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/btf_ids.h>
   6 #include <linux/filter.h>
   7 #include <linux/errno.h>
   8 #include <linux/file.h>
   9 #include <linux/net.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/skmsg.h>
  12 #include <linux/list.h>
  13 #include <linux/jhash.h>
  14 #include <linux/sock_diag.h>
  15 #include <net/udp.h>
  16
  17 struct bpf_stab {
  18         struct bpf_map map;
  19         struct sock **sks;
  20         struct sk_psock_progs progs;
  21         raw_spinlock_t lock;
  22 };
  23
  24 #define SOCK_CREATE_FLAG_MASK                           \
  25         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
  26
  27 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  28 {
  29         struct bpf_stab *stab;
  30         u64 cost;
  31         int err;
  32
  33         if (!capable(CAP_NET_ADMIN))
  34                 return ERR_PTR(-EPERM);
  35         if (attr->max_entries == 0 ||
  36             attr->key_size    != 4 ||
  37             (attr->value_size != sizeof(u32) &&
  38              attr->value_size != sizeof(u64)) ||
  39             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  40                 return ERR_PTR(-EINVAL);
  41
  42         stab = kzalloc(sizeof(*stab), GFP_USER);
  43         if (!stab)
  44                 return ERR_PTR(-ENOMEM);
  45
  46         bpf_map_init_from_attr(&stab->map, attr);
  47         raw_spin_lock_init(&stab->lock);
  48
  49         /* Make sure page count doesn't overflow. */
  50         cost = (u64) stab->map.max_entries * sizeof(struct sock *);
  51         err = bpf_map_charge_init(&stab->map.memory, cost);
  52         if (err)
  53                 goto free_stab;
  54
  55         stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
  56                                        sizeof(struct sock *),
  57                                        stab->map.numa_node);
  58         if (stab->sks)
  59                 return &stab->map;
  60         err = -ENOMEM;
  61         bpf_map_charge_finish(&stab->map.memory);
  62 free_stab:
  63         kfree(stab);
  64         return ERR_PTR(err);
  65 }
  66
  67 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  68 {
  69         u32 ufd = attr->target_fd;
  70         struct bpf_map *map;
  71         struct fd f;
  72         int ret;
  73
  74         if (attr->attach_flags || attr->replace_bpf_fd)
  75                 return -EINVAL;
  76
  77         f = fdget(ufd);
  78         map = __bpf_map_get(f);
  79         if (IS_ERR(map))
  80                 return PTR_ERR(map);
  81         ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
  82         fdput(f);
  83         return ret;
  84 }
  85
  86 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  87 {
  88         u32 ufd = attr->target_fd;
  89         struct bpf_prog *prog;
  90         struct bpf_map *map;
  91         struct fd f;
  92         int ret;
  93
  94         if (attr->attach_flags || attr->replace_bpf_fd)
  95                 return -EINVAL;
  96
  97         f = fdget(ufd);
  98         map = __bpf_map_get(f);
  99         if (IS_ERR(map))
 100                 return PTR_ERR(map);
 101
 102         prog = bpf_prog_get(attr->attach_bpf_fd);
 103         if (IS_ERR(prog)) {
 104                 ret = PTR_ERR(prog);
 105                 goto put_map;
 106         }
 107
 108         if (prog->type != ptype) {
 109                 ret = -EINVAL;
 110                 goto put_prog;
 111         }
 112
 113         ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
 114 put_prog:
 115         bpf_prog_put(prog);
 116 put_map:
 117         fdput(f);
 118         return ret;
 119 }
 120
 121 static void sock_map_sk_acquire(struct sock *sk)
 122         __acquires(&sk->sk_lock.slock)
 123 {
 124         lock_sock(sk);
 125         rcu_read_lock();
 126 }
 127
 128 static void sock_map_sk_release(struct sock *sk)
 129         __releases(&sk->sk_lock.slock)
 130 {
 131         rcu_read_unlock();
 132         release_sock(sk);
 133 }
 134
 135 static void sock_map_add_link(struct sk_psock *psock,
 136                               struct sk_psock_link *link,
 137                               struct bpf_map *map, void *link_raw)
 138 {
 139         link->link_raw = link_raw;
 140         link->map = map;
 141         spin_lock_bh(&psock->link_lock);
 142         list_add_tail(&link->list, &psock->link);
 143         spin_unlock_bh(&psock->link_lock);
 144 }
 145
 146 static void sock_map_del_link(struct sock *sk,
 147                               struct sk_psock *psock, void *link_raw)
 148 {
 149         bool strp_stop = false, verdict_stop = false;
 150         struct sk_psock_link *link, *tmp;
 151
 152         spin_lock_bh(&psock->link_lock);
 153         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 154                 if (link->link_raw == link_raw) {
 155                         struct bpf_map *map = link->map;
 156                         struct bpf_stab *stab = container_of(map, struct bpf_stab,
 157                                                              map);
 158                         if (psock->parser.enabled && stab->progs.skb_parser)
 159                                 strp_stop = true;
 160                         if (psock->parser.enabled && stab->progs.skb_verdict)
 161                                 verdict_stop = true;
 162                         list_del(&link->list);
 163                         sk_psock_free_link(link);
 164                 }
 165         }
 166         spin_unlock_bh(&psock->link_lock);
 167         if (strp_stop || verdict_stop) {
 168                 write_lock_bh(&sk->sk_callback_lock);
 169                 if (strp_stop)
 170                         sk_psock_stop_strp(sk, psock);
 171                 else
 172                         sk_psock_stop_verdict(sk, psock);
 173                 write_unlock_bh(&sk->sk_callback_lock);
 174         }
 175 }
 176
 177 static void sock_map_unref(struct sock *sk, void *link_raw)
