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/bpf_verifier.h>
   8 #include <linux/btf.h>
   9 #include <linux/syscalls.h>
  10 #include <linux/slab.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/vmalloc.h>
  13 #include <linux/mmzone.h>
  14 #include <linux/anon_inodes.h>
  15 #include <linux/fdtable.h>
  16 #include <linux/file.h>
  17 #include <linux/fs.h>
  18 #include <linux/license.h>
  19 #include <linux/filter.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 #include <linux/pgtable.h>
  29 #include <linux/bpf_lsm.h>
  30 #include <linux/poll.h>
  31 #include <linux/bpf-netns.h>
  32 #include <linux/rcupdate_trace.h>
  33 #include <linux/memcontrol.h>
  34
  35 #define IS_FD_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PERF_EVENT_ARRAY || \
  36                           (map)->map_type == BPF_MAP_TYPE_CGROUP_ARRAY || \
  37                           (map)->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
  38 #define IS_FD_PROG_ARRAY(map) ((map)->map_type == BPF_MAP_TYPE_PROG_ARRAY)
  39 #define IS_FD_HASH(map) ((map)->map_type == BPF_MAP_TYPE_HASH_OF_MAPS)
  40 #define IS_FD_MAP(map) (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map) || \
  41                         IS_FD_HASH(map))
  42
  43 #define BPF_OBJ_FLAG_MASK   (BPF_F_RDONLY | BPF_F_WRONLY)
  44
  45 DEFINE_PER_CPU(int, bpf_prog_active);
  46 static DEFINE_IDR(prog_idr);
  47 static DEFINE_SPINLOCK(prog_idr_lock);
  48 static DEFINE_IDR(map_idr);
  49 static DEFINE_SPINLOCK(map_idr_lock);
  50 static DEFINE_IDR(link_idr);
  51 static DEFINE_SPINLOCK(link_idr_lock);
  52
  53 int sysctl_unprivileged_bpf_disabled __read_mostly;
  54
  55 static const struct bpf_map_ops * const bpf_map_types[] = {
  56 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
  57 #define BPF_MAP_TYPE(_id, _ops) \
  58         [_id] = &_ops,
  59 #define BPF_LINK_TYPE(_id, _name)
  60 #include <linux/bpf_types.h>
  61 #undef BPF_PROG_TYPE
  62 #undef BPF_MAP_TYPE
  63 #undef BPF_LINK_TYPE
  64 };
  65
  66 /*
  67  * If we're handed a bigger struct than we know of, ensure all the unknown bits
  68  * are 0 - i.e. new user-space does not rely on any kernel feature extensions
  69  * we don't know about yet.
  70  *
  71  * There is a ToCToU between this function call and the following
  72  * copy_from_user() call. However, this is not a concern since this function is
  73  * meant to be a future-proofing of bits.
  74  */
  75 int bpf_check_uarg_tail_zero(void __user *uaddr,
  76                              size_t expected_size,
  77                              size_t actual_size)
  78 {
  79         unsigned char __user *addr = uaddr + expected_size;
  80         int res;
  81
  82         if (unlikely(actual_size > PAGE_SIZE))  /* silly large */
  83                 return -E2BIG;
  84
  85         if (actual_size <= expected_size)
  86                 return 0;
  87
  88         res = check_zeroed_user(addr, actual_size - expected_size);
  89         if (res < 0)
  90                 return res;
  91         return res ? 0 : -E2BIG;
  92 }
  93
  94 const struct bpf_map_ops bpf_map_offload_ops = {
  95         .map_meta_equal = bpf_map_meta_equal,
  96         .map_alloc = bpf_map_offload_map_alloc,
  97         .map_free = bpf_map_offload_map_free,
  98         .map_check_btf = map_check_no_btf,
  99 };
 100
 101 static struct bpf_map *find_and_alloc_map(union bpf_attr *attr)
 102 {
 103         const struct bpf_map_ops *ops;
 104         u32 type = attr->map_type;
 105         struct bpf_map *map;
 106         int err;
 107
 108         if (type >= ARRAY_SIZE(bpf_map_types))
 109                 return ERR_PTR(-EINVAL);
 110         type = array_index_nospec(type, ARRAY_SIZE(bpf_map_types));
 111         ops = bpf_map_types[type];
 112         if (!ops)
 113                 return ERR_PTR(-EINVAL);
 114
 115         if (ops->map_alloc_check) {
 116                 err = ops->map_alloc_check(attr);
 117                 if (err)
 118                         return ERR_PTR(err);
 119         }
 120         if (attr->map_ifindex)
 121                 ops = &bpf_map_offload_ops;
 122         map = ops->map_alloc(attr);
 123         if (IS_ERR(map))
 124                 return map;
 125         map->ops = ops;
 126         map->map_type = type;
 127         return map;
 128 }
 129
 130 static u32 bpf_map_value_size(const struct bpf_map *map)
 131 {
 132         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 133             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
 134             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
 135             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
 136                 return round_up(map->value_size, 8) * num_possible_cpus();
 137         else if (IS_FD_MAP(map))
 138                 return sizeof(u32);
 139         else
 140                 return  map->value_size;
 141 }
 142
 143 static void maybe_wait_bpf_programs(struct bpf_map *map)
 144 {
 145         /* Wait for any running BPF programs to complete so that
 146          * userspace, when we return to it, knows that all programs
 147          * that could be running use the new map value.
 148          */
 149         if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS ||
 150             map->map_type == BPF_MAP_TYPE_ARRAY_OF_MAPS)
 151                 synchronize_rcu();
 152 }
 153
 154 static int bpf_map_update_value(struct bpf_map *map, struct fd f, void *key,
 155                                 void *value, __u64 flags)
 156 {
 157         int err;
 158
 159         /* Need to create a kthread, thus must support schedule */
 160         if (bpf_map_is_dev_bound(map)) {
 161                 return bpf_map_offload_update_elem(map, key, value, flags);
 162         } else if (map->map_type == BPF_MAP_TYPE_CPUMAP ||
 163                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 164                 return map->ops->map_update_elem(map, key, value, flags);
 165         } else if (map->map_type == BPF_MAP_TYPE_SOCKHASH ||
 166                    map->map_type == BPF_MAP_TYPE_SOCKMAP) {
 167                 return sock_map_update_elem_sys(map, key, value, flags);
 168         } else if (IS_FD_PROG_ARRAY(map)) {
 169                 return bpf_fd_array_map_update_elem(map, f.file, key, value,
 170                                                     flags);
 171         }
 172
 173         bpf_disable_instrumentation();
 174         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 175             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 176                 err = bpf_percpu_hash_update(map, key, value, flags);
 177         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 178                 err = bpf_percpu_array_update(map, key, value, flags);
 179         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 180                 err = bpf_percpu_cgroup_storage_update(map, key, value,
 181                                                        flags);
 182         } else if (IS_FD_ARRAY(map)) {
 183                 rcu_read_lock();
 184                 err = bpf_fd_array_map_update_elem(map, f.file, key, value,
 185                                                    flags);
 186                 rcu_read_unlock();
 187         } else if (map->map_type == BPF_MAP_TYPE_HASH_OF_MAPS) {
 188                 rcu_read_lock();
 189                 err = bpf_fd_htab_map_update_elem(map, f.file, key, value,
 190                                                   flags);
 191                 rcu_read_unlock();
 192         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 193                 /* rcu_read_lock() is not needed */
 194                 err = bpf_fd_reuseport_array_update_elem(map, key, value,
 195                                                          flags);
 196         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 197                    map->map_type == BPF_MAP_TYPE_STACK) {
 198                 err = map->ops->map_push_elem(map, value, flags);
 199         } else {
 200                 rcu_read_lock();
 201                 err = map->ops->map_update_elem(map, key, value, flags);
 202                 rcu_read_unlock();
 203         }
 204         bpf_enable_instrumentation();
 205         maybe_wait_bpf_programs(map);
 206
 207         return err;
 208 }
 209
 210 static int bpf_map_copy_value(struct bpf_map *map, void *key, void *value,
 211                               __u64 flags)
 212 {
 213         void *ptr;
 214         int err;
 215
 216         if (bpf_map_is_dev_bound(map))
 217                 return bpf_map_offload_lookup_elem(map, key, value);
 218
 219         bpf_disable_instrumentation();
 220         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 221             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 222                 err = bpf_percpu_hash_copy(map, key, value);
 223         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY) {
 224                 err = bpf_percpu_array_copy(map, key, value);
 225         } else if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) {
 226                 err = bpf_percpu_cgroup_storage_copy(map, key, value);
 227         } else if (map->map_type == BPF_MAP_TYPE_STACK_TRACE) {
 228                 err = bpf_stackmap_copy(map, key, value);
 229         } else if (IS_FD_ARRAY(map) || IS_FD_PROG_ARRAY(map)) {
 230                 err = bpf_fd_array_map_lookup_elem(map, key, value);
 231         } else if (IS_FD_HASH(map)) {
 232                 err = bpf_fd_htab_map_lookup_elem(map, key, value);
 233         } else if (map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) {
 234                 err = bpf_fd_reuseport_array_lookup_elem(map, key, value);
 235         } else if (map->map_type == BPF_MAP_TYPE_QUEUE ||
 236                    map->map_type == BPF_MAP_TYPE_STACK) {
 237                 err = map->ops->map_peek_elem(map, value);
 238         } else if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
 239                 /* struct_ops map requires directly updating "value" */
 240                 err = bpf_struct_ops_map_sys_lookup_elem(map, key, value);
 241         } else {
 242                 rcu_read_lock();
 243                 if (map->ops->map_lookup_elem_sys_only)
 244                         ptr = map->ops->map_lookup_elem_sys_only(map, key);
 245                 else
 246                         ptr = map->ops->map_lookup_elem(map, key);
 247                 if (IS_ERR(ptr)) {
 248                         err = PTR_ERR(ptr);
 249                 } else if (!ptr) {
 250                         err = -ENOENT;
 251                 } else {
 252                         err = 0;
 253                         if (flags & BPF_F_LOCK)
 254                                 /* lock 'ptr' and copy everything but lock */
 255                                 copy_map_value_locked(map, value, ptr, true);
 256                         else
 257                                 copy_map_value(map, value, ptr);
 258                         /* mask lock, since value wasn't zero inited */
 259                         check_and_init_map_lock(map, value);
 260                 }
 261                 rcu_read_unlock();
 262         }
 263
 264         bpf_enable_instrumentation();
 265         maybe_wait_bpf_programs(map);
 266
 267         return err;
 268 }
 269
 270 /* Please, do not use this function outside from the map creation path
 271  * (e.g. in map update path) without taking care of setting the active
 272  * memory cgroup (see at bpf_map_kmalloc_node() for example).
 273  */
 274 static void *__bpf_map_area_alloc(u64 size, int numa_node, bool mmapable)
 275 {
 276         /* We really just want to fail instead of triggering OOM killer
 277          * under memory pressure, therefore we set __GFP_NORETRY to kmalloc,
 278          * which is used for lower order allocation requests.
 279          *
 280          * It has been observed that higher order allocation requests done by
 281          * vmalloc with __GFP_NORETRY being set might fail due to not trying
 282          * to reclaim memory from the page cache, thus we set
 283          * __GFP_RETRY_MAYFAIL to avoid such situations.
 284          */
 285
 286         const gfp_t gfp = __GFP_NOWARN | __GFP_ZERO | __GFP_ACCOUNT;
 287         unsigned int flags = 0;
 288         unsigned long align = 1;
 289         void *area;
 290
 291         if (size >= SIZE_MAX)
 292                 return NULL;
 293
 294         /* kmalloc()'ed memory can't be mmap()'ed */
 295         if (mmapable) {
 296                 BUG_ON(!PAGE_ALIGNED(size));
 297                 align = SHMLBA;
 298                 flags = VM_USERMAP;
 299         } else if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
 300                 area = kmalloc_node(size, gfp | GFP_USER | __GFP_NORETRY,
 301                                     numa_node);
 302                 if (area != NULL)
 303                         return area;
 304         }
 305
 306         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
 307                         gfp | GFP_KERNEL | __GFP_RETRY_MAYFAIL, PAGE_KERNEL,
 308                         flags, numa_node, __builtin_return_address(0));
 309 }
 310
 311 void *bpf_map_area_alloc(u64 size, int numa_node)
 312 {
 313         return __bpf_map_area_alloc(size, numa_node, false);
 314 }
 315
 316 void *bpf_map_area_mmapable_alloc(u64 size, int numa_node)
 317 {
 318         return __bpf_map_area_alloc(size, numa_node, true);
 319 }
 320
 321 void bpf_map_area_free(void *area)
 322 {
 323         kvfree(area);
 324 }
 325
 326 static u32 bpf_map_flags_retain_permanent(u32 flags)
 327 {
 328         /* Some map creation flags are not tied to the map object but
 329          * rather to the map fd instead, so they have no meaning upon
 330          * map object inspection since multiple file descriptors with
 331          * different (access) properties can exist here. Thus, given
 332          * this has zero meaning for the map itself, lets clear these
 333          * from here.
 334          */
 335         return flags & ~(BPF_F_RDONLY | BPF_F_WRONLY);
 336 }
 337
 338 void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr)
 339 {
 340         map->map_type = attr->map_type;
 341         map->key_size = attr->key_size;
 342         map->value_size = attr->value_size;
 343         map->max_entries = attr->max_entries;
 344         map->map_flags = bpf_map_flags_retain_permanent(attr->map_flags);
 345         map->numa_node = bpf_map_attr_numa_node(attr);
 346 }
 347
 348 static int bpf_map_alloc_id(struct bpf_map *map)
 349 {
 350         int id;
 351
 352         idr_preload(GFP_KERNEL);
 353         spin_lock_bh(&map_idr_lock);
 354         id = idr_alloc_cyclic(&map_idr, map, 1, INT_MAX, GFP_ATOMIC);
 355         if (id > 0)
 356                 map->id = id;
 357         spin_unlock_bh(&map_idr_lock);
 358         idr_preload_end();
 359
 360         if (WARN_ON_ONCE(!id))
 361                 return -ENOSPC;
 362
 363         return id > 0 ? 0 : id;
 364 }
 365
 366 void bpf_map_free_id(struct bpf_map *map, bool do_idr_lock)
 367 {
 368         unsigned long flags;
 369
 370         /* Offloaded maps are removed from the IDR store when their device
 371          * disappears - even if someone holds an fd to them they are unusable,
 372          * the memory is gone, all ops will fail; they are simply waiting for
 373          * refcnt to drop to be freed.
 374          */
 375         if (!map->id)
 376                 return;
 377
 378         if (do_idr_lock)
 379                 spin_lock_irqsave(&map_idr_lock, flags);
 380         else
 381                 __acquire(&map_idr_lock);
 382
 383         idr_remove(&map_idr, map->id);
 384         map->id = 0;
 385
 386         if (do_idr_lock)
 387                 spin_unlock_irqrestore(&map_idr_lock, flags);
 388         else
 389                 __release(&map_idr_lock);
 390 }
 391
 392 #ifdef CONFIG_MEMCG_KMEM
 393 static void bpf_map_save_memcg(struct bpf_map *map)
 394 {
 395         map->memcg = get_mem_cgroup_from_mm(current->mm);
 396 }
 397
 398 static void bpf_map_release_memcg(struct bpf_map *map)
 399 {
 400         mem_cgroup_put(map->memcg);
 401 }
 402
 403 void *bpf_map_kmalloc_node(const struct bpf_map *map, size_t size, gfp_t flags,
 404                            int node)
 405 {
 406         struct mem_cgroup *old_memcg;
 407         void *ptr;
 408
 409         old_memcg = set_active_memcg(map->memcg);
 410         ptr = kmalloc_node(size, flags | __GFP_ACCOUNT, node);
 411         set_active_memcg(old_memcg);
 412
 413         return ptr;
 414 }
 415
 416 void *bpf_map_kzalloc(const struct bpf_map *map, size_t size, gfp_t flags)
 417 {
 418         struct mem_cgroup *old_memcg;
 419         void *ptr;
 420
 421         old_memcg = set_active_memcg(map->memcg);
 422         ptr = kzalloc(size, flags | __GFP_ACCOUNT);
 423         set_active_memcg(old_memcg);
 424
 425         return ptr;
 426 }
 427
 428 void __percpu *bpf_map_alloc_percpu(const struct bpf_map *map, size_t size,
 429                                     size_t align, gfp_t flags)
 430 {
 431         struct mem_cgroup *old_memcg;
 432         void __percpu *ptr;
 433
 434         old_memcg = set_active_memcg(map->memcg);
 435         ptr = __alloc_percpu_gfp(size, align, flags | __GFP_ACCOUNT);
 436         set_active_memcg(old_memcg);
 437
 438         return ptr;
 439 }
 440
 441 #else
 442 static void bpf_map_save_memcg(struct bpf_map *map)
 443 {
 444 }
 445
 446 static void bpf_map_release_memcg(struct bpf_map *map)
 447 {
 448 }
 449 #endif
 450
 451 /* called from workqueue */
 452 static void bpf_map_free_deferred(struct work_struct *work)
 453 {
 454         struct bpf_map *map = container_of(work, struct bpf_map, work);
 455
 456         security_bpf_map_free(map);
 457         bpf_map_release_memcg(map);
 458         /* implementation dependent freeing */
 459         map->ops->map_free(map);
 460 }
 461
 462 static void bpf_map_put_uref(struct bpf_map *map)
 463 {
 464         if (atomic64_dec_and_test(&map->usercnt)) {
 465                 if (map->ops->map_release_uref)
 466                         map->ops->map_release_uref(map);
 467         }
 468 }
 469
 470 /* decrement map refcnt and schedule it for freeing via workqueue
 471  * (unrelying map implementation ops->map_free() might sleep)
 472  */
 473 static void __bpf_map_put(struct bpf_map *map, bool do_idr_lock)
 474 {
 475         if (atomic64_dec_and_test(&map->refcnt)) {
 476                 /* bpf_map_free_id() must be called first */
 477                 bpf_map_free_id(map, do_idr_lock);
 478                 btf_put(map->btf);
 479                 INIT_WORK(&map->work, bpf_map_free_deferred);
 480                 schedule_work(&map->work);
 481         }
 482 }
 483
 484 void bpf_map_put(struct bpf_map *map)
 485 {
 486         __bpf_map_put(map, true);
 487 }
 488 EXPORT_SYMBOL_GPL(bpf_map_put);
 489
 490 void bpf_map_put_with_uref(struct bpf_map *map)
 491 {
 492         bpf_map_put_uref(map);
 493         bpf_map_put(map);
 494 }
 495
 496 static int bpf_map_release(struct inode *inode, struct file *filp)
 497 {
 498         struct bpf_map *map = filp->private_data;
 499
 500         if (map->ops->map_release)
 501                 map->ops->map_release(map, filp);
 502
 503         bpf_map_put_with_uref(map);
 504         return 0;
 505 }
 506
 507 static fmode_t map_get_sys_perms(struct bpf_map *map, struct fd f)
 508 {
 509         fmode_t mode = f.file->f_mode;
 510
 511         /* Our file permissions may have been overridden by global
 512          * map permissions facing syscall side.
