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, misses = 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, tmisses;
1741
1742                 st = per_cpu_ptr(prog->stats, cpu);
1743                 do {
1744                         start = u64_stats_fetch_begin_irq(&st->syncp);
1745                         tnsecs = st->nsecs;
1746                         tcnt = st->cnt;
1747                         tmisses = st->misses;
1748                 } while (u64_stats_fetch_retry_irq(&st->syncp, start));
1749                 nsecs += tnsecs;
1750                 cnt += tcnt;
1751                 misses += tmisses;
1752         }
1753         stats->nsecs = nsecs;
1754         stats->cnt = cnt;
1755         stats->misses = misses;
1756 }
1757
1758 #ifdef CONFIG_PROC_FS
1759 static void bpf_prog_show_fdinfo(struct seq_file *m, struct file *filp)
1760 {
1761         const struct bpf_prog *prog = filp->private_data;
1762         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
1763         struct bpf_prog_stats stats;
1764
1765         bpf_prog_get_stats(prog, &stats);
1766         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
1767         seq_printf(m,
1768                    "prog_type:\t%u\n"
1769                    "prog_jited:\t%u\n"
1770                    "prog_tag:\t%s\n"
1771                    "memlock:\t%llu\n"
1772                    "prog_id:\t%u\n"
1773                    "run_time_ns:\t%llu\n"
1774                    "run_cnt:\t%llu\n"
1775                    "recursion_misses:\t%llu\n",
1776                    prog->type,
1777                    prog->jited,
1778                    prog_tag,
1779                    prog->pages * 1ULL << PAGE_SHIFT,
1780                    prog->aux->id,
1781                    stats.nsecs,
1782                    stats.cnt,
1783                    stats.misses);
1784 }
1785 #endif
1786
1787 const struct file_operations bpf_prog_fops = {
1788 #ifdef CONFIG_PROC_FS
1789         .show_fdinfo    = bpf_prog_show_fdinfo,
1790 #endif
1791         .release        = bpf_prog_release,
1792         .read           = bpf_dummy_read,
1793         .write          = bpf_dummy_write,
1794 };
1795
1796 int bpf_prog_new_fd(struct bpf_prog *prog)
1797 {
1798         int ret;
1799
1800         ret = security_bpf_prog(prog);
1801         if (ret < 0)
1802                 return ret;
1803
1804         return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
1805                                 O_RDWR | O_CLOEXEC);
1806 }
1807
1808 static struct bpf_prog *____bpf_prog_get(struct fd f)
1809 {
1810         if (!f.file)
1811                 return ERR_PTR(-EBADF);
1812         if (f.file->f_op != &bpf_prog_fops) {
1813                 fdput(f);
1814                 return ERR_PTR(-EINVAL);
1815         }
1816
1817         return f.file->private_data;
1818 }
1819
1820 void bpf_prog_add(struct bpf_prog *prog, int i)
1821 {
1822         atomic64_add(i, &prog->aux->refcnt);
1823 }
1824 EXPORT_SYMBOL_GPL(bpf_prog_add);
1825
1826 void bpf_prog_sub(struct bpf_prog *prog, int i)
1827 {
1828         /* Only to be used for undoing previous bpf_prog_add() in some
1829          * error path. We still know that another entity in our call
1830          * path holds a reference to the program, thus atomic_sub() can
1831          * be safely used in such cases!
1832          */
1833         WARN_ON(atomic64_sub_return(i, &prog->aux->refcnt) == 0);
1834 }
1835 EXPORT_SYMBOL_GPL(bpf_prog_sub);
1836
1837 void bpf_prog_inc(struct bpf_prog *prog)
1838 {
1839         atomic64_inc(&prog->aux->refcnt);
1840 }
1841 EXPORT_SYMBOL_GPL(bpf_prog_inc);
1842
1843 /* prog_idr_lock should have been held */
1844 struct bpf_prog *bpf_prog_inc_not_zero(struct bpf_prog *prog)
1845 {
1846         int refold;
1847
1848         refold = atomic64_fetch_add_unless(&prog->aux->refcnt, 1, 0);
1849
1850         if (!refold)
1851                 return ERR_PTR(-ENOENT);
1852
1853         return prog;
1854 }
1855 EXPORT_SYMBOL_GPL(bpf_prog_inc_not_zero);
1856
1857 bool bpf_prog_get_ok(struct bpf_prog *prog,
1858                             enum bpf_prog_type *attach_type, bool attach_drv)
1859 {
1860         /* not an attachment, just a refcount inc, always allow */
1861         if (!attach_type)
1862                 return true;
1863
1864         if (prog->type != *attach_type)
1865                 return false;
1866         if (bpf_prog_is_dev_bound(prog->aux) && !attach_drv)
1867                 return false;
1868
1869         return true;
1870 }
1871
1872 static struct bpf_prog *__bpf_prog_get(u32 ufd, enum bpf_prog_type *attach_type,
1873                                        bool attach_drv)
1874 {
1875         struct fd f = fdget(ufd);
1876         struct bpf_prog *prog;
1877
1878         prog = ____bpf_prog_get(f);
1879         if (IS_ERR(prog))
1880                 return prog;
1881         if (!bpf_prog_get_ok(prog, attach_type, attach_drv)) {
1882                 prog = ERR_PTR(-EINVAL);
1883                 goto out;
1884         }
1885
1886         bpf_prog_inc(prog);
1887 out:
1888         fdput(f);
1889         return prog;
1890 }
1891
1892 struct bpf_prog *bpf_prog_get(u32 ufd)
1893 {
1894         return __bpf_prog_get(ufd, NULL, false);
1895 }
1896
1897 struct bpf_prog *bpf_prog_get_type_dev(u32 ufd, enum bpf_prog_type type,
1898                                        bool attach_drv)
1899 {
1900         return __bpf_prog_get(ufd, &type, attach_drv);
1901 }
1902 EXPORT_SYMBOL_GPL(bpf_prog_get_type_dev);
1903
1904 /* Initially all BPF programs could be loaded w/o specifying
1905  * expected_attach_type. Later for some of them specifying expected_attach_type
1906  * at load time became required so that program could be validated properly.
1907  * Programs of types that are allowed to be loaded both w/ and w/o (for
1908  * backward compatibility) expected_attach_type, should have the default attach
1909  * type assigned to expected_attach_type for the latter case, so that it can be
1910  * validated later at attach time.
1911  *
1912  * bpf_prog_load_fixup_attach_type() sets expected_attach_type in @attr if
1913  * prog type requires it but has some attach types that have to be backward
1914  * compatible.
1915  */
1916 static void bpf_prog_load_fixup_attach_type(union bpf_attr *attr)
1917 {
1918         switch (attr->prog_type) {
1919         case BPF_PROG_TYPE_CGROUP_SOCK:
1920                 /* Unfortunately BPF_ATTACH_TYPE_UNSPEC enumeration doesn't
1921                  * exist so checking for non-zero is the way to go here.
1922                  */
1923                 if (!attr->expected_attach_type)
1924                         attr->expected_attach_type =
1925                                 BPF_CGROUP_INET_SOCK_CREATE;
1926                 break;
1927         }
1928 }
1929
1930 static int
1931 bpf_prog_load_check_attach(enum bpf_prog_type prog_type,
1932                            enum bpf_attach_type expected_attach_type,
1933                            struct btf *attach_btf, u32 btf_id,
1934                            struct bpf_prog *dst_prog)
1935 {
1936         if (btf_id) {
1937                 if (btf_id > BTF_MAX_TYPE)
1938                         return -EINVAL;
1939
1940                 if (!attach_btf && !dst_prog)
1941                         return -EINVAL;
1942
1943                 switch (prog_type) {
1944                 case BPF_PROG_TYPE_TRACING:
1945                 case BPF_PROG_TYPE_LSM:
1946                 case BPF_PROG_TYPE_STRUCT_OPS:
1947                 case BPF_PROG_TYPE_EXT:
1948                         break;
1949                 default:
1950                         return -EINVAL;
1951                 }
1952         }
1953
1954         if (attach_btf && (!btf_id || dst_prog))
1955                 return -EINVAL;
1956
1957         if (dst_prog && prog_type != BPF_PROG_TYPE_TRACING &&
1958             prog_type != BPF_PROG_TYPE_EXT)
1959                 return -EINVAL;
1960
1961         switch (prog_type) {
1962         case BPF_PROG_TYPE_CGROUP_SOCK:
1963                 switch (expected_attach_type) {
1964                 case BPF_CGROUP_INET_SOCK_CREATE:
1965                 case BPF_CGROUP_INET_SOCK_RELEASE:
1966                 case BPF_CGROUP_INET4_POST_BIND:
1967                 case BPF_CGROUP_INET6_POST_BIND:
1968                         return 0;
1969                 default:
1970                         return -EINVAL;
1971                 }
1972         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
1973                 switch (expected_attach_type) {
1974                 case BPF_CGROUP_INET4_BIND:
1975                 case BPF_CGROUP_INET6_BIND:
1976                 case BPF_CGROUP_INET4_CONNECT:
1977                 case BPF_CGROUP_INET6_CONNECT:
1978                 case BPF_CGROUP_INET4_GETPEERNAME:
1979                 case BPF_CGROUP_INET6_GETPEERNAME:
1980                 case BPF_CGROUP_INET4_GETSOCKNAME:
1981                 case BPF_CGROUP_INET6_GETSOCKNAME:
1982                 case BPF_CGROUP_UDP4_SENDMSG:
1983                 case BPF_CGROUP_UDP6_SENDMSG:
1984                 case BPF_CGROUP_UDP4_RECVMSG:
1985                 case BPF_CGROUP_UDP6_RECVMSG:
1986                         return 0;
1987                 default:
1988                         return -EINVAL;
1989                 }
1990         case BPF_PROG_TYPE_CGROUP_SKB:
1991                 switch (expected_attach_type) {
1992                 case BPF_CGROUP_INET_INGRESS:
1993                 case BPF_CGROUP_INET_EGRESS:
1994                         return 0;
1995                 default:
1996                         return -EINVAL;
1997                 }
1998         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
1999                 switch (expected_attach_type) {
2000                 case BPF_CGROUP_SETSOCKOPT:
2001                 case BPF_CGROUP_GETSOCKOPT:
2002                         return 0;
2003                 default:
2004                         return -EINVAL;
2005                 }
2006         case BPF_PROG_TYPE_SK_LOOKUP:
2007                 if (expected_attach_type == BPF_SK_LOOKUP)
2008                         return 0;
2009                 return -EINVAL;
2010         case BPF_PROG_TYPE_EXT:
2011                 if (expected_attach_type)
2012                         return -EINVAL;
2013                 fallthrough;
2014         default:
2015                 return 0;
2016         }
2017 }
2018
2019 static bool is_net_admin_prog_type(enum bpf_prog_type prog_type)
2020 {
2021         switch (prog_type) {
2022         case BPF_PROG_TYPE_SCHED_CLS:
2023         case BPF_PROG_TYPE_SCHED_ACT:
2024         case BPF_PROG_TYPE_XDP:
2025         case BPF_PROG_TYPE_LWT_IN:
2026         case BPF_PROG_TYPE_LWT_OUT:
2027         case BPF_PROG_TYPE_LWT_XMIT:
2028         case BPF_PROG_TYPE_LWT_SEG6LOCAL:
2029         case BPF_PROG_TYPE_SK_SKB:
2030         case BPF_PROG_TYPE_SK_MSG:
2031         case BPF_PROG_TYPE_LIRC_MODE2:
2032         case BPF_PROG_TYPE_FLOW_DISSECTOR:
2033         case BPF_PROG_TYPE_CGROUP_DEVICE:
2034         case BPF_PROG_TYPE_CGROUP_SOCK:
2035         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2036         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2037         case BPF_PROG_TYPE_CGROUP_SYSCTL:
2038         case BPF_PROG_TYPE_SOCK_OPS:
2039         case BPF_PROG_TYPE_EXT: /* extends any prog */
2040                 return true;
2041         case BPF_PROG_TYPE_CGROUP_SKB:
2042                 /* always unpriv */
2043         case BPF_PROG_TYPE_SK_REUSEPORT:
2044                 /* equivalent to SOCKET_FILTER. need CAP_BPF only */
2045         default:
2046                 return false;
2047         }
2048 }
2049
2050 static bool is_perfmon_prog_type(enum bpf_prog_type prog_type)
2051 {
2052         switch (prog_type) {
2053         case BPF_PROG_TYPE_KPROBE:
2054         case BPF_PROG_TYPE_TRACEPOINT:
2055         case BPF_PROG_TYPE_PERF_EVENT:
2056         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2057         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2058         case BPF_PROG_TYPE_TRACING:
2059         case BPF_PROG_TYPE_LSM:
2060         case BPF_PROG_TYPE_STRUCT_OPS: /* has access to struct sock */
2061         case BPF_PROG_TYPE_EXT: /* extends any prog */
2062                 return true;
2063         default:
2064                 return false;
2065         }
2066 }
2067
2068 /* last field in 'union bpf_attr' used by this command */
2069 #define BPF_PROG_LOAD_LAST_FIELD attach_prog_fd
2070
2071 static int bpf_prog_load(union bpf_attr *attr, union bpf_attr __user *uattr)
2072 {
2073         enum bpf_prog_type type = attr->prog_type;
2074         struct bpf_prog *prog, *dst_prog = NULL;
2075         struct btf *attach_btf = NULL;
2076         int err;
2077         char license[128];
2078         bool is_gpl;
2079
2080         if (CHECK_ATTR(BPF_PROG_LOAD))
2081                 return -EINVAL;
2082
2083         if (attr->prog_flags & ~(BPF_F_STRICT_ALIGNMENT |
2084                                  BPF_F_ANY_ALIGNMENT |
2085                                  BPF_F_TEST_STATE_FREQ |
2086                                  BPF_F_SLEEPABLE |