 178 {
 179         struct sk_psock *psock = sk_psock(sk);
 180
 181         if (likely(psock)) {
 182                 sock_map_del_link(sk, psock, link_raw);
 183                 sk_psock_put(sk, psock);
 184         }
 185 }
 186
 187 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
 188 {
 189         struct proto *prot;
 190
 191         switch (sk->sk_type) {
 192         case SOCK_STREAM:
 193                 prot = tcp_bpf_get_proto(sk, psock);
 194                 break;
 195
 196         case SOCK_DGRAM:
 197                 prot = udp_bpf_get_proto(sk, psock);
 198                 break;
 199
 200         default:
 201                 return -EINVAL;
 202         }
 203
 204         if (IS_ERR(prot))
 205                 return PTR_ERR(prot);
 206
 207         sk_psock_update_proto(sk, psock, prot);
 208         return 0;
 209 }
 210
 211 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 212 {
 213         struct sk_psock *psock;
 214
 215         rcu_read_lock();
 216         psock = sk_psock(sk);
 217         if (psock) {
 218                 if (sk->sk_prot->close != sock_map_close) {
 219                         psock = ERR_PTR(-EBUSY);
 220                         goto out;
 221                 }
 222
 223                 if (!refcount_inc_not_zero(&psock->refcnt))
 224                         psock = ERR_PTR(-EBUSY);
 225         }
 226 out:
 227         rcu_read_unlock();
 228         return psock;
 229 }
 230
 231 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
 232                          struct sock *sk)
 233 {
 234         struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
 235         struct sk_psock *psock;
 236         int ret;
 237
 238         skb_verdict = READ_ONCE(progs->skb_verdict);
 239         if (skb_verdict) {
 240                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 241                 if (IS_ERR(skb_verdict))
 242                         return PTR_ERR(skb_verdict);
 243         }
 244
 245         skb_parser = READ_ONCE(progs->skb_parser);
 246         if (skb_parser) {
 247                 skb_parser = bpf_prog_inc_not_zero(skb_parser);
 248                 if (IS_ERR(skb_parser)) {
 249                         ret = PTR_ERR(skb_parser);
 250                         goto out_put_skb_verdict;
 251                 }
 252         }
 253
 254         msg_parser = READ_ONCE(progs->msg_parser);
 255         if (msg_parser) {
 256                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 257                 if (IS_ERR(msg_parser)) {
 258                         ret = PTR_ERR(msg_parser);
 259                         goto out_put_skb_parser;
 260                 }
 261         }
 262
 263         psock = sock_map_psock_get_checked(sk);
 264         if (IS_ERR(psock)) {
 265                 ret = PTR_ERR(psock);
 266                 goto out_progs;
 267         }
 268
 269         if (psock) {
 270                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 271                     (skb_parser  && READ_ONCE(psock->progs.skb_parser)) ||
 272                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict))) {
 273                         sk_psock_put(sk, psock);
 274                         ret = -EBUSY;
 275                         goto out_progs;
 276                 }
 277         } else {
 278                 psock = sk_psock_init(sk, map->numa_node);
 279                 if (IS_ERR(psock)) {
 280                         ret = PTR_ERR(psock);
 281                         goto out_progs;
 282                 }
 283         }
 284
 285         if (msg_parser)
 286                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 287
 288         ret = sock_map_init_proto(sk, psock);
 289         if (ret < 0)
 290                 goto out_drop;
 291
 292         write_lock_bh(&sk->sk_callback_lock);
 293         if (skb_parser && skb_verdict && !psock->parser.enabled) {
 294                 ret = sk_psock_init_strp(sk, psock);
 295                 if (ret)
 296                         goto out_unlock_drop;
 297                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 298                 psock_set_prog(&psock->progs.skb_parser, skb_parser);
 299                 sk_psock_start_strp(sk, psock);
 300         } else if (!skb_parser && skb_verdict && !psock->parser.enabled) {
 301                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 302                 sk_psock_start_verdict(sk,psock);
 303         }
 304         write_unlock_bh(&sk->sk_callback_lock);
 305         return 0;
 306 out_unlock_drop:
 307         write_unlock_bh(&sk->sk_callback_lock);
 308 out_drop:
 309         sk_psock_put(sk, psock);
 310 out_progs:
 311         if (msg_parser)
 312                 bpf_prog_put(msg_parser);
 313 out_put_skb_parser:
 314         if (skb_parser)
 315                 bpf_prog_put(skb_parser);
 316 out_put_skb_verdict:
 317         if (skb_verdict)
 318                 bpf_prog_put(skb_verdict);
 319         return ret;
 320 }
 321
 322 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
 323 {
 324         struct sk_psock *psock;
 325         int ret;
 326
 327         psock = sock_map_psock_get_checked(sk);
 328         if (IS_ERR(psock))
 329                 return PTR_ERR(psock);
 330
 331         if (!psock) {
 332                 psock = sk_psock_init(sk, map->numa_node);
 333                 if (IS_ERR(psock))
 334                         return PTR_ERR(psock);
 335         }
 336
 337         ret = sock_map_init_proto(sk, psock);
 338         if (ret < 0)
 339                 sk_psock_put(sk, psock);
 340         return ret;
 341 }
 342
 343 static void sock_map_free(struct bpf_map *map)
 344 {
 345         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 346         int i;
 347
 348         /* After the sync no updates or deletes will be in-flight so it
 349          * is safe to walk map and remove entries without risking a race
 350          * in EEXIST update case.