 513          */
 514         if (READ_ONCE(map->frozen))
 515                 mode &= ~FMODE_CAN_WRITE;
 516         return mode;
 517 }
 518
 519 #ifdef CONFIG_PROC_FS
 520 /* Provides an approximation of the map's memory footprint.
 521  * Used only to provide a backward compatibility and display
 522  * a reasonable "memlock" info.
 523  */
 524 static unsigned long bpf_map_memory_footprint(const struct bpf_map *map)
 525 {
 526         unsigned long size;
 527
 528         size = round_up(map->key_size + bpf_map_value_size(map), 8);
 529
 530         return round_up(map->max_entries * size, PAGE_SIZE);
 531 }
 532
 533 static void bpf_map_show_fdinfo(struct seq_file *m, struct file *filp)
 534 {
 535         const struct bpf_map *map = filp->private_data;
 536         const struct bpf_array *array;
 537         u32 type = 0, jited = 0;
 538
 539         if (map->map_type == BPF_MAP_TYPE_PROG_ARRAY) {
 540                 array = container_of(map, struct bpf_array, map);
 541                 type  = array->aux->type;
 542                 jited = array->aux->jited;
 543         }
 544
 545         seq_printf(m,
 546                    "map_type:\t%u\n"
 547                    "key_size:\t%u\n"
 548                    "value_size:\t%u\n"
 549                    "max_entries:\t%u\n"
 550                    "map_flags:\t%#x\n"
 551                    "memlock:\t%lu\n"
 552                    "map_id:\t%u\n"
 553                    "frozen:\t%u\n",
 554                    map->map_type,
 555                    map->key_size,
 556                    map->value_size,
 557                    map->max_entries,
 558                    map->map_flags,
 559                    bpf_map_memory_footprint(map),
 560                    map->id,
 561                    READ_ONCE(map->frozen));
 562         if (type) {
 563                 seq_printf(m, "owner_prog_type:\t%u\n", type);
 564                 seq_printf(m, "owner_jited:\t%u\n", jited);
 565         }
 566 }
 567 #endif
 568
 569 static ssize_t bpf_dummy_read(struct file *filp, char __user *buf, size_t siz,
 570                               loff_t *ppos)
 571 {
 572         /* We need this handler such that alloc_file() enables
 573          * f_mode with FMODE_CAN_READ.
 574          */
 575         return -EINVAL;
 576 }
 577
 578 static ssize_t bpf_dummy_write(struct file *filp, const char __user *buf,
 579                                size_t siz, loff_t *ppos)
 580 {
 581         /* We need this handler such that alloc_file() enables
 582          * f_mode with FMODE_CAN_WRITE.
 583          */
 584         return -EINVAL;
 585 }
 586
 587 /* called for any extra memory-mapped regions (except initial) */
 588 static void bpf_map_mmap_open(struct vm_area_struct *vma)
 589 {
 590         struct bpf_map *map = vma->vm_file->private_data;
 591
 592         if (vma->vm_flags & VM_MAYWRITE) {
 593                 mutex_lock(&map->freeze_mutex);
 594                 map->writecnt++;
 595                 mutex_unlock(&map->freeze_mutex);
 596         }
 597 }
 598
 599 /* called for all unmapped memory region (including initial) */
 600 static void bpf_map_mmap_close(struct vm_area_struct *vma)
 601 {
 602         struct bpf_map *map = vma->vm_file->private_data;
 603
 604         if (vma->vm_flags & VM_MAYWRITE) {
 605                 mutex_lock(&map->freeze_mutex);
 606                 map->writecnt--;
 607                 mutex_unlock(&map->freeze_mutex);
 608         }
 609 }
 610
 611 static const struct vm_operations_struct bpf_map_default_vmops = {
 612         .open           = bpf_map_mmap_open,
 613         .close          = bpf_map_mmap_close,
 614 };
 615
 616 static int bpf_map_mmap(struct file *filp, struct vm_area_struct *vma)
 617 {
 618         struct bpf_map *map = filp->private_data;
 619         int err;
 620
 621         if (!map->ops->map_mmap || map_value_has_spin_lock(map))
 622                 return -ENOTSUPP;
 623
 624         if (!(vma->vm_flags & VM_SHARED))
 625                 return -EINVAL;
 626
 627         mutex_lock(&map->freeze_mutex);
 628
 629         if (vma->vm_flags & VM_WRITE) {
 630                 if (map->frozen) {
 631                         err = -EPERM;
 632                         goto out;
 633                 }
 634                 /* map is meant to be read-only, so do not allow mapping as
 635                  * writable, because it's possible to leak a writable page
 636                  * reference and allows user-space to still modify it after
 637                  * freezing, while verifier will assume contents do not change
 638                  */
 639                 if (map->map_flags & BPF_F_RDONLY_PROG) {
 640                         err = -EACCES;
 641                         goto out;
 642                 }
 643         }
 644
 645         /* set default open/close callbacks */
 646         vma->vm_ops = &bpf_map_default_vmops;
 647         vma->vm_private_data = map;
 648         vma->vm_flags &= ~VM_MAYEXEC;
 649         if (!(vma->vm_flags & VM_WRITE))
 650                 /* disallow re-mapping with PROT_WRITE */
 651                 vma->vm_flags &= ~VM_MAYWRITE;
 652
 653         err = map->ops->map_mmap(map, vma);
 654         if (err)
 655                 goto out;
 656
 657         if (vma->vm_flags & VM_MAYWRITE)
 658                 map->writecnt++;
 659 out:
 660         mutex_unlock(&map->freeze_mutex);
 661         return err;
 662 }
 663
 664 static __poll_t bpf_map_poll(struct file *filp, struct poll_table_struct *pts)
 665 {
 666         struct bpf_map *map = filp->private_data;
 667
 668         if (map->ops->map_poll)
 669                 return map->ops->map_poll(map, filp, pts);
 670
 671         return EPOLLERR;
 672 }
 673
 674 const struct file_operations bpf_map_fops = {
 675 #ifdef CONFIG_PROC_FS
 676         .show_fdinfo    = bpf_map_show_fdinfo,
 677 #endif
 678         .release        = bpf_map_release,
 679         .read           = bpf_dummy_read,
 680         .write          = bpf_dummy_write,
 681         .mmap           = bpf_map_mmap,
 682         .poll           = bpf_map_poll,
 683 };
 684
 685 int bpf_map_new_fd(struct bpf_map *map, int flags)
 686 {
 687         int ret;
 688
 689         ret = security_bpf_map(map, OPEN_FMODE(flags));
 690         if (ret < 0)
 691                 return ret;
 692
 693         return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
 694                                 flags | O_CLOEXEC);
 695 }
 696
 697 int bpf_get_file_flag(int flags)
 698 {
 699         if ((flags & BPF_F_RDONLY) && (flags & BPF_F_WRONLY))
 700                 return -EINVAL;
 701         if (flags & BPF_F_RDONLY)
 702                 return O_RDONLY;
 703         if (flags & BPF_F_WRONLY)
 704                 return O_WRONLY;
 705         return O_RDWR;
 706 }
 707
 708 /* helper macro to check that unused fields 'union bpf_attr' are zero */
 709 #define CHECK_ATTR(CMD) \
 710         memchr_inv((void *) &attr->CMD##_LAST_FIELD + \
 711                    sizeof(attr->CMD##_LAST_FIELD), 0, \
 712                    sizeof(*attr) - \
 713                    offsetof(union bpf_attr, CMD##_LAST_FIELD) - \
 714                    sizeof(attr->CMD##_LAST_FIELD)) != NULL
 715
 716 /* dst and src must have at least "size" number of bytes.
 717  * Return strlen on success and < 0 on error.
 718  */
 719 int bpf_obj_name_cpy(char *dst, const char *src, unsigned int size)
 720 {
 721         const char *end = src + size;
 722         const char *orig_src = src;
 723
 724         memset(dst, 0, size);
 725         /* Copy all isalnum(), '_' and '.' chars. */
 726         while (src < end && *src) {
 727                 if (!isalnum(*src) &&
 728                     *src != '_' && *src != '.')
 729                         return -EINVAL;
 730                 *dst++ = *src++;
 731         }
 732
 733         /* No '\0' found in "size" number of bytes */
 734         if (src == end)
 735                 return -EINVAL;
 736
 737         return src - orig_src;
 738 }
 739
 740 int map_check_no_btf(const struct bpf_map *map,
 741                      const struct btf *btf,
 742                      const struct btf_type *key_type,
 743                      const struct btf_type *value_type)
 744 {
 745         return -ENOTSUPP;
 746 }
 747
 748 static int map_check_btf(struct bpf_map *map, const struct btf *btf,
 749                          u32 btf_key_id, u32 btf_value_id)
 750 {
 751         const struct btf_type *key_type, *value_type;
 752         u32 key_size, value_size;
 753         int ret = 0;
 754
 755         /* Some maps allow key to be unspecified. */
 756         if (btf_key_id) {
 757                 key_type = btf_type_id_size(btf, &btf_key_id, &key_size);
 758                 if (!key_type || key_size != map->key_size)
 759                         return -EINVAL;
 760         } else {
 761                 key_type = btf_type_by_id(btf, 0);
 762                 if (!map->ops->map_check_btf)
 763                         return -EINVAL;
 764         }
 765
 766         value_type = btf_type_id_size(btf, &btf_value_id, &value_size);
 767         if (!value_type || value_size != map->value_size)
 768                 return -EINVAL;
 769
 770         map->spin_lock_off = btf_find_spin_lock(btf, value_type);
 771
 772         if (map_value_has_spin_lock(map)) {
 773                 if (map->map_flags & BPF_F_RDONLY_PROG)
 774                         return -EACCES;
 775                 if (map->map_type != BPF_MAP_TYPE_HASH &&
 776                     map->map_type != BPF_MAP_TYPE_ARRAY &&
 777                     map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE &&
 778                     map->map_type != BPF_MAP_TYPE_SK_STORAGE &&
 779                     map->map_type != BPF_MAP_TYPE_INODE_STORAGE &&
 780                     map->map_type != BPF_MAP_TYPE_TASK_STORAGE)
 781                         return -ENOTSUPP;
 782                 if (map->spin_lock_off + sizeof(struct bpf_spin_lock) >
 783                     map->value_size) {
 784                         WARN_ONCE(1,
 785                                   "verifier bug spin_lock_off %d value_size %d\n",
 786                                   map->spin_lock_off, map->value_size);
 787                         return -EFAULT;
 788                 }
 789         }
 790
 791         if (map->ops->map_check_btf)
 792                 ret = map->ops->map_check_btf(map, btf, key_type, value_type);
 793
 794         return ret;
 795 }
 796
 797 #define BPF_MAP_CREATE_LAST_FIELD btf_vmlinux_value_type_id
 798 /* called via syscall */
 799 static int map_create(union bpf_attr *attr)
 800 {
 801         int numa_node = bpf_map_attr_numa_node(attr);
 802         struct bpf_map *map;
 803         int f_flags;
 804         int err;
 805
 806         err = CHECK_ATTR(BPF_MAP_CREATE);
 807         if (err)
 808                 return -EINVAL;
 809
 810         if (attr->btf_vmlinux_value_type_id) {
 811                 if (attr->map_type != BPF_MAP_TYPE_STRUCT_OPS ||
 812                     attr->btf_key_type_id || attr->btf_value_type_id)
 813                         return -EINVAL;
 814         } else if (attr->btf_key_type_id && !attr->btf_value_type_id) {
 815                 return -EINVAL;
 816         }
 817
 818         f_flags = bpf_get_file_flag(attr->map_flags);
 819         if (f_flags < 0)
 820                 return f_flags;
 821
 822         if (numa_node != NUMA_NO_NODE &&
 823             ((unsigned int)numa_node >= nr_node_ids ||
 824              !node_online(numa_node)))
 825                 return -EINVAL;
 826
 827         /* find map type and init map: hashtable vs rbtree vs bloom vs ... */
 828         map = find_and_alloc_map(attr);
 829         if (IS_ERR(map))
 830                 return PTR_ERR(map);
 831
 832         err = bpf_obj_name_cpy(map->name, attr->map_name,
 833                                sizeof(attr->map_name));
 834         if (err < 0)
 835                 goto free_map;
 836
 837         atomic64_set(&map->refcnt, 1);
 838         atomic64_set(&map->usercnt, 1);
 839         mutex_init(&map->freeze_mutex);
 840
 841         map->spin_lock_off = -EINVAL;
 842         if (attr->btf_key_type_id || attr->btf_value_type_id ||
 843             /* Even the map's value is a kernel's struct,
 844              * the bpf_prog.o must have BTF to begin with
 845              * to figure out the corresponding kernel's
 846              * counter part.  Thus, attr->btf_fd has
 847              * to be valid also.
 848              */
 849             attr->btf_vmlinux_value_type_id) {
 850                 struct btf *btf;
 851
 852                 btf = btf_get_by_fd(attr->btf_fd);
 853                 if (IS_ERR(btf)) {
 854                         err = PTR_ERR(btf);
 855                         goto free_map;
 856                 }
 857                 if (btf_is_kernel(btf)) {
 858                         btf_put(btf);
 859                         err = -EACCES;
 860                         goto free_map;
 861                 }
 862                 map->btf = btf;
 863
 864                 if (attr->btf_value_type_id) {
 865                         err = map_check_btf(map, btf, attr->btf_key_type_id,
 866                                             attr->btf_value_type_id);
 867                         if (err)
 868                                 goto free_map;
 869                 }
 870
 871                 map->btf_key_type_id = attr->btf_key_type_id;
 872                 map->btf_value_type_id = attr->btf_value_type_id;
 873                 map->btf_vmlinux_value_type_id =
 874                         attr->btf_vmlinux_value_type_id;
 875         }
 876
 877         err = security_bpf_map_alloc(map);
 878         if (err)
 879                 goto free_map;
 880
 881         err = bpf_map_alloc_id(map);
 882         if (err)
 883                 goto free_map_sec;
 884
 885         bpf_map_save_memcg(map);
 886
 887         err = bpf_map_new_fd(map, f_flags);
 888         if (err < 0) {
 889                 /* failed to allocate fd.
 890                  * bpf_map_put_with_uref() is needed because the above
 891                  * bpf_map_alloc_id() has published the map
 892                  * to the userspace and the userspace may
 893                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 894                  */
 895                 bpf_map_put_with_uref(map);
 896                 return err;
 897         }
 898
 899         return err;
 900
 901 free_map_sec:
 902         security_bpf_map_free(map);
 903 free_map:
 904         btf_put(map->btf);
 905         map->ops->map_free(map);
 906         return err;
 907 }
 908
 909 /* if error is returned, fd is released.