2087                                  BPF_F_TEST_RND_HI32))
2088                 return -EINVAL;
2089
2090         if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) &&
2091             (attr->prog_flags & BPF_F_ANY_ALIGNMENT) &&
2092             !bpf_capable())
2093                 return -EPERM;
2094
2095         /* copy eBPF program license from user space */
2096         if (strncpy_from_user(license, u64_to_user_ptr(attr->license),
2097                               sizeof(license) - 1) < 0)
2098                 return -EFAULT;
2099         license[sizeof(license) - 1] = 0;
2100
2101         /* eBPF programs must be GPL compatible to use GPL-ed functions */
2102         is_gpl = license_is_gpl_compatible(license);
2103
2104         if (attr->insn_cnt == 0 ||
2105             attr->insn_cnt > (bpf_capable() ? BPF_COMPLEXITY_LIMIT_INSNS : BPF_MAXINSNS))
2106                 return -E2BIG;
2107         if (type != BPF_PROG_TYPE_SOCKET_FILTER &&
2108             type != BPF_PROG_TYPE_CGROUP_SKB &&
2109             !bpf_capable())
2110                 return -EPERM;
2111
2112         if (is_net_admin_prog_type(type) && !capable(CAP_NET_ADMIN) && !capable(CAP_SYS_ADMIN))
2113                 return -EPERM;
2114         if (is_perfmon_prog_type(type) && !perfmon_capable())
2115                 return -EPERM;
2116
2117         /* attach_prog_fd/attach_btf_obj_fd can specify fd of either bpf_prog
2118          * or btf, we need to check which one it is
2119          */
2120         if (attr->attach_prog_fd) {
2121                 dst_prog = bpf_prog_get(attr->attach_prog_fd);
2122                 if (IS_ERR(dst_prog)) {
2123                         dst_prog = NULL;
2124                         attach_btf = btf_get_by_fd(attr->attach_btf_obj_fd);
2125                         if (IS_ERR(attach_btf))
2126                                 return -EINVAL;
2127                         if (!btf_is_kernel(attach_btf)) {
2128                                 /* attaching through specifying bpf_prog's BTF
2129                                  * objects directly might be supported eventually
2130                                  */
2131                                 btf_put(attach_btf);
2132                                 return -ENOTSUPP;
2133                         }
2134                 }
2135         } else if (attr->attach_btf_id) {
2136                 /* fall back to vmlinux BTF, if BTF type ID is specified */
2137                 attach_btf = bpf_get_btf_vmlinux();
2138                 if (IS_ERR(attach_btf))
2139                         return PTR_ERR(attach_btf);
2140                 if (!attach_btf)
2141                         return -EINVAL;
2142                 btf_get(attach_btf);
2143         }
2144
2145         bpf_prog_load_fixup_attach_type(attr);
2146         if (bpf_prog_load_check_attach(type, attr->expected_attach_type,
2147                                        attach_btf, attr->attach_btf_id,
2148                                        dst_prog)) {
2149                 if (dst_prog)
2150                         bpf_prog_put(dst_prog);
2151                 if (attach_btf)
2152                         btf_put(attach_btf);
2153                 return -EINVAL;
2154         }
2155
2156         /* plain bpf_prog allocation */
2157         prog = bpf_prog_alloc(bpf_prog_size(attr->insn_cnt), GFP_USER);
2158         if (!prog) {
2159                 if (dst_prog)
2160                         bpf_prog_put(dst_prog);
2161                 if (attach_btf)
2162                         btf_put(attach_btf);
2163                 return -ENOMEM;
2164         }
2165
2166         prog->expected_attach_type = attr->expected_attach_type;
2167         prog->aux->attach_btf = attach_btf;
2168         prog->aux->attach_btf_id = attr->attach_btf_id;
2169         prog->aux->dst_prog = dst_prog;
2170         prog->aux->offload_requested = !!attr->prog_ifindex;
2171         prog->aux->sleepable = attr->prog_flags & BPF_F_SLEEPABLE;
2172
2173         err = security_bpf_prog_alloc(prog->aux);
2174         if (err)
2175                 goto free_prog;
2176
2177         prog->aux->user = get_current_user();
2178         prog->len = attr->insn_cnt;
2179
2180         err = -EFAULT;
2181         if (copy_from_user(prog->insns, u64_to_user_ptr(attr->insns),
2182                            bpf_prog_insn_size(prog)) != 0)
2183                 goto free_prog_sec;
2184
2185         prog->orig_prog = NULL;
2186         prog->jited = 0;
2187
2188         atomic64_set(&prog->aux->refcnt, 1);
2189         prog->gpl_compatible = is_gpl ? 1 : 0;
2190
2191         if (bpf_prog_is_dev_bound(prog->aux)) {
2192                 err = bpf_prog_offload_init(prog, attr);
2193                 if (err)
2194                         goto free_prog_sec;
2195         }
2196
2197         /* find program type: socket_filter vs tracing_filter */
2198         err = find_prog_type(type, prog);
2199         if (err < 0)
2200                 goto free_prog_sec;
2201
2202         prog->aux->load_time = ktime_get_boottime_ns();
2203         err = bpf_obj_name_cpy(prog->aux->name, attr->prog_name,
2204                                sizeof(attr->prog_name));
2205         if (err < 0)
2206                 goto free_prog_sec;
2207
2208         /* run eBPF verifier */
2209         err = bpf_check(&prog, attr, uattr);
2210         if (err < 0)
2211                 goto free_used_maps;
2212
2213         prog = bpf_prog_select_runtime(prog, &err);
2214         if (err < 0)
2215                 goto free_used_maps;
2216
2217         err = bpf_prog_alloc_id(prog);
2218         if (err)
2219                 goto free_used_maps;
2220
2221         /* Upon success of bpf_prog_alloc_id(), the BPF prog is
2222          * effectively publicly exposed. However, retrieving via
2223          * bpf_prog_get_fd_by_id() will take another reference,
2224          * therefore it cannot be gone underneath us.
2225          *
2226          * Only for the time /after/ successful bpf_prog_new_fd()
2227          * and before returning to userspace, we might just hold
2228          * one reference and any parallel close on that fd could
2229          * rip everything out. Hence, below notifications must
2230          * happen before bpf_prog_new_fd().
2231          *
2232          * Also, any failure handling from this point onwards must
2233          * be using bpf_prog_put() given the program is exposed.
2234          */
2235         bpf_prog_kallsyms_add(prog);
2236         perf_event_bpf_event(prog, PERF_BPF_EVENT_PROG_LOAD, 0);
2237         bpf_audit_prog(prog, BPF_AUDIT_LOAD);
2238
2239         err = bpf_prog_new_fd(prog);
2240         if (err < 0)
2241                 bpf_prog_put(prog);
2242         return err;
2243
2244 free_used_maps:
2245         /* In case we have subprogs, we need to wait for a grace
2246          * period before we can tear down JIT memory since symbols
2247          * are already exposed under kallsyms.
2248          */
2249         __bpf_prog_put_noref(prog, prog->aux->func_cnt);
2250         return err;
2251 free_prog_sec:
2252         free_uid(prog->aux->user);
2253         security_bpf_prog_free(prog->aux);
2254 free_prog:
2255         if (prog->aux->attach_btf)
2256                 btf_put(prog->aux->attach_btf);
2257         bpf_prog_free(prog);
2258         return err;
2259 }
2260
2261 #define BPF_OBJ_LAST_FIELD file_flags
2262
2263 static int bpf_obj_pin(const union bpf_attr *attr)
2264 {
2265         if (CHECK_ATTR(BPF_OBJ) || attr->file_flags != 0)
2266                 return -EINVAL;
2267
2268         return bpf_obj_pin_user(attr->bpf_fd, u64_to_user_ptr(attr->pathname));
2269 }
2270
2271 static int bpf_obj_get(const union bpf_attr *attr)
2272 {
2273         if (CHECK_ATTR(BPF_OBJ) || attr->bpf_fd != 0 ||
2274             attr->file_flags & ~BPF_OBJ_FLAG_MASK)
2275                 return -EINVAL;
2276
2277         return bpf_obj_get_user(u64_to_user_ptr(attr->pathname),
2278                                 attr->file_flags);
2279 }
2280
2281 void bpf_link_init(struct bpf_link *link, enum bpf_link_type type,
2282                    const struct bpf_link_ops *ops, struct bpf_prog *prog)
2283 {
2284         atomic64_set(&link->refcnt, 1);
2285         link->type = type;
2286         link->id = 0;
2287         link->ops = ops;
2288         link->prog = prog;
2289 }
2290
2291 static void bpf_link_free_id(int id)
2292 {
2293         if (!id)
2294                 return;
2295
2296         spin_lock_bh(&link_idr_lock);
2297         idr_remove(&link_idr, id);
2298         spin_unlock_bh(&link_idr_lock);
2299 }
2300
2301 /* Clean up bpf_link and corresponding anon_inode file and FD. After
2302  * anon_inode is created, bpf_link can't be just kfree()'d due to deferred
2303  * anon_inode's release() call. This helper marksbpf_link as
2304  * defunct, releases anon_inode file and puts reserved FD. bpf_prog's refcnt
2305  * is not decremented, it's the responsibility of a calling code that failed
2306  * to complete bpf_link initialization.
2307  */
2308 void bpf_link_cleanup(struct bpf_link_primer *primer)
2309 {
2310         primer->link->prog = NULL;
2311         bpf_link_free_id(primer->id);
2312         fput(primer->file);
2313         put_unused_fd(primer->fd);
2314 }
2315
2316 void bpf_link_inc(struct bpf_link *link)
2317 {
2318         atomic64_inc(&link->refcnt);
2319 }
2320
2321 /* bpf_link_free is guaranteed to be called from process context */
2322 static void bpf_link_free(struct bpf_link *link)
2323 {
2324         bpf_link_free_id(link->id);
2325         if (link->prog) {
2326                 /* detach BPF program, clean up used resources */
2327                 link->ops->release(link);
2328                 bpf_prog_put(link->prog);
2329         }
2330         /* free bpf_link and its containing memory */
2331         link->ops->dealloc(link);
2332 }
2333
2334 static void bpf_link_put_deferred(struct work_struct *work)
2335 {
2336         struct bpf_link *link = container_of(work, struct bpf_link, work);
2337
2338         bpf_link_free(link);
2339 }
2340
2341 /* bpf_link_put can be called from atomic context, but ensures that resources
2342  * are freed from process context
2343  */
2344 void bpf_link_put(struct bpf_link *link)
2345 {
2346         if (!atomic64_dec_and_test(&link->refcnt))
2347                 return;
2348
2349         if (in_atomic()) {
2350                 INIT_WORK(&link->work, bpf_link_put_deferred);
2351                 schedule_work(&link->work);
2352         } else {
2353                 bpf_link_free(link);
2354         }
2355 }
2356
2357 static int bpf_link_release(struct inode *inode, struct file *filp)
2358 {
2359         struct bpf_link *link = filp->private_data;
2360
2361         bpf_link_put(link);
2362         return 0;
2363 }
2364
2365 #ifdef CONFIG_PROC_FS
2366 #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type)
2367 #define BPF_MAP_TYPE(_id, _ops)
2368 #define BPF_LINK_TYPE(_id, _name) [_id] = #_name,
2369 static const char *bpf_link_type_strs[] = {
2370         [BPF_LINK_TYPE_UNSPEC] = "<invalid>",
2371 #include <linux/bpf_types.h>
2372 };
2373 #undef BPF_PROG_TYPE
2374 #undef BPF_MAP_TYPE
2375 #undef BPF_LINK_TYPE
2376
2377 static void bpf_link_show_fdinfo(struct seq_file *m, struct file *filp)
2378 {
2379         const struct bpf_link *link = filp->private_data;
2380         const struct bpf_prog *prog = link->prog;
2381         char prog_tag[sizeof(prog->tag) * 2 + 1] = { };
2382
2383         bin2hex(prog_tag, prog->tag, sizeof(prog->tag));
2384         seq_printf(m,
2385                    "link_type:\t%s\n"
2386                    "link_id:\t%u\n"
2387                    "prog_tag:\t%s\n"
2388                    "prog_id:\t%u\n",
2389                    bpf_link_type_strs[link->type],
2390                    link->id,
2391                    prog_tag,
2392                    prog->aux->id);
2393         if (link->ops->show_fdinfo)
2394                 link->ops->show_fdinfo(link, m);
2395 }
2396 #endif
2397
2398 static const struct file_operations bpf_link_fops = {
2399 #ifdef CONFIG_PROC_FS
2400         .show_fdinfo    = bpf_link_show_fdinfo,
2401 #endif
2402         .release        = bpf_link_release,
2403         .read           = bpf_dummy_read,
2404         .write          = bpf_dummy_write,
2405 };
2406
2407 static int bpf_link_alloc_id(struct bpf_link *link)
2408 {
2409         int id;
2410
2411         idr_preload(GFP_KERNEL);
2412         spin_lock_bh(&link_idr_lock);
2413         id = idr_alloc_cyclic(&link_idr, link, 1, INT_MAX, GFP_ATOMIC);
2414         spin_unlock_bh(&link_idr_lock);
2415         idr_preload_end();
2416
2417         return id;
2418 }
2419
2420 /* Prepare bpf_link to be exposed to user-space by allocating anon_inode file,
2421  * reserving unused FD and allocating ID from link_idr. This is to be paired
2422  * with bpf_link_settle() to install FD and ID and expose bpf_link to
2423  * user-space, if bpf_link is successfully attached. If not, bpf_link and
2424  * pre-allocated resources are to be freed with bpf_cleanup() call. All the
2425  * transient state is passed around in struct bpf_link_primer.
