net/core/sock_map.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
   3
   4 #include <linux/bpf.h>
   5 #include <linux/filter.h>
   6 #include <linux/errno.h>
   7 #include <linux/file.h>
   8 #include <linux/net.h>
   9 #include <linux/workqueue.h>
  10 #include <linux/skmsg.h>
  11 #include <linux/list.h>
  12 #include <linux/jhash.h>
  13
  14 struct bpf_stab {
  15         struct bpf_map map;
  16         struct sock **sks;
  17         struct sk_psock_progs progs;
  18         raw_spinlock_t lock;
  19 };
  20
  21 #define SOCK_CREATE_FLAG_MASK                           \
  22         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
  23
  24 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  25 {
  26         struct bpf_stab *stab;
  27         u64 cost;
  28         int err;
  29
  30         if (!capable(CAP_NET_ADMIN))
  31                 return ERR_PTR(-EPERM);
  32         if (attr->max_entries == 0 ||
  33             attr->key_size    != 4 ||
  34             attr->value_size  != 4 ||
  35             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  36                 return ERR_PTR(-EINVAL);
  37
  38         stab = kzalloc(sizeof(*stab), GFP_USER);
  39         if (!stab)
  40                 return ERR_PTR(-ENOMEM);
  41
  42         bpf_map_init_from_attr(&stab->map, attr);
  43         raw_spin_lock_init(&stab->lock);
  44
  45         /* Make sure page count doesn't overflow. */
  46         cost = (u64) stab->map.max_entries * sizeof(struct sock *);
  47         err = bpf_map_charge_init(&stab->map.memory, cost);
  48         if (err)
  49                 goto free_stab;
  50
  51         stab->sks = bpf_map_area_alloc(stab->map.max_entries *
  52                                        sizeof(struct sock *),
  53                                        stab->map.numa_node);
  54         if (stab->sks)
  55                 return &stab->map;
  56         err = -ENOMEM;
  57         bpf_map_charge_finish(&stab->map.memory);
  58 free_stab:
  59         kfree(stab);
  60         return ERR_PTR(err);
  61 }
  62
  63 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  64 {
  65         u32 ufd = attr->target_fd;
  66         struct bpf_map *map;
  67         struct fd f;
  68         int ret;
  69
  70         f = fdget(ufd);
  71         map = __bpf_map_get(f);
  72         if (IS_ERR(map))
  73                 return PTR_ERR(map);
  74         ret = sock_map_prog_update(map, prog, attr->attach_type);
  75         fdput(f);
  76         return ret;
  77 }
  78
  79 static void sock_map_sk_acquire(struct sock *sk)
  80         __acquires(&sk->sk_lock.slock)
  81 {
  82         lock_sock(sk);
  83         preempt_disable();
  84         rcu_read_lock();
  85 }
  86
  87 static void sock_map_sk_release(struct sock *sk)
  88         __releases(&sk->sk_lock.slock)
  89 {
  90         rcu_read_unlock();
  91         preempt_enable();
  92         release_sock(sk);
  93 }
  94
  95 static void sock_map_add_link(struct sk_psock *psock,
  96                               struct sk_psock_link *link,
  97                               struct bpf_map *map, void *link_raw)
  98 {
  99         link->link_raw = link_raw;
 100         link->map = map;
 101         spin_lock_bh(&psock->link_lock);
 102         list_add_tail(&link->list, &psock->link);
 103         spin_unlock_bh(&psock->link_lock);
 104 }
 105
 106 static void sock_map_del_link(struct sock *sk,
 107                               struct sk_psock *psock, void *link_raw)
 108 {
 109         struct sk_psock_link *link, *tmp;
 110         bool strp_stop = false;
 111
 112         spin_lock_bh(&psock->link_lock);
 113         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 114                 if (link->link_raw == link_raw) {
 115                         struct bpf_map *map = link->map;
 116                         struct bpf_stab *stab = container_of(map, struct bpf_stab,
 117                                                              map);
 118                         if (psock->parser.enabled && stab->progs.skb_parser)
 119                                 strp_stop = true;
 120                         list_del(&link->list);
 121                         sk_psock_free_link(link);
 122                 }
 123         }
 124         spin_unlock_bh(&psock->link_lock);
 125         if (strp_stop) {
 126                 write_lock_bh(&sk->sk_callback_lock);
 127                 sk_psock_stop_strp(sk, psock);
 128                 write_unlock_bh(&sk->sk_callback_lock);
 129         }
 130 }
