kernel/bpf/syscall.c

   1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  *
   7  * This program is distributed in the hope that it will be useful, but
   8  * WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  10  * General Public License for more details.
  11  */
  12 #include <linux/bpf.h>
  13 #include <linux/bpf_trace.h>
  14 #include <linux/syscalls.h>
  15 #include <linux/slab.h>
  16 #include <linux/sched/signal.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/mmzone.h>
  19 #include <linux/anon_inodes.h>
  20 #include <linux/file.h>
  21 #include <linux/license.h>
  22 #include <linux/filter.h>
  23 #include <linux/version.h>
  24 #include <linux/kernel.h>
  25 #include <linux/idr.h>
  26 #include <linux/cred.h>
  27 #include <linux/timekeeping.h>
  28 #include <linux/ctype.h>
  29
  30 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY || \
  31                            (map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  32                            (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  33                            (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  34 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  35 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_HASH(map))
  36
  37 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  38
  39 DEFINE_PER_CPU(int, bpf_prog_active);
  40 static DEFINE_IDR(prog_idr);
  41 static DEFINE_SPINLOCK(prog_idr_lock);
  42 static DEFINE_IDR(map_idr);
  43 static DEFINE_SPINLOCK(map_idr_lock);
  44
  45 int sysctl_unprivileged_bpf_disabled __read_mostly;
  46
  47 static const struct bpf_map_ops * const bpf_map_types[] = {
  48 #define BPF_PROG_TYPE(_id, _ops)
  49 #define BPF_MAP_TYPE(_id, _ops) \
  50         [_id] = &_ops,
  51 #include <linux/bpf_types.h>
  52 #undef BPF_PROG_TYPE
  53 #undef BPF_MAP_TYPE
  54 };
  55
  56 /*
  57  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  58  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  59  * we don't know about yet.
  60  *
  61  * There is a ToCToU between this function call and the following
  62  * copy_from_user() call. However, this is not a concern since this function is
  63  * meant to be a future-proofing of bits.
  64  */
  65 static int check_uarg_tail_zero(void __user *uaddr,
  66                                 size_t expected_size,
  67                                 size_t actual_size)
  68 {
  69         unsigned char __user *addr;
  70         unsigned char __user *end;
  71         unsigned char val;
  72         int err;
  73
  74         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  75                 return -E2BIG;
  76
  77         if (unlikely(!access_ok(VERIFY_READ, uaddr, actual_size)))
  78                 return -EFAULT;
  79
  80         if (actual_size <= expected_size)
  81                 return 0;
  82
  83         addr = uaddr + expected_size;
  84         end  = uaddr + actual_size;
  85
  86         for (; addr < end; addr++) {
  87                 err = get_user(val, addr);
  88                 if (err)
  89                         return err;
  90                 if (val)
  91                         return -E2BIG;
  92         }
  93
  94         return 0;
  95 }
  96
  97 const struct bpf_map_ops bpf_map_offload_ops = {
  98         .map_alloc = bpf_map_offload_map_alloc,
  99         .map_free = bpf_map_offload_map_free,
 100 };
 101
 102 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 103 {
 104         const struct bpf_map_ops *ops;
 105         struct bpf_map *map;
 106         int err;
 107
 108         if (attr->map_type >= ARRAY_SIZE(bpf_map_types))
 109                 return ERR_PTR(-EINVAL);
 110         ops = bpf_map_types[attr->map_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 = attr->map_type;
 126         return map;
 127 }
 128
 129 void *bpf_map_area_alloc(size_t size, int numa_node)
 130 {
 131         /* We definitely need __GFP_NORETRY, so OOM killer doesn't
 132          * trigger under memory pressure as we really just want to
 133          * fail instead.
 134          */
 135         const gfp_t flags = __GFP_NOWARN | __GFP_NORETRY | __GFP_ZERO;
 136         void *area;
 137
 138         if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 139                 area = kmalloc_node(size, GFP_USER | flags, numa_node);
 140                 if (area != NULL)
 141                         return area;
 142         }
 143
 144         return __vmalloc_node_flags_caller(size, numa_node, GFP_KERNEL | flags,
 145                                            __builtin_return_address(0));
 146 }
 147
 148 void bpf_map_area_free(void *area)
 149 {
 150         kvfree(area);
 151 }
 152
 153 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 154 {
 155         map->map_type = attr->map_type;
 156         map->key_size = attr->key_size;
 157         map->value_size = attr->value_size;
 158         map->max_entries = attr->max_entries;
 159         map->map_flags = attr->map_flags;
 160         map->numa_node = bpf_map_attr_numa_node(attr);
 161 }
 162
 163 int bpf_map_precharge_memlock(u32 pages)
 164 {
 165         struct user_struct *user = get_current_user();
 166         unsigned long memlock_limit, cur;
 167
 168         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 169         cur = atomic_long_read(&user->locked_vm);
 170         free_uid(user);
 171         if (cur + pages > memlock_limit)
 172                 return -EPERM;
 173         return 0;
 174 }
 175
 176 static int bpf_map_charge_memlock(struct bpf_map *map)
 177 {
 178         struct user_struct *user = get_current_user();
 179         unsigned long memlock_limit;
 180
 181         memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 182
 183         atomic_long_add(map->pages, &user->locked_vm);
 184
 185         if (atomic_long_read(&user->locked_vm) > memlock_limit) {
 186                 atomic_long_sub(map->pages, &user->locked_vm);
 187                 free_uid(user);
 188                 return -EPERM;
 189         }
 190         map->user = user;
 191         return 0;
 192 }
 193
 194 static void bpf_map_uncharge_memlock(struct bpf_map *map)
 195 {
 196         struct user_struct *user = map->user;
 197
 198         atomic_long_sub(map->pages, &user->locked_vm);
 199         free_uid(user);
 200 }
 201
 202 static int bpf_map_alloc_id(struct bpf_map *map)
 203 {
 204         int id;
 205
 206         spin_lock_bh(&map_idr_lock);
 207         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 208         if (id > 0)
 209                 map->id = id;
 210         spin_unlock_bh(&map_idr_lock);
 211
 212         if (WARN_ON_ONCE(!id))
 213                 return -ENOSPC;
 214
 215         return id > 0 ? 0 : id;
 216 }
 217
 218 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 219 {
 220         unsigned long flags;
 221
 222         /* Offloaded maps are removed from the IDR store when their device
 223          * disappears - even if someone holds an fd to them they are unusable,
 224          * the memory is gone, all ops will fail; they are simply waiting for
 225          * refcnt to drop to be freed.