2426  * This is preferred way to create and initialize bpf_link, especially when
2427  * there are complicated and expensive operations inbetween creating bpf_link
2428  * itself and attaching it to BPF hook. By using bpf_link_prime() and
2429  * bpf_link_settle() kernel code using bpf_link doesn't have to perform
2430  * expensive (and potentially failing) roll back operations in a rare case
2431  * that file, FD, or ID can't be allocated.
2432  */
2433 int bpf_link_prime(struct bpf_link *link, struct bpf_link_primer *primer)
2434 {
2435         struct file *file;
2436         int fd, id;
2437
2438         fd = get_unused_fd_flags(O_CLOEXEC);
2439         if (fd < 0)
2440                 return fd;
2441
2442
2443         id = bpf_link_alloc_id(link);
2444         if (id < 0) {
2445                 put_unused_fd(fd);
2446                 return id;
2447         }
2448
2449         file = anon_inode_getfile("bpf_link", &bpf_link_fops, link, O_CLOEXEC);
2450         if (IS_ERR(file)) {
2451                 bpf_link_free_id(id);
2452                 put_unused_fd(fd);
2453                 return PTR_ERR(file);
2454         }
2455
2456         primer->link = link;
2457         primer->file = file;
2458         primer->fd = fd;
2459         primer->id = id;
2460         return 0;
2461 }
2462
2463 int bpf_link_settle(struct bpf_link_primer *primer)
2464 {
2465         /* make bpf_link fetchable by ID */
2466         spin_lock_bh(&link_idr_lock);
2467         primer->link->id = primer->id;
2468         spin_unlock_bh(&link_idr_lock);
2469         /* make bpf_link fetchable by FD */
2470         fd_install(primer->fd, primer->file);
2471         /* pass through installed FD */
2472         return primer->fd;
2473 }
2474
2475 int bpf_link_new_fd(struct bpf_link *link)
2476 {
2477         return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
2478 }
2479
2480 struct bpf_link *bpf_link_get_from_fd(u32 ufd)
2481 {
2482         struct fd f = fdget(ufd);
2483         struct bpf_link *link;
2484
2485         if (!f.file)
2486                 return ERR_PTR(-EBADF);
2487         if (f.file->f_op != &bpf_link_fops) {
2488                 fdput(f);
2489                 return ERR_PTR(-EINVAL);
2490         }
2491
2492         link = f.file->private_data;
2493         bpf_link_inc(link);
2494         fdput(f);
2495
2496         return link;
2497 }
2498
2499 struct bpf_tracing_link {
2500         struct bpf_link link;
2501         enum bpf_attach_type attach_type;
2502         struct bpf_trampoline *trampoline;
2503         struct bpf_prog *tgt_prog;
2504 };
2505
2506 static void bpf_tracing_link_release(struct bpf_link *link)
2507 {
2508         struct bpf_tracing_link *tr_link =
2509                 container_of(link, struct bpf_tracing_link, link);
2510
2511         WARN_ON_ONCE(bpf_trampoline_unlink_prog(link->prog,
2512                                                 tr_link->trampoline));
2513
2514         bpf_trampoline_put(tr_link->trampoline);
2515
2516         /* tgt_prog is NULL if target is a kernel function */
2517         if (tr_link->tgt_prog)
2518                 bpf_prog_put(tr_link->tgt_prog);
2519 }
2520
2521 static void bpf_tracing_link_dealloc(struct bpf_link *link)
2522 {
2523         struct bpf_tracing_link *tr_link =
2524                 container_of(link, struct bpf_tracing_link, link);
2525
2526         kfree(tr_link);
2527 }
2528
2529 static void bpf_tracing_link_show_fdinfo(const struct bpf_link *link,
2530                                          struct seq_file *seq)
2531 {
2532         struct bpf_tracing_link *tr_link =
2533                 container_of(link, struct bpf_tracing_link, link);
2534
2535         seq_printf(seq,
2536                    "attach_type:\t%d\n",
2537                    tr_link->attach_type);
2538 }
2539
2540 static int bpf_tracing_link_fill_link_info(const struct bpf_link *link,
2541                                            struct bpf_link_info *info)
2542 {
2543         struct bpf_tracing_link *tr_link =
2544                 container_of(link, struct bpf_tracing_link, link);
2545
2546         info->tracing.attach_type = tr_link->attach_type;
2547
2548         return 0;
2549 }
2550
2551 static const struct bpf_link_ops bpf_tracing_link_lops = {
2552         .release = bpf_tracing_link_release,
2553         .dealloc = bpf_tracing_link_dealloc,
2554         .show_fdinfo = bpf_tracing_link_show_fdinfo,
2555         .fill_link_info = bpf_tracing_link_fill_link_info,
2556 };
2557
2558 static int bpf_tracing_prog_attach(struct bpf_prog *prog,
2559                                    int tgt_prog_fd,
2560                                    u32 btf_id)
2561 {
2562         struct bpf_link_primer link_primer;
2563         struct bpf_prog *tgt_prog = NULL;
2564         struct bpf_trampoline *tr = NULL;
2565         struct bpf_tracing_link *link;
2566         u64 key = 0;
2567         int err;
2568
2569         switch (prog->type) {
2570         case BPF_PROG_TYPE_TRACING:
2571                 if (prog->expected_attach_type != BPF_TRACE_FENTRY &&
2572                     prog->expected_attach_type != BPF_TRACE_FEXIT &&
2573                     prog->expected_attach_type != BPF_MODIFY_RETURN) {
2574                         err = -EINVAL;
2575                         goto out_put_prog;
2576                 }
2577                 break;
2578         case BPF_PROG_TYPE_EXT:
2579                 if (prog->expected_attach_type != 0) {
2580                         err = -EINVAL;
2581                         goto out_put_prog;
2582                 }
2583                 break;
2584         case BPF_PROG_TYPE_LSM:
2585                 if (prog->expected_attach_type != BPF_LSM_MAC) {
2586                         err = -EINVAL;
2587                         goto out_put_prog;
2588                 }
2589                 break;
2590         default:
2591                 err = -EINVAL;
2592                 goto out_put_prog;
2593         }
2594
2595         if (!!tgt_prog_fd != !!btf_id) {
2596                 err = -EINVAL;
2597                 goto out_put_prog;
2598         }
2599
2600         if (tgt_prog_fd) {
2601                 /* For now we only allow new targets for BPF_PROG_TYPE_EXT */
2602                 if (prog->type != BPF_PROG_TYPE_EXT) {
2603                         err = -EINVAL;
2604                         goto out_put_prog;
2605                 }
2606
2607                 tgt_prog = bpf_prog_get(tgt_prog_fd);
2608                 if (IS_ERR(tgt_prog)) {
2609                         err = PTR_ERR(tgt_prog);
2610                         tgt_prog = NULL;
2611                         goto out_put_prog;
2612                 }
2613
2614                 key = bpf_trampoline_compute_key(tgt_prog, NULL, btf_id);
2615         }
2616
2617         link = kzalloc(sizeof(*link), GFP_USER);
2618         if (!link) {
2619                 err = -ENOMEM;
2620                 goto out_put_prog;
2621         }
2622         bpf_link_init(&link->link, BPF_LINK_TYPE_TRACING,
2623                       &bpf_tracing_link_lops, prog);
2624         link->attach_type = prog->expected_attach_type;
2625
2626         mutex_lock(&prog->aux->dst_mutex);
2627
2628         /* There are a few possible cases here:
2629          *
2630          * - if prog->aux->dst_trampoline is set, the program was just loaded
2631          *   and not yet attached to anything, so we can use the values stored
2632          *   in prog->aux
2633          *
2634          * - if prog->aux->dst_trampoline is NULL, the program has already been
2635          *   attached to a target and its initial target was cleared (below)
2636          *
2637          * - if tgt_prog != NULL, the caller specified tgt_prog_fd +
2638          *   target_btf_id using the link_create API.
2639          *
2640          * - if tgt_prog == NULL when this function was called using the old
2641          *   raw_tracepoint_open API, and we need a target from prog->aux
2642          *
2643          * The combination of no saved target in prog->aux, and no target
2644          * specified on load is illegal, and we reject that here.
2645          */
2646         if (!prog->aux->dst_trampoline && !tgt_prog) {
2647                 err = -ENOENT;
2648                 goto out_unlock;
2649         }
2650
2651         if (!prog->aux->dst_trampoline ||
2652             (key && key != prog->aux->dst_trampoline->key)) {
2653                 /* If there is no saved target, or the specified target is
2654                  * different from the destination specified at load time, we
2655                  * need a new trampoline and a check for compatibility
2656                  */
2657                 struct bpf_attach_target_info tgt_info = {};
2658
2659                 err = bpf_check_attach_target(NULL, prog, tgt_prog, btf_id,
2660                                               &tgt_info);
2661                 if (err)
2662                         goto out_unlock;
2663
2664                 tr = bpf_trampoline_get(key, &tgt_info);
2665                 if (!tr) {
2666                         err = -ENOMEM;
2667                         goto out_unlock;
2668                 }
2669         } else {
2670                 /* The caller didn't specify a target, or the target was the
2671                  * same as the destination supplied during program load. This
2672                  * means we can reuse the trampoline and reference from program
2673                  * load time, and there is no need to allocate a new one. This
2674                  * can only happen once for any program, as the saved values in
2675                  * prog->aux are cleared below.
2676                  */
2677                 tr = prog->aux->dst_trampoline;
2678                 tgt_prog = prog->aux->dst_prog;
2679         }
2680
2681         err = bpf_link_prime(&link->link, &link_primer);
2682         if (err)
2683                 goto out_unlock;
2684
2685         err = bpf_trampoline_link_prog(prog, tr);
2686         if (err) {
2687                 bpf_link_cleanup(&link_primer);
2688                 link = NULL;
2689                 goto out_unlock;
2690         }
2691
2692         link->tgt_prog = tgt_prog;
2693         link->trampoline = tr;
2694
2695         /* Always clear the trampoline and target prog from prog->aux to make
2696          * sure the original attach destination is not kept alive after a
2697          * program is (re-)attached to another target.
2698          */
2699         if (prog->aux->dst_prog &&
2700             (tgt_prog_fd || tr != prog->aux->dst_trampoline))
2701                 /* got extra prog ref from syscall, or attaching to different prog */
2702                 bpf_prog_put(prog->aux->dst_prog);
2703         if (prog->aux->dst_trampoline && tr != prog->aux->dst_trampoline)
2704                 /* we allocated a new trampoline, so free the old one */
2705                 bpf_trampoline_put(prog->aux->dst_trampoline);
2706
2707         prog->aux->dst_prog = NULL;
2708         prog->aux->dst_trampoline = NULL;
2709         mutex_unlock(&prog->aux->dst_mutex);
2710
2711         return bpf_link_settle(&link_primer);
2712 out_unlock:
2713         if (tr && tr != prog->aux->dst_trampoline)
2714                 bpf_trampoline_put(tr);
2715         mutex_unlock(&prog->aux->dst_mutex);
2716         kfree(link);
2717 out_put_prog:
2718         if (tgt_prog_fd && tgt_prog)
2719                 bpf_prog_put(tgt_prog);
2720         return err;
2721 }
2722
2723 struct bpf_raw_tp_link {
2724         struct bpf_link link;
2725         struct bpf_raw_event_map *btp;
2726 };
2727
2728 static void bpf_raw_tp_link_release(struct bpf_link *link)
2729 {
2730         struct bpf_raw_tp_link *raw_tp =
2731                 container_of(link, struct bpf_raw_tp_link, link);
2732
2733         bpf_probe_unregister(raw_tp->btp, raw_tp->link.prog);
2734         bpf_put_raw_tracepoint(raw_tp->btp);
2735 }
2736
2737 static void bpf_raw_tp_link_dealloc(struct bpf_link *link)
2738 {
2739         struct bpf_raw_tp_link *raw_tp =
2740                 container_of(link, struct bpf_raw_tp_link, link);
2741
2742         kfree(raw_tp);
2743 }
2744
2745 static void bpf_raw_tp_link_show_fdinfo(const struct bpf_link *link,
2746                                         struct seq_file *seq)
2747 {
2748         struct bpf_raw_tp_link *raw_tp_link =
2749                 container_of(link, struct bpf_raw_tp_link, link);
2750
2751         seq_printf(seq,
2752                    "tp_name:\t%s\n",
2753                    raw_tp_link->btp->tp->name);
2754 }
2755
2756 static int bpf_raw_tp_link_fill_link_info(const struct bpf_link *link,
2757                                           struct bpf_link_info *info)
2758 {
2759         struct bpf_raw_tp_link *raw_tp_link =
2760                 container_of(link, struct bpf_raw_tp_link, link);
2761         char __user *ubuf = u64_to_user_ptr(info->raw_tracepoint.tp_name);
2762         const char *tp_name = raw_tp_link->btp->tp->name;
2763         u32 ulen = info->raw_tracepoint.tp_name_len;
2764         size_t tp_len = strlen(tp_name);
2765
2766         if (!ulen ^ !ubuf)
2767                 return -EINVAL;
2768
2769         info->raw_tracepoint.tp_name_len = tp_len + 1;
2770
2771         if (!ubuf)
2772                 return 0;
2773
2774         if (ulen >= tp_len + 1) {
2775                 if (copy_to_user(ubuf, tp_name, tp_len + 1))
2776                         return -EFAULT;
2777         } else {
2778                 char zero = '\0';
2779
2780                 if (copy_to_user(ubuf, tp_name, ulen - 1))
2781                         return -EFAULT;
2782                 if (put_user(zero, ubuf + ulen - 1))
2783                         return -EFAULT;
2784                 return -ENOSPC;
2785         }
2786
2787         return 0;
2788 }
2789
2790 static const struct bpf_link_ops bpf_raw_tp_link_lops = {
2791         .release = bpf_raw_tp_link_release,
2792         .dealloc = bpf_raw_tp_link_dealloc,
2793         .show_fdinfo = bpf_raw_tp_link_show_fdinfo,
2794         .fill_link_info = bpf_raw_tp_link_fill_link_info,
2795 };
2796
2797 #define BPF_RAW_TRACEPOINT_OPEN_LAST_FIELD raw_tracepoint.prog_fd
2798
2799 static int bpf_raw_tracepoint_open(const union bpf_attr *attr)
2800 {
2801         struct bpf_link_primer link_primer;
2802         struct bpf_raw_tp_link *link;
2803         struct bpf_raw_event_map *btp;
2804         struct bpf_prog *prog;
2805         const char *tp_name;
2806         char buf[128];
2807         int err;
2808
2809         if (CHECK_ATTR(BPF_RAW_TRACEPOINT_OPEN))
2810                 return -EINVAL;
2811
2812         prog = bpf_prog_get(attr->raw_tracepoint.prog_fd);
2813         if (IS_ERR(prog))
2814                 return PTR_ERR(prog);
2815
2816         switch (prog->type) {
2817         case BPF_PROG_TYPE_TRACING:
2818         case BPF_PROG_TYPE_EXT:
2819         case BPF_PROG_TYPE_LSM:
2820                 if (attr->raw_tracepoint.name) {
2821                         /* The attach point for this category of programs
2822                          * should be specified via btf_id during program load.