 351          */
 352         synchronize_rcu();
 353         for (i = 0; i < stab->map.max_entries; i++) {
 354                 struct sock **psk = &stab->sks[i];
 355                 struct sock *sk;
 356
 357                 sk = xchg(psk, NULL);
 358                 if (sk) {
 359                         sock_hold(sk);
 360                         lock_sock(sk);
 361                         rcu_read_lock();
 362                         sock_map_unref(sk, psk);
 363                         rcu_read_unlock();
 364                         release_sock(sk);
 365                         sock_put(sk);
 366                 }
 367         }
 368
 369         /* wait for psock readers accessing its map link */
 370         synchronize_rcu();
 371
 372         bpf_map_area_free(stab->sks);
 373         kfree(stab);
 374 }
 375
 376 static void sock_map_release_progs(struct bpf_map *map)
 377 {
 378         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 379 }
 380
 381 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 382 {
 383         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 384
 385         WARN_ON_ONCE(!rcu_read_lock_held());
 386
 387         if (unlikely(key >= map->max_entries))
 388                 return NULL;
 389         return READ_ONCE(stab->sks[key]);
 390 }
 391
 392 static void *sock_map_lookup(struct bpf_map *map, void *key)
 393 {
 394         struct sock *sk;
 395
 396         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 397         if (!sk)
 398                 return NULL;
 399         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 400                 return NULL;
 401         return sk;
 402 }
 403
 404 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 405 {
 406         struct sock *sk;
 407
 408         if (map->value_size != sizeof(u64))
 409                 return ERR_PTR(-ENOSPC);
 410
 411         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 412         if (!sk)
 413                 return ERR_PTR(-ENOENT);
 414
 415         __sock_gen_cookie(sk);
 416         return &sk->sk_cookie;
 417 }
 418
 419 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 420                              struct sock **psk)
 421 {
 422         struct sock *sk;
 423         int err = 0;
 424
 425         if (irqs_disabled())
 426                 return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
 427
 428         raw_spin_lock_bh(&stab->lock);
 429         sk = *psk;
 430         if (!sk_test || sk_test == sk)
 431                 sk = xchg(psk, NULL);
 432
 433         if (likely(sk))
 434                 sock_map_unref(sk, psk);
 435         else
 436                 err = -EINVAL;
 437
 438         raw_spin_unlock_bh(&stab->lock);
 439         return err;
 440 }
 441
 442 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 443                                       void *link_raw)
 444 {
 445         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 446
 447         __sock_map_delete(stab, sk, link_raw);
 448 }
 449
 450 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 451 {
 452         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 453         u32 i = *(u32 *)key;
 454         struct sock **psk;
 455
 456         if (unlikely(i >= map->max_entries))
 457                 return -EINVAL;
 458
 459         psk = &stab->sks[i];
 460         return __sock_map_delete(stab, NULL, psk);
 461 }
 462
 463 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 464 {
 465         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 466         u32 i = key ? *(u32 *)key : U32_MAX;
 467         u32 *key_next = next;
 468
 469         if (i == stab->map.max_entries - 1)
 470                 return -ENOENT;
 471         if (i >= stab->map.max_entries)
 472                 *key_next = 0;
 473         else
 474                 *key_next = i + 1;
 475         return 0;
 476 }
 477
 478 static bool sock_map_redirect_allowed(const struct sock *sk);
 479
 480 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 481                                   struct sock *sk, u64 flags)
 482 {
 483         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 484         struct sk_psock_link *link;
 485         struct sk_psock *psock;
 486         struct sock *osk;
 487         int ret;
 488
 489         WARN_ON_ONCE(!rcu_read_lock_held());
 490         if (unlikely(flags > BPF_EXIST))
 491                 return -EINVAL;
 492         if (unlikely(idx >= map->max_entries))
 493                 return -E2BIG;
 494
 495         link = sk_psock_init_link();
 496         if (!link)
 497                 return -ENOMEM;
 498
 499         /* Only sockets we can redirect into/from in BPF need to hold
 500          * refs to parser/verdict progs and have their sk_data_ready
 501          * and sk_write_space callbacks overridden.
 502          */
 503         if (sock_map_redirect_allowed(sk))
 504                 ret = sock_map_link(map, &stab->progs, sk);
 505         else
 506                 ret = sock_map_link_no_progs(map, sk);
 507         if (ret < 0)
 508                 goto out_free;
 509
 510         psock = sk_psock(sk);
 511         WARN_ON_ONCE(!psock);
 512
 513         raw_spin_lock_bh(&stab->lock);
 514         osk = stab->sks[idx];
 515         if (osk && flags == BPF_NOEXIST) {
 516                 ret = -EEXIST;
 517                 goto out_unlock;
 518         } else if (!osk && flags == BPF_EXIST) {
 519                 ret = -ENOENT;
 520                 goto out_unlock;
 521         }
 522
 523         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 524         stab->sks[idx] = sk;
 525         if (osk)
 526                 sock_map_unref(osk, &stab->sks[idx]);
 527         raw_spin_unlock_bh(&stab->lock);
 528         return 0;
 529 out_unlock:
 530         raw_spin_unlock_bh(&stab->lock);
 531         if (psock)
 532                 sk_psock_put(sk, psock);
 533 out_free:
 534         sk_psock_free_link(link);
 535         return ret;
 536 }
 537
 538 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 539 {
 540         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 541                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 542                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 543 }
 544
 545 static bool sk_is_tcp(const struct sock *sk)
 546 {
 547         return sk->sk_type == SOCK_STREAM &&
 548                sk->sk_protocol == IPPROTO_TCP;
 549 }
 550
 551 static bool sk_is_udp(const struct sock *sk)
 552 {
 553         return sk->sk_type == SOCK_DGRAM &&
 554                sk->sk_protocol == IPPROTO_UDP;
 555 }
 556
 557 static bool sock_map_redirect_allowed(const struct sock *sk)
 558 {
 559         return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
 560 }
 561
 562 static bool sock_map_sk_is_suitable(const struct sock *sk)
 563 {
 564         return sk_is_tcp(sk) || sk_is_udp(sk);
 565 }
 566
 567 static bool sock_map_sk_state_allowed(const struct sock *sk)
 568 {
 569         if (sk_is_tcp(sk))
 570                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 571         else if (sk_is_udp(sk))
 572                 return sk_hashed(sk);
 573
 574         return false;
 575 }
 576
 577 static int sock_hash_update_common(struct bpf_map *map, void *key,
 578                                    struct sock *sk, u64 flags);
 579
 580 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 581                              u64 flags)
 582 {
 583         struct socket *sock;
 584         struct sock *sk;
 585         int ret;
 586         u64 ufd;
 587
 588         if (map->value_size == sizeof(u64))