 226          */
 227         if (!map->id)
 228                 return;
 229
 230         if (do_idr_lock)
 231                 spin_lock_irqsave(&map_idr_lock, flags);
 232         else
 233                 __acquire(&map_idr_lock);
 234
 235         idr_remove(&map_idr, map->id);
 236         map->id = 0;
 237
 238         if (do_idr_lock)
 239                 spin_unlock_irqrestore(&map_idr_lock, flags);
 240         else
 241                 __release(&map_idr_lock);
 242 }
 243
 244 /* called from workqueue */
 245 static void bpf_map_free_deferred(struct work_struct *work)
 246 {
 247         struct bpf_map *map = container_of(work, struct bpf_map, work);
 248
 249         bpf_map_uncharge_memlock(map);
 250         security_bpf_map_free(map);
 251         /* implementation dependent freeing */
 252         map->ops->map_free(map);
 253 }
 254
 255 static void bpf_map_put_uref(struct bpf_map *map)
 256 {
 257         if (atomic_dec_and_test(&map->usercnt)) {
 258                 if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY)
 259                         bpf_fd_array_map_clear(map);
 260         }
 261 }
 262
 263 /* decrement map refcnt and schedule it for freeing via workqueue
 264  * (unrelying map implementation ops->map_free() might sleep)
 265  */
 266 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 267 {
 268         if (atomic_dec_and_test(&map->refcnt)) {
 269                 /* bpf_map_free_id() must be called first */
 270                 bpf_map_free_id(map, do_idr_lock);
 271                 INIT_WORK(&map->work, bpf_map_free_deferred);
 272                 schedule_work(&map->work);
 273         }
 274 }
 275
 276 void bpf_map_put(struct bpf_map *map)
 277 {
 278         __bpf_map_put(map, true);
 279 }
 280
 281 void bpf_map_put_with_uref(struct bpf_map *map)
 282 {
 283         bpf_map_put_uref(map);
 284         bpf_map_put(map);
 285 }
 286
 287 static int bpf_map_release(struct inode *inode, struct file *filp)
 288 {
 289         struct bpf_map *map = filp->private_data;
 290
 291         if (map->ops->map_release)
 292                 map->ops->map_release(map, filp);
 293
 294         bpf_map_put_with_uref(map);
 295         return 0;
 296 }
 297
 298 #ifdef CONFIG_PROC_FS
 299 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 300 {
 301         const struct bpf_map *map = filp->private_data;
 302         const struct bpf_array *array;
 303         u32 owner_prog_type = 0;
 304         u32 owner_jited = 0;
 305
 306         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 307                 array = container_of(map, struct bpf_array, map);
 308                 owner_prog_type = array->owner_prog_type;
 309                 owner_jited = array->owner_jited;
 310         }
 311
 312         seq_printf(m,
 313                    "map_type:\t%u\n"
 314                    "key_size:\t%u\n"
 315                    "value_size:\t%u\n"
 316                    "max_entries:\t%u\n"
 317                    "map_flags:\t%#x\n"
 318                    "memlock:\t%llu\n",
 319                    map->map_type,
 320                    map->key_size,
 321                    map->value_size,
 322                    map->max_entries,
 323                    map->map_flags,
 324                    map->pages * 1ULL << PAGE_SHIFT);
 325
 326         if (owner_prog_type) {
 327                 seq_printf(m, "owner_prog_type:\t%u\n",
 328                            owner_prog_type);
 329                 seq_printf(m, "owner_jited:\t%u\n",
 330                            owner_jited);
 331         }
 332 }
 333 #endif
 334
 335 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 336                               loff_t *ppos)
 337 {
 338         /* We need this handler such that alloc_file() enables
 339          * f_mode with FMODE_CAN_READ.
 340          */
 341         return -EINVAL;
 342 }
 343
 344 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 345                                size_t siz, loff_t *ppos)
 346 {
 347         /* We need this handler such that alloc_file() enables
 348          * f_mode with FMODE_CAN_WRITE.
 349          */
 350         return -EINVAL;
 351 }
 352
 353 const struct file_operations bpf_map_fops = {
 354 #ifdef CONFIG_PROC_FS
 355         .show_fdinfo    = bpf_map_show_fdinfo,
 356 #endif
 357         .release        = bpf_map_release,
 358         .read           = bpf_dummy_read,
 359         .write          = bpf_dummy_write,
 360 };
 361
 362 int bpf_map_new_fd(struct bpf_map *map, int flags)
 363 {
 364         int ret;
 365
 366         ret = security_bpf_map(map, OPEN_FMODE(flags));
 367         if (ret < 0)
 368                 return ret;
 369
 370         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 371                                 flags | O_CLOEXEC);
 372 }
 373
 374 int bpf_get_file_flag(int flags)
 375 {
 376         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 377                 return -EINVAL;
 378         if (flags & BPF_F_RDONLY)
 379                 return O_RDONLY;
 380         if (flags & BPF_F_WRONLY)
 381                 return O_WRONLY;
 382         return O_RDWR;
 383 }
 384
 385 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 386 #define CHECK_ATTR(CMD) \
 387         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 388                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 389                    sizeof(*attr) - \
 390                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 391                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 392
 393 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 394  * Return 0 on success and < 0 on error.
 395  */
 396 static int bpf_obj_name_cpy(char *dst, const char *src)
 397 {
 398         const char *end = src + BPF_OBJ_NAME_LEN;
 399
 400         memset(dst, 0, BPF_OBJ_NAME_LEN);
 401
 402         /* Copy all isalnum() and '_' char */
 403         while (src < end && *src) {
 404                 if (!isalnum(*src) && *src != '_')
 405                         return -EINVAL;
 406                 *dst++ = *src++;
 407         }
 408
 409         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 410         if (src == end)
 411                 return -EINVAL;
 412
 413         return 0;
 414 }
 415
 416 #define BPF_MAP_CREATE_LAST_FIELD map_ifindex
 417 /* called via syscall */
 418 static int map_create(union bpf_attr *attr)
 419 {
 420         int numa_node = bpf_map_attr_numa_node(attr);
 421         struct bpf_map *map;
 422         int f_flags;
 423         int err;
 424
 425         err = CHECK_ATTR(BPF_MAP_CREATE);
 426         if (err)
 427                 return -EINVAL;
 428
 429         f_flags = bpf_get_file_flag(attr->map_flags);
 430         if (f_flags < 0)
 431                 return f_flags;
 432
 433         if (numa_node != NUMA_NO_NODE &&
 434             ((unsigned int)numa_node >= nr_node_ids ||
 435              !node_online(numa_node)))
 436                 return -EINVAL;
 437
 438         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 439         map = find_and_alloc_map(attr);
 440         if (IS_ERR(map))
 441                 return PTR_ERR(map);
 442
 443         err = bpf_obj_name_cpy(map->name, attr->map_name);
 444         if (err)
 445                 goto free_map_nouncharge;
 446
 447         atomic_set(&map->refcnt, 1);
 448         atomic_set(&map->usercnt, 1);
 449
 450         err = security_bpf_map_alloc(map);
 451         if (err)
 452                 goto free_map_nouncharge;
 453
 454         err = bpf_map_charge_memlock(map);
 455         if (err)
 456                 goto free_map_sec;
 457
 458         err = bpf_map_alloc_id(map);
 459         if (err)
 460                 goto free_map;
 461
 462         err = bpf_map_new_fd(map, f_flags);
 463         if (err < 0) {
 464                 /* failed to allocate fd.
 465                  * bpf_map_put() is needed because the above
 466                  * bpf_map_alloc_id() has published the map
 467                  * to the userspace and the userspace may
 468                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 469                  */
 470                 bpf_map_put(map);
 471                 return err;
 472         }
 473
 474         trace_bpf_map_create(map, err);
 475         return err;
 476
 477 free_map:
 478         bpf_map_uncharge_memlock(map);
 479 free_map_sec:
 480         security_bpf_map_free(map);
 481 free_map_nouncharge:
 482         map->ops->map_free(map);
 483         return err;
 484 }
 485
 486 /* if error is returned, fd is released.
