kernel/bpf/syscall.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   3  */
   4 #include <linux/bpf.h>
   5 #include <linux/bpf_trace.h>
   6 #include <linux/bpf_lirc.h>
   7 #include <linux/btf.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/slab.h>
  10 #include <linux/sched/signal.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/mmzone.h>
  13 #include <linux/anon_inodes.h>
  14 #include <linux/fdtable.h>
  15 #include <linux/file.h>
  16 #include <linux/fs.h>
  17 #include <linux/license.h>
  18 #include <linux/filter.h>
  19 #include <linux/version.h>
  20 #include <linux/kernel.h>
  21 #include <linux/idr.h>
  22 #include <linux/cred.h>
  23 #include <linux/timekeeping.h>
  24 #include <linux/ctype.h>
  25 #include <linux/nospec.h>
  26 #include <linux/audit.h>
  27 #include <uapi/linux/btf.h>
  28
  29 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  30                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  31                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  32 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  33 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  34 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  35                         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, _name, prog_ctx_type, kern_ctx_type)
  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 int bpf_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(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         .map_check_btf = map_check_no_btf,
 101 };
 102
 103 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 104 {
 105         const struct bpf_map_ops *ops;
 106         u32 type = attr->map_type;
 107         struct bpf_map *map;
 108         int err;
 109
 110         if (type >= ARRAY_SIZE(bpf_map_types))
 111                 return ERR_PTR(-EINVAL);
 112         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 113         ops = bpf_map_types[type];
 114         if (!ops)
 115                 return ERR_PTR(-EINVAL);
 116
 117         if (ops->map_alloc_check) {
 118                 err = ops->map_alloc_check(attr);
 119                 if (err)
 120                         return ERR_PTR(err);
 121         }
 122         if (attr->map_ifindex)
 123                 ops = &bpf_map_offload_ops;
 124         map = ops->map_alloc(attr);
 125         if (IS_ERR(map))
 126                 return map;
 127         map->ops = ops;
 128         map->map_type = type;
 129         return map;
 130 }
 131
 132 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 133 {
 134         /* We really just want to fail instead of triggering OOM killer
 135          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 136          * which is used for lower order allocation requests.
 137          *
 138          * It has been observed that higher order allocation requests done by
 139          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 140          * to reclaim memory from the page cache, thus we set
 141          * __GFP_RETRY_MAYFAIL to avoid such situations.
 142          */
 143
 144         const gfp_t flags = __GFP_NOWARN | __GFP_ZERO;
 145         void *area;
 146
 147         if (size >= SIZE_MAX)
 148                 return NULL;
 149
 150         /* kmalloc()'ed memory can't be mmap()'ed */
 151         if (!mmapable && size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 152                 area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags,
 153                                     numa_node);
 154                 if (area != NULL)
 155                         return area;
 156         }
 157         if (mmapable) {
 158                 BUG_ON(!PAGE_ALIGNED(size));
 159                 return vmalloc_user_node_flags(size, numa_node, GFP_KERNEL |
 160                                                __GFP_RETRY_MAYFAIL | flags);
 161         }
 162         return __vmalloc_node_flags_caller(size, numa_node,
 163                                            GFP_KERNEL | __GFP_RETRY_MAYFAIL |
 164                                            flags, __builtin_return_address(0));
 165 }
 166
 167 void *bpf_map_area_alloc(u64 size, int numa_node)
 168 {
 169         return __bpf_map_area_alloc(size, numa_node, false);
 170 }
 171
 172 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 173 {
 174         return __bpf_map_area_alloc(size, numa_node, true);
 175 }
 176
 177 void bpf_map_area_free(void *area)
 178 {
 179         kvfree(area);
 180 }
 181
 182 static u32 bpf_map_flags_retain_permanent(u32 flags)
 183 {
 184         /* Some map creation flags are not tied to the map object but
 185          * rather to the map fd instead, so they have no meaning upon
 186          * map object inspection since multiple file descriptors with
 187          * different (access) properties can exist here. Thus, given
 188          * this has zero meaning for the map itself, lets clear these
 189          * from here.
 190          */
 191         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 192 }
 193
 194 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 195 {
 196         map->map_type = attr->map_type;
 197         map->key_size = attr->key_size;
 198         map->value_size = attr->value_size;
 199         map->max_entries = attr->max_entries;
 200         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 201         map->numa_node = bpf_map_attr_numa_node(attr);
 202 }
 203
 204 static int bpf_charge_memlock(struct user_struct *user, u32 pages)
 205 {
 206         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 207
 208         if (atomic_long_add_return(pages, &user->locked_vm) > memlock_limit) {
 209                 atomic_long_sub(pages, &user->locked_vm);
 210                 return -EPERM;
 211         }
 212         return 0;
 213 }
 214
 215 static void bpf_uncharge_memlock(struct user_struct *user, u32 pages)
 216 {
 217         if (user)
 218                 atomic_long_sub(pages, &user->locked_vm);
 219 }
 220
 221 int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size)
 222 {
 223         u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT;
 224         struct user_struct *user;
 225         int ret;
 226
 227         if (size >= U32_MAX - PAGE_SIZE)
 228                 return -E2BIG;
 229
 230         user = get_current_user();
 231         ret = bpf_charge_memlock(user, pages);
 232         if (ret) {
 233                 free_uid(user);
 234                 return ret;
 235         }
 236
 237         mem->pages = pages;
 238         mem->user = user;
 239
 240         return 0;
 241 }
 242
 243 void bpf_map_charge_finish(struct bpf_map_memory *mem)
 244 {
 245         bpf_uncharge_memlock(mem->user, mem->pages);
 246         free_uid(mem->user);
 247 }
 248
 249 void bpf_map_charge_move(struct bpf_map_memory *dst,
 250                          struct bpf_map_memory *src)
 251 {
 252         *dst = *src;
 253
 254         /* Make sure src will not be used for the redundant uncharging. */
 255         memset(src, 0, sizeof(struct bpf_map_memory));
 256 }
 257
 258 int bpf_map_charge_memlock(struct bpf_map *map, u32 pages)
 259 {
 260         int ret;
 261
 262         ret = bpf_charge_memlock(map->memory.user, pages);
 263         if (ret)
 264                 return ret;
 265         map->memory.pages += pages;
 266         return ret;
 267 }
 268
 269 void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages)
 270 {
 271         bpf_uncharge_memlock(map->memory.user, pages);
 272         map->memory.pages -= pages;
 273 }
 274
 275 static int bpf_map_alloc_id(struct bpf_map *map)
 276 {
 277         int id;
 278
 279         idr_preload(GFP_KERNEL);
 280         spin_lock_bh(&map_idr_lock);
 281         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 282         if (id > 0)
 283                 map->id = id;
 284         spin_unlock_bh(&map_idr_lock);
 285         idr_preload_end();
 286
 287         if (WARN_ON_ONCE(!id))
 288                 return -ENOSPC;
 289
 290         return id > 0 ? 0 : id;
 291 }
 292
 293 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 294 {
 295         unsigned long flags;
 296
 297         /* Offloaded maps are removed from the IDR store when their device
 298          * disappears - even if someone holds an fd to them they are unusable,
 299          * the memory is gone, all ops will fail; they are simply waiting for
 300          * refcnt to drop to be freed.
 301          */
 302         if (!map->id)
 303                 return;
 304
 305         if (do_idr_lock)
 306                 spin_lock_irqsave(&map_idr_lock, flags);
 307         else
 308                 __acquire(&map_idr_lock);
 309
 310         idr_remove(&map_idr, map->id);
 311         map->id = 0;
 312
 313         if (do_idr_lock)
 314                 spin_unlock_irqrestore(&map_idr_lock, flags);
 315         else
 316                 __release(&map_idr_lock);
 317 }
 318
 319 /* called from workqueue */
 320 static void bpf_map_free_deferred(struct work_struct *work)
 321 {
 322         struct bpf_map *map = container_of(work, struct bpf_map, work);
 323         struct bpf_map_memory mem;
 324
 325         bpf_map_charge_move(&mem, &map->memory);
 326         security_bpf_map_free(map);
 327         /* implementation dependent freeing */
 328         map->ops->map_free(map);
 329         bpf_map_charge_finish(&mem);
 330 }
 331
 332 static void bpf_map_put_uref(struct bpf_map *map)
 333 {
 334         if (atomic64_dec_and_test(&map->usercnt)) {
 335                 if (map->ops->map_release_uref)
 336                         map->ops->map_release_uref(map);
 337         }
 338 }
 339
 340 /* decrement map refcnt and schedule it for freeing via workqueue
 341  * (unrelying map implementation ops->map_free() might sleep)
 342  */
 343 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 344 {
 345         if (atomic64_dec_and_test(&map->refcnt)) {
 346                 /* bpf_map_free_id() must be called first */
 347                 bpf_map_free_id(map, do_idr_lock);
 348                 btf_put(map->btf);
 349                 INIT_WORK(&map->work, bpf_map_free_deferred);
 350                 schedule_work(&map->work);
 351         }
 352 }
 353
 354 void bpf_map_put(struct bpf_map *map)
 355 {
 356         __bpf_map_put(map, true);
 357 }
 358 EXPORT_SYMBOL_GPL(bpf_map_put);
 359
 360 void bpf_map_put_with_uref(struct bpf_map *map)
 361 {
 362         bpf_map_put_uref(map);
 363         bpf_map_put(map);
 364 }
 365
 366 static int bpf_map_release(struct inode *inode, struct file *filp)
 367 {
 368         struct bpf_map *map = filp->private_data;
 369
 370         if (map->ops->map_release)
 371                 map->ops->map_release(map, filp);
 372
 373         bpf_map_put_with_uref(map);
 374         return 0;
 375 }
 376
 377 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 378 {
 379         fmode_t mode = f.file->f_mode;
 380
 381         /* Our file permissions may have been overridden by global
 382          * map permissions facing syscall side.
 383          */
 384         if (READ_ONCE(map->frozen))
 385                 mode &= ~FMODE_CAN_WRITE;
 386         return mode;
 387 }
 388
 389 #ifdef CONFIG_PROC_FS
 390 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 391 {
 392         const struct bpf_map *map = filp->private_data;
 393         const struct bpf_array *array;
 394         u32 type = 0, jited = 0;
 395
 396         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 397                 array = container_of(map, struct bpf_array, map);
 398                 type  = array->aux->type;
 399                 jited = array->aux->jited;
 400         }
 401
 402         seq_printf(m,
 403                    "map_type:\t%u\n"
 404                    "key_size:\t%u\n"
 405                    "value_size:\t%u\n"
 406                    "max_entries:\t%u\n"
 407                    "map_flags:\t%#x\n"
 408                    "memlock:\t%llu\n"
 409                    "map_id:\t%u\n"
 410                    "frozen:\t%u\n",
 411                    map->map_type,
 412                    map->key_size,
 413                    map->value_size,
 414                    map->max_entries,
 415                    map->map_flags,
 416                    map->memory.pages * 1ULL << PAGE_SHIFT,
 417                    map->id,
 418                    READ_ONCE(map->frozen));
 419         if (type) {
 420                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 421                 seq_printf(m, "owner_jited:\t%u\n", jited);
 422         }
 423 }
 424 #endif
 425
 426 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 427                               loff_t *ppos)
 428 {
 429         /* We need this handler such that alloc_file() enables
 430          * f_mode with FMODE_CAN_READ.
 431          */
 432         return -EINVAL;
 433 }
 434
 435 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 436                                size_t siz, loff_t *ppos)
 437 {
 438         /* We need this handler such that alloc_file() enables
 439          * f_mode with FMODE_CAN_WRITE.