 910  * On success caller should complete fd access with matching fdput()
 911  */
 912 struct bpf_map *__bpf_map_get(struct fd f)
 913 {
 914         if (!f.file)
 915                 return ERR_PTR(-EBADF);
 916         if (f.file->f_op != &bpf_map_fops) {
 917                 fdput(f);
 918                 return ERR_PTR(-EINVAL);
 919         }
 920
 921         return f.file->private_data;
 922 }
 923
 924 void bpf_map_inc(struct bpf_map *map)
 925 {
 926         atomic64_inc(&map->refcnt);
 927 }
 928 EXPORT_SYMBOL_GPL(bpf_map_inc);
 929
 930 void bpf_map_inc_with_uref(struct bpf_map *map)
 931 {
 932         atomic64_inc(&map->refcnt);
 933         atomic64_inc(&map->usercnt);
 934 }
 935 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
 936
 937 struct bpf_map *bpf_map_get(u32 ufd)
 938 {
 939         struct fd f = fdget(ufd);
 940         struct bpf_map *map;
 941
 942         map = __bpf_map_get(f);
 943         if (IS_ERR(map))
 944                 return map;
 945
 946         bpf_map_inc(map);
 947         fdput(f);
 948
 949         return map;
 950 }
 951
 952 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 953 {
 954         struct fd f = fdget(ufd);
 955         struct bpf_map *map;
 956
 957         map = __bpf_map_get(f);
 958         if (IS_ERR(map))
 959                 return map;
 960
 961         bpf_map_inc_with_uref(map);
 962         fdput(f);
 963
 964         return map;
 965 }
 966
 967 /* map_idr_lock should have been held */
 968 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 969 {
 970         int refold;
 971
 972         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
 973         if (!refold)
 974                 return ERR_PTR(-ENOENT);
 975         if (uref)
 976                 atomic64_inc(&map->usercnt);
 977
 978         return map;
 979 }
 980
 981 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
 982 {
 983         spin_lock_bh(&map_idr_lock);
 984         map = __bpf_map_inc_not_zero(map, false);
 985         spin_unlock_bh(&map_idr_lock);
 986
 987         return map;
 988 }
 989 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 990
 991 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 992 {
 993         return -ENOTSUPP;
 994 }
 995
 996 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 997 {
 998         if (key_size)
 999                 return memdup_user(ukey, key_size);
1000
1001         if (ukey)
1002                 return ERR_PTR(-EINVAL);
1003
1004         return NULL;
1005 }
1006
1007 /* last field in 'union bpf_attr' used by this command */
1008 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1009
1010 static int map_lookup_elem(union bpf_attr *attr)
1011 {
1012         void __user *ukey = u64_to_user_ptr(attr->key);
1013         void __user *uvalue = u64_to_user_ptr(attr->value);
1014         int ufd = attr->map_fd;
1015         struct bpf_map *map;
1016         void *key, *value;
1017         u32 value_size;
1018         struct fd f;
1019         int err;
1020
1021         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1022                 return -EINVAL;
1023
1024         if (attr->flags & ~BPF_F_LOCK)
1025                 return -EINVAL;
1026
1027         f = fdget(ufd);
1028         map = __bpf_map_get(f);
1029         if (IS_ERR(map))
1030                 return PTR_ERR(map);
1031         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1032                 err = -EPERM;
1033                 goto err_put;
1034         }
1035
1036         if ((attr->flags & BPF_F_LOCK) &&
1037             !map_value_has_spin_lock(map)) {
1038                 err = -EINVAL;
1039                 goto err_put;
1040         }
1041
1042         key = __bpf_copy_key(ukey, map->key_size);
1043         if (IS_ERR(key)) {
1044                 err = PTR_ERR(key);
1045                 goto err_put;
1046         }
1047
1048         value_size = bpf_map_value_size(map);
1049
1050         err = -ENOMEM;
1051         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1052         if (!value)
1053                 goto free_key;
1054
1055         err = bpf_map_copy_value(map, key, value, attr->flags);
1056         if (err)
1057                 goto free_value;
1058
1059         err = -EFAULT;
1060         if (copy_to_user(uvalue, value, value_size) != 0)
1061                 goto free_value;
1062
1063         err = 0;
1064
1065 free_value:
1066         kfree(value);
1067 free_key:
1068         kfree(key);
1069 err_put:
1070         fdput(f);
1071         return err;
1072 }
1073
1074
1075 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1076
1077 static int map_update_elem(union bpf_attr *attr)
1078 {
1079         void __user *ukey = u64_to_user_ptr(attr->key);
1080         void __user *uvalue = u64_to_user_ptr(attr->value);
1081         int ufd = attr->map_fd;
1082         struct bpf_map *map;
1083         void *key, *value;
1084         u32 value_size;
1085         struct fd f;
1086         int err;
1087
1088         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1089                 return -EINVAL;
1090
1091         f = fdget(ufd);
1092         map = __bpf_map_get(f);
1093         if (IS_ERR(map))
1094                 return PTR_ERR(map);
1095         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1096                 err = -EPERM;
1097                 goto err_put;
1098         }
1099
1100         if ((attr->flags & BPF_F_LOCK) &&
1101             !map_value_has_spin_lock(map)) {
1102                 err = -EINVAL;
1103                 goto err_put;
1104         }
1105
1106         key = __bpf_copy_key(ukey, map->key_size);
1107         if (IS_ERR(key)) {
1108                 err = PTR_ERR(key);
1109                 goto err_put;
1110         }
1111
1112         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1113             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1114             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1115             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1116                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1117         else
1118                 value_size = map->value_size;
1119
1120         err = -ENOMEM;
1121         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1122         if (!value)
1123                 goto free_key;
1124
1125         err = -EFAULT;
1126         if (copy_from_user(value, uvalue, value_size) != 0)
1127                 goto free_value;
1128
1129         err = bpf_map_update_value(map, f, key, value, attr->flags);
1130
1131 free_value:
1132         kfree(value);
1133 free_key:
1134         kfree(key);
1135 err_put:
1136         fdput(f);
1137         return err;
1138 }
1139
1140 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1141
1142 static int map_delete_elem(union bpf_attr *attr)
1143 {
1144         void __user *ukey = u64_to_user_ptr(attr->key);
1145         int ufd = attr->map_fd;
1146         struct bpf_map *map;
1147         struct fd f;
1148         void *key;
1149         int err;
1150
1151         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1152                 return -EINVAL;
1153
1154         f = fdget(ufd);
1155         map = __bpf_map_get(f);
1156         if (IS_ERR(map))
1157                 return PTR_ERR(map);
1158         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1159                 err = -EPERM;
1160                 goto err_put;
1161         }
1162
1163         key = __bpf_copy_key(ukey, map->key_size);
1164         if (IS_ERR(key)) {
1165                 err = PTR_ERR(key);
1166                 goto err_put;
1167         }
1168
1169         if (bpf_map_is_dev_bound(map)) {
1170                 err = bpf_map_offload_delete_elem(map, key);
1171                 goto out;
1172         } else if (IS_FD_PROG_ARRAY(map) ||
1173                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1174                 /* These maps require sleepable context */
1175                 err = map->ops->map_delete_elem(map, key);
1176                 goto out;
1177         }
1178
1179         bpf_disable_instrumentation();
1180         rcu_read_lock();
1181         err = map->ops->map_delete_elem(map, key);
1182         rcu_read_unlock();
1183         bpf_enable_instrumentation();
1184         maybe_wait_bpf_programs(map);
1185 out:
1186         kfree(key);
1187 err_put:
1188         fdput(f);
1189         return err;
1190 }
1191
1192 /* last field in 'union bpf_attr' used by this command */
1193 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1194
1195 static int map_get_next_key(union bpf_attr *attr)
1196 {
1197         void __user *ukey = u64_to_user_ptr(attr->key);
1198         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1199         int ufd = attr->map_fd;
1200         struct bpf_map *map;
1201         void *key, *next_key;
1202         struct fd f;
1203         int err;
1204
1205         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1206                 return -EINVAL;
1207
1208         f = fdget(ufd);
1209         map = __bpf_map_get(f);
1210         if (IS_ERR(map))
1211                 return PTR_ERR(map);
1212         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1213                 err = -EPERM;
1214                 goto err_put;
1215         }
1216
1217         if (ukey) {
1218                 key = __bpf_copy_key(ukey, map->key_size);
1219                 if (IS_ERR(key)) {
1220                         err = PTR_ERR(key);
1221                         goto err_put;
1222                 }
1223         } else {
1224                 key = NULL;
1225         }
1226
1227         err = -ENOMEM;
1228         next_key = kmalloc(map->key_size, GFP_USER);
1229         if (!next_key)
1230                 goto free_key;
1231
1232         if (bpf_map_is_dev_bound(map)) {
1233                 err = bpf_map_offload_get_next_key(map, key, next_key);
1234                 goto out;
1235         }
1236
1237         rcu_read_lock();
1238         err = map->ops->map_get_next_key(map, key, next_key);
1239         rcu_read_unlock();
1240 out:
1241         if (err)
1242                 goto free_next_key;
1243
1244         err = -EFAULT;
1245         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1246                 goto free_next_key;
1247
1248         err = 0;
1249
1250 free_next_key:
1251         kfree(next_key);
1252 free_key:
1253         kfree(key);
1254 err_put:
1255         fdput(f);
1256         return err;
1257 }
1258
1259 int generic_map_delete_batch(struct bpf_map *map,
1260                              const union bpf_attr *attr,
1261                              union bpf_attr __user *uattr)
1262 {
1263         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1264         u32 cp, max_count;
1265         int err = 0;
1266         void *key;
1267
1268         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1269                 return -EINVAL;
1270
1271         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1272             !map_value_has_spin_lock(map)) {
1273                 return -EINVAL;
1274         }
1275
1276         max_count = attr->batch.count;
1277         if (!max_count)
1278                 return 0;
1279
1280         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1281         if (!key)
1282                 return -ENOMEM;
1283
1284         for (cp = 0; cp < max_count; cp++) {
1285                 err = -EFAULT;
1286                 if (copy_from_user(key, keys + cp * map->key_size,
1287                                    map->key_size))
1288                         break;
1289
1290                 if (bpf_map_is_dev_bound(map)) {
1291                         err = bpf_map_offload_delete_elem(map, key);
1292                         break;
1293                 }
1294
1295                 bpf_disable_instrumentation();
1296                 rcu_read_lock();
1297                 err = map->ops->map_delete_elem(map, key);
1298                 rcu_read_unlock();
1299                 bpf_enable_instrumentation();
1300                 maybe_wait_bpf_programs(map);
1301                 if (err)
1302                         break;
1303         }
1304         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1305                 err = -EFAULT;
1306
1307         kfree(key);
1308         return err;
1309 }
1310
1311 int generic_map_update_batch(struct bpf_map *map,
1312                              const union bpf_attr *attr,
1313                              union bpf_attr __user *uattr)
1314 {
1315         void __user *values = u64_to_user_ptr(attr->batch.values);
1316         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1317         u32 value_size, cp, max_count;
1318         int ufd = attr->map_fd;
1319         void *key, *value;
1320         struct fd f;
1321         int err = 0;
1322
1323         f = fdget(ufd);
1324         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1325                 return -EINVAL;
1326
1327         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1328             !map_value_has_spin_lock(map)) {
1329                 return -EINVAL;
1330         }
1331
1332         value_size = bpf_map_value_size(map);
1333
1334         max_count = attr->batch.count;
1335         if (!max_count)
1336                 return 0;
1337
1338         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1339         if (!key)
1340                 return -ENOMEM;
1341
1342         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1343         if (!value) {
1344                 kfree(key);
1345                 return -ENOMEM;
1346         }
1347
1348         for (cp = 0; cp < max_count; cp++) {
1349                 err = -EFAULT;
1350                 if (copy_from_user(key, keys + cp * map->key_size,
1351                     map->key_size) ||
1352                     copy_from_user(value, values + cp * value_size, value_size))
1353                         break;
1354
1355                 err = bpf_map_update_value(map, f, key, value,
1356                                            attr->batch.elem_flags);
1357
1358                 if (err)
1359                         break;
1360         }
1361
1362         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1363                 err = -EFAULT;
1364
1365         kfree(value);
1366         kfree(key);
1367         return err;
1368 }
1369
1370 #define MAP_LOOKUP_RETRIES 3
1371
1372 int generic_map_lookup_batch(struct bpf_map *map,
1373                                     const union bpf_attr *attr,
1374                                     union bpf_attr __user *uattr)
1375 {
1376         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1377         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1378         void __user *values = u64_to_user_ptr(attr->batch.values);
1379         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1380         void *buf, *buf_prevkey, *prev_key, *key, *value;
1381         int err, retry = MAP_LOOKUP_RETRIES;
1382         u32 value_size, cp, max_count;
1383
1384         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1385                 return -EINVAL;
1386
1387         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1388             !map_value_has_spin_lock(map))
1389                 return -EINVAL;
1390
1391         value_size = bpf_map_value_size(map);
1392
1393         max_count = attr->batch.count;
1394         if (!max_count)
1395                 return 0;
1396
1397         if (put_user(0, &uattr->batch.count))
1398                 return -EFAULT;
1399
1400         buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1401         if (!buf_prevkey)
1402                 return -ENOMEM;
1403
1404         buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1405         if (!buf) {
1406                 kfree(buf_prevkey);
1407                 return -ENOMEM;
1408         }
1409
1410         err = -EFAULT;
1411         prev_key = NULL;
1412         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1413                 goto free_buf;
1414         key = buf;
1415         value = key + map->key_size;
1416         if (ubatch)
1417                 prev_key = buf_prevkey;
1418
1419         for (cp = 0; cp < max_count;) {
1420                 rcu_read_lock();
1421                 err = map->ops->map_get_next_key(map, prev_key, key);
1422                 rcu_read_unlock();
1423                 if (err)
1424                         break;
1425                 err = bpf_map_copy_value(map, key, value,
1426                                          attr->batch.elem_flags);
1427
1428                 if (err == -ENOENT) {
1429                         if (retry) {
1430                                 retry--;
1431                                 continue;
1432                         }
1433                         err = -EINTR;
1434                         break;
1435                 }
1436
1437                 if (err)
1438                         goto free_buf;
1439
1440                 if (copy_to_user(keys + cp * map->key_size, key,
1441                                  map->key_size)) {
1442                         err = -EFAULT;
1443                         goto free_buf;
1444                 }
1445                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1446                         err = -EFAULT;
1447                         goto free_buf;
1448                 }
1449
1450                 if (!prev_key)
1451                         prev_key = buf_prevkey;
1452
1453                 swap(prev_key, key);
1454                 retry = MAP_LOOKUP_RETRIES;
1455                 cp++;
1456         }
1457
1458         if (err == -EFAULT)
1459                 goto free_buf;
1460
1461         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1462                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1463                 err = -EFAULT;
1464
1465 free_buf:
1466         kfree(buf_prevkey);
1467         kfree(buf);
1468         return err;
1469 }
1470
1471 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1472
1473 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1474 {
1475         void __user *ukey = u64_to_user_ptr(attr->key);
1476         void __user *uvalue = u64_to_user_ptr(attr->value);
1477         int ufd = attr->map_fd;
1478         struct bpf_map *map;
1479         void *key, *value;
1480         u32 value_size;
1481         struct fd f;
1482         int err;
1483
1484         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1485                 return -EINVAL;
1486
1487         f = fdget(ufd);
1488         map = __bpf_map_get(f);
1489         if (IS_ERR(map))
1490                 return PTR_ERR(map);
1491         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1492             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1493                 err = -EPERM;
1494                 goto err_put;
1495         }
1496
1497         key = __bpf_copy_key(ukey, map->key_size);
1498         if (IS_ERR(key)) {
1499                 err = PTR_ERR(key);
1500                 goto err_put;
1501         }
1502
1503         value_size = map->value_size;
1504
1505         err = -ENOMEM;
1506         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1507         if (!value)
1508                 goto free_key;
1509
1510         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1511             map->map_type == BPF_MAP_TYPE_STACK) {
1512                 err = map->ops->map_pop_elem(map, value);
1513         } else {
1514                 err = -ENOTSUPP;
1515         }
1516
1517         if (err)
1518                 goto free_value;
1519
1520         if (copy_to_user(uvalue, value, value_size) != 0) {
1521                 err = -EFAULT;
1522                 goto free_value;
1523         }
1524
1525         err = 0;
1526
1527 free_value:
1528         kfree(value);
1529 free_key:
1530         kfree(key);
1531 err_put:
1532         fdput(f);
1533         return err;
1534 }
1535
1536 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1537
1538 static int map_freeze(const union bpf_attr *attr)
1539 {
1540         int err = 0, ufd = attr->map_fd;
1541         struct bpf_map *map;
1542         struct fd f;
1543
1544         if (CHECK_ATTR(BPF_MAP_FREEZE))
1545                 return -EINVAL;
1546
1547         f = fdget(ufd);
1548         map = __bpf_map_get(f);
1549         if (IS_ERR(map))
1550                 return PTR_ERR(map);
1551
1552         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1553                 fdput(f);
1554                 return -ENOTSUPP;
1555         }
1556
1557         mutex_lock(&map->freeze_mutex);
1558
1559         if (map->writecnt) {
1560                 err = -EBUSY;
1561                 goto err_put;
1562         }
1563         if (READ_ONCE(map->frozen)) {
1564                 err = -EBUSY;
1565                 goto err_put;
1566         }
1567         if (!bpf_capable()) {
1568                 err = -EPERM;
1569                 goto err_put;
1570         }
1571
1572         WRITE_ONCE(map->frozen, true);
1573 err_put:
1574         mutex_unlock(&map->freeze_mutex);
1575         fdput(f);
1576         return err;
1577 }
1578
1579 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1580 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1581         [_id] = & _name ## _prog_ops,
1582 #define BPF_MAP_TYPE(_id, _ops)
1583 #define BPF_LINK_TYPE(_id, _name)
1584 #include <linux/bpf_types.h>
1585 #undef BPF_PROG_TYPE
1586 #undef BPF_MAP_TYPE
1587 #undef BPF_LINK_TYPE
1588 };
1589
1590 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1591 {
1592         const struct bpf_prog_ops *ops;
1593
1594         if (type >= ARRAY_SIZE(bpf_prog_types))
1595                 return -EINVAL;
1596         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1597         ops = bpf_prog_types[type];
1598         if (!ops)
1599                 return -EINVAL;
1600
1601         if (!bpf_prog_is_dev_bound(prog->aux))
1602                 prog->aux->ops = ops;
1603         else
1604                 prog->aux->ops = &bpf_offload_prog_ops;
1605         prog->type = type;
1606         return 0;
1607 }
1608
1609 enum bpf_audit {
1610         BPF_AUDIT_LOAD,
1611         BPF_AUDIT_UNLOAD,
1612         BPF_AUDIT_MAX,
1613 };
1614
1615 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1616         [BPF_AUDIT_LOAD]   = "LOAD",
1617         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1618 };
1619
1620 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1621 {
1622         struct audit_context *ctx = NULL;
1623         struct audit_buffer *ab;
1624
1625         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1626                 return;
1627         if (audit_enabled == AUDIT_OFF)
1628                 return;
1629         if (op == BPF_AUDIT_LOAD)
1630                 ctx = audit_context();
1631         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1632         if (unlikely(!ab))
1633                 return;
1634         audit_log_format(ab, "prog-id=%u op=%s",
1635                          prog->aux->id, bpf_audit_str[op]);
1636         audit_log_end(ab);
1637 }
1638
1639 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1640 {
1641         int id;
1642
1643         idr_preload(GFP_KERNEL);
1644         spin_lock_bh(&prog_idr_lock);
1645         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1646         if (id > 0)
1647                 prog->aux->id = id;
1648         spin_unlock_bh(&prog_idr_lock);
1649         idr_preload_end();
1650
1651         /* id is in [1, INT_MAX) */
1652         if (WARN_ON_ONCE(!id))
1653                 return -ENOSPC;
1654
1655         return id > 0 ? 0 : id;
1656 }
1657
1658 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1659 {
1660         /* cBPF to eBPF migrations are currently not in the idr store.
1661          * Offloaded programs are removed from the store when their device
1662          * disappears - even if someone grabs an fd to them they are unusable,
1663          * simply waiting for refcnt to drop to be freed.
