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                 map->btf = btf;
 858
 859                 if (attr->btf_value_type_id) {
 860                         err = map_check_btf(map, btf, attr->btf_key_type_id,
 861                                             attr->btf_value_type_id);
 862                         if (err)
 863                                 goto free_map;
 864                 }
 865
 866                 map->btf_key_type_id = attr->btf_key_type_id;
 867                 map->btf_value_type_id = attr->btf_value_type_id;
 868                 map->btf_vmlinux_value_type_id =
 869                         attr->btf_vmlinux_value_type_id;
 870         }
 871
 872         err = security_bpf_map_alloc(map);
 873         if (err)
 874                 goto free_map;
 875
 876         err = bpf_map_alloc_id(map);
 877         if (err)
 878                 goto free_map_sec;
 879
 880         bpf_map_save_memcg(map);
 881
 882         err = bpf_map_new_fd(map, f_flags);
 883         if (err < 0) {
 884                 /* failed to allocate fd.
 885                  * bpf_map_put_with_uref() is needed because the above
 886                  * bpf_map_alloc_id() has published the map
 887                  * to the userspace and the userspace may
 888                  * have refcnt-ed it through BPF_MAP_GET_FD_BY_ID.
 889                  */
 890                 bpf_map_put_with_uref(map);
 891                 return err;
 892         }
 893
 894         return err;
 895
 896 free_map_sec:
 897         security_bpf_map_free(map);
 898 free_map:
 899         btf_put(map->btf);
 900         map->ops->map_free(map);
 901         return err;
 902 }
 903
 904 /* if error is returned, fd is released.
 905  * On success caller should complete fd access with matching fdput()
 906  */
 907 struct bpf_map *__bpf_map_get(struct fd f)
 908 {
 909         if (!f.file)
 910                 return ERR_PTR(-EBADF);
 911         if (f.file->f_op != &bpf_map_fops) {
 912                 fdput(f);
 913                 return ERR_PTR(-EINVAL);
 914         }
 915
 916         return f.file->private_data;
 917 }
 918
 919 void bpf_map_inc(struct bpf_map *map)
 920 {
 921         atomic64_inc(&map->refcnt);
 922 }
 923 EXPORT_SYMBOL_GPL(bpf_map_inc);
 924
 925 void bpf_map_inc_with_uref(struct bpf_map *map)
 926 {
 927         atomic64_inc(&map->refcnt);
 928         atomic64_inc(&map->usercnt);
 929 }
 930 EXPORT_SYMBOL_GPL(bpf_map_inc_with_uref);
 931
 932 struct bpf_map *bpf_map_get(u32 ufd)
 933 {
 934         struct fd f = fdget(ufd);
 935         struct bpf_map *map;
 936
 937         map = __bpf_map_get(f);
 938         if (IS_ERR(map))
 939                 return map;
 940
 941         bpf_map_inc(map);
 942         fdput(f);
 943
 944         return map;
 945 }
 946
 947 struct bpf_map *bpf_map_get_with_uref(u32 ufd)
 948 {
 949         struct fd f = fdget(ufd);
 950         struct bpf_map *map;
 951
 952         map = __bpf_map_get(f);
 953         if (IS_ERR(map))
 954                 return map;
 955
 956         bpf_map_inc_with_uref(map);
 957         fdput(f);
 958
 959         return map;
 960 }
 961
 962 /* map_idr_lock should have been held */
 963 static struct bpf_map *__bpf_map_inc_not_zero(struct bpf_map *map, bool uref)
 964 {
 965         int refold;
 966
 967         refold = atomic64_fetch_add_unless(&map->refcnt, 1, 0);
 968         if (!refold)
 969                 return ERR_PTR(-ENOENT);
 970         if (uref)
 971                 atomic64_inc(&map->usercnt);
 972
 973         return map;
 974 }
 975
 976 struct bpf_map *bpf_map_inc_not_zero(struct bpf_map *map)
 977 {
 978         spin_lock_bh(&map_idr_lock);
 979         map = __bpf_map_inc_not_zero(map, false);
 980         spin_unlock_bh(&map_idr_lock);
 981
 982         return map;
 983 }
 984 EXPORT_SYMBOL_GPL(bpf_map_inc_not_zero);
 985
 986 int __weak bpf_stackmap_copy(struct bpf_map *map, void *key, void *value)
 987 {
 988         return -ENOTSUPP;
 989 }
 990
 991 static void *__bpf_copy_key(void __user *ukey, u64 key_size)
 992 {
 993         if (key_size)
 994                 return memdup_user(ukey, key_size);
 995
 996         if (ukey)
 997                 return ERR_PTR(-EINVAL);
 998
 999         return NULL;
1000 }
1001
1002 /* last field in 'union bpf_attr' used by this command */
1003 #define BPF_MAP_LOOKUP_ELEM_LAST_FIELD flags
1004
1005 static int map_lookup_elem(union bpf_attr *attr)
1006 {
1007         void __user *ukey = u64_to_user_ptr(attr->key);
1008         void __user *uvalue = u64_to_user_ptr(attr->value);
1009         int ufd = attr->map_fd;
1010         struct bpf_map *map;
1011         void *key, *value;
1012         u32 value_size;
1013         struct fd f;
1014         int err;
1015
1016         if (CHECK_ATTR(BPF_MAP_LOOKUP_ELEM))
1017                 return -EINVAL;
1018
1019         if (attr->flags & ~BPF_F_LOCK)
1020                 return -EINVAL;
1021
1022         f = fdget(ufd);
1023         map = __bpf_map_get(f);
1024         if (IS_ERR(map))
1025                 return PTR_ERR(map);
1026         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1027                 err = -EPERM;
1028                 goto err_put;
1029         }
1030
1031         if ((attr->flags & BPF_F_LOCK) &&
1032             !map_value_has_spin_lock(map)) {
1033                 err = -EINVAL;
1034                 goto err_put;
1035         }
1036
1037         key = __bpf_copy_key(ukey, map->key_size);
1038         if (IS_ERR(key)) {
1039                 err = PTR_ERR(key);
1040                 goto err_put;
1041         }
1042
1043         value_size = bpf_map_value_size(map);
1044
1045         err = -ENOMEM;
1046         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1047         if (!value)
1048                 goto free_key;
1049
1050         err = bpf_map_copy_value(map, key, value, attr->flags);
1051         if (err)
1052                 goto free_value;
1053
1054         err = -EFAULT;
1055         if (copy_to_user(uvalue, value, value_size) != 0)
1056                 goto free_value;
1057
1058         err = 0;
1059
1060 free_value:
1061         kfree(value);
1062 free_key:
1063         kfree(key);
1064 err_put:
1065         fdput(f);
1066         return err;
1067 }
1068
1069
1070 #define BPF_MAP_UPDATE_ELEM_LAST_FIELD flags
1071
1072 static int map_update_elem(union bpf_attr *attr)
1073 {
1074         void __user *ukey = u64_to_user_ptr(attr->key);
1075         void __user *uvalue = u64_to_user_ptr(attr->value);
1076         int ufd = attr->map_fd;
1077         struct bpf_map *map;
1078         void *key, *value;
1079         u32 value_size;
1080         struct fd f;
1081         int err;
1082
1083         if (CHECK_ATTR(BPF_MAP_UPDATE_ELEM))
1084                 return -EINVAL;
1085
1086         f = fdget(ufd);
1087         map = __bpf_map_get(f);
1088         if (IS_ERR(map))
1089                 return PTR_ERR(map);
1090         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1091                 err = -EPERM;
1092                 goto err_put;
1093         }
1094
1095         if ((attr->flags & BPF_F_LOCK) &&
1096             !map_value_has_spin_lock(map)) {
1097                 err = -EINVAL;
1098                 goto err_put;
1099         }
1100
1101         key = __bpf_copy_key(ukey, map->key_size);
1102         if (IS_ERR(key)) {
1103                 err = PTR_ERR(key);
1104                 goto err_put;
1105         }
1106
1107         if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
1108             map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH ||
1109             map->map_type == BPF_MAP_TYPE_PERCPU_ARRAY ||
1110             map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
1111                 value_size = round_up(map->value_size, 8) * num_possible_cpus();
1112         else
1113                 value_size = map->value_size;
1114
1115         err = -ENOMEM;
1116         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1117         if (!value)
1118                 goto free_key;
1119
1120         err = -EFAULT;
1121         if (copy_from_user(value, uvalue, value_size) != 0)
1122                 goto free_value;
1123
1124         err = bpf_map_update_value(map, f, key, value, attr->flags);
1125
1126 free_value:
1127         kfree(value);
1128 free_key:
1129         kfree(key);
1130 err_put:
1131         fdput(f);
1132         return err;
1133 }
1134
1135 #define BPF_MAP_DELETE_ELEM_LAST_FIELD key
1136
1137 static int map_delete_elem(union bpf_attr *attr)
1138 {
1139         void __user *ukey = u64_to_user_ptr(attr->key);
1140         int ufd = attr->map_fd;
1141         struct bpf_map *map;
1142         struct fd f;
1143         void *key;
1144         int err;
1145
1146         if (CHECK_ATTR(BPF_MAP_DELETE_ELEM))
1147                 return -EINVAL;
1148
1149         f = fdget(ufd);
1150         map = __bpf_map_get(f);
1151         if (IS_ERR(map))
1152                 return PTR_ERR(map);
1153         if (!(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1154                 err = -EPERM;
1155                 goto err_put;
1156         }
1157
1158         key = __bpf_copy_key(ukey, map->key_size);
1159         if (IS_ERR(key)) {
1160                 err = PTR_ERR(key);
1161                 goto err_put;
1162         }
1163
1164         if (bpf_map_is_dev_bound(map)) {
1165                 err = bpf_map_offload_delete_elem(map, key);
1166                 goto out;
1167         } else if (IS_FD_PROG_ARRAY(map) ||
1168                    map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1169                 /* These maps require sleepable context */
1170                 err = map->ops->map_delete_elem(map, key);
1171                 goto out;
1172         }
1173
1174         bpf_disable_instrumentation();
1175         rcu_read_lock();
1176         err = map->ops->map_delete_elem(map, key);
1177         rcu_read_unlock();
1178         bpf_enable_instrumentation();
1179         maybe_wait_bpf_programs(map);
1180 out:
1181         kfree(key);
1182 err_put:
1183         fdput(f);
1184         return err;
1185 }
1186
1187 /* last field in 'union bpf_attr' used by this command */
1188 #define BPF_MAP_GET_NEXT_KEY_LAST_FIELD next_key
1189
1190 static int map_get_next_key(union bpf_attr *attr)
1191 {
1192         void __user *ukey = u64_to_user_ptr(attr->key);
1193         void __user *unext_key = u64_to_user_ptr(attr->next_key);
1194         int ufd = attr->map_fd;
1195         struct bpf_map *map;
1196         void *key, *next_key;
1197         struct fd f;
1198         int err;
1199
1200         if (CHECK_ATTR(BPF_MAP_GET_NEXT_KEY))
1201                 return -EINVAL;
1202
1203         f = fdget(ufd);
1204         map = __bpf_map_get(f);
1205         if (IS_ERR(map))
1206                 return PTR_ERR(map);
1207         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
1208                 err = -EPERM;
1209                 goto err_put;
1210         }
1211
1212         if (ukey) {
1213                 key = __bpf_copy_key(ukey, map->key_size);
1214                 if (IS_ERR(key)) {
1215                         err = PTR_ERR(key);
1216                         goto err_put;
1217                 }
1218         } else {
1219                 key = NULL;
1220         }
1221
1222         err = -ENOMEM;
1223         next_key = kmalloc(map->key_size, GFP_USER);
1224         if (!next_key)
1225                 goto free_key;
1226
1227         if (bpf_map_is_dev_bound(map)) {
1228                 err = bpf_map_offload_get_next_key(map, key, next_key);
1229                 goto out;
1230         }
1231
1232         rcu_read_lock();
1233         err = map->ops->map_get_next_key(map, key, next_key);
1234         rcu_read_unlock();
1235 out:
1236         if (err)
1237                 goto free_next_key;
1238
1239         err = -EFAULT;
1240         if (copy_to_user(unext_key, next_key, map->key_size) != 0)
1241                 goto free_next_key;
1242
1243         err = 0;
1244
1245 free_next_key:
1246         kfree(next_key);
1247 free_key:
1248         kfree(key);
1249 err_put:
1250         fdput(f);
1251         return err;
1252 }
1253
1254 int generic_map_delete_batch(struct bpf_map *map,
1255                              const union bpf_attr *attr,
1256                              union bpf_attr __user *uattr)
1257 {
1258         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1259         u32 cp, max_count;
1260         int err = 0;
1261         void *key;
1262
1263         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1264                 return -EINVAL;
1265
1266         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1267             !map_value_has_spin_lock(map)) {
1268                 return -EINVAL;
1269         }
1270
1271         max_count = attr->batch.count;
1272         if (!max_count)
1273                 return 0;
1274
1275         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1276         if (!key)
1277                 return -ENOMEM;
1278
1279         for (cp = 0; cp < max_count; cp++) {
1280                 err = -EFAULT;
1281                 if (copy_from_user(key, keys + cp * map->key_size,
1282                                    map->key_size))
1283                         break;
1284
1285                 if (bpf_map_is_dev_bound(map)) {
1286                         err = bpf_map_offload_delete_elem(map, key);
1287                         break;
1288                 }
1289
1290                 bpf_disable_instrumentation();
1291                 rcu_read_lock();
1292                 err = map->ops->map_delete_elem(map, key);
1293                 rcu_read_unlock();
1294                 bpf_enable_instrumentation();
1295                 maybe_wait_bpf_programs(map);
1296                 if (err)
1297                         break;
1298         }
1299         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1300                 err = -EFAULT;
1301
1302         kfree(key);
1303         return err;
1304 }
1305
1306 int generic_map_update_batch(struct bpf_map *map,
1307                              const union bpf_attr *attr,
1308                              union bpf_attr __user *uattr)
1309 {
1310         void __user *values = u64_to_user_ptr(attr->batch.values);
1311         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1312         u32 value_size, cp, max_count;
1313         int ufd = attr->map_fd;
1314         void *key, *value;
1315         struct fd f;
1316         int err = 0;
1317
1318         f = fdget(ufd);
1319         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1320                 return -EINVAL;
1321
1322         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1323             !map_value_has_spin_lock(map)) {
1324                 return -EINVAL;
1325         }
1326
1327         value_size = bpf_map_value_size(map);
1328
1329         max_count = attr->batch.count;
1330         if (!max_count)
1331                 return 0;
1332
1333         key = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1334         if (!key)
1335                 return -ENOMEM;
1336
1337         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1338         if (!value) {
1339                 kfree(key);
1340                 return -ENOMEM;
1341         }
1342
1343         for (cp = 0; cp < max_count; cp++) {
1344                 err = -EFAULT;
1345                 if (copy_from_user(key, keys + cp * map->key_size,
1346                     map->key_size) ||
1347                     copy_from_user(value, values + cp * value_size, value_size))
1348                         break;
1349
1350                 err = bpf_map_update_value(map, f, key, value,
1351                                            attr->batch.elem_flags);
1352
