        = bpf_raw_tracepoint_release,
1758         .read           = bpf_dummy_read,
1759         .write          = bpf_dummy_write,
1760 };
1761
1762 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1763
1764 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1765 {
1766         struct bpf_raw_tracepoint *raw_tp;
1767         struct bpf_raw_event_map *btp;
1768         struct bpf_prog *prog;
1769         char tp_name[128];
1770         int tp_fd, err;
1771
1772         if (strncpy_from_user(tp_name, u64_to_user_ptr(attr->raw_tracepoint.name),
1773                               sizeof(tp_name) - 1) < 0)
1774                 return -EFAULT;
1775         tp_name[sizeof(tp_name) - 1] = 0;
1776
1777         btp = bpf_get_raw_tracepoint(tp_name);
1778         if (!btp)
1779                 return -ENOENT;
1780
1781         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
1782         if (!raw_tp) {
1783                 err = -ENOMEM;
1784                 goto out_put_btp;
1785         }
1786         raw_tp->btp = btp;
1787
1788         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1789         if (IS_ERR(prog)) {
1790                 err = PTR_ERR(prog);
1791                 goto out_free_tp;
1792         }
1793         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1794             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1795                 err = -EINVAL;
1796                 goto out_put_prog;
1797         }
1798
1799         err = bpf_probe_register(raw_tp->btp, prog);
1800         if (err)
1801                 goto out_put_prog;
1802
1803         raw_tp->prog = prog;
1804         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
1805                                  O_CLOEXEC);
1806         if (tp_fd < 0) {
1807                 bpf_probe_unregister(raw_tp->btp, prog);
1808                 err = tp_fd;
1809                 goto out_put_prog;
1810         }
1811         return tp_fd;
1812
1813 out_put_prog:
1814         bpf_prog_put(prog);
1815 out_free_tp:
1816         kfree(raw_tp);
1817 out_put_btp:
1818         bpf_put_raw_tracepoint(btp);
1819         return err;
1820 }
1821
1822 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
1823                                              enum bpf_attach_type attach_type)
1824 {
1825         switch (prog->type) {
1826         case BPF_PROG_TYPE_CGROUP_SOCK:
1827         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1828                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
1829         default:
1830                 return 0;
1831         }
1832 }
1833
1834 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1835
1836 #define BPF_F_ATTACH_MASK \
1837         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1838
1839 static int bpf_prog_attach(const union bpf_attr *attr)
1840 {
1841         enum bpf_prog_type ptype;
1842         struct bpf_prog *prog;
1843         int ret;
1844
1845         if (!capable(CAP_NET_ADMIN))
1846                 return -EPERM;
1847
1848         if (CHECK_ATTR(BPF_PROG_ATTACH))
1849                 return -EINVAL;
1850
1851         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1852                 return -EINVAL;
1853
1854         switch (attr->attach_type) {
1855         case BPF_CGROUP_INET_INGRESS:
1856         case BPF_CGROUP_INET_EGRESS:
1857                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1858                 break;
1859         case BPF_CGROUP_INET_SOCK_CREATE:
1860         case BPF_CGROUP_INET4_POST_BIND:
1861         case BPF_CGROUP_INET6_POST_BIND:
1862                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1863                 break;
1864         case BPF_CGROUP_INET4_BIND:
1865         case BPF_CGROUP_INET6_BIND:
1866         case BPF_CGROUP_INET4_CONNECT:
1867         case BPF_CGROUP_INET6_CONNECT:
1868         case BPF_CGROUP_UDP4_SENDMSG:
1869         case BPF_CGROUP_UDP6_SENDMSG:
1870                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1871                 break;
1872         case BPF_CGROUP_SOCK_OPS:
1873                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1874                 break;
1875         case BPF_CGROUP_DEVICE:
1876                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1877                 break;
1878         case BPF_SK_MSG_VERDICT:
1879                 ptype = BPF_PROG_TYPE_SK_MSG;
1880                 break;
1881         case BPF_SK_SKB_STREAM_PARSER:
1882         case BPF_SK_SKB_STREAM_VERDICT:
1883                 ptype = BPF_PROG_TYPE_SK_SKB;
1884                 break;
1885         case BPF_LIRC_MODE2:
1886                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
1887                 break;
1888         case BPF_FLOW_DISSECTOR:
1889                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
1890                 break;
1891         case BPF_CGROUP_SYSCTL:
1892                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1893                 break;
1894         default:
1895                 return -EINVAL;
1896         }
1897
1898         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1899         if (IS_ERR(prog))
1900                 return PTR_ERR(prog);
1901
1902         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