 440          */
 441         return -EINVAL;
 442 }
 443
 444 /* called for any extra memory-mapped regions (except initial) */
 445 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 446 {
 447         struct bpf_map *map = vma->vm_file->private_data;
 448
 449         bpf_map_inc_with_uref(map);
 450
 451         if (vma->vm_flags & VM_WRITE) {
 452                 mutex_lock(&map->freeze_mutex);
 453                 map->writecnt++;
 454                 mutex_unlock(&map->freeze_mutex);
 455         }
 456 }
 457
 458 /* called for all unmapped memory region (including initial) */
 459 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 460 {
 461         struct bpf_map *map = vma->vm_file->private_data;
 462
 463         if (vma->vm_flags & VM_WRITE) {
 464                 mutex_lock(&map->freeze_mutex);
 465                 map->writecnt--;
 466                 mutex_unlock(&map->freeze_mutex);
 467         }
 468
 469         bpf_map_put_with_uref(map);
 470 }
 471
 472 static const struct vm_operations_struct bpf_map_default_vmops = {
 473         .open           = bpf_map_mmap_open,
 474         .close          = bpf_map_mmap_close,
 475 };
 476
 477 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 478 {
 479         struct bpf_map *map = filp->private_data;
 480         int err;
 481
 482         if (!map->ops->map_mmap || map_value_has_spin_lock(map))
 483                 return -ENOTSUPP;
 484
 485         if (!(vma->vm_flags & VM_SHARED))
 486                 return -EINVAL;
 487
 488         mutex_lock(&map->freeze_mutex);
 489
 490         if ((vma->vm_flags & VM_WRITE) && map->frozen) {
 491                 err = -EPERM;
 492                 goto out;
 493         }
 494
 495         /* set default open/close callbacks */
 496         vma->vm_ops = &bpf_map_default_vmops;
 497         vma->vm_private_data = map;
 498
 499         err = map->ops->map_mmap(map, vma);
 500         if (err)
 501                 goto out;
 502
 503         bpf_map_inc_with_uref(map);
 504
 505         if (vma->vm_flags & VM_WRITE)
 506                 map->writecnt++;
 507 out:
 508         mutex_unlock(&map->freeze_mutex);
 509         return err;
 510 }
 511
 512 const struct file_operations bpf_map_fops = {
 513 #ifdef CONFIG_PROC_FS
 514         .show_fdinfo    = bpf_map_show_fdinfo,
 515 #endif
 516         .release        = bpf_map_release,
 517         .read           = bpf_dummy_read,
 518         .write          = bpf_dummy_write,
 519         .mmap           = bpf_map_mmap,
 520 };
 521
 522 int bpf_map_new_fd(struct bpf_map *map, int flags)
 523 {
 524         int ret;
 525
 526         ret = security_bpf_map(map, OPEN_FMODE(flags));
 527         if (ret < 0)
 528                 return ret;
 529
 530         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 531                                 flags | O_CLOEXEC);
 532 }
 533
 534 int bpf_get_file_flag(int flags)
 535 {
 536         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 537                 return -EINVAL;
 538         if (flags & BPF_F_RDONLY)
 539                 return O_RDONLY;
 540         if (flags & BPF_F_WRONLY)
 541                 return O_WRONLY;
 542         return O_RDWR;
 543 }
 544
 545 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 546 #define CHECK_ATTR(CMD) \
 547         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 548                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 549                    sizeof(*attr) - \
 550                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 551                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 552
 553 /* dst and src must have at least BPF_OBJ_NAME_LEN number of bytes.
 554  * Return 0 on success and < 0 on error.
 555  */
 556 static int bpf_obj_name_cpy(char *dst, const char *src)
 557 {
 558         const char *end = src + BPF_OBJ_NAME_LEN;
 559
 560         memset(dst, 0, BPF_OBJ_NAME_LEN);
 561         /* Copy all isalnum(), '_' and '.' chars. */
 562         while (src < end && *src) {
 563                 if (!isalnum(*src) &&
 564                     *src != '_' && *src != '.')
 565                         return -EINVAL;
 566                 *dst++ = *src++;
 567         }
 568
 569         /* No '\0' found in BPF_OBJ_NAME_LEN number of bytes */
 570         if (src == end)
 571                 return -EINVAL;
 572
 573         return 0;
 574 }
 575
 576 int map_check_no_btf(const struct bpf_map *map,
 577                      const struct btf *btf,
 578                      const struct btf_type *key_type,
 579                      const struct btf_type *value_type)
 580 {
 581         return -ENOTSUPP;
 582 }
 583
 584 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 585                          u32 btf_key_id, u32 btf_value_id)
 586 {
 587         const struct btf_type *key_type, *value_type;
 588         u32 key_size, value_size;
 589         int ret = 0;
 590
 591         /* Some maps allow key to be unspecified. */
 592         if (btf_key_id) {
 593                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 594                 if (!key_type || key_size != map->key_size)
 595                         return -EINVAL;
 596         } else {
 597                 key_type = btf_type_by_id(btf, 0);
 598                 if (!map->ops->map_check_btf)
 599                         return -EINVAL;
 600         }
 601
 602         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 603         if (!value_type || value_size != map->value_size)
 604                 return -EINVAL;
 605
 606         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 607
 608         if (map_value_has_spin_lock(map)) {
 609                 if (map->map_flags & BPF_F_RDONLY_PROG)
 610                         return -EACCES;
 611                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 612                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 613                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 614                     map->map_type != BPF_MAP_TYPE_SK_STORAGE)
 615                         return -ENOTSUPP;
 616                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 617                     map->value_size) {
 618                         WARN_ONCE(1,
 619                                   "verifier bug spin_lock_off %d value_size %d\n",
 620                                   map->spin_lock_off, map->value_size);
 621                         return -EFAULT;
 622                 }
 623         }
 624
 625         if (map->ops->map_check_btf)
 626                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 627
 628         return ret;
 629 }
 630
 631 #define BPF_MAP_CREATE_LAST_FIELD btf_value_type_id
 632 /* called via syscall */
 633 static int map_create(union bpf_attr *attr)
 634 {
 635         int numa_node = bpf_map_attr_numa_node(attr);
 636         struct bpf_map_memory mem;
 637         struct bpf_map *map;
 638         int f_flags;
 639         int err;
 640
 641         err = CHECK_ATTR(BPF_MAP_CREATE);
 642         if (err)
 643                 return -EINVAL;
 644
 645         f_flags = bpf_get_file_flag(attr->map_flags);
 646         if (f_flags < 0)
 647                 return f_flags;
 648
 649         if (numa_node != NUMA_NO_NODE &&
 650             ((unsigned int)numa_node >= nr_node_ids ||
 651              !node_online(numa_node)))
 652                 return -EINVAL;
 653
 654         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 655         map = find_and_alloc_map(attr);
 656         if (IS_ERR(map))
 657                 return PTR_ERR(map);
 658
 659         err = bpf_obj_name_cpy(map->name, attr->map_name);
 660         if (err)
 661                 goto free_map;
 662
 663         atomic64_set(&map->refcnt, 1);
 664         atomic64_set(&map->usercnt, 1);
 665         mutex_init(&map->freeze_mutex);
 666
 667         if (attr->btf_key_type_id || attr->btf_value_type_id) {
 668                 struct btf *btf;
 669
 670                 if (!attr->btf_value_type_id) {
 671                         err = -EINVAL;
 672                         goto free_map;
 673                 }
 674
 675                 btf = btf_get_by_fd(attr->btf_fd);
 676                 if (IS_ERR(btf)) {
 677                         err = PTR_ERR(btf);
 678                         goto free_map;
 679                 }
 680
 681                 err = map_check_btf(map, btf, attr->btf_key_type_id,
 682                                     attr->btf_value_type_id);
 683                 if (err) {
 684                         btf_put(btf);
 685                         goto free_map;
 686                 }
 687
 688                 map->btf = btf;
 689                 map->btf_key_type_id = attr->btf_key_type_id;
 690                 map->btf_value_type_id = attr->btf_value_type_id;
 691         } else {
 692                 map->spin_lock_off = -EINVAL;
 693         }
 694
 695         err = security_bpf_map_alloc(map);
 696         if (err)
 697                 goto free_map;
 698
 699         err = bpf_map_alloc_id(map);
 700         if (err)
 701                 goto free_map_sec;
 702
 703         err = bpf_map_new_fd(map, f_flags);
 704         if (err < 0) {
 705                 /* failed to allocate fd.
 706                  * bpf_map_put_with_uref() is needed because the above
 707                  * bpf_map_alloc_id() has published the map
 708                  * to the userspace and the userspace may
 709                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 710                  */
 711                 bpf_map_put_with_uref(map);
 712                 return err;
 713         }
 714
 715         return err;
 716
 717 free_map_sec:
 718         security_bpf_map_free(map);
 719 free_map:
 720         btf_put(map->btf);
 721         bpf_map_charge_move(&mem, &map->memory);
 722         map->ops->map_free(map);
 723         bpf_map_charge_finish(&mem);
 724         return err;
 725 }
 726
 727 /* if error is returned, fd is released.
 728  * On success caller should complete fd access with matching fdput()
 729  */
 730 struct bpf_map *__bpf_map_get(struct fd f)
 731 {
 732         if (!f.file)
 733                 return ERR_PTR(-EBADF);
 734         if (f.file->f_op != &bpf_map_fops) {
 735                 fdput(f);
 736                 return ERR_PTR(-EINVAL);
 737         }
 738
 739         return f.file->private_data;
 740 }
 741
 742 void bpf_map_inc(struct bpf_map *map)
 743 {
 744         atomic64_inc(&map->refcnt);
 745 }
 746 EXPORT_SYMBOL_GPL(bpf_map_inc);
 747
 748 void bpf_map_inc_with_uref(struct bpf_map *map)
 749 {
 750         atomic64_inc(&map->refcnt);
 751         atomic64_inc(&map->usercnt);
 752 }
 753 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
 754
 755 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 756 {
 757         struct fd f = fdget(ufd);
 758         struct bpf_map *map;
 759
 760         map = __bpf_map_get(f);
 761         if (IS_ERR(map))
 762                 return map;
 763
 764         bpf_map_inc_with_uref(map);
 765         fdput(f);
 766
 767         return map;
 768 }
 769
 770 /* map_idr_lock should have been held */
 771 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 772 {
 773         int refold;
 774
 775         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
 776         if (!refold)
 777                 return ERR_PTR(-ENOENT);
 778         if (uref)
 779                 atomic64_inc(&map->usercnt);
 780
 781         return map;
 782 }
 783
 784 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
 785 {
 786         spin_lock_bh(&map_idr_lock);
 787         map = __bpf_map_inc_not_zero(map, false);
 788         spin_unlock_bh(&map_idr_lock);
 789
 790         return map;
 791 }
 792 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 793
 794 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 795 {
 796         return -ENOTSUPP;
 797 }
 798
 799 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 800 {
 801         if (key_size)
 802                 return memdup_user(ukey, key_size);
 803
 804         if (ukey)
 805                 return ERR_PTR(-EINVAL);
 806
 807         return NULL;
 808 }
 809
 810 /* last field in 'union bpf_attr' used by this command */
 811 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
 812
 813 static int map_lookup_elem(union bpf_attr *attr)
 814 {
 815         void __user *ukey = u64_to_user_ptr(attr->key);
 816         void __user *uvalue = u64_to_user_ptr(attr->value);
 817         int ufd = attr->map_fd;
 818         struct bpf_map *map;
 819         void *key, *value, *ptr;
 820         u32 value_size;
 821         struct fd f;
 822         int err;
 823
 824         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
 825                 return -EINVAL;
 826
 827         if (attr->flags & ~BPF_F_LOCK)
 828                 return -EINVAL;
 829
 830         f = fdget(ufd);
 831         map = __bpf_map_get(f);
 832         if (IS_ERR(map))
 833                 return PTR_ERR(map);
 834         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
 835                 err = -EPERM;
 836                 goto err_put;
 837         }
 838
 839         if ((attr->flags & BPF_F_LOCK) &&
 840             !map_value_has_spin_lock(map)) {
 841                 err = -EINVAL;
 842                 goto err_put;
 843         }
 844
 845         key = __bpf_copy_key(ukey, map->key_size);
 846         if (IS_ERR(key)) {
 847                 err = PTR_ERR(key);
 848                 goto err_put;
 849         }
 850
 851         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 852             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 853             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 854             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 855                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 856         else if (IS_FD_MAP(map))
 857                 value_size = sizeof(u32);
 858         else
 859                 value_size = map->value_size;
 860
 861         err = -ENOMEM;
 862         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 863         if (!value)
 864                 goto free_key;
 865
 866         if (bpf_map_is_dev_bound(map)) {
 867                 err = bpf_map_offload_lookup_elem(map, key, value);
 868                 goto done;
 869         }
 870
 871         preempt_disable();
 872         this_cpu_inc(bpf_prog_active);
 873         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 874             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 875                 err = bpf_percpu_hash_copy(map, key, value);
 876         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 877                 err = bpf_percpu_array_copy(map, key, value);
 878         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 879                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 880         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 881                 err = bpf_stackmap_copy(map, key, value);
 882         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 883                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 884         } else if (IS_FD_HASH(map)) {
 885                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 886         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 887                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 888         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 889                    map->map_type == BPF_MAP_TYPE_STACK) {
 890                 err = map->ops->map_peek_elem(map, value);
 891         } else {
 892                 rcu_read_lock();
 893                 if (map->ops->map_lookup_elem_sys_only)
 894                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 895                 else
 896                         ptr = map->ops->map_lookup_elem(map, key);
 897                 if (IS_ERR(ptr)) {
 898                         err = PTR_ERR(ptr);
 899                 } else if (!ptr) {
 900                         err = -ENOENT;
 901                 } else {
 902                         err = 0;
 903                         if (attr->flags & BPF_F_LOCK)
 904                                 /* lock 'ptr' and copy everything but lock */
 905                                 copy_map_value_locked(map, value, ptr, true);
 906                         else
 907                                 copy_map_value(map, value, ptr);
 908                         /* mask lock, since value wasn't zero inited */
 909                         check_and_init_map_lock(map, value);
 910                 }
 911                 rcu_read_unlock();
 912         }
 913         this_cpu_dec(bpf_prog_active);
 914         preempt_enable();
 915
 916 done:
 917         if (err)
 918                 goto free_value;
 919
 920         err = -EFAULT;
 921         if (copy_to_user(uvalue, value, value_size) != 0)
 922                 goto free_value;
 923
 924         err = 0;
 925
 926 free_value:
 927         kfree(value);
 928 free_key:
 929         kfree(key);
 930 err_put:
 931         fdput(f);
 932         return err;
 933 }
 934
 935 static void maybe_wait_bpf_programs(struct bpf_map *map)
 936 {
 937         /* Wait for any running BPF programs to complete so that
 938          * userspace, when we return to it, knows that all programs
 939          * that could be running use the new map value.