1664          */
1665         if (!prog->aux->id)
1666                 return;
1667
1668         if (do_idr_lock)
1669                 spin_lock_bh(&prog_idr_lock);
1670         else
1671                 __acquire(&prog_idr_lock);
1672
1673         idr_remove(&prog_idr, prog->aux->id);
1674         prog->aux->id = 0;
1675
1676         if (do_idr_lock)
1677                 spin_unlock_bh(&prog_idr_lock);
1678         else
1679                 __release(&prog_idr_lock);
1680 }
1681
1682 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1683 {
1684         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1685
1686         kvfree(aux->func_info);
1687         kfree(aux->func_info_aux);
1688         free_uid(aux->user);
1689         security_bpf_prog_free(aux);
1690         bpf_prog_free(aux->prog);
1691 }
1692
1693 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1694 {
1695         bpf_prog_kallsyms_del_all(prog);
1696         btf_put(prog->aux->btf);
1697         kvfree(prog->aux->jited_linfo);
1698         kvfree(prog->aux->linfo);
1699         kfree(prog->aux->kfunc_tab);
1700         if (prog->aux->attach_btf)
1701                 btf_put(prog->aux->attach_btf);
1702
1703         if (deferred) {
1704                 if (prog->aux->sleepable)
1705                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1706                 else
1707                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1708         } else {
1709                 __bpf_prog_put_rcu(&prog->aux->rcu);
1710         }
1711 }
1712
1713 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1714 {
1715         if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1716                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1717                 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1718                 /* bpf_prog_free_id() must be called first */
1719                 bpf_prog_free_id(prog, do_idr_lock);
1720                 __bpf_prog_put_noref(prog, true);
1721         }
1722 }
1723
1724 void bpf_prog_put(struct bpf_prog *prog)
1725 {
1726         __bpf_prog_put(prog, true);
1727 }
1728 EXPORT_SYMBOL_GPL(bpf_prog_put);
1729
1730 static int bpf_prog_release(struct inode *inode, struct file *filp)
1731 {
1732         struct bpf_prog *prog = filp->private_data;
1733
1734         bpf_prog_put(prog);
1735         return 0;
1736 }
1737
1738 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1739                                struct bpf_prog_stats *stats)
1740 {
1741         u64 nsecs = 0, cnt = 0, misses = 0;
1742         int cpu;
1743
1744         for_each_possible_cpu(cpu) {
1745                 const struct bpf_prog_stats *st;
1746                 unsigned int start;
1747                 u64 tnsecs, tcnt, tmisses;
1748
1749                 st = per_cpu_ptr(prog->stats, cpu);
1750                 do {
1751                         start = u64_stats_fetch_begin_irq(&st->syncp);
1752                         tnsecs = st->nsecs;
1753                         tcnt = st->cnt;
1754                         tmisses = st->misses;
1755                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1756                 nsecs += tnsecs;
1757                 cnt += tcnt;
1758                 misses += tmisses;
1759         }
1760         stats->nsecs = nsecs;
1761         stats->cnt = cnt;
1762         stats->misses = misses;
1763 }
1764
1765 #ifdef CONFIG_PROC_FS
1766 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1767 {
1768         const struct bpf_prog *prog = filp->private_data;
1769         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1770         struct bpf_prog_stats stats;
1771
1772         bpf_prog_get_stats(prog, &stats);
1773         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1774         seq_printf(m,
1775                    "prog_type:\t%u\n"
1776                    "prog_jited:\t%u\n"
1777                    "prog_tag:\t%s\n"
1778                    "memlock:\t%llu\n"
1779                    "prog_id:\t%u\n"
1780                    "run_time_ns:\t%llu\n"
1781                    "run_cnt:\t%llu\n"
1782                    "recursion_misses:\t%llu\n",
1783                    prog->type,
1784                    prog->jited,
1785                    prog_tag,
1786                    prog->pages * 1ULL << PAGE_SHIFT,
1787                    prog->aux->id,
1788                    stats.nsecs,
1789                    stats.cnt,
1790                    stats.misses);
1791 }
1792 #endif
1793
1794 const struct file_operations bpf_prog_fops = {
1795 #ifdef CONFIG_PROC_FS
1796         .show_fdinfo    = bpf_prog_show_fdinfo,
1797 #endif
1798         .release        = bpf_prog_release,
1799         .read           = bpf_dummy_read,
1800         .write          = bpf_dummy_write,
1801 };
1802
1803 int bpf_prog_new_fd(struct bpf_prog *prog)
1804 {
1805         int ret;
1806
1807         ret = security_bpf_prog(prog);
1808         if (ret < 0)
1809                 return ret;
1810
1811         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1812                                 O_RDWR | O_CLOEXEC);
1813 }
1814
1815 static struct bpf_prog *____bpf_prog_get(struct fd f)
1816 {
1817         if (!f.file)
1818                 return ERR_PTR(-EBADF);
1819         if (f.file->f_op != &bpf_prog_fops) {
1820                 fdput(f);
1821                 return ERR_PTR(-EINVAL);
1822         }
1823
1824         return f.file->private_data;
1825 }
1826
1827 void bpf_prog_add(struct bpf_prog *prog, int i)
1828 {
1829         atomic64_add(i, &prog->aux->refcnt);
1830 }
1831 EXPORT_SYMBOL_GPL(bpf_prog_add);
1832
1833 void bpf_prog_sub(struct bpf_prog *prog, int i)
1834 {
1835         /* Only to be used for undoing previous bpf_prog_add() in some
1836          * error path. We still know that another entity in our call
1837          * path holds a reference to the program, thus atomic_sub() can
1838          * be safely used in such cases!
1839          */
1840         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1841 }
1842 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1843
1844 void bpf_prog_inc(struct bpf_prog *prog)
1845 {
1846         atomic64_inc(&prog->aux->refcnt);
1847 }
1848 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1849
1850 /* prog_idr_lock should have been held */
1851 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1852 {
1853         int refold;
1854
1855         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1856
1857         if (!refold)
1858                 return ERR_PTR(-ENOENT);
1859
1860         return prog;
1861 }
1862 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1863
1864 bool bpf_prog_get_ok(struct bpf_prog *prog,
1865                             enum bpf_prog_type *attach_type, bool attach_drv)
1866 {
1867         /* not an attachment, just a refcount inc, always allow */
1868         if (!attach_type)
1869                 return true;
1870
1871         if (prog->type != *attach_type)
1872                 return false;
1873         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1874                 return false;
1875
1876         return true;
1877 }
1878
1879 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1880                                        bool attach_drv)
1881 {
1882         struct fd f = fdget(ufd);
1883         struct bpf_prog *prog;
1884
1885         prog = ____bpf_prog_get(f);
1886         if (IS_ERR(prog))
1887                 return prog;
1888         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1889                 prog = ERR_PTR(-EINVAL);
1890                 goto out;
1891         }
1892
1893         bpf_prog_inc(prog);
1894 out:
1895         fdput(f);
1896         return prog;
1897 }
1898
1899 struct bpf_prog *bpf_prog_get(u32 ufd)
1900 {
1901         return __bpf_prog_get(ufd, NULL, false);
1902 }
1903
1904 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1905                                        bool attach_drv)
1906 {
1907         return __bpf_prog_get(ufd, &type, attach_drv);
1908 }
1909 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1910
1911 /* Initially all BPF programs could be loaded w/o specifying
1912  * expected_attach_type. Later for some of them specifying expected_attach_type
1913  * at load time became required so that program could be validated properly.
1914  * Programs of types that are allowed to be loaded both w/ and w/o (for
1915  * backward compatibility) expected_attach_type, should have the default attach
1916  * type assigned to expected_attach_type for the latter case, so that it can be
1917  * validated later at attach time.
1918  *
1919  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1920  * prog type requires it but has some attach types that have to be backward
1921  * compatible.
1922  */
1923 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1924 {
1925         switch (attr->prog_type) {
1926         case BPF_PROG_TYPE_CGROUP_SOCK:
1927                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1928                  * exist so checking for non-zero is the way to go here.
1929                  */
1930                 if (!attr->expected_attach_type)
1931                         attr->expected_attach_type =
1932                                 BPF_CGROUP_INET_SOCK_CREATE;
1933                 break;
1934         }
1935 }
1936
1937 static int
1938 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1939                            enum bpf_attach_type expected_attach_type,
1940                            struct btf *attach_btf, u32 btf_id,
1941                            struct bpf_prog *dst_prog)
1942 {
1943         if (btf_id) {
1944                 if (btf_id > BTF_MAX_TYPE)
1945                         return -EINVAL;
1946
1947                 if (!attach_btf && !dst_prog)
1948                         return -EINVAL;
1949
1950                 switch (prog_type) {
1951                 case BPF_PROG_TYPE_TRACING:
1952                 case BPF_PROG_TYPE_LSM:
1953                 case BPF_PROG_TYPE_STRUCT_OPS:
1954                 case BPF_PROG_TYPE_EXT:
1955                         break;
1956                 default:
1957                         return -EINVAL;
1958                 }
1959         }
1960
1961         if (attach_btf && (!btf_id || dst_prog))
1962                 return -EINVAL;
1963
1964         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1965             prog_type != BPF_PROG_TYPE_EXT)
1966                 return -EINVAL;
1967
1968         switch (prog_type) {
1969         case BPF_PROG_TYPE_CGROUP_SOCK:
1970                 switch (expected_attach_type) {
1971                 case BPF_CGROUP_INET_SOCK_CREATE:
1972                 case BPF_CGROUP_INET_SOCK_RELEASE:
1973                 case BPF_CGROUP_INET4_POST_BIND:
1974                 case BPF_CGROUP_INET6_POST_BIND:
1975                         return 0;
1976                 default:
1977                         return -EINVAL;
1978                 }
1979         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1980                 switch (expected_attach_type) {
1981                 case BPF_CGROUP_INET4_BIND:
1982                 case BPF_CGROUP_INET6_BIND:
1983                 case BPF_CGROUP_INET4_CONNECT:
1984                 case BPF_CGROUP_INET6_CONNECT:
1985                 case BPF_CGROUP_INET4_GETPEERNAME:
1986                 case BPF_CGROUP_INET6_GETPEERNAME:
1987                 case BPF_CGROUP_INET4_GETSOCKNAME:
1988                 case BPF_CGROUP_INET6_GETSOCKNAME:
1989                 case BPF_CGROUP_UDP4_SENDMSG:
1990                 case BPF_CGROUP_UDP6_SENDMSG:
1991                 case BPF_CGROUP_UDP4_RECVMSG:
1992                 case BPF_CGROUP_UDP6_RECVMSG:
1993                         return 0;
1994                 default:
1995                         return -EINVAL;
1996                 }
1997         case BPF_PROG_TYPE_CGROUP_SKB:
1998                 switch (expected_attach_type) {
1999                 case BPF_CGROUP_INET_INGRESS:
2000                 case BPF_CGROUP_INET_EGRESS:
2001                         return 0;
2002                 default:
2003                         return -EINVAL;
2004                 }
2005         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2006                 switch (expected_attach_type) {
2007                 case BPF_CGROUP_SETSOCKOPT:
2008                 case BPF_CGROUP_GETSOCKOPT:
2009                         return 0;
2010                 default:
2011                         return -EINVAL;
2012                 }
2013         case BPF_PROG_TYPE_SK_LOOKUP:
2014                 if (expected_attach_type == BPF_SK_LOOKUP)
2015                         return 0;
2016                 return -EINVAL;
2017         case BPF_PROG_TYPE_EXT:
2018                 if (expected_attach_type)
2019                         return -EINVAL;
2020                 fallthrough;
2021         default:
2022                 return 0;
2023         }
2024 }
2025
2026 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2027 {
2028         switch (prog_type) {
2029         case BPF_PROG_TYPE_SCHED_CLS:
2030         case BPF_PROG_TYPE_SCHED_ACT:
2031         case BPF_PROG_TYPE_XDP:
2032         case BPF_PROG_TYPE_LWT_IN:
2033         case BPF_PROG_TYPE_LWT_OUT:
2034         case BPF_PROG_TYPE_LWT_XMIT:
2035         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2036         case BPF_PROG_TYPE_SK_SKB:
2037         case BPF_PROG_TYPE_SK_MSG:
2038         case BPF_PROG_TYPE_LIRC_MODE2:
2039         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2040         case BPF_PROG_TYPE_CGROUP_DEVICE:
2041         case BPF_PROG_TYPE_CGROUP_SOCK:
2042         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2043         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2044         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2045         case BPF_PROG_TYPE_SOCK_OPS:
2046         case BPF_PROG_TYPE_EXT: /* extends any prog */
2047                 return true;
2048         case BPF_PROG_TYPE_CGROUP_SKB:
2049                 /* always unpriv */
2050         case BPF_PROG_TYPE_SK_REUSEPORT:
2051                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2052         default:
2053                 return false;
2054         }
2055 }
2056
2057 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2058 {
2059         switch (prog_type) {
2060         case BPF_PROG_TYPE_KPROBE:
2061         case BPF_PROG_TYPE_TRACEPOINT:
2062         case BPF_PROG_TYPE_PERF_EVENT:
2063         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2064         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2065         case BPF_PROG_TYPE_TRACING:
2066         case BPF_PROG_TYPE_LSM:
2067         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2068         case BPF_PROG_TYPE_EXT: /* extends any prog */
2069                 return true;
2070         default:
2071                 return false;
2072         }
2073 }
2074
2075 /* last field in 'union bpf_attr' used by this command */
2076 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2077
2078 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2079 {
2080         enum bpf_prog_type type = attr->prog_type;
2081         struct bpf_prog *prog, *dst_prog = NULL;
2082         struct btf *attach_btf = NULL;
2083         int err;
2084         char license[128];
2085         bool is_gpl;
2086
2087         if (CHECK_ATTR(BPF_PROG_LOAD))
2088                 return -EINVAL;
2089
2090         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2091                                  BPF_F_ANY_ALIGNMENT |
2092                                  BPF_F_TEST_STATE_FREQ |
2093                                  BPF_F_SLEEPABLE |
2094                                  BPF_F_TEST_RND_HI32))
2095                 return -EINVAL;
2096
2097         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2098             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2099             !bpf_capable())
2100                 return -EPERM;
2101
2102         /* copy eBPF program license from user space */
2103         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2104                               sizeof(license) - 1) < 0)
2105                 return -EFAULT;
2106         license[sizeof(license) - 1] = 0;
2107
2108         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2109         is_gpl = license_is_gpl_compatible(license);
2110
2111         if (attr->insn_cnt == 0 ||
2112             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2113                 return -E2BIG;
2114         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2115             type != BPF_PROG_TYPE_CGROUP_SKB &&
2116             !bpf_capable())
2117                 return -EPERM;
2118
2119         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2120                 return -EPERM;
2121         if (is_perfmon_prog_type(type) && !perfmon_capable())
2122                 return -EPERM;
2123
2124         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2125          * or btf, we need to check which one it is
2126          */
2127         if (attr->attach_prog_fd) {
2128                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2129                 if (IS_ERR(dst_prog)) {
2130                         dst_prog = NULL;
2131                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2132                         if (IS_ERR(attach_btf))
2133                                 return -EINVAL;
2134                         if (!btf_is_kernel(attach_btf)) {
2135                                 /* attaching through specifying bpf_prog's BTF
2136                                  * objects directly might be supported eventually
2137                                  */
2138                                 btf_put(attach_btf);
2139                                 return -ENOTSUPP;
2140                         }
2141                 }
2142         } else if (attr->attach_btf_id) {
2143                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2144                 attach_btf = bpf_get_btf_vmlinux();
2145                 if (IS_ERR(attach_btf))
2146                         return PTR_ERR(attach_btf);
2147                 if (!attach_btf)
2148                         return -EINVAL;
2149                 btf_get(attach_btf);
2150         }
2151
2152         bpf_prog_load_fixup_attach_type(attr);
2153         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2154                                        attach_btf, attr->attach_btf_id,
2155                                        dst_prog)) {
2156                 if (dst_prog)
2157                         bpf_prog_put(dst_prog);
2158                 if (attach_btf)
2159                         btf_put(attach_btf);
2160                 return -EINVAL;
2161         }
2162
2163         /* plain bpf_prog allocation */
2164         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2165         if (!prog) {
2166                 if (dst_prog)
2167                         bpf_prog_put(dst_prog);
2168                 if (attach_btf)
2169                         btf_put(attach_btf);
2170                 return -ENOMEM;
2171         }
2172
2173         prog->expected_attach_type = attr->expected_attach_type;
2174         prog->aux->attach_btf = attach_btf;
2175         prog->aux->attach_btf_id = attr->attach_btf_id;
2176         prog->aux->dst_prog = dst_prog;
2177         prog->aux->offload_requested = !!attr->prog_ifindex;
2178         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2179
2180         err = security_bpf_prog_alloc(prog->aux);
2181         if (err)
2182                 goto free_prog;
2183
2184         prog->aux->user = get_current_user();
2185         prog->len = attr->insn_cnt;
2186
2187         err = -EFAULT;
2188         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2189                            bpf_prog_insn_size(prog)) != 0)
2190                 goto free_prog_sec;
2191
2192         prog->orig_prog = NULL;
2193         prog->jited = 0;
2194
2195         atomic64_set(&prog->aux->refcnt, 1);
2196         prog->gpl_compatible = is_gpl ? 1 : 0;
2197
2198         if (bpf_prog_is_dev_bound(prog->aux)) {
2199                 err = bpf_prog_offload_init(prog, attr);
2200                 if (err)
2201                         goto free_prog_sec;
2202         }
2203
2204         /* find program type: socket_filter vs tracing_filter */
2205         err = find_prog_type(type, prog);
2206         if (err < 0)
2207                 goto free_prog_sec;
2208
2209         prog->aux->load_time = ktime_get_boottime_ns();
2210         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2211                                sizeof(attr->prog_name));
2212         if (err < 0)
2213                 goto free_prog_sec;
2214
2215         /* run eBPF verifier */
2216         err = bpf_check(&prog, attr, uattr);
2217         if (err < 0)
2218                 goto free_used_maps;
2219
2220         prog = bpf_prog_select_runtime(prog, &err);
2221         if (err < 0)
2222                 goto free_used_maps;
2223
2224         err = bpf_prog_alloc_id(prog);
2225         if (err)
2226                 goto free_used_maps;
2227
2228         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2229          * effectively publicly exposed. However, retrieving via
2230          * bpf_prog_get_fd_by_id() will take another reference,
2231          * therefore it cannot be gone underneath us.