1903                 bpf_prog_put(prog);
1904                 return -EINVAL;
1905         }
1906
1907         switch (ptype) {
1908         case BPF_PROG_TYPE_SK_SKB:
1909         case BPF_PROG_TYPE_SK_MSG:
1910                 ret = sock_map_get_from_fd(attr, prog);
1911                 break;
1912         case BPF_PROG_TYPE_LIRC_MODE2:
1913                 ret = lirc_prog_attach(attr, prog);
1914                 break;
1915         case BPF_PROG_TYPE_FLOW_DISSECTOR:
1916                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
1917                 break;
1918         default:
1919                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
1920         }
1921
1922         if (ret)
1923                 bpf_prog_put(prog);
1924         return ret;
1925 }
1926
1927 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1928
1929 static int bpf_prog_detach(const union bpf_attr *attr)
1930 {
1931         enum bpf_prog_type ptype;
1932
1933         if (!capable(CAP_NET_ADMIN))
1934                 return -EPERM;
1935
1936         if (CHECK_ATTR(BPF_PROG_DETACH))
1937                 return -EINVAL;
1938
1939         switch (attr->attach_type) {
1940         case BPF_CGROUP_INET_INGRESS:
1941         case BPF_CGROUP_INET_EGRESS:
1942                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1943                 break;
1944         case BPF_CGROUP_INET_SOCK_CREATE:
1945         case BPF_CGROUP_INET4_POST_BIND:
1946         case BPF_CGROUP_INET6_POST_BIND:
1947                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1948                 break;
1949         case BPF_CGROUP_INET4_BIND:
1950         case BPF_CGROUP_INET6_BIND:
1951         case BPF_CGROUP_INET4_CONNECT:
1952         case BPF_CGROUP_INET6_CONNECT:
1953         case BPF_CGROUP_UDP4_SENDMSG:
1954         case BPF_CGROUP_UDP6_SENDMSG:
1955                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
1956                 break;
1957         case BPF_CGROUP_SOCK_OPS:
1958                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1959                 break;
1960         case BPF_CGROUP_DEVICE:
1961                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1962                 break;
1963         case BPF_SK_MSG_VERDICT:
1964                 return sock_map_get_from_fd(attr, NULL);
1965         case BPF_SK_SKB_STREAM_PARSER:
1966         case BPF_SK_SKB_STREAM_VERDICT:
1967                 return sock_map_get_from_fd(attr, NULL);
1968         case BPF_LIRC_MODE2:
1969                 return lirc_prog_detach(attr);
1970         case BPF_FLOW_DISSECTOR:
1971                 return skb_flow_dissector_bpf_prog_detach(attr);
1972         case BPF_CGROUP_SYSCTL:
1973                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
1974                 break;
1975         default:
1976                 return -EINVAL;
1977         }
1978
1979         return cgroup_bpf_prog_detach(attr, ptype);
1980 }
1981
1982 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1983
1984 static int bpf_prog_query(const union bpf_attr *attr,
1985                           union bpf_attr __user *uattr)
1986 {
1987         if (!capable(CAP_NET_ADMIN))
1988                 return -EPERM;
1989         if (CHECK_ATTR(BPF_PROG_QUERY))
1990                 return -EINVAL;
1991         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1992                 return -EINVAL;
1993
1994         switch (attr->query.attach_type) {
1995         case BPF_CGROUP_INET_INGRESS:
1996         case BPF_CGROUP_INET_EGRESS:
1997         case BPF_CGROUP_INET_SOCK_CREATE:
1998         case BPF_CGROUP_INET4_BIND:
1999         case BPF_CGROUP_INET6_BIND:
2000         case BPF_CGROUP_INET4_POST_BIND:
2001         case BPF_CGROUP_INET6_POST_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_SOCK_OPS:
2007         case BPF_CGROUP_DEVICE:
2008         case BPF_CGROUP_SYSCTL:
2009                 break;
2010         case BPF_LIRC_MODE2:
2011                 return lirc_prog_query(attr, uattr);
2012         case BPF_FLOW_DISSECTOR:
2013                 return skb_flow_dissector_prog_query(attr, uattr);
2014         default:
2015                 return -EINVAL;
2016         }
2017
2018         return cgroup_bpf_prog_query(attr, uattr);
2019 }
2020
2021 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2022
2023 static int bpf_prog_test_run(const union bpf_attr *attr,
2024                              union bpf_attr __user *uattr)
2025 {
2026         struct bpf_prog *prog;
2027         int ret = -ENOTSUPP;
2028
2029         if (!capable(CAP_SYS_ADMIN))
2030                 return -EPERM;
2031         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2032                 return -EINVAL;
2033
2034         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2035             (!attr->test.ctx_size_in && attr->test.ctx_in))
2036                 return -EINVAL;
2037
2038         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2039             (!attr->test.ctx_size_out && attr->test.ctx_out))
2040                 return -EINVAL;
2041
2042         prog = bpf_prog_get(attr->test.prog_fd);
2043         if (IS_ERR(prog))
2044                 return PTR_ERR(prog);
2045
2046         if (prog->aux->ops->test_run)
2047                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2048