1353                 if (err)
1354                         break;
1355         }
1356
1357         if (copy_to_user(&uattr->batch.count, &cp, sizeof(cp)))
1358                 err = -EFAULT;
1359
1360         kfree(value);
1361         kfree(key);
1362         return err;
1363 }
1364
1365 #define MAP_LOOKUP_RETRIES 3
1366
1367 int generic_map_lookup_batch(struct bpf_map *map,
1368                                     const union bpf_attr *attr,
1369                                     union bpf_attr __user *uattr)
1370 {
1371         void __user *uobatch = u64_to_user_ptr(attr->batch.out_batch);
1372         void __user *ubatch = u64_to_user_ptr(attr->batch.in_batch);
1373         void __user *values = u64_to_user_ptr(attr->batch.values);
1374         void __user *keys = u64_to_user_ptr(attr->batch.keys);
1375         void *buf, *buf_prevkey, *prev_key, *key, *value;
1376         int err, retry = MAP_LOOKUP_RETRIES;
1377         u32 value_size, cp, max_count;
1378
1379         if (attr->batch.elem_flags & ~BPF_F_LOCK)
1380                 return -EINVAL;
1381
1382         if ((attr->batch.elem_flags & BPF_F_LOCK) &&
1383             !map_value_has_spin_lock(map))
1384                 return -EINVAL;
1385
1386         value_size = bpf_map_value_size(map);
1387
1388         max_count = attr->batch.count;
1389         if (!max_count)
1390                 return 0;
1391
1392         if (put_user(0, &uattr->batch.count))
1393                 return -EFAULT;
1394
1395         buf_prevkey = kmalloc(map->key_size, GFP_USER | __GFP_NOWARN);
1396         if (!buf_prevkey)
1397                 return -ENOMEM;
1398
1399         buf = kmalloc(map->key_size + value_size, GFP_USER | __GFP_NOWARN);
1400         if (!buf) {
1401                 kfree(buf_prevkey);
1402                 return -ENOMEM;
1403         }
1404
1405         err = -EFAULT;
1406         prev_key = NULL;
1407         if (ubatch && copy_from_user(buf_prevkey, ubatch, map->key_size))
1408                 goto free_buf;
1409         key = buf;
1410         value = key + map->key_size;
1411         if (ubatch)
1412                 prev_key = buf_prevkey;
1413
1414         for (cp = 0; cp < max_count;) {
1415                 rcu_read_lock();
1416                 err = map->ops->map_get_next_key(map, prev_key, key);
1417                 rcu_read_unlock();
1418                 if (err)
1419                         break;
1420                 err = bpf_map_copy_value(map, key, value,
1421                                          attr->batch.elem_flags);
1422
1423                 if (err == -ENOENT) {
1424                         if (retry) {
1425                                 retry--;
1426                                 continue;
1427                         }
1428                         err = -EINTR;
1429                         break;
1430                 }
1431
1432                 if (err)
1433                         goto free_buf;
1434
1435                 if (copy_to_user(keys + cp * map->key_size, key,
1436                                  map->key_size)) {
1437                         err = -EFAULT;
1438                         goto free_buf;
1439                 }
1440                 if (copy_to_user(values + cp * value_size, value, value_size)) {
1441                         err = -EFAULT;
1442                         goto free_buf;
1443                 }
1444
1445                 if (!prev_key)
1446                         prev_key = buf_prevkey;
1447
1448                 swap(prev_key, key);
1449                 retry = MAP_LOOKUP_RETRIES;
1450                 cp++;
1451         }
1452
1453         if (err == -EFAULT)
1454                 goto free_buf;
1455
1456         if ((copy_to_user(&uattr->batch.count, &cp, sizeof(cp)) ||
1457                     (cp && copy_to_user(uobatch, prev_key, map->key_size))))
1458                 err = -EFAULT;
1459
1460 free_buf:
1461         kfree(buf_prevkey);
1462         kfree(buf);
1463         return err;
1464 }
1465
1466 #define BPF_MAP_LOOKUP_AND_DELETE_ELEM_LAST_FIELD value
1467
1468 static int map_lookup_and_delete_elem(union bpf_attr *attr)
1469 {
1470         void __user *ukey = u64_to_user_ptr(attr->key);
1471         void __user *uvalue = u64_to_user_ptr(attr->value);
1472         int ufd = attr->map_fd;
1473         struct bpf_map *map;
1474         void *key, *value;
1475         u32 value_size;
1476         struct fd f;
1477         int err;
1478
1479         if (CHECK_ATTR(BPF_MAP_LOOKUP_AND_DELETE_ELEM))
1480                 return -EINVAL;
1481
1482         f = fdget(ufd);
1483         map = __bpf_map_get(f);
1484         if (IS_ERR(map))
1485                 return PTR_ERR(map);
1486         if (!(map_get_sys_perms(map, f) & FMODE_CAN_READ) ||
1487             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
1488                 err = -EPERM;
1489                 goto err_put;
1490         }
1491
1492         key = __bpf_copy_key(ukey, map->key_size);
1493         if (IS_ERR(key)) {
1494                 err = PTR_ERR(key);
1495                 goto err_put;
1496         }
1497
1498         value_size = map->value_size;
1499
1500         err = -ENOMEM;
1501         value = kmalloc(value_size, GFP_USER | __GFP_NOWARN);
1502         if (!value)
1503                 goto free_key;
1504
1505         if (map->map_type == BPF_MAP_TYPE_QUEUE ||
1506             map->map_type == BPF_MAP_TYPE_STACK) {
1507                 err = map->ops->map_pop_elem(map, value);
1508         } else {
1509                 err = -ENOTSUPP;
1510         }
1511
1512         if (err)
1513                 goto free_value;
1514
1515         if (copy_to_user(uvalue, value, value_size) != 0) {
1516                 err = -EFAULT;
1517                 goto free_value;
1518         }
1519
1520         err = 0;
1521
1522 free_value:
1523         kfree(value);
1524 free_key:
1525         kfree(key);
1526 err_put:
1527         fdput(f);
1528         return err;
1529 }
1530
1531 #define BPF_MAP_FREEZE_LAST_FIELD map_fd
1532
1533 static int map_freeze(const union bpf_attr *attr)
1534 {
1535         int err = 0, ufd = attr->map_fd;
1536         struct bpf_map *map;
1537         struct fd f;
1538
1539         if (CHECK_ATTR(BPF_MAP_FREEZE))
1540                 return -EINVAL;
1541
1542         f = fdget(ufd);
1543         map = __bpf_map_get(f);
1544         if (IS_ERR(map))
1545                 return PTR_ERR(map);
1546
1547         if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) {
1548                 fdput(f);
1549                 return -ENOTSUPP;
1550         }
1551
1552         mutex_lock(&map->freeze_mutex);
1553
1554         if (map->writecnt) {
1555                 err = -EBUSY;
1556                 goto err_put;
1557         }
1558         if (READ_ONCE(map->frozen)) {
1559                 err = -EBUSY;
1560                 goto err_put;
1561         }
1562         if (!bpf_capable()) {
1563                 err = -EPERM;
1564                 goto err_put;
1565         }
1566
1567         WRITE_ONCE(map->frozen, true);
1568 err_put:
1569         mutex_unlock(&map->freeze_mutex);
1570         fdput(f);
1571         return err;
1572 }
1573
1574 static const struct bpf_prog_ops * const bpf_prog_types[] = {
1575 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
1576         [_id] = & _name ## _prog_ops,
1577 #define BPF_MAP_TYPE(_id, _ops)
1578 #define BPF_LINK_TYPE(_id, _name)
1579 #include <linux/bpf_types.h>
1580 #undef BPF_PROG_TYPE
1581 #undef BPF_MAP_TYPE
1582 #undef BPF_LINK_TYPE
1583 };
1584
1585 static int find_prog_type(enum bpf_prog_type type, struct bpf_prog *prog)
1586 {
1587         const struct bpf_prog_ops *ops;
1588
1589         if (type >= ARRAY_SIZE(bpf_prog_types))
1590                 return -EINVAL;
1591         type = array_index_nospec(type, ARRAY_SIZE(bpf_prog_types));
1592         ops = bpf_prog_types[type];
1593         if (!ops)
1594                 return -EINVAL;
1595
1596         if (!bpf_prog_is_dev_bound(prog->aux))
1597                 prog->aux->ops = ops;
1598         else
1599                 prog->aux->ops = &bpf_offload_prog_ops;
1600         prog->type = type;
1601         return 0;
1602 }
1603
1604 enum bpf_audit {
1605         BPF_AUDIT_LOAD,
1606         BPF_AUDIT_UNLOAD,
1607         BPF_AUDIT_MAX,
1608 };
1609
1610 static const char * const bpf_audit_str[BPF_AUDIT_MAX] = {
1611         [BPF_AUDIT_LOAD]   = "LOAD",
1612         [BPF_AUDIT_UNLOAD] = "UNLOAD",
1613 };
1614
1615 static void bpf_audit_prog(const struct bpf_prog *prog, unsigned int op)
1616 {
1617         struct audit_context *ctx = NULL;
1618         struct audit_buffer *ab;
1619
1620         if (WARN_ON_ONCE(op >= BPF_AUDIT_MAX))
1621                 return;
1622         if (audit_enabled == AUDIT_OFF)
1623                 return;
1624         if (op == BPF_AUDIT_LOAD)
1625                 ctx = audit_context();
1626         ab = audit_log_start(ctx, GFP_ATOMIC, AUDIT_BPF);
1627         if (unlikely(!ab))
1628                 return;
1629         audit_log_format(ab, "prog-id=%u op=%s",
1630                          prog->aux->id, bpf_audit_str[op]);
1631         audit_log_end(ab);
1632 }
1633
1634 static int bpf_prog_alloc_id(struct bpf_prog *prog)
1635 {
1636         int id;
1637
1638         idr_preload(GFP_KERNEL);
1639         spin_lock_bh(&prog_idr_lock);
1640         id = idr_alloc_cyclic(&prog_idr, prog, 1, INT_MAX, GFP_ATOMIC);
1641         if (id > 0)
1642                 prog->aux->id = id;
1643         spin_unlock_bh(&prog_idr_lock);
1644         idr_preload_end();
1645
1646         /* id is in [1, INT_MAX) */
1647         if (WARN_ON_ONCE(!id))
1648                 return -ENOSPC;
1649
1650         return id > 0 ? 0 : id;
1651 }
1652
1653 void bpf_prog_free_id(struct bpf_prog *prog, bool do_idr_lock)
1654 {
1655         /* cBPF to eBPF migrations are currently not in the idr store.
1656          * Offloaded programs are removed from the store when their device
1657          * disappears - even if someone grabs an fd to them they are unusable,
1658          * simply waiting for refcnt to drop to be freed.
1659          */
1660         if (!prog->aux->id)
1661                 return;
1662
1663         if (do_idr_lock)
1664                 spin_lock_bh(&prog_idr_lock);
1665         else
1666                 __acquire(&prog_idr_lock);
1667
1668         idr_remove(&prog_idr, prog->aux->id);
1669         prog->aux->id = 0;
1670
1671         if (do_idr_lock)
1672                 spin_unlock_bh(&prog_idr_lock);
1673         else
1674                 __release(&prog_idr_lock);
1675 }
1676
1677 static void __bpf_prog_put_rcu(struct rcu_head *rcu)
1678 {
1679         struct bpf_prog_aux *aux = container_of(rcu, struct bpf_prog_aux, rcu);
1680
1681         kvfree(aux->func_info);
1682         kfree(aux->func_info_aux);
1683         free_uid(aux->user);
1684         security_bpf_prog_free(aux);
1685         bpf_prog_free(aux->prog);
1686 }
1687
1688 static void __bpf_prog_put_noref(struct bpf_prog *prog, bool deferred)
1689 {
1690         bpf_prog_kallsyms_del_all(prog);
1691         btf_put(prog->aux->btf);
1692         bpf_prog_free_linfo(prog);
1693         if (prog->aux->attach_btf)
1694                 btf_put(prog->aux->attach_btf);
1695
1696         if (deferred) {
1697                 if (prog->aux->sleepable)
1698                         call_rcu_tasks_trace(&prog->aux->rcu, __bpf_prog_put_rcu);
1699                 else
1700                         call_rcu(&prog->aux->rcu, __bpf_prog_put_rcu);
1701         } else {
1702                 __bpf_prog_put_rcu(&prog->aux->rcu);
1703         }
1704 }
1705
1706 static void __bpf_prog_put(struct bpf_prog *prog, bool do_idr_lock)
1707 {
1708         if (atomic64_dec_and_test(&prog->aux->refcnt)) {
1709                 perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_UNLOAD, 0);
1710                 bpf_audit_prog(prog, BPF_AUDIT_UNLOAD);
1711                 /* bpf_prog_free_id() must be called first */
1712                 bpf_prog_free_id(prog, do_idr_lock);
1713                 __bpf_prog_put_noref(prog, true);
1714         }
1715 }
1716
1717 void bpf_prog_put(struct bpf_prog *prog)
1718 {
1719         __bpf_prog_put(prog, true);
1720 }
1721 EXPORT_SYMBOL_GPL(bpf_prog_put);
1722
1723 static int bpf_prog_release(struct inode *inode, struct file *filp)
1724 {
1725         struct bpf_prog *prog = filp->private_data;
1726
1727         bpf_prog_put(prog);
1728         return 0;
1729 }
1730
1731 static void bpf_prog_get_stats(const struct bpf_prog *prog,
1732                                struct bpf_prog_stats *stats)
1733 {
1734         u64 nsecs = 0, cnt = 0;
1735         int cpu;
1736
1737         for_each_possible_cpu(cpu) {
1738                 const struct bpf_prog_stats *st;
1739                 unsigned int start;
1740                 u64 tnsecs, tcnt;
1741
1742                 st = per_cpu_ptr(prog->aux->stats, cpu);
1743                 do {
1744                         start = u64_stats_fetch_begin_irq(&st->syncp);
1745                         tnsecs = st->nsecs;
1746                         tcnt = st->cnt;
1747                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1748                 nsecs += tnsecs;
1749                 cnt += tcnt;
1750         }
1751         stats->nsecs = nsecs;
1752         stats->cnt = cnt;
1753 }
1754
1755 #ifdef CONFIG_PROC_FS
1756 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1757 {
1758         const struct bpf_prog *prog = filp->private_data;
1759         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1760         struct bpf_prog_stats stats;
1761
1762         bpf_prog_get_stats(prog, &stats);
1763         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1764         seq_printf(m,
1765                    "prog_type:\t%u\n"
1766                    "prog_jited:\t%u\n"
1767                    "prog_tag:\t%s\n"
1768                    "memlock:\t%llu\n"
1769                    "prog_id:\t%u\n"
1770                    "run_time_ns:\t%llu\n"
1771                    "run_cnt:\t%llu\n",
1772                    prog->type,
1773                    prog->jited,
1774                    prog_tag,
1775                    prog->pages * 1ULL << PAGE_SHIFT,
1776                    prog->aux->id,
1777                    stats.nsecs,
1778                    stats.cnt);
1779 }
1780 #endif
1781
1782 const struct file_operations bpf_prog_fops = {
1783 #ifdef CONFIG_PROC_FS
1784         .show_fdinfo    = bpf_prog_show_fdinfo,
1785 #endif
1786         .release        = bpf_prog_release,
1787         .read           = bpf_dummy_read,
1788         .write          = bpf_dummy_write,
1789 };
1790
1791 int bpf_prog_new_fd(struct bpf_prog *prog)
1792 {
1793         int ret;
1794
1795         ret = security_bpf_prog(prog);
1796         if (ret < 0)
1797                 return ret;
1798
1799         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1800                                 O_RDWR | O_CLOEXEC);
1801 }
1802
1803 static struct bpf_prog *____bpf_prog_get(struct fd f)
1804 {
1805         if (!f.file)
1806                 return ERR_PTR(-EBADF);
1807         if (f.file->f_op != &bpf_prog_fops) {
1808                 fdput(f);
1809                 return ERR_PTR(-EINVAL);
1810         }
1811
1812         return f.file->private_data;
1813 }
1814
1815 void bpf_prog_add(struct bpf_prog *prog, int i)
1816 {
1817         atomic64_add(i, &prog->aux->refcnt);
1818 }
1819 EXPORT_SYMBOL_GPL(bpf_prog_add);
1820
1821 void bpf_prog_sub(struct bpf_prog *prog, int i)
1822 {
1823         /* Only to be used for undoing previous bpf_prog_add() in some
1824          * error path. We still know that another entity in our call
1825          * path holds a reference to the program, thus atomic_sub() can
1826          * be safely used in such cases!