 589                 ufd = *(u64 *)value;
 590         else
 591                 ufd = *(u32 *)value;
 592         if (ufd > S32_MAX)
 593                 return -EINVAL;
 594
 595         sock = sockfd_lookup(ufd, &ret);
 596         if (!sock)
 597                 return ret;
 598         sk = sock->sk;
 599         if (!sk) {
 600                 ret = -EINVAL;
 601                 goto out;
 602         }
 603         if (!sock_map_sk_is_suitable(sk)) {
 604                 ret = -EOPNOTSUPP;
 605                 goto out;
 606         }
 607
 608         sock_map_sk_acquire(sk);
 609         if (!sock_map_sk_state_allowed(sk))
 610                 ret = -EOPNOTSUPP;
 611         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 612                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 613         else
 614                 ret = sock_hash_update_common(map, key, sk, flags);
 615         sock_map_sk_release(sk);
 616 out:
 617         fput(sock->file);
 618         return ret;
 619 }
 620
 621 static int sock_map_update_elem(struct bpf_map *map, void *key,
 622                                 void *value, u64 flags)
 623 {
 624         struct sock *sk = (struct sock *)value;
 625         int ret;
 626
 627         if (unlikely(!sk || !sk_fullsock(sk)))
 628                 return -EINVAL;
 629
 630         if (!sock_map_sk_is_suitable(sk))
 631                 return -EOPNOTSUPP;
 632
 633         local_bh_disable();
 634         bh_lock_sock(sk);
 635         if (!sock_map_sk_state_allowed(sk))
 636                 ret = -EOPNOTSUPP;
 637         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 638                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 639         else
 640                 ret = sock_hash_update_common(map, key, sk, flags);
 641         bh_unlock_sock(sk);
 642         local_bh_enable();
 643         return ret;
 644 }
 645
 646 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 647            struct bpf_map *, map, void *, key, u64, flags)
 648 {
 649         WARN_ON_ONCE(!rcu_read_lock_held());
 650
 651         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 652                    sock_map_op_okay(sops)))
 653                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 654                                               flags);
 655         return -EOPNOTSUPP;
 656 }
 657
 658 const struct bpf_func_proto bpf_sock_map_update_proto = {
 659         .func           = bpf_sock_map_update,
 660         .gpl_only       = false,
 661         .pkt_access     = true,
 662         .ret_type       = RET_INTEGER,
 663         .arg1_type      = ARG_PTR_TO_CTX,
 664         .arg2_type      = ARG_CONST_MAP_PTR,
 665         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 666         .arg4_type      = ARG_ANYTHING,
 667 };
 668
 669 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 670            struct bpf_map *, map, u32, key, u64, flags)
 671 {
 672         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 673         struct sock *sk;
 674
 675         if (unlikely(flags & ~(BPF_F_INGRESS)))
 676                 return SK_DROP;
 677
 678         sk = __sock_map_lookup_elem(map, key);
 679         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 680                 return SK_DROP;
 681
 682         tcb->bpf.flags = flags;
 683         tcb->bpf.sk_redir = sk;
 684         return SK_PASS;
 685 }
 686
 687 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 688         .func           = bpf_sk_redirect_map,
 689         .gpl_only       = false,
 690         .ret_type       = RET_INTEGER,
 691         .arg1_type      = ARG_PTR_TO_CTX,
 692         .arg2_type      = ARG_CONST_MAP_PTR,
 693         .arg3_type      = ARG_ANYTHING,
 694         .arg4_type      = ARG_ANYTHING,
 695 };
 696
 697 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 698            struct bpf_map *, map, u32, key, u64, flags)
 699 {
 700         struct sock *sk;
 701
 702         if (unlikely(flags & ~(BPF_F_INGRESS)))
 703                 return SK_DROP;
 704
 705         sk = __sock_map_lookup_elem(map, key);
 706         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 707                 return SK_DROP;
 708
 709         msg->flags = flags;
 710         msg->sk_redir = sk;
 711         return SK_PASS;
 712 }
 713
 714 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 715         .func           = bpf_msg_redirect_map,
 716         .gpl_only       = false,
 717         .ret_type       = RET_INTEGER,
 718         .arg1_type      = ARG_PTR_TO_CTX,
 719         .arg2_type      = ARG_CONST_MAP_PTR,
 720         .arg3_type      = ARG_ANYTHING,
 721         .arg4_type      = ARG_ANYTHING,
 722 };
 723
 724 struct sock_map_seq_info {
 725         struct bpf_map *map;
 726         struct sock *sk;
 727         u32 index;
 728 };
 729
 730 struct bpf_iter__sockmap {
 731         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 732         __bpf_md_ptr(struct bpf_map *, map);
 733         __bpf_md_ptr(void *, key);
 734         __bpf_md_ptr(struct sock *, sk);
 735 };
 736
 737 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 738                      struct bpf_map *map, void *key,
 739                      struct sock *sk)
 740
 741 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 742 {
 743         if (unlikely(info->index >= info->map->max_entries))
 744                 return NULL;
 745
 746         info->sk = __sock_map_lookup_elem(info->map, info->index);
 747
 748         /* can't return sk directly, since that might be NULL */
 749         return info;
 750 }
 751
 752 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 753         __acquires(rcu)
 754 {
 755         struct sock_map_seq_info *info = seq->private;
 756
 757         if (*pos == 0)
 758                 ++*pos;
 759
 760         /* pairs with sock_map_seq_stop */
 761         rcu_read_lock();
 762         return sock_map_seq_lookup_elem(info);
 763 }
 764
 765 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 766         __must_hold(rcu)
 767 {
 768         struct sock_map_seq_info *info = seq->private;
 769
 770         ++*pos;
 771         ++info->index;
 772
 773         return sock_map_seq_lookup_elem(info);
 774 }
 775
 776 static int sock_map_seq_show(struct seq_file *seq, void *v)
 777         __must_hold(rcu)
 778 {
 779         struct sock_map_seq_info *info = seq->private;
 780         struct bpf_iter__sockmap ctx = {};
 781         struct bpf_iter_meta meta;
 782         struct bpf_prog *prog;
 783
 784         meta.seq = seq;
 785         prog = bpf_iter_get_info(&meta, !v);
 786         if (!prog)
 787                 return 0;
 788
 789         ctx.meta = &meta;
 790         ctx.map = info->map;
 791         if (v) {
 792                 ctx.key = &info->index;
 793                 ctx.sk = info->sk;
 794         }
 795
 796         return bpf_iter_run_prog(prog, &ctx);
 797 }
 798
 799 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 800         __releases(rcu)
 801 {
 802         if (!v)
 803                 (void)sock_map_seq_show(seq, NULL);
 804
 805         /* pairs with sock_map_seq_start */
 806         rcu_read_unlock();
 807 }
 808
 809 static const struct seq_operations sock_map_seq_ops = {
 810         .start  = sock_map_seq_start,
 811         .next   = sock_map_seq_next,
 812         .stop   = sock_map_seq_stop,
 813         .show   = sock_map_seq_show,
 814 };
 815
 816 static int sock_map_init_seq_private(void *priv_data,
 817                                      struct bpf_iter_aux_info *aux)
 818 {
 819         struct sock_map_seq_info *info = priv_data;