 940          */
 941         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 942             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 943                 synchronize_rcu();
 944 }
 945
 946 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
 947
 948 static int map_update_elem(union bpf_attr *attr)
 949 {
 950         void __user *ukey = u64_to_user_ptr(attr->key);
 951         void __user *uvalue = u64_to_user_ptr(attr->value);
 952         int ufd = attr->map_fd;
 953         struct bpf_map *map;
 954         void *key, *value;
 955         u32 value_size;
 956         struct fd f;
 957         int err;
 958
 959         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
 960                 return -EINVAL;
 961
 962         f = fdget(ufd);
 963         map = __bpf_map_get(f);
 964         if (IS_ERR(map))
 965                 return PTR_ERR(map);
 966         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
 967                 err = -EPERM;
 968                 goto err_put;
 969         }
 970
 971         if ((attr->flags & BPF_F_LOCK) &&
 972             !map_value_has_spin_lock(map)) {
 973                 err = -EINVAL;
 974                 goto err_put;
 975         }
 976
 977         key = __bpf_copy_key(ukey, map->key_size);
 978         if (IS_ERR(key)) {
 979                 err = PTR_ERR(key);
 980                 goto err_put;
 981         }
 982
 983         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 984             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 985             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 986             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 987                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
 988         else
 989                 value_size = map->value_size;
 990
 991         err = -ENOMEM;
 992         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
 993         if (!value)
 994                 goto free_key;
 995
 996         err = -EFAULT;
 997         if (copy_from_user(value, uvalue, value_size) != 0)
 998                 goto free_value;
 999
1000         /* Need to create a kthread, thus must support schedule */
1001         if (bpf_map_is_dev_bound(map)) {
1002                 err = bpf_map_offload_update_elem(map, key, value, attr->flags);
1003                 goto out;
1004         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
1005                    map->map_type == BPF_MAP_TYPE_SOCKHASH ||
1006                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
1007                 err = map->ops->map_update_elem(map, key, value, attr->flags);
1008                 goto out;
1009         } else if (IS_FD_PROG_ARRAY(map)) {
1010                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1011                                                    attr->flags);
1012                 goto out;
1013         }
1014
1015         /* must increment bpf_prog_active to avoid kprobe+bpf triggering from
1016          * inside bpf map update or delete otherwise deadlocks are possible
1017          */
1018         preempt_disable();
1019         __this_cpu_inc(bpf_prog_active);
1020         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1021             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
1022                 err = bpf_percpu_hash_update(map, key, value, attr->flags);
1023         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
1024                 err = bpf_percpu_array_update(map, key, value, attr->flags);
1025         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
1026                 err = bpf_percpu_cgroup_storage_update(map, key, value,
1027                                                        attr->flags);
1028         } else if (IS_FD_ARRAY(map)) {
1029                 rcu_read_lock();
1030                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
1031                                                    attr->flags);
1032                 rcu_read_unlock();
1033         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
1034                 rcu_read_lock();
1035                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
1036                                                   attr->flags);
1037                 rcu_read_unlock();
1038         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
1039                 /* rcu_read_lock() is not needed */
1040                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
1041                                                          attr->flags);
1042         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1043                    map->map_type == BPF_MAP_TYPE_STACK) {
1044                 err = map->ops->map_push_elem(map, value, attr->flags);
1045         } else {
1046                 rcu_read_lock();
1047                 err = map->ops->map_update_elem(map, key, value, attr->flags);
1048                 rcu_read_unlock();
1049         }
1050         __this_cpu_dec(bpf_prog_active);
1051         preempt_enable();
1052         maybe_wait_bpf_programs(map);
1053 out:
1054 free_value:
1055         kfree(value);
1056 free_key:
1057         kfree(key);
1058 err_put:
1059         fdput(f);
1060         return err;
1061 }
1062
1063 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1064
1065 static int map_delete_elem(union bpf_attr *attr)
1066 {
1067         void __user *ukey = u64_to_user_ptr(attr->key);
1068         int ufd = attr->map_fd;
1069         struct bpf_map *map;
1070         struct fd f;
1071         void *key;
1072         int err;
1073
1074         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1075                 return -EINVAL;
1076
1077         f = fdget(ufd);
1078         map = __bpf_map_get(f);
1079         if (IS_ERR(map))
1080                 return PTR_ERR(map);
1081         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1082                 err = -EPERM;
1083                 goto err_put;
1084         }
1085
1086         key = __bpf_copy_key(ukey, map->key_size);
1087         if (IS_ERR(key)) {
1088                 err = PTR_ERR(key);
1089                 goto err_put;
1090         }
1091
1092         if (bpf_map_is_dev_bound(map)) {
1093                 err = bpf_map_offload_delete_elem(map, key);
1094                 goto out;
1095         } else if (IS_FD_PROG_ARRAY(map)) {
1096                 err = map->ops->map_delete_elem(map, key);
1097                 goto out;
1098         }
1099
1100         preempt_disable();
1101         __this_cpu_inc(bpf_prog_active);
1102         rcu_read_lock();
1103         err = map->ops->map_delete_elem(map, key);
1104         rcu_read_unlock();
1105         __this_cpu_dec(bpf_prog_active);
1106         preempt_enable();
1107         maybe_wait_bpf_programs(map);
1108 out:
1109         kfree(key);
1110 err_put:
1111         fdput(f);
1112         return err;
1113 }
1114
1115 /* last field in 'union bpf_attr' used by this command */
1116 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1117
1118 static int map_get_next_key(union bpf_attr *attr)
1119 {
1120         void __user *ukey = u64_to_user_ptr(attr->key);
1121         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1122         int ufd = attr->map_fd;
1123         struct bpf_map *map;
1124         void *key, *next_key;
1125         struct fd f;
1126         int err;
1127
1128         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1129                 return -EINVAL;
1130
1131         f = fdget(ufd);
1132         map = __bpf_map_get(f);
1133         if (IS_ERR(map))
1134                 return PTR_ERR(map);
1135         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1136                 err = -EPERM;
1137                 goto err_put;
1138         }
1139
1140         if (ukey) {
1141                 key = __bpf_copy_key(ukey, map->key_size);
1142                 if (IS_ERR(key)) {
1143                         err = PTR_ERR(key);
1144                         goto err_put;
1145                 }
1146         } else {
1147                 key = NULL;
1148         }
1149
1150         err = -ENOMEM;
1151         next_key = kmalloc(map->key_size, GFP_USER);
1152         if (!next_key)
1153                 goto free_key;
1154
1155         if (bpf_map_is_dev_bound(map)) {
1156                 err = bpf_map_offload_get_next_key(map, key, next_key);
1157                 goto out;
1158         }
1159
1160         rcu_read_lock();
1161         err = map->ops->map_get_next_key(map, key, next_key);
1162         rcu_read_unlock();
1163 out:
1164         if (err)
1165                 goto free_next_key;
1166
1167         err = -EFAULT;
1168         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1169                 goto free_next_key;
1170
1171         err = 0;
1172
1173 free_next_key:
1174         kfree(next_key);
1175 free_key:
1176         kfree(key);
1177 err_put:
1178         fdput(f);
1179         return err;
1180 }
1181
1182 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1183
1184 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1185 {
1186         void __user *ukey = u64_to_user_ptr(attr->key);
1187         void __user *uvalue = u64_to_user_ptr(attr->value);
1188         int ufd = attr->map_fd;
1189         struct bpf_map *map;
1190         void *key, *value;
1191         u32 value_size;
1192         struct fd f;
1193         int err;
1194
1195         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1196                 return -EINVAL;
1197
1198         f = fdget(ufd);
1199         map = __bpf_map_get(f);
1200         if (IS_ERR(map))
1201                 return PTR_ERR(map);
1202         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1203                 err = -EPERM;
1204                 goto err_put;
1205         }
1206
1207         key = __bpf_copy_key(ukey, map->key_size);
1208         if (IS_ERR(key)) {
1209                 err = PTR_ERR(key);
1210                 goto err_put;
1211         }
1212
1213         value_size = map->value_size;
1214
1215         err = -ENOMEM;
1216         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1217         if (!value)
1218                 goto free_key;
1219
1220         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1221             map->map_type == BPF_MAP_TYPE_STACK) {
1222                 err = map->ops->map_pop_elem(map, value);
1223         } else {
1224                 err = -ENOTSUPP;
1225         }
1226
1227         if (err)
1228                 goto free_value;
1229
1230         if (copy_to_user(uvalue, value, value_size) != 0)
1231                 goto free_value;
1232
1233         err = 0;
1234
1235 free_value:
1236         kfree(value);
1237 free_key:
1238         kfree(key);
1239 err_put:
1240         fdput(f);
1241         return err;
1242 }
1243
1244 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1245
1246 static int map_freeze(const union bpf_attr *attr)
1247 {
1248         int err = 0, ufd = attr->map_fd;
1249         struct bpf_map *map;
1250         struct fd f;
1251
1252         if (CHECK_ATTR(BPF_MAP_FREEZE))
1253                 return -EINVAL;
1254
1255         f = fdget(ufd);
1256         map = __bpf_map_get(f);
1257         if (IS_ERR(map))
1258                 return PTR_ERR(map);
1259
1260         mutex_lock(&map->freeze_mutex);
1261
1262         if (map->writecnt) {
1263                 err = -EBUSY;
1264                 goto err_put;
1265         }
1266         if (READ_ONCE(map->frozen)) {
1267                 err = -EBUSY;
1268                 goto err_put;
1269         }
1270         if (!capable(CAP_SYS_ADMIN)) {
1271                 err = -EPERM;
1272                 goto err_put;
1273         }
1274
1275         WRITE_ONCE(map->frozen, true);
1276 err_put:
1277         mutex_unlock(&map->freeze_mutex);
1278         fdput(f);
1279         return err;
1280 }
1281
1282 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1283 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1284         [_id] = & _name ## _prog_ops,
1285 #define BPF_MAP_TYPE(_id, _ops)
1286 #include <linux/bpf_types.h>
1287 #undef BPF_PROG_TYPE
1288 #undef BPF_MAP_TYPE
1289 };
1290
1291 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1292 {
1293         const struct bpf_prog_ops *ops;
1294
1295         if (type >= ARRAY_SIZE(bpf_prog_types))
1296                 return -EINVAL;
1297         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1298         ops = bpf_prog_types[type];
1299         if (!ops)
1300                 return -EINVAL;
1301
1302         if (!bpf_prog_is_dev_bound(prog->aux))
1303                 prog->aux->ops = ops;
1304         else
1305                 prog->aux->ops = &bpf_offload_prog_ops;
1306         prog->type = type;
1307         return 0;
1308 }
1309
1310 enum bpf_audit {
1311         BPF_AUDIT_LOAD,
1312         BPF_AUDIT_UNLOAD,
1313         BPF_AUDIT_MAX,
1314 };
1315
1316 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1317         [BPF_AUDIT_LOAD]   = "LOAD",
1318         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1319 };
1320
1321 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1322 {
1323         struct audit_context *ctx = NULL;
1324         struct audit_buffer *ab;
1325
1326         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1327                 return;
1328         if (audit_enabled == AUDIT_OFF)
1329                 return;
1330         if (op == BPF_AUDIT_LOAD)
1331                 ctx = audit_context();
1332         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1333         if (unlikely(!ab))
1334                 return;
1335         audit_log_format(ab, "prog-id=%u op=%s",
1336                          prog->aux->id, bpf_audit_str[op]);
1337         audit_log_end(ab);
1338 }
1339
1340 int __bpf_prog_charge(struct user_struct *user, u32 pages)
1341 {
1342         unsigned long memlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1343         unsigned long user_bufs;
1344
1345         if (user) {
1346                 user_bufs = atomic_long_add_return(pages, &user->locked_vm);
1347                 if (user_bufs > memlock_limit) {
1348                         atomic_long_sub(pages, &user->locked_vm);
1349                         return -EPERM;
1350                 }
1351         }
1352
1353         return 0;
1354 }
1355
1356 void __bpf_prog_uncharge(struct user_struct *user, u32 pages)
1357 {
1358         if (user)
1359                 atomic_long_sub(pages, &user->locked_vm);
1360 }
1361
1362 static int bpf_prog_charge_memlock(struct bpf_prog *prog)
1363 {
1364         struct user_struct *user = get_current_user();
1365         int ret;
1366
1367         ret = __bpf_prog_charge(user, prog->pages);
1368         if (ret) {
1369                 free_uid(user);
1370                 return ret;
1371         }
1372
1373         prog->aux->user = user;
1374         return 0;
1375 }
1376
1377 static void bpf_prog_uncharge_memlock(struct bpf_prog *prog)
1378 {
1379         struct user_struct *user = prog->aux->user;
1380
1381         __bpf_prog_uncharge(user, prog->pages);
1382         free_uid(user);
1383 }
1384
1385 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1386 {
1387         int id;
1388
1389         idr_preload(GFP_KERNEL);
1390         spin_lock_bh(&prog_idr_lock);
1391         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1392         if (id > 0)
1393                 prog->aux->id = id;
1394         spin_unlock_bh(&prog_idr_lock);
1395         idr_preload_end();
1396
1397         /* id is in [1, INT_MAX) */
1398         if (WARN_ON_ONCE(!id))
1399                 return -ENOSPC;
1400
1401         return id > 0 ? 0 : id;
1402 }
1403
1404 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1405 {
1406         /* cBPF to eBPF migrations are currently not in the idr store.