2232          *
2233          * Only for the time /after/ successful bpf_prog_new_fd()
2234          * and before returning to userspace, we might just hold
2235          * one reference and any parallel close on that fd could
2236          * rip everything out. Hence, below notifications must
2237          * happen before bpf_prog_new_fd().
2238          *
2239          * Also, any failure handling from this point onwards must
2240          * be using bpf_prog_put() given the program is exposed.
2241          */
2242         bpf_prog_kallsyms_add(prog);
2243         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2244         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2245
2246         err = bpf_prog_new_fd(prog);
2247         if (err < 0)
2248                 bpf_prog_put(prog);
2249         return err;
2250
2251 free_used_maps:
2252         /* In case we have subprogs, we need to wait for a grace
2253          * period before we can tear down JIT memory since symbols
2254          * are already exposed under kallsyms.
2255          */
2256         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2257         return err;
2258 free_prog_sec:
2259         free_uid(prog->aux->user);
2260         security_bpf_prog_free(prog->aux);
2261 free_prog:
2262         if (prog->aux->attach_btf)
2263                 btf_put(prog->aux->attach_btf);
2264         bpf_prog_free(prog);
2265         return err;
2266 }
2267
2268 #define BPF_OBJ_LAST_FIELD file_flags
2269
2270 static int bpf_obj_pin(const union bpf_attr *attr)
2271 {
2272         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2273                 return -EINVAL;
2274
2275         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2276 }
2277
2278 static int bpf_obj_get(const union bpf_attr *attr)
2279 {
2280         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2281             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2282                 return -EINVAL;
2283
2284         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2285                                 attr->file_flags);
2286 }
2287
2288 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2289                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2290 {
2291         atomic64_set(&link->refcnt, 1);
2292         link->type = type;
2293         link->id = 0;
2294         link->ops = ops;
2295         link->prog = prog;
2296 }
2297
2298 static void bpf_link_free_id(int id)
2299 {
2300         if (!id)
2301                 return;
2302
2303         spin_lock_bh(&link_idr_lock);
2304         idr_remove(&link_idr, id);
2305         spin_unlock_bh(&link_idr_lock);
2306 }
2307
2308 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2309  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2310  * anon_inode's release() call. This helper marksbpf_link as
2311  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2312  * is not decremented, it's the responsibility of a calling code that failed
2313  * to complete bpf_link initialization.
2314  */
2315 void bpf_link_cleanup(struct bpf_link_primer *primer)
2316 {
2317         primer->link->prog = NULL;
2318         bpf_link_free_id(primer->id);
2319         fput(primer->file);
2320         put_unused_fd(primer->fd);
2321 }
2322
2323 void bpf_link_inc(struct bpf_link *link)
2324 {
2325         atomic64_inc(&link->refcnt);
2326 }
2327
2328 /* bpf_link_free is guaranteed to be called from process context */
2329 static void bpf_link_free(struct bpf_link *link)
2330 {
2331         bpf_link_free_id(link->id);
2332         if (link->prog) {
2333                 /* detach BPF program, clean up used resources */
2334                 link->ops->release(link);
2335                 bpf_prog_put(link->prog);
2336         }
2337         /* free bpf_link and its containing memory */
2338         link->ops->dealloc(link);
2339 }
2340
2341 static void bpf_link_put_deferred(struct work_struct *work)
2342 {
2343         struct bpf_link *link = container_of(work, struct bpf_link, work);
2344
2345         bpf_link_free(link);
2346 }
2347
2348 /* bpf_link_put can be called from atomic context, but ensures that resources
2349  * are freed from process context
2350  */
2351 void bpf_link_put(struct bpf_link *link)
2352 {
2353         if (!atomic64_dec_and_test(&link->refcnt))
2354                 return;
2355
2356         if (in_atomic()) {
2357                 INIT_WORK(&link->work, bpf_link_put_deferred);
2358                 schedule_work(&link->work);
2359         } else {
2360                 bpf_link_free(link);
2361         }
2362 }
2363
2364 static int bpf_link_release(struct inode *inode, struct file *filp)
2365 {
2366         struct bpf_link *link = filp->private_data;
2367
2368         bpf_link_put(link);
2369         return 0;
2370 }
2371
2372 #ifdef CONFIG_PROC_FS
2373 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2374 #define BPF_MAP_TYPE(_id, _ops)
2375 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2376 static const char *bpf_link_type_strs[] = {
2377         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2378 #include <linux/bpf_types.h>
2379 };
2380 #undef BPF_PROG_TYPE
2381 #undef BPF_MAP_TYPE
2382 #undef BPF_LINK_TYPE
2383
2384 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2385 {
2386         const struct bpf_link *link = filp->private_data;
2387         const struct bpf_prog *prog = link->prog;
2388         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2389
2390         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2391         seq_printf(m,
2392                    "link_type:\t%s\n"
2393                    "link_id:\t%u\n"
2394                    "prog_tag:\t%s\n"
2395                    "prog_id:\t%u\n",
2396                    bpf_link_type_strs[link->type],
2397                    link->id,
2398                    prog_tag,
2399                    prog->aux->id);
2400         if (link->ops->show_fdinfo)
2401                 link->ops->show_fdinfo(link, m);
2402 }
2403 #endif
2404
2405 static const struct file_operations bpf_link_fops = {
2406 #ifdef CONFIG_PROC_FS
2407         .show_fdinfo    = bpf_link_show_fdinfo,
2408 #endif
2409         .release        = bpf_link_release,
2410         .read           = bpf_dummy_read,
2411         .write          = bpf_dummy_write,
2412 };
2413
2414 static int bpf_link_alloc_id(struct bpf_link *link)
2415 {
2416         int id;
2417
2418         idr_preload(GFP_KERNEL);
2419         spin_lock_bh(&link_idr_lock);
2420         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2421         spin_unlock_bh(&link_idr_lock);
2422         idr_preload_end();
2423
2424         return id;
2425 }
2426
2427 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2428  * reserving unused FD and allocating ID from link_idr. This is to be paired
2429  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2430  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2431  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2432  * transient state is passed around in struct bpf_link_primer.
2433  * This is preferred way to create and initialize bpf_link, especially when
2434  * there are complicated and expensive operations inbetween creating bpf_link
2435  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2436  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2437  * expensive (and potentially failing) roll back operations in a rare case
2438  * that file, FD, or ID can't be allocated.
2439  */
2440 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2441 {
2442         struct file *file;
2443         int fd, id;
2444
2445         fd = get_unused_fd_flags(O_CLOEXEC);
2446         if (fd < 0)
2447                 return fd;
2448
2449
2450         id = bpf_link_alloc_id(link);
2451         if (id < 0) {
2452                 put_unused_fd(fd);
2453                 return id;
2454         }
2455
2456         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2457         if (IS_ERR(file)) {
2458                 bpf_link_free_id(id);
2459                 put_unused_fd(fd);
2460                 return PTR_ERR(file);
2461         }
2462
2463         primer->link = link;
2464         primer->file = file;
2465         primer->fd = fd;
2466         primer->id = id;
2467         return 0;
2468 }
2469
2470 int bpf_link_settle(struct bpf_link_primer *primer)
2471 {
2472         /* make bpf_link fetchable by ID */
2473         spin_lock_bh(&link_idr_lock);
2474         primer->link->id = primer->id;
2475         spin_unlock_bh(&link_idr_lock);
2476         /* make bpf_link fetchable by FD */
2477         fd_install(primer->fd, primer->file);
2478         /* pass through installed FD */
2479         return primer->fd;
2480 }
2481
2482 int bpf_link_new_fd(struct bpf_link *link)
2483 {
2484         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2485 }
2486
2487 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2488 {
2489         struct fd f = fdget(ufd);
2490         struct bpf_link *link;
2491
2492         if (!f.file)
2493                 return ERR_PTR(-EBADF);
2494         if (f.file->f_op != &bpf_link_fops) {
2495                 fdput(f);
2496                 return ERR_PTR(-EINVAL);
2497         }
2498
2499         link = f.file->private_data;
2500         bpf_link_inc(link);
2501         fdput(f);
2502
2503         return link;
2504 }
2505
2506 struct bpf_tracing_link {
2507         struct bpf_link link;
2508         enum bpf_attach_type attach_type;
2509         struct bpf_trampoline *trampoline;
2510         struct bpf_prog *tgt_prog;
2511 };
2512
2513 static void bpf_tracing_link_release(struct bpf_link *link)
2514 {
2515         struct bpf_tracing_link *tr_link =
2516                 container_of(link, struct bpf_tracing_link, link);
2517
2518         WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2519                                                 tr_link->trampoline));
2520
2521         bpf_trampoline_put(tr_link->trampoline);
2522
2523         /* tgt_prog is NULL if target is a kernel function */
2524         if (tr_link->tgt_prog)
2525                 bpf_prog_put(tr_link->tgt_prog);
2526 }
2527
2528 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2529 {
2530         struct bpf_tracing_link *tr_link =
2531                 container_of(link, struct bpf_tracing_link, link);
2532
2533         kfree(tr_link);
2534 }
2535
2536 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2537                                          struct seq_file *seq)
2538 {
2539         struct bpf_tracing_link *tr_link =
2540                 container_of(link, struct bpf_tracing_link, link);
2541
2542         seq_printf(seq,
2543                    "attach_type:\t%d\n",
2544                    tr_link->attach_type);
2545 }
2546
2547 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2548                                            struct bpf_link_info *info)
2549 {
2550         struct bpf_tracing_link *tr_link =
2551                 container_of(link, struct bpf_tracing_link, link);
2552
2553         info->tracing.attach_type = tr_link->attach_type;
2554         bpf_trampoline_unpack_key(tr_link->trampoline->key,
2555                                   &info->tracing.target_obj_id,
2556                                   &info->tracing.target_btf_id);
2557
2558         return 0;
2559 }
2560
2561 static const struct bpf_link_ops bpf_tracing_link_lops = {
2562         .release = bpf_tracing_link_release,
2563         .dealloc = bpf_tracing_link_dealloc,
2564         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2565         .fill_link_info = bpf_tracing_link_fill_link_info,
2566 };
2567
2568 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2569                                    int tgt_prog_fd,
2570                                    u32 btf_id)
2571 {
2572         struct bpf_link_primer link_primer;
2573         struct bpf_prog *tgt_prog = NULL;
2574         struct bpf_trampoline *tr = NULL;
2575         struct bpf_tracing_link *link;
2576         u64 key = 0;
2577         int err;
2578
2579         switch (prog->type) {
2580         case BPF_PROG_TYPE_TRACING:
2581                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2582                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2583                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2584                         err = -EINVAL;
2585                         goto out_put_prog;
2586                 }
2587                 break;
2588         case BPF_PROG_TYPE_EXT:
2589                 if (prog->expected_attach_type != 0) {
2590                         err = -EINVAL;
2591                         goto out_put_prog;
2592                 }
2593                 break;
2594         case BPF_PROG_TYPE_LSM:
2595                 if (prog->expected_attach_type != BPF_LSM_MAC) {
2596                         err = -EINVAL;
2597                         goto out_put_prog;
2598                 }
2599                 break;
2600         default:
2601                 err = -EINVAL;
2602                 goto out_put_prog;
2603         }
2604
2605         if (!!tgt_prog_fd != !!btf_id) {
2606                 err = -EINVAL;
2607                 goto out_put_prog;
2608         }
2609
2610         if (tgt_prog_fd) {
2611                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2612                 if (prog->type != BPF_PROG_TYPE_EXT) {
2613                         err = -EINVAL;
2614                         goto out_put_prog;
2615                 }
2616
2617                 tgt_prog = bpf_prog_get(tgt_prog_fd);
2618                 if (IS_ERR(tgt_prog)) {
2619                         err = PTR_ERR(tgt_prog);
2620                         tgt_prog = NULL;
2621                         goto out_put_prog;
2622                 }
2623
2624                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2625         }
2626
2627         link = kzalloc(sizeof(*link), GFP_USER);
2628         if (!link) {
2629                 err = -ENOMEM;
2630                 goto out_put_prog;
2631         }
2632         bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2633                       &bpf_tracing_link_lops, prog);
2634         link->attach_type = prog->expected_attach_type;
2635
2636         mutex_lock(&prog->aux->dst_mutex);
2637
2638         /* There are a few possible cases here:
2639          *
2640          * - if prog->aux->dst_trampoline is set, the program was just loaded
2641          *   and not yet attached to anything, so we can use the values stored
2642          *   in prog->aux
2643          *
2644          * - if prog->aux->dst_trampoline is NULL, the program has already been
2645          *   attached to a target and its initial target was cleared (below)
2646          *
2647          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2648          *   target_btf_id using the link_create API.
2649          *
2650          * - if tgt_prog == NULL when this function was called using the old
2651          *   raw_tracepoint_open API, and we need a target from prog->aux
2652          *
2653          * - if prog->aux->dst_trampoline and tgt_prog is NULL, the program
2654          *   was detached and is going for re-attachment.
2655          */
2656         if (!prog->aux->dst_trampoline && !tgt_prog) {
2657                 /*
2658                  * Allow re-attach for TRACING and LSM programs. If it's
2659                  * currently linked, bpf_trampoline_link_prog will fail.
2660                  * EXT programs need to specify tgt_prog_fd, so they
2661                  * re-attach in separate code path.
2662                  */
2663                 if (prog->type != BPF_PROG_TYPE_TRACING &&
2664                     prog->type != BPF_PROG_TYPE_LSM) {
2665                         err = -EINVAL;
2666                         goto out_unlock;
2667                 }
2668                 btf_id = prog->aux->attach_btf_id;
2669                 key = bpf_trampoline_compute_key(NULL, prog->aux->attach_btf, btf_id);
2670         }
2671
2672         if (!prog->aux->dst_trampoline ||
2673             (key && key != prog->aux->dst_trampoline->key)) {
2674                 /* If there is no saved target, or the specified target is
2675                  * different from the destination specified at load time, we
2676                  * need a new trampoline and a check for compatibility
2677                  */
2678                 struct bpf_attach_target_info tgt_info = {};
2679
2680                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2681                                               &tgt_info);
2682                 if (err)
2683                         goto out_unlock;
2684
2685                 tr = bpf_trampoline_get(key, &tgt_info);
2686                 if (!tr) {
2687                         err = -ENOMEM;
2688                         goto out_unlock;
2689                 }
2690         } else {
2691                 /* The caller didn't specify a target, or the target was the
2692                  * same as the destination supplied during program load. This
2693                  * means we can reuse the trampoline and reference from program
2694                  * load time, and there is no need to allocate a new one. This
2695                  * can only happen once for any program, as the saved values in
2696                  * prog->aux are cleared below.
2697                  */
2698                 tr = prog->aux->dst_trampoline;
2699                 tgt_prog = prog->aux->dst_prog;
2700         }
2701
2702         err = bpf_link_prime(&link->link, &link_primer);
2703         if (err)
2704                 goto out_unlock;
2705
2706         err = bpf_trampoline_link_prog(prog, tr);
2707         if (err) {
2708                 bpf_link_cleanup(&link_primer);
2709                 link = NULL;
2710                 goto out_unlock;
2711         }
2712
2713         link->tgt_prog = tgt_prog;
2714         link->trampoline = tr;
2715
2716         /* Always clear the trampoline and target prog from prog->aux to make
2717          * sure the original attach destination is not kept alive after a
2718          * program is (re-)attached to another target.
2719          */
2720         if (prog->aux->dst_prog &&
2721             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2722                 /* got extra prog ref from syscall, or attaching to different prog */
2723                 bpf_prog_put(prog->aux->dst_prog);
2724         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2725                 /* we allocated a new trampoline, so free the old one */
2726                 bpf_trampoline_put(prog->aux->dst_trampoline);
2727
2728         prog->aux->dst_prog = NULL;
2729         prog->aux->dst_trampoline = NULL;
2730         mutex_unlock(&prog->aux->dst_mutex);
2731
2732         return bpf_link_settle(&link_primer);
2733 out_unlock:
2734         if (tr && tr != prog->aux->dst_trampoline)
2735                 bpf_trampoline_put(tr);
2736         mutex_unlock(&prog->aux->dst_mutex);
2737         kfree(link);
2738 out_put_prog:
2739         if (tgt_prog_fd && tgt_prog)
2740                 bpf_prog_put(tgt_prog);
2741         return err;
2742 }
2743
2744 struct bpf_raw_tp_link {
2745         struct bpf_link link;
2746         struct bpf_raw_event_map *btp;
2747 };
2748
2749 static void bpf_raw_tp_link_release(struct bpf_link *link)
2750 {
2751         struct bpf_raw_tp_link *raw_tp =
2752                 container_of(link, struct bpf_raw_tp_link, link);
2753
2754         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2755         bpf_put_raw_tracepoint(raw_tp->btp);
2756 }
2757
2758 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2759 {
2760         struct bpf_raw_tp_link *raw_tp =
2761                 container_of(link, struct bpf_raw_tp_link, link);
2762
2763         kfree(raw_tp);
2764 }
2765
2766 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2767                                         struct seq_file *seq)
2768 {
2769         struct bpf_raw_tp_link *raw_tp_link =
2770                 container_of(link, struct bpf_raw_tp_link, link);
2771
2772         seq_printf(seq,
2773                    "tp_name:\t%s\n",
2774                    raw_tp_link->btp->tp->name);
2775 }
2776
2777 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2778                                           struct bpf_link_info *info)
2779 {
2780         struct bpf_raw_tp_link *raw_tp_link =
2781                 container_of(link, struct bpf_raw_tp_link, link);
2782         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2783         const char *tp_name = raw_tp_link->btp->tp->name;
2784         u32 ulen = info->raw_tracepoint.tp_name_len;
2785         size_t tp_len = strlen(tp_name);
2786
2787         if (!ulen ^ !ubuf)
2788                 return -EINVAL;
2789
2790         info->raw_tracepoint.tp_name_len = tp_len + 1;
2791
2792         if (!ubuf)
2793                 return 0;
2794
2795         if (ulen >= tp_len + 1) {
2796                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2797                         return -EFAULT;
2798         } else {
2799                 char zero = '\0';
2800
2801                 if (copy_to_user(ubuf, tp_name, ulen - 1))
2802                         return -EFAULT;
2803                 if (put_user(zero, ubuf + ulen - 1))
2804                         return -EFAULT;
2805                 return -ENOSPC;
2806         }
2807
2808         return 0;
2809 }
2810
2811 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2812         .release = bpf_raw_tp_link_release,
2813         .dealloc = bpf_raw_tp_link_dealloc,
2814         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2815         .fill_link_info = bpf_raw_tp_link_fill_link_info,
2816 };
2817
2818 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2819
2820 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2821 {
2822         struct bpf_link_primer link_primer;
2823         struct bpf_raw_tp_link *link;
2824         struct bpf_raw_event_map *btp;
2825         struct bpf_prog *prog;
2826         const char *tp_name;
2827         char buf[128];
2828         int err;
2829
2830         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2831                 return -EINVAL;
2832
2833         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2834         if (IS_ERR(prog))
2835                 return PTR_ERR(prog);
2836
2837         switch (prog->type) {
2838         case BPF_PROG_TYPE_TRACING:
2839         case BPF_PROG_TYPE_EXT:
2840         case BPF_PROG_TYPE_LSM:
2841                 if (attr->raw_tracepoint.name) {
2842                         /* The attach point for this category of programs
2843                          * should be specified via btf_id during program load.