 820
 821         bpf_map_inc_with_uref(aux->map);
 822         info->map = aux->map;
 823         return 0;
 824 }
 825
 826 static void sock_map_fini_seq_private(void *priv_data)
 827 {
 828         struct sock_map_seq_info *info = priv_data;
 829
 830         bpf_map_put_with_uref(info->map);
 831 }
 832
 833 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 834         .seq_ops                = &sock_map_seq_ops,
 835         .init_seq_private       = sock_map_init_seq_private,
 836         .fini_seq_private       = sock_map_fini_seq_private,
 837         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 838 };
 839
 840 static int sock_map_btf_id;
 841 const struct bpf_map_ops sock_map_ops = {
 842         .map_meta_equal         = bpf_map_meta_equal,
 843         .map_alloc              = sock_map_alloc,
 844         .map_free               = sock_map_free,
 845         .map_get_next_key       = sock_map_get_next_key,
 846         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 847         .map_update_elem        = sock_map_update_elem,
 848         .map_delete_elem        = sock_map_delete_elem,
 849         .map_lookup_elem        = sock_map_lookup,
 850         .map_release_uref       = sock_map_release_progs,
 851         .map_check_btf          = map_check_no_btf,
 852         .map_btf_name           = "bpf_stab",
 853         .map_btf_id             = &sock_map_btf_id,
 854         .iter_seq_info          = &sock_map_iter_seq_info,
 855 };
 856
 857 struct bpf_shtab_elem {
 858         struct rcu_head rcu;
 859         u32 hash;
 860         struct sock *sk;
 861         struct hlist_node node;
 862         u8 key[];
 863 };
 864
 865 struct bpf_shtab_bucket {
 866         struct hlist_head head;
 867         raw_spinlock_t lock;
 868 };
 869
 870 struct bpf_shtab {
 871         struct bpf_map map;
 872         struct bpf_shtab_bucket *buckets;
 873         u32 buckets_num;
 874         u32 elem_size;
 875         struct sk_psock_progs progs;
 876         atomic_t count;
 877 };
 878
 879 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 880 {
 881         return jhash(key, len, 0);
 882 }
 883
 884 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 885                                                         u32 hash)
 886 {
 887         return &htab->buckets[hash & (htab->buckets_num - 1)];
 888 }
 889
 890 static struct bpf_shtab_elem *
 891 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 892                           u32 key_size)
 893 {
 894         struct bpf_shtab_elem *elem;
 895
 896         hlist_for_each_entry_rcu(elem, head, node) {
 897                 if (elem->hash == hash &&
 898                     !memcmp(&elem->key, key, key_size))
 899                         return elem;
 900         }
 901
 902         return NULL;
 903 }
 904
 905 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 906 {
 907         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 908         u32 key_size = map->key_size, hash;
 909         struct bpf_shtab_bucket *bucket;
 910         struct bpf_shtab_elem *elem;
 911
 912         WARN_ON_ONCE(!rcu_read_lock_held());
 913
 914         hash = sock_hash_bucket_hash(key, key_size);
 915         bucket = sock_hash_select_bucket(htab, hash);
 916         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 917
 918         return elem ? elem->sk : NULL;
 919 }
 920
 921 static void sock_hash_free_elem(struct bpf_shtab *htab,
 922                                 struct bpf_shtab_elem *elem)
 923 {
 924         atomic_dec(&htab->count);
 925         kfree_rcu(elem, rcu);
 926 }
 927
 928 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 929                                        void *link_raw)
 930 {
 931         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 932         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 933         struct bpf_shtab_bucket *bucket;
 934
 935         WARN_ON_ONCE(!rcu_read_lock_held());
 936         bucket = sock_hash_select_bucket(htab, elem->hash);
 937
 938         /* elem may be deleted in parallel from the map, but access here
 939          * is okay since it's going away only after RCU grace period.
 940          * However, we need to check whether it's still present.
 941          */
 942         raw_spin_lock_bh(&bucket->lock);
 943         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 944                                                elem->key, map->key_size);
 945         if (elem_probe && elem_probe == elem) {
 946                 hlist_del_rcu(&elem->node);
 947                 sock_map_unref(elem->sk, elem);
 948                 sock_hash_free_elem(htab, elem);
 949         }
 950         raw_spin_unlock_bh(&bucket->lock);
 951 }
 952
 953 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 954 {
 955         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 956         u32 hash, key_size = map->key_size;
 957         struct bpf_shtab_bucket *bucket;
 958         struct bpf_shtab_elem *elem;
 959         int ret = -ENOENT;
 960
 961         if (irqs_disabled())
 962                 return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
 963
 964         hash = sock_hash_bucket_hash(key, key_size);
 965         bucket = sock_hash_select_bucket(htab, hash);
 966
 967         raw_spin_lock_bh(&bucket->lock);
 968         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 969         if (elem) {
 970                 hlist_del_rcu(&elem->node);
 971                 sock_map_unref(elem->sk, elem);
 972                 sock_hash_free_elem(htab, elem);
 973                 ret = 0;
 974         }
 975         raw_spin_unlock_bh(&bucket->lock);
 976         return ret;
 977 }
 978
 979 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 980                                                    void *key, u32 key_size,
 981                                                    u32 hash, struct sock *sk,
 982                                                    struct bpf_shtab_elem *old)
 983 {
 984         struct bpf_shtab_elem *new;
 985
 986         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 987                 if (!old) {
 988                         atomic_dec(&htab->count);
 989                         return ERR_PTR(-E2BIG);
 990                 }
 991         }
 992
 993         new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
 994                            htab->map.numa_node);
 995         if (!new) {
 996                 atomic_dec(&htab->count);
 997                 return ERR_PTR(-ENOMEM);
 998         }
 999         memcpy(new->key, key, key_size);
1000         new->sk = sk;
1001         new->hash = hash;
1002         return new;
1003 }
1004
1005 static int sock_hash_update_common(struct bpf_map *map, void *key,
1006                                    struct sock *sk, u64 flags)
1007 {
1008         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1009         u32 key_size = map->key_size, hash;
1010         struct bpf_shtab_elem *elem, *elem_new;
1011         struct bpf_shtab_bucket *bucket;
1012         struct sk_psock_link *link;
1013         struct sk_psock *psock;
1014         int ret;
1015
1016         WARN_ON_ONCE(!rcu_read_lock_held());
1017         if (unlikely(flags > BPF_EXIST))
1018                 return -EINVAL;
1019
1020         link = sk_psock_init_link();
1021         if (!link)
1022                 return -ENOMEM;
1023
1024         /* Only sockets we can redirect into/from in BPF need to hold
1025          * refs to parser/verdict progs and have their sk_data_ready
1026          * and sk_write_space callbacks overridden.