2049         bpf_prog_put(prog);
2050         return ret;
2051 }
2052
2053 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2054
2055 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2056                                union bpf_attr __user *uattr,
2057                                struct idr *idr,
2058                                spinlock_t *lock)
2059 {
2060         u32 next_id = attr->start_id;
2061         int err = 0;
2062
2063         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2064                 return -EINVAL;
2065
2066         if (!capable(CAP_SYS_ADMIN))
2067                 return -EPERM;
2068
2069         next_id++;
2070         spin_lock_bh(lock);
2071         if (!idr_get_next(idr, &next_id))
2072                 err = -ENOENT;
2073         spin_unlock_bh(lock);
2074
2075         if (!err)
2076                 err = put_user(next_id, &uattr->next_id);
2077
2078         return err;
2079 }
2080
2081 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2082
2083 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2084 {
2085         struct bpf_prog *prog;
2086         u32 id = attr->prog_id;
2087         int fd;
2088
2089         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2090                 return -EINVAL;
2091
2092         if (!capable(CAP_SYS_ADMIN))
2093                 return -EPERM;
2094
2095         spin_lock_bh(&prog_idr_lock);
2096         prog = idr_find(&prog_idr, id);
2097         if (prog)
2098                 prog = bpf_prog_inc_not_zero(prog);
2099         else
2100                 prog = ERR_PTR(-ENOENT);
2101         spin_unlock_bh(&prog_idr_lock);
2102
2103         if (IS_ERR(prog))
2104                 return PTR_ERR(prog);
2105
2106         fd = bpf_prog_new_fd(prog);
2107         if (fd < 0)
2108                 bpf_prog_put(prog);
2109
2110         return fd;
2111 }
2112
2113 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2114
2115 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2116 {
2117         struct bpf_map *map;
2118         u32 id = attr->map_id;
2119         int f_flags;
2120         int fd;
2121
2122         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2123             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2124                 return -EINVAL;
2125
2126         if (!capable(CAP_SYS_ADMIN))
2127                 return -EPERM;
2128
2129         f_flags = bpf_get_file_flag(attr->open_flags);
2130         if (f_flags < 0)
2131                 return f_flags;
2132
2133         spin_lock_bh(&map_idr_lock);
2134         map = idr_find(&map_idr, id);
2135         if (map)
2136                 map = bpf_map_inc_not_zero(map, true);
2137         else
2138                 map = ERR_PTR(-ENOENT);
2139         spin_unlock_bh(&map_idr_lock);
2140
2141         if (IS_ERR(map))
2142                 return PTR_ERR(map);
2143
2144         fd = bpf_map_new_fd(map, f_flags);
2145         if (fd < 0)
2146                 bpf_map_put_with_uref(map);
2147
2148         return fd;
2149 }
2150
2151 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2152                                               unsigned long addr, u32 *off,
2153                                               u32 *type)
2154 {
2155         const struct bpf_map *map;
2156         int i;
2157
2158         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2159                 map = prog->aux->used_maps[i];
2160                 if (map == (void *)addr) {
2161                         *type = BPF_PSEUDO_MAP_FD;
2162                         return map;
2163                 }
2164                 if (!map->ops->map_direct_value_meta)
2165                         continue;
2166                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2167                         *type = BPF_PSEUDO_MAP_VALUE;
2168                         return map;
2169                 }
2170         }
2171
2172         return NULL;
2173 }
2174
2175 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2176 {
2177         const struct bpf_map *map;
2178         struct bpf_insn *insns;
2179         u32 off, type;
2180         u64 imm;
2181         int i;
2182
2183         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2184                         GFP_USER);
2185         if (!insns)
2186                 return insns;
2187
2188         for (i = 0; i < prog->len; i++) {
2189                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2190                         insns[i].code = BPF_JMP | BPF_CALL;
2191                         insns[i].imm = BPF_FUNC_tail_call;
2192                         /* fall-through */
2193                 }
2194                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2195                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2196                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2197                                 insns[i].code = BPF_JMP | BPF_CALL;
2198                         if (!bpf_dump_raw_ok())
2199                                 insns[i].imm = 0;
2200                         continue;
2201                 }
2202
2203                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2204                         continue;
2205