1407          * Offloaded programs are removed from the store when their device
1408          * disappears - even if someone grabs an fd to them they are unusable,
1409          * simply waiting for refcnt to drop to be freed.
1410          */
1411         if (!prog->aux->id)
1412                 return;
1413
1414         if (do_idr_lock)
1415                 spin_lock_bh(&prog_idr_lock);
1416         else
1417                 __acquire(&prog_idr_lock);
1418
1419         idr_remove(&prog_idr, prog->aux->id);
1420         prog->aux->id = 0;
1421
1422         if (do_idr_lock)
1423                 spin_unlock_bh(&prog_idr_lock);
1424         else
1425                 __release(&prog_idr_lock);
1426 }
1427
1428 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1429 {
1430         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1431
1432         kvfree(aux->func_info);
1433         kfree(aux->func_info_aux);
1434         bpf_prog_uncharge_memlock(aux->prog);
1435         security_bpf_prog_free(aux);
1436         bpf_prog_free(aux->prog);
1437 }
1438
1439 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1440 {
1441         bpf_prog_kallsyms_del_all(prog);
1442         btf_put(prog->aux->btf);
1443         bpf_prog_free_linfo(prog);
1444
1445         if (deferred)
1446                 call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1447         else
1448                 __bpf_prog_put_rcu(&prog->aux->rcu);
1449 }
1450
1451 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1452 {
1453         if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1454                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1455                 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1456                 /* bpf_prog_free_id() must be called first */
1457                 bpf_prog_free_id(prog, do_idr_lock);
1458                 __bpf_prog_put_noref(prog, true);
1459         }
1460 }
1461
1462 void bpf_prog_put(struct bpf_prog *prog)
1463 {
1464         __bpf_prog_put(prog, true);
1465 }
1466 EXPORT_SYMBOL_GPL(bpf_prog_put);
1467
1468 static int bpf_prog_release(struct inode *inode, struct file *filp)
1469 {
1470         struct bpf_prog *prog = filp->private_data;
1471
1472         bpf_prog_put(prog);
1473         return 0;
1474 }
1475
1476 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1477                                struct bpf_prog_stats *stats)
1478 {
1479         u64 nsecs = 0, cnt = 0;
1480         int cpu;
1481
1482         for_each_possible_cpu(cpu) {
1483                 const struct bpf_prog_stats *st;
1484                 unsigned int start;
1485                 u64 tnsecs, tcnt;
1486
1487                 st = per_cpu_ptr(prog->aux->stats, cpu);
1488                 do {
1489                         start = u64_stats_fetch_begin_irq(&st->syncp);
1490                         tnsecs = st->nsecs;
1491                         tcnt = st->cnt;
1492                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1493                 nsecs += tnsecs;
1494                 cnt += tcnt;
1495         }
1496         stats->nsecs = nsecs;
1497         stats->cnt = cnt;
1498 }
1499
1500 #ifdef CONFIG_PROC_FS
1501 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1502 {
1503         const struct bpf_prog *prog = filp->private_data;
1504         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1505         struct bpf_prog_stats stats;
1506
1507         bpf_prog_get_stats(prog, &stats);
1508         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1509         seq_printf(m,
1510                    "prog_type:\t%u\n"
1511                    "prog_jited:\t%u\n"
1512                    "prog_tag:\t%s\n"
1513                    "memlock:\t%llu\n"
1514                    "prog_id:\t%u\n"
1515                    "run_time_ns:\t%llu\n"
1516                    "run_cnt:\t%llu\n",
1517                    prog->type,
1518                    prog->jited,
1519                    prog_tag,
1520                    prog->pages * 1ULL << PAGE_SHIFT,
1521                    prog->aux->id,
1522                    stats.nsecs,
1523                    stats.cnt);
1524 }
1525 #endif
1526
1527 const struct file_operations bpf_prog_fops = {
1528 #ifdef CONFIG_PROC_FS
1529         .show_fdinfo    = bpf_prog_show_fdinfo,
1530 #endif
1531         .release        = bpf_prog_release,
1532         .read           = bpf_dummy_read,
1533         .write          = bpf_dummy_write,
1534 };
1535
1536 int bpf_prog_new_fd(struct bpf_prog *prog)
1537 {
1538         int ret;
1539
1540         ret = security_bpf_prog(prog);
1541         if (ret < 0)
1542                 return ret;
1543
1544         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1545                                 O_RDWR | O_CLOEXEC);
1546 }
1547
1548 static struct bpf_prog *____bpf_prog_get(struct fd f)
1549 {
1550         if (!f.file)
1551                 return ERR_PTR(-EBADF);
1552         if (f.file->f_op != &bpf_prog_fops) {
1553                 fdput(f);
1554                 return ERR_PTR(-EINVAL);
1555         }
1556
1557         return f.file->private_data;
1558 }
1559
1560 void bpf_prog_add(struct bpf_prog *prog, int i)
1561 {
1562         atomic64_add(i, &prog->aux->refcnt);
1563 }
1564 EXPORT_SYMBOL_GPL(bpf_prog_add);
1565
1566 void bpf_prog_sub(struct bpf_prog *prog, int i)
1567 {
1568         /* Only to be used for undoing previous bpf_prog_add() in some
1569          * error path. We still know that another entity in our call
1570          * path holds a reference to the program, thus atomic_sub() can
1571          * be safely used in such cases!
1572          */
1573         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1574 }
1575 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1576
1577 void bpf_prog_inc(struct bpf_prog *prog)
1578 {
1579         atomic64_inc(&prog->aux->refcnt);
1580 }
1581 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1582
1583 /* prog_idr_lock should have been held */
1584 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1585 {
1586         int refold;
1587
1588         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1589
1590         if (!refold)
1591                 return ERR_PTR(-ENOENT);
1592
1593         return prog;
1594 }
1595 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1596
1597 bool bpf_prog_get_ok(struct bpf_prog *prog,
1598                             enum bpf_prog_type *attach_type, bool attach_drv)
1599 {
1600         /* not an attachment, just a refcount inc, always allow */
1601         if (!attach_type)
1602                 return true;
1603
1604         if (prog->type != *attach_type)
1605                 return false;
1606         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1607                 return false;
1608
1609         return true;
1610 }
1611
1612 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1613                                        bool attach_drv)
1614 {
1615         struct fd f = fdget(ufd);
1616         struct bpf_prog *prog;
1617
1618         prog = ____bpf_prog_get(f);
1619         if (IS_ERR(prog))
1620                 return prog;
1621         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1622                 prog = ERR_PTR(-EINVAL);
1623                 goto out;
1624         }
1625
1626         bpf_prog_inc(prog);
1627 out:
1628         fdput(f);
1629         return prog;
1630 }
1631
1632 struct bpf_prog *bpf_prog_get(u32 ufd)
1633 {
1634         return __bpf_prog_get(ufd, NULL, false);
1635 }
1636
1637 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1638                                        bool attach_drv)
1639 {
1640         return __bpf_prog_get(ufd, &type, attach_drv);
1641 }
1642 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1643
1644 /* Initially all BPF programs could be loaded w/o specifying
1645  * expected_attach_type. Later for some of them specifying expected_attach_type
1646  * at load time became required so that program could be validated properly.
1647  * Programs of types that are allowed to be loaded both w/ and w/o (for
1648  * backward compatibility) expected_attach_type, should have the default attach
1649  * type assigned to expected_attach_type for the latter case, so that it can be
1650  * validated later at attach time.
1651  *
1652  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1653  * prog type requires it but has some attach types that have to be backward
1654  * compatible.
1655  */
1656 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1657 {
1658         switch (attr->prog_type) {
1659         case BPF_PROG_TYPE_CGROUP_SOCK:
1660                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1661                  * exist so checking for non-zero is the way to go here.