1827          */
1828         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1829 }
1830 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1831
1832 void bpf_prog_inc(struct bpf_prog *prog)
1833 {
1834         atomic64_inc(&prog->aux->refcnt);
1835 }
1836 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1837
1838 /* prog_idr_lock should have been held */
1839 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1840 {
1841         int refold;
1842
1843         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1844
1845         if (!refold)
1846                 return ERR_PTR(-ENOENT);
1847
1848         return prog;
1849 }
1850 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1851
1852 bool bpf_prog_get_ok(struct bpf_prog *prog,
1853                             enum bpf_prog_type *attach_type, bool attach_drv)
1854 {
1855         /* not an attachment, just a refcount inc, always allow */
1856         if (!attach_type)
1857                 return true;
1858
1859         if (prog->type != *attach_type)
1860                 return false;
1861         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1862                 return false;
1863
1864         return true;
1865 }
1866
1867 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1868                                        bool attach_drv)
1869 {
1870         struct fd f = fdget(ufd);
1871         struct bpf_prog *prog;
1872
1873         prog = ____bpf_prog_get(f);
1874         if (IS_ERR(prog))
1875                 return prog;
1876         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1877                 prog = ERR_PTR(-EINVAL);
1878                 goto out;
1879         }
1880
1881         bpf_prog_inc(prog);
1882 out:
1883         fdput(f);
1884         return prog;
1885 }
1886
1887 struct bpf_prog *bpf_prog_get(u32 ufd)
1888 {
1889         return __bpf_prog_get(ufd, NULL, false);
1890 }
1891
1892 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1893                                        bool attach_drv)
1894 {
1895         return __bpf_prog_get(ufd, &type, attach_drv);
1896 }
1897 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1898
1899 /* Initially all BPF programs could be loaded w/o specifying
1900  * expected_attach_type. Later for some of them specifying expected_attach_type
1901  * at load time became required so that program could be validated properly.
1902  * Programs of types that are allowed to be loaded both w/ and w/o (for
1903  * backward compatibility) expected_attach_type, should have the default attach
1904  * type assigned to expected_attach_type for the latter case, so that it can be
1905  * validated later at attach time.
1906  *
1907  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1908  * prog type requires it but has some attach types that have to be backward
1909  * compatible.
1910  */
1911 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1912 {
1913         switch (attr->prog_type) {
1914         case BPF_PROG_TYPE_CGROUP_SOCK:
1915                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1916                  * exist so checking for non-zero is the way to go here.
1917                  */
1918                 if (!attr->expected_attach_type)
1919                         attr->expected_attach_type =
1920                                 BPF_CGROUP_INET_SOCK_CREATE;
1921                 break;
1922         }
1923 }
1924
1925 static int
1926 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1927                            enum bpf_attach_type expected_attach_type,
1928                            struct btf *attach_btf, u32 btf_id,
1929                            struct bpf_prog *dst_prog)
1930 {
1931         if (btf_id) {
1932                 if (btf_id > BTF_MAX_TYPE)
1933                         return -EINVAL;
1934
1935                 if (!attach_btf && !dst_prog)
1936                         return -EINVAL;
1937
1938                 switch (prog_type) {
1939                 case BPF_PROG_TYPE_TRACING:
1940                 case BPF_PROG_TYPE_LSM:
1941                 case BPF_PROG_TYPE_STRUCT_OPS:
1942                 case BPF_PROG_TYPE_EXT:
1943                         break;
1944                 default:
1945                         return -EINVAL;
1946                 }
1947         }
1948
1949         if (attach_btf && (!btf_id || dst_prog))
1950                 return -EINVAL;
1951
1952         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1953             prog_type != BPF_PROG_TYPE_EXT)
1954                 return -EINVAL;
1955
1956         switch (prog_type) {
1957         case BPF_PROG_TYPE_CGROUP_SOCK:
1958                 switch (expected_attach_type) {
1959                 case BPF_CGROUP_INET_SOCK_CREATE:
1960                 case BPF_CGROUP_INET_SOCK_RELEASE:
1961                 case BPF_CGROUP_INET4_POST_BIND:
1962                 case BPF_CGROUP_INET6_POST_BIND:
1963                         return 0;
1964                 default:
1965                         return -EINVAL;
1966                 }
1967         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1968                 switch (expected_attach_type) {
1969                 case BPF_CGROUP_INET4_BIND:
1970                 case BPF_CGROUP_INET6_BIND:
1971                 case BPF_CGROUP_INET4_CONNECT:
1972                 case BPF_CGROUP_INET6_CONNECT:
1973                 case BPF_CGROUP_INET4_GETPEERNAME:
1974                 case BPF_CGROUP_INET6_GETPEERNAME:
1975                 case BPF_CGROUP_INET4_GETSOCKNAME:
1976                 case BPF_CGROUP_INET6_GETSOCKNAME:
1977                 case BPF_CGROUP_UDP4_SENDMSG:
1978                 case BPF_CGROUP_UDP6_SENDMSG:
1979                 case BPF_CGROUP_UDP4_RECVMSG:
1980                 case BPF_CGROUP_UDP6_RECVMSG:
1981                         return 0;
1982                 default:
1983                         return -EINVAL;
1984                 }
1985         case BPF_PROG_TYPE_CGROUP_SKB:
1986                 switch (expected_attach_type) {
1987                 case BPF_CGROUP_INET_INGRESS:
1988                 case BPF_CGROUP_INET_EGRESS:
1989                         return 0;
1990                 default:
1991                         return -EINVAL;
1992                 }
1993         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1994                 switch (expected_attach_type) {
1995                 case BPF_CGROUP_SETSOCKOPT:
1996                 case BPF_CGROUP_GETSOCKOPT:
1997                         return 0;
1998                 default:
1999                         return -EINVAL;
2000                 }
2001         case BPF_PROG_TYPE_SK_LOOKUP:
2002                 if (expected_attach_type == BPF_SK_LOOKUP)
2003                         return 0;
2004                 return -EINVAL;
2005         case BPF_PROG_TYPE_EXT:
2006                 if (expected_attach_type)
2007                         return -EINVAL;
2008                 fallthrough;
2009         default:
2010                 return 0;
2011         }
2012 }
2013
2014 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2015 {
2016         switch (prog_type) {
2017         case BPF_PROG_TYPE_SCHED_CLS:
2018         case BPF_PROG_TYPE_SCHED_ACT:
2019         case BPF_PROG_TYPE_XDP:
2020         case BPF_PROG_TYPE_LWT_IN:
2021         case BPF_PROG_TYPE_LWT_OUT:
2022         case BPF_PROG_TYPE_LWT_XMIT:
2023         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2024         case BPF_PROG_TYPE_SK_SKB:
2025         case BPF_PROG_TYPE_SK_MSG:
2026         case BPF_PROG_TYPE_LIRC_MODE2:
2027         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2028         case BPF_PROG_TYPE_CGROUP_DEVICE:
2029         case BPF_PROG_TYPE_CGROUP_SOCK:
2030         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2031         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2032         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2033         case BPF_PROG_TYPE_SOCK_OPS:
2034         case BPF_PROG_TYPE_EXT: /* extends any prog */
2035                 return true;
2036         case BPF_PROG_TYPE_CGROUP_SKB:
2037                 /* always unpriv */
2038         case BPF_PROG_TYPE_SK_REUSEPORT:
2039                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2040         default:
2041                 return false;
2042         }
2043 }
2044
2045 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2046 {
2047         switch (prog_type) {
2048         case BPF_PROG_TYPE_KPROBE:
2049         case BPF_PROG_TYPE_TRACEPOINT:
2050         case BPF_PROG_TYPE_PERF_EVENT:
2051         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2052         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2053         case BPF_PROG_TYPE_TRACING:
2054         case BPF_PROG_TYPE_LSM:
2055         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2056         case BPF_PROG_TYPE_EXT: /* extends any prog */
2057                 return true;
2058         default:
2059                 return false;
2060         }
2061 }
2062
2063 /* last field in 'union bpf_attr' used by this command */
2064 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2065
2066 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2067 {
2068         enum bpf_prog_type type = attr->prog_type;
2069         struct bpf_prog *prog, *dst_prog = NULL;
2070         struct btf *attach_btf = NULL;
2071         int err;
2072         char license[128];
2073         bool is_gpl;
2074
2075         if (CHECK_ATTR(BPF_PROG_LOAD))
2076                 return -EINVAL;
2077
2078         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2079                                  BPF_F_ANY_ALIGNMENT |
2080                                  BPF_F_TEST_STATE_FREQ |
2081                                  BPF_F_SLEEPABLE |
2082                                  BPF_F_TEST_RND_HI32))
2083                 return -EINVAL;
2084
2085         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2086             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2087             !bpf_capable())
2088                 return -EPERM;
2089
2090         /* copy eBPF program license from user space */
2091         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2092                               sizeof(license) - 1) < 0)
2093                 return -EFAULT;
2094         license[sizeof(license) - 1] = 0;
2095
2096         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2097         is_gpl = license_is_gpl_compatible(license);
2098
2099         if (attr->insn_cnt == 0 ||
2100             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2101                 return -E2BIG;
2102         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2103             type != BPF_PROG_TYPE_CGROUP_SKB &&
2104             !bpf_capable())
2105                 return -EPERM;
2106
2107         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2108                 return -EPERM;
2109         if (is_perfmon_prog_type(type) && !perfmon_capable())
2110                 return -EPERM;
2111
2112         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2113          * or btf, we need to check which one it is
2114          */
2115         if (attr->attach_prog_fd) {
2116                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2117                 if (IS_ERR(dst_prog)) {
2118                         dst_prog = NULL;
2119                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2120                         if (IS_ERR(attach_btf))
2121                                 return -EINVAL;
2122                         if (!btf_is_kernel(attach_btf)) {
2123                                 /* attaching through specifying bpf_prog's BTF
2124                                  * objects directly might be supported eventually
2125                                  */
2126                                 btf_put(attach_btf);
2127                                 return -ENOTSUPP;
2128                         }
2129                 }
2130         } else if (attr->attach_btf_id) {
2131                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2132                 attach_btf = bpf_get_btf_vmlinux();
2133                 if (IS_ERR(attach_btf))
2134                         return PTR_ERR(attach_btf);
2135                 if (!attach_btf)
2136                         return -EINVAL;
2137                 btf_get(attach_btf);
2138         }
2139
2140         bpf_prog_load_fixup_attach_type(attr);
2141         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2142                                        attach_btf, attr->attach_btf_id,
2143                                        dst_prog)) {
2144                 if (dst_prog)
2145                         bpf_prog_put(dst_prog);
2146                 if (attach_btf)
2147                         btf_put(attach_btf);
2148                 return -EINVAL;
2149         }
2150
2151         /* plain bpf_prog allocation */
2152         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2153         if (!prog) {
2154                 if (dst_prog)
2155                         bpf_prog_put(dst_prog);
2156                 if (attach_btf)
2157                         btf_put(attach_btf);
2158                 return -ENOMEM;
2159         }
2160
2161         prog->expected_attach_type = attr->expected_attach_type;
2162         prog->aux->attach_btf = attach_btf;
2163         prog->aux->attach_btf_id = attr->attach_btf_id;
2164         prog->aux->dst_prog = dst_prog;
2165         prog->aux->offload_requested = !!attr->prog_ifindex;
2166         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2167
2168         err = security_bpf_prog_alloc(prog->aux);
2169         if (err)
2170                 goto free_prog;
2171
2172         prog->aux->user = get_current_user();
2173         prog->len = attr->insn_cnt;
2174
2175         err = -EFAULT;
2176         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2177                            bpf_prog_insn_size(prog)) != 0)
2178                 goto free_prog_sec;
2179
2180         prog->orig_prog = NULL;
2181         prog->jited = 0;
2182
2183         atomic64_set(&prog->aux->refcnt, 1);
2184         prog->gpl_compatible = is_gpl ? 1 : 0;
2185
2186         if (bpf_prog_is_dev_bound(prog->aux)) {
2187                 err = bpf_prog_offload_init(prog, attr);
2188                 if (err)
2189                         goto free_prog_sec;
2190         }
2191
2192         /* find program type: socket_filter vs tracing_filter */
2193         err = find_prog_type(type, prog);
2194         if (err < 0)
2195                 goto free_prog_sec;
2196
2197         prog->aux->load_time = ktime_get_boottime_ns();
2198         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2199                                sizeof(attr->prog_name));
2200         if (err < 0)
2201                 goto free_prog_sec;
2202
2203         /* run eBPF verifier */
2204         err = bpf_check(&prog, attr, uattr);
2205         if (err < 0)
2206                 goto free_used_maps;
2207
2208         prog = bpf_prog_select_runtime(prog, &err);
2209         if (err < 0)
2210                 goto free_used_maps;
2211
2212         err = bpf_prog_alloc_id(prog);
2213         if (err)
2214                 goto free_used_maps;
2215
2216         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2217          * effectively publicly exposed. However, retrieving via
2218          * bpf_prog_get_fd_by_id() will take another reference,
2219          * therefore it cannot be gone underneath us.