1027          */
1028         if (sock_map_redirect_allowed(sk))
1029                 ret = sock_map_link(map, &htab->progs, sk);
1030         else
1031                 ret = sock_map_link_no_progs(map, sk);
1032         if (ret < 0)
1033                 goto out_free;
1034
1035         psock = sk_psock(sk);
1036         WARN_ON_ONCE(!psock);
1037
1038         hash = sock_hash_bucket_hash(key, key_size);
1039         bucket = sock_hash_select_bucket(htab, hash);
1040
1041         raw_spin_lock_bh(&bucket->lock);
1042         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1043         if (elem && flags == BPF_NOEXIST) {
1044                 ret = -EEXIST;
1045                 goto out_unlock;
1046         } else if (!elem && flags == BPF_EXIST) {
1047                 ret = -ENOENT;
1048                 goto out_unlock;
1049         }
1050
1051         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1052         if (IS_ERR(elem_new)) {
1053                 ret = PTR_ERR(elem_new);
1054                 goto out_unlock;
1055         }
1056
1057         sock_map_add_link(psock, link, map, elem_new);
1058         /* Add new element to the head of the list, so that
1059          * concurrent search will find it before old elem.
1060          */
1061         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1062         if (elem) {
1063                 hlist_del_rcu(&elem->node);
1064                 sock_map_unref(elem->sk, elem);
1065                 sock_hash_free_elem(htab, elem);
1066         }
1067         raw_spin_unlock_bh(&bucket->lock);
1068         return 0;
1069 out_unlock:
1070         raw_spin_unlock_bh(&bucket->lock);
1071         sk_psock_put(sk, psock);
1072 out_free:
1073         sk_psock_free_link(link);
1074         return ret;
1075 }
1076
1077 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1078                                   void *key_next)
1079 {
1080         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1081         struct bpf_shtab_elem *elem, *elem_next;
1082         u32 hash, key_size = map->key_size;
1083         struct hlist_head *head;
1084         int i = 0;
1085
1086         if (!key)
1087                 goto find_first_elem;
1088         hash = sock_hash_bucket_hash(key, key_size);
1089         head = &sock_hash_select_bucket(htab, hash)->head;
1090         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1091         if (!elem)
1092                 goto find_first_elem;
1093
1094         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1095                                      struct bpf_shtab_elem, node);
1096         if (elem_next) {
1097                 memcpy(key_next, elem_next->key, key_size);
1098                 return 0;
1099         }
1100
1101         i = hash & (htab->buckets_num - 1);
1102         i++;
1103 find_first_elem:
1104         for (; i < htab->buckets_num; i++) {
1105                 head = &sock_hash_select_bucket(htab, i)->head;
1106                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1107                                              struct bpf_shtab_elem, node);
1108                 if (elem_next) {
1109                         memcpy(key_next, elem_next->key, key_size);
1110                         return 0;
1111                 }
1112         }
1113
1114         return -ENOENT;
1115 }
1116
1117 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1118 {
1119         struct bpf_shtab *htab;
1120         int i, err;
1121         u64 cost;
1122
1123         if (!capable(CAP_NET_ADMIN))
1124                 return ERR_PTR(-EPERM);
1125         if (attr->max_entries == 0 ||
1126             attr->key_size    == 0 ||
1127             (attr->value_size != sizeof(u32) &&
1128              attr->value_size != sizeof(u64)) ||
1129             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1130                 return ERR_PTR(-EINVAL);
1131         if (attr->key_size > MAX_BPF_STACK)
1132                 return ERR_PTR(-E2BIG);
1133
1134         htab = kzalloc(sizeof(*htab), GFP_USER);
1135         if (!htab)
1136                 return ERR_PTR(-ENOMEM);
1137
1138         bpf_map_init_from_attr(&htab->map, attr);
1139
1140         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1141         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1142                           round_up(htab->map.key_size, 8);
1143         if (htab->buckets_num == 0 ||
1144             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1145                 err = -EINVAL;
1146                 goto free_htab;
1147         }
1148
1149         cost = (u64) htab->buckets_num * sizeof(struct bpf_shtab_bucket) +
1150                (u64) htab->elem_size * htab->map.max_entries;
1151         if (cost >= U32_MAX - PAGE_SIZE) {
1152                 err = -EINVAL;
1153                 goto free_htab;
1154         }
1155         err = bpf_map_charge_init(&htab->map.memory, cost);
1156         if (err)
1157                 goto free_htab;
1158
1159         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1160                                            sizeof(struct bpf_shtab_bucket),
1161                                            htab->map.numa_node);
1162         if (!htab->buckets) {
1163                 bpf_map_charge_finish(&htab->map.memory);
1164                 err = -ENOMEM;
1165                 goto free_htab;
1166         }
1167
1168         for (i = 0; i < htab->buckets_num; i++) {
1169                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1170                 raw_spin_lock_init(&htab->buckets[i].lock);
1171         }
1172
1173         return &htab->map;
1174 free_htab:
1175         kfree(htab);
1176         return ERR_PTR(err);
1177 }
1178
1179 static void sock_hash_free(struct bpf_map *map)
1180 {
1181         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1182         struct bpf_shtab_bucket *bucket;
1183         struct hlist_head unlink_list;
1184         struct bpf_shtab_elem *elem;
1185         struct hlist_node *node;
1186         int i;
1187
1188         /* After the sync no updates or deletes will be in-flight so it
1189          * is safe to walk map and remove entries without risking a race
1190          * in EEXIST update case.