2206                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2207                 map = bpf_map_from_imm(prog, imm, &off, &type);
2208                 if (map) {
2209                         insns[i].src_reg = type;
2210                         insns[i].imm = map->id;
2211                         insns[i + 1].imm = off;
2212                         continue;
2213                 }
2214         }
2215
2216         return insns;
2217 }
2218
2219 static int set_info_rec_size(struct bpf_prog_info *info)
2220 {
2221         /*
2222          * Ensure info.*_rec_size is the same as kernel expected size
2223          *
2224          * or
2225          *
2226          * Only allow zero *_rec_size if both _rec_size and _cnt are
2227          * zero.  In this case, the kernel will set the expected
2228          * _rec_size back to the info.
2229          */
2230
2231         if ((info->nr_func_info || info->func_info_rec_size) &&
2232             info->func_info_rec_size != sizeof(struct bpf_func_info))
2233                 return -EINVAL;
2234
2235         if ((info->nr_line_info || info->line_info_rec_size) &&
2236             info->line_info_rec_size != sizeof(struct bpf_line_info))
2237                 return -EINVAL;
2238
2239         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2240             info->jited_line_info_rec_size != sizeof(__u64))
2241                 return -EINVAL;
2242
2243         info->func_info_rec_size = sizeof(struct bpf_func_info);
2244         info->line_info_rec_size = sizeof(struct bpf_line_info);
2245         info->jited_line_info_rec_size = sizeof(__u64);
2246
2247         return 0;
2248 }
2249
2250 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2251                                    const union bpf_attr *attr,
2252                                    union bpf_attr __user *uattr)
2253 {
2254         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2255         struct bpf_prog_info info = {};
2256         u32 info_len = attr->info.info_len;
2257         struct bpf_prog_stats stats;
2258         char __user *uinsns;
2259         u32 ulen;
2260         int err;
2261
2262         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2263         if (err)
2264                 return err;
2265         info_len = min_t(u32, sizeof(info), info_len);
2266
2267         if (copy_from_user(&info, uinfo, info_len))
2268                 return -EFAULT;
2269
2270         info.type = prog->type;
2271         info.id = prog->aux->id;
2272         info.load_time = prog->aux->load_time;
2273         info.created_by_uid = from_kuid_munged(current_user_ns(),
2274                                                prog->aux->user->uid);
2275         info.gpl_compatible = prog->gpl_compatible;
2276
2277         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2278         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2279
2280         ulen = info.nr_map_ids;
2281         info.nr_map_ids = prog->aux->used_map_cnt;
2282         ulen = min_t(u32, info.nr_map_ids, ulen);
2283         if (ulen) {
2284                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2285                 u32 i;
2286
2287                 for (i = 0; i < ulen; i++)
2288                         if (put_user(prog->aux->used_maps[i]->id,
2289                                      &user_map_ids[i]))
2290                                 return -EFAULT;
2291         }
2292
2293         err = set_info_rec_size(&info);
2294         if (err)
2295                 return err;
2296
2297         bpf_prog_get_stats(prog, &stats);
2298         info.run_time_ns = stats.nsecs;
2299         info.run_cnt = stats.cnt;
2300
2301         if (!capable(CAP_SYS_ADMIN)) {
2302                 info.jited_prog_len = 0;
2303                 info.xlated_prog_len = 0;
2304                 info.nr_jited_ksyms = 0;
2305                 info.nr_jited_func_lens = 0;
2306                 info.nr_func_info = 0;
2307                 info.nr_line_info = 0;
2308                 info.nr_jited_line_info = 0;
2309                 goto done;
2310         }
2311
2312         ulen = info.xlated_prog_len;
2313         info.xlated_prog_len = bpf_prog_insn_size(prog);
2314         if (info.xlated_prog_len && ulen) {
2315                 struct bpf_insn *insns_sanitized;
2316                 bool fault;
2317
2318                 if (prog->blinded && !bpf_dump_raw_ok()) {
2319                         info.xlated_prog_insns = 0;
2320                         goto done;
2321                 }
2322                 insns_sanitized = bpf_insn_prepare_dump(prog);
2323                 if (!insns_sanitized)
2324                         return -ENOMEM;
2325                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2326                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2327                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2328                 kfree(insns_sanitized);
2329                 if (fault)
2330                         return -EFAULT;
2331         }
2332
2333         if (bpf_prog_is_dev_bound(prog->aux)) {
2334                 err = bpf_prog_offload_info_fill(&info, prog);
2335                 if (err)
2336                         return err;
2337                 goto done;
2338         }
2339
2340         /* NOTE: the following code is supposed to be skipped for offload.