2220          *
2221          * Only for the time /after/ successful bpf_prog_new_fd()
2222          * and before returning to userspace, we might just hold
2223          * one reference and any parallel close on that fd could
2224          * rip everything out. Hence, below notifications must
2225          * happen before bpf_prog_new_fd().
2226          *
2227          * Also, any failure handling from this point onwards must
2228          * be using bpf_prog_put() given the program is exposed.
2229          */
2230         bpf_prog_kallsyms_add(prog);
2231         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2232         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2233
2234         err = bpf_prog_new_fd(prog);
2235         if (err < 0)
2236                 bpf_prog_put(prog);
2237         return err;
2238
2239 free_used_maps:
2240         /* In case we have subprogs, we need to wait for a grace
2241          * period before we can tear down JIT memory since symbols
2242          * are already exposed under kallsyms.
2243          */
2244         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2245         return err;
2246 free_prog_sec:
2247         free_uid(prog->aux->user);
2248         security_bpf_prog_free(prog->aux);
2249 free_prog:
2250         if (prog->aux->attach_btf)
2251                 btf_put(prog->aux->attach_btf);
2252         bpf_prog_free(prog);
2253         return err;
2254 }
2255
2256 #define BPF_OBJ_LAST_FIELD file_flags
2257
2258 static int bpf_obj_pin(const union bpf_attr *attr)
2259 {
2260         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2261                 return -EINVAL;
2262
2263         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2264 }
2265
2266 static int bpf_obj_get(const union bpf_attr *attr)
2267 {
2268         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2269             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2270                 return -EINVAL;
2271
2272         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2273                                 attr->file_flags);
2274 }
2275
2276 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2277                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2278 {
2279         atomic64_set(&link->refcnt, 1);
2280         link->type = type;
2281         link->id = 0;
2282         link->ops = ops;
2283         link->prog = prog;
2284 }
2285
2286 static void bpf_link_free_id(int id)
2287 {
2288         if (!id)
2289                 return;
2290
2291         spin_lock_bh(&link_idr_lock);
2292         idr_remove(&link_idr, id);
2293         spin_unlock_bh(&link_idr_lock);
2294 }
2295
2296 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2297  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2298  * anon_inode's release() call. This helper marksbpf_link as
2299  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2300  * is not decremented, it's the responsibility of a calling code that failed
2301  * to complete bpf_link initialization.
2302  */
2303 void bpf_link_cleanup(struct bpf_link_primer *primer)
2304 {
2305         primer->link->prog = NULL;
2306         bpf_link_free_id(primer->id);
2307         fput(primer->file);
2308         put_unused_fd(primer->fd);
2309 }
2310
2311 void bpf_link_inc(struct bpf_link *link)
2312 {
2313         atomic64_inc(&link->refcnt);
2314 }
2315
2316 /* bpf_link_free is guaranteed to be called from process context */
2317 static void bpf_link_free(struct bpf_link *link)
2318 {
2319         bpf_link_free_id(link->id);
2320         if (link->prog) {
2321                 /* detach BPF program, clean up used resources */
2322                 link->ops->release(link);
2323                 bpf_prog_put(link->prog);
2324         }
2325         /* free bpf_link and its containing memory */
2326         link->ops->dealloc(link);
2327 }
2328
2329 static void bpf_link_put_deferred(struct work_struct *work)
2330 {
2331         struct bpf_link *link = container_of(work, struct bpf_link, work);
2332
2333         bpf_link_free(link);
2334 }
2335
2336 /* bpf_link_put can be called from atomic context, but ensures that resources
2337  * are freed from process context
2338  */
2339 void bpf_link_put(struct bpf_link *link)
2340 {
2341         if (!atomic64_dec_and_test(&link->refcnt))
2342                 return;
2343
2344         if (in_atomic()) {
2345                 INIT_WORK(&link->work, bpf_link_put_deferred);
2346                 schedule_work(&link->work);
2347         } else {
2348                 bpf_link_free(link);
2349         }
2350 }
2351
2352 static int bpf_link_release(struct inode *inode, struct file *filp)
2353 {
2354         struct bpf_link *link = filp->private_data;
2355
2356         bpf_link_put(link);
2357         return 0;
2358 }
2359
2360 #ifdef CONFIG_PROC_FS
2361 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2362 #define BPF_MAP_TYPE(_id, _ops)
2363 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2364 static const char *bpf_link_type_strs[] = {
2365         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2366 #include <linux/bpf_types.h>
2367 };
2368 #undef BPF_PROG_TYPE
2369 #undef BPF_MAP_TYPE
2370 #undef BPF_LINK_TYPE
2371
2372 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2373 {
2374         const struct bpf_link *link = filp->private_data;
2375         const struct bpf_prog *prog = link->prog;
2376         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2377
2378         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2379         seq_printf(m,
2380                    "link_type:\t%s\n"
2381                    "link_id:\t%u\n"
2382                    "prog_tag:\t%s\n"
2383                    "prog_id:\t%u\n",
2384                    bpf_link_type_strs[link->type],
2385                    link->id,
2386                    prog_tag,
2387                    prog->aux->id);
2388         if (link->ops->show_fdinfo)
2389                 link->ops->show_fdinfo(link, m);
2390 }
2391 #endif
2392
2393 static const struct file_operations bpf_link_fops = {
2394 #ifdef CONFIG_PROC_FS
2395         .show_fdinfo    = bpf_link_show_fdinfo,
2396 #endif
2397         .release        = bpf_link_release,
2398         .read           = bpf_dummy_read,
2399         .write          = bpf_dummy_write,
2400 };
2401
2402 static int bpf_link_alloc_id(struct bpf_link *link)
2403 {
2404         int id;
2405
2406         idr_preload(GFP_KERNEL);
2407         spin_lock_bh(&link_idr_lock);
2408         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2409         spin_unlock_bh(&link_idr_lock);
2410         idr_preload_end();
2411
2412         return id;
2413 }
2414
2415 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2416  * reserving unused FD and allocating ID from link_idr. This is to be paired
2417  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2418  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2419  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2420  * transient state is passed around in struct bpf_link_primer.
2421  * This is preferred way to create and initialize bpf_link, especially when
2422  * there are complicated and expensive operations inbetween creating bpf_link
2423  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2424  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2425  * expensive (and potentially failing) roll back operations in a rare case
2426  * that file, FD, or ID can't be allocated.
2427  */
2428 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2429 {
2430         struct file *file;
2431         int fd, id;
2432
2433         fd = get_unused_fd_flags(O_CLOEXEC);
2434         if (fd < 0)
2435                 return fd;
2436
2437
2438         id = bpf_link_alloc_id(link);
2439         if (id < 0) {
2440                 put_unused_fd(fd);
2441                 return id;
2442         }
2443
2444         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2445         if (IS_ERR(file)) {
2446                 bpf_link_free_id(id);
2447                 put_unused_fd(fd);
2448                 return PTR_ERR(file);
2449         }
2450
2451         primer->link = link;
2452         primer->file = file;
2453         primer->fd = fd;
2454         primer->id = id;
2455         return 0;
2456 }
2457
2458 int bpf_link_settle(struct bpf_link_primer *primer)
2459 {
2460         /* make bpf_link fetchable by ID */
2461         spin_lock_bh(&link_idr_lock);
2462         primer->link->id = primer->id;
2463         spin_unlock_bh(&link_idr_lock);
2464         /* make bpf_link fetchable by FD */
2465         fd_install(primer->fd, primer->file);
2466         /* pass through installed FD */
2467         return primer->fd;
2468 }
2469
2470 int bpf_link_new_fd(struct bpf_link *link)
2471 {
2472         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2473 }
2474
2475 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2476 {
2477         struct fd f = fdget(ufd);
2478         struct bpf_link *link;
2479
2480         if (!f.file)
2481                 return ERR_PTR(-EBADF);
2482         if (f.file->f_op != &bpf_link_fops) {
2483                 fdput(f);
2484                 return ERR_PTR(-EINVAL);
2485         }
2486
2487         link = f.file->private_data;
2488         bpf_link_inc(link);
2489         fdput(f);
2490
2491         return link;
2492 }
2493
2494 struct bpf_tracing_link {
2495         struct bpf_link link;
2496         enum bpf_attach_type attach_type;
2497         struct bpf_trampoline *trampoline;
2498         struct bpf_prog *tgt_prog;
2499 };
2500
2501 static void bpf_tracing_link_release(struct bpf_link *link)
2502 {
2503         struct bpf_tracing_link *tr_link =
2504                 container_of(link, struct bpf_tracing_link, link);
2505
2506         WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2507                                                 tr_link->trampoline));
2508
2509         bpf_trampoline_put(tr_link->trampoline);
2510
2511         /* tgt_prog is NULL if target is a kernel function */
2512         if (tr_link->tgt_prog)
2513                 bpf_prog_put(tr_link->tgt_prog);
2514 }
2515
2516 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2517 {
2518         struct bpf_tracing_link *tr_link =
2519                 container_of(link, struct bpf_tracing_link, link);
2520
2521         kfree(tr_link);
2522 }
2523
2524 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2525                                          struct seq_file *seq)
2526 {
2527         struct bpf_tracing_link *tr_link =
2528                 container_of(link, struct bpf_tracing_link, link);
2529
2530         seq_printf(seq,
2531                    "attach_type:\t%d\n",
2532                    tr_link->attach_type);
2533 }
2534
2535 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2536                                            struct bpf_link_info *info)
2537 {
2538         struct bpf_tracing_link *tr_link =
2539                 container_of(link, struct bpf_tracing_link, link);
2540
2541         info->tracing.attach_type = tr_link->attach_type;
2542
2543         return 0;
2544 }
2545
2546 static const struct bpf_link_ops bpf_tracing_link_lops = {
2547         .release = bpf_tracing_link_release,
2548         .dealloc = bpf_tracing_link_dealloc,
2549         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2550         .fill_link_info = bpf_tracing_link_fill_link_info,
2551 };
2552
2553 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2554                                    int tgt_prog_fd,
2555                                    u32 btf_id)
2556 {
2557         struct bpf_link_primer link_primer;
2558         struct bpf_prog *tgt_prog = NULL;
2559         struct bpf_trampoline *tr = NULL;
2560         struct bpf_tracing_link *link;
2561         u64 key = 0;
2562         int err;
2563
2564         switch (prog->type) {
2565         case BPF_PROG_TYPE_TRACING:
2566                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2567                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2568                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2569                         err = -EINVAL;
2570                         goto out_put_prog;
2571                 }
2572                 break;
2573         case BPF_PROG_TYPE_EXT:
2574                 if (prog->expected_attach_type != 0) {
2575                         err = -EINVAL;
2576                         goto out_put_prog;
2577                 }
2578                 break;
2579         case BPF_PROG_TYPE_LSM:
2580                 if (prog->expected_attach_type != BPF_LSM_MAC) {
2581                         err = -EINVAL;
2582                         goto out_put_prog;
2583                 }
2584                 break;
2585         default:
2586                 err = -EINVAL;
2587                 goto out_put_prog;
2588         }
2589
2590         if (!!tgt_prog_fd != !!btf_id) {
2591                 err = -EINVAL;
2592                 goto out_put_prog;
2593         }
2594
2595         if (tgt_prog_fd) {
2596                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2597                 if (prog->type != BPF_PROG_TYPE_EXT) {
2598                         err = -EINVAL;
2599                         goto out_put_prog;
2600                 }
2601
2602                 tgt_prog = bpf_prog_get(tgt_prog_fd);
2603                 if (IS_ERR(tgt_prog)) {
2604                         err = PTR_ERR(tgt_prog);
2605                         tgt_prog = NULL;
2606                         goto out_put_prog;
2607                 }
2608
2609                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2610         }
2611
2612         link = kzalloc(sizeof(*link), GFP_USER);
2613         if (!link) {
2614                 err = -ENOMEM;
2615                 goto out_put_prog;
2616         }
2617         bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2618                       &bpf_tracing_link_lops, prog);
2619         link->attach_type = prog->expected_attach_type;
2620
2621         mutex_lock(&prog->aux->dst_mutex);
2622
2623         /* There are a few possible cases here:
2624          *
2625          * - if prog->aux->dst_trampoline is set, the program was just loaded
2626          *   and not yet attached to anything, so we can use the values stored
2627          *   in prog->aux
2628          *
2629          * - if prog->aux->dst_trampoline is NULL, the program has already been
2630          *   attached to a target and its initial target was cleared (below)
2631          *
2632          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2633          *   target_btf_id using the link_create API.
2634          *
2635          * - if tgt_prog == NULL when this function was called using the old
2636          *   raw_tracepoint_open API, and we need a target from prog->aux
2637          *
2638          * The combination of no saved target in prog->aux, and no target
2639          * specified on load is illegal, and we reject that here.
2640          */
2641         if (!prog->aux->dst_trampoline && !tgt_prog) {
2642                 err = -ENOENT;
2643                 goto out_unlock;
2644         }
2645
2646         if (!prog->aux->dst_trampoline ||
2647             (key && key != prog->aux->dst_trampoline->key)) {
2648                 /* If there is no saved target, or the specified target is
2649                  * different from the destination specified at load time, we
2650                  * need a new trampoline and a check for compatibility
2651                  */
2652                 struct bpf_attach_target_info tgt_info = {};
2653
2654                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2655                                               &tgt_info);
2656                 if (err)
2657                         goto out_unlock;
2658
2659                 tr = bpf_trampoline_get(key, &tgt_info);
2660                 if (!tr) {
2661                         err = -ENOMEM;
2662                         goto out_unlock;
2663                 }
2664         } else {
2665                 /* The caller didn't specify a target, or the target was the
2666                  * same as the destination supplied during program load. This
2667                  * means we can reuse the trampoline and reference from program
2668                  * load time, and there is no need to allocate a new one. This
2669                  * can only happen once for any program, as the saved values in
2670                  * prog->aux are cleared below.