1191          */
1192         synchronize_rcu();
1193         for (i = 0; i < htab->buckets_num; i++) {
1194                 bucket = sock_hash_select_bucket(htab, i);
1195
1196                 /* We are racing with sock_hash_delete_from_link to
1197                  * enter the spin-lock critical section. Every socket on
1198                  * the list is still linked to sockhash. Since link
1199                  * exists, psock exists and holds a ref to socket. That
1200                  * lets us to grab a socket ref too.
1201                  */
1202                 raw_spin_lock_bh(&bucket->lock);
1203                 hlist_for_each_entry(elem, &bucket->head, node)
1204                         sock_hold(elem->sk);
1205                 hlist_move_list(&bucket->head, &unlink_list);
1206                 raw_spin_unlock_bh(&bucket->lock);
1207
1208                 /* Process removed entries out of atomic context to
1209                  * block for socket lock before deleting the psock's
1210                  * link to sockhash.
1211                  */
1212                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1213                         hlist_del(&elem->node);
1214                         lock_sock(elem->sk);
1215                         rcu_read_lock();
1216                         sock_map_unref(elem->sk, elem);
1217                         rcu_read_unlock();
1218                         release_sock(elem->sk);
1219                         sock_put(elem->sk);
1220                         sock_hash_free_elem(htab, elem);
1221                 }
1222         }
1223
1224         /* wait for psock readers accessing its map link */
1225         synchronize_rcu();
1226
1227         bpf_map_area_free(htab->buckets);
1228         kfree(htab);
1229 }
1230
1231 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1232 {
1233         struct sock *sk;
1234
1235         if (map->value_size != sizeof(u64))
1236                 return ERR_PTR(-ENOSPC);
1237
1238         sk = __sock_hash_lookup_elem(map, key);
1239         if (!sk)
1240                 return ERR_PTR(-ENOENT);
1241
1242         __sock_gen_cookie(sk);
1243         return &sk->sk_cookie;
1244 }
1245
1246 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1247 {
1248         struct sock *sk;
1249
1250         sk = __sock_hash_lookup_elem(map, key);
1251         if (!sk)
1252                 return NULL;
1253         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1254                 return NULL;
1255         return sk;
1256 }
1257
1258 static void sock_hash_release_progs(struct bpf_map *map)
1259 {
1260         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1261 }
1262
1263 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1264            struct bpf_map *, map, void *, key, u64, flags)
1265 {
1266         WARN_ON_ONCE(!rcu_read_lock_held());
1267
1268         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1269                    sock_map_op_okay(sops)))
1270                 return sock_hash_update_common(map, key, sops->sk, flags);
1271         return -EOPNOTSUPP;
1272 }
1273
1274 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1275         .func           = bpf_sock_hash_update,
1276         .gpl_only       = false,
1277         .pkt_access     = true,
1278         .ret_type       = RET_INTEGER,
1279         .arg1_type      = ARG_PTR_TO_CTX,
1280         .arg2_type      = ARG_CONST_MAP_PTR,
1281         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1282         .arg4_type      = ARG_ANYTHING,
1283 };
1284
1285 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1286            struct bpf_map *, map, void *, key, u64, flags)
1287 {
1288         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1289         struct sock *sk;
1290
1291         if (unlikely(flags & ~(BPF_F_INGRESS)))
1292                 return SK_DROP;
1293
1294         sk = __sock_hash_lookup_elem(map, key);
1295         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1296                 return SK_DROP;
1297
1298         tcb->bpf.flags = flags;
1299         tcb->bpf.sk_redir = sk;
1300         return SK_PASS;
1301 }
1302
1303 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1304         .func           = bpf_sk_redirect_hash,
1305         .gpl_only       = false,
1306         .ret_type       = RET_INTEGER,
1307         .arg1_type      = ARG_PTR_TO_CTX,
1308         .arg2_type      = ARG_CONST_MAP_PTR,
1309         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1310         .arg4_type      = ARG_ANYTHING,
1311 };
1312
1313 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1314            struct bpf_map *, map, void *, key, u64, flags)
1315 {
1316         struct sock *sk;
1317
1318         if (unlikely(flags & ~(BPF_F_INGRESS)))
1319                 return SK_DROP;
1320
1321         sk = __sock_hash_lookup_elem(map, key);
1322         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1323                 return SK_DROP;
1324
1325         msg->flags = flags;
1326         msg->sk_redir = sk;
1327         return SK_PASS;
1328 }
1329
1330 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1331         .func           = bpf_msg_redirect_hash,
1332         .gpl_only       = false,
1333         .ret_type       = RET_INTEGER,
1334         .arg1_type      = ARG_PTR_TO_CTX,
1335         .arg2_type      = ARG_CONST_MAP_PTR,
1336         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1337         .arg4_type      = ARG_ANYTHING,
1338 };
1339
1340 struct sock_hash_seq_info {
1341         struct bpf_map *map;
1342         struct bpf_shtab *htab;
1343         u32 bucket_id;
1344 };
1345
1346 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1347                                      struct bpf_shtab_elem *prev_elem)
1348 {
1349         const struct bpf_shtab *htab = info->htab;
1350         struct bpf_shtab_bucket *bucket;
1351         struct bpf_shtab_elem *elem;
1352         struct hlist_node *node;
1353
1354         /* try to find next elem in the same bucket */
1355         if (prev_elem) {
1356                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1357                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1358                 if (elem)
1359                         return elem;
1360
1361                 /* no more elements, continue in the next bucket */
1362                 info->bucket_id++;
1363         }
1364
1365         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1366                 bucket = &htab->buckets[info->bucket_id];
1367                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1368                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1369                 if (elem)
1370                         return elem;
1371         }
1372
1373         return NULL;
1374 }
1375
1376 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1377         __acquires(rcu)