 487  * On success caller should complete fd access with matching fdput()
 488  */
 489 struct bpf_map *__bpf_map_get(struct fd f)
 490 {
 491         if (!f.file)
 492                 return ERR_PTR(-EBADF);
 493         if (f.file->f_op != &bpf_map_fops) {
 494                 fdput(f);
 495                 return ERR_PTR(-EINVAL);
 496         }
 497
 498         return f.file->private_data;
 499 }
 500
 501 /* prog's and map's refcnt limit */
 502 #define BPF_MAX_REFCNT 32768
 503
 504 struct bpf_map *bpf_map_inc(struct bpf_map *map, bool uref)
 505 {
 506         if (atomic_inc_return(&map->refcnt) > BPF_MAX_REFCNT) {
 507                 atomic_dec(&map->refcnt);
 508                 return ERR_PTR(-EBUSY);
 509         }
 510         if (uref)
 511                 atomic_inc(&map->usercnt);
 512         return map;
 513 }
 514
 515 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 516 {
 517         struct fd f = fdget(ufd);
 518         struct bpf_map *map;
 519
 520         map = __bpf_map_get(f);
 521         if (IS_ERR(map))
 522                 return map;
 523
 524         map = bpf_map_inc(map, true);
 525         fdput(f);
 526
 527         return map;
 528 }
 529
 530 /* map_idr_lock should have been held */
 531 static struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map,
 532                                             bool uref)
 533 {
 534         int refold;
 535
 536         refold = __atomic_add_unless(&map->refcnt, 1, 0);
 537
 538         if (refold >= BPF_MAX_REFCNT) {
 539                 __bpf_map_put(map, false);
 540                 return ERR_PTR(-EBUSY);
 541         }
 542
 543         if (!refold)
 544                 return ERR_PTR(-ENOENT);
 545
 546         if (uref)
 547                 atomic_inc(&map->usercnt);
 548
 549         return map;
 550 }
 551
 552 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 553 {
 554         return -ENOTSUPP;
 555 }
 556
 557 /* last field in 'union bpf_attr' used by this command */
 558 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD value
 559
 560 static int map_lookup_elem(union bpf_attr *attr)
 561 {
 562         void __user *ukey = u64_to_user_ptr(attr->key);
 563         void __user *uvalue = u64_to_user_ptr(attr->value);
 564         int ufd = attr->map_fd;
 565         struct bpf_map *map;
 566         void *key, *value, *ptr;
 567         u32 value_size;
 568         struct fd f;
 569         int err;
 570
 571         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 572                 return -EINVAL;
 573
 574         f = fdget(ufd);
 575         map = __bpf_map_get(f);
 576         if (IS_ERR(map))
 577                 return PTR_ERR(map);
 578
 579         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 580                 err = -EPERM;
 581                 goto err_put;
 582         }
 583
 584         key = memdup_user(ukey, map->key_size);
 585         if (IS_ERR(key)) {
 586                 err = PTR_ERR(key);
 587                 goto err_put;
 588         }
 589
 590         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 591             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 592             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 593                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 594         else if (IS_FD_MAP(map))
 595                 value_size = sizeof(u32);
 596         else
 597                 value_size = map->value_size;
 598
 599         err = -ENOMEM;
 600         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 601         if (!value)
 602                 goto free_key;
 603
 604         if (bpf_map_is_dev_bound(map)) {
 605                 err = bpf_map_offload_lookup_elem(map, key, value);
 606         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 607                    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 608                 err = bpf_percpu_hash_copy(map, key, value);
 609         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 610                 err = bpf_percpu_array_copy(map, key, value);
 611         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 612                 err = bpf_stackmap_copy(map, key, value);
 613         } else if (IS_FD_ARRAY(map)) {
 614                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 615         } else if (IS_FD_HASH(map)) {
 616                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 617         } else {
 618                 rcu_read_lock();
 619                 ptr = map->ops->map_lookup_elem(map, key);
 620                 if (ptr)
 621                         memcpy(value, ptr, value_size);
 622                 rcu_read_unlock();
 623                 err = ptr ? 0 : -ENOENT;
 624         }
 625
 626         if (err)
 627                 goto free_value;
 628
 629         err = -EFAULT;
 630         if (copy_to_user(uvalue, value, value_size) != 0)
 631                 goto free_value;
 632
 633         trace_bpf_map_lookup_elem(map, ufd, key, value);
 634         err = 0;
 635
 636 free_value:
 637         kfree(value);
 638 free_key:
 639         kfree(key);
 640 err_put:
 641         fdput(f);
 642         return err;
 643 }
 644
 645 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 646
 647 static int map_update_elem(union bpf_attr *attr)
 648 {
 649         void __user *ukey = u64_to_user_ptr(attr->key);
 650         void __user *uvalue = u64_to_user_ptr(attr->value);
 651         int ufd = attr->map_fd;
 652         struct bpf_map *map;
 653         void *key, *value;
 654         u32 value_size;
 655         struct fd f;
 656         int err;
 657
 658         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 659                 return -EINVAL;
 660
 661         f = fdget(ufd);
 662         map = __bpf_map_get(f);
 663         if (IS_ERR(map))
 664                 return PTR_ERR(map);
 665
 666         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 667                 err = -EPERM;
 668                 goto err_put;
 669         }
 670
 671         key = memdup_user(ukey, map->key_size);
 672         if (IS_ERR(key)) {
 673                 err = PTR_ERR(key);
 674                 goto err_put;
 675         }
 676
 677         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 678             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 679             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY)
 680                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 681         else
 682                 value_size = map->value_size;
 683
 684         err = -ENOMEM;
 685         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 686         if (!value)
 687                 goto free_key;
 688
 689         err = -EFAULT;
 690         if (copy_from_user(value, uvalue, value_size) != 0)
 691                 goto free_value;
 692
 693         /* Need to create a kthread, thus must support schedule */
 694         if (bpf_map_is_dev_bound(map)) {
 695                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
 696                 goto out;
 697         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP) {
 698                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 699                 goto out;
 700         }
 701
 702         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
 703          * inside bpf map update or delete otherwise deadlocks are possible
 704          */
 705         preempt_disable();
 706         __this_cpu_inc(bpf_prog_active);
 707         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 708             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 709                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
 710         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 711                 err = bpf_percpu_array_update(map, key, value, attr->flags);
 712         } else if (IS_FD_ARRAY(map)) {