 131
 132 static void sock_map_unref(struct sock *sk, void *link_raw)
 133 {
 134         struct sk_psock *psock = sk_psock(sk);
 135
 136         if (likely(psock)) {
 137                 sock_map_del_link(sk, psock, link_raw);
 138                 sk_psock_put(sk, psock);
 139         }
 140 }
 141
 142 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
 143                          struct sock *sk)
 144 {
 145         struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
 146         bool skb_progs, sk_psock_is_new = false;
 147         struct sk_psock *psock;
 148         int ret;
 149
 150         skb_verdict = READ_ONCE(progs->skb_verdict);
 151         skb_parser = READ_ONCE(progs->skb_parser);
 152         skb_progs = skb_parser && skb_verdict;
 153         if (skb_progs) {
 154                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 155                 if (IS_ERR(skb_verdict))
 156                         return PTR_ERR(skb_verdict);
 157                 skb_parser = bpf_prog_inc_not_zero(skb_parser);
 158                 if (IS_ERR(skb_parser)) {
 159                         bpf_prog_put(skb_verdict);
 160                         return PTR_ERR(skb_parser);
 161                 }
 162         }
 163
 164         msg_parser = READ_ONCE(progs->msg_parser);
 165         if (msg_parser) {
 166                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 167                 if (IS_ERR(msg_parser)) {
 168                         ret = PTR_ERR(msg_parser);
 169                         goto out;
 170                 }
 171         }
 172
 173         psock = sk_psock_get_checked(sk);
 174         if (IS_ERR(psock)) {
 175                 ret = PTR_ERR(psock);
 176                 goto out_progs;
 177         }
 178
 179         if (psock) {
 180                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 181                     (skb_progs  && READ_ONCE(psock->progs.skb_parser))) {
 182                         sk_psock_put(sk, psock);
 183                         ret = -EBUSY;
 184                         goto out_progs;
 185                 }
 186         } else {
 187                 psock = sk_psock_init(sk, map->numa_node);
 188                 if (!psock) {
 189                         ret = -ENOMEM;
 190                         goto out_progs;
 191                 }
 192                 sk_psock_is_new = true;
 193         }
 194
 195         if (msg_parser)
 196                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 197         if (sk_psock_is_new) {
 198                 ret = tcp_bpf_init(sk);
 199                 if (ret < 0)
 200                         goto out_drop;
 201         } else {
 202                 tcp_bpf_reinit(sk);
 203         }
 204
 205         write_lock_bh(&sk->sk_callback_lock);
 206         if (skb_progs && !psock->parser.enabled) {
 207                 ret = sk_psock_init_strp(sk, psock);
 208                 if (ret) {
 209                         write_unlock_bh(&sk->sk_callback_lock);
 210                         goto out_drop;
 211                 }
 212                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 213                 psock_set_prog(&psock->progs.skb_parser, skb_parser);
 214                 sk_psock_start_strp(sk, psock);
 215         }
 216         write_unlock_bh(&sk->sk_callback_lock);
 217         return 0;
 218 out_drop:
 219         sk_psock_put(sk, psock);
 220 out_progs:
 221         if (msg_parser)
 222                 bpf_prog_put(msg_parser);
 223 out:
 224         if (skb_progs) {
 225                 bpf_prog_put(skb_verdict);
 226                 bpf_prog_put(skb_parser);
 227         }
 228         return ret;
 229 }
 230
 231 static void sock_map_free(struct bpf_map *map)
 232 {
 233         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 234         int i;
 235
 236         synchronize_rcu();
 237         raw_spin_lock_bh(&stab->lock);
 238         for (i = 0; i < stab->map.max_entries; i++) {
 239                 struct sock **psk = &stab->sks[i];
 240                 struct sock *sk;
 241
 242                 sk = xchg(psk, NULL);
 243                 if (sk) {
 244                         lock_sock(sk);
 245                         rcu_read_lock();
 246                         sock_map_unref(sk, psk);
 247                         rcu_read_unlock();
 248                         release_sock(sk);
 249                 }
 250         }
 251         raw_spin_unlock_bh(&stab->lock);
 252
 253         synchronize_rcu();
 254
 255         bpf_map_area_free(stab->sks);