2823                          */
2824                         err = -EINVAL;
2825                         goto out_put_prog;
2826                 }
2827                 if (prog->type == BPF_PROG_TYPE_TRACING &&
2828                     prog->expected_attach_type == BPF_TRACE_RAW_TP) {
2829                         tp_name = prog->aux->attach_func_name;
2830                         break;
2831                 }
2832                 err = bpf_tracing_prog_attach(prog, 0, 0);
2833                 if (err >= 0)
2834                         return err;
2835                 goto out_put_prog;
2836         case BPF_PROG_TYPE_RAW_TRACEPOINT:
2837         case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE:
2838                 if (strncpy_from_user(buf,
2839                                       u64_to_user_ptr(attr->raw_tracepoint.name),
2840                                       sizeof(buf) - 1) < 0) {
2841                         err = -EFAULT;
2842                         goto out_put_prog;
2843                 }
2844                 buf[sizeof(buf) - 1] = 0;
2845                 tp_name = buf;
2846                 break;
2847         default:
2848                 err = -EINVAL;
2849                 goto out_put_prog;
2850         }
2851
2852         btp = bpf_get_raw_tracepoint(tp_name);
2853         if (!btp) {
2854                 err = -ENOENT;
2855                 goto out_put_prog;
2856         }
2857
2858         link = kzalloc(sizeof(*link), GFP_USER);
2859         if (!link) {
2860                 err = -ENOMEM;
2861                 goto out_put_btp;
2862         }
2863         bpf_link_init(&link->link, BPF_LINK_TYPE_RAW_TRACEPOINT,
2864                       &bpf_raw_tp_link_lops, prog);
2865         link->btp = btp;
2866
2867         err = bpf_link_prime(&link->link, &link_primer);
2868         if (err) {
2869                 kfree(link);
2870                 goto out_put_btp;
2871         }
2872
2873         err = bpf_probe_register(link->btp, prog);
2874         if (err) {
2875                 bpf_link_cleanup(&link_primer);
2876                 goto out_put_btp;
2877         }
2878
2879         return bpf_link_settle(&link_primer);
2880
2881 out_put_btp:
2882         bpf_put_raw_tracepoint(btp);
2883 out_put_prog:
2884         bpf_prog_put(prog);
2885         return err;
2886 }
2887
2888 static int bpf_prog_attach_check_attach_type(const struct bpf_prog *prog,
2889                                              enum bpf_attach_type attach_type)
2890 {
2891         switch (prog->type) {
2892         case BPF_PROG_TYPE_CGROUP_SOCK:
2893         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
2894         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
2895         case BPF_PROG_TYPE_SK_LOOKUP:
2896                 return attach_type == prog->expected_attach_type ? 0 : -EINVAL;
2897         case BPF_PROG_TYPE_CGROUP_SKB:
2898                 if (!capable(CAP_NET_ADMIN))
2899                         /* cg-skb progs can be loaded by unpriv user.
2900                          * check permissions at attach time.
2901                          */
2902                         return -EPERM;
2903                 return prog->enforce_expected_attach_type &&
2904                         prog->expected_attach_type != attach_type ?
2905                         -EINVAL : 0;
2906         default:
2907                 return 0;
2908         }
2909 }
2910
2911 static enum bpf_prog_type
2912 attach_type_to_prog_type(enum bpf_attach_type attach_type)
2913 {
2914         switch (attach_type) {
2915         case BPF_CGROUP_INET_INGRESS:
2916         case BPF_CGROUP_INET_EGRESS:
2917                 return BPF_PROG_TYPE_CGROUP_SKB;
2918         case BPF_CGROUP_INET_SOCK_CREATE:
2919         case BPF_CGROUP_INET_SOCK_RELEASE:
2920         case BPF_CGROUP_INET4_POST_BIND:
2921         case BPF_CGROUP_INET6_POST_BIND:
2922                 return BPF_PROG_TYPE_CGROUP_SOCK;
2923         case BPF_CGROUP_INET4_BIND:
2924         case BPF_CGROUP_INET6_BIND:
2925         case BPF_CGROUP_INET4_CONNECT:
2926         case BPF_CGROUP_INET6_CONNECT:
2927         case BPF_CGROUP_INET4_GETPEERNAME:
2928         case BPF_CGROUP_INET6_GETPEERNAME:
2929         case BPF_CGROUP_INET4_GETSOCKNAME:
2930         case BPF_CGROUP_INET6_GETSOCKNAME:
2931         case BPF_CGROUP_UDP4_SENDMSG:
2932         case BPF_CGROUP_UDP6_SENDMSG:
2933         case BPF_CGROUP_UDP4_RECVMSG:
2934         case BPF_CGROUP_UDP6_RECVMSG:
2935                 return BPF_PROG_TYPE_CGROUP_SOCK_ADDR;
2936         case BPF_CGROUP_SOCK_OPS:
2937                 return BPF_PROG_TYPE_SOCK_OPS;
2938         case BPF_CGROUP_DEVICE:
2939                 return BPF_PROG_TYPE_CGROUP_DEVICE;
2940         case BPF_SK_MSG_VERDICT:
2941                 return BPF_PROG_TYPE_SK_MSG;
2942         case BPF_SK_SKB_STREAM_PARSER:
2943         case BPF_SK_SKB_STREAM_VERDICT:
2944                 return BPF_PROG_TYPE_SK_SKB;
2945         case BPF_LIRC_MODE2:
2946                 return BPF_PROG_TYPE_LIRC_MODE2;
2947         case BPF_FLOW_DISSECTOR:
2948                 return BPF_PROG_TYPE_FLOW_DISSECTOR;
2949         case BPF_CGROUP_SYSCTL:
2950                 return BPF_PROG_TYPE_CGROUP_SYSCTL;
2951         case BPF_CGROUP_GETSOCKOPT:
2952         case BPF_CGROUP_SETSOCKOPT:
2953                 return BPF_PROG_TYPE_CGROUP_SOCKOPT;
2954         case BPF_TRACE_ITER:
2955                 return BPF_PROG_TYPE_TRACING;
2956         case BPF_SK_LOOKUP:
2957                 return BPF_PROG_TYPE_SK_LOOKUP;
2958         case BPF_XDP:
2959                 return BPF_PROG_TYPE_XDP;
2960         default:
2961                 return BPF_PROG_TYPE_UNSPEC;
2962         }
2963 }
2964
2965 #define BPF_PROG_ATTACH_LAST_FIELD replace_bpf_fd
2966
2967 #define BPF_F_ATTACH_MASK \
2968         (BPF_F_ALLOW_OVERRIDE | BPF_F_ALLOW_MULTI | BPF_F_REPLACE)
2969
2970 static int bpf_prog_attach(const union bpf_attr *attr)
2971 {
2972         enum bpf_prog_type ptype;
2973         struct bpf_prog *prog;
2974         int ret;
2975
2976         if (CHECK_ATTR(BPF_PROG_ATTACH))
2977                 return -EINVAL;
2978
2979         if (attr->attach_flags & ~BPF_F_ATTACH_MASK)
2980                 return -EINVAL;
2981
2982         ptype = attach_type_to_prog_type(attr->attach_type);
2983         if (ptype == BPF_PROG_TYPE_UNSPEC)
2984                 return -EINVAL;
2985
2986         prog = bpf_prog_get_type(attr->attach_bpf_fd, ptype);
2987         if (IS_ERR(prog))
2988                 return PTR_ERR(prog);
2989
2990         if (bpf_prog_attach_check_attach_type(prog, attr->attach_type)) {
2991                 bpf_prog_put(prog);
2992                 return -EINVAL;
2993         }
2994
2995         switch (ptype) {
2996         case BPF_PROG_TYPE_SK_SKB:
2997         case BPF_PROG_TYPE_SK_MSG:
2998                 ret = sock_map_get_from_fd(attr, prog);
2999                 break;
3000         case BPF_PROG_TYPE_LIRC_MODE2:
3001                 ret = lirc_prog_attach(attr, prog);
3002                 break;
3003         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3004                 ret = netns_bpf_prog_attach(attr, prog);
3005                 break;
3006         case BPF_PROG_TYPE_CGROUP_DEVICE:
3007         case BPF_PROG_TYPE_CGROUP_SKB:
3008         case BPF_PROG_TYPE_CGROUP_SOCK:
3009         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3010         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3011         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3012         case BPF_PROG_TYPE_SOCK_OPS:
3013                 ret = cgroup_bpf_prog_attach(attr, ptype, prog);
3014                 break;
3015         default:
3016                 ret = -EINVAL;
3017         }
3018
3019         if (ret)
3020                 bpf_prog_put(prog);
3021         return ret;
3022 }
3023
3024 #define BPF_PROG_DETACH_LAST_FIELD attach_type
3025
3026 static int bpf_prog_detach(const union bpf_attr *attr)
3027 {
3028         enum bpf_prog_type ptype;
3029
3030         if (CHECK_ATTR(BPF_PROG_DETACH))
3031                 return -EINVAL;
3032
3033         ptype = attach_type_to_prog_type(attr->attach_type);
3034
3035         switch (ptype) {
3036         case BPF_PROG_TYPE_SK_MSG:
3037         case BPF_PROG_TYPE_SK_SKB:
3038                 return sock_map_prog_detach(attr, ptype);
3039         case BPF_PROG_TYPE_LIRC_MODE2:
3040                 return lirc_prog_detach(attr);
3041         case BPF_PROG_TYPE_FLOW_DISSECTOR:
3042                 return netns_bpf_prog_detach(attr, ptype);
3043         case BPF_PROG_TYPE_CGROUP_DEVICE:
3044         case BPF_PROG_TYPE_CGROUP_SKB:
3045         case BPF_PROG_TYPE_CGROUP_SOCK:
3046         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
3047         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
3048         case BPF_PROG_TYPE_CGROUP_SYSCTL:
3049         case BPF_PROG_TYPE_SOCK_OPS:
3050                 return cgroup_bpf_prog_detach(attr, ptype);
3051         default:
3052                 return -EINVAL;
3053         }
3054 }
3055
3056 #define BPF_PROG_QUERY_LAST_FIELD query.prog_cnt
3057
3058 static int bpf_prog_query(const union bpf_attr *attr,
3059                           union bpf_attr __user *uattr)
3060 {
3061         if (!capable(CAP_NET_ADMIN))
3062                 return -EPERM;
3063         if (CHECK_ATTR(BPF_PROG_QUERY))
3064                 return -EINVAL;
3065         if (attr->query.query_flags & ~BPF_F_QUERY_EFFECTIVE)
3066                 return -EINVAL;
3067
3068         switch (attr->query.attach_type) {
3069         case BPF_CGROUP_INET_INGRESS:
3070         case BPF_CGROUP_INET_EGRESS:
3071         case BPF_CGROUP_INET_SOCK_CREATE:
3072         case BPF_CGROUP_INET_SOCK_RELEASE:
3073         case BPF_CGROUP_INET4_BIND:
3074         case BPF_CGROUP_INET6_BIND:
3075         case BPF_CGROUP_INET4_POST_BIND:
3076         case BPF_CGROUP_INET6_POST_BIND:
3077         case BPF_CGROUP_INET4_CONNECT:
3078         case BPF_CGROUP_INET6_CONNECT:
3079         case BPF_CGROUP_INET4_GETPEERNAME:
3080         case BPF_CGROUP_INET6_GETPEERNAME:
3081         case BPF_CGROUP_INET4_GETSOCKNAME:
3082         case BPF_CGROUP_INET6_GETSOCKNAME:
3083         case BPF_CGROUP_UDP4_SENDMSG:
3084         case BPF_CGROUP_UDP6_SENDMSG:
3085         case BPF_CGROUP_UDP4_RECVMSG:
3086         case BPF_CGROUP_UDP6_RECVMSG:
3087         case BPF_CGROUP_SOCK_OPS:
3088         case BPF_CGROUP_DEVICE:
3089         case BPF_CGROUP_SYSCTL:
3090         case BPF_CGROUP_GETSOCKOPT:
3091         case BPF_CGROUP_SETSOCKOPT:
3092                 return cgroup_bpf_prog_query(attr, uattr);
3093         case BPF_LIRC_MODE2:
3094                 return lirc_prog_query(attr, uattr);
3095         case BPF_FLOW_DISSECTOR:
3096         case BPF_SK_LOOKUP:
3097                 return netns_bpf_prog_query(attr, uattr);
3098         default:
3099                 return -EINVAL;
3100         }
3101 }
3102
3103 #define BPF_PROG_TEST_RUN_LAST_FIELD test.cpu
3104
3105 static int bpf_prog_test_run(const union bpf_attr *attr,
3106                              union bpf_attr __user *uattr)
3107 {
3108         struct bpf_prog *prog;
3109         int ret = -ENOTSUPP;
3110
3111         if (CHECK_ATTR(BPF_PROG_TEST_RUN))
3112                 return -EINVAL;
3113
3114         if ((attr->test.ctx_size_in && !attr->test.ctx_in) ||
3115             (!attr->test.ctx_size_in && attr->test.ctx_in))
3116                 return -EINVAL;
3117
3118         if ((attr->test.ctx_size_out && !attr->test.ctx_out) ||
3119             (!attr->test.ctx_size_out && attr->test.ctx_out))
3120                 return -EINVAL;
3121
3122         prog = bpf_prog_get(attr->test.prog_fd);
3123         if (IS_ERR(prog))
3124                 return PTR_ERR(prog);
3125
3126         if (prog->aux->ops->test_run)
3127                 ret = prog->aux->ops->test_run(prog, attr, uattr);
3128
3129         bpf_prog_put(prog);
3130         return ret;
3131 }
3132
3133 #define BPF_OBJ_GET_NEXT_ID_LAST_FIELD next_id
3134
3135 static int bpf_obj_get_next_id(const union bpf_attr *attr,
3136                                union bpf_attr __user *uattr,
3137                                struct idr *idr,
3138                                spinlock_t *lock)
3139 {
3140         u32 next_id = attr->start_id;
3141         int err = 0;
3142