 713                 rcu_read_lock();
 714                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 715                                                    attr->flags);
 716                 rcu_read_unlock();
 717         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 718                 rcu_read_lock();
 719                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 720                                                   attr->flags);
 721                 rcu_read_unlock();
 722         } else {
 723                 rcu_read_lock();
 724                 err = map->ops->map_update_elem(map, key, value, attr->flags);
 725                 rcu_read_unlock();
 726         }
 727         __this_cpu_dec(bpf_prog_active);
 728         preempt_enable();
 729 out:
 730         if (!err)
 731                 trace_bpf_map_update_elem(map, ufd, key, value);
 732 free_value:
 733         kfree(value);
 734 free_key:
 735         kfree(key);
 736 err_put:
 737         fdput(f);
 738         return err;
 739 }
 740
 741 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
 742
 743 static int map_delete_elem(union bpf_attr *attr)
 744 {
 745         void __user *ukey = u64_to_user_ptr(attr->key);
 746         int ufd = attr->map_fd;
 747         struct bpf_map *map;
 748         struct fd f;
 749         void *key;
 750         int err;
 751
 752         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
 753                 return -EINVAL;
 754
 755         f = fdget(ufd);
 756         map = __bpf_map_get(f);
 757         if (IS_ERR(map))
 758                 return PTR_ERR(map);
 759
 760         if (!(f.file->f_mode & FMODE_CAN_WRITE)) {
 761                 err = -EPERM;
 762                 goto err_put;
 763         }
 764
 765         key = memdup_user(ukey, map->key_size);
 766         if (IS_ERR(key)) {
 767                 err = PTR_ERR(key);
 768                 goto err_put;
 769         }
 770
 771         if (bpf_map_is_dev_bound(map)) {
 772                 err = bpf_map_offload_delete_elem(map, key);
 773                 goto out;
 774         }
 775
 776         preempt_disable();
 777         __this_cpu_inc(bpf_prog_active);
 778         rcu_read_lock();
 779         err = map->ops->map_delete_elem(map, key);
 780         rcu_read_unlock();
 781         __this_cpu_dec(bpf_prog_active);
 782         preempt_enable();
 783 out:
 784         if (!err)
 785                 trace_bpf_map_delete_elem(map, ufd, key);
 786         kfree(key);
 787 err_put:
 788         fdput(f);
 789         return err;
 790 }
 791
 792 /* last field in 'union bpf_attr' used by this command */
 793 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
 794
 795 static int map_get_next_key(union bpf_attr *attr)
 796 {
 797         void __user *ukey = u64_to_user_ptr(attr->key);
 798         void __user *unext_key = u64_to_user_ptr(attr->next_key);
 799         int ufd = attr->map_fd;
 800         struct bpf_map *map;
 801         void *key, *next_key;
 802         struct fd f;
 803         int err;
 804
 805         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
 806                 return -EINVAL;
 807
 808         f = fdget(ufd);
 809         map = __bpf_map_get(f);
 810         if (IS_ERR(map))
 811                 return PTR_ERR(map);
 812
 813         if (!(f.file->f_mode & FMODE_CAN_READ)) {
 814                 err = -EPERM;
 815                 goto err_put;
 816         }
 817
 818         if (ukey) {
 819                 key = memdup_user(ukey, map->key_size);
 820                 if (IS_ERR(key)) {
 821                         err = PTR_ERR(key);
 822                         goto err_put;
 823                 }
 824         } else {
 825                 key = NULL;
 826         }
 827
 828         err = -ENOMEM;
 829         next_key = kmalloc(map->key_size, GFP_USER);
 830         if (!next_key)
 831                 goto free_key;
 832
 833         if (bpf_map_is_dev_bound(map)) {
 834                 err = bpf_map_offload_get_next_key(map, key, next_key);
 835                 goto out;
 836         }
 837
 838         rcu_read_lock();
 839         err = map->ops->map_get_next_key(map, key, next_key);
 840         rcu_read_unlock();
 841 out:
 842         if (err)
 843                 goto free_next_key;
 844
 845         err = -EFAULT;
 846         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
 847                 goto free_next_key;
 848
 849         trace_bpf_map_next_key(map, ufd, key, next_key);
 850         err = 0;
 851
 852 free_next_key:
 853         kfree(next_key);
 854 free_key:
 855         kfree(key);
 856 err_put:
 857         fdput(f);
 858         return err;
 859 }
 860
 861 static const struct bpf_prog_ops * const bpf_prog_types[] = {
 862 #define BPF_PROG_TYPE(_id, _name) \
 863         [_id] = & _name ## _prog_ops,
 864 #define BPF_MAP_TYPE(_id, _ops)
 865 #include <linux/bpf_types.h>
 866 #undef BPF_PROG_TYPE
 867 #undef BPF_MAP_TYPE
 868 };
 869
 870 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
 871 {
 872         if (type >= ARRAY_SIZE(bpf_prog_types) || !bpf_prog_types[type])
 873                 return -EINVAL;
 874
 875         if (!bpf_prog_is_dev_bound(prog->aux))
 876                 prog->aux->ops = bpf_prog_types[type];
 877         else
 878                 prog->aux->ops = &bpf_offload_prog_ops;
 879         prog->type = type;
 880         return 0;
 881 }
 882
 883 /* drop refcnt on maps used by eBPF program and free auxilary data */
 884 static void free_used_maps(struct bpf_prog_aux *aux)
 885 {
 886         int i;
 887
 888         for (i = 0; i < aux->used_map_cnt; i++)
 889                 bpf_map_put(aux->used_maps[i]);
 890
 891         kfree(aux->used_maps);
 892 }
 893
 894 int __bpf_prog_charge(struct user_struct *user, u32 pages)
 895 {
 896         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 897         unsigned long user_bufs;
 898
 899         if (user) {
 900                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
 901                 if (user_bufs > memlock_limit) {
 902                         atomic_long_sub(pages, &user->locked_vm);
 903                         return -EPERM;
 904                 }
 905         }
 906
 907         return 0;
 908 }
 909
 910 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
 911 {
 912         if (user)
 913                 atomic_long_sub(pages, &user->locked_vm);
 914 }
 915
 916 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
 917 {
 918         struct user_struct *user = get_current_user();
 919         int ret;
 920
 921         ret = __bpf_prog_charge(user, prog->pages);
 922         if (ret) {
 923                 free_uid(user);
 924                 return ret;
 925         }
 926
 927         prog->aux->user = user;
 928         return 0;
 929 }
 930
 931 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
 932 {
 933         struct user_struct *user = prog->aux->user;
 934
 935         __bpf_prog_uncharge(user, prog->pages);
 936         free_uid(user);
 937 }
 938
 939 static int bpf_prog_alloc_id(struct bpf_prog *prog)
 940 {
 941         int id;
 942
 943         spin_lock_bh(&prog_idr_lock);
 944         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
 945         if (id > 0)
 946                 prog->aux->id = id;
 947         spin_unlock_bh(&prog_idr_lock);
 948
 949         /* id is in [1, INT_MAX) */
 950         if (WARN_ON_ONCE(!id))
 951                 return -ENOSPC;
 952
 953         return id > 0 ? 0 : id;
 954 }
 955
 956 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
 957 {
 958         /* cBPF to eBPF migrations are currently not in the idr store.
 959          * Offloaded programs are removed from the store when their device
 960          * disappears - even if someone grabs an fd to them they are unusable,
 961          * simply waiting for refcnt to drop to be freed.