2341          * bpf_prog_offload_info_fill() is the place to fill similar fields
2342          * for offload.
2343          */
2344         ulen = info.jited_prog_len;
2345         if (prog->aux->func_cnt) {
2346                 u32 i;
2347
2348                 info.jited_prog_len = 0;
2349                 for (i = 0; i < prog->aux->func_cnt; i++)
2350                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2351         } else {
2352                 info.jited_prog_len = prog->jited_len;
2353         }
2354
2355         if (info.jited_prog_len && ulen) {
2356                 if (bpf_dump_raw_ok()) {
2357                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2358                         ulen = min_t(u32, info.jited_prog_len, ulen);
2359
2360                         /* for multi-function programs, copy the JITed
2361                          * instructions for all the functions
2362                          */
2363                         if (prog->aux->func_cnt) {
2364                                 u32 len, free, i;
2365                                 u8 *img;
2366
2367                                 free = ulen;
2368                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2369                                         len = prog->aux->func[i]->jited_len;
2370                                         len = min_t(u32, len, free);
2371                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2372                                         if (copy_to_user(uinsns, img, len))
2373                                                 return -EFAULT;
2374                                         uinsns += len;
2375                                         free -= len;
2376                                         if (!free)
2377                                                 break;
2378                                 }
2379                         } else {
2380                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2381                                         return -EFAULT;
2382                         }
2383                 } else {
2384                         info.jited_prog_insns = 0;
2385                 }
2386         }
2387
2388         ulen = info.nr_jited_ksyms;
2389         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2390         if (ulen) {
2391                 if (bpf_dump_raw_ok()) {
2392                         unsigned long ksym_addr;
2393                         u64 __user *user_ksyms;
2394                         u32 i;
2395
2396                         /* copy the address of the kernel symbol
2397                          * corresponding to each function
2398                          */
2399                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2400                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2401                         if (prog->aux->func_cnt) {
2402                                 for (i = 0; i < ulen; i++) {
2403                                         ksym_addr = (unsigned long)
2404                                                 prog->aux->func[i]->bpf_func;
2405                                         if (put_user((u64) ksym_addr,
2406                                                      &user_ksyms[i]))
2407                                                 return -EFAULT;
2408                                 }
2409                         } else {
2410                                 ksym_addr = (unsigned long) prog->bpf_func;
2411                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2412                                         return -EFAULT;
2413                         }
2414                 } else {
2415                         info.jited_ksyms = 0;
2416                 }
2417         }
2418
2419         ulen = info.nr_jited_func_lens;
2420         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2421         if (ulen) {
2422                 if (bpf_dump_raw_ok()) {
2423                         u32 __user *user_lens;
2424                         u32 func_len, i;
2425
2426                         /* copy the JITed image lengths for each function */
2427                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2428                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2429                         if (prog->aux->func_cnt) {
2430                                 for (i = 0; i < ulen; i++) {
2431                                         func_len =
2432                                                 prog->aux->func[i]->jited_len;
2433                                         if (put_user(func_len, &user_lens[i]))
2434                                                 return -EFAULT;
2435                                 }
2436                         } else {
2437                                 func_len = prog->jited_len;
2438                                 if (put_user(func_len, &user_lens[0]))
2439                                         return -EFAULT;
2440                         }
2441                 } else {
2442                         info.jited_func_lens = 0;
2443                 }
2444         }
2445
2446         if (prog->aux->btf)
2447                 info.btf_id = btf_id(prog->aux->btf);
2448
2449         ulen = info.nr_func_info;