2844                          */
2845                         err = -EINVAL;
2846                         goto out_put_prog;
2847                 }
2848                 if (prog->type == BPF_PROG_TYPE_TRACING &&
2849                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2850                         tp_name = prog->aux->attach_func_name;
2851                         break;
2852                 }
2853                 err = bpf_tracing_prog_attach(prog, 0, 0);
2854                 if (err >= 0)
2855                         return err;
2856                 goto out_put_prog;
2857         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2858         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2859                 if (strncpy_from_user(buf,
2860                                       u64_to_user_ptr(attr->raw_tracepoint.name),
2861                                       sizeof(buf) - 1) < 0) {
2862                         err = -EFAULT;
2863                         goto out_put_prog;
2864                 }
2865                 buf[sizeof(buf) - 1] = 0;
2866                 tp_name = buf;
2867                 break;
2868         default:
2869                 err = -EINVAL;
2870                 goto out_put_prog;
2871         }
2872
2873         btp = bpf_get_raw_tracepoint(tp_name);
2874         if (!btp) {
2875                 err = -ENOENT;
2876                 goto out_put_prog;
2877         }
2878
2879         link = kzalloc(sizeof(*link), GFP_USER);
2880         if (!link) {
2881                 err = -ENOMEM;
2882                 goto out_put_btp;
2883         }
2884         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2885                       &bpf_raw_tp_link_lops, prog);
2886         link->btp = btp;
2887
2888         err = bpf_link_prime(&link->link, &link_primer);
2889         if (err) {
2890                 kfree(link);
2891                 goto out_put_btp;
2892         }
2893
2894         err = bpf_probe_register(link->btp, prog);
2895         if (err) {
2896                 bpf_link_cleanup(&link_primer);
2897                 goto out_put_btp;
2898         }
2899
2900         return bpf_link_settle(&link_primer);
2901
2902 out_put_btp:
2903         bpf_put_raw_tracepoint(btp);
2904 out_put_prog:
2905         bpf_prog_put(prog);
2906         return err;
2907 }
2908
2909 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2910                                              enum bpf_attach_type attach_type)
2911 {
2912         switch (prog->type) {
2913         case BPF_PROG_TYPE_CGROUP_SOCK:
2914         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2915         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2916         case BPF_PROG_TYPE_SK_LOOKUP:
2917                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2918         case BPF_PROG_TYPE_CGROUP_SKB:
2919                 if (!capable(CAP_NET_ADMIN))
2920                         /* cg-skb progs can be loaded by unpriv user.
2921                          * check permissions at attach time.
2922                          */
2923                         return -EPERM;
2924                 return prog->enforce_expected_attach_type &&
2925                         prog->expected_attach_type != attach_type ?
2926                         -EINVAL : 0;
2927         default:
2928                 return 0;
2929         }
2930 }
2931
2932 static enum bpf_prog_type
2933 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2934 {
2935         switch (attach_type) {
2936         case BPF_CGROUP_INET_INGRESS:
2937         case BPF_CGROUP_INET_EGRESS:
2938                 return BPF_PROG_TYPE_CGROUP_SKB;
2939         case BPF_CGROUP_INET_SOCK_CREATE:
2940         case BPF_CGROUP_INET_SOCK_RELEASE:
2941         case BPF_CGROUP_INET4_POST_BIND:
2942         case BPF_CGROUP_INET6_POST_BIND:
2943                 return BPF_PROG_TYPE_CGROUP_SOCK;
2944         case BPF_CGROUP_INET4_BIND:
2945         case BPF_CGROUP_INET6_BIND:
2946         case BPF_CGROUP_INET4_CONNECT:
2947         case BPF_CGROUP_INET6_CONNECT:
2948         case BPF_CGROUP_INET4_GETPEERNAME:
2949         case BPF_CGROUP_INET6_GETPEERNAME:
2950         case BPF_CGROUP_INET4_GETSOCKNAME:
2951         case BPF_CGROUP_INET6_GETSOCKNAME:
2952         case BPF_CGROUP_UDP4_SENDMSG:
2953         case BPF_CGROUP_UDP6_SENDMSG:
2954         case BPF_CGROUP_UDP4_RECVMSG:
2955         case BPF_CGROUP_UDP6_RECVMSG:
2956                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2957         case BPF_CGROUP_SOCK_OPS:
2958                 return BPF_PROG_TYPE_SOCK_OPS;
2959         case BPF_CGROUP_DEVICE:
2960                 return BPF_PROG_TYPE_CGROUP_DEVICE;
2961         case BPF_SK_MSG_VERDICT:
2962                 return BPF_PROG_TYPE_SK_MSG;
2963         case BPF_SK_SKB_STREAM_PARSER:
2964         case BPF_SK_SKB_STREAM_VERDICT:
2965         case BPF_SK_SKB_VERDICT:
2966                 return BPF_PROG_TYPE_SK_SKB;
2967         case BPF_LIRC_MODE2:
2968                 return BPF_PROG_TYPE_LIRC_MODE2;
2969         case BPF_FLOW_DISSECTOR:
2970                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2971         case BPF_CGROUP_SYSCTL:
2972                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2973         case BPF_CGROUP_GETSOCKOPT:
2974         case BPF_CGROUP_SETSOCKOPT:
2975                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2976         case BPF_TRACE_ITER:
2977                 return BPF_PROG_TYPE_TRACING;
2978         case BPF_SK_LOOKUP:
2979                 return BPF_PROG_TYPE_SK_LOOKUP;
2980         case BPF_XDP:
2981                 return BPF_PROG_TYPE_XDP;
2982         default:
2983                 return BPF_PROG_TYPE_UNSPEC;
2984         }
2985 }
2986
2987 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2988
2989 #define BPF_F_ATTACH_MASK \
2990         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2991
2992 static int bpf_prog_attach(const union bpf_attr *attr)
2993 {
2994         enum bpf_prog_type ptype;
2995         struct bpf_prog *prog;
2996         int ret;
2997
2998         if (CHECK_ATTR(BPF_PROG_ATTACH))
2999                 return -EINVAL;
3000
3001         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
3002                 return -EINVAL;
3003
3004         ptype = attach_type_to_prog_type(attr->attach_type);
3005         if (ptype == BPF_PROG_TYPE_UNSPEC)
3006                 return -EINVAL;
3007
3008         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
3009         if (IS_ERR(prog))
3010                 return PTR_ERR(prog);
3011
3012         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
3013                 bpf_prog_put(prog);
3014                 return -EINVAL;
3015         }
3016
3017         switch (ptype) {
3018         case BPF_PROG_TYPE_SK_SKB:
3019         case BPF_PROG_TYPE_SK_MSG:
3020                 ret = sock_map_get_from_fd(attr, prog);
3021                 break;
3022         case BPF_PROG_TYPE_LIRC_MODE2:
3023                 ret = lirc_prog_attach(attr, prog);
3024                 break;
3025         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3026                 ret = netns_bpf_prog_attach(attr, prog);
3027                 break;
3028         case BPF_PROG_TYPE_CGROUP_DEVICE:
3029         case BPF_PROG_TYPE_CGROUP_SKB:
3030         case BPF_PROG_TYPE_CGROUP_SOCK:
3031         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3032         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3033         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3034         case BPF_PROG_TYPE_SOCK_OPS:
3035                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3036                 break;
3037         default:
3038                 ret = -EINVAL;
3039         }
3040
3041         if (ret)
3042                 bpf_prog_put(prog);
3043         return ret;
3044 }
3045
3046 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3047
3048 static int bpf_prog_detach(const union bpf_attr *attr)
3049 {
3050         enum bpf_prog_type ptype;
3051
3052         if (CHECK_ATTR(BPF_PROG_DETACH))
3053                 return -EINVAL;
3054
3055         ptype = attach_type_to_prog_type(attr->attach_type);
3056
3057         switch (ptype) {
3058         case BPF_PROG_TYPE_SK_MSG:
3059         case BPF_PROG_TYPE_SK_SKB:
3060                 return sock_map_prog_detach(attr, ptype);
3061         case BPF_PROG_TYPE_LIRC_MODE2:
3062                 return lirc_prog_detach(attr);
3063         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3064                 return netns_bpf_prog_detach(attr, ptype);
3065         case BPF_PROG_TYPE_CGROUP_DEVICE:
3066         case BPF_PROG_TYPE_CGROUP_SKB:
3067         case BPF_PROG_TYPE_CGROUP_SOCK:
3068         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3069         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3070         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3071         case BPF_PROG_TYPE_SOCK_OPS:
3072                 return cgroup_bpf_prog_detach(attr, ptype);
3073         default:
3074                 return -EINVAL;
3075         }
3076 }
3077
3078 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3079
3080 static int bpf_prog_query(const union bpf_attr *attr,
3081                           union bpf_attr __user *uattr)
3082 {
3083         if (!capable(CAP_NET_ADMIN))
3084                 return -EPERM;
3085         if (CHECK_ATTR(BPF_PROG_QUERY))
3086                 return -EINVAL;
3087         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3088                 return -EINVAL;
3089
3090         switch (attr->query.attach_type) {
3091         case BPF_CGROUP_INET_INGRESS:
3092         case BPF_CGROUP_INET_EGRESS:
3093         case BPF_CGROUP_INET_SOCK_CREATE:
3094         case BPF_CGROUP_INET_SOCK_RELEASE:
3095         case BPF_CGROUP_INET4_BIND:
3096         case BPF_CGROUP_INET6_BIND:
3097         case BPF_CGROUP_INET4_POST_BIND:
3098         case BPF_CGROUP_INET6_POST_BIND:
3099         case BPF_CGROUP_INET4_CONNECT:
3100         case BPF_CGROUP_INET6_CONNECT:
3101         case BPF_CGROUP_INET4_GETPEERNAME:
3102         case BPF_CGROUP_INET6_GETPEERNAME:
3103         case BPF_CGROUP_INET4_GETSOCKNAME:
3104         case BPF_CGROUP_INET6_GETSOCKNAME:
3105         case BPF_CGROUP_UDP4_SENDMSG:
3106         case BPF_CGROUP_UDP6_SENDMSG:
3107         case BPF_CGROUP_UDP4_RECVMSG:
3108         case BPF_CGROUP_UDP6_RECVMSG:
3109         case BPF_CGROUP_SOCK_OPS:
3110         case BPF_CGROUP_DEVICE:
3111         case BPF_CGROUP_SYSCTL:
3112         case BPF_CGROUP_GETSOCKOPT:
3113         case BPF_CGROUP_SETSOCKOPT:
3114                 return cgroup_bpf_prog_query(attr, uattr);
3115         case BPF_LIRC_MODE2:
3116                 return lirc_prog_query(attr, uattr);
3117         case BPF_FLOW_DISSECTOR:
3118         case BPF_SK_LOOKUP:
3119                 return netns_bpf_prog_query(attr, uattr);
3120         default:
3121                 return -EINVAL;
3122         }
3123 }
3124
3125 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3126
3127 static int bpf_prog_test_run(const union bpf_attr *attr,
3128                              union bpf_attr __user *uattr)
3129 {
3130         struct bpf_prog *prog;
3131         int ret = -ENOTSUPP;
3132
3133         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3134                 return -EINVAL;
3135
3136         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3137             (!attr->test.ctx_size_in && attr->test.ctx_in))
3138                 return -EINVAL;
3139
3140         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3141             (!attr->test.ctx_size_out && attr->test.ctx_out))
3142                 return -EINVAL;
3143
3144         prog = bpf_prog_get(attr->test.prog_fd);
3145         if (IS_ERR(prog))
3146                 return PTR_ERR(prog);
3147
3148         if (prog->aux->ops->test_run)
3149                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3150
3151         bpf_prog_put(prog);
3152         return ret;
3153 }
3154
3155 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3156
3157 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3158                                union bpf_attr __user *uattr,
3159                                struct idr *idr,
3160                                spinlock_t *lock)
3161 {
3162         u32 next_id = attr->start_id;
3163         int err = 0;
3164
3165         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3166                 return -EINVAL;
3167
3168         if (!capable(CAP_SYS_ADMIN))
3169                 return -EPERM;
3170
3171         next_id++;
3172         spin_lock_bh(lock);
3173         if (!idr_get_next(idr, &next_id))
3174                 err = -ENOENT;
3175         spin_unlock_bh(lock);
3176
3177         if (!err)
3178                 err = put_user(next_id, &uattr->next_id);
3179
3180         return err;
3181 }
3182
3183 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3184 {
3185         struct bpf_map *map;
3186
3187         spin_lock_bh(&map_idr_lock);
3188 again:
3189         map = idr_get_next(&map_idr, id);
3190         if (map) {
3191                 map = __bpf_map_inc_not_zero(map, false);
3192                 if (IS_ERR(map)) {
3193                         (*id)++;
3194                         goto again;
3195                 }
3196         }
3197         spin_unlock_bh(&map_idr_lock);
3198
3199         return map;
3200 }
3201
3202 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3203 {
3204         struct bpf_prog *prog;
3205
3206         spin_lock_bh(&prog_idr_lock);
3207 again:
3208         prog = idr_get_next(&prog_idr, id);
3209         if (prog) {
3210                 prog = bpf_prog_inc_not_zero(prog);
3211                 if (IS_ERR(prog)) {
3212                         (*id)++;
3213                         goto again;
3214                 }
3215         }
3216         spin_unlock_bh(&prog_idr_lock);
3217
3218         return prog;
3219 }
3220
3221 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3222
3223 struct bpf_prog *bpf_prog_by_id(u32 id)
3224 {
3225         struct bpf_prog *prog;
3226
3227         if (!id)
3228                 return ERR_PTR(-ENOENT);
3229
3230         spin_lock_bh(&prog_idr_lock);
3231         prog = idr_find(&prog_idr, id);
3232         if (prog)
3233                 prog = bpf_prog_inc_not_zero(prog);
3234         else
3235                 prog = ERR_PTR(-ENOENT);
3236         spin_unlock_bh(&prog_idr_lock);
3237         return prog;
3238 }
3239
3240 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3241 {
3242         struct bpf_prog *prog;
3243         u32 id = attr->prog_id;
3244         int fd;
3245
3246         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3247                 return -EINVAL;
3248
3249         if (!capable(CAP_SYS_ADMIN))
3250                 return -EPERM;
3251
3252         prog = bpf_prog_by_id(id);
3253         if (IS_ERR(prog))
3254                 return PTR_ERR(prog);
3255
3256         fd = bpf_prog_new_fd(prog);
3257         if (fd < 0)
3258                 bpf_prog_put(prog);
3259
3260         return fd;
3261 }
3262
3263 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3264
3265 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3266 {
3267         struct bpf_map *map;
3268         u32 id = attr->map_id;
3269         int f_flags;
3270         int fd;
3271
3272         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3273             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3274                 return -EINVAL;
3275
3276         if (!capable(CAP_SYS_ADMIN))
3277                 return -EPERM;
3278
3279         f_flags = bpf_get_file_flag(attr->open_flags);
3280         if (f_flags < 0)
3281                 return f_flags;
3282
3283         spin_lock_bh(&map_idr_lock);
3284         map = idr_find(&map_idr, id);
3285         if (map)
3286                 map = __bpf_map_inc_not_zero(map, true);
3287         else
3288                 map = ERR_PTR(-ENOENT);
3289         spin_unlock_bh(&map_idr_lock);
3290
3291         if (IS_ERR(map))
3292                 return PTR_ERR(map);
3293
3294         fd = bpf_map_new_fd(map, f_flags);
3295         if (fd < 0)
3296                 bpf_map_put_with_uref(map);
3297
3298         return fd;
3299 }
3300
3301 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3302                                               unsigned long addr, u32 *off,
3303                                               u32 *type)
3304 {
3305         const struct bpf_map *map;
3306         int i;
3307
3308         mutex_lock(&prog->aux->used_maps_mutex);
3309         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3310                 map = prog->aux->used_maps[i];
3311                 if (map == (void *)addr) {
3312                         *type = BPF_PSEUDO_MAP_FD;
3313                         goto out;
3314                 }
3315                 if (!map->ops->map_direct_value_meta)
3316                         continue;
3317                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3318                         *type = BPF_PSEUDO_MAP_VALUE;
3319                         goto out;
3320                 }
3321         }
3322         map = NULL;
3323
3324 out:
3325         mutex_unlock(&prog->aux->used_maps_mutex);
3326         return map;
3327 }
3328
3329 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3330                                               const struct cred *f_cred)
3331 {
3332         const struct bpf_map *map;
3333         struct bpf_insn *insns;
3334         u32 off, type;
3335         u64 imm;
3336         u8 code;
3337         int i;
3338
3339         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3340                         GFP_USER);
3341         if (!insns)
3342                 return insns;
3343
3344         for (i = 0; i < prog->len; i++) {
3345                 code = insns[i].code;
3346
3347                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3348                         insns[i].code = BPF_JMP | BPF_CALL;
3349                         insns[i].imm = BPF_FUNC_tail_call;
3350                         /* fall-through */
3351                 }
3352                 if (code == (BPF_JMP | BPF_CALL) ||
3353                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3354                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3355                                 insns[i].code = BPF_JMP | BPF_CALL;
3356                         if (!bpf_dump_raw_ok(f_cred))
3357                                 insns[i].imm = 0;
3358                         continue;
3359                 }
3360                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3361                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3362                         continue;
3363                 }
3364
3365                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3366                         continue;
3367
3368                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3369                 map = bpf_map_from_imm(prog, imm, &off, &type);
3370                 if (map) {
3371                         insns[i].src_reg = type;
3372                         insns[i].imm = map->id;
3373                         insns[i + 1].imm = off;
3374                         continue;
3375                 }
3376         }
3377
3378         return insns;
3379 }
3380
3381 static int set_info_rec_size(struct bpf_prog_info *info)
3382 {
3383         /*
3384          * Ensure info.*_rec_size is the same as kernel expected size
3385          *
3386          * or
3387          *
3388          * Only allow zero *_rec_size if both _rec_size and _cnt are
3389          * zero.  In this case, the kernel will set the expected
3390          * _rec_size back to the info.