 256         kfree(stab);
 257 }
 258
 259 static void sock_map_release_progs(struct bpf_map *map)
 260 {
 261         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 262 }
 263
 264 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 265 {
 266         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 267
 268         WARN_ON_ONCE(!rcu_read_lock_held());
 269
 270         if (unlikely(key >= map->max_entries))
 271                 return NULL;
 272         return READ_ONCE(stab->sks[key]);
 273 }
 274
 275 static void *sock_map_lookup(struct bpf_map *map, void *key)
 276 {
 277         return ERR_PTR(-EOPNOTSUPP);
 278 }
 279
 280 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 281                              struct sock **psk)
 282 {
 283         struct sock *sk;
 284         int err = 0;
 285
 286         raw_spin_lock_bh(&stab->lock);
 287         sk = *psk;
 288         if (!sk_test || sk_test == sk)
 289                 sk = xchg(psk, NULL);
 290
 291         if (likely(sk))
 292                 sock_map_unref(sk, psk);
 293         else
 294                 err = -EINVAL;
 295
 296         raw_spin_unlock_bh(&stab->lock);
 297         return err;
 298 }
 299
 300 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 301                                       void *link_raw)
 302 {
 303         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 304
 305         __sock_map_delete(stab, sk, link_raw);
 306 }
 307
 308 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 309 {
 310         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 311         u32 i = *(u32 *)key;
 312         struct sock **psk;
 313
 314         if (unlikely(i >= map->max_entries))
 315                 return -EINVAL;
 316
 317         psk = &stab->sks[i];
 318         return __sock_map_delete(stab, NULL, psk);
 319 }
 320
 321 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 322 {
 323         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 324         u32 i = key ? *(u32 *)key : U32_MAX;
 325         u32 *key_next = next;
 326
 327         if (i == stab->map.max_entries - 1)
 328                 return -ENOENT;
 329         if (i >= stab->map.max_entries)
 330                 *key_next = 0;
 331         else
 332                 *key_next = i + 1;
 333         return 0;
 334 }
 335
 336 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 337                                   struct sock *sk, u64 flags)
 338 {
 339         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 340         struct inet_connection_sock *icsk = inet_csk(sk);
 341         struct sk_psock_link *link;
 342         struct sk_psock *psock;
 343         struct sock *osk;
 344         int ret;
 345
 346         WARN_ON_ONCE(!rcu_read_lock_held());
 347         if (unlikely(flags > BPF_EXIST))
 348                 return -EINVAL;
 349         if (unlikely(idx >= map->max_entries))
 350                 return -E2BIG;
 351         if (unlikely(rcu_access_pointer(icsk->icsk_ulp_data)))
 352                 return -EINVAL;
 353
 354         link = sk_psock_init_link();
 355         if (!link)
 356                 return -ENOMEM;
 357
 358         ret = sock_map_link(map, &stab->progs, sk);
 359         if (ret < 0)
 360                 goto out_free;
 361
 362         psock = sk_psock(sk);
 363         WARN_ON_ONCE(!psock);
 364
 365         raw_spin_lock_bh(&stab->lock);
 366         osk = stab->sks[idx];
 367         if (osk && flags == BPF_NOEXIST) {
 368                 ret = -EEXIST;
 369                 goto out_unlock;
 370         } else if (!osk && flags == BPF_EXIST) {
 371                 ret = -ENOENT;
 372                 goto out_unlock;
 373         }
 374
 375         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 376         stab->sks[idx] = sk;
 377         if (osk)
 378                 sock_map_unref(osk, &stab->sks[idx]);
 379         raw_spin_unlock_bh(&stab->lock);
 380         return 0;
 381 out_unlock:
 382         raw_spin_unlock_bh(&stab->lock);
 383         if (psock)
 384                 sk_psock_put(sk, psock);
 385 out_free:
 386         sk_psock_free_link(link);
 387         return ret;
 388 }
 389
 390 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 391 {
 392         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 393                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB;
 394 }
 395
 396 static bool sock_map_sk_is_suitable(const struct sock *sk)
 397 {
 398         return sk->sk_type == SOCK_STREAM &&