2450         info.nr_func_info = prog->aux->func_info_cnt;
2451         if (info.nr_func_info && ulen) {
2452                 char __user *user_finfo;
2453
2454                 user_finfo = u64_to_user_ptr(info.func_info);
2455                 ulen = min_t(u32, info.nr_func_info, ulen);
2456                 if (copy_to_user(user_finfo, prog->aux->func_info,
2457                                  info.func_info_rec_size * ulen))
2458                         return -EFAULT;
2459         }
2460
2461         ulen = info.nr_line_info;
2462         info.nr_line_info = prog->aux->nr_linfo;
2463         if (info.nr_line_info && ulen) {
2464                 __u8 __user *user_linfo;
2465
2466                 user_linfo = u64_to_user_ptr(info.line_info);
2467                 ulen = min_t(u32, info.nr_line_info, ulen);
2468                 if (copy_to_user(user_linfo, prog->aux->linfo,
2469                                  info.line_info_rec_size * ulen))
2470                         return -EFAULT;
2471         }
2472
2473         ulen = info.nr_jited_line_info;
2474         if (prog->aux->jited_linfo)
2475                 info.nr_jited_line_info = prog->aux->nr_linfo;
2476         else
2477                 info.nr_jited_line_info = 0;
2478         if (info.nr_jited_line_info && ulen) {
2479                 if (bpf_dump_raw_ok()) {
2480                         __u64 __user *user_linfo;
2481                         u32 i;
2482
2483                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2484                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2485                         for (i = 0; i < ulen; i++) {
2486                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2487                                              &user_linfo[i]))
2488                                         return -EFAULT;
2489                         }
2490                 } else {
2491                         info.jited_line_info = 0;
2492                 }
2493         }
2494
2495         ulen = info.nr_prog_tags;
2496         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2497         if (ulen) {
2498                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2499                 u32 i;
2500
2501                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2502                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2503                 if (prog->aux->func_cnt) {
2504                         for (i = 0; i < ulen; i++) {
2505                                 if (copy_to_user(user_prog_tags[i],
2506                                                  prog->aux->func[i]->tag,
2507                                                  BPF_TAG_SIZE))
2508                                         return -EFAULT;
2509                         }
2510                 } else {
2511                         if (copy_to_user(user_prog_tags[0],
2512                                          prog->tag, BPF_TAG_SIZE))
2513                                 return -EFAULT;
2514                 }
2515         }
2516
2517 done:
2518         if (copy_to_user(uinfo, &info, info_len) ||
2519             put_user(info_len, &uattr->info.info_len))
2520                 return -EFAULT;
2521
2522         return 0;
2523 }
2524
2525 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2526                                   const union bpf_attr *attr,
2527                                   union bpf_attr __user *uattr)
2528 {
2529         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2530         struct bpf_map_info info = {};
2531         u32 info_len = attr->info.info_len;
2532         int err;
2533
2534         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2535         if (err)
2536                 return err;
2537         info_len = min_t(u32, sizeof(info), info_len);
2538
2539         info.type = map->map_type;
2540         info.id = map->id;
2541         info.key_size = map->key_size;
2542         info.value_size = map->value_size;
2543         info.max_entries = map->max_entries;
2544         info.map_flags = map->map_flags;
2545         memcpy(info.name, map->name, sizeof(map->name));
2546
2547         if (map->btf) {
2548                 info.btf_id = btf_id(map->btf);
2549                 info.btf_key_type_id = map->btf_key_type_id;
2550                 info.btf_value_type_id = map->btf_value_type_id;
2551         }
2552
2553         if (bpf_map_is_dev_bound(map)) {
2554                 err = bpf_map_offload_info_fill(&info, map);
2555                 if (err)
2556                         return err;
2557         }
2558
2559         if (copy_to_user(uinfo, &info, info_len) ||
2560             put_user(info_len, &uattr->info.info_len))
2561                 return -EFAULT;
2562
2563         return 0;
2564 }
2565
2566 static int bpf_btf_get_info_by_fd(struct btf *btf,
2567                                   const union bpf_attr *attr,
2568                                   union bpf_attr __user *uattr)
2569 {
2570         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2571         u32 info_len = attr->info.info_len;
2572         int err;
2573
2574         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2575         if (err)
2576                 return err;
2577
2578         return btf_get_info_by_fd(btf, attr, uattr);
2579 }
2580