3143         if (CHECK_ATTR(BPF_OBJ_GET_NEXT_ID) || next_id >= INT_MAX)
3144                 return -EINVAL;
3145
3146         if (!capable(CAP_SYS_ADMIN))
3147                 return -EPERM;
3148
3149         next_id++;
3150         spin_lock_bh(lock);
3151         if (!idr_get_next(idr, &next_id))
3152                 err = -ENOENT;
3153         spin_unlock_bh(lock);
3154
3155         if (!err)
3156                 err = put_user(next_id, &uattr->next_id);
3157
3158         return err;
3159 }
3160
3161 struct bpf_map *bpf_map_get_curr_or_next(u32 *id)
3162 {
3163         struct bpf_map *map;
3164
3165         spin_lock_bh(&map_idr_lock);
3166 again:
3167         map = idr_get_next(&map_idr, id);
3168         if (map) {
3169                 map = __bpf_map_inc_not_zero(map, false);
3170                 if (IS_ERR(map)) {
3171                         (*id)++;
3172                         goto again;
3173                 }
3174         }
3175         spin_unlock_bh(&map_idr_lock);
3176
3177         return map;
3178 }
3179
3180 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id)
3181 {
3182         struct bpf_prog *prog;
3183
3184         spin_lock_bh(&prog_idr_lock);
3185 again:
3186         prog = idr_get_next(&prog_idr, id);
3187         if (prog) {
3188                 prog = bpf_prog_inc_not_zero(prog);
3189                 if (IS_ERR(prog)) {
3190                         (*id)++;
3191                         goto again;
3192                 }
3193         }
3194         spin_unlock_bh(&prog_idr_lock);
3195
3196         return prog;
3197 }
3198
3199 #define BPF_PROG_GET_FD_BY_ID_LAST_FIELD prog_id
3200
3201 struct bpf_prog *bpf_prog_by_id(u32 id)
3202 {
3203         struct bpf_prog *prog;
3204
3205         if (!id)
3206                 return ERR_PTR(-ENOENT);
3207
3208         spin_lock_bh(&prog_idr_lock);
3209         prog = idr_find(&prog_idr, id);
3210         if (prog)
3211                 prog = bpf_prog_inc_not_zero(prog);
3212         else
3213                 prog = ERR_PTR(-ENOENT);
3214         spin_unlock_bh(&prog_idr_lock);
3215         return prog;
3216 }
3217
3218 static int bpf_prog_get_fd_by_id(const union bpf_attr *attr)
3219 {
3220         struct bpf_prog *prog;
3221         u32 id = attr->prog_id;
3222         int fd;
3223
3224         if (CHECK_ATTR(BPF_PROG_GET_FD_BY_ID))
3225                 return -EINVAL;
3226
3227         if (!capable(CAP_SYS_ADMIN))
3228                 return -EPERM;
3229
3230         prog = bpf_prog_by_id(id);
3231         if (IS_ERR(prog))
3232                 return PTR_ERR(prog);
3233
3234         fd = bpf_prog_new_fd(prog);
3235         if (fd < 0)
3236                 bpf_prog_put(prog);
3237
3238         return fd;
3239 }
3240
3241 #define BPF_MAP_GET_FD_BY_ID_LAST_FIELD open_flags
3242
3243 static int bpf_map_get_fd_by_id(const union bpf_attr *attr)
3244 {
3245         struct bpf_map *map;
3246         u32 id = attr->map_id;
3247         int f_flags;
3248         int fd;
3249
3250         if (CHECK_ATTR(BPF_MAP_GET_FD_BY_ID) ||
3251             attr->open_flags & ~BPF_OBJ_FLAG_MASK)
3252                 return -EINVAL;
3253
3254         if (!capable(CAP_SYS_ADMIN))
3255                 return -EPERM;
3256
3257         f_flags = bpf_get_file_flag(attr->open_flags);
3258         if (f_flags < 0)
3259                 return f_flags;
3260
3261         spin_lock_bh(&map_idr_lock);
3262         map = idr_find(&map_idr, id);
3263         if (map)
3264                 map = __bpf_map_inc_not_zero(map, true);
3265         else
3266                 map = ERR_PTR(-ENOENT);
3267         spin_unlock_bh(&map_idr_lock);
3268
3269         if (IS_ERR(map))
3270                 return PTR_ERR(map);
3271
3272         fd = bpf_map_new_fd(map, f_flags);
3273         if (fd < 0)
3274                 bpf_map_put_with_uref(map);
3275
3276         return fd;
3277 }
3278
3279 static const struct bpf_map *bpf_map_from_imm(const struct bpf_prog *prog,
3280                                               unsigned long addr, u32 *off,
3281                                               u32 *type)
3282 {
3283         const struct bpf_map *map;
3284         int i;
3285
3286         mutex_lock(&prog->aux->used_maps_mutex);
3287         for (i = 0, *off = 0; i < prog->aux->used_map_cnt; i++) {
3288                 map = prog->aux->used_maps[i];
3289                 if (map == (void *)addr) {
3290                         *type = BPF_PSEUDO_MAP_FD;
3291                         goto out;
3292                 }
3293                 if (!map->ops->map_direct_value_meta)
3294                         continue;
3295                 if (!map->ops->map_direct_value_meta(map, addr, off)) {
3296                         *type = BPF_PSEUDO_MAP_VALUE;
3297                         goto out;
3298                 }
3299         }
3300         map = NULL;
3301
3302 out:
3303         mutex_unlock(&prog->aux->used_maps_mutex);
3304         return map;
3305 }
3306
3307 static struct bpf_insn *bpf_insn_prepare_dump(const struct bpf_prog *prog,
3308                                               const struct cred *f_cred)
3309 {
3310         const struct bpf_map *map;
3311         struct bpf_insn *insns;
3312         u32 off, type;
3313         u64 imm;
3314         u8 code;
3315         int i;
3316
3317         insns = kmemdup(prog->insnsi, bpf_prog_insn_size(prog),
3318                         GFP_USER);
3319         if (!insns)
3320                 return insns;
3321
3322         for (i = 0; i < prog->len; i++) {
3323                 code = insns[i].code;
3324
3325                 if (code == (BPF_JMP | BPF_TAIL_CALL)) {
3326                         insns[i].code = BPF_JMP | BPF_CALL;
3327                         insns[i].imm = BPF_FUNC_tail_call;
3328                         /* fall-through */
3329                 }
3330                 if (code == (BPF_JMP | BPF_CALL) ||
3331                     code == (BPF_JMP | BPF_CALL_ARGS)) {
3332                         if (code == (BPF_JMP | BPF_CALL_ARGS))
3333                                 insns[i].code = BPF_JMP | BPF_CALL;
3334                         if (!bpf_dump_raw_ok(f_cred))
3335                                 insns[i].imm = 0;
3336                         continue;
3337                 }
3338                 if (BPF_CLASS(code) == BPF_LDX && BPF_MODE(code) == BPF_PROBE_MEM) {
3339                         insns[i].code = BPF_LDX | BPF_SIZE(code) | BPF_MEM;
3340                         continue;
3341                 }
3342
3343                 if (code != (BPF_LD | BPF_IMM | BPF_DW))
3344                         continue;
3345
3346                 imm = ((u64)insns[i + 1].imm << 32) | (u32)insns[i].imm;
3347                 map = bpf_map_from_imm(prog, imm, &off, &type);
3348                 if (map) {
3349                         insns[i].src_reg = type;
3350                         insns[i].imm = map->id;
3351                         insns[i + 1].imm = off;
3352                         continue;
3353                 }
3354         }
3355
3356         return insns;
3357 }
3358
3359 static int set_info_rec_size(struct bpf_prog_info *info)
3360 {
3361         /*
3362          * Ensure info.*_rec_size is the same as kernel expected size
3363          *
3364          * or
3365          *
3366          * Only allow zero *_rec_size if both _rec_size and _cnt are
3367          * zero.  In this case, the kernel will set the expected
3368          * _rec_size back to the info.
3369          */
3370
3371         if ((info->nr_func_info || info->func_info_rec_size) &&
3372             info->func_info_rec_size != sizeof(struct bpf_func_info))
3373                 return -EINVAL;
3374
3375         if ((info->nr_line_info || info->line_info_rec_size) &&
3376             info->line_info_rec_size != sizeof(struct bpf_line_info))
3377                 return -EINVAL;
3378
3379         if ((info->nr_jited_line_info || info->jited_line_info_rec_size) &&
3380             info->jited_line_info_rec_size != sizeof(__u64))
3381                 return -EINVAL;
3382
3383         info->func_info_rec_size = sizeof(struct bpf_func_info);
3384         info->line_info_rec_size = sizeof(struct bpf_line_info);
3385         info->jited_line_info_rec_size = sizeof(__u64);
3386
3387         return 0;
3388 }
3389
3390 static int bpf_prog_get_info_by_fd(struct file *file,
3391                                    struct bpf_prog *prog,
3392                                    const union bpf_attr *attr,
3393                                    union bpf_attr __user *uattr)
3394 {
3395         struct bpf_prog_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3396         struct bpf_prog_info info;
3397         u32 info_len = attr->info.info_len;
3398         struct bpf_prog_stats stats;
3399         char __user *uinsns;
3400         u32 ulen;
3401         int err;
3402
3403         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3404         if (err)
3405                 return err;
3406         info_len = min_t(u32, sizeof(info), info_len);
3407
3408         memset(&info, 0, sizeof(info));
3409         if (copy_from_user(&info, uinfo, info_len))
3410                 return -EFAULT;
3411
3412         info.type = prog->type;
3413         info.id = prog->aux->id;
3414         info.load_time = prog->aux->load_time;
3415         info.created_by_uid = from_kuid_munged(current_user_ns(),
3416                                                prog->aux->user->uid);
3417         info.gpl_compatible = prog->gpl_compatible;
3418
3419         memcpy(info.tag, prog->tag, sizeof(prog->tag));
3420         memcpy(info.name, prog->aux->name, sizeof(prog->aux->name));
3421
3422         mutex_lock(&prog->aux->used_maps_mutex);
3423         ulen = info.nr_map_ids;
3424         info.nr_map_ids = prog->aux->used_map_cnt;
3425         ulen = min_t(u32, info.nr_map_ids, ulen);
3426         if (ulen) {
3427                 u32 __user *user_map_ids = u64_to_user_ptr(info.map_ids);
3428                 u32 i;
3429
3430                 for (i = 0; i < ulen; i++)
3431                         if (put_user(prog->aux->used_maps[i]->id,
3432                                      &user_map_ids[i])) {
3433                                 mutex_unlock(&prog->aux->used_maps_mutex);
3434                                 return -EFAULT;
3435                         }
3436         }
3437         mutex_unlock(&prog->aux->used_maps_mutex);
3438
3439         err = set_info_rec_size(&info);
3440         if (err)
3441                 return err;
3442
3443         bpf_prog_get_stats(prog, &stats);
3444         info.run_time_ns = stats.nsecs;
3445         info.run_cnt = stats.cnt;
3446         info.recursion_misses = stats.misses;
3447
3448         if (!bpf_capable()) {
3449                 info.jited_prog_len = 0;
3450                 info.xlated_prog_len = 0;
3451                 info.nr_jited_ksyms = 0;
3452                 info.nr_jited_func_lens = 0;
3453                 info.nr_func_info = 0;
3454                 info.nr_line_info = 0;
3455                 info.nr_jited_line_info = 0;
3456                 goto done;
3457         }
3458
3459         ulen = info.xlated_prog_len;
3460         info.xlated_prog_len = bpf_prog_insn_size(prog);
3461         if (info.xlated_prog_len && ulen) {
3462                 struct bpf_insn *insns_sanitized;
3463                 bool fault;
3464
3465                 if (prog->blinded && !bpf_dump_raw_ok(file->f_cred)) {
3466                         info.xlated_prog_insns = 0;
3467                         goto done;
3468                 }
3469                 insns_sanitized = bpf_insn_prepare_dump(prog, file->f_cred);
3470                 if (!insns_sanitized)
3471                         return -ENOMEM;
3472                 uinsns = u64_to_user_ptr(info.xlated_prog_insns);
3473                 ulen = min_t(u32, info.xlated_prog_len, ulen);
3474                 fault = copy_to_user(uinsns, insns_sanitized, ulen);
3475                 kfree(insns_sanitized);
3476                 if (fault)
3477                         return -EFAULT;
3478         }
3479
3480         if (bpf_prog_is_dev_bound(prog->aux)) {
3481                 err = bpf_prog_offload_info_fill(&info, prog);
3482                 if (err)
3483                         return err;
3484                 goto done;
3485         }
3486
3487         /* NOTE: the following code is supposed to be skipped for offload.