 962          */
 963         if (!prog->aux->id)
 964                 return;
 965
 966         if (do_idr_lock)
 967                 spin_lock_bh(&prog_idr_lock);
 968         else
 969                 __acquire(&prog_idr_lock);
 970
 971         idr_remove(&prog_idr, prog->aux->id);
 972         prog->aux->id = 0;
 973
 974         if (do_idr_lock)
 975                 spin_unlock_bh(&prog_idr_lock);
 976         else
 977                 __release(&prog_idr_lock);
 978 }
 979
 980 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
 981 {
 982         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
 983
 984         free_used_maps(aux);
 985         bpf_prog_uncharge_memlock(aux->prog);
 986         security_bpf_prog_free(aux);
 987         bpf_prog_free(aux->prog);
 988 }
 989
 990 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
 991 {
 992         if (atomic_dec_and_test(&prog->aux->refcnt)) {
 993                 int i;
 994
 995                 trace_bpf_prog_put_rcu(prog);
 996                 /* bpf_prog_free_id() must be called first */
 997                 bpf_prog_free_id(prog, do_idr_lock);
 998
 999                 for (i = 0; i < prog->aux->func_cnt; i++)
1000                         bpf_prog_kallsyms_del(prog->aux->func[i]);
1001                 bpf_prog_kallsyms_del(prog);
1002
1003                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1004         }
1005 }
1006
1007 void bpf_prog_put(struct bpf_prog *prog)
1008 {
1009         __bpf_prog_put(prog, true);
1010 }
1011 EXPORT_SYMBOL_GPL(bpf_prog_put);
1012
1013 static int bpf_prog_release(struct inode *inode, struct file *filp)
1014 {
1015         struct bpf_prog *prog = filp->private_data;
1016
1017         bpf_prog_put(prog);
1018         return 0;
1019 }
1020
1021 #ifdef CONFIG_PROC_FS
1022 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1023 {
1024         const struct bpf_prog *prog = filp->private_data;
1025         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1026
1027         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1028         seq_printf(m,
1029                    "prog_type:\t%u\n"
1030                    "prog_jited:\t%u\n"
1031                    "prog_tag:\t%s\n"
1032                    "memlock:\t%llu\n",
1033                    prog->type,
1034                    prog->jited,
1035                    prog_tag,
1036                    prog->pages * 1ULL << PAGE_SHIFT);
1037 }
1038 #endif
1039
1040 const struct file_operations bpf_prog_fops = {
1041 #ifdef CONFIG_PROC_FS
1042         .show_fdinfo    = bpf_prog_show_fdinfo,
1043 #endif
1044         .release        = bpf_prog_release,
1045         .read           = bpf_dummy_read,
1046         .write          = bpf_dummy_write,
1047 };
1048
1049 int bpf_prog_new_fd(struct bpf_prog *prog)
1050 {
1051         int ret;
1052
1053         ret = security_bpf_prog(prog);
1054         if (ret < 0)
1055                 return ret;
1056
1057         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1058                                 O_RDWR | O_CLOEXEC);
1059 }
1060
1061 static struct bpf_prog *____bpf_prog_get(struct fd f)
1062 {
1063         if (!f.file)
1064                 return ERR_PTR(-EBADF);
1065         if (f.file->f_op != &bpf_prog_fops) {
1066                 fdput(f);
1067                 return ERR_PTR(-EINVAL);
1068         }
1069
1070         return f.file->private_data;
1071 }
1072
1073 struct bpf_prog *bpf_prog_add(struct bpf_prog *prog, int i)
1074 {
1075         if (atomic_add_return(i, &prog->aux->refcnt) > BPF_MAX_REFCNT) {
1076                 atomic_sub(i, &prog->aux->refcnt);
1077                 return ERR_PTR(-EBUSY);
1078         }
1079         return prog;
1080 }
1081 EXPORT_SYMBOL_GPL(bpf_prog_add);
1082
1083 void bpf_prog_sub(struct bpf_prog *prog, int i)
1084 {
1085         /* Only to be used for undoing previous bpf_prog_add() in some
1086          * error path. We still know that another entity in our call
1087          * path holds a reference to the program, thus atomic_sub() can
1088          * be safely used in such cases!
1089          */
1090         WARN_ON(atomic_sub_return(i, &prog->aux->refcnt) == 0);
1091 }
1092 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1093
1094 struct bpf_prog *bpf_prog_inc(struct bpf_prog *prog)
1095 {
1096         return bpf_prog_add(prog, 1);
1097 }
1098 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1099
1100 /* prog_idr_lock should have been held */
1101 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1102 {
1103         int refold;
1104
1105         refold = __atomic_add_unless(&prog->aux->refcnt, 1, 0);
1106
1107         if (refold >= BPF_MAX_REFCNT) {
1108                 __bpf_prog_put(prog, false);
1109                 return ERR_PTR(-EBUSY);
1110         }
1111
1112         if (!refold)
1113                 return ERR_PTR(-ENOENT);
1114
1115         return prog;
1116 }
1117 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1118
1119 bool bpf_prog_get_ok(struct bpf_prog *prog,
1120                             enum bpf_prog_type *attach_type, bool attach_drv)
1121 {
1122         /* not an attachment, just a refcount inc, always allow */
1123         if (!attach_type)
1124                 return true;
1125
1126         if (prog->type != *attach_type)
1127                 return false;
1128         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1129                 return false;
1130
1131         return true;
1132 }
1133
1134 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1135                                        bool attach_drv)
1136 {
1137         struct fd f = fdget(ufd);
1138         struct bpf_prog *prog;
1139
1140         prog = ____bpf_prog_get(f);
1141         if (IS_ERR(prog))
1142                 return prog;
1143         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1144                 prog = ERR_PTR(-EINVAL);
1145                 goto out;
1146         }
1147
1148         prog = bpf_prog_inc(prog);
1149 out:
1150         fdput(f);
1151         return prog;
1152 }
1153
1154 struct bpf_prog *bpf_prog_get(u32 ufd)
1155 {
1156         return __bpf_prog_get(ufd, NULL, false);
1157 }
1158
1159 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1160                                        bool attach_drv)
1161 {
1162         struct bpf_prog *prog = __bpf_prog_get(ufd, &type, attach_drv);
1163
1164         if (!IS_ERR(prog))
1165                 trace_bpf_prog_get_type(prog);
1166         return prog;
1167 }
1168 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1169
1170 /* last field in 'union bpf_attr' used by this command */
1171 #define BPF_PROG_LOAD_LAST_FIELD prog_ifindex
1172
1173 static int bpf_prog_load(union bpf_attr *attr)
1174 {
1175         enum bpf_prog_type type = attr->prog_type;
1176         struct bpf_prog *prog;
1177         int err;
1178         char license[128];
1179         bool is_gpl;
1180
1181         if (CHECK_ATTR(BPF_PROG_LOAD))
1182                 return -EINVAL;
1183
1184         if (attr->prog_flags & ~BPF_F_STRICT_ALIGNMENT)
1185                 return -EINVAL;
1186
1187         /* copy eBPF program license from user space */
1188         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1189                               sizeof(license) - 1) < 0)
1190                 return -EFAULT;
1191         license[sizeof(license) - 1] = 0;
1192
1193         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1194         is_gpl = license_is_gpl_compatible(license);
1195
1196         if (attr->insn_cnt == 0 || attr->insn_cnt > BPF_MAXINSNS)
1197                 return -E2BIG;
1198
1199         if (type == BPF_PROG_TYPE_KPROBE &&
1200             attr->kern_version != LINUX_VERSION_CODE)
1201                 return -EINVAL;
1202
1203         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1204             type != BPF_PROG_TYPE_CGROUP_SKB &&
1205             !capable(CAP_SYS_ADMIN))
1206                 return -EPERM;
1207
1208         /* plain bpf_prog allocation */
1209         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1210         if (!prog)
1211                 return -ENOMEM;
1212
1213         prog->aux->offload_requested = !!attr->prog_ifindex;
1214
1215         err = security_bpf_prog_alloc(prog->aux);
1216         if (err)
1217                 goto free_prog_nouncharge;
1218
1219         err = bpf_prog_charge_memlock(prog);
1220         if (err)
1221                 goto free_prog_sec;
1222
1223         prog->len = attr->insn_cnt;
1224
1225         err = -EFAULT;
1226         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1227                            bpf_prog_insn_size(prog)) != 0)
1228                 goto free_prog;
1229
1230         prog->orig_prog = NULL;
1231         prog->jited = 0;
1232
1233         atomic_set(&prog->aux->refcnt, 1);
1234         prog->gpl_compatible = is_gpl ? 1 : 0;
1235
1236         if (bpf_prog_is_dev_bound(prog->aux)) {
1237                 err = bpf_prog_offload_init(prog, attr);
1238                 if (err)
1239                         goto free_prog;
1240         }
1241
1242         /* find program type: socket_filter vs tracing_filter */
1243         err = find_prog_type(type, prog);
1244         if (err < 0)
1245                 goto free_prog;
1246
1247         prog->aux->load_time = ktime_get_boot_ns();
1248         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1249         if (err)
1250                 goto free_prog;
1251
1252         /* run eBPF verifier */
1253         err = bpf_check(&prog, attr);
1254         if (err < 0)
1255                 goto free_used_maps;
1256
1257         /* eBPF program is ready to be JITed */
1258         if (!prog->bpf_func)
1259                 prog = bpf_prog_select_runtime(prog, &err);
1260         if (err < 0)
1261                 goto free_used_maps;
1262
1263         err = bpf_prog_alloc_id(prog);
1264         if (err)
1265                 goto free_used_maps;
1266
1267         err = bpf_prog_new_fd(prog);
1268         if (err < 0) {
1269                 /* failed to allocate fd.