1662                  */
1663                 if (!attr->expected_attach_type)
1664                         attr->expected_attach_type =
1665                                 BPF_CGROUP_INET_SOCK_CREATE;
1666                 break;
1667         }
1668 }
1669
1670 static int
1671 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1672                            enum bpf_attach_type expected_attach_type,
1673                            u32 btf_id, u32 prog_fd)
1674 {
1675         switch (prog_type) {
1676         case BPF_PROG_TYPE_TRACING:
1677                 if (btf_id > BTF_MAX_TYPE)
1678                         return -EINVAL;
1679                 break;
1680         default:
1681                 if (btf_id || prog_fd)
1682                         return -EINVAL;
1683                 break;
1684         }
1685
1686         switch (prog_type) {
1687         case BPF_PROG_TYPE_CGROUP_SOCK:
1688                 switch (expected_attach_type) {
1689                 case BPF_CGROUP_INET_SOCK_CREATE:
1690                 case BPF_CGROUP_INET4_POST_BIND:
1691                 case BPF_CGROUP_INET6_POST_BIND:
1692                         return 0;
1693                 default:
1694                         return -EINVAL;
1695                 }
1696         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1697                 switch (expected_attach_type) {
1698                 case BPF_CGROUP_INET4_BIND:
1699                 case BPF_CGROUP_INET6_BIND:
1700                 case BPF_CGROUP_INET4_CONNECT:
1701                 case BPF_CGROUP_INET6_CONNECT:
1702                 case BPF_CGROUP_UDP4_SENDMSG:
1703                 case BPF_CGROUP_UDP6_SENDMSG:
1704                 case BPF_CGROUP_UDP4_RECVMSG:
1705                 case BPF_CGROUP_UDP6_RECVMSG:
1706                         return 0;
1707                 default:
1708                         return -EINVAL;
1709                 }
1710         case BPF_PROG_TYPE_CGROUP_SKB:
1711                 switch (expected_attach_type) {
1712                 case BPF_CGROUP_INET_INGRESS:
1713                 case BPF_CGROUP_INET_EGRESS:
1714                         return 0;
1715                 default:
1716                         return -EINVAL;
1717                 }
1718         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1719                 switch (expected_attach_type) {
1720                 case BPF_CGROUP_SETSOCKOPT:
1721                 case BPF_CGROUP_GETSOCKOPT:
1722                         return 0;
1723                 default:
1724                         return -EINVAL;
1725                 }
1726         default:
1727                 return 0;
1728         }
1729 }
1730
1731 /* last field in 'union bpf_attr' used by this command */
1732 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
1733
1734 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
1735 {
1736         enum bpf_prog_type type = attr->prog_type;
1737         struct bpf_prog *prog;
1738         int err;
1739         char license[128];
1740         bool is_gpl;
1741
1742         if (CHECK_ATTR(BPF_PROG_LOAD))
1743                 return -EINVAL;
1744
1745         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
1746                                  BPF_F_ANY_ALIGNMENT |
1747                                  BPF_F_TEST_STATE_FREQ |
1748                                  BPF_F_TEST_RND_HI32))
1749                 return -EINVAL;
1750
1751         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
1752             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
1753             !capable(CAP_SYS_ADMIN))
1754                 return -EPERM;
1755
1756         /* copy eBPF program license from user space */
1757         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
1758                               sizeof(license) - 1) < 0)
1759                 return -EFAULT;
1760         license[sizeof(license) - 1] = 0;
1761
1762         /* eBPF programs must be GPL compatible to use GPL-ed functions */
1763         is_gpl = license_is_gpl_compatible(license);
1764
1765         if (attr->insn_cnt == 0 ||
1766             attr->insn_cnt > (capable(CAP_SYS_ADMIN) ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
1767                 return -E2BIG;
1768         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
1769             type != BPF_PROG_TYPE_CGROUP_SKB &&
1770             !capable(CAP_SYS_ADMIN))
1771                 return -EPERM;
1772
1773         bpf_prog_load_fixup_attach_type(attr);
1774         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
1775                                        attr->attach_btf_id,
1776                                        attr->attach_prog_fd))
1777                 return -EINVAL;
1778
1779         /* plain bpf_prog allocation */
1780         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
1781         if (!prog)
1782                 return -ENOMEM;
1783
1784         prog->expected_attach_type = attr->expected_attach_type;
1785         prog->aux->attach_btf_id = attr->attach_btf_id;
1786         if (attr->attach_prog_fd) {
1787                 struct bpf_prog *tgt_prog;
1788
1789                 tgt_prog = bpf_prog_get(attr->attach_prog_fd);
1790                 if (IS_ERR(tgt_prog)) {
1791                         err = PTR_ERR(tgt_prog);
1792                         goto free_prog_nouncharge;
1793                 }
1794                 prog->aux->linked_prog = tgt_prog;
1795         }
1796
1797         prog->aux->offload_requested = !!attr->prog_ifindex;
1798
1799         err = security_bpf_prog_alloc(prog->aux);
1800         if (err)
1801                 goto free_prog_nouncharge;
1802
1803         err = bpf_prog_charge_memlock(prog);
1804         if (err)
1805                 goto free_prog_sec;
1806
1807         prog->len = attr->insn_cnt;
1808
1809         err = -EFAULT;
1810         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
1811                            bpf_prog_insn_size(prog)) != 0)
1812                 goto free_prog;
1813
1814         prog->orig_prog = NULL;
1815         prog->jited = 0;
1816
1817         atomic64_set(&prog->aux->refcnt, 1);
1818         prog->gpl_compatible = is_gpl ? 1 : 0;
1819
1820         if (bpf_prog_is_dev_bound(prog->aux)) {
1821                 err = bpf_prog_offload_init(prog, attr);
1822                 if (err)
1823                         goto free_prog;
1824         }
1825
1826         /* find program type: socket_filter vs tracing_filter */
1827         err = find_prog_type(type, prog);
1828         if (err < 0)
1829                 goto free_prog;
1830
1831         prog->aux->load_time = ktime_get_boottime_ns();
1832         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name);
1833         if (err)
1834                 goto free_prog;
1835
1836         /* run eBPF verifier */
1837         err = bpf_check(&prog, attr, uattr);
1838         if (err < 0)
1839                 goto free_used_maps;
1840
1841         prog = bpf_prog_select_runtime(prog, &err);
1842         if (err < 0)
1843                 goto free_used_maps;
1844
1845         err = bpf_prog_alloc_id(prog);
1846         if (err)
1847                 goto free_used_maps;
1848
1849         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
1850          * effectively publicly exposed. However, retrieving via
1851          * bpf_prog_get_fd_by_id() will take another reference,
1852          * therefore it cannot be gone underneath us.
1853          *
1854          * Only for the time /after/ successful bpf_prog_new_fd()
1855          * and before returning to userspace, we might just hold
1856          * one reference and any parallel close on that fd could
1857          * rip everything out. Hence, below notifications must
1858          * happen before bpf_prog_new_fd().
1859          *
1860          * Also, any failure handling from this point onwards must
1861          * be using bpf_prog_put() given the program is exposed.
1862          */
1863         bpf_prog_kallsyms_add(prog);
1864         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
1865         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
1866
1867         err = bpf_prog_new_fd(prog);
1868         if (err < 0)
1869                 bpf_prog_put(prog);
1870         return err;
1871
1872 free_used_maps:
1873         /* In case we have subprogs, we need to wait for a grace
1874          * period before we can tear down JIT memory since symbols
1875          * are already exposed under kallsyms.
1876          */
1877         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
1878         return err;
1879 free_prog:
1880         bpf_prog_uncharge_memlock(prog);
1881 free_prog_sec:
1882         security_bpf_prog_free(prog->aux);
1883 free_prog_nouncharge:
1884         bpf_prog_free(prog);
1885         return err;
1886 }
1887
1888 #define BPF_OBJ_LAST_FIELD file_flags
1889
1890 static int bpf_obj_pin(const union bpf_attr *attr)
1891 {
1892         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
1893                 return -EINVAL;
1894
1895         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
1896 }
1897
1898 static int bpf_obj_get(const union bpf_attr *attr)
1899 {
1900         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
1901             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
1902                 return -EINVAL;
1903
1904         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
1905                                 attr->file_flags);
1906 }
1907
1908 static int bpf_tracing_prog_release(struct inode *inode, struct file *filp)
1909 {
1910         struct bpf_prog *prog = filp->private_data;
1911
1912         WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1913         bpf_prog_put(prog);
1914         return 0;
1915 }
1916
1917 static const struct file_operations bpf_tracing_prog_fops = {
1918         .release        = bpf_tracing_prog_release,
1919         .read           = bpf_dummy_read,
1920         .write          = bpf_dummy_write,
1921 };
1922
1923 static int bpf_tracing_prog_attach(struct bpf_prog *prog)
1924 {
1925         int tr_fd, err;
1926
1927         if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
1928             prog->expected_attach_type != BPF_TRACE_FEXIT) {
1929                 err = -EINVAL;
1930                 goto out_put_prog;
1931         }
1932
1933         err = bpf_trampoline_link_prog(prog);
1934         if (err)
1935                 goto out_put_prog;
1936
1937         tr_fd = anon_inode_getfd("bpf-tracing-prog", &bpf_tracing_prog_fops,
1938                                  prog, O_CLOEXEC);
1939         if (tr_fd < 0) {
1940                 WARN_ON_ONCE(bpf_trampoline_unlink_prog(prog));
1941                 err = tr_fd;
1942                 goto out_put_prog;
1943         }
1944         return tr_fd;
1945
1946 out_put_prog:
1947         bpf_prog_put(prog);
1948         return err;
1949 }
1950
1951 struct bpf_raw_tracepoint {
1952         struct bpf_raw_event_map *btp;
1953         struct bpf_prog *prog;
1954 };
1955
1956 static int bpf_raw_tracepoint_release(struct inode *inode, struct file *filp)
1957 {
1958         struct bpf_raw_tracepoint *raw_tp = filp->private_data;
1959
1960         if (raw_tp->prog) {
1961                 bpf_probe_unregister(raw_tp->btp, raw_tp->prog);
1962                 bpf_prog_put(raw_tp->prog);
1963         }
1964         bpf_put_raw_tracepoint(raw_tp->btp);
1965         kfree(raw_tp);
1966         return 0;
1967 }
1968
1969 static const struct file_operations bpf_raw_tp_fops = {
1970         .release        = bpf_raw_tracepoint_release,
1971         .read           = bpf_dummy_read,
1972         .write          = bpf_dummy_write,
1973 };
1974
1975 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
1976
1977 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
1978 {
1979         struct bpf_raw_tracepoint *raw_tp;
1980         struct bpf_raw_event_map *btp;
1981         struct bpf_prog *prog;
1982         const char *tp_name;
1983         char buf[128];
1984         int tp_fd, err;
1985
1986         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
1987                 return -EINVAL;
1988
1989         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
1990         if (IS_ERR(prog))
1991                 return PTR_ERR(prog);
1992
1993         if (prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT &&
1994             prog->type != BPF_PROG_TYPE_TRACING &&
1995             prog->type != BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE) {
1996                 err = -EINVAL;
1997                 goto out_put_prog;
1998         }
1999
2000         if (prog->type == BPF_PROG_TYPE_TRACING) {
2001                 if (attr->raw_tracepoint.name) {
2002                         /* The attach point for this category of programs
2003                          * should be specified via btf_id during program load.
2004                          */
2005                         err = -EINVAL;
2006                         goto out_put_prog;
2007                 }
2008                 if (prog->expected_attach_type == BPF_TRACE_RAW_TP)
2009                         tp_name = prog->aux->attach_func_name;
2010                 else
2011                         return bpf_tracing_prog_attach(prog);
2012         } else {
2013                 if (strncpy_from_user(buf,
2014                                       u64_to_user_ptr(attr->raw_tracepoint.name),
2015                                       sizeof(buf) - 1) < 0) {
2016                         err = -EFAULT;
2017                         goto out_put_prog;
2018                 }
2019                 buf[sizeof(buf) - 1] = 0;
2020                 tp_name = buf;
2021         }
2022
2023         btp = bpf_get_raw_tracepoint(tp_name);
2024         if (!btp) {
2025                 err = -ENOENT;
2026                 goto out_put_prog;
2027         }
2028
2029         raw_tp = kzalloc(sizeof(*raw_tp), GFP_USER);
2030         if (!raw_tp) {
2031                 err = -ENOMEM;
2032                 goto out_put_btp;
2033         }
2034         raw_tp->btp = btp;
2035         raw_tp->prog = prog;
2036
2037         err = bpf_probe_register(raw_tp->btp, prog);
2038         if (err)
2039                 goto out_free_tp;
2040
2041         tp_fd = anon_inode_getfd("bpf-raw-tracepoint", &bpf_raw_tp_fops, raw_tp,
2042                                  O_CLOEXEC);
2043         if (tp_fd < 0) {
2044                 bpf_probe_unregister(raw_tp->btp, prog);
2045                 err = tp_fd;
2046                 goto out_free_tp;
2047         }
2048         return tp_fd;
2049
2050 out_free_tp:
2051         kfree(raw_tp);
2052 out_put_btp:
2053         bpf_put_raw_tracepoint(btp);
2054 out_put_prog:
2055         bpf_prog_put(prog);
2056         return err;
2057 }
2058
2059 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2060                                              enum bpf_attach_type attach_type)
2061 {
2062         switch (prog->type) {
2063         case BPF_PROG_TYPE_CGROUP_SOCK:
2064         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2065         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2066                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2067         case BPF_PROG_TYPE_CGROUP_SKB:
2068                 return prog->enforce_expected_attach_type &&
2069                         prog->expected_attach_type != attach_type ?