3391          */
3392
3393         if ((info->nr_func_info || info->func_info_rec_size) &&
3394             info->func_info_rec_size != sizeof(struct bpf_func_info))
3395                 return -EINVAL;
3396
3397         if ((info->nr_line_info || info->line_info_rec_size) &&
3398             info->line_info_rec_size != sizeof(struct bpf_line_info))
3399                 return -EINVAL;
3400
3401         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3402             info->jited_line_info_rec_size != sizeof(__u64))
3403                 return -EINVAL;
3404
3405         info->func_info_rec_size = sizeof(struct bpf_func_info);
3406         info->line_info_rec_size = sizeof(struct bpf_line_info);
3407         info->jited_line_info_rec_size = sizeof(__u64);
3408
3409         return 0;
3410 }
3411
3412 static int bpf_prog_get_info_by_fd(struct file *file,
3413                                    struct bpf_prog *prog,
3414                                    const union bpf_attr *attr,
3415                                    union bpf_attr __user *uattr)
3416 {
3417         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3418         struct bpf_prog_info info;
3419         u32 info_len = attr->info.info_len;
3420         struct bpf_prog_stats stats;
3421         char __user *uinsns;
3422         u32 ulen;
3423         int err;
3424
3425         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3426         if (err)
3427                 return err;
3428         info_len = min_t(u32, sizeof(info), info_len);
3429
3430         memset(&info, 0, sizeof(info));
3431         if (copy_from_user(&info, uinfo, info_len))
3432                 return -EFAULT;
3433
3434         info.type = prog->type;
3435         info.id = prog->aux->id;
3436         info.load_time = prog->aux->load_time;
3437         info.created_by_uid = from_kuid_munged(current_user_ns(),
3438                                                prog->aux->user->uid);
3439         info.gpl_compatible = prog->gpl_compatible;
3440
3441         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3442         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3443
3444         mutex_lock(&prog->aux->used_maps_mutex);
3445         ulen = info.nr_map_ids;
3446         info.nr_map_ids = prog->aux->used_map_cnt;
3447         ulen = min_t(u32, info.nr_map_ids, ulen);
3448         if (ulen) {
3449                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3450                 u32 i;
3451
3452                 for (i = 0; i < ulen; i++)
3453                         if (put_user(prog->aux->used_maps[i]->id,
3454                                      &user_map_ids[i])) {
3455                                 mutex_unlock(&prog->aux->used_maps_mutex);
3456                                 return -EFAULT;
3457                         }
3458         }
3459         mutex_unlock(&prog->aux->used_maps_mutex);
3460
3461         err = set_info_rec_size(&info);
3462         if (err)
3463                 return err;
3464
3465         bpf_prog_get_stats(prog, &stats);
3466         info.run_time_ns = stats.nsecs;
3467         info.run_cnt = stats.cnt;
3468         info.recursion_misses = stats.misses;
3469
3470         if (!bpf_capable()) {
3471                 info.jited_prog_len = 0;
3472                 info.xlated_prog_len = 0;
3473                 info.nr_jited_ksyms = 0;
3474                 info.nr_jited_func_lens = 0;
3475                 info.nr_func_info = 0;
3476                 info.nr_line_info = 0;
3477                 info.nr_jited_line_info = 0;
3478                 goto done;
3479         }
3480
3481         ulen = info.xlated_prog_len;
3482         info.xlated_prog_len = bpf_prog_insn_size(prog);
3483         if (info.xlated_prog_len && ulen) {
3484                 struct bpf_insn *insns_sanitized;
3485                 bool fault;
3486
3487                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3488                         info.xlated_prog_insns = 0;
3489                         goto done;
3490                 }
3491                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3492                 if (!insns_sanitized)
3493                         return -ENOMEM;
3494                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3495                 ulen = min_t(u32, info.xlated_prog_len, ulen);
3496                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3497                 kfree(insns_sanitized);
3498                 if (fault)
3499                         return -EFAULT;
3500         }
3501
3502         if (bpf_prog_is_dev_bound(prog->aux)) {
3503                 err = bpf_prog_offload_info_fill(&info, prog);
3504                 if (err)
3505                         return err;
3506                 goto done;
3507         }
3508
3509         /* NOTE: the following code is supposed to be skipped for offload.
3510          * bpf_prog_offload_info_fill() is the place to fill similar fields
3511          * for offload.
3512          */
3513         ulen = info.jited_prog_len;
3514         if (prog->aux->func_cnt) {
3515                 u32 i;
3516
3517                 info.jited_prog_len = 0;
3518                 for (i = 0; i < prog->aux->func_cnt; i++)
3519                         info.jited_prog_len += prog->aux->func[i]->jited_len;
3520         } else {
3521                 info.jited_prog_len = prog->jited_len;
3522         }
3523
3524         if (info.jited_prog_len && ulen) {
3525                 if (bpf_dump_raw_ok(file->f_cred)) {
3526                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
3527                         ulen = min_t(u32, info.jited_prog_len, ulen);
3528
3529                         /* for multi-function programs, copy the JITed
3530                          * instructions for all the functions
3531                          */
3532                         if (prog->aux->func_cnt) {
3533                                 u32 len, free, i;
3534                                 u8 *img;
3535
3536                                 free = ulen;
3537                                 for (i = 0; i < prog->aux->func_cnt; i++) {
3538                                         len = prog->aux->func[i]->jited_len;
3539                                         len = min_t(u32, len, free);
3540                                         img = (u8 *) prog->aux->func[i]->bpf_func;
3541                                         if (copy_to_user(uinsns, img, len))
3542                                                 return -EFAULT;
3543                                         uinsns += len;
3544                                         free -= len;
3545                                         if (!free)
3546                                                 break;
3547                                 }
3548                         } else {
3549                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3550                                         return -EFAULT;
3551                         }
3552                 } else {
3553                         info.jited_prog_insns = 0;
3554                 }
3555         }
3556
3557         ulen = info.nr_jited_ksyms;
3558         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3559         if (ulen) {
3560                 if (bpf_dump_raw_ok(file->f_cred)) {
3561                         unsigned long ksym_addr;
3562                         u64 __user *user_ksyms;
3563                         u32 i;
3564
3565                         /* copy the address of the kernel symbol
3566                          * corresponding to each function
3567                          */
3568                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3569                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3570                         if (prog->aux->func_cnt) {
3571                                 for (i = 0; i < ulen; i++) {
3572                                         ksym_addr = (unsigned long)
3573                                                 prog->aux->func[i]->bpf_func;
3574                                         if (put_user((u64) ksym_addr,
3575                                                      &user_ksyms[i]))
3576                                                 return -EFAULT;
3577                                 }
3578                         } else {
3579                                 ksym_addr = (unsigned long) prog->bpf_func;
3580                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3581                                         return -EFAULT;
3582                         }
3583                 } else {
3584                         info.jited_ksyms = 0;
3585                 }
3586         }
3587
3588         ulen = info.nr_jited_func_lens;
3589         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3590         if (ulen) {
3591                 if (bpf_dump_raw_ok(file->f_cred)) {
3592                         u32 __user *user_lens;
3593                         u32 func_len, i;
3594
3595                         /* copy the JITed image lengths for each function */
3596                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3597                         user_lens = u64_to_user_ptr(info.jited_func_lens);
3598                         if (prog->aux->func_cnt) {
3599                                 for (i = 0; i < ulen; i++) {
3600                                         func_len =
3601                                                 prog->aux->func[i]->jited_len;
3602                                         if (put_user(func_len, &user_lens[i]))
3603                                                 return -EFAULT;
3604                                 }
3605                         } else {
3606                                 func_len = prog->jited_len;
3607                                 if (put_user(func_len, &user_lens[0]))
3608                                         return -EFAULT;
3609                         }
3610                 } else {
3611                         info.jited_func_lens = 0;
3612                 }
3613         }
3614
3615         if (prog->aux->btf)
3616                 info.btf_id = btf_obj_id(prog->aux->btf);
3617
3618         ulen = info.nr_func_info;
3619         info.nr_func_info = prog->aux->func_info_cnt;
3620         if (info.nr_func_info && ulen) {
3621                 char __user *user_finfo;
3622
3623                 user_finfo = u64_to_user_ptr(info.func_info);
3624                 ulen = min_t(u32, info.nr_func_info, ulen);
3625                 if (copy_to_user(user_finfo, prog->aux->func_info,
3626                                  info.func_info_rec_size * ulen))
3627                         return -EFAULT;
3628         }
3629
3630         ulen = info.nr_line_info;
3631         info.nr_line_info = prog->aux->nr_linfo;
3632         if (info.nr_line_info && ulen) {
3633                 __u8 __user *user_linfo;
3634
3635                 user_linfo = u64_to_user_ptr(info.line_info);
3636                 ulen = min_t(u32, info.nr_line_info, ulen);
3637                 if (copy_to_user(user_linfo, prog->aux->linfo,
3638                                  info.line_info_rec_size * ulen))
3639                         return -EFAULT;
3640         }
3641
3642         ulen = info.nr_jited_line_info;
3643         if (prog->aux->jited_linfo)
3644                 info.nr_jited_line_info = prog->aux->nr_linfo;
3645         else
3646                 info.nr_jited_line_info = 0;
3647         if (info.nr_jited_line_info && ulen) {
3648                 if (bpf_dump_raw_ok(file->f_cred)) {
3649                         __u64 __user *user_linfo;
3650                         u32 i;
3651
3652                         user_linfo = u64_to_user_ptr(info.jited_line_info);
3653                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
3654                         for (i = 0; i < ulen; i++) {
3655                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3656                                              &user_linfo[i]))
3657                                         return -EFAULT;
3658                         }
3659                 } else {
3660                         info.jited_line_info = 0;
3661                 }
3662         }
3663
3664         ulen = info.nr_prog_tags;
3665         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3666         if (ulen) {
3667                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3668                 u32 i;
3669
3670                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3671                 ulen = min_t(u32, info.nr_prog_tags, ulen);
3672                 if (prog->aux->func_cnt) {
3673                         for (i = 0; i < ulen; i++) {
3674                                 if (copy_to_user(user_prog_tags[i],
3675                                                  prog->aux->func[i]->tag,
3676                                                  BPF_TAG_SIZE))
3677                                         return -EFAULT;
3678                         }
3679                 } else {
3680                         if (copy_to_user(user_prog_tags[0],
3681                                          prog->tag, BPF_TAG_SIZE))
3682                                 return -EFAULT;
3683                 }
3684         }
3685
3686 done:
3687         if (copy_to_user(uinfo, &info, info_len) ||
3688             put_user(info_len, &uattr->info.info_len))
3689                 return -EFAULT;
3690
3691         return 0;
3692 }
3693
3694 static int bpf_map_get_info_by_fd(struct file *file,
3695                                   struct bpf_map *map,
3696                                   const union bpf_attr *attr,
3697                                   union bpf_attr __user *uattr)
3698 {
3699         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3700         struct bpf_map_info info;
3701         u32 info_len = attr->info.info_len;
3702         int err;
3703
3704         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3705         if (err)
3706                 return err;
3707         info_len = min_t(u32, sizeof(info), info_len);
3708
3709         memset(&info, 0, sizeof(info));
3710         info.type = map->map_type;
3711         info.id = map->id;
3712         info.key_size = map->key_size;
3713         info.value_size = map->value_size;
3714         info.max_entries = map->max_entries;
3715         info.map_flags = map->map_flags;
3716         memcpy(info.name, map->name, sizeof(map->name));
3717
3718         if (map->btf) {
3719                 info.btf_id = btf_obj_id(map->btf);
3720                 info.btf_key_type_id = map->btf_key_type_id;
3721                 info.btf_value_type_id = map->btf_value_type_id;
3722         }
3723         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3724
3725         if (bpf_map_is_dev_bound(map)) {
3726                 err = bpf_map_offload_info_fill(&info, map);
3727                 if (err)
3728                         return err;
3729         }
3730
3731         if (copy_to_user(uinfo, &info, info_len) ||
3732             put_user(info_len, &uattr->info.info_len))
3733                 return -EFAULT;
3734
3735         return 0;
3736 }
3737
3738 static int bpf_btf_get_info_by_fd(struct file *file,
3739                                   struct btf *btf,
3740                                   const union bpf_attr *attr,
3741                                   union bpf_attr __user *uattr)
3742 {
3743         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3744         u32 info_len = attr->info.info_len;
3745         int err;
3746
3747         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3748         if (err)
3749                 return err;
3750
3751         return btf_get_info_by_fd(btf, attr, uattr);
3752 }
3753
3754 static int bpf_link_get_info_by_fd(struct file *file,
3755                                   struct bpf_link *link,
3756                                   const union bpf_attr *attr,
3757                                   union bpf_attr __user *uattr)
3758 {
3759         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3760         struct bpf_link_info info;
3761         u32 info_len = attr->info.info_len;
3762         int err;
3763
3764         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3765         if (err)
3766                 return err;
3767         info_len = min_t(u32, sizeof(info), info_len);
3768
3769         memset(&info, 0, sizeof(info));
3770         if (copy_from_user(&info, uinfo, info_len))
3771                 return -EFAULT;
3772
3773         info.type = link->type;
3774         info.id = link->id;
3775         info.prog_id = link->prog->aux->id;
3776
3777         if (link->ops->fill_link_info) {
3778                 err = link->ops->fill_link_info(link, &info);
3779                 if (err)
3780                         return err;
3781         }
3782
3783         if (copy_to_user(uinfo, &info, info_len) ||
3784             put_user(info_len, &uattr->info.info_len))
3785                 return -EFAULT;
3786
3787         return 0;
3788 }
3789
3790
3791 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3792
3793 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3794                                   union bpf_attr __user *uattr)
3795 {
3796         int ufd = attr->info.bpf_fd;
3797         struct fd f;
3798         int err;
3799
3800         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3801                 return -EINVAL;
3802
3803         f = fdget(ufd);
3804         if (!f.file)
3805                 return -EBADFD;
3806
3807         if (f.file->f_op == &bpf_prog_fops)
3808                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3809                                               uattr);
3810         else if (f.file->f_op == &bpf_map_fops)
3811                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3812                                              uattr);
3813         else if (f.file->f_op == &btf_fops)
3814                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3815         else if (f.file->f_op == &bpf_link_fops)
3816                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3817                                               attr, uattr);
3818         else
3819                 err = -EINVAL;
3820
3821         fdput(f);
3822         return err;
3823 }
3824
3825 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3826
3827 static int bpf_btf_load(const union bpf_attr *attr)
3828 {
3829         if (CHECK_ATTR(BPF_BTF_LOAD))
3830                 return -EINVAL;
3831
3832         if (!bpf_capable())
3833                 return -EPERM;
3834
3835         return btf_new_fd(attr);
3836 }
3837
3838 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3839
3840 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3841 {
3842         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3843                 return -EINVAL;
3844
3845         if (!capable(CAP_SYS_ADMIN))
3846                 return -EPERM;
3847
3848         return btf_get_fd_by_id(attr->btf_id);
3849 }
3850
3851 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3852                                     union bpf_attr __user *uattr,
3853                                     u32 prog_id, u32 fd_type,
3854                                     const char *buf, u64 probe_offset,
3855                                     u64 probe_addr)
3856 {
3857         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3858         u32 len = buf ? strlen(buf) : 0, input_len;
3859         int err = 0;
3860
3861         if (put_user(len, &uattr->task_fd_query.buf_len))
3862                 return -EFAULT;
3863         input_len = attr->task_fd_query.buf_len;
3864         if (input_len && ubuf) {
3865                 if (!len) {
3866                         /* nothing to copy, just make ubuf NULL terminated */
3867                         char zero = '\0';
3868
3869                         if (put_user(zero, ubuf))
3870                                 return -EFAULT;
3871                 } else if (input_len >= len + 1) {
3872                         /* ubuf can hold the string with NULL terminator */
3873                         if (copy_to_user(ubuf, buf, len + 1))
3874                                 return -EFAULT;
3875                 } else {
3876                         /* ubuf cannot hold the string with NULL terminator,
3877                          * do a partial copy with NULL terminator.