1270                  * bpf_prog_put() is needed because the above
1271                  * bpf_prog_alloc_id() has published the prog
1272                  * to the userspace and the userspace may
1273                  * have refcnt-ed it through BPF_PROG_GET_FD_BY_ID.
1274                  */
1275                 bpf_prog_put(prog);
1276                 return err;
1277         }
1278
1279         bpf_prog_kallsyms_add(prog);
1280         trace_bpf_prog_load(prog, err);
1281         return err;
1282
1283 free_used_maps:
1284         free_used_maps(prog->aux);
1285 free_prog:
1286         bpf_prog_uncharge_memlock(prog);
1287 free_prog_sec:
1288         security_bpf_prog_free(prog->aux);
1289 free_prog_nouncharge:
1290         bpf_prog_free(prog);
1291         return err;
1292 }
1293
1294 #define BPF_OBJ_LAST_FIELD file_flags
1295
1296 static int bpf_obj_pin(const union bpf_attr *attr)
1297 {
1298         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1299                 return -EINVAL;
1300
1301         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1302 }
1303
1304 static int bpf_obj_get(const union bpf_attr *attr)
1305 {
1306         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1307             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1308                 return -EINVAL;
1309
1310         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1311                                 attr->file_flags);
1312 }
1313
1314 #ifdef CONFIG_CGROUP_BPF
1315
1316 #define BPF_PROG_ATTACH_LAST_FIELD attach_flags
1317
1318 static int sockmap_get_from_fd(const union bpf_attr *attr, bool attach)
1319 {
1320         struct bpf_prog *prog = NULL;
1321         int ufd = attr->target_fd;
1322         struct bpf_map *map;
1323         struct fd f;
1324         int err;
1325
1326         f = fdget(ufd);
1327         map = __bpf_map_get(f);
1328         if (IS_ERR(map))
1329                 return PTR_ERR(map);
1330
1331         if (attach) {
1332                 prog = bpf_prog_get_type(attr->attach_bpf_fd,
1333                                          BPF_PROG_TYPE_SK_SKB);
1334                 if (IS_ERR(prog)) {
1335                         fdput(f);
1336                         return PTR_ERR(prog);
1337                 }
1338         }
1339
1340         err = sock_map_prog(map, prog, attr->attach_type);
1341         if (err) {
1342                 fdput(f);
1343                 if (prog)
1344                         bpf_prog_put(prog);
1345                 return err;
1346         }
1347
1348         fdput(f);
1349         return 0;
1350 }
1351
1352 #define BPF_F_ATTACH_MASK \
1353         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI)
1354
1355 static int bpf_prog_attach(const union bpf_attr *attr)
1356 {
1357         enum bpf_prog_type ptype;
1358         struct bpf_prog *prog;
1359         struct cgroup *cgrp;
1360         int ret;
1361
1362         if (!capable(CAP_NET_ADMIN))
1363                 return -EPERM;
1364
1365         if (CHECK_ATTR(BPF_PROG_ATTACH))
1366                 return -EINVAL;
1367
1368         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
1369                 return -EINVAL;
1370
1371         switch (attr->attach_type) {
1372         case BPF_CGROUP_INET_INGRESS:
1373         case BPF_CGROUP_INET_EGRESS:
1374                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1375                 break;
1376         case BPF_CGROUP_INET_SOCK_CREATE:
1377                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1378                 break;
1379         case BPF_CGROUP_SOCK_OPS:
1380                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1381                 break;
1382         case BPF_CGROUP_DEVICE:
1383                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1384                 break;
1385         case BPF_SK_SKB_STREAM_PARSER:
1386         case BPF_SK_SKB_STREAM_VERDICT:
1387                 return sockmap_get_from_fd(attr, true);
1388         default:
1389                 return -EINVAL;
1390         }
1391
1392         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1393         if (IS_ERR(prog))
1394                 return PTR_ERR(prog);
1395
1396         cgrp = cgroup_get_from_fd(attr->target_fd);
1397         if (IS_ERR(cgrp)) {
1398                 bpf_prog_put(prog);
1399                 return PTR_ERR(cgrp);
1400         }
1401
1402         ret = cgroup_bpf_attach(cgrp, prog, attr->attach_type,
1403                                 attr->attach_flags);
1404         if (ret)
1405                 bpf_prog_put(prog);
1406         cgroup_put(cgrp);
1407
1408         return ret;
1409 }
1410
1411 #define BPF_PROG_DETACH_LAST_FIELD attach_type
1412
1413 static int bpf_prog_detach(const union bpf_attr *attr)
1414 {
1415         enum bpf_prog_type ptype;
1416         struct bpf_prog *prog;
1417         struct cgroup *cgrp;
1418         int ret;
1419
1420         if (!capable(CAP_NET_ADMIN))
1421                 return -EPERM;
1422
1423         if (CHECK_ATTR(BPF_PROG_DETACH))
1424                 return -EINVAL;
1425
1426         switch (attr->attach_type) {
1427         case BPF_CGROUP_INET_INGRESS:
1428         case BPF_CGROUP_INET_EGRESS:
1429                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
1430                 break;
1431         case BPF_CGROUP_INET_SOCK_CREATE:
1432                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
1433                 break;
1434         case BPF_CGROUP_SOCK_OPS:
1435                 ptype = BPF_PROG_TYPE_SOCK_OPS;
1436                 break;
1437         case BPF_CGROUP_DEVICE:
1438                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
1439                 break;
1440         case BPF_SK_SKB_STREAM_PARSER:
1441         case BPF_SK_SKB_STREAM_VERDICT:
1442                 return sockmap_get_from_fd(attr, false);
1443         default:
1444                 return -EINVAL;
1445         }
1446
1447         cgrp = cgroup_get_from_fd(attr->target_fd);
1448         if (IS_ERR(cgrp))
1449                 return PTR_ERR(cgrp);
1450
1451         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
1452         if (IS_ERR(prog))
1453                 prog = NULL;
1454
1455         ret = cgroup_bpf_detach(cgrp, prog, attr->attach_type, 0);
1456         if (prog)
1457                 bpf_prog_put(prog);
1458         cgroup_put(cgrp);
1459         return ret;
1460 }
1461
1462 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
1463
1464 static int bpf_prog_query(const union bpf_attr *attr,
1465                           union bpf_attr __user *uattr)
1466 {
1467         struct cgroup *cgrp;
1468         int ret;
1469
1470         if (!capable(CAP_NET_ADMIN))
1471                 return -EPERM;
1472         if (CHECK_ATTR(BPF_PROG_QUERY))
1473                 return -EINVAL;
1474         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
1475                 return -EINVAL;
1476
1477         switch (attr->query.attach_type) {
1478         case BPF_CGROUP_INET_INGRESS:
1479         case BPF_CGROUP_INET_EGRESS:
1480         case BPF_CGROUP_INET_SOCK_CREATE:
1481         case BPF_CGROUP_SOCK_OPS:
1482         case BPF_CGROUP_DEVICE:
1483                 break;
1484         default:
1485                 return -EINVAL;
1486         }
1487         cgrp = cgroup_get_from_fd(attr->query.target_fd);
1488         if (IS_ERR(cgrp))
1489                 return PTR_ERR(cgrp);
1490         ret = cgroup_bpf_query(cgrp, attr, uattr);
1491         cgroup_put(cgrp);
1492         return ret;
1493 }
1494 #endif /* CONFIG_CGROUP_BPF */
1495
1496 #define BPF_PROG_TEST_RUN_LAST_FIELD test.duration
1497
1498 static int bpf_prog_test_run(const union bpf_attr *attr,
1499                              union bpf_attr __user *uattr)
1500 {
1501         struct bpf_prog *prog;
1502         int ret = -ENOTSUPP;
1503
1504         if (!capable(CAP_SYS_ADMIN))
1505                 return -EPERM;
1506         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
1507                 return -EINVAL;
1508
1509         prog = bpf_prog_get(attr->test.prog_fd);
1510         if (IS_ERR(prog))
1511                 return PTR_ERR(prog);
1512
1513         if (prog->aux->ops->test_run)
1514                 ret = prog->aux->ops->test_run(prog, attr, uattr);
1515
1516         bpf_prog_put(prog);
1517         return ret;
1518 }
1519
1520 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