2671                  */
2672                 tr = prog->aux->dst_trampoline;
2673                 tgt_prog = prog->aux->dst_prog;
2674         }
2675
2676         err = bpf_link_prime(&link->link, &link_primer);
2677         if (err)
2678                 goto out_unlock;
2679
2680         err = bpf_trampoline_link_prog(prog, tr);
2681         if (err) {
2682                 bpf_link_cleanup(&link_primer);
2683                 link = NULL;
2684                 goto out_unlock;
2685         }
2686
2687         link->tgt_prog = tgt_prog;
2688         link->trampoline = tr;
2689
2690         /* Always clear the trampoline and target prog from prog->aux to make
2691          * sure the original attach destination is not kept alive after a
2692          * program is (re-)attached to another target.
2693          */
2694         if (prog->aux->dst_prog &&
2695             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2696                 /* got extra prog ref from syscall, or attaching to different prog */
2697                 bpf_prog_put(prog->aux->dst_prog);
2698         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2699                 /* we allocated a new trampoline, so free the old one */
2700                 bpf_trampoline_put(prog->aux->dst_trampoline);
2701
2702         prog->aux->dst_prog = NULL;
2703         prog->aux->dst_trampoline = NULL;
2704         mutex_unlock(&prog->aux->dst_mutex);
2705
2706         return bpf_link_settle(&link_primer);
2707 out_unlock:
2708         if (tr && tr != prog->aux->dst_trampoline)
2709                 bpf_trampoline_put(tr);
2710         mutex_unlock(&prog->aux->dst_mutex);
2711         kfree(link);
2712 out_put_prog:
2713         if (tgt_prog_fd && tgt_prog)
2714                 bpf_prog_put(tgt_prog);
2715         bpf_prog_put(prog);
2716         return err;
2717 }
2718
2719 struct bpf_raw_tp_link {
2720         struct bpf_link link;
2721         struct bpf_raw_event_map *btp;
2722 };
2723
2724 static void bpf_raw_tp_link_release(struct bpf_link *link)
2725 {
2726         struct bpf_raw_tp_link *raw_tp =
2727                 container_of(link, struct bpf_raw_tp_link, link);
2728
2729         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2730         bpf_put_raw_tracepoint(raw_tp->btp);
2731 }
2732
2733 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2734 {
2735         struct bpf_raw_tp_link *raw_tp =
2736                 container_of(link, struct bpf_raw_tp_link, link);
2737
2738         kfree(raw_tp);
2739 }
2740
2741 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2742                                         struct seq_file *seq)
2743 {
2744         struct bpf_raw_tp_link *raw_tp_link =
2745                 container_of(link, struct bpf_raw_tp_link, link);
2746
2747         seq_printf(seq,
2748                    "tp_name:\t%s\n",
2749                    raw_tp_link->btp->tp->name);
2750 }
2751
2752 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2753                                           struct bpf_link_info *info)
2754 {
2755         struct bpf_raw_tp_link *raw_tp_link =
2756                 container_of(link, struct bpf_raw_tp_link, link);
2757         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2758         const char *tp_name = raw_tp_link->btp->tp->name;
2759         u32 ulen = info->raw_tracepoint.tp_name_len;
2760         size_t tp_len = strlen(tp_name);
2761
2762         if (!ulen ^ !ubuf)
2763                 return -EINVAL;
2764
2765         info->raw_tracepoint.tp_name_len = tp_len + 1;
2766
2767         if (!ubuf)
2768                 return 0;
2769
2770         if (ulen >= tp_len + 1) {
2771                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2772                         return -EFAULT;
2773         } else {
2774                 char zero = '\0';
2775
2776                 if (copy_to_user(ubuf, tp_name, ulen - 1))
2777                         return -EFAULT;
2778                 if (put_user(zero, ubuf + ulen - 1))
2779                         return -EFAULT;
2780                 return -ENOSPC;
2781         }
2782
2783         return 0;
2784 }
2785
2786 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2787         .release = bpf_raw_tp_link_release,
2788         .dealloc = bpf_raw_tp_link_dealloc,
2789         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2790         .fill_link_info = bpf_raw_tp_link_fill_link_info,
2791 };
2792
2793 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2794
2795 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2796 {
2797         struct bpf_link_primer link_primer;
2798         struct bpf_raw_tp_link *link;
2799         struct bpf_raw_event_map *btp;
2800         struct bpf_prog *prog;
2801         const char *tp_name;
2802         char buf[128];
2803         int err;
2804
2805         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2806                 return -EINVAL;
2807
2808         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2809         if (IS_ERR(prog))
2810                 return PTR_ERR(prog);
2811
2812         switch (prog->type) {
2813         case BPF_PROG_TYPE_TRACING:
2814         case BPF_PROG_TYPE_EXT:
2815         case BPF_PROG_TYPE_LSM:
2816                 if (attr->raw_tracepoint.name) {
2817                         /* The attach point for this category of programs
2818                          * should be specified via btf_id during program load.
2819                          */
2820                         err = -EINVAL;
2821                         goto out_put_prog;
2822                 }
2823                 if (prog->type == BPF_PROG_TYPE_TRACING &&
2824                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2825                         tp_name = prog->aux->attach_func_name;
2826                         break;
2827                 }
2828                 return bpf_tracing_prog_attach(prog, 0, 0);
2829         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2830         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2831                 if (strncpy_from_user(buf,
2832                                       u64_to_user_ptr(attr->raw_tracepoint.name),
2833                                       sizeof(buf) - 1) < 0) {
2834                         err = -EFAULT;
2835                         goto out_put_prog;
2836                 }
2837                 buf[sizeof(buf) - 1] = 0;
2838                 tp_name = buf;
2839                 break;
2840         default:
2841                 err = -EINVAL;
2842                 goto out_put_prog;
2843         }
2844
2845         btp = bpf_get_raw_tracepoint(tp_name);
2846         if (!btp) {
2847                 err = -ENOENT;
2848                 goto out_put_prog;
2849         }
2850
2851         link = kzalloc(sizeof(*link), GFP_USER);
2852         if (!link) {
2853                 err = -ENOMEM;
2854                 goto out_put_btp;
2855         }
2856         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2857                       &bpf_raw_tp_link_lops, prog);
2858         link->btp = btp;
2859
2860         err = bpf_link_prime(&link->link, &link_primer);
2861         if (err) {
2862                 kfree(link);
2863                 goto out_put_btp;
2864         }
2865
2866         err = bpf_probe_register(link->btp, prog);
2867         if (err) {
2868                 bpf_link_cleanup(&link_primer);
2869                 goto out_put_btp;
2870         }
2871
2872         return bpf_link_settle(&link_primer);
2873
2874 out_put_btp:
2875         bpf_put_raw_tracepoint(btp);
2876 out_put_prog:
2877         bpf_prog_put(prog);
2878         return err;
2879 }
2880
2881 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2882                                              enum bpf_attach_type attach_type)
2883 {
2884         switch (prog->type) {
2885         case BPF_PROG_TYPE_CGROUP_SOCK:
2886         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2887         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2888         case BPF_PROG_TYPE_SK_LOOKUP:
2889                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2890         case BPF_PROG_TYPE_CGROUP_SKB:
2891                 if (!capable(CAP_NET_ADMIN))
2892                         /* cg-skb progs can be loaded by unpriv user.
2893                          * check permissions at attach time.
2894                          */
2895                         return -EPERM;
2896                 return prog->enforce_expected_attach_type &&
2897                         prog->expected_attach_type != attach_type ?
2898                         -EINVAL : 0;
2899         default:
2900                 return 0;
2901         }
2902 }
2903
2904 static enum bpf_prog_type
2905 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2906 {
2907         switch (attach_type) {
2908         case BPF_CGROUP_INET_INGRESS:
2909         case BPF_CGROUP_INET_EGRESS:
2910                 return BPF_PROG_TYPE_CGROUP_SKB;
2911         case BPF_CGROUP_INET_SOCK_CREATE:
2912         case BPF_CGROUP_INET_SOCK_RELEASE:
2913         case BPF_CGROUP_INET4_POST_BIND:
2914         case BPF_CGROUP_INET6_POST_BIND:
2915                 return BPF_PROG_TYPE_CGROUP_SOCK;
2916         case BPF_CGROUP_INET4_BIND:
2917         case BPF_CGROUP_INET6_BIND:
2918         case BPF_CGROUP_INET4_CONNECT:
2919         case BPF_CGROUP_INET6_CONNECT:
2920         case BPF_CGROUP_INET4_GETPEERNAME:
2921         case BPF_CGROUP_INET6_GETPEERNAME:
2922         case BPF_CGROUP_INET4_GETSOCKNAME:
2923         case BPF_CGROUP_INET6_GETSOCKNAME:
2924         case BPF_CGROUP_UDP4_SENDMSG:
2925         case BPF_CGROUP_UDP6_SENDMSG:
2926         case BPF_CGROUP_UDP4_RECVMSG:
2927         case BPF_CGROUP_UDP6_RECVMSG:
2928                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2929         case BPF_CGROUP_SOCK_OPS:
2930                 return BPF_PROG_TYPE_SOCK_OPS;
2931         case BPF_CGROUP_DEVICE:
2932                 return BPF_PROG_TYPE_CGROUP_DEVICE;
2933         case BPF_SK_MSG_VERDICT:
2934                 return BPF_PROG_TYPE_SK_MSG;
2935         case BPF_SK_SKB_STREAM_PARSER:
2936         case BPF_SK_SKB_STREAM_VERDICT:
2937                 return BPF_PROG_TYPE_SK_SKB;
2938         case BPF_LIRC_MODE2:
2939                 return BPF_PROG_TYPE_LIRC_MODE2;
2940         case BPF_FLOW_DISSECTOR:
2941                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2942         case BPF_CGROUP_SYSCTL:
2943                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2944         case BPF_CGROUP_GETSOCKOPT:
2945         case BPF_CGROUP_SETSOCKOPT:
2946                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2947         case BPF_TRACE_ITER:
2948                 return BPF_PROG_TYPE_TRACING;
2949         case BPF_SK_LOOKUP:
2950                 return BPF_PROG_TYPE_SK_LOOKUP;
2951         case BPF_XDP:
2952                 return BPF_PROG_TYPE_XDP;
2953         default:
2954                 return BPF_PROG_TYPE_UNSPEC;
2955         }
2956 }
2957
2958 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2959
2960 #define BPF_F_ATTACH_MASK \
2961         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2962
2963 static int bpf_prog_attach(const union bpf_attr *attr)
2964 {
2965         enum bpf_prog_type ptype;
2966         struct bpf_prog *prog;
2967         int ret;
2968
2969         if (CHECK_ATTR(BPF_PROG_ATTACH))
2970                 return -EINVAL;
2971
2972         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2973                 return -EINVAL;
2974
2975         ptype = attach_type_to_prog_type(attr->attach_type);
2976         if (ptype == BPF_PROG_TYPE_UNSPEC)
2977                 return -EINVAL;
2978
2979         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2980         if (IS_ERR(prog))
2981                 return PTR_ERR(prog);
2982
2983         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2984                 bpf_prog_put(prog);
2985                 return -EINVAL;
2986         }
2987
2988         switch (ptype) {
2989         case BPF_PROG_TYPE_SK_SKB:
2990         case BPF_PROG_TYPE_SK_MSG:
2991                 ret = sock_map_get_from_fd(attr, prog);
2992                 break;
2993         case BPF_PROG_TYPE_LIRC_MODE2:
2994                 ret = lirc_prog_attach(attr, prog);
2995                 break;
2996         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2997                 ret = netns_bpf_prog_attach(attr, prog);
2998                 break;
2999         case BPF_PROG_TYPE_CGROUP_DEVICE:
3000         case BPF_PROG_TYPE_CGROUP_SKB:
3001         case BPF_PROG_TYPE_CGROUP_SOCK:
3002         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3003         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3004         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3005         case BPF_PROG_TYPE_SOCK_OPS:
3006                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3007                 break;
3008         default:
3009                 ret = -EINVAL;
3010         }
3011
3012         if (ret)
3013                 bpf_prog_put(prog);
3014         return ret;
3015 }
3016
3017 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3018
3019 static int bpf_prog_detach(const union bpf_attr *attr)
3020 {
3021         enum bpf_prog_type ptype;
3022
3023         if (CHECK_ATTR(BPF_PROG_DETACH))
3024                 return -EINVAL;
3025
3026         ptype = attach_type_to_prog_type(attr->attach_type);
3027
3028         switch (ptype) {
3029         case BPF_PROG_TYPE_SK_MSG:
3030         case BPF_PROG_TYPE_SK_SKB:
3031                 return sock_map_prog_detach(attr, ptype);
3032         case BPF_PROG_TYPE_LIRC_MODE2:
3033                 return lirc_prog_detach(attr);
3034         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3035                 return netns_bpf_prog_detach(attr, ptype);
3036         case BPF_PROG_TYPE_CGROUP_DEVICE:
3037         case BPF_PROG_TYPE_CGROUP_SKB:
3038         case BPF_PROG_TYPE_CGROUP_SOCK:
3039         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3040         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3041         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3042         case BPF_PROG_TYPE_SOCK_OPS:
3043                 return cgroup_bpf_prog_detach(attr, ptype);
3044         default:
3045                 return -EINVAL;
3046         }
3047 }
3048
3049 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3050
3051 static int bpf_prog_query(const union bpf_attr *attr,
3052                           union bpf_attr __user *uattr)
3053 {
3054         if (!capable(CAP_NET_ADMIN))
3055                 return -EPERM;
3056         if (CHECK_ATTR(BPF_PROG_QUERY))
3057                 return -EINVAL;
3058         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3059                 return -EINVAL;
3060
3061         switch (attr->query.attach_type) {
3062         case BPF_CGROUP_INET_INGRESS:
3063         case BPF_CGROUP_INET_EGRESS:
3064         case BPF_CGROUP_INET_SOCK_CREATE:
3065         case BPF_CGROUP_INET_SOCK_RELEASE:
3066         case BPF_CGROUP_INET4_BIND:
3067         case BPF_CGROUP_INET6_BIND:
3068         case BPF_CGROUP_INET4_POST_BIND:
3069         case BPF_CGROUP_INET6_POST_BIND:
3070         case BPF_CGROUP_INET4_CONNECT:
3071         case BPF_CGROUP_INET6_CONNECT:
3072         case BPF_CGROUP_INET4_GETPEERNAME:
3073         case BPF_CGROUP_INET6_GETPEERNAME:
3074         case BPF_CGROUP_INET4_GETSOCKNAME:
3075         case BPF_CGROUP_INET6_GETSOCKNAME:
3076         case BPF_CGROUP_UDP4_SENDMSG:
3077         case BPF_CGROUP_UDP6_SENDMSG:
3078         case BPF_CGROUP_UDP4_RECVMSG:
3079         case BPF_CGROUP_UDP6_RECVMSG:
3080         case BPF_CGROUP_SOCK_OPS:
3081         case BPF_CGROUP_DEVICE:
3082         case BPF_CGROUP_SYSCTL:
3083         case BPF_CGROUP_GETSOCKOPT:
3084         case BPF_CGROUP_SETSOCKOPT:
3085                 return cgroup_bpf_prog_query(attr, uattr);
3086         case BPF_LIRC_MODE2:
3087                 return lirc_prog_query(attr, uattr);
3088         case BPF_FLOW_DISSECTOR:
3089         case BPF_SK_LOOKUP:
3090                 return netns_bpf_prog_query(attr, uattr);
3091         default:
3092                 return -EINVAL;
3093         }
3094 }
3095
3096 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3097
3098 static int bpf_prog_test_run(const union bpf_attr *attr,
3099                              union bpf_attr __user *uattr)
3100 {
3101         struct bpf_prog *prog;
3102         int ret = -ENOTSUPP;
3103
3104         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3105                 return -EINVAL;
3106
3107         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3108             (!attr->test.ctx_size_in && attr->test.ctx_in))
3109                 return -EINVAL;
3110
3111         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3112             (!attr->test.ctx_size_out && attr->test.ctx_out))
3113                 return -EINVAL;
3114
3115         prog = bpf_prog_get(attr->test.prog_fd);
3116         if (IS_ERR(prog))
3117                 return PTR_ERR(prog);
3118
3119         if (prog->aux->ops->test_run)
3120                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3121
3122         bpf_prog_put(prog);
3123         return ret;
3124 }
3125
3126 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3127
3128 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3129                                union bpf_attr __user *uattr,
3130                                struct idr *idr,
3131                                spinlock_t *lock)
3132 {
3133         u32 next_id = attr->start_id;
3134         int err = 0;
3135
3136         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3137                 return -EINVAL;
3138
3139         if (!capable(CAP_SYS_ADMIN))
3140                 return -EPERM;
3141
3142         next_id++;
3143         spin_lock_bh(lock);
3144         if (!idr_get_next(idr, &next_id))
3145                 err = -ENOENT;
3146         spin_unlock_bh(lock);
3147
3148         if (!err)
3149                 err = put_user(next_id, &uattr->next_id);
3150
3151         return err;
3152 }
3153
3154 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3155 {
3156         struct bpf_map *map;
3157
3158         spin_lock_bh(&map_idr_lock);
3159 again:
3160         map = idr_get_next(&map_idr, id);
3161         if (map) {
3162                 map = __bpf_map_inc_not_zero(map, false);
3163                 if (IS_ERR(map)) {
3164                         (*id)++;
3165                         goto again;
3166                 }
3167         }
3168         spin_unlock_bh(&map_idr_lock);
3169
3170         return map;
3171 }
3172
3173 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3174 {
3175         struct bpf_prog *prog;
3176
3177         spin_lock_bh(&prog_idr_lock);
3178 again:
3179         prog = idr_get_next(&prog_idr, id);
3180         if (prog) {
3181                 prog = bpf_prog_inc_not_zero(prog);
3182                 if (IS_ERR(prog)) {
3183                         (*id)++;
3184                         goto again;
3185                 }
3186         }
3187         spin_unlock_bh(&prog_idr_lock);
3188
3189         return prog;
3190 }
3191
3192 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3193
3194 struct bpf_prog *bpf_prog_by_id(u32 id)
3195 {
3196         struct bpf_prog *prog;
3197
3198         if (!id)
3199                 return ERR_PTR(-ENOENT);
3200
3201         spin_lock_bh(&prog_idr_lock);
3202         prog = idr_find(&prog_idr, id);
3203         if (prog)
3204                 prog = bpf_prog_inc_not_zero(prog);
3205         else
3206                 prog = ERR_PTR(-ENOENT);
3207         spin_unlock_bh(&prog_idr_lock);
3208         return prog;
3209 }
3210
3211 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3212 {
3213         struct bpf_prog *prog;
3214         u32 id = attr->prog_id;
3215         int fd;
3216
3217         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3218                 return -EINVAL;
3219
3220         if (!capable(CAP_SYS_ADMIN))
3221                 return -EPERM;
3222
3223         prog = bpf_prog_by_id(id);
3224         if (IS_ERR(prog))
3225                 return PTR_ERR(prog);
3226
3227         fd = bpf_prog_new_fd(prog);
3228         if (fd < 0)
3229                 bpf_prog_put(prog);
3230
3231         return fd;
3232 }
3233
3234 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3235
3236 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3237 {
3238         struct bpf_map *map;
3239         u32 id = attr->map_id;
3240         int f_flags;
3241         int fd;
3242
3243         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3244             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3245                 return -EINVAL;
3246
3247         if (!capable(CAP_SYS_ADMIN))
3248                 return -EPERM;
3249
3250         f_flags = bpf_get_file_flag(attr->open_flags);
3251         if (f_flags < 0)
3252                 return f_flags;
3253
3254         spin_lock_bh(&map_idr_lock);
3255         map = idr_find(&map_idr, id);
3256         if (map)
3257                 map = __bpf_map_inc_not_zero(map, true);
3258         else
3259                 map = ERR_PTR(-ENOENT);
3260         spin_unlock_bh(&map_idr_lock);
3261
3262         if (IS_ERR(map))
3263                 return PTR_ERR(map);
3264
3265         fd = bpf_map_new_fd(map, f_flags);
3266         if (fd < 0)
3267                 bpf_map_put_with_uref(map);
3268
3269         return fd;
3270 }
3271
3272 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3273                                               unsigned long addr, u32 *off,
3274                                               u32 *type)
3275 {
3276         const struct bpf_map *map;
3277         int i;
3278
3279         mutex_lock(&prog->aux->used_maps_mutex);
3280         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3281                 map = prog->aux->used_maps[i];
3282                 if (map == (void *)addr) {
3283                         *type = BPF_PSEUDO_MAP_FD;
3284                         goto out;
3285                 }
3286                 if (!map->ops->map_direct_value_meta)
3287                         continue;
3288                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3289                         *type = BPF_PSEUDO_MAP_VALUE;
3290                         goto out;
3291                 }
3292         }
3293         map = NULL;
3294
3295 out:
3296         mutex_unlock(&prog->aux->used_maps_mutex);
3297         return map;
3298 }
3299
3300 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3301                                               const struct cred *f_cred)
3302 {
3303         const struct bpf_map *map;
3304         struct bpf_insn *insns;
3305         u32 off, type;
3306         u64 imm;
3307         u8 code;
3308         int i;
3309
3310         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3311                         GFP_USER);
3312         if (!insns)
3313                 return insns;
3314
3315         for (i = 0; i < prog->len; i++) {
3316                 code = insns[i].code;
3317
3318                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3319                         insns[i].code = BPF_JMP | BPF_CALL;
3320                         insns[i].imm = BPF_FUNC_tail_call;
3321                         /* fall-through */
3322                 }
3323                 if (code == (BPF_JMP | BPF_CALL) ||
3324                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3325                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3326                                 insns[i].code = BPF_JMP | BPF_CALL;
3327                         if (!bpf_dump_raw_ok(f_cred))
3328                                 insns[i].imm = 0;
3329                         continue;
3330                 }
3331                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3332                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3333                         continue;
3334                 }
3335
3336                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3337                         continue;
3338
3339                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3340                 map = bpf_map_from_imm(prog, imm, &off, &type);
3341                 if (map) {
3342                         insns[i].src_reg = type;
3343                         insns[i].imm = map->id;
3344                         insns[i + 1].imm = off;
3345                         continue;
3346                 }
3347         }
3348
3349         return insns;
3350 }
3351
3352 static int set_info_rec_size(struct bpf_prog_info *info)
3353 {
3354         /*
3355          * Ensure info.*_rec_size is the same as kernel expected size
3356          *
3357          * or
3358          *
3359          * Only allow zero *_rec_size if both _rec_size and _cnt are
3360          * zero.  In this case, the kernel will set the expected
3361          * _rec_size back to the info.
3362          */
3363
3364         if ((info->nr_func_info || info->func_info_rec_size) &&
3365             info->func_info_rec_size != sizeof(struct bpf_func_info))
3366                 return -EINVAL;
3367
3368         if ((info->nr_line_info || info->line_info_rec_size) &&
3369             info->line_info_rec_size != sizeof(struct bpf_line_info))
3370                 return -EINVAL;
3371
3372         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3373             info->jited_line_info_rec_size != sizeof(__u64))
3374                 return -EINVAL;
3375
3376         info->func_info_rec_size = sizeof(struct bpf_func_info);
3377         info->line_info_rec_size = sizeof(struct bpf_line_info);
3378         info->jited_line_info_rec_size = sizeof(__u64);
3379
3380         return 0;
3381 }
3382
3383 static int bpf_prog_get_info_by_fd(struct file *file,
3384                                    struct bpf_prog *prog,
3385                                    const union bpf_attr *attr,
3386                                    union bpf_attr __user *uattr)
3387 {
3388         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3389         struct bpf_prog_info info;
3390         u32 info_len = attr->info.info_len;
3391         struct bpf_prog_stats stats;
3392         char __user *uinsns;
3393         u32 ulen;
3394         int err;
3395
3396         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3397         if (err)
3398                 return err;
3399         info_len = min_t(u32, sizeof(info), info_len);
3400
3401         memset(&info, 0, sizeof(info));
3402         if (copy_from_user(&info, uinfo, info_len))
3403                 return -EFAULT;
3404
3405         info.type = prog->type;
3406         info.id = prog->aux->id;
3407         info.load_time = prog->aux->load_time;
3408         info.created_by_uid = from_kuid_munged(current_user_ns(),
3409                                                prog->aux->user->uid);
3410         info.gpl_compatible = prog->gpl_compatible;
3411
3412         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3413         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3414
3415         mutex_lock(&prog->aux->used_maps_mutex);
3416         ulen = info.nr_map_ids;
3417         info.nr_map_ids = prog->aux->used_map_cnt;
3418         ulen = min_t(u32, info.nr_map_ids, ulen);
3419         if (ulen) {
3420                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3421                 u32 i;
3422
3423                 for (i = 0; i < ulen; i++)
3424                         if (put_user(prog->aux->used_maps[i]->id,
3425                                      &user_map_ids[i])) {
3426                                 mutex_unlock(&prog->aux->used_maps_mutex);
3427                                 return -EFAULT;
3428                         }
3429         }
3430         mutex_unlock(&prog->aux->used_maps_mutex);
3431
3432         err = set_info_rec_size(&info);
3433         if (err)
3434                 return err;
3435
3436         bpf_prog_get_stats(prog, &stats);
3437         info.run_time_ns = stats.nsecs;
3438         info.run_cnt = stats.cnt;
3439
3440         if (!bpf_capable()) {
3441                 info.jited_prog_len = 0;
3442                 info.xlated_prog_len = 0;
3443                 info.nr_jited_ksyms = 0;
3444                 info.nr_jited_func_lens = 0;
3445                 info.nr_func_info = 0;
3446                 info.nr_line_info = 0;
3447                 info.nr_jited_line_info = 0;
3448                 goto done;
3449         }
3450
3451         ulen = info.xlated_prog_len;
3452         info.xlated_prog_len = bpf_prog_insn_size(prog);
3453         if (info.xlated_prog_len && ulen) {
3454                 struct bpf_insn *insns_sanitized;
3455                 bool fault;
3456
3457                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3458                         info.xlated_prog_insns = 0;
3459                         goto done;
3460                 }
3461                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3462                 if (!insns_sanitized)
3463                         return -ENOMEM;
3464                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3465                 ulen = min_t(u32, info.xlated_prog_len, ulen);
3466                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3467                 kfree(insns_sanitized);
3468                 if (fault)
3469                         return -EFAULT;
3470         }
3471
3472         if (bpf_prog_is_dev_bound(prog->aux)) {
3473                 err = bpf_prog_offload_info_fill(&info, prog);
3474                 if (err)
3475                         return err;
3476                 goto done;
3477         }
3478
3479         /* NOTE: the following code is supposed to be skipped for offload.
3480          * bpf_prog_offload_info_fill() is the place to fill similar fields
3481          * for offload.