2070                         -EINVAL : 0;
2071         default:
2072                 return 0;
2073         }
2074 }
2075
2076 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2077
2078 #define BPF_F_ATTACH_MASK \
2079         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2080
2081 static int bpf_prog_attach(const union bpf_attr *attr)
2082 {
2083         enum bpf_prog_type ptype;
2084         struct bpf_prog *prog;
2085         int ret;
2086
2087         if (!capable(CAP_NET_ADMIN))
2088                 return -EPERM;
2089
2090         if (CHECK_ATTR(BPF_PROG_ATTACH))
2091                 return -EINVAL;
2092
2093         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2094                 return -EINVAL;
2095
2096         switch (attr->attach_type) {
2097         case BPF_CGROUP_INET_INGRESS:
2098         case BPF_CGROUP_INET_EGRESS:
2099                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2100                 break;
2101         case BPF_CGROUP_INET_SOCK_CREATE:
2102         case BPF_CGROUP_INET4_POST_BIND:
2103         case BPF_CGROUP_INET6_POST_BIND:
2104                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2105                 break;
2106         case BPF_CGROUP_INET4_BIND:
2107         case BPF_CGROUP_INET6_BIND:
2108         case BPF_CGROUP_INET4_CONNECT:
2109         case BPF_CGROUP_INET6_CONNECT:
2110         case BPF_CGROUP_UDP4_SENDMSG:
2111         case BPF_CGROUP_UDP6_SENDMSG:
2112         case BPF_CGROUP_UDP4_RECVMSG:
2113         case BPF_CGROUP_UDP6_RECVMSG:
2114                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2115                 break;
2116         case BPF_CGROUP_SOCK_OPS:
2117                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2118                 break;
2119         case BPF_CGROUP_DEVICE:
2120                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2121                 break;
2122         case BPF_SK_MSG_VERDICT:
2123                 ptype = BPF_PROG_TYPE_SK_MSG;
2124                 break;
2125         case BPF_SK_SKB_STREAM_PARSER:
2126         case BPF_SK_SKB_STREAM_VERDICT:
2127                 ptype = BPF_PROG_TYPE_SK_SKB;
2128                 break;
2129         case BPF_LIRC_MODE2:
2130                 ptype = BPF_PROG_TYPE_LIRC_MODE2;
2131                 break;
2132         case BPF_FLOW_DISSECTOR:
2133                 ptype = BPF_PROG_TYPE_FLOW_DISSECTOR;
2134                 break;
2135         case BPF_CGROUP_SYSCTL:
2136                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2137                 break;
2138         case BPF_CGROUP_GETSOCKOPT:
2139         case BPF_CGROUP_SETSOCKOPT:
2140                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2141                 break;
2142         default:
2143                 return -EINVAL;
2144         }
2145
2146         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2147         if (IS_ERR(prog))
2148                 return PTR_ERR(prog);
2149
2150         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2151                 bpf_prog_put(prog);
2152                 return -EINVAL;
2153         }
2154
2155         switch (ptype) {
2156         case BPF_PROG_TYPE_SK_SKB:
2157         case BPF_PROG_TYPE_SK_MSG:
2158                 ret = sock_map_get_from_fd(attr, prog);
2159                 break;
2160         case BPF_PROG_TYPE_LIRC_MODE2:
2161                 ret = lirc_prog_attach(attr, prog);
2162                 break;
2163         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2164                 ret = skb_flow_dissector_bpf_prog_attach(attr, prog);
2165                 break;
2166         default:
2167                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
2168         }
2169
2170         if (ret)
2171                 bpf_prog_put(prog);
2172         return ret;
2173 }
2174
2175 #define BPF_PROG_DETACH_LAST_FIELD attach_type
2176
2177 static int bpf_prog_detach(const union bpf_attr *attr)
2178 {
2179         enum bpf_prog_type ptype;
2180
2181         if (!capable(CAP_NET_ADMIN))
2182                 return -EPERM;
2183
2184         if (CHECK_ATTR(BPF_PROG_DETACH))
2185                 return -EINVAL;
2186
2187         switch (attr->attach_type) {
2188         case BPF_CGROUP_INET_INGRESS:
2189         case BPF_CGROUP_INET_EGRESS:
2190                 ptype = BPF_PROG_TYPE_CGROUP_SKB;
2191                 break;
2192         case BPF_CGROUP_INET_SOCK_CREATE:
2193         case BPF_CGROUP_INET4_POST_BIND:
2194         case BPF_CGROUP_INET6_POST_BIND:
2195                 ptype = BPF_PROG_TYPE_CGROUP_SOCK;
2196                 break;
2197         case BPF_CGROUP_INET4_BIND:
2198         case BPF_CGROUP_INET6_BIND:
2199         case BPF_CGROUP_INET4_CONNECT:
2200         case BPF_CGROUP_INET6_CONNECT:
2201         case BPF_CGROUP_UDP4_SENDMSG:
2202         case BPF_CGROUP_UDP6_SENDMSG:
2203         case BPF_CGROUP_UDP4_RECVMSG:
2204         case BPF_CGROUP_UDP6_RECVMSG:
2205                 ptype = BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2206                 break;
2207         case BPF_CGROUP_SOCK_OPS:
2208                 ptype = BPF_PROG_TYPE_SOCK_OPS;
2209                 break;
2210         case BPF_CGROUP_DEVICE:
2211                 ptype = BPF_PROG_TYPE_CGROUP_DEVICE;
2212                 break;
2213         case BPF_SK_MSG_VERDICT:
2214                 return sock_map_get_from_fd(attr, NULL);
2215         case BPF_SK_SKB_STREAM_PARSER:
2216         case BPF_SK_SKB_STREAM_VERDICT:
2217                 return sock_map_get_from_fd(attr, NULL);
2218         case BPF_LIRC_MODE2:
2219                 return lirc_prog_detach(attr);
2220         case BPF_FLOW_DISSECTOR:
2221                 return skb_flow_dissector_bpf_prog_detach(attr);
2222         case BPF_CGROUP_SYSCTL:
2223                 ptype = BPF_PROG_TYPE_CGROUP_SYSCTL;
2224                 break;
2225         case BPF_CGROUP_GETSOCKOPT:
2226         case BPF_CGROUP_SETSOCKOPT:
2227                 ptype = BPF_PROG_TYPE_CGROUP_SOCKOPT;
2228                 break;
2229         default:
2230                 return -EINVAL;
2231         }
2232
2233         return cgroup_bpf_prog_detach(attr, ptype);
2234 }
2235
2236 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
2237
2238 static int bpf_prog_query(const union bpf_attr *attr,
2239                           union bpf_attr __user *uattr)
2240 {
2241         if (!capable(CAP_NET_ADMIN))
2242                 return -EPERM;
2243         if (CHECK_ATTR(BPF_PROG_QUERY))
2244                 return -EINVAL;
2245         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
2246                 return -EINVAL;
2247
2248         switch (attr->query.attach_type) {
2249         case BPF_CGROUP_INET_INGRESS:
2250         case BPF_CGROUP_INET_EGRESS:
2251         case BPF_CGROUP_INET_SOCK_CREATE:
2252         case BPF_CGROUP_INET4_BIND:
2253         case BPF_CGROUP_INET6_BIND:
2254         case BPF_CGROUP_INET4_POST_BIND:
2255         case BPF_CGROUP_INET6_POST_BIND:
2256         case BPF_CGROUP_INET4_CONNECT:
2257         case BPF_CGROUP_INET6_CONNECT:
2258         case BPF_CGROUP_UDP4_SENDMSG:
2259         case BPF_CGROUP_UDP6_SENDMSG:
2260         case BPF_CGROUP_UDP4_RECVMSG:
2261         case BPF_CGROUP_UDP6_RECVMSG:
2262         case BPF_CGROUP_SOCK_OPS:
2263         case BPF_CGROUP_DEVICE:
2264         case BPF_CGROUP_SYSCTL:
2265         case BPF_CGROUP_GETSOCKOPT:
2266         case BPF_CGROUP_SETSOCKOPT:
2267                 break;
2268         case BPF_LIRC_MODE2:
2269                 return lirc_prog_query(attr, uattr);
2270         case BPF_FLOW_DISSECTOR:
2271                 return skb_flow_dissector_prog_query(attr, uattr);
2272         default:
2273                 return -EINVAL;
2274         }
2275
2276         return cgroup_bpf_prog_query(attr, uattr);
2277 }
2278
2279 #define BPF_PROG_TEST_RUN_LAST_FIELD test.ctx_out
2280
2281 static int bpf_prog_test_run(const union bpf_attr *attr,
2282                              union bpf_attr __user *uattr)
2283 {
2284         struct bpf_prog *prog;
2285         int ret = -ENOTSUPP;
2286
2287         if (!capable(CAP_SYS_ADMIN))
2288                 return -EPERM;
2289         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
2290                 return -EINVAL;
2291
2292         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
2293             (!attr->test.ctx_size_in && attr->test.ctx_in))
2294                 return -EINVAL;
2295
2296         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
2297             (!attr->test.ctx_size_out && attr->test.ctx_out))
2298                 return -EINVAL;
2299
2300         prog = bpf_prog_get(attr->test.prog_fd);
2301         if (IS_ERR(prog))
2302                 return PTR_ERR(prog);
2303
2304         if (prog->aux->ops->test_run)
2305                 ret = prog->aux->ops->test_run(prog, attr, uattr);
2306
2307         bpf_prog_put(prog);
2308         return ret;
2309 }
2310
2311 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
2312
2313 static int bpf_obj_get_next_id(const union bpf_attr *attr,
2314                                union bpf_attr __user *uattr,
2315                                struct idr *idr,
2316                                spinlock_t *lock)
2317 {
2318         u32 next_id = attr->start_id;
2319         int err = 0;
2320
2321         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
2322                 return -EINVAL;
2323
2324         if (!capable(CAP_SYS_ADMIN))
2325                 return -EPERM;
2326
2327         next_id++;
2328         spin_lock_bh(lock);
2329         if (!idr_get_next(idr, &next_id))
2330                 err = -ENOENT;
2331         spin_unlock_bh(lock);
2332
2333         if (!err)
2334                 err = put_user(next_id, &uattr->next_id);
2335
2336         return err;
2337 }
2338
2339 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
2340
2341 struct bpf_prog *bpf_prog_by_id(u32 id)
2342 {
2343         struct bpf_prog *prog;
2344
2345         if (!id)
2346                 return ERR_PTR(-ENOENT);
2347
2348         spin_lock_bh(&prog_idr_lock);
2349         prog = idr_find(&prog_idr, id);
2350         if (prog)
2351                 prog = bpf_prog_inc_not_zero(prog);
2352         else
2353                 prog = ERR_PTR(-ENOENT);
2354         spin_unlock_bh(&prog_idr_lock);
2355         return prog;
2356 }
2357
2358 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
2359 {
2360         struct bpf_prog *prog;
2361         u32 id = attr->prog_id;
2362         int fd;
2363
2364         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
2365                 return -EINVAL;
2366
2367         if (!capable(CAP_SYS_ADMIN))
2368                 return -EPERM;
2369
2370         prog = bpf_prog_by_id(id);
2371         if (IS_ERR(prog))
2372                 return PTR_ERR(prog);
2373
2374         fd = bpf_prog_new_fd(prog);
2375         if (fd < 0)
2376                 bpf_prog_put(prog);
2377
2378         return fd;
2379 }
2380
2381 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
2382
2383 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
2384 {
2385         struct bpf_map *map;
2386         u32 id = attr->map_id;
2387         int f_flags;
2388         int fd;
2389
2390         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
2391             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
2392                 return -EINVAL;
2393
2394         if (!capable(CAP_SYS_ADMIN))
2395                 return -EPERM;
2396
2397         f_flags = bpf_get_file_flag(attr->open_flags);
2398         if (f_flags < 0)
2399                 return f_flags;
2400
2401         spin_lock_bh(&map_idr_lock);
2402         map = idr_find(&map_idr, id);
2403         if (map)
2404                 map = __bpf_map_inc_not_zero(map, true);
2405         else
2406                 map = ERR_PTR(-ENOENT);
2407         spin_unlock_bh(&map_idr_lock);
2408
2409         if (IS_ERR(map))
2410                 return PTR_ERR(map);
2411
2412         fd = bpf_map_new_fd(map, f_flags);
2413         if (fd < 0)
2414                 bpf_map_put_with_uref(map);
2415
2416         return fd;
2417 }
2418
2419 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
2420                                               unsigned long addr, u32 *off,
2421                                               u32 *type)
2422 {
2423         const struct bpf_map *map;
2424         int i;
2425
2426         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
2427                 map = prog->aux->used_maps[i];
2428                 if (map == (void *)addr) {
2429                         *type = BPF_PSEUDO_MAP_FD;
2430                         return map;
2431                 }
2432                 if (!map->ops->map_direct_value_meta)
2433                         continue;
2434                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
2435                         *type = BPF_PSEUDO_MAP_VALUE;
2436                         return map;
2437                 }
2438         }
2439
2440         return NULL;
2441 }
2442
2443 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog)
2444 {
2445         const struct bpf_map *map;
2446         struct bpf_insn *insns;
2447         u32 off, type;
2448         u64 imm;
2449         int i;
2450
2451         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
2452                         GFP_USER);
2453         if (!insns)
2454                 return insns;
2455
2456         for (i = 0; i < prog->len; i++) {
2457                 if (insns[i].code == (BPF_JMP | BPF_TAIL_CALL)) {
2458                         insns[i].code = BPF_JMP | BPF_CALL;
2459                         insns[i].imm = BPF_FUNC_tail_call;
2460                         /* fall-through */
2461                 }
2462                 if (insns[i].code == (BPF_JMP | BPF_CALL) ||
2463                     insns[i].code == (BPF_JMP | BPF_CALL_ARGS)) {
2464                         if (insns[i].code == (BPF_JMP | BPF_CALL_ARGS))
2465                                 insns[i].code = BPF_JMP | BPF_CALL;
2466                         if (!bpf_dump_raw_ok())
2467                                 insns[i].imm = 0;
2468                         continue;
2469                 }
2470
2471                 if (insns[i].code != (BPF_LD | BPF_IMM | BPF_DW))
2472                         continue;
2473
2474                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
2475                 map = bpf_map_from_imm(prog, imm, &off, &type);
2476                 if (map) {
2477                         insns[i].src_reg = type;
2478                         insns[i].imm = map->id;
2479                         insns[i + 1].imm = off;
2480                         continue;
2481                 }
2482         }
2483
2484         return insns;
2485 }
2486
2487 static int set_info_rec_size(struct bpf_prog_info *info)
2488 {
2489         /*
2490          * Ensure info.*_rec_size is the same as kernel expected size
2491          *
2492          * or
2493          *
2494          * Only allow zero *_rec_size if both _rec_size and _cnt are
2495          * zero.  In this case, the kernel will set the expected
2496          * _rec_size back to the info.