 399                sk->sk_protocol == IPPROTO_TCP;
 400 }
 401
 402 static int sock_map_update_elem(struct bpf_map *map, void *key,
 403                                 void *value, u64 flags)
 404 {
 405         u32 ufd = *(u32 *)value;
 406         u32 idx = *(u32 *)key;
 407         struct socket *sock;
 408         struct sock *sk;
 409         int ret;
 410
 411         sock = sockfd_lookup(ufd, &ret);
 412         if (!sock)
 413                 return ret;
 414         sk = sock->sk;
 415         if (!sk) {
 416                 ret = -EINVAL;
 417                 goto out;
 418         }
 419         if (!sock_map_sk_is_suitable(sk)) {
 420                 ret = -EOPNOTSUPP;
 421                 goto out;
 422         }
 423
 424         sock_map_sk_acquire(sk);
 425         if (sk->sk_state != TCP_ESTABLISHED)
 426                 ret = -EOPNOTSUPP;
 427         else
 428                 ret = sock_map_update_common(map, idx, sk, flags);
 429         sock_map_sk_release(sk);
 430 out:
 431         fput(sock->file);
 432         return ret;
 433 }
 434
 435 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 436            struct bpf_map *, map, void *, key, u64, flags)
 437 {
 438         WARN_ON_ONCE(!rcu_read_lock_held());
 439
 440         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 441                    sock_map_op_okay(sops)))
 442                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 443                                               flags);
 444         return -EOPNOTSUPP;
 445 }
 446
 447 const struct bpf_func_proto bpf_sock_map_update_proto = {
 448         .func           = bpf_sock_map_update,
 449         .gpl_only       = false,
 450         .pkt_access     = true,
 451         .ret_type       = RET_INTEGER,
 452         .arg1_type      = ARG_PTR_TO_CTX,
 453         .arg2_type      = ARG_CONST_MAP_PTR,
 454         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 455         .arg4_type      = ARG_ANYTHING,
 456 };
 457
 458 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 459            struct bpf_map *, map, u32, key, u64, flags)
 460 {
 461         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 462
 463         if (unlikely(flags & ~(BPF_F_INGRESS)))
 464                 return SK_DROP;
 465         tcb->bpf.flags = flags;
 466         tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
 467         if (!tcb->bpf.sk_redir)
 468                 return SK_DROP;
 469         return SK_PASS;
 470 }
 471
 472 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 473         .func           = bpf_sk_redirect_map,
 474         .gpl_only       = false,
 475         .ret_type       = RET_INTEGER,
 476         .arg1_type      = ARG_PTR_TO_CTX,
 477         .arg2_type      = ARG_CONST_MAP_PTR,
 478         .arg3_type      = ARG_ANYTHING,
 479         .arg4_type      = ARG_ANYTHING,
 480 };
 481
 482 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 483            struct bpf_map *, map, u32, key, u64, flags)
 484 {
 485         if (unlikely(flags & ~(BPF_F_INGRESS)))
 486                 return SK_DROP;
 487         msg->flags = flags;
 488         msg->sk_redir = __sock_map_lookup_elem(map, key);
 489         if (!msg->sk_redir)
 490                 return SK_DROP;
 491         return SK_PASS;
 492 }
 493
 494 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 495         .func           = bpf_msg_redirect_map,
 496         .gpl_only       = false,
 497         .ret_type       = RET_INTEGER,
 498         .arg1_type      = ARG_PTR_TO_CTX,
 499         .arg2_type      = ARG_CONST_MAP_PTR,
 500         .arg3_type      = ARG_ANYTHING,
 501         .arg4_type      = ARG_ANYTHING,
 502 };
 503
 504 const struct bpf_map_ops sock_map_ops = {
 505         .map_alloc              = sock_map_alloc,
 506         .map_free               = sock_map_free,
 507         .map_get_next_key       = sock_map_get_next_key,
 508         .map_update_elem        = sock_map_update_elem,
 509         .map_delete_elem        = sock_map_delete_elem,
 510         .map_lookup_elem        = sock_map_lookup,
 511         .map_release_uref       = sock_map_release_progs,
 512         .map_check_btf          = map_check_no_btf,
 513 };
 514
 515 struct bpf_htab_elem {
 516         struct rcu_head rcu;
 517         u32 hash;
 518         struct sock *sk;
 519         struct hlist_node node;
 520         u8 key[0];
 521 };
 522
 523 struct bpf_htab_bucket {
 524         struct hlist_head head;
 525         raw_spinlock_t lock;
 526 };
 527
 528 struct bpf_htab {