3878                          */
3879                         char zero = '\0';
3880
3881                         err = -ENOSPC;
3882                         if (copy_to_user(ubuf, buf, input_len - 1))
3883                                 return -EFAULT;
3884                         if (put_user(zero, ubuf + input_len - 1))
3885                                 return -EFAULT;
3886                 }
3887         }
3888
3889         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3890             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3891             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3892             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3893                 return -EFAULT;
3894
3895         return err;
3896 }
3897
3898 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3899
3900 static int bpf_task_fd_query(const union bpf_attr *attr,
3901                              union bpf_attr __user *uattr)
3902 {
3903         pid_t pid = attr->task_fd_query.pid;
3904         u32 fd = attr->task_fd_query.fd;
3905         const struct perf_event *event;
3906         struct task_struct *task;
3907         struct file *file;
3908         int err;
3909
3910         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3911                 return -EINVAL;
3912
3913         if (!capable(CAP_SYS_ADMIN))
3914                 return -EPERM;
3915
3916         if (attr->task_fd_query.flags != 0)
3917                 return -EINVAL;
3918
3919         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3920         if (!task)
3921                 return -ENOENT;
3922
3923         err = 0;
3924         file = fget_task(task, fd);
3925         put_task_struct(task);
3926         if (!file)
3927                 return -EBADF;
3928
3929         if (file->f_op == &bpf_link_fops) {
3930                 struct bpf_link *link = file->private_data;
3931
3932                 if (link->ops == &bpf_raw_tp_link_lops) {
3933                         struct bpf_raw_tp_link *raw_tp =
3934                                 container_of(link, struct bpf_raw_tp_link, link);
3935                         struct bpf_raw_event_map *btp = raw_tp->btp;
3936
3937                         err = bpf_task_fd_query_copy(attr, uattr,
3938                                                      raw_tp->link.prog->aux->id,
3939                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
3940                                                      btp->tp->name, 0, 0);
3941                         goto put_file;
3942                 }
3943                 goto out_not_supp;
3944         }
3945
3946         event = perf_get_event(file);
3947         if (!IS_ERR(event)) {
3948                 u64 probe_offset, probe_addr;
3949                 u32 prog_id, fd_type;
3950                 const char *buf;
3951
3952                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3953                                               &buf, &probe_offset,
3954                                               &probe_addr);
3955                 if (!err)
3956                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3957                                                      fd_type, buf,
3958                                                      probe_offset,
3959                                                      probe_addr);
3960                 goto put_file;
3961         }
3962
3963 out_not_supp:
3964         err = -ENOTSUPP;
3965 put_file:
3966         fput(file);
3967         return err;
3968 }
3969
3970 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3971
3972 #define BPF_DO_BATCH(fn)                        \
3973         do {                                    \
3974                 if (!fn) {                      \
3975                         err = -ENOTSUPP;        \
3976                         goto err_put;           \
3977                 }                               \
3978                 err = fn(map, attr, uattr);     \
3979         } while (0)
3980
3981 static int bpf_map_do_batch(const union bpf_attr *attr,
3982                             union bpf_attr __user *uattr,
3983                             int cmd)
3984 {
3985         struct bpf_map *map;
3986         int err, ufd;
3987         struct fd f;
3988
3989         if (CHECK_ATTR(BPF_MAP_BATCH))
3990                 return -EINVAL;
3991
3992         ufd = attr->batch.map_fd;
3993         f = fdget(ufd);
3994         map = __bpf_map_get(f);
3995         if (IS_ERR(map))
3996                 return PTR_ERR(map);
3997
3998         if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3999              cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
4000             !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
4001                 err = -EPERM;
4002                 goto err_put;
4003         }
4004
4005         if (cmd != BPF_MAP_LOOKUP_BATCH &&
4006             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
4007                 err = -EPERM;
4008                 goto err_put;
4009         }
4010
4011         if (cmd == BPF_MAP_LOOKUP_BATCH)
4012                 BPF_DO_BATCH(map->ops->map_lookup_batch);
4013         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
4014                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
4015         else if (cmd == BPF_MAP_UPDATE_BATCH)
4016                 BPF_DO_BATCH(map->ops->map_update_batch);
4017         else
4018                 BPF_DO_BATCH(map->ops->map_delete_batch);
4019
4020 err_put:
4021         fdput(f);
4022         return err;
4023 }
4024
4025 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4026 {
4027         if (attr->link_create.attach_type != prog->expected_attach_type)
4028                 return -EINVAL;
4029
4030         if (prog->expected_attach_type == BPF_TRACE_ITER)
4031                 return bpf_iter_link_attach(attr, prog);
4032         else if (prog->type == BPF_PROG_TYPE_EXT)
4033                 return bpf_tracing_prog_attach(prog,
4034                                                attr->link_create.target_fd,
4035                                                attr->link_create.target_btf_id);
4036         return -EINVAL;
4037 }
4038
4039 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4040 static int link_create(union bpf_attr *attr)
4041 {
4042         enum bpf_prog_type ptype;
4043         struct bpf_prog *prog;
4044         int ret;
4045
4046         if (CHECK_ATTR(BPF_LINK_CREATE))
4047                 return -EINVAL;
4048
4049         prog = bpf_prog_get(attr->link_create.prog_fd);
4050         if (IS_ERR(prog))
4051                 return PTR_ERR(prog);
4052
4053         ret = bpf_prog_attach_check_attach_type(prog,
4054                                                 attr->link_create.attach_type);
4055         if (ret)
4056                 goto out;
4057
4058         if (prog->type == BPF_PROG_TYPE_EXT) {
4059                 ret = tracing_bpf_link_attach(attr, prog);
4060                 goto out;
4061         }
4062
4063         ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4064         if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4065                 ret = -EINVAL;
4066                 goto out;
4067         }
4068
4069         switch (ptype) {
4070         case BPF_PROG_TYPE_CGROUP_SKB:
4071         case BPF_PROG_TYPE_CGROUP_SOCK:
4072         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4073         case BPF_PROG_TYPE_SOCK_OPS:
4074         case BPF_PROG_TYPE_CGROUP_DEVICE:
4075         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4076         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4077                 ret = cgroup_bpf_link_attach(attr, prog);
4078                 break;
4079         case BPF_PROG_TYPE_TRACING:
4080                 ret = tracing_bpf_link_attach(attr, prog);
4081                 break;
4082         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4083         case BPF_PROG_TYPE_SK_LOOKUP:
4084                 ret = netns_bpf_link_create(attr, prog);
4085                 break;
4086 #ifdef CONFIG_NET
4087         case BPF_PROG_TYPE_XDP:
4088                 ret = bpf_xdp_link_attach(attr, prog);
4089                 break;
4090 #endif
4091         default:
4092                 ret = -EINVAL;
4093         }
4094
4095 out:
4096         if (ret < 0)
4097                 bpf_prog_put(prog);
4098         return ret;
4099 }
4100
4101 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4102
4103 static int link_update(union bpf_attr *attr)
4104 {
4105         struct bpf_prog *old_prog = NULL, *new_prog;
4106         struct bpf_link *link;
4107         u32 flags;
4108         int ret;
4109
4110         if (CHECK_ATTR(BPF_LINK_UPDATE))
4111                 return -EINVAL;
4112
4113         flags = attr->link_update.flags;
4114         if (flags & ~BPF_F_REPLACE)
4115                 return -EINVAL;
4116
4117         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4118         if (IS_ERR(link))
4119                 return PTR_ERR(link);
4120
4121         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4122         if (IS_ERR(new_prog)) {
4123                 ret = PTR_ERR(new_prog);
4124                 goto out_put_link;
4125         }
4126
4127         if (flags & BPF_F_REPLACE) {
4128                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4129                 if (IS_ERR(old_prog)) {
4130                         ret = PTR_ERR(old_prog);
4131                         old_prog = NULL;
4132                         goto out_put_progs;
4133                 }
4134         } else if (attr->link_update.old_prog_fd) {
4135                 ret = -EINVAL;
4136                 goto out_put_progs;
4137         }
4138
4139         if (link->ops->update_prog)
4140                 ret = link->ops->update_prog(link, new_prog, old_prog);
4141         else
4142                 ret = -EINVAL;
4143
4144 out_put_progs:
4145         if (old_prog)
4146                 bpf_prog_put(old_prog);
4147         if (ret)
4148                 bpf_prog_put(new_prog);
4149 out_put_link:
4150         bpf_link_put(link);
4151         return ret;
4152 }
4153
4154 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4155
4156 static int link_detach(union bpf_attr *attr)
4157 {
4158         struct bpf_link *link;
4159         int ret;
4160
4161         if (CHECK_ATTR(BPF_LINK_DETACH))
4162                 return -EINVAL;
4163
4164         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4165         if (IS_ERR(link))
4166                 return PTR_ERR(link);
4167
4168         if (link->ops->detach)
4169                 ret = link->ops->detach(link);
4170         else
4171                 ret = -EOPNOTSUPP;
4172
4173         bpf_link_put(link);
4174         return ret;
4175 }
4176
4177 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4178 {
4179         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4180 }
4181
4182 struct bpf_link *bpf_link_by_id(u32 id)
4183 {
4184         struct bpf_link *link;
4185
4186         if (!id)
4187                 return ERR_PTR(-ENOENT);
4188
4189         spin_lock_bh(&link_idr_lock);
4190         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4191         link = idr_find(&link_idr, id);
4192         if (link) {
4193                 if (link->id)
4194                         link = bpf_link_inc_not_zero(link);
4195                 else
4196                         link = ERR_PTR(-EAGAIN);
4197         } else {
4198                 link = ERR_PTR(-ENOENT);
4199         }
4200         spin_unlock_bh(&link_idr_lock);
4201         return link;
4202 }
4203
4204 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4205
4206 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4207 {
4208         struct bpf_link *link;
4209         u32 id = attr->link_id;
4210         int fd;
4211
4212         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4213                 return -EINVAL;
4214
4215         if (!capable(CAP_SYS_ADMIN))
4216                 return -EPERM;
4217
4218         link = bpf_link_by_id(id);
4219         if (IS_ERR(link))
4220                 return PTR_ERR(link);
4221
4222         fd = bpf_link_new_fd(link);
4223         if (fd < 0)
4224                 bpf_link_put(link);
4225
4226         return fd;
4227 }
4228
4229 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4230
4231 static int bpf_stats_release(struct inode *inode, struct file *file)
4232 {
4233         mutex_lock(&bpf_stats_enabled_mutex);
4234         static_key_slow_dec(&bpf_stats_enabled_key.key);
4235         mutex_unlock(&bpf_stats_enabled_mutex);
4236         return 0;
4237 }
4238
4239 static const struct file_operations bpf_stats_fops = {
4240         .release = bpf_stats_release,
4241 };
4242
4243 static int bpf_enable_runtime_stats(void)
4244 {
4245         int fd;
4246
4247         mutex_lock(&bpf_stats_enabled_mutex);
4248
4249         /* Set a very high limit to avoid overflow */
4250         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4251                 mutex_unlock(&bpf_stats_enabled_mutex);
4252                 return -EBUSY;
4253         }
4254
4255         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4256         if (fd >= 0)
4257                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4258
4259         mutex_unlock(&bpf_stats_enabled_mutex);
4260         return fd;
4261 }
4262
4263 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4264
4265 static int bpf_enable_stats(union bpf_attr *attr)
4266 {
4267
4268         if (CHECK_ATTR(BPF_ENABLE_STATS))
4269                 return -EINVAL;
4270
4271         if (!capable(CAP_SYS_ADMIN))
4272                 return -EPERM;
4273
4274         switch (attr->enable_stats.type) {
4275         case BPF_STATS_RUN_TIME:
4276                 return bpf_enable_runtime_stats();
4277         default:
4278                 break;
4279         }
4280         return -EINVAL;
4281 }
4282
4283 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4284
4285 static int bpf_iter_create(union bpf_attr *attr)
4286 {
4287         struct bpf_link *link;
4288         int err;
4289
4290         if (CHECK_ATTR(BPF_ITER_CREATE))
4291                 return -EINVAL;
4292
4293         if (attr->iter_create.flags)
4294                 return -EINVAL;
4295
4296         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4297         if (IS_ERR(link))
4298                 return PTR_ERR(link);
4299
4300         err = bpf_iter_new_fd(link);
4301         bpf_link_put(link);
4302
4303         return err;
4304 }
4305
4306 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4307
4308 static int bpf_prog_bind_map(union bpf_attr *attr)
4309 {
4310         struct bpf_prog *prog;
4311         struct bpf_map *map;
4312         struct bpf_map **used_maps_old, **used_maps_new;
4313         int i, ret = 0;
4314
4315         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4316                 return -EINVAL;
4317
4318         if (attr->prog_bind_map.flags)
4319                 return -EINVAL;
4320
4321         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4322         if (IS_ERR(prog))
4323                 return PTR_ERR(prog);
4324
4325         map = bpf_map_get(attr->prog_bind_map.map_fd);
4326         if (IS_ERR(map)) {
4327                 ret = PTR_ERR(map);
4328                 goto out_prog_put;
4329         }
4330
4331         mutex_lock(&prog->aux->used_maps_mutex);
4332
4333         used_maps_old = prog->aux->used_maps;
4334
4335         for (i = 0; i < prog->aux->used_map_cnt; i++)
4336                 if (used_maps_old[i] == map) {
4337                         bpf_map_put(map);
4338                         goto out_unlock;
4339                 }
4340
4341         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4342                                       sizeof(used_maps_new[0]),
4343                                       GFP_KERNEL);
4344         if (!used_maps_new) {
4345                 ret = -ENOMEM;
4346                 goto out_unlock;
4347         }
4348
4349         memcpy(used_maps_new, used_maps_old,
4350                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4351         used_maps_new[prog->aux->used_map_cnt] = map;
4352
4353         prog->aux->used_map_cnt++;
4354         prog->aux->used_maps = used_maps_new;
4355
4356         kfree(used_maps_old);
4357
4358 out_unlock:
4359         mutex_unlock(&prog->aux->used_maps_mutex);
4360
4361         if (ret)
4362                 bpf_map_put(map);
4363 out_prog_put:
4364         bpf_prog_put(prog);
4365         return ret;
4366 }
4367
4368 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4369 {
4370         union bpf_attr attr;
4371         int err;
4372
4373         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4374                 return -EPERM;
4375
4376         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4377         if (err)
4378                 return err;
4379         size = min_t(u32, size, sizeof(attr));
4380
4381         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4382         memset(&attr, 0, sizeof(attr));
4383         if (copy_from_user(&attr, uattr, size) != 0)
4384                 return -EFAULT;
4385
4386         err = security_bpf(cmd, &attr, size);
4387         if (err < 0)
4388                 return err;
4389
4390         switch (cmd) {
4391         case BPF_MAP_CREATE:
4392                 err = map_create(&attr);
4393                 break;
4394         case BPF_MAP_LOOKUP_ELEM:
4395                 err = map_lookup_elem(&attr);
4396                 break;
4397         case BPF_MAP_UPDATE_ELEM:
4398                 err = map_update_elem(&attr);
4399                 break;
4400         case BPF_MAP_DELETE_ELEM:
4401                 err = map_delete_elem(&attr);
4402                 break;
4403         case BPF_MAP_GET_NEXT_KEY:
4404                 err = map_get_next_key(&attr);
4405                 break;
4406         case BPF_MAP_FREEZE:
4407                 err = map_freeze(&attr);
4408                 break;
4409         case BPF_PROG_LOAD:
4410                 err = bpf_prog_load(&attr, uattr);
4411                 break;
4412         case BPF_OBJ_PIN:
4413                 err = bpf_obj_pin(&attr);
4414                 break;
4415         case BPF_OBJ_GET:
4416                 err = bpf_obj_get(&attr);
4417                 break;
4418         case BPF_PROG_ATTACH:
4419                 err = bpf_prog_attach(&attr);
4420                 break;
4421         case BPF_PROG_DETACH:
4422                 err = bpf_prog_detach(&attr);
4423                 break;
4424         case BPF_PROG_QUERY:
4425                 err = bpf_prog_query(&attr, uattr);
4426                 break;
4427         case BPF_PROG_TEST_RUN:
4428                 err = bpf_prog_test_run(&attr, uattr);
4429                 break;
4430         case BPF_PROG_GET_NEXT_ID:
4431                 err = bpf_obj_get_next_id(&attr, uattr,
4432                                           &prog_idr, &prog_idr_lock);
4433                 break;
4434         case BPF_MAP_GET_NEXT_ID:
4435                 err = bpf_obj_get_next_id(&attr, uattr,
4436                                           &map_idr, &map_idr_lock);
4437                 break;
4438         case BPF_BTF_GET_NEXT_ID:
4439                 err = bpf_obj_get_next_id(&attr, uattr,
4440                                           &btf_idr, &btf_idr_lock);
4441                 break;
4442         case BPF_PROG_GET_FD_BY_ID:
4443                 err = bpf_prog_get_fd_by_id(&attr);
4444                 break;
4445         case BPF_MAP_GET_FD_BY_ID:
4446                 err = bpf_map_get_fd_by_id(&attr);
4447                 break;
4448         case BPF_OBJ_GET_INFO_BY_FD:
4449                 err = bpf_obj_get_info_by_fd(&attr, uattr);
4450                 break;
4451         case BPF_RAW_TRACEPOINT_OPEN:
4452                 err = bpf_raw_tracepoint_open(&attr);
4453                 break;
4454         case BPF_BTF_LOAD:
4455                 err = bpf_btf_load(&attr);
4456                 break;
4457         case BPF_BTF_GET_FD_BY_ID:
4458                 err = bpf_btf_get_fd_by_id(&attr);
4459                 break;
4460         case BPF_TASK_FD_QUERY:
4461                 err = bpf_task_fd_query(&attr, uattr);
4462                 break;
4463         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4464                 err = map_lookup_and_delete_elem(&attr);
4465                 break;
4466         case BPF_MAP_LOOKUP_BATCH:
4467                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4468                 break;
4469         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4470                 err = bpf_map_do_batch(&attr, uattr,
4471                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4472                 break;
4473         case BPF_MAP_UPDATE_BATCH:
4474                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4475                 break;
4476         case BPF_MAP_DELETE_BATCH:
4477                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4478                 break;
4479         case BPF_LINK_CREATE:
4480                 err = link_create(&attr);
4481                 break;
4482         case BPF_LINK_UPDATE:
4483                 err = link_update(&attr);
4484                 break;
4485         case BPF_LINK_GET_FD_BY_ID:
4486                 err = bpf_link_get_fd_by_id(&attr);
4487                 break;
4488         case BPF_LINK_GET_NEXT_ID:
4489                 err = bpf_obj_get_next_id(&attr, uattr,
4490                                           &link_idr, &link_idr_lock);
4491                 break;
4492         case BPF_ENABLE_STATS:
4493                 err = bpf_enable_stats(&attr);
4494                 break;
4495         case BPF_ITER_CREATE:
4496                 err = bpf_iter_create(&attr);
4497                 break;
4498         case BPF_LINK_DETACH:
4499                 err = link_detach(&attr);
4500                 break;
4501         case BPF_PROG_BIND_MAP:
4502                 err = bpf_prog_bind_map(&attr);
4503                 break;
4504         default:
4505                 err = -EINVAL;
4506                 break;
4507         }
4508
4509         return err;
4510 }