1521
1522 static int bpf_obj_get_next_id(const union bpf_attr *attr,
1523                                union bpf_attr __user *uattr,
1524                                struct idr *idr,
1525                                spinlock_t *lock)
1526 {
1527         u32 next_id = attr->start_id;
1528         int err = 0;
1529
1530         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
1531                 return -EINVAL;
1532
1533         if (!capable(CAP_SYS_ADMIN))
1534                 return -EPERM;
1535
1536         next_id++;
1537         spin_lock_bh(lock);
1538         if (!idr_get_next(idr, &next_id))
1539                 err = -ENOENT;
1540         spin_unlock_bh(lock);
1541
1542         if (!err)
1543                 err = put_user(next_id, &uattr->next_id);
1544
1545         return err;
1546 }
1547
1548 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
1549
1550 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
1551 {
1552         struct bpf_prog *prog;
1553         u32 id = attr->prog_id;
1554         int fd;
1555
1556         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
1557                 return -EINVAL;
1558
1559         if (!capable(CAP_SYS_ADMIN))
1560                 return -EPERM;
1561
1562         spin_lock_bh(&prog_idr_lock);
1563         prog = idr_find(&prog_idr, id);
1564         if (prog)
1565                 prog = bpf_prog_inc_not_zero(prog);
1566         else
1567                 prog = ERR_PTR(-ENOENT);
1568         spin_unlock_bh(&prog_idr_lock);
1569
1570         if (IS_ERR(prog))
1571                 return PTR_ERR(prog);
1572
1573         fd = bpf_prog_new_fd(prog);
1574         if (fd < 0)
1575                 bpf_prog_put(prog);
1576
1577         return fd;
1578 }
1579
1580 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
1581
1582 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
1583 {
1584         struct bpf_map *map;
1585         u32 id = attr->map_id;
1586         int f_flags;
1587         int fd;
1588
1589         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
1590             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
1591                 return -EINVAL;
1592
1593         if (!capable(CAP_SYS_ADMIN))
1594                 return -EPERM;
1595
1596         f_flags = bpf_get_file_flag(attr->open_flags);
1597         if (f_flags < 0)
1598                 return f_flags;
1599
1600         spin_lock_bh(&map_idr_lock);
1601         map = idr_find(&map_idr, id);
1602         if (map)
1603                 map = bpf_map_inc_not_zero(map, true);
1604         else
1605                 map = ERR_PTR(-ENOENT);
1606         spin_unlock_bh(&map_idr_lock);
1607
1608         if (IS_ERR(map))
1609                 return PTR_ERR(map);
1610
1611         fd = bpf_map_new_fd(map, f_flags);
1612         if (fd < 0)
1613                 bpf_map_put(map);
1614
1615         return fd;
1616 }
1617
1618 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
1619                                               unsigned long addr)
1620 {
1621         int i;
1622
1623         for (i = 0; i < prog->aux->used_map_cnt; i++)
1624                 if (prog->aux->used_maps[i] == (void *)addr)
1625                         return prog->aux->used_maps[i];
1626         return NULL;
1627 }
1628
1629 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
1630 {
1631         const struct bpf_map *map;
1632         struct bpf_insn *insns;
1633         u64 imm;
1634         int i;
1635
1636         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
1637                         GFP_USER);
1638         if (!insns)
1639                 return insns;
1640
1641         for (i = 0; i < prog->len; i++) {
1642                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
1643                         insns[i].code = BPF_JMP | BPF_CALL;
1644                         insns[i].imm = BPF_FUNC_tail_call;
1645                         /* fall-through */
1646                 }
1647                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
1648                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
1649                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
1650                                 insns[i].code = BPF_JMP | BPF_CALL;
1651                         if (!bpf_dump_raw_ok())
1652                                 insns[i].imm = 0;
1653                         continue;
1654                 }
1655
1656                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
1657                         continue;
1658
1659                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
1660                 map = bpf_map_from_imm(prog, imm);
1661                 if (map) {
1662                         insns[i].src_reg = BPF_PSEUDO_MAP_FD;
1663                         insns[i].imm = map->id;
1664                         insns[i + 1].imm = 0;
1665                         continue;
1666                 }
1667
1668                 if (!bpf_dump_raw_ok() &&
1669                     imm == (unsigned long)prog->aux) {
1670                         insns[i].imm = 0;
1671                         insns[i + 1].imm = 0;
1672                         continue;
1673                 }
1674         }
1675
1676         return insns;
1677 }
1678
1679 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
1680                                    const union bpf_attr *attr,
1681                                    union bpf_attr __user *uattr)
1682 {
1683         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1684         struct bpf_prog_info info = {};
1685         u32 info_len = attr->info.info_len;
1686         char __user *uinsns;
1687         u32 ulen;
1688         int err;
1689
1690         err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1691         if (err)
1692                 return err;
1693         info_len = min_t(u32, sizeof(info), info_len);
1694
1695         if (copy_from_user(&info, uinfo, info_len))
1696                 return -EFAULT;
1697
1698         info.type = prog->type;
1699         info.id = prog->aux->id;
1700         info.load_time = prog->aux->load_time;
1701         info.created_by_uid = from_kuid_munged(current_user_ns(),
1702                                                prog->aux->user->uid);
1703
1704         memcpy(info.tag, prog->tag, sizeof(prog->tag));
1705         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
1706
1707         ulen = info.nr_map_ids;
1708         info.nr_map_ids = prog->aux->used_map_cnt;
1709         ulen = min_t(u32, info.nr_map_ids, ulen);
1710         if (ulen) {
1711                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
1712                 u32 i;
1713
1714                 for (i = 0; i < ulen; i++)
1715                         if (put_user(prog->aux->used_maps[i]->id,
1716                                      &user_map_ids[i]))
1717                                 return -EFAULT;
1718         }
1719
1720         if (!capable(CAP_SYS_ADMIN)) {
1721                 info.jited_prog_len = 0;
1722                 info.xlated_prog_len = 0;
1723                 goto done;
1724         }
1725
1726         ulen = info.xlated_prog_len;
1727         info.xlated_prog_len = bpf_prog_insn_size(prog);
1728         if (info.xlated_prog_len && ulen) {
1729                 struct bpf_insn *insns_sanitized;
1730                 bool fault;
1731
1732                 if (prog->blinded && !bpf_dump_raw_ok()) {
1733                         info.xlated_prog_insns = 0;
1734                         goto done;
1735                 }
1736                 insns_sanitized = bpf_insn_prepare_dump(prog);
1737                 if (!insns_sanitized)
1738                         return -ENOMEM;
1739                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
1740                 ulen = min_t(u32, info.xlated_prog_len, ulen);
1741                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
1742                 kfree(insns_sanitized);
1743                 if (fault)
1744                         return -EFAULT;
1745         }
1746
1747         if (bpf_prog_is_dev_bound(prog->aux)) {
1748                 err = bpf_prog_offload_info_fill(&info, prog);
1749                 if (err)
1750                         return err;
1751                 goto done;
1752         }
1753
1754         /* NOTE: the following code is supposed to be skipped for offload.