3488          * bpf_prog_offload_info_fill() is the place to fill similar fields
3489          * for offload.
3490          */
3491         ulen = info.jited_prog_len;
3492         if (prog->aux->func_cnt) {
3493                 u32 i;
3494
3495                 info.jited_prog_len = 0;
3496                 for (i = 0; i < prog->aux->func_cnt; i++)
3497                         info.jited_prog_len += prog->aux->func[i]->jited_len;
3498         } else {
3499                 info.jited_prog_len = prog->jited_len;
3500         }
3501
3502         if (info.jited_prog_len && ulen) {
3503                 if (bpf_dump_raw_ok(file->f_cred)) {
3504                         uinsns = u64_to_user_ptr(info.jited_prog_insns);
3505                         ulen = min_t(u32, info.jited_prog_len, ulen);
3506
3507                         /* for multi-function programs, copy the JITed
3508                          * instructions for all the functions
3509                          */
3510                         if (prog->aux->func_cnt) {
3511                                 u32 len, free, i;
3512                                 u8 *img;
3513
3514                                 free = ulen;
3515                                 for (i = 0; i < prog->aux->func_cnt; i++) {
3516                                         len = prog->aux->func[i]->jited_len;
3517                                         len = min_t(u32, len, free);
3518                                         img = (u8 *) prog->aux->func[i]->bpf_func;
3519                                         if (copy_to_user(uinsns, img, len))
3520                                                 return -EFAULT;
3521                                         uinsns += len;
3522                                         free -= len;
3523                                         if (!free)
3524                                                 break;
3525                                 }
3526                         } else {
3527                                 if (copy_to_user(uinsns, prog->bpf_func, ulen))
3528                                         return -EFAULT;
3529                         }
3530                 } else {
3531                         info.jited_prog_insns = 0;
3532                 }
3533         }
3534
3535         ulen = info.nr_jited_ksyms;
3536         info.nr_jited_ksyms = prog->aux->func_cnt ? : 1;
3537         if (ulen) {
3538                 if (bpf_dump_raw_ok(file->f_cred)) {
3539                         unsigned long ksym_addr;
3540                         u64 __user *user_ksyms;
3541                         u32 i;
3542
3543                         /* copy the address of the kernel symbol
3544                          * corresponding to each function
3545                          */
3546                         ulen = min_t(u32, info.nr_jited_ksyms, ulen);
3547                         user_ksyms = u64_to_user_ptr(info.jited_ksyms);
3548                         if (prog->aux->func_cnt) {
3549                                 for (i = 0; i < ulen; i++) {
3550                                         ksym_addr = (unsigned long)
3551                                                 prog->aux->func[i]->bpf_func;
3552                                         if (put_user((u64) ksym_addr,
3553                                                      &user_ksyms[i]))
3554                                                 return -EFAULT;
3555                                 }
3556                         } else {
3557                                 ksym_addr = (unsigned long) prog->bpf_func;
3558                                 if (put_user((u64) ksym_addr, &user_ksyms[0]))
3559                                         return -EFAULT;
3560                         }
3561                 } else {
3562                         info.jited_ksyms = 0;
3563                 }
3564         }
3565
3566         ulen = info.nr_jited_func_lens;
3567         info.nr_jited_func_lens = prog->aux->func_cnt ? : 1;
3568         if (ulen) {
3569                 if (bpf_dump_raw_ok(file->f_cred)) {
3570                         u32 __user *user_lens;
3571                         u32 func_len, i;
3572
3573                         /* copy the JITed image lengths for each function */
3574                         ulen = min_t(u32, info.nr_jited_func_lens, ulen);
3575                         user_lens = u64_to_user_ptr(info.jited_func_lens);
3576                         if (prog->aux->func_cnt) {
3577                                 for (i = 0; i < ulen; i++) {
3578                                         func_len =
3579                                                 prog->aux->func[i]->jited_len;
3580                                         if (put_user(func_len, &user_lens[i]))
3581                                                 return -EFAULT;
3582                                 }
3583                         } else {
3584                                 func_len = prog->jited_len;
3585                                 if (put_user(func_len, &user_lens[0]))
3586                                         return -EFAULT;
3587                         }
3588                 } else {
3589                         info.jited_func_lens = 0;
3590                 }
3591         }
3592
3593         if (prog->aux->btf)
3594                 info.btf_id = btf_obj_id(prog->aux->btf);
3595
3596         ulen = info.nr_func_info;
3597         info.nr_func_info = prog->aux->func_info_cnt;
3598         if (info.nr_func_info && ulen) {
3599                 char __user *user_finfo;
3600
3601                 user_finfo = u64_to_user_ptr(info.func_info);
3602                 ulen = min_t(u32, info.nr_func_info, ulen);
3603                 if (copy_to_user(user_finfo, prog->aux->func_info,
3604                                  info.func_info_rec_size * ulen))
3605                         return -EFAULT;
3606         }
3607
3608         ulen = info.nr_line_info;
3609         info.nr_line_info = prog->aux->nr_linfo;
3610         if (info.nr_line_info && ulen) {
3611                 __u8 __user *user_linfo;
3612
3613                 user_linfo = u64_to_user_ptr(info.line_info);
3614                 ulen = min_t(u32, info.nr_line_info, ulen);
3615                 if (copy_to_user(user_linfo, prog->aux->linfo,
3616                                  info.line_info_rec_size * ulen))
3617                         return -EFAULT;
3618         }
3619
3620         ulen = info.nr_jited_line_info;
3621         if (prog->aux->jited_linfo)
3622                 info.nr_jited_line_info = prog->aux->nr_linfo;
3623         else
3624                 info.nr_jited_line_info = 0;
3625         if (info.nr_jited_line_info && ulen) {
3626                 if (bpf_dump_raw_ok(file->f_cred)) {
3627                         __u64 __user *user_linfo;
3628                         u32 i;
3629
3630                         user_linfo = u64_to_user_ptr(info.jited_line_info);
3631                         ulen = min_t(u32, info.nr_jited_line_info, ulen);
3632                         for (i = 0; i < ulen; i++) {
3633                                 if (put_user((__u64)(long)prog->aux->jited_linfo[i],
3634                                              &user_linfo[i]))
3635                                         return -EFAULT;
3636                         }
3637                 } else {
3638                         info.jited_line_info = 0;
3639                 }
3640         }
3641
3642         ulen = info.nr_prog_tags;
3643         info.nr_prog_tags = prog->aux->func_cnt ? : 1;
3644         if (ulen) {
3645                 __u8 __user (*user_prog_tags)[BPF_TAG_SIZE];
3646                 u32 i;
3647
3648                 user_prog_tags = u64_to_user_ptr(info.prog_tags);
3649                 ulen = min_t(u32, info.nr_prog_tags, ulen);
3650                 if (prog->aux->func_cnt) {
3651                         for (i = 0; i < ulen; i++) {
3652                                 if (copy_to_user(user_prog_tags[i],
3653                                                  prog->aux->func[i]->tag,
3654                                                  BPF_TAG_SIZE))
3655                                         return -EFAULT;
3656                         }
3657                 } else {
3658                         if (copy_to_user(user_prog_tags[0],
3659                                          prog->tag, BPF_TAG_SIZE))
3660                                 return -EFAULT;
3661                 }
3662         }
3663
3664 done:
3665         if (copy_to_user(uinfo, &info, info_len) ||
3666             put_user(info_len, &uattr->info.info_len))
3667                 return -EFAULT;
3668
3669         return 0;
3670 }
3671
3672 static int bpf_map_get_info_by_fd(struct file *file,
3673                                   struct bpf_map *map,
3674                                   const union bpf_attr *attr,
3675                                   union bpf_attr __user *uattr)
3676 {
3677         struct bpf_map_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3678         struct bpf_map_info info;
3679         u32 info_len = attr->info.info_len;
3680         int err;
3681
3682         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3683         if (err)
3684                 return err;
3685         info_len = min_t(u32, sizeof(info), info_len);
3686
3687         memset(&info, 0, sizeof(info));
3688         info.type = map->map_type;
3689         info.id = map->id;
3690         info.key_size = map->key_size;
3691         info.value_size = map->value_size;
3692         info.max_entries = map->max_entries;
3693         info.map_flags = map->map_flags;
3694         memcpy(info.name, map->name, sizeof(map->name));
3695
3696         if (map->btf) {
3697                 info.btf_id = btf_obj_id(map->btf);
3698                 info.btf_key_type_id = map->btf_key_type_id;
3699                 info.btf_value_type_id = map->btf_value_type_id;
3700         }
3701         info.btf_vmlinux_value_type_id = map->btf_vmlinux_value_type_id;
3702
3703         if (bpf_map_is_dev_bound(map)) {
3704                 err = bpf_map_offload_info_fill(&info, map);
3705                 if (err)
3706                         return err;
3707         }
3708
3709         if (copy_to_user(uinfo, &info, info_len) ||
3710             put_user(info_len, &uattr->info.info_len))
3711                 return -EFAULT;
3712
3713         return 0;
3714 }
3715
3716 static int bpf_btf_get_info_by_fd(struct file *file,
3717                                   struct btf *btf,
3718                                   const union bpf_attr *attr,
3719                                   union bpf_attr __user *uattr)
3720 {
3721         struct bpf_btf_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3722         u32 info_len = attr->info.info_len;
3723         int err;
3724
3725         err = bpf_check_uarg_tail_zero(uinfo, sizeof(*uinfo), info_len);
3726         if (err)
3727                 return err;
3728
3729         return btf_get_info_by_fd(btf, attr, uattr);
3730 }
3731
3732 static int bpf_link_get_info_by_fd(struct file *file,
3733                                   struct bpf_link *link,
3734                                   const union bpf_attr *attr,
3735                                   union bpf_attr __user *uattr)
3736 {
3737         struct bpf_link_info __user *uinfo = u64_to_user_ptr(attr->info.info);
3738         struct bpf_link_info info;
3739         u32 info_len = attr->info.info_len;
3740         int err;
3741
3742         err = bpf_check_uarg_tail_zero(uinfo, sizeof(info), info_len);
3743         if (err)
3744                 return err;
3745         info_len = min_t(u32, sizeof(info), info_len);
3746
3747         memset(&info, 0, sizeof(info));
3748         if (copy_from_user(&info, uinfo, info_len))
3749                 return -EFAULT;
3750
3751         info.type = link->type;
3752         info.id = link->id;
3753         info.prog_id = link->prog->aux->id;
3754
3755         if (link->ops->fill_link_info) {
3756                 err = link->ops->fill_link_info(link, &info);
3757                 if (err)
3758                         return err;
3759         }
3760
3761         if (copy_to_user(uinfo, &info, info_len) ||
3762             put_user(info_len, &uattr->info.info_len))
3763                 return -EFAULT;
3764
3765         return 0;
3766 }
3767
3768
3769 #define BPF_OBJ_GET_INFO_BY_FD_LAST_FIELD info.info
3770
3771 static int bpf_obj_get_info_by_fd(const union bpf_attr *attr,
3772                                   union bpf_attr __user *uattr)
3773 {
3774         int ufd = attr->info.bpf_fd;
3775         struct fd f;
3776         int err;
3777
3778         if (CHECK_ATTR(BPF_OBJ_GET_INFO_BY_FD))
3779                 return -EINVAL;
3780
3781         f = fdget(ufd);
3782         if (!f.file)
3783                 return -EBADFD;
3784
3785         if (f.file->f_op == &bpf_prog_fops)
3786                 err = bpf_prog_get_info_by_fd(f.file, f.file->private_data, attr,
3787                                               uattr);
3788         else if (f.file->f_op == &bpf_map_fops)
3789                 err = bpf_map_get_info_by_fd(f.file, f.file->private_data, attr,
3790                                              uattr);
3791         else if (f.file->f_op == &btf_fops)
3792                 err = bpf_btf_get_info_by_fd(f.file, f.file->private_data, attr, uattr);
3793         else if (f.file->f_op == &bpf_link_fops)
3794                 err = bpf_link_get_info_by_fd(f.file, f.file->private_data,
3795                                               attr, uattr);
3796         else
3797                 err = -EINVAL;
3798
3799         fdput(f);
3800         return err;
3801 }
3802
3803 #define BPF_BTF_LOAD_LAST_FIELD btf_log_level
3804
3805 static int bpf_btf_load(const union bpf_attr *attr)
3806 {
3807         if (CHECK_ATTR(BPF_BTF_LOAD))
3808                 return -EINVAL;
3809
3810         if (!bpf_capable())
3811                 return -EPERM;
3812
3813         return btf_new_fd(attr);
3814 }
3815
3816 #define BPF_BTF_GET_FD_BY_ID_LAST_FIELD btf_id
3817
3818 static int bpf_btf_get_fd_by_id(const union bpf_attr *attr)
3819 {
3820         if (CHECK_ATTR(BPF_BTF_GET_FD_BY_ID))
3821                 return -EINVAL;
3822
3823         if (!capable(CAP_SYS_ADMIN))
3824                 return -EPERM;
3825
3826         return btf_get_fd_by_id(attr->btf_id);
3827 }
3828
3829 static int bpf_task_fd_query_copy(const union bpf_attr *attr,
3830                                     union bpf_attr __user *uattr,
3831                                     u32 prog_id, u32 fd_type,
3832                                     const char *buf, u64 probe_offset,
3833                                     u64 probe_addr)
3834 {
3835         char __user *ubuf = u64_to_user_ptr(attr->task_fd_query.buf);
3836         u32 len = buf ? strlen(buf) : 0, input_len;
3837         int err = 0;
3838
3839         if (put_user(len, &uattr->task_fd_query.buf_len))
3840                 return -EFAULT;
3841         input_len = attr->task_fd_query.buf_len;
3842         if (input_len && ubuf) {
3843                 if (!len) {
3844                         /* nothing to copy, just make ubuf NULL terminated */
3845                         char zero = '\0';
3846
3847                         if (put_user(zero, ubuf))
3848                                 return -EFAULT;
3849                 } else if (input_len >= len + 1) {
3850                         /* ubuf can hold the string with NULL terminator */
3851                         if (copy_to_user(ubuf, buf, len + 1))
3852                                 return -EFAULT;
3853                 } else {
3854                         /* ubuf cannot hold the string with NULL terminator,
3855                          * do a partial copy with NULL terminator.