 529         struct bpf_map map;
 530         struct bpf_htab_bucket *buckets;
 531         u32 buckets_num;
 532         u32 elem_size;
 533         struct sk_psock_progs progs;
 534         atomic_t count;
 535 };
 536
 537 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 538 {
 539         return jhash(key, len, 0);
 540 }
 541
 542 static struct bpf_htab_bucket *sock_hash_select_bucket(struct bpf_htab *htab,
 543                                                        u32 hash)
 544 {
 545         return &htab->buckets[hash & (htab->buckets_num - 1)];
 546 }
 547
 548 static struct bpf_htab_elem *
 549 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 550                           u32 key_size)
 551 {
 552         struct bpf_htab_elem *elem;
 553
 554         hlist_for_each_entry_rcu(elem, head, node) {
 555                 if (elem->hash == hash &&
 556                     !memcmp(&elem->key, key, key_size))
 557                         return elem;
 558         }
 559
 560         return NULL;
 561 }
 562
 563 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 564 {
 565         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 566         u32 key_size = map->key_size, hash;
 567         struct bpf_htab_bucket *bucket;
 568         struct bpf_htab_elem *elem;
 569
 570         WARN_ON_ONCE(!rcu_read_lock_held());
 571
 572         hash = sock_hash_bucket_hash(key, key_size);
 573         bucket = sock_hash_select_bucket(htab, hash);
 574         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 575
 576         return elem ? elem->sk : NULL;
 577 }
 578
 579 static void sock_hash_free_elem(struct bpf_htab *htab,
 580                                 struct bpf_htab_elem *elem)
 581 {
 582         atomic_dec(&htab->count);
 583         kfree_rcu(elem, rcu);
 584 }
 585
 586 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 587                                        void *link_raw)
 588 {
 589         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 590         struct bpf_htab_elem *elem_probe, *elem = link_raw;
 591         struct bpf_htab_bucket *bucket;
 592
 593         WARN_ON_ONCE(!rcu_read_lock_held());
 594         bucket = sock_hash_select_bucket(htab, elem->hash);
 595
 596         /* elem may be deleted in parallel from the map, but access here
 597          * is okay since it's going away only after RCU grace period.
 598          * However, we need to check whether it's still present.
 599          */
 600         raw_spin_lock_bh(&bucket->lock);
 601         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 602                                                elem->key, map->key_size);
 603         if (elem_probe && elem_probe == elem) {
 604                 hlist_del_rcu(&elem->node);
 605                 sock_map_unref(elem->sk, elem);
 606                 sock_hash_free_elem(htab, elem);
 607         }
 608         raw_spin_unlock_bh(&bucket->lock);
 609 }
 610
 611 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 612 {
 613         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 614         u32 hash, key_size = map->key_size;
 615         struct bpf_htab_bucket *bucket;
 616         struct bpf_htab_elem *elem;
 617         int ret = -ENOENT;
 618
 619         hash = sock_hash_bucket_hash(key, key_size);
 620         bucket = sock_hash_select_bucket(htab, hash);
 621
 622         raw_spin_lock_bh(&bucket->lock);
 623         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 624         if (elem) {
 625                 hlist_del_rcu(&elem->node);
 626                 sock_map_unref(elem->sk, elem);
 627                 sock_hash_free_elem(htab, elem);
 628                 ret = 0;
 629         }
 630         raw_spin_unlock_bh(&bucket->lock);
 631         return ret;
 632 }
 633
 634 static struct bpf_htab_elem *sock_hash_alloc_elem(struct bpf_htab *htab,
 635                                                   void *key, u32 key_size,
 636                                                   u32 hash, struct sock *sk,
 637                                                   struct bpf_htab_elem *old)
 638 {
 639         struct bpf_htab_elem *new;
 640
 641         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 642                 if (!old) {
 643                         atomic_dec(&htab->count);
 644                         return ERR_PTR(-E2BIG);
 645                 }
 646         }
 647
 648         new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
 649                            htab->map.numa_node);