3482          */
3483         ulen = info.jited_prog_len;
3484         if (prog->aux->func_cnt) {
3485                 u32 i;
3486
3487                 info.jited_prog_len = 0;
3488                 for (i = 0; i < prog->aux->func_cnt; i++)
3489                         info.jited_prog_len += prog->aux->func[i]->jited_len;
3490         } else {
3491                 info.jited_prog_len = prog->jited_len;
3492         }
3493
3494         if (info.jited_prog_len && ulen) {
3495                 if (bpf_dump_raw_ok(file->f_cred)) {
3496                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
3497                         ulen = min_t(u32, info.jited_prog_len, ulen);
3498
3499                         /* for multi-function programs, copy the JITed
3500                          * instructions for all the functions
3501                          */
3502                         if (prog->aux->func_cnt) {
3503                                 u32 len, free, i;
3504                                 u8 *img;
3505
3506                                 free = ulen;
3507                                 for (i = 0; i < prog->aux->func_cnt; i++) {
3508                                         len = prog->aux->func[i]->jited_len;
3509                                         len = min_t(u32, len, free);
3510                                         img = (u8 *) prog->aux->func[i]->bpf_func;
3511                                         if (copy_to_user(uinsns, img, len))
3512                                                 return -EFAULT;
3513                                         uinsns += len;
3514                                         free -= len;
3515                                         if (!free)
3516                                                 break;
3517                                 }
3518                         } else {
3519                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3520                                         return -EFAULT;
3521                         }
3522                 } else {
3523                         info.jited_prog_insns = 0;
3524                 }
3525         }
3526
3527         ulen = info.nr_jited_ksyms;
3528         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3529         if (ulen) {
3530                 if (bpf_dump_raw_ok(file->f_cred)) {
3531                         unsigned long ksym_addr;
3532                         u64 __user *user_ksyms;
3533                         u32 i;
3534
3535                         /* copy the address of the kernel symbol
3536                          * corresponding to each function
3537                          */
3538                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3539                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3540                         if (prog->aux->func_cnt) {
3541                                 for (i = 0; i < ulen; i++) {
3542                                         ksym_addr = (unsigned long)
3543                                                 prog->aux->func[i]->bpf_func;
3544                                         if (put_user((u64) ksym_addr,
3545                                                      &user_ksyms[i]))
3546                                                 return -EFAULT;
3547                                 }
3548                         } else {
3549                                 ksym_addr = (unsigned long) prog->bpf_func;
3550                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3551                                         return -EFAULT;
3552                         }
3553                 } else {
3554                         info.jited_ksyms = 0;
3555                 }
3556         }
3557
3558         ulen = info.nr_jited_func_lens;
3559         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3560         if (ulen) {
3561                 if (bpf_dump_raw_ok(file->f_cred)) {
3562                         u32 __user *user_lens;
3563                         u32 func_len, i;
3564
3565                         /* copy the JITed image lengths for each function */
3566                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3567                         user_lens = u64_to_user_ptr(info.jited_func_lens);
3568                         if (prog->aux->func_cnt) {
3569                                 for (i = 0; i < ulen; i++) {
3570                                         func_len =
3571                                                 prog->aux->func[i]->jited_len;
3572                                         if (put_user(func_len, &user_lens[i]))
3573                                                 return -EFAULT;
3574                                 }
3575                         } else {
3576                                 func_len = prog->jited_len;
3577                                 if (put_user(func_len, &user_lens[0]))
3578                                         return -EFAULT;
3579                         }
3580                 } else {
3581                         info.jited_func_lens = 0;
3582                 }
3583         }
3584
3585         if (prog->aux->btf)
3586                 info.btf_id = btf_obj_id(prog->aux->btf);
3587
3588         ulen = info.nr_func_info;
3589         info.nr_func_info = prog->aux->func_info_cnt;
3590         if (info.nr_func_info && ulen) {
3591                 char __user *user_finfo;
3592
3593                 user_finfo = u64_to_user_ptr(info.func_info);
3594                 ulen = min_t(u32, info.nr_func_info, ulen);
3595                 if (copy_to_user(user_finfo, prog->aux->func_info,
3596                                  info.func_info_rec_size * ulen))
3597                         return -EFAULT;
3598         }
3599
3600         ulen = info.nr_line_info;
3601         info.nr_line_info = prog->aux->nr_linfo;
3602         if (info.nr_line_info && ulen) {
3603                 __u8 __user *user_linfo;
3604
3605                 user_linfo = u64_to_user_ptr(info.line_info);
3606                 ulen = min_t(u32, info.nr_line_info, ulen);
3607                 if (copy_to_user(user_linfo, prog->aux->linfo,
3608                                  info.line_info_rec_size * ulen))
3609                         return -EFAULT;
3610         }
3611
3612         ulen = info.nr_jited_line_info;
3613         if (prog->aux->jited_linfo)
3614                 info.nr_jited_line_info = prog->aux->nr_linfo;
3615         else
3616                 info.nr_jited_line_info = 0;
3617         if (info.nr_jited_line_info && ulen) {
3618                 if (bpf_dump_raw_ok(file->f_cred)) {
3619                         __u64 __user *user_linfo;
3620                         u32 i;
3621
3622                         user_linfo = u64_to_user_ptr(info.jited_line_info);
3623                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
3624                         for (i = 0; i < ulen; i++) {
3625                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3626                                              &user_linfo[i]))
3627                                         return -EFAULT;
3628                         }
3629                 } else {
3630                         info.jited_line_info = 0;
3631                 }
3632         }
3633
3634         ulen = info.nr_prog_tags;
3635         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3636         if (ulen) {
3637                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3638                 u32 i;
3639
3640                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3641                 ulen = min_t(u32, info.nr_prog_tags, ulen);
3642                 if (prog->aux->func_cnt) {
3643                         for (i = 0; i < ulen; i++) {
3644                                 if (copy_to_user(user_prog_tags[i],
3645                                                  prog->aux->func[i]->tag,
3646                                                  BPF_TAG_SIZE))
3647                                         return -EFAULT;
3648                         }
3649                 } else {
3650                         if (copy_to_user(user_prog_tags[0],
3651                                          prog->tag, BPF_TAG_SIZE))
3652                                 return -EFAULT;
3653                 }
3654         }
3655
3656 done:
3657         if (copy_to_user(uinfo, &info, info_len) ||
3658             put_user(info_len, &uattr->info.info_len))
3659                 return -EFAULT;
3660
3661         return 0;
3662 }
3663
3664 static int bpf_map_get_info_by_fd(struct file *file,
3665                                   struct bpf_map *map,
3666                                   const union bpf_attr *attr,
3667                                   union bpf_attr __user *uattr)
3668 {
3669         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3670         struct bpf_map_info info;
3671         u32 info_len = attr->info.info_len;
3672         int err;
3673
3674         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3675         if (err)
3676                 return err;
3677         info_len = min_t(u32, sizeof(info), info_len);
3678
3679         memset(&info, 0, sizeof(info));
3680         info.type = map->map_type;
3681         info.id = map->id;
3682         info.key_size = map->key_size;
3683         info.value_size = map->value_size;
3684         info.max_entries = map->max_entries;
3685         info.map_flags = map->map_flags;
3686         memcpy(info.name, map->name, sizeof(map->name));
3687
3688         if (map->btf) {
3689                 info.btf_id = btf_obj_id(map->btf);
3690                 info.btf_key_type_id = map->btf_key_type_id;
3691                 info.btf_value_type_id = map->btf_value_type_id;
3692         }
3693         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3694
3695         if (bpf_map_is_dev_bound(map)) {
3696                 err = bpf_map_offload_info_fill(&info, map);
3697                 if (err)
3698                         return err;
3699         }
3700
3701         if (copy_to_user(uinfo, &info, info_len) ||
3702             put_user(info_len, &uattr->info.info_len))
3703                 return -EFAULT;
3704
3705         return 0;
3706 }
3707
3708 static int bpf_btf_get_info_by_fd(struct file *file,
3709                                   struct btf *btf,
3710                                   const union bpf_attr *attr,
3711                                   union bpf_attr __user *uattr)
3712 {
3713         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3714         u32 info_len = attr->info.info_len;
3715         int err;
3716
3717         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3718         if (err)
3719                 return err;
3720
3721         return btf_get_info_by_fd(btf, attr, uattr);
3722 }
3723
3724 static int bpf_link_get_info_by_fd(struct file *file,
3725                                   struct bpf_link *link,
3726                                   const union bpf_attr *attr,
3727                                   union bpf_attr __user *uattr)
3728 {
3729         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3730         struct bpf_link_info info;
3731         u32 info_len = attr->info.info_len;
3732         int err;
3733
3734         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3735         if (err)
3736                 return err;
3737         info_len = min_t(u32, sizeof(info), info_len);
3738
3739         memset(&info, 0, sizeof(info));
3740         if (copy_from_user(&info, uinfo, info_len))
3741                 return -EFAULT;
3742
3743         info.type = link->type;
3744         info.id = link->id;
3745         info.prog_id = link->prog->aux->id;
3746
3747         if (link->ops->fill_link_info) {
3748                 err = link->ops->fill_link_info(link, &info);
3749                 if (err)
3750                         return err;
3751         }
3752
3753         if (copy_to_user(uinfo, &info, info_len) ||
3754             put_user(info_len, &uattr->info.info_len))
3755                 return -EFAULT;
3756
3757         return 0;
3758 }
3759
3760
3761 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3762
3763 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3764                                   union bpf_attr __user *uattr)
3765 {
3766         int ufd = attr->info.bpf_fd;
3767         struct fd f;
3768         int err;
3769
3770         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3771                 return -EINVAL;
3772
3773         f = fdget(ufd);
3774         if (!f.file)
3775                 return -EBADFD;
3776
3777         if (f.file->f_op == &bpf_prog_fops)
3778                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3779                                               uattr);
3780         else if (f.file->f_op == &bpf_map_fops)
3781                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3782                                              uattr);
3783         else if (f.file->f_op == &btf_fops)
3784                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3785         else if (f.file->f_op == &bpf_link_fops)
3786                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3787                                               attr, uattr);
3788         else
3789                 err = -EINVAL;
3790
3791         fdput(f);
3792         return err;
3793 }
3794
3795 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3796
3797 static int bpf_btf_load(const union bpf_attr *attr)
3798 {
3799         if (CHECK_ATTR(BPF_BTF_LOAD))
3800                 return -EINVAL;
3801
3802         if (!bpf_capable())
3803                 return -EPERM;
3804
3805         return btf_new_fd(attr);
3806 }
3807
3808 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3809
3810 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3811 {
3812         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3813                 return -EINVAL;
3814
3815         if (!capable(CAP_SYS_ADMIN))
3816                 return -EPERM;
3817
3818         return btf_get_fd_by_id(attr->btf_id);
3819 }
3820
3821 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3822                                     union bpf_attr __user *uattr,
3823                                     u32 prog_id, u32 fd_type,
3824                                     const char *buf, u64 probe_offset,
3825                                     u64 probe_addr)
3826 {
3827         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3828         u32 len = buf ? strlen(buf) : 0, input_len;
3829         int err = 0;
3830
3831         if (put_user(len, &uattr->task_fd_query.buf_len))
3832                 return -EFAULT;
3833         input_len = attr->task_fd_query.buf_len;
3834         if (input_len && ubuf) {
3835                 if (!len) {
3836                         /* nothing to copy, just make ubuf NULL terminated */
3837                         char zero = '\0';
3838
3839                         if (put_user(zero, ubuf))
3840                                 return -EFAULT;
3841                 } else if (input_len >= len + 1) {
3842                         /* ubuf can hold the string with NULL terminator */
3843                         if (copy_to_user(ubuf, buf, len + 1))
3844                                 return -EFAULT;
3845                 } else {
3846                         /* ubuf cannot hold the string with NULL terminator,
3847                          * do a partial copy with NULL terminator.
3848                          */
3849                         char zero = '\0';
3850
3851                         err = -ENOSPC;
3852                         if (copy_to_user(ubuf, buf, input_len - 1))
3853                                 return -EFAULT;
3854                         if (put_user(zero, ubuf + input_len - 1))
3855                                 return -EFAULT;
3856                 }
3857         }
3858
3859         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3860             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3861             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3862             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3863                 return -EFAULT;
3864
3865         return err;
3866 }
3867
3868 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3869
3870 static int bpf_task_fd_query(const union bpf_attr *attr,
3871                              union bpf_attr __user *uattr)
3872 {
3873         pid_t pid = attr->task_fd_query.pid;
3874         u32 fd = attr->task_fd_query.fd;
3875         const struct perf_event *event;
3876         struct task_struct *task;
3877         struct file *file;
3878         int err;
3879
3880         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3881                 return -EINVAL;
3882
3883         if (!capable(CAP_SYS_ADMIN))
3884                 return -EPERM;
3885
3886         if (attr->task_fd_query.flags != 0)
3887                 return -EINVAL;
3888
3889         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3890         if (!task)
3891                 return -ENOENT;
3892
3893         err = 0;
3894         file = fget_task(task, fd);
3895         put_task_struct(task);
3896         if (!file)
3897                 return -EBADF;
3898
3899         if (file->f_op == &bpf_link_fops) {
3900                 struct bpf_link *link = file->private_data;
3901
3902                 if (link->ops == &bpf_raw_tp_link_lops) {
3903                         struct bpf_raw_tp_link *raw_tp =
3904                                 container_of(link, struct bpf_raw_tp_link, link);
3905                         struct bpf_raw_event_map *btp = raw_tp->btp;
3906
3907                         err = bpf_task_fd_query_copy(attr, uattr,
3908                                                      raw_tp->link.prog->aux->id,
3909                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
3910                                                      btp->tp->name, 0, 0);
3911                         goto put_file;
3912                 }
3913                 goto out_not_supp;
3914         }
3915
3916         event = perf_get_event(file);
3917         if (!IS_ERR(event)) {
3918                 u64 probe_offset, probe_addr;
3919                 u32 prog_id, fd_type;
3920                 const char *buf;
3921
3922                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3923                                               &buf, &probe_offset,
3924                                               &probe_addr);
3925                 if (!err)
3926                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3927                                                      fd_type, buf,
3928                                                      probe_offset,
3929                                                      probe_addr);