3856                          */
3857                         char zero = '\0';
3858
3859                         err = -ENOSPC;
3860                         if (copy_to_user(ubuf, buf, input_len - 1))
3861                                 return -EFAULT;
3862                         if (put_user(zero, ubuf + input_len - 1))
3863                                 return -EFAULT;
3864                 }
3865         }
3866
3867         if (put_user(prog_id, &uattr->task_fd_query.prog_id) ||
3868             put_user(fd_type, &uattr->task_fd_query.fd_type) ||
3869             put_user(probe_offset, &uattr->task_fd_query.probe_offset) ||
3870             put_user(probe_addr, &uattr->task_fd_query.probe_addr))
3871                 return -EFAULT;
3872
3873         return err;
3874 }
3875
3876 #define BPF_TASK_FD_QUERY_LAST_FIELD task_fd_query.probe_addr
3877
3878 static int bpf_task_fd_query(const union bpf_attr *attr,
3879                              union bpf_attr __user *uattr)
3880 {
3881         pid_t pid = attr->task_fd_query.pid;
3882         u32 fd = attr->task_fd_query.fd;
3883         const struct perf_event *event;
3884         struct task_struct *task;
3885         struct file *file;
3886         int err;
3887
3888         if (CHECK_ATTR(BPF_TASK_FD_QUERY))
3889                 return -EINVAL;
3890
3891         if (!capable(CAP_SYS_ADMIN))
3892                 return -EPERM;
3893
3894         if (attr->task_fd_query.flags != 0)
3895                 return -EINVAL;
3896
3897         task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3898         if (!task)
3899                 return -ENOENT;
3900
3901         err = 0;
3902         file = fget_task(task, fd);
3903         put_task_struct(task);
3904         if (!file)
3905                 return -EBADF;
3906
3907         if (file->f_op == &bpf_link_fops) {
3908                 struct bpf_link *link = file->private_data;
3909
3910                 if (link->ops == &bpf_raw_tp_link_lops) {
3911                         struct bpf_raw_tp_link *raw_tp =
3912                                 container_of(link, struct bpf_raw_tp_link, link);
3913                         struct bpf_raw_event_map *btp = raw_tp->btp;
3914
3915                         err = bpf_task_fd_query_copy(attr, uattr,
3916                                                      raw_tp->link.prog->aux->id,
3917                                                      BPF_FD_TYPE_RAW_TRACEPOINT,
3918                                                      btp->tp->name, 0, 0);
3919                         goto put_file;
3920                 }
3921                 goto out_not_supp;
3922         }
3923
3924         event = perf_get_event(file);
3925         if (!IS_ERR(event)) {
3926                 u64 probe_offset, probe_addr;
3927                 u32 prog_id, fd_type;
3928                 const char *buf;
3929
3930                 err = bpf_get_perf_event_info(event, &prog_id, &fd_type,
3931                                               &buf, &probe_offset,
3932                                               &probe_addr);
3933                 if (!err)
3934                         err = bpf_task_fd_query_copy(attr, uattr, prog_id,
3935                                                      fd_type, buf,
3936                                                      probe_offset,
3937                                                      probe_addr);
3938                 goto put_file;
3939         }
3940
3941 out_not_supp:
3942         err = -ENOTSUPP;
3943 put_file:
3944         fput(file);
3945         return err;
3946 }
3947
3948 #define BPF_MAP_BATCH_LAST_FIELD batch.flags
3949
3950 #define BPF_DO_BATCH(fn)                        \
3951         do {                                    \
3952                 if (!fn) {                      \
3953                         err = -ENOTSUPP;        \
3954                         goto err_put;           \
3955                 }                               \
3956                 err = fn(map, attr, uattr);     \
3957         } while (0)
3958
3959 static int bpf_map_do_batch(const union bpf_attr *attr,
3960                             union bpf_attr __user *uattr,
3961                             int cmd)
3962 {
3963         struct bpf_map *map;
3964         int err, ufd;
3965         struct fd f;
3966
3967         if (CHECK_ATTR(BPF_MAP_BATCH))
3968                 return -EINVAL;
3969
3970         ufd = attr->batch.map_fd;
3971         f = fdget(ufd);
3972         map = __bpf_map_get(f);
3973         if (IS_ERR(map))
3974                 return PTR_ERR(map);
3975
3976         if ((cmd == BPF_MAP_LOOKUP_BATCH ||
3977              cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH) &&
3978             !(map_get_sys_perms(map, f) & FMODE_CAN_READ)) {
3979                 err = -EPERM;
3980                 goto err_put;
3981         }
3982
3983         if (cmd != BPF_MAP_LOOKUP_BATCH &&
3984             !(map_get_sys_perms(map, f) & FMODE_CAN_WRITE)) {
3985                 err = -EPERM;
3986                 goto err_put;
3987         }
3988
3989         if (cmd == BPF_MAP_LOOKUP_BATCH)
3990                 BPF_DO_BATCH(map->ops->map_lookup_batch);
3991         else if (cmd == BPF_MAP_LOOKUP_AND_DELETE_BATCH)
3992                 BPF_DO_BATCH(map->ops->map_lookup_and_delete_batch);
3993         else if (cmd == BPF_MAP_UPDATE_BATCH)
3994                 BPF_DO_BATCH(map->ops->map_update_batch);
3995         else
3996                 BPF_DO_BATCH(map->ops->map_delete_batch);
3997
3998 err_put:
3999         fdput(f);
4000         return err;
4001 }
4002
4003 static int tracing_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
4004 {
4005         if (attr->link_create.attach_type != prog->expected_attach_type)
4006                 return -EINVAL;
4007
4008         if (prog->expected_attach_type == BPF_TRACE_ITER)
4009                 return bpf_iter_link_attach(attr, prog);
4010         else if (prog->type == BPF_PROG_TYPE_EXT)
4011                 return bpf_tracing_prog_attach(prog,
4012                                                attr->link_create.target_fd,
4013                                                attr->link_create.target_btf_id);
4014         return -EINVAL;
4015 }
4016
4017 #define BPF_LINK_CREATE_LAST_FIELD link_create.iter_info_len
4018 static int link_create(union bpf_attr *attr)
4019 {
4020         enum bpf_prog_type ptype;
4021         struct bpf_prog *prog;
4022         int ret;
4023
4024         if (CHECK_ATTR(BPF_LINK_CREATE))
4025                 return -EINVAL;
4026
4027         prog = bpf_prog_get(attr->link_create.prog_fd);
4028         if (IS_ERR(prog))
4029                 return PTR_ERR(prog);
4030
4031         ret = bpf_prog_attach_check_attach_type(prog,
4032                                                 attr->link_create.attach_type);
4033         if (ret)
4034                 goto out;
4035
4036         if (prog->type == BPF_PROG_TYPE_EXT) {
4037                 ret = tracing_bpf_link_attach(attr, prog);
4038                 goto out;
4039         }
4040
4041         ptype = attach_type_to_prog_type(attr->link_create.attach_type);
4042         if (ptype == BPF_PROG_TYPE_UNSPEC || ptype != prog->type) {
4043                 ret = -EINVAL;
4044                 goto out;
4045         }
4046
4047         switch (ptype) {
4048         case BPF_PROG_TYPE_CGROUP_SKB:
4049         case BPF_PROG_TYPE_CGROUP_SOCK:
4050         case BPF_PROG_TYPE_CGROUP_SOCK_ADDR:
4051         case BPF_PROG_TYPE_SOCK_OPS:
4052         case BPF_PROG_TYPE_CGROUP_DEVICE:
4053         case BPF_PROG_TYPE_CGROUP_SYSCTL:
4054         case BPF_PROG_TYPE_CGROUP_SOCKOPT:
4055                 ret = cgroup_bpf_link_attach(attr, prog);
4056                 break;
4057         case BPF_PROG_TYPE_TRACING:
4058                 ret = tracing_bpf_link_attach(attr, prog);
4059                 break;
4060         case BPF_PROG_TYPE_FLOW_DISSECTOR:
4061         case BPF_PROG_TYPE_SK_LOOKUP:
4062                 ret = netns_bpf_link_create(attr, prog);
4063                 break;
4064 #ifdef CONFIG_NET
4065         case BPF_PROG_TYPE_XDP:
4066                 ret = bpf_xdp_link_attach(attr, prog);
4067                 break;
4068 #endif
4069         default:
4070                 ret = -EINVAL;
4071         }
4072
4073 out:
4074         if (ret < 0)
4075                 bpf_prog_put(prog);
4076         return ret;
4077 }
4078
4079 #define BPF_LINK_UPDATE_LAST_FIELD link_update.old_prog_fd
4080
4081 static int link_update(union bpf_attr *attr)
4082 {
4083         struct bpf_prog *old_prog = NULL, *new_prog;
4084         struct bpf_link *link;
4085         u32 flags;
4086         int ret;
4087
4088         if (CHECK_ATTR(BPF_LINK_UPDATE))
4089                 return -EINVAL;
4090
4091         flags = attr->link_update.flags;
4092         if (flags & ~BPF_F_REPLACE)
4093                 return -EINVAL;
4094
4095         link = bpf_link_get_from_fd(attr->link_update.link_fd);
4096         if (IS_ERR(link))
4097                 return PTR_ERR(link);
4098
4099         new_prog = bpf_prog_get(attr->link_update.new_prog_fd);
4100         if (IS_ERR(new_prog)) {
4101                 ret = PTR_ERR(new_prog);
4102                 goto out_put_link;
4103         }
4104
4105         if (flags & BPF_F_REPLACE) {
4106                 old_prog = bpf_prog_get(attr->link_update.old_prog_fd);
4107                 if (IS_ERR(old_prog)) {
4108                         ret = PTR_ERR(old_prog);
4109                         old_prog = NULL;
4110                         goto out_put_progs;
4111                 }
4112         } else if (attr->link_update.old_prog_fd) {
4113                 ret = -EINVAL;
4114                 goto out_put_progs;
4115         }
4116
4117         if (link->ops->update_prog)
4118                 ret = link->ops->update_prog(link, new_prog, old_prog);
4119         else
4120                 ret = -EINVAL;
4121
4122 out_put_progs:
4123         if (old_prog)
4124                 bpf_prog_put(old_prog);
4125         if (ret)
4126                 bpf_prog_put(new_prog);
4127 out_put_link:
4128         bpf_link_put(link);
4129         return ret;
4130 }
4131
4132 #define BPF_LINK_DETACH_LAST_FIELD link_detach.link_fd
4133
4134 static int link_detach(union bpf_attr *attr)
4135 {
4136         struct bpf_link *link;
4137         int ret;
4138
4139         if (CHECK_ATTR(BPF_LINK_DETACH))
4140                 return -EINVAL;
4141
4142         link = bpf_link_get_from_fd(attr->link_detach.link_fd);
4143         if (IS_ERR(link))
4144                 return PTR_ERR(link);
4145
4146         if (link->ops->detach)
4147                 ret = link->ops->detach(link);
4148         else
4149                 ret = -EOPNOTSUPP;
4150
4151         bpf_link_put(link);
4152         return ret;
4153 }
4154
4155 static struct bpf_link *bpf_link_inc_not_zero(struct bpf_link *link)
4156 {
4157         return atomic64_fetch_add_unless(&link->refcnt, 1, 0) ? link : ERR_PTR(-ENOENT);
4158 }
4159
4160 struct bpf_link *bpf_link_by_id(u32 id)
4161 {
4162         struct bpf_link *link;
4163
4164         if (!id)
4165                 return ERR_PTR(-ENOENT);
4166
4167         spin_lock_bh(&link_idr_lock);