2497          */
2498
2499         if ((info->nr_func_info || info->func_info_rec_size) &&
2500             info->func_info_rec_size != sizeof(struct bpf_func_info))
2501                 return -EINVAL;
2502
2503         if ((info->nr_line_info || info->line_info_rec_size) &&
2504             info->line_info_rec_size != sizeof(struct bpf_line_info))
2505                 return -EINVAL;
2506
2507         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
2508             info->jited_line_info_rec_size != sizeof(__u64))
2509                 return -EINVAL;
2510
2511         info->func_info_rec_size = sizeof(struct bpf_func_info);
2512         info->line_info_rec_size = sizeof(struct bpf_line_info);
2513         info->jited_line_info_rec_size = sizeof(__u64);
2514
2515         return 0;
2516 }
2517
2518 static int bpf_prog_get_info_by_fd(struct bpf_prog *prog,
2519                                    const union bpf_attr *attr,
2520                                    union bpf_attr __user *uattr)
2521 {
2522         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2523         struct bpf_prog_info info = {};
2524         u32 info_len = attr->info.info_len;
2525         struct bpf_prog_stats stats;
2526         char __user *uinsns;
2527         u32 ulen;
2528         int err;
2529
2530         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2531         if (err)
2532                 return err;
2533         info_len = min_t(u32, sizeof(info), info_len);
2534
2535         if (copy_from_user(&info, uinfo, info_len))
2536                 return -EFAULT;
2537
2538         info.type = prog->type;
2539         info.id = prog->aux->id;
2540         info.load_time = prog->aux->load_time;
2541         info.created_by_uid = from_kuid_munged(current_user_ns(),
2542                                                prog->aux->user->uid);
2543         info.gpl_compatible = prog->gpl_compatible;
2544
2545         memcpy(info.tag, prog->tag, sizeof(prog->tag));
2546         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
2547
2548         ulen = info.nr_map_ids;
2549         info.nr_map_ids = prog->aux->used_map_cnt;
2550         ulen = min_t(u32, info.nr_map_ids, ulen);
2551         if (ulen) {
2552                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
2553                 u32 i;
2554
2555                 for (i = 0; i < ulen; i++)
2556                         if (put_user(prog->aux->used_maps[i]->id,
2557                                      &user_map_ids[i]))
2558                                 return -EFAULT;
2559         }
2560
2561         err = set_info_rec_size(&info);
2562         if (err)
2563                 return err;
2564
2565         bpf_prog_get_stats(prog, &stats);
2566         info.run_time_ns = stats.nsecs;
2567         info.run_cnt = stats.cnt;
2568
2569         if (!capable(CAP_SYS_ADMIN)) {
2570                 info.jited_prog_len = 0;
2571                 info.xlated_prog_len = 0;
2572                 info.nr_jited_ksyms = 0;
2573                 info.nr_jited_func_lens = 0;
2574                 info.nr_func_info = 0;
2575                 info.nr_line_info = 0;
2576                 info.nr_jited_line_info = 0;
2577                 goto done;
2578         }
2579
2580         ulen = info.xlated_prog_len;
2581         info.xlated_prog_len = bpf_prog_insn_size(prog);
2582         if (info.xlated_prog_len && ulen) {
2583                 struct bpf_insn *insns_sanitized;
2584                 bool fault;
2585
2586                 if (prog->blinded && !bpf_dump_raw_ok()) {
2587                         info.xlated_prog_insns = 0;
2588                         goto done;
2589                 }
2590                 insns_sanitized = bpf_insn_prepare_dump(prog);
2591                 if (!insns_sanitized)
2592                         return -ENOMEM;
2593                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
2594                 ulen = min_t(u32, info.xlated_prog_len, ulen);
2595                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
2596                 kfree(insns_sanitized);
2597                 if (fault)
2598                         return -EFAULT;
2599         }
2600
2601         if (bpf_prog_is_dev_bound(prog->aux)) {
2602                 err = bpf_prog_offload_info_fill(&info, prog);
2603                 if (err)
2604                         return err;
2605                 goto done;
2606         }
2607
2608         /* NOTE: the following code is supposed to be skipped for offload.
2609          * bpf_prog_offload_info_fill() is the place to fill similar fields
2610          * for offload.
2611          */
2612         ulen = info.jited_prog_len;
2613         if (prog->aux->func_cnt) {
2614                 u32 i;
2615
2616                 info.jited_prog_len = 0;
2617                 for (i = 0; i < prog->aux->func_cnt; i++)
2618                         info.jited_prog_len += prog->aux->func[i]->jited_len;
2619         } else {
2620                 info.jited_prog_len = prog->jited_len;
2621         }
2622
2623         if (info.jited_prog_len && ulen) {
2624                 if (bpf_dump_raw_ok()) {
2625                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
2626                         ulen = min_t(u32, info.jited_prog_len, ulen);
2627
2628                         /* for multi-function programs, copy the JITed
2629                          * instructions for all the functions
2630                          */
2631                         if (prog->aux->func_cnt) {
2632                                 u32 len, free, i;
2633                                 u8 *img;
2634
2635                                 free = ulen;
2636                                 for (i = 0; i < prog->aux->func_cnt; i++) {
2637                                         len = prog->aux->func[i]->jited_len;
2638                                         len = min_t(u32, len, free);
2639                                         img = (u8 *) prog->aux->func[i]->bpf_func;
2640                                         if (copy_to_user(uinsns, img, len))
2641                                                 return -EFAULT;
2642                                         uinsns += len;
2643                                         free -= len;
2644                                         if (!free)
2645                                                 break;
2646                                 }
2647                         } else {
2648                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
2649                                         return -EFAULT;
2650                         }
2651                 } else {
2652                         info.jited_prog_insns = 0;
2653                 }
2654         }
2655
2656         ulen = info.nr_jited_ksyms;
2657         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
2658         if (ulen) {
2659                 if (bpf_dump_raw_ok()) {
2660                         unsigned long ksym_addr;
2661                         u64 __user *user_ksyms;
2662                         u32 i;
2663
2664                         /* copy the address of the kernel symbol
2665                          * corresponding to each function
2666                          */
2667                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
2668                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
2669                         if (prog->aux->func_cnt) {
2670                                 for (i = 0; i < ulen; i++) {
2671                                         ksym_addr = (unsigned long)
2672                                                 prog->aux->func[i]->bpf_func;
2673                                         if (put_user((u64) ksym_addr,
2674                                                      &user_ksyms[i]))
2675                                                 return -EFAULT;
2676                                 }
2677                         } else {
2678                                 ksym_addr = (unsigned long) prog->bpf_func;
2679                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
2680                                         return -EFAULT;
2681                         }
2682                 } else {
2683                         info.jited_ksyms = 0;
2684                 }
2685         }
2686
2687         ulen = info.nr_jited_func_lens;
2688         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
2689         if (ulen) {
2690                 if (bpf_dump_raw_ok()) {
2691                         u32 __user *user_lens;
2692                         u32 func_len, i;
2693
2694                         /* copy the JITed image lengths for each function */
2695                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
2696                         user_lens = u64_to_user_ptr(info.jited_func_lens);
2697                         if (prog->aux->func_cnt) {
2698                                 for (i = 0; i < ulen; i++) {
2699                                         func_len =
2700                                                 prog->aux->func[i]->jited_len;
2701                                         if (put_user(func_len, &user_lens[i]))
2702                                                 return -EFAULT;
2703                                 }
2704                         } else {
2705                                 func_len = prog->jited_len;
2706                                 if (put_user(func_len, &user_lens[0]))
2707                                         return -EFAULT;
2708                         }
2709                 } else {
2710                         info.jited_func_lens = 0;
2711                 }
2712         }
2713
2714         if (prog->aux->btf)
2715                 info.btf_id = btf_id(prog->aux->btf);
2716
2717         ulen = info.nr_func_info;
2718         info.nr_func_info = prog->aux->func_info_cnt;
2719         if (info.nr_func_info && ulen) {
2720                 char __user *user_finfo;
2721
2722                 user_finfo = u64_to_user_ptr(info.func_info);
2723                 ulen = min_t(u32, info.nr_func_info, ulen);
2724                 if (copy_to_user(user_finfo, prog->aux->func_info,
2725                                  info.func_info_rec_size * ulen))
2726                         return -EFAULT;
2727         }
2728
2729         ulen = info.nr_line_info;
2730         info.nr_line_info = prog->aux->nr_linfo;
2731         if (info.nr_line_info && ulen) {
2732                 __u8 __user *user_linfo;
2733
2734                 user_linfo = u64_to_user_ptr(info.line_info);
2735                 ulen = min_t(u32, info.nr_line_info, ulen);
2736                 if (copy_to_user(user_linfo, prog->aux->linfo,
2737                                  info.line_info_rec_size * ulen))
2738                         return -EFAULT;
2739         }
2740
2741         ulen = info.nr_jited_line_info;
2742         if (prog->aux->jited_linfo)
2743                 info.nr_jited_line_info = prog->aux->nr_linfo;
2744         else
2745                 info.nr_jited_line_info = 0;
2746         if (info.nr_jited_line_info && ulen) {
2747                 if (bpf_dump_raw_ok()) {
2748                         __u64 __user *user_linfo;
2749                         u32 i;
2750
2751                         user_linfo = u64_to_user_ptr(info.jited_line_info);
2752                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
2753                         for (i = 0; i < ulen; i++) {
2754                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
2755                                              &user_linfo[i]))
2756                                         return -EFAULT;
2757                         }
2758                 } else {
2759                         info.jited_line_info = 0;
2760                 }
2761         }
2762
2763         ulen = info.nr_prog_tags;
2764         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
2765         if (ulen) {
2766                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
2767                 u32 i;
2768
2769                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
2770                 ulen = min_t(u32, info.nr_prog_tags, ulen);
2771                 if (prog->aux->func_cnt) {
2772                         for (i = 0; i < ulen; i++) {
2773                                 if (copy_to_user(user_prog_tags[i],
2774                                                  prog->aux->func[i]->tag,
2775                                                  BPF_TAG_SIZE))
2776                                         return -EFAULT;
2777                         }
2778                 } else {
2779                         if (copy_to_user(user_prog_tags[0],
2780                                          prog->tag, BPF_TAG_SIZE))
2781                                 return -EFAULT;
2782                 }
2783         }
2784
2785 done:
2786         if (copy_to_user(uinfo, &info, info_len) ||
2787             put_user(info_len, &uattr->info.info_len))