2581 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2582
2583 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2584                                   union bpf_attr __user *uattr)
2585 {
2586         int ufd = attr->info.bpf_fd;
2587         struct fd f;
2588         int err;
2589
2590         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2591                 return -EINVAL;
2592
2593         f = fdget(ufd);
2594         if (!f.file)
2595                 return -EBADFD;
2596
2597         if (f.file->f_op == &bpf_prog_fops)
2598                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2599                                               uattr);
2600         else if (f.file->f_op == &bpf_map_fops)
2601                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2602                                              uattr);
2603         else if (f.file->f_op == &btf_fops)
2604                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2605         else
2606                 err = -EINVAL;
2607
2608         fdput(f);
2609         return err;
2610 }
2611
2612 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2613
2614 static int bpf_btf_load(const union bpf_attr *attr)
2615 {
2616         if (CHECK_ATTR(BPF_BTF_LOAD))
2617                 return -EINVAL;
2618
2619         if (!capable(CAP_SYS_ADMIN))
2620                 return -EPERM;
2621
2622         return btf_new_fd(attr);
2623 }
2624
2625 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2626
2627 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2628 {
2629         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2630                 return -EINVAL;
2631
2632         if (!capable(CAP_SYS_ADMIN))
2633                 return -EPERM;
2634
2635         return btf_get_fd_by_id(attr->btf_id);
2636 }
2637
2638 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2639                                     union bpf_attr __user *uattr,
2640                                     u32 prog_id, u32 fd_type,
2641                                     const char *buf, u64 probe_offset,
2642                                     u64 probe_addr)
2643 {
2644         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2645         u32 len = buf ? strlen(buf) : 0, input_len;
2646         int err = 0;
2647
2648         if (put_user(len, &uattr->task_fd_query.buf_len))
2649                 return -EFAULT;
2650         input_len = attr->task_fd_query.buf_len;
2651         if (input_len && ubuf) {
2652                 if (!len) {
2653                         /* nothing to copy, just make ubuf NULL terminated */
2654                         char zero = '\0';
2655
2656                         if (put_user(zero, ubuf))
2657                                 return -EFAULT;
2658                 } else if (input_len >= len + 1) {
2659                         /* ubuf can hold the string with NULL terminator */
2660                         if (copy_to_user(ubuf, buf, len + 1))
2661                                 return -EFAULT;
2662                 } else {
2663                         /* ubuf cannot hold the string with NULL terminator,
2664                          * do a partial copy with NULL terminator.
2665                          */
2666                         char zero = '\0';
2667
2668                         err = -ENOSPC;
2669                         if (copy_to_user(ubuf, buf, input_len - 1))
2670                                 return -EFAULT;
2671                         if (put_user(zero, ubuf + input_len - 1))
2672                                 return -EFAULT;
2673                 }
2674         }
2675
2676         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2677             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2678             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2679             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2680                 return -EFAULT;
2681
2682         return err;
2683 }
2684
2685 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2686
2687 static int bpf_task_fd_query(const union bpf_attr *attr,
2688                              union bpf_attr __user *uattr)
2689 {
2690         pid_t pid = attr->task_fd_query.pid;
2691         u32 fd = attr->task_fd_query.fd;
2692         const struct perf_event *event;
2693         struct files_struct *files;
2694         struct task_struct *task;
2695         struct file *file;
2696         int err;
2697
2698         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2699                 return -EINVAL;
2700
2701         if (!capable(CAP_SYS_ADMIN))
2702                 return -EPERM;
2703
2704         if (attr->task_fd_query.flags != 0)
2705                 return -EINVAL;
2706
2707         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2708         if (!task)
2709                 return -ENOENT;
2710
2711         files = get_files_struct(task);
2712         put_task_struct(task);
2713         if (!files)
2714                 return -ENOENT;
2715
2716         err = 0;
2717         spin_lock(&files->file_lock);
2718         file = fcheck_files(files, fd);
2719         if (!file)
2720                 err = -EBADF;
2721         else
2722                 get_file(file);
2723         spin_unlock(&files->file_lock);
2724         put_files_struct(files);