3930                 goto put_file;
3931         }
3932
3933 out_not_supp:
3934         err = -ENOTSUPP;
3935 put_file:
3936         fput(file);
3937         return err;
3938 }
3939
3940 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3941
3942 #define BPF_DO_BATCH(fn)                        \
3943         do {                                    \
3944                 if (!fn) {                      \
3945                         err = -ENOTSUPP;        \
3946                         goto err_put;           \
3947                 }                               \
3948                 err = fn(map, attr, uattr);     \
3949         } while (0)
3950
3951 static int bpf_map_do_batch(const union bpf_attr *attr,
3952                             union bpf_attr __user *uattr,
3953                             int cmd)
3954 {
3955         struct bpf_map *map;
3956         int err, ufd;
3957         struct fd f;
3958
3959         if (CHECK_ATTR(BPF_MAP_BATCH))
3960                 return -EINVAL;
3961
3962         ufd = attr->batch.map_fd;
3963         f = fdget(ufd);
3964         map = __bpf_map_get(f);
3965         if (IS_ERR(map))
3966                 return PTR_ERR(map);
3967
3968         if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3969              cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3970             !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3971                 err = -EPERM;
3972                 goto err_put;
3973         }
3974
3975         if (cmd != BPF_MAP_LOOKUP_BATCH &&
3976             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3977                 err = -EPERM;
3978                 goto err_put;
3979         }
3980
3981         if (cmd == BPF_MAP_LOOKUP_BATCH)
3982                 BPF_DO_BATCH(map->ops->map_lookup_batch);
3983         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3984                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3985         else if (cmd == BPF_MAP_UPDATE_BATCH)
3986                 BPF_DO_BATCH(map->ops->map_update_batch);
3987         else
3988                 BPF_DO_BATCH(map->ops->map_delete_batch);
3989
3990 err_put:
3991         fdput(f);
3992         return err;
3993 }
3994
3995 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3996 {
3997         if (attr->link_create.attach_type != prog->expected_attach_type)
3998                 return -EINVAL;
3999
4000         if (prog->expected_attach_type == BPF_TRACE_ITER)
4001                 return bpf_iter_link_attach(attr, prog);
4002         else if (prog->type == BPF_PROG_TYPE_EXT)
4003                 return bpf_tracing_prog_attach(prog,
4004                                                attr->link_create.target_fd,
4005                                                attr->link_create.target_btf_id);
4006         return -EINVAL;
4007 }
4008
4009 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4010 static int link_create(union bpf_attr *attr)
4011 {
4012         enum bpf_prog_type ptype;
4013         struct bpf_prog *prog;
4014         int ret;
4015
4016         if (CHECK_ATTR(BPF_LINK_CREATE))
4017                 return -EINVAL;
4018
4019         prog = bpf_prog_get(attr->link_create.prog_fd);
4020         if (IS_ERR(prog))
4021                 return PTR_ERR(prog);
4022
4023         ret = bpf_prog_attach_check_attach_type(prog,
4024                                                 attr->link_create.attach_type);
4025         if (ret)
4026                 goto out;
4027
4028         if (prog->type == BPF_PROG_TYPE_EXT) {
4029                 ret = tracing_bpf_link_attach(attr, prog);
4030                 goto out;
4031         }
4032
4033         ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4034         if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4035                 ret = -EINVAL;
4036                 goto out;
4037         }
4038
4039         switch (ptype) {
4040         case BPF_PROG_TYPE_CGROUP_SKB:
4041         case BPF_PROG_TYPE_CGROUP_SOCK:
4042         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4043         case BPF_PROG_TYPE_SOCK_OPS:
4044         case BPF_PROG_TYPE_CGROUP_DEVICE:
4045         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4046         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4047                 ret = cgroup_bpf_link_attach(attr, prog);
4048                 break;
4049         case BPF_PROG_TYPE_TRACING:
4050                 ret = tracing_bpf_link_attach(attr, prog);
4051                 break;
4052         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4053         case BPF_PROG_TYPE_SK_LOOKUP:
4054                 ret = netns_bpf_link_create(attr, prog);
4055                 break;
4056 #ifdef CONFIG_NET
4057         case BPF_PROG_TYPE_XDP:
4058                 ret = bpf_xdp_link_attach(attr, prog);
4059                 break;
4060 #endif
4061         default:
4062                 ret = -EINVAL;
4063         }
4064
4065 out:
4066         if (ret < 0)
4067                 bpf_prog_put(prog);
4068         return ret;
4069 }
4070
4071 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4072
4073 static int link_update(union bpf_attr *attr)
4074 {
4075         struct bpf_prog *old_prog = NULL, *new_prog;
4076         struct bpf_link *link;
4077         u32 flags;
4078         int ret;
4079
4080         if (CHECK_ATTR(BPF_LINK_UPDATE))
4081                 return -EINVAL;
4082
4083         flags = attr->link_update.flags;
4084         if (flags & ~BPF_F_REPLACE)
4085                 return -EINVAL;
4086
4087         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4088         if (IS_ERR(link))
4089                 return PTR_ERR(link);
4090
4091         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4092         if (IS_ERR(new_prog)) {
4093                 ret = PTR_ERR(new_prog);
4094                 goto out_put_link;
4095         }
4096
4097         if (flags & BPF_F_REPLACE) {
4098                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4099                 if (IS_ERR(old_prog)) {
4100                         ret = PTR_ERR(old_prog);
4101                         old_prog = NULL;
4102                         goto out_put_progs;
4103                 }
4104         } else if (attr->link_update.old_prog_fd) {
4105                 ret = -EINVAL;
4106                 goto out_put_progs;
4107         }
4108
4109         if (link->ops->update_prog)
4110                 ret = link->ops->update_prog(link, new_prog, old_prog);
4111         else
4112                 ret = -EINVAL;
4113
4114 out_put_progs:
4115         if (old_prog)
4116                 bpf_prog_put(old_prog);
4117         if (ret)
4118                 bpf_prog_put(new_prog);
4119 out_put_link:
4120         bpf_link_put(link);
4121         return ret;
4122 }
4123
4124 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4125
4126 static int link_detach(union bpf_attr *attr)
4127 {
4128         struct bpf_link *link;
4129         int ret;
4130
4131         if (CHECK_ATTR(BPF_LINK_DETACH))
4132                 return -EINVAL;
4133
4134         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4135         if (IS_ERR(link))
4136                 return PTR_ERR(link);
4137
4138         if (link->ops->detach)
4139                 ret = link->ops->detach(link);
4140         else
4141                 ret = -EOPNOTSUPP;
4142
4143         bpf_link_put(link);
4144         return ret;
4145 }
4146
4147 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4148 {
4149         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4150 }
4151
4152 struct bpf_link *bpf_link_by_id(u32 id)
4153 {
4154         struct bpf_link *link;
4155
4156         if (!id)
4157                 return ERR_PTR(-ENOENT);
4158
4159         spin_lock_bh(&link_idr_lock);
4160         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4161         link = idr_find(&link_idr, id);
4162         if (link) {
4163                 if (link->id)
4164                         link = bpf_link_inc_not_zero(link);
4165                 else
4166                         link = ERR_PTR(-EAGAIN);
4167         } else {
4168                 link = ERR_PTR(-ENOENT);
4169         }
4170         spin_unlock_bh(&link_idr_lock);
4171         return link;
4172 }
4173
4174 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4175
4176 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4177 {
4178         struct bpf_link *link;
4179         u32 id = attr->link_id;
4180         int fd;
4181
4182         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4183                 return -EINVAL;
4184
4185         if (!capable(CAP_SYS_ADMIN))
4186                 return -EPERM;
4187
4188         link = bpf_link_by_id(id);
4189         if (IS_ERR(link))
4190                 return PTR_ERR(link);
4191
4192         fd = bpf_link_new_fd(link);
4193         if (fd < 0)
4194                 bpf_link_put(link);
4195
4196         return fd;
4197 }
4198
4199 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4200
4201 static int bpf_stats_release(struct inode *inode, struct file *file)
4202 {
4203         mutex_lock(&bpf_stats_enabled_mutex);
4204         static_key_slow_dec(&bpf_stats_enabled_key.key);
4205         mutex_unlock(&bpf_stats_enabled_mutex);
4206         return 0;
4207 }
4208
4209 static const struct file_operations bpf_stats_fops = {
4210         .release = bpf_stats_release,
4211 };
4212
4213 static int bpf_enable_runtime_stats(void)
4214 {
4215         int fd;
4216
4217         mutex_lock(&bpf_stats_enabled_mutex);
4218
4219         /* Set a very high limit to avoid overflow */
4220         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4221                 mutex_unlock(&bpf_stats_enabled_mutex);
4222                 return -EBUSY;
4223         }
4224
4225         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4226         if (fd >= 0)
4227                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4228
4229         mutex_unlock(&bpf_stats_enabled_mutex);
4230         return fd;
4231 }
4232
4233 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4234
4235 static int bpf_enable_stats(union bpf_attr *attr)
4236 {
4237
4238         if (CHECK_ATTR(BPF_ENABLE_STATS))
4239                 return -EINVAL;
4240
4241         if (!capable(CAP_SYS_ADMIN))
4242                 return -EPERM;
4243
4244         switch (attr->enable_stats.type) {
4245         case BPF_STATS_RUN_TIME:
4246                 return bpf_enable_runtime_stats();
4247         default:
4248                 break;
4249         }
4250         return -EINVAL;
4251 }
4252
4253 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4254
4255 static int bpf_iter_create(union bpf_attr *attr)
4256 {
4257         struct bpf_link *link;
4258         int err;
4259
4260         if (CHECK_ATTR(BPF_ITER_CREATE))
4261                 return -EINVAL;
4262
4263         if (attr->iter_create.flags)
4264                 return -EINVAL;
4265
4266         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4267         if (IS_ERR(link))
4268                 return PTR_ERR(link);
4269
4270         err = bpf_iter_new_fd(link);
4271         bpf_link_put(link);
4272
4273         return err;
4274 }
4275
4276 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4277
4278 static int bpf_prog_bind_map(union bpf_attr *attr)
4279 {
4280         struct bpf_prog *prog;
4281         struct bpf_map *map;
4282         struct bpf_map **used_maps_old, **used_maps_new;
4283         int i, ret = 0;
4284
4285         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4286                 return -EINVAL;
4287
4288         if (attr->prog_bind_map.flags)
4289                 return -EINVAL;
4290
4291         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4292         if (IS_ERR(prog))
4293                 return PTR_ERR(prog);
4294
4295         map = bpf_map_get(attr->prog_bind_map.map_fd);
4296         if (IS_ERR(map)) {
4297                 ret = PTR_ERR(map);
4298                 goto out_prog_put;
4299         }
4300
4301         mutex_lock(&prog->aux->used_maps_mutex);
4302
4303         used_maps_old = prog->aux->used_maps;
4304
4305         for (i = 0; i < prog->aux->used_map_cnt; i++)
4306                 if (used_maps_old[i] == map) {
4307                         bpf_map_put(map);
4308                         goto out_unlock;
4309                 }
4310
4311         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4312                                       sizeof(used_maps_new[0]),
4313                                       GFP_KERNEL);
4314         if (!used_maps_new) {
4315                 ret = -ENOMEM;
4316                 goto out_unlock;
4317         }
4318
4319         memcpy(used_maps_new, used_maps_old,
4320                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4321         used_maps_new[prog->aux->used_map_cnt] = map;
4322
4323         prog->aux->used_map_cnt++;
4324         prog->aux->used_maps = used_maps_new;
4325
4326         kfree(used_maps_old);
4327
4328 out_unlock:
4329         mutex_unlock(&prog->aux->used_maps_mutex);
4330
4331         if (ret)
4332                 bpf_map_put(map);
4333 out_prog_put:
4334         bpf_prog_put(prog);
4335         return ret;
4336 }
4337
4338 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4339 {
4340         union bpf_attr attr;
4341         int err;
4342
4343         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4344                 return -EPERM;
4345
4346         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4347         if (err)
4348                 return err;
4349         size = min_t(u32, size, sizeof(attr));
4350
4351         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4352         memset(&attr, 0, sizeof(attr));
4353         if (copy_from_user(&attr, uattr, size) != 0)
4354                 return -EFAULT;
4355
4356         err = security_bpf(cmd, &attr, size);
4357         if (err < 0)
4358                 return err;
4359
4360         switch (cmd) {
4361         case BPF_MAP_CREATE:
4362                 err = map_create(&attr);
4363                 break;
4364         case BPF_MAP_LOOKUP_ELEM:
4365                 err = map_lookup_elem(&attr);
4366                 break;
4367         case BPF_MAP_UPDATE_ELEM:
4368                 err = map_update_elem(&attr);
4369                 break;
4370         case BPF_MAP_DELETE_ELEM:
4371                 err = map_delete_elem(&attr);
4372                 break;
4373         case BPF_MAP_GET_NEXT_KEY:
4374                 err = map_get_next_key(&attr);
4375                 break;
4376         case BPF_MAP_FREEZE:
4377                 err = map_freeze(&attr);
4378                 break;
4379         case BPF_PROG_LOAD:
4380                 err = bpf_prog_load(&attr, uattr);
4381                 break;
4382         case BPF_OBJ_PIN:
4383                 err = bpf_obj_pin(&attr);
4384                 break;
4385         case BPF_OBJ_GET:
4386                 err = bpf_obj_get(&attr);
4387                 break;
4388         case BPF_PROG_ATTACH:
4389                 err = bpf_prog_attach(&attr);
4390                 break;
4391         case BPF_PROG_DETACH:
4392                 err = bpf_prog_detach(&attr);
4393                 break;
4394         case BPF_PROG_QUERY:
4395                 err = bpf_prog_query(&attr, uattr);
4396                 break;
4397         case BPF_PROG_TEST_RUN:
4398                 err = bpf_prog_test_run(&attr, uattr);
4399                 break;
4400         case BPF_PROG_GET_NEXT_ID:
4401                 err = bpf_obj_get_next_id(&attr, uattr,
4402                                           &prog_idr, &prog_idr_lock);
4403                 break;
4404         case BPF_MAP_GET_NEXT_ID:
4405                 err = bpf_obj_get_next_id(&attr, uattr,
4406                                           &map_idr, &map_idr_lock);
4407                 break;
4408         case BPF_BTF_GET_NEXT_ID:
4409                 err = bpf_obj_get_next_id(&attr, uattr,
4410                                           &btf_idr, &btf_idr_lock);
4411                 break;
4412         case BPF_PROG_GET_FD_BY_ID:
4413                 err = bpf_prog_get_fd_by_id(&attr);
4414                 break;
4415         case BPF_MAP_GET_FD_BY_ID:
4416                 err = bpf_map_get_fd_by_id(&attr);
4417                 break;
4418         case BPF_OBJ_GET_INFO_BY_FD:
4419                 err = bpf_obj_get_info_by_fd(&attr, uattr);
4420                 break;
4421         case BPF_RAW_TRACEPOINT_OPEN:
4422                 err = bpf_raw_tracepoint_open(&attr);
4423                 break;
4424         case BPF_BTF_LOAD:
4425                 err = bpf_btf_load(&attr);
4426                 break;
4427         case BPF_BTF_GET_FD_BY_ID:
4428                 err = bpf_btf_get_fd_by_id(&attr);
4429                 break;
4430         case BPF_TASK_FD_QUERY:
4431                 err = bpf_task_fd_query(&attr, uattr);
4432                 break;
4433         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4434                 err = map_lookup_and_delete_elem(&attr);
4435                 break;
4436         case BPF_MAP_LOOKUP_BATCH:
4437                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4438                 break;
4439         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4440                 err = bpf_map_do_batch(&attr, uattr,
4441                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4442                 break;
4443         case BPF_MAP_UPDATE_BATCH:
4444                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4445                 break;
4446         case BPF_MAP_DELETE_BATCH:
4447                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4448                 break;
4449         case BPF_LINK_CREATE:
4450                 err = link_create(&attr);
4451                 break;
4452         case BPF_LINK_UPDATE:
4453                 err = link_update(&attr);
4454                 break;
4455         case BPF_LINK_GET_FD_BY_ID:
4456                 err = bpf_link_get_fd_by_id(&attr);
4457                 break;
4458         case BPF_LINK_GET_NEXT_ID:
4459                 err = bpf_obj_get_next_id(&attr, uattr,
4460                                           &link_idr, &link_idr_lock);
4461                 break;
4462         case BPF_ENABLE_STATS:
4463                 err = bpf_enable_stats(&attr);
4464                 break;
4465         case BPF_ITER_CREATE:
4466                 err = bpf_iter_create(&attr);
4467                 break;
4468         case BPF_LINK_DETACH:
4469                 err = link_detach(&attr);
4470                 break;
4471         case BPF_PROG_BIND_MAP:
4472                 err = bpf_prog_bind_map(&attr);
4473                 break;
4474         default:
4475                 err = -EINVAL;
4476                 break;
4477         }
4478
4479         return err;
4480 }