 650         if (!new) {
 651                 atomic_dec(&htab->count);
 652                 return ERR_PTR(-ENOMEM);
 653         }
 654         memcpy(new->key, key, key_size);
 655         new->sk = sk;
 656         new->hash = hash;
 657         return new;
 658 }
 659
 660 static int sock_hash_update_common(struct bpf_map *map, void *key,
 661                                    struct sock *sk, u64 flags)
 662 {
 663         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 664         struct inet_connection_sock *icsk = inet_csk(sk);
 665         u32 key_size = map->key_size, hash;
 666         struct bpf_htab_elem *elem, *elem_new;
 667         struct bpf_htab_bucket *bucket;
 668         struct sk_psock_link *link;
 669         struct sk_psock *psock;
 670         int ret;
 671
 672         WARN_ON_ONCE(!rcu_read_lock_held());
 673         if (unlikely(flags > BPF_EXIST))
 674                 return -EINVAL;
 675         if (unlikely(icsk->icsk_ulp_data))
 676                 return -EINVAL;
 677
 678         link = sk_psock_init_link();
 679         if (!link)
 680                 return -ENOMEM;
 681
 682         ret = sock_map_link(map, &htab->progs, sk);
 683         if (ret < 0)
 684                 goto out_free;
 685
 686         psock = sk_psock(sk);
 687         WARN_ON_ONCE(!psock);
 688
 689         hash = sock_hash_bucket_hash(key, key_size);
 690         bucket = sock_hash_select_bucket(htab, hash);
 691
 692         raw_spin_lock_bh(&bucket->lock);
 693         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 694         if (elem && flags == BPF_NOEXIST) {
 695                 ret = -EEXIST;
 696                 goto out_unlock;
 697         } else if (!elem && flags == BPF_EXIST) {
 698                 ret = -ENOENT;
 699                 goto out_unlock;
 700         }
 701
 702         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
 703         if (IS_ERR(elem_new)) {
 704                 ret = PTR_ERR(elem_new);
 705                 goto out_unlock;
 706         }
 707
 708         sock_map_add_link(psock, link, map, elem_new);
 709         /* Add new element to the head of the list, so that
 710          * concurrent search will find it before old elem.
 711          */
 712         hlist_add_head_rcu(&elem_new->node, &bucket->head);
 713         if (elem) {
 714                 hlist_del_rcu(&elem->node);
 715                 sock_map_unref(elem->sk, elem);
 716                 sock_hash_free_elem(htab, elem);
 717         }
 718         raw_spin_unlock_bh(&bucket->lock);
 719         return 0;
 720 out_unlock:
 721         raw_spin_unlock_bh(&bucket->lock);
 722         sk_psock_put(sk, psock);
 723 out_free:
 724         sk_psock_free_link(link);
 725         return ret;
 726 }
 727
 728 static int sock_hash_update_elem(struct bpf_map *map, void *key,
 729                                  void *value, u64 flags)
 730 {
 731         u32 ufd = *(u32 *)value;
 732         struct socket *sock;
 733         struct sock *sk;
 734         int ret;
 735
 736         sock = sockfd_lookup(ufd, &ret);
 737         if (!sock)
 738                 return ret;
 739         sk = sock->sk;
 740         if (!sk) {
 741                 ret = -EINVAL;
 742                 goto out;
 743         }
 744         if (!sock_map_sk_is_suitable(sk)) {
 745                 ret = -EOPNOTSUPP;
 746                 goto out;
 747         }
 748
 749         sock_map_sk_acquire(sk);
 750         if (sk->sk_state != TCP_ESTABLISHED)
 751                 ret = -EOPNOTSUPP;
 752         else
 753                 ret = sock_hash_update_common(map, key, sk, flags);
 754         sock_map_sk_release(sk);
 755 out:
 756         fput(sock->file);
 757         return ret;
 758 }
 759
 760 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
 761                                   void *key_next)
 762 {
 763         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 764         struct bpf_htab_elem *elem, *elem_next;
 765         u32 hash, key_size = map->key_size;
 766         struct hlist_head *head;
 767         int i = 0;
 768
 769         if (!key)
 770                 goto find_first_elem;
 771         hash = sock_hash_bucket_hash(key, key_size);
 772         head = &sock_hash_select_bucket(htab, hash)->head;
 773         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
 774         if (!elem)
 775                 goto find_first_elem;
 776
 777         elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&elem->node)),
 778                                      struct bpf_htab_elem, node);