2725
2726         if (err)
2727                 goto out;
2728
2729         if (file->f_op == &bpf_raw_tp_fops) {
2730                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2731                 struct bpf_raw_event_map *btp = raw_tp->btp;
2732
2733                 err = bpf_task_fd_query_copy(attr, uattr,
2734                                              raw_tp->prog->aux->id,
2735                                              BPF_FD_TYPE_RAW_TRACEPOINT,
2736                                              btp->tp->name, 0, 0);
2737                 goto put_file;
2738         }
2739
2740         event = perf_get_event(file);
2741         if (!IS_ERR(event)) {
2742                 u64 probe_offset, probe_addr;
2743                 u32 prog_id, fd_type;
2744                 const char *buf;
2745
2746                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
2747                                               &buf, &probe_offset,
2748                                               &probe_addr);
2749                 if (!err)
2750                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
2751                                                      fd_type, buf,
2752                                                      probe_offset,
2753                                                      probe_addr);
2754                 goto put_file;
2755         }
2756
2757         err = -ENOTSUPP;
2758 put_file:
2759         fput(file);
2760 out:
2761         return err;
2762 }
2763
2764 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
2765 {
2766         union bpf_attr attr = {};
2767         int err;
2768
2769         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
2770                 return -EPERM;
2771
2772         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
2773         if (err)
2774                 return err;
2775         size = min_t(u32, size, sizeof(attr));
2776
2777         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
2778         if (copy_from_user(&attr, uattr, size) != 0)
2779                 return -EFAULT;
2780
2781         err = security_bpf(cmd, &attr, size);
2782         if (err < 0)
2783                 return err;
2784
2785         switch (cmd) {
2786         case BPF_MAP_CREATE:
2787                 err = map_create(&attr);
2788                 break;
2789         case BPF_MAP_LOOKUP_ELEM:
2790                 err = map_lookup_elem(&attr);
2791                 break;
2792         case BPF_MAP_UPDATE_ELEM:
2793                 err = map_update_elem(&attr);
2794                 break;
2795         case BPF_MAP_DELETE_ELEM:
2796                 err = map_delete_elem(&attr);
2797                 break;
2798         case BPF_MAP_GET_NEXT_KEY:
2799                 err = map_get_next_key(&attr);
2800                 break;
2801         case BPF_MAP_FREEZE:
2802                 err = map_freeze(&attr);
2803                 break;
2804         case BPF_PROG_LOAD:
2805                 err = bpf_prog_load(&attr, uattr);
2806                 break;
2807         case BPF_OBJ_PIN:
2808                 err = bpf_obj_pin(&attr);
2809                 break;
2810         case BPF_OBJ_GET:
2811                 err = bpf_obj_get(&attr);
2812                 break;
2813         case BPF_PROG_ATTACH:
2814                 err = bpf_prog_attach(&attr);
2815                 break;
2816         case BPF_PROG_DETACH:
2817                 err = bpf_prog_detach(&attr);
2818                 break;
2819         case BPF_PROG_QUERY:
2820                 err = bpf_prog_query(&attr, uattr);
2821                 break;
2822         case BPF_PROG_TEST_RUN:
2823                 err = bpf_prog_test_run(&attr, uattr);
2824                 break;
2825         case BPF_PROG_GET_NEXT_ID:
2826                 err = bpf_obj_get_next_id(&attr, uattr,
2827                                           &prog_idr, &prog_idr_lock);
2828                 break;
2829         case BPF_MAP_GET_NEXT_ID:
2830                 err = bpf_obj_get_next_id(&attr, uattr,
2831                                           &map_idr, &map_idr_lock);
2832                 break;
2833         case BPF_PROG_GET_FD_BY_ID:
2834                 err = bpf_prog_get_fd_by_id(&attr);
2835                 break;
2836         case BPF_MAP_GET_FD_BY_ID:
2837                 err = bpf_map_get_fd_by_id(&attr);
2838                 break;
2839         case BPF_OBJ_GET_INFO_BY_FD:
2840                 err = bpf_obj_get_info_by_fd(&attr, uattr);
2841                 break;
2842         case BPF_RAW_TRACEPOINT_OPEN:
2843                 err = bpf_raw_tracepoint_open(&attr);
2844                 break;
2845         case BPF_BTF_LOAD:
2846                 err = bpf_btf_load(&attr);
2847                 break;
2848         case BPF_BTF_GET_FD_BY_ID:
2849                 err = bpf_btf_get_fd_by_id(&attr);
2850                 break;
2851         case BPF_TASK_FD_QUERY:
2852                 err = bpf_task_fd_query(&attr, uattr);
2853                 break;
2854         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
2855                 err = map_lookup_and_delete_elem(&attr);
2856                 break;
2857         default:
2858                 err = -EINVAL;
2859                 break;
2860         }
2861
2862         return err;
2863 }