2788                 return -EFAULT;
2789
2790         return 0;
2791 }
2792
2793 static int bpf_map_get_info_by_fd(struct bpf_map *map,
2794                                   const union bpf_attr *attr,
2795                                   union bpf_attr __user *uattr)
2796 {
2797         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2798         struct bpf_map_info info = {};
2799         u32 info_len = attr->info.info_len;
2800         int err;
2801
2802         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
2803         if (err)
2804                 return err;
2805         info_len = min_t(u32, sizeof(info), info_len);
2806
2807         info.type = map->map_type;
2808         info.id = map->id;
2809         info.key_size = map->key_size;
2810         info.value_size = map->value_size;
2811         info.max_entries = map->max_entries;
2812         info.map_flags = map->map_flags;
2813         memcpy(info.name, map->name, sizeof(map->name));
2814
2815         if (map->btf) {
2816                 info.btf_id = btf_id(map->btf);
2817                 info.btf_key_type_id = map->btf_key_type_id;
2818                 info.btf_value_type_id = map->btf_value_type_id;
2819         }
2820
2821         if (bpf_map_is_dev_bound(map)) {
2822                 err = bpf_map_offload_info_fill(&info, map);
2823                 if (err)
2824                         return err;
2825         }
2826
2827         if (copy_to_user(uinfo, &info, info_len) ||
2828             put_user(info_len, &uattr->info.info_len))
2829                 return -EFAULT;
2830
2831         return 0;
2832 }
2833
2834 static int bpf_btf_get_info_by_fd(struct btf *btf,
2835                                   const union bpf_attr *attr,
2836                                   union bpf_attr __user *uattr)
2837 {
2838         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
2839         u32 info_len = attr->info.info_len;
2840         int err;
2841
2842         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
2843         if (err)
2844                 return err;
2845
2846         return btf_get_info_by_fd(btf, attr, uattr);
2847 }
2848
2849 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
2850
2851 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
2852                                   union bpf_attr __user *uattr)
2853 {
2854         int ufd = attr->info.bpf_fd;
2855         struct fd f;
2856         int err;
2857
2858         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
2859                 return -EINVAL;
2860
2861         f = fdget(ufd);
2862         if (!f.file)
2863                 return -EBADFD;
2864
2865         if (f.file->f_op == &bpf_prog_fops)
2866                 err = bpf_prog_get_info_by_fd(f.file->private_data, attr,
2867                                               uattr);
2868         else if (f.file->f_op == &bpf_map_fops)
2869                 err = bpf_map_get_info_by_fd(f.file->private_data, attr,
2870                                              uattr);
2871         else if (f.file->f_op == &btf_fops)
2872                 err = bpf_btf_get_info_by_fd(f.file->private_data, attr, uattr);
2873         else
2874                 err = -EINVAL;
2875
2876         fdput(f);
2877         return err;
2878 }
2879
2880 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
2881
2882 static int bpf_btf_load(const union bpf_attr *attr)
2883 {
2884         if (CHECK_ATTR(BPF_BTF_LOAD))
2885                 return -EINVAL;
2886
2887         if (!capable(CAP_SYS_ADMIN))
2888                 return -EPERM;
2889
2890         return btf_new_fd(attr);
2891 }
2892
2893 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
2894
2895 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
2896 {
2897         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
2898                 return -EINVAL;
2899
2900         if (!capable(CAP_SYS_ADMIN))
2901                 return -EPERM;
2902
2903         return btf_get_fd_by_id(attr->btf_id);
2904 }
2905
2906 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
2907                                     union bpf_attr __user *uattr,
2908                                     u32 prog_id, u32 fd_type,
2909                                     const char *buf, u64 probe_offset,
2910                                     u64 probe_addr)
2911 {
2912         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
2913         u32 len = buf ? strlen(buf) : 0, input_len;
2914         int err = 0;
2915
2916         if (put_user(len, &uattr->task_fd_query.buf_len))
2917                 return -EFAULT;
2918         input_len = attr->task_fd_query.buf_len;
2919         if (input_len && ubuf) {
2920                 if (!len) {
2921                         /* nothing to copy, just make ubuf NULL terminated */
2922                         char zero = '\0';
2923
2924                         if (put_user(zero, ubuf))
2925                                 return -EFAULT;
2926                 } else if (input_len >= len + 1) {
2927                         /* ubuf can hold the string with NULL terminator */
2928                         if (copy_to_user(ubuf, buf, len + 1))
2929                                 return -EFAULT;
2930                 } else {
2931                         /* ubuf cannot hold the string with NULL terminator,
2932                          * do a partial copy with NULL terminator.
2933                          */
2934                         char zero = '\0';
2935
2936                         err = -ENOSPC;
2937                         if (copy_to_user(ubuf, buf, input_len - 1))
2938                                 return -EFAULT;
2939                         if (put_user(zero, ubuf + input_len - 1))
2940                                 return -EFAULT;
2941                 }
2942         }
2943
2944         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
2945             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
2946             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
2947             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
2948                 return -EFAULT;
2949
2950         return err;
2951 }
2952
2953 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
2954
2955 static int bpf_task_fd_query(const union bpf_attr *attr,
2956                              union bpf_attr __user *uattr)
2957 {
2958         pid_t pid = attr->task_fd_query.pid;
2959         u32 fd = attr->task_fd_query.fd;
2960         const struct perf_event *event;
2961         struct files_struct *files;
2962         struct task_struct *task;
2963         struct file *file;
2964         int err;
2965
2966         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
2967                 return -EINVAL;
2968
2969         if (!capable(CAP_SYS_ADMIN))
2970                 return -EPERM;
2971
2972         if (attr->task_fd_query.flags != 0)
2973                 return -EINVAL;
2974
2975         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
2976         if (!task)
2977                 return -ENOENT;
2978
2979         files = get_files_struct(task);
2980         put_task_struct(task);
2981         if (!files)
2982                 return -ENOENT;
2983
2984         err = 0;
2985         spin_lock(&files->file_lock);
2986         file = fcheck_files(files, fd);
2987         if (!file)
2988                 err = -EBADF;
2989         else
2990                 get_file(file);
2991         spin_unlock(&files->file_lock);
2992         put_files_struct(files);
2993
2994         if (err)
2995                 goto out;
2996
2997         if (file->f_op == &bpf_raw_tp_fops) {
2998                 struct bpf_raw_tracepoint *raw_tp = file->private_data;
2999                 struct bpf_raw_event_map *btp = raw_tp->btp;
3000
3001                 err = bpf_task_fd_query_copy(attr, uattr,
3002                                              raw_tp->prog->aux->id,
3003                                              BPF_FD_TYPE_RAW_TRACEPOINT,
3004                                              btp->tp->name, 0, 0);
3005                 goto put_file;
3006         }
3007
3008         event = perf_get_event(file);
3009         if (!IS_ERR(event)) {
3010                 u64 probe_offset, probe_addr;
3011                 u32 prog_id, fd_type;
3012                 const char *buf;
3013
3014                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3015                                               &buf, &probe_offset,
3016                                               &probe_addr);
3017                 if (!err)
3018                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3019                                                      fd_type, buf,
3020                                                      probe_offset,
3021                                                      probe_addr);
3022                 goto put_file;
3023         }
3024
3025         err = -ENOTSUPP;
3026 put_file:
3027         fput(file);
3028 out:
3029         return err;
3030 }
3031
3032 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
3033 {
3034         union bpf_attr attr = {};
3035         int err;
3036
3037         if (sysctl_unprivileged_bpf_disabled && !capable(CAP_SYS_ADMIN))
3038                 return -EPERM;
3039
3040         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
3041         if (err)
3042                 return err;
3043         size = min_t(u32, size, sizeof(attr));
3044
3045         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
3046         if (copy_from_user(&attr, uattr, size) != 0)
3047                 return -EFAULT;
3048
3049         err = security_bpf(cmd, &attr, size);
3050         if (err < 0)
3051                 return err;
3052
3053         switch (cmd) {
3054         case BPF_MAP_CREATE:
3055                 err = map_create(&attr);
3056                 break;
3057         case BPF_MAP_LOOKUP_ELEM:
3058                 err = map_lookup_elem(&attr);
3059                 break;
3060         case BPF_MAP_UPDATE_ELEM:
3061                 err = map_update_elem(&attr);
3062                 break;
3063         case BPF_MAP_DELETE_ELEM:
3064                 err = map_delete_elem(&attr);
3065                 break;
3066         case BPF_MAP_GET_NEXT_KEY:
3067                 err = map_get_next_key(&attr);
3068                 break;
3069         case BPF_MAP_FREEZE:
3070                 err = map_freeze(&attr);
3071                 break;
3072         case BPF_PROG_LOAD:
3073                 err = bpf_prog_load(&attr, uattr);
3074                 break;
3075         case BPF_OBJ_PIN:
3076                 err = bpf_obj_pin(&attr);
3077                 break;
3078         case BPF_OBJ_GET:
3079                 err = bpf_obj_get(&attr);
3080                 break;
3081         case BPF_PROG_ATTACH:
3082                 err = bpf_prog_attach(&attr);
3083                 break;
3084         case BPF_PROG_DETACH:
3085                 err = bpf_prog_detach(&attr);
3086                 break;
3087         case BPF_PROG_QUERY:
3088                 err = bpf_prog_query(&attr, uattr);
3089                 break;
3090         case BPF_PROG_TEST_RUN:
3091                 err = bpf_prog_test_run(&attr, uattr);
3092                 break;
3093         case BPF_PROG_GET_NEXT_ID:
3094                 err = bpf_obj_get_next_id(&attr, uattr,
3095                                           &prog_idr, &prog_idr_lock);
3096                 break;
3097         case BPF_MAP_GET_NEXT_ID:
3098                 err = bpf_obj_get_next_id(&attr, uattr,
3099                                           &map_idr, &map_idr_lock);
3100                 break;
3101         case BPF_BTF_GET_NEXT_ID:
3102                 err = bpf_obj_get_next_id(&attr, uattr,
3103                                           &btf_idr, &btf_idr_lock);
3104                 break;
3105         case BPF_PROG_GET_FD_BY_ID:
3106                 err = bpf_prog_get_fd_by_id(&attr);
3107                 break;
3108         case BPF_MAP_GET_FD_BY_ID:
3109                 err = bpf_map_get_fd_by_id(&attr);
3110                 break;
3111         case BPF_OBJ_GET_INFO_BY_FD:
3112                 err = bpf_obj_get_info_by_fd(&attr, uattr);
3113                 break;
3114         case BPF_RAW_TRACEPOINT_OPEN:
3115                 err = bpf_raw_tracepoint_open(&attr);
3116                 break;
3117         case BPF_BTF_LOAD:
3118                 err = bpf_btf_load(&attr);
3119                 break;
3120         case BPF_BTF_GET_FD_BY_ID:
3121                 err = bpf_btf_get_fd_by_id(&attr);
3122                 break;
3123         case BPF_TASK_FD_QUERY:
3124                 err = bpf_task_fd_query(&attr, uattr);
3125                 break;
3126         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
3127                 err = map_lookup_and_delete_elem(&attr);
3128                 break;
3129         default:
3130                 err = -EINVAL;
3131                 break;
3132         }
3133
3134         return err;
3135 }