 779         if (elem_next) {
 780                 memcpy(key_next, elem_next->key, key_size);
 781                 return 0;
 782         }
 783
 784         i = hash & (htab->buckets_num - 1);
 785         i++;
 786 find_first_elem:
 787         for (; i < htab->buckets_num; i++) {
 788                 head = &sock_hash_select_bucket(htab, i)->head;
 789                 elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
 790                                              struct bpf_htab_elem, node);
 791                 if (elem_next) {
 792                         memcpy(key_next, elem_next->key, key_size);
 793                         return 0;
 794                 }
 795         }
 796
 797         return -ENOENT;
 798 }
 799
 800 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
 801 {
 802         struct bpf_htab *htab;
 803         int i, err;
 804         u64 cost;
 805
 806         if (!capable(CAP_NET_ADMIN))
 807                 return ERR_PTR(-EPERM);
 808         if (attr->max_entries == 0 ||
 809             attr->key_size    == 0 ||
 810             attr->value_size  != 4 ||
 811             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
 812                 return ERR_PTR(-EINVAL);
 813         if (attr->key_size > MAX_BPF_STACK)
 814                 return ERR_PTR(-E2BIG);
 815
 816         htab = kzalloc(sizeof(*htab), GFP_USER);
 817         if (!htab)
 818                 return ERR_PTR(-ENOMEM);
 819
 820         bpf_map_init_from_attr(&htab->map, attr);
 821
 822         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
 823         htab->elem_size = sizeof(struct bpf_htab_elem) +
 824                           round_up(htab->map.key_size, 8);
 825         if (htab->buckets_num == 0 ||
 826             htab->buckets_num > U32_MAX / sizeof(struct bpf_htab_bucket)) {
 827                 err = -EINVAL;
 828                 goto free_htab;
 829         }
 830
 831         cost = (u64) htab->buckets_num * sizeof(struct bpf_htab_bucket) +
 832                (u64) htab->elem_size * htab->map.max_entries;
 833         if (cost >= U32_MAX - PAGE_SIZE) {
 834                 err = -EINVAL;
 835                 goto free_htab;
 836         }
 837
 838         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
 839                                            sizeof(struct bpf_htab_bucket),
 840                                            htab->map.numa_node);
 841         if (!htab->buckets) {
 842                 err = -ENOMEM;
 843                 goto free_htab;
 844         }
 845
 846         for (i = 0; i < htab->buckets_num; i++) {
 847                 INIT_HLIST_HEAD(&htab->buckets[i].head);
 848                 raw_spin_lock_init(&htab->buckets[i].lock);
 849         }
 850
 851         return &htab->map;
 852 free_htab:
 853         kfree(htab);
 854         return ERR_PTR(err);
 855 }
 856
 857 static void sock_hash_free(struct bpf_map *map)
 858 {
 859         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 860         struct bpf_htab_bucket *bucket;
 861         struct bpf_htab_elem *elem;
 862         struct hlist_node *node;
 863         int i;
 864
 865         synchronize_rcu();
 866         for (i = 0; i < htab->buckets_num; i++) {
 867                 bucket = sock_hash_select_bucket(htab, i);
 868                 raw_spin_lock_bh(&bucket->lock);
 869                 hlist_for_each_entry_safe(elem, node, &bucket->head, node) {
 870                         hlist_del_rcu(&elem->node);
 871                         lock_sock(elem->sk);
 872                         rcu_read_lock();
 873                         sock_map_unref(elem->sk, elem);
 874                         rcu_read_unlock();
 875                         release_sock(elem->sk);
 876                 }
 877                 raw_spin_unlock_bh(&bucket->lock);
 878         }
 879
 880         bpf_map_area_free(htab->buckets);
 881         kfree(htab);
 882 }
 883
 884 static void sock_hash_release_progs(struct bpf_map *map)
 885 {
 886         psock_progs_drop(&container_of(map, struct bpf_htab, map)->progs);
 887 }
 888
 889 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
 890            struct bpf_map *, map, void *, key, u64, flags)
 891 {
 892         WARN_ON_ONCE(!rcu_read_lock_held());
 893
 894         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 895                    sock_map_op_okay(sops)))
 896                 return sock_hash_update_common(map, key, sops->sk, flags);
 897         return -EOPNOTSUPP;
 898 }
 899