1755          * bpf_prog_offload_info_fill() is the place to fill similar fields
1756          * for offload.
1757          */
1758         ulen = info.jited_prog_len;
1759         info.jited_prog_len = prog->jited_len;
1760         if (info.jited_prog_len && ulen) {
1761                 if (bpf_dump_raw_ok()) {
1762                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
1763                         ulen = min_t(u32, info.jited_prog_len, ulen);
1764                         if (copy_to_user(uinsns, prog->bpf_func, ulen))
1765                                 return -EFAULT;
1766                 } else {
1767                         info.jited_prog_insns = 0;
1768                 }
1769         }
1770
1771 done:
1772         if (copy_to_user(uinfo, &info, info_len) ||
1773             put_user(info_len, &uattr->info.info_len))
1774                 return -EFAULT;
1775
1776         return 0;
1777 }
1778
1779 static int bpf_map_get_info_by_fd(struct bpf_map *map,
1780                                   const union bpf_attr *attr,
1781                                   union bpf_attr __user *uattr)
1782 {
1783         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
1784         struct bpf_map_info info = {};
1785         u32 info_len = attr->info.info_len;
1786         int err;
1787
1788         err = check_uarg_tail_zero(uinfo, sizeof(info), info_len);
1789         if (err)
1790                 return err;
1791         info_len = min_t(u32, sizeof(info), info_len);
1792
1793         info.type = map->map_type;
1794         info.id = map->id;
1795         info.key_size = map->key_size;
1796         info.value_size = map->value_size;
1797         info.max_entries = map->max_entries;
1798         info.map_flags = map->map_flags;
1799         memcpy(info.name, map->name, sizeof(map->name));
1800
1801         if (bpf_map_is_dev_bound(map)) {
1802                 err = bpf_map_offload_info_fill(&info, map);
1803                 if (err)
1804                         return err;
1805         }
1806
1807         if (copy_to_user(uinfo, &info, info_len) ||
1808             put_user(info_len, &uattr->info.info_len))
1809                 return -EFAULT;
1810
1811         return 0;
1812 }
1813
1814 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
1815
1816 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
1817                                   union bpf_attr __user *uattr)
1818 {
1819         int ufd = attr->info.bpf_fd;
1820         struct fd f;
1821         int err;
1822
1823         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
1824                 return -EINVAL;
1825
1826         f = fdget(ufd);
1827         if (!f.file)
1828                 return -EBADFD;
1829
1830         if (f.file->f_op == &bpf_prog_fops)
1831                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
1832                                               uattr);
1833         else if (f.file->f_op == &bpf_map_fops)
1834                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
1835                                              uattr);
1836         else
1837                 err = -EINVAL;
1838
1839         fdput(f);
1840         return err;
1841 }
1842
1843 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
1844 {
1845         union bpf_attr attr = {};
1846         int err;
1847
1848         if (!capable(CAP_SYS_ADMIN) && sysctl_unprivileged_bpf_disabled)
1849                 return -EPERM;
1850
1851         err = check_uarg_tail_zero(uattr, sizeof(attr), size);
1852         if (err)
1853                 return err;
1854         size = min_t(u32, size, sizeof(attr));
1855
1856         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
1857         if (copy_from_user(&attr, uattr, size) != 0)
1858                 return -EFAULT;
1859
1860         err = security_bpf(cmd, &attr, size);
1861         if (err < 0)
1862                 return err;
1863
1864         switch (cmd) {
1865         case BPF_MAP_CREATE:
1866                 err = map_create(&attr);
1867                 break;
1868         case BPF_MAP_LOOKUP_ELEM:
1869                 err = map_lookup_elem(&attr);
1870                 break;
1871         case BPF_MAP_UPDATE_ELEM:
1872                 err = map_update_elem(&attr);
1873                 break;
1874         case BPF_MAP_DELETE_ELEM:
1875                 err = map_delete_elem(&attr);
1876                 break;
1877         case BPF_MAP_GET_NEXT_KEY:
1878                 err = map_get_next_key(&attr);
1879                 break;
1880         case BPF_PROG_LOAD:
1881                 err = bpf_prog_load(&attr);
1882                 break;
1883         case BPF_OBJ_PIN:
1884                 err = bpf_obj_pin(&attr);
1885                 break;
1886         case BPF_OBJ_GET:
1887                 err = bpf_obj_get(&attr);
1888                 break;
1889 #ifdef CONFIG_CGROUP_BPF
1890         case BPF_PROG_ATTACH:
1891                 err = bpf_prog_attach(&attr);
1892                 break;
1893         case BPF_PROG_DETACH:
1894                 err = bpf_prog_detach(&attr);
1895                 break;
1896         case BPF_PROG_QUERY:
1897                 err = bpf_prog_query(&attr, uattr);
1898                 break;
1899 #endif
1900         case BPF_PROG_TEST_RUN:
1901                 err = bpf_prog_test_run(&attr, uattr);
1902                 break;
1903         case BPF_PROG_GET_NEXT_ID:
1904                 err = bpf_obj_get_next_id(&attr, uattr,
1905                                           &prog_idr, &prog_idr_lock);
1906                 break;
1907         case BPF_MAP_GET_NEXT_ID:
1908                 err = bpf_obj_get_next_id(&attr, uattr,
1909                                           &map_idr, &map_idr_lock);
1910                 break;
1911         case BPF_PROG_GET_FD_BY_ID:
1912                 err = bpf_prog_get_fd_by_id(&attr);
1913                 break;
1914         case BPF_MAP_GET_FD_BY_ID:
1915                 err = bpf_map_get_fd_by_id(&attr);
1916                 break;
1917         case BPF_OBJ_GET_INFO_BY_FD:
1918                 err = bpf_obj_get_info_by_fd(&attr, uattr);
1919                 break;
1920         default:
1921                 err = -EINVAL;
1922                 break;
1923         }
1924
1925         return err;
1926 }