4168         /* before link is "settled", ID is 0, pretend it doesn't exist yet */
4169         link = idr_find(&link_idr, id);
4170         if (link) {
4171                 if (link->id)
4172                         link = bpf_link_inc_not_zero(link);
4173                 else
4174                         link = ERR_PTR(-EAGAIN);
4175         } else {
4176                 link = ERR_PTR(-ENOENT);
4177         }
4178         spin_unlock_bh(&link_idr_lock);
4179         return link;
4180 }
4181
4182 #define BPF_LINK_GET_FD_BY_ID_LAST_FIELD link_id
4183
4184 static int bpf_link_get_fd_by_id(const union bpf_attr *attr)
4185 {
4186         struct bpf_link *link;
4187         u32 id = attr->link_id;
4188         int fd;
4189
4190         if (CHECK_ATTR(BPF_LINK_GET_FD_BY_ID))
4191                 return -EINVAL;
4192
4193         if (!capable(CAP_SYS_ADMIN))
4194                 return -EPERM;
4195
4196         link = bpf_link_by_id(id);
4197         if (IS_ERR(link))
4198                 return PTR_ERR(link);
4199
4200         fd = bpf_link_new_fd(link);
4201         if (fd < 0)
4202                 bpf_link_put(link);
4203
4204         return fd;
4205 }
4206
4207 DEFINE_MUTEX(bpf_stats_enabled_mutex);
4208
4209 static int bpf_stats_release(struct inode *inode, struct file *file)
4210 {
4211         mutex_lock(&bpf_stats_enabled_mutex);
4212         static_key_slow_dec(&bpf_stats_enabled_key.key);
4213         mutex_unlock(&bpf_stats_enabled_mutex);
4214         return 0;
4215 }
4216
4217 static const struct file_operations bpf_stats_fops = {
4218         .release = bpf_stats_release,
4219 };
4220
4221 static int bpf_enable_runtime_stats(void)
4222 {
4223         int fd;
4224
4225         mutex_lock(&bpf_stats_enabled_mutex);
4226
4227         /* Set a very high limit to avoid overflow */
4228         if (static_key_count(&bpf_stats_enabled_key.key) > INT_MAX / 2) {
4229                 mutex_unlock(&bpf_stats_enabled_mutex);
4230                 return -EBUSY;
4231         }
4232
4233         fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
4234         if (fd >= 0)
4235                 static_key_slow_inc(&bpf_stats_enabled_key.key);
4236
4237         mutex_unlock(&bpf_stats_enabled_mutex);
4238         return fd;
4239 }
4240
4241 #define BPF_ENABLE_STATS_LAST_FIELD enable_stats.type
4242
4243 static int bpf_enable_stats(union bpf_attr *attr)
4244 {
4245
4246         if (CHECK_ATTR(BPF_ENABLE_STATS))
4247                 return -EINVAL;
4248
4249         if (!capable(CAP_SYS_ADMIN))
4250                 return -EPERM;
4251
4252         switch (attr->enable_stats.type) {
4253         case BPF_STATS_RUN_TIME:
4254                 return bpf_enable_runtime_stats();
4255         default:
4256                 break;
4257         }
4258         return -EINVAL;
4259 }
4260
4261 #define BPF_ITER_CREATE_LAST_FIELD iter_create.flags
4262
4263 static int bpf_iter_create(union bpf_attr *attr)
4264 {
4265         struct bpf_link *link;
4266         int err;
4267
4268         if (CHECK_ATTR(BPF_ITER_CREATE))
4269                 return -EINVAL;
4270
4271         if (attr->iter_create.flags)
4272                 return -EINVAL;
4273
4274         link = bpf_link_get_from_fd(attr->iter_create.link_fd);
4275         if (IS_ERR(link))
4276                 return PTR_ERR(link);
4277
4278         err = bpf_iter_new_fd(link);
4279         bpf_link_put(link);
4280
4281         return err;
4282 }
4283
4284 #define BPF_PROG_BIND_MAP_LAST_FIELD prog_bind_map.flags
4285
4286 static int bpf_prog_bind_map(union bpf_attr *attr)
4287 {
4288         struct bpf_prog *prog;
4289         struct bpf_map *map;
4290         struct bpf_map **used_maps_old, **used_maps_new;
4291         int i, ret = 0;
4292
4293         if (CHECK_ATTR(BPF_PROG_BIND_MAP))
4294                 return -EINVAL;
4295
4296         if (attr->prog_bind_map.flags)
4297                 return -EINVAL;
4298
4299         prog = bpf_prog_get(attr->prog_bind_map.prog_fd);
4300         if (IS_ERR(prog))
4301                 return PTR_ERR(prog);
4302
4303         map = bpf_map_get(attr->prog_bind_map.map_fd);
4304         if (IS_ERR(map)) {
4305                 ret = PTR_ERR(map);
4306                 goto out_prog_put;
4307         }
4308
4309         mutex_lock(&prog->aux->used_maps_mutex);
4310
4311         used_maps_old = prog->aux->used_maps;
4312
4313         for (i = 0; i < prog->aux->used_map_cnt; i++)
4314                 if (used_maps_old[i] == map) {
4315                         bpf_map_put(map);
4316                         goto out_unlock;
4317                 }
4318
4319         used_maps_new = kmalloc_array(prog->aux->used_map_cnt + 1,
4320                                       sizeof(used_maps_new[0]),
4321                                       GFP_KERNEL);
4322         if (!used_maps_new) {
4323                 ret = -ENOMEM;
4324                 goto out_unlock;
4325         }
4326
4327         memcpy(used_maps_new, used_maps_old,
4328                sizeof(used_maps_old[0]) * prog->aux->used_map_cnt);
4329         used_maps_new[prog->aux->used_map_cnt] = map;
4330
4331         prog->aux->used_map_cnt++;
4332         prog->aux->used_maps = used_maps_new;
4333
4334         kfree(used_maps_old);
4335
4336 out_unlock:
4337         mutex_unlock(&prog->aux->used_maps_mutex);
4338
4339         if (ret)
4340                 bpf_map_put(map);
4341 out_prog_put:
4342         bpf_prog_put(prog);
4343         return ret;
4344 }
4345
4346 SYSCALL_DEFINE3(bpf, int, cmd, union bpf_attr __user *, uattr, unsigned int, size)
4347 {
4348         union bpf_attr attr;
4349         int err;
4350
4351         if (sysctl_unprivileged_bpf_disabled && !bpf_capable())
4352                 return -EPERM;
4353
4354         err = bpf_check_uarg_tail_zero(uattr, sizeof(attr), size);
4355         if (err)
4356                 return err;
4357         size = min_t(u32, size, sizeof(attr));
4358
4359         /* copy attributes from user space, may be less than sizeof(bpf_attr) */
4360         memset(&attr, 0, sizeof(attr));
4361         if (copy_from_user(&attr, uattr, size) != 0)
4362                 return -EFAULT;
4363
4364         err = security_bpf(cmd, &attr, size);
4365         if (err < 0)
4366                 return err;
4367
4368         switch (cmd) {
4369         case BPF_MAP_CREATE:
4370                 err = map_create(&attr);
4371                 break;
4372         case BPF_MAP_LOOKUP_ELEM:
4373                 err = map_lookup_elem(&attr);
4374                 break;
4375         case BPF_MAP_UPDATE_ELEM:
4376                 err = map_update_elem(&attr);
4377                 break;
4378         case BPF_MAP_DELETE_ELEM:
4379                 err = map_delete_elem(&attr);
4380                 break;
4381         case BPF_MAP_GET_NEXT_KEY:
4382                 err = map_get_next_key(&attr);
4383                 break;
4384         case BPF_MAP_FREEZE:
4385                 err = map_freeze(&attr);
4386                 break;
4387         case BPF_PROG_LOAD:
4388                 err = bpf_prog_load(&attr, uattr);
4389                 break;
4390         case BPF_OBJ_PIN:
4391                 err = bpf_obj_pin(&attr);
4392                 break;
4393         case BPF_OBJ_GET:
4394                 err = bpf_obj_get(&attr);
4395                 break;
4396         case BPF_PROG_ATTACH:
4397                 err = bpf_prog_attach(&attr);
4398                 break;
4399         case BPF_PROG_DETACH:
4400                 err = bpf_prog_detach(&attr);
4401                 break;
4402         case BPF_PROG_QUERY:
4403                 err = bpf_prog_query(&attr, uattr);
4404                 break;
4405         case BPF_PROG_TEST_RUN:
4406                 err = bpf_prog_test_run(&attr, uattr);
4407                 break;
4408         case BPF_PROG_GET_NEXT_ID:
4409                 err = bpf_obj_get_next_id(&attr, uattr,
4410                                           &prog_idr, &prog_idr_lock);
4411                 break;
4412         case BPF_MAP_GET_NEXT_ID:
4413                 err = bpf_obj_get_next_id(&attr, uattr,
4414                                           &map_idr, &map_idr_lock);
4415                 break;
4416         case BPF_BTF_GET_NEXT_ID:
4417                 err = bpf_obj_get_next_id(&attr, uattr,
4418                                           &btf_idr, &btf_idr_lock);
4419                 break;
4420         case BPF_PROG_GET_FD_BY_ID:
4421                 err = bpf_prog_get_fd_by_id(&attr);
4422                 break;
4423         case BPF_MAP_GET_FD_BY_ID:
4424                 err = bpf_map_get_fd_by_id(&attr);
4425                 break;
4426         case BPF_OBJ_GET_INFO_BY_FD:
4427                 err = bpf_obj_get_info_by_fd(&attr, uattr);
4428                 break;
4429         case BPF_RAW_TRACEPOINT_OPEN:
4430                 err = bpf_raw_tracepoint_open(&attr);
4431                 break;
4432         case BPF_BTF_LOAD:
4433                 err = bpf_btf_load(&attr);
4434                 break;
4435         case BPF_BTF_GET_FD_BY_ID:
4436                 err = bpf_btf_get_fd_by_id(&attr);
4437                 break;
4438         case BPF_TASK_FD_QUERY:
4439                 err = bpf_task_fd_query(&attr, uattr);
4440                 break;
4441         case BPF_MAP_LOOKUP_AND_DELETE_ELEM:
4442                 err = map_lookup_and_delete_elem(&attr);
4443                 break;
4444         case BPF_MAP_LOOKUP_BATCH:
4445                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_LOOKUP_BATCH);
4446                 break;
4447         case BPF_MAP_LOOKUP_AND_DELETE_BATCH:
4448                 err = bpf_map_do_batch(&attr, uattr,
4449                                        BPF_MAP_LOOKUP_AND_DELETE_BATCH);
4450                 break;
4451         case BPF_MAP_UPDATE_BATCH:
4452                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_UPDATE_BATCH);
4453                 break;
4454         case BPF_MAP_DELETE_BATCH:
4455                 err = bpf_map_do_batch(&attr, uattr, BPF_MAP_DELETE_BATCH);
4456                 break;
4457         case BPF_LINK_CREATE:
4458                 err = link_create(&attr);
4459                 break;
4460         case BPF_LINK_UPDATE:
4461                 err = link_update(&attr);
4462                 break;
4463         case BPF_LINK_GET_FD_BY_ID:
4464                 err = bpf_link_get_fd_by_id(&attr);
4465                 break;
4466         case BPF_LINK_GET_NEXT_ID:
4467                 err = bpf_obj_get_next_id(&attr, uattr,
4468                                           &link_idr, &link_idr_lock);
4469                 break;
4470         case BPF_ENABLE_STATS:
4471                 err = bpf_enable_stats(&attr);
4472                 break;
4473         case BPF_ITER_CREATE:
4474                 err = bpf_iter_create(&attr);
4475                 break;
4476         case BPF_LINK_DETACH:
4477                 err = link_detach(&attr);
4478                 break;
4479         case BPF_PROG_BIND_MAP:
4480                 err = bpf_prog_bind_map(&attr);
4481                 break;
4482         default:
4483                 err = -EINVAL;
4484                 break;
4485         }
4486
4487         return err;
4488 }