 900 const struct bpf_func_proto bpf_sock_hash_update_proto = {
 901         .func           = bpf_sock_hash_update,
 902         .gpl_only       = false,
 903         .pkt_access     = true,
 904         .ret_type       = RET_INTEGER,
 905         .arg1_type      = ARG_PTR_TO_CTX,
 906         .arg2_type      = ARG_CONST_MAP_PTR,
 907         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 908         .arg4_type      = ARG_ANYTHING,
 909 };
 910
 911 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
 912            struct bpf_map *, map, void *, key, u64, flags)
 913 {
 914         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 915
 916         if (unlikely(flags & ~(BPF_F_INGRESS)))
 917                 return SK_DROP;
 918         tcb->bpf.flags = flags;
 919         tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
 920         if (!tcb->bpf.sk_redir)
 921                 return SK_DROP;
 922         return SK_PASS;
 923 }
 924
 925 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
 926         .func           = bpf_sk_redirect_hash,
 927         .gpl_only       = false,
 928         .ret_type       = RET_INTEGER,
 929         .arg1_type      = ARG_PTR_TO_CTX,
 930         .arg2_type      = ARG_CONST_MAP_PTR,
 931         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 932         .arg4_type      = ARG_ANYTHING,
 933 };
 934
 935 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
 936            struct bpf_map *, map, void *, key, u64, flags)
 937 {
 938         if (unlikely(flags & ~(BPF_F_INGRESS)))
 939                 return SK_DROP;
 940         msg->flags = flags;
 941         msg->sk_redir = __sock_hash_lookup_elem(map, key);
 942         if (!msg->sk_redir)
 943                 return SK_DROP;
 944         return SK_PASS;
 945 }
 946
 947 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
 948         .func           = bpf_msg_redirect_hash,
 949         .gpl_only       = false,
 950         .ret_type       = RET_INTEGER,
 951         .arg1_type      = ARG_PTR_TO_CTX,
 952         .arg2_type      = ARG_CONST_MAP_PTR,
 953         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 954         .arg4_type      = ARG_ANYTHING,
 955 };
 956
 957 const struct bpf_map_ops sock_hash_ops = {
 958         .map_alloc              = sock_hash_alloc,
 959         .map_free               = sock_hash_free,
 960         .map_get_next_key       = sock_hash_get_next_key,
 961         .map_update_elem        = sock_hash_update_elem,
 962         .map_delete_elem        = sock_hash_delete_elem,
 963         .map_lookup_elem        = sock_map_lookup,
 964         .map_release_uref       = sock_hash_release_progs,
 965         .map_check_btf          = map_check_no_btf,
 966 };
 967
 968 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
 969 {
 970         switch (map->map_type) {
 971         case BPF_MAP_TYPE_SOCKMAP:
 972                 return &container_of(map, struct bpf_stab, map)->progs;
 973         case BPF_MAP_TYPE_SOCKHASH:
 974                 return &container_of(map, struct bpf_htab, map)->progs;
 975         default:
 976                 break;
 977         }
 978
 979         return NULL;
 980 }
 981
 982 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
 983                          u32 which)
 984 {
 985         struct sk_psock_progs *progs = sock_map_progs(map);
 986
 987         if (!progs)
 988                 return -EOPNOTSUPP;
 989
 990         switch (which) {
 991         case BPF_SK_MSG_VERDICT:
 992                 psock_set_prog(&progs->msg_parser, prog);
 993                 break;
 994         case BPF_SK_SKB_STREAM_PARSER:
 995                 psock_set_prog(&progs->skb_parser, prog);
 996                 break;
 997         case BPF_SK_SKB_STREAM_VERDICT:
 998                 psock_set_prog(&progs->skb_verdict, prog);
 999                 break;
1000         default:
1001                 return -EOPNOTSUPP;
1002         }
1003
1004         return 0;
1005 }
1006
1007 void sk_psock_unlink(struct sock *sk, struct sk_psock_link *link)
1008 {
1009         switch (link->map->map_type) {
1010         case BPF_MAP_TYPE_SOCKMAP:
1011                 return sock_map_delete_from_link(link->map, sk,
1012                                                  link->link_raw);
1013         case BPF_MAP_TYPE_SOCKHASH:
1014                 return sock_hash_delete_from_link(link->map, sk,
1015                                                   link->link_raw);
1016         default:
1017                 break;
1018         }
1019 }