1378 {
1379         struct sock_hash_seq_info *info = seq->private;
1380
1381         if (*pos == 0)
1382                 ++*pos;
1383
1384         /* pairs with sock_hash_seq_stop */
1385         rcu_read_lock();
1386         return sock_hash_seq_find_next(info, NULL);
1387 }
1388
1389 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1390         __must_hold(rcu)
1391 {
1392         struct sock_hash_seq_info *info = seq->private;
1393
1394         ++*pos;
1395         return sock_hash_seq_find_next(info, v);
1396 }
1397
1398 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1399         __must_hold(rcu)
1400 {
1401         struct sock_hash_seq_info *info = seq->private;
1402         struct bpf_iter__sockmap ctx = {};
1403         struct bpf_shtab_elem *elem = v;
1404         struct bpf_iter_meta meta;
1405         struct bpf_prog *prog;
1406
1407         meta.seq = seq;
1408         prog = bpf_iter_get_info(&meta, !elem);
1409         if (!prog)
1410                 return 0;
1411
1412         ctx.meta = &meta;
1413         ctx.map = info->map;
1414         if (elem) {
1415                 ctx.key = elem->key;
1416                 ctx.sk = elem->sk;
1417         }
1418
1419         return bpf_iter_run_prog(prog, &ctx);
1420 }
1421
1422 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1423         __releases(rcu)
1424 {
1425         if (!v)
1426                 (void)sock_hash_seq_show(seq, NULL);
1427
1428         /* pairs with sock_hash_seq_start */
1429         rcu_read_unlock();
1430 }
1431
1432 static const struct seq_operations sock_hash_seq_ops = {
1433         .start  = sock_hash_seq_start,
1434         .next   = sock_hash_seq_next,
1435         .stop   = sock_hash_seq_stop,
1436         .show   = sock_hash_seq_show,
1437 };
1438
1439 static int sock_hash_init_seq_private(void *priv_data,
1440                                       struct bpf_iter_aux_info *aux)
1441 {
1442         struct sock_hash_seq_info *info = priv_data;
1443
1444         bpf_map_inc_with_uref(aux->map);
1445         info->map = aux->map;
1446         info->htab = container_of(aux->map, struct bpf_shtab, map);
1447         return 0;
1448 }
1449
1450 static void sock_hash_fini_seq_private(void *priv_data)
1451 {
1452         struct sock_hash_seq_info *info = priv_data;
1453
1454         bpf_map_put_with_uref(info->map);
1455 }
1456
1457 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1458         .seq_ops                = &sock_hash_seq_ops,
1459         .init_seq_private       = sock_hash_init_seq_private,
1460         .fini_seq_private       = sock_hash_fini_seq_private,
1461         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1462 };
1463
1464 static int sock_hash_map_btf_id;
1465 const struct bpf_map_ops sock_hash_ops = {
1466         .map_meta_equal         = bpf_map_meta_equal,
1467         .map_alloc              = sock_hash_alloc,
1468         .map_free               = sock_hash_free,
1469         .map_get_next_key       = sock_hash_get_next_key,
1470         .map_update_elem        = sock_map_update_elem,
1471         .map_delete_elem        = sock_hash_delete_elem,
1472         .map_lookup_elem        = sock_hash_lookup,
1473         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1474         .map_release_uref       = sock_hash_release_progs,
1475         .map_check_btf          = map_check_no_btf,
1476         .map_btf_name           = "bpf_shtab",
1477         .map_btf_id             = &sock_hash_map_btf_id,
1478         .iter_seq_info          = &sock_hash_iter_seq_info,
1479 };
1480
1481 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1482 {
1483         switch (map->map_type) {
1484         case BPF_MAP_TYPE_SOCKMAP:
1485                 return &container_of(map, struct bpf_stab, map)->progs;
1486         case BPF_MAP_TYPE_SOCKHASH:
1487                 return &container_of(map, struct bpf_shtab, map)->progs;
1488         default:
1489                 break;
1490         }
1491
1492         return NULL;
1493 }
1494
1495 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1496                          struct bpf_prog *old, u32 which)
1497 {
1498         struct sk_psock_progs *progs = sock_map_progs(map);
1499         struct bpf_prog **pprog;
1500
1501         if (!progs)
1502                 return -EOPNOTSUPP;
1503
1504         switch (which) {
1505         case BPF_SK_MSG_VERDICT:
1506                 pprog = &progs->msg_parser;
1507                 break;
1508         case BPF_SK_SKB_STREAM_PARSER:
1509                 pprog = &progs->skb_parser;
1510                 break;
1511         case BPF_SK_SKB_STREAM_VERDICT:
1512                 pprog = &progs->skb_verdict;
1513                 break;
1514         default:
1515                 return -EOPNOTSUPP;
1516         }
1517
1518         if (old)
1519                 return psock_replace_prog(pprog, prog, old);
1520
1521         psock_set_prog(pprog, prog);
1522         return 0;
1523 }
1524
1525 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1526 {
1527         switch (link->map->map_type) {
1528         case BPF_MAP_TYPE_SOCKMAP:
1529                 return sock_map_delete_from_link(link->map, sk,
1530                                                  link->link_raw);
1531         case BPF_MAP_TYPE_SOCKHASH:
1532                 return sock_hash_delete_from_link(link->map, sk,
1533                                                   link->link_raw);
1534         default:
1535                 break;
1536         }
1537 }
1538
1539 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1540 {
1541         struct sk_psock_link *link;
1542
1543         while ((link = sk_psock_link_pop(psock))) {
1544                 sock_map_unlink(sk, link);
1545                 sk_psock_free_link(link);
1546         }
1547 }
1548
1549 void sock_map_unhash(struct sock *sk)
1550 {
1551         void (*saved_unhash)(struct sock *sk);
1552         struct sk_psock *psock;
1553
1554         rcu_read_lock();
1555         psock = sk_psock(sk);
1556         if (unlikely(!psock)) {
1557                 rcu_read_unlock();
1558                 if (sk->sk_prot->unhash)
1559                         sk->sk_prot->unhash(sk);
1560                 return;
1561         }
1562
1563         saved_unhash = psock->saved_unhash;
1564         sock_map_remove_links(sk, psock);
1565         rcu_read_unlock();
1566         saved_unhash(sk);
1567 }
1568
1569 void sock_map_close(struct sock *sk, long timeout)
1570 {
1571         void (*saved_close)(struct sock *sk, long timeout);
1572         struct sk_psock *psock;
1573
1574         lock_sock(sk);
1575         rcu_read_lock();
1576         psock = sk_psock(sk);
1577         if (unlikely(!psock)) {
1578                 rcu_read_unlock();
1579                 release_sock(sk);
1580                 return sk->sk_prot->close(sk, timeout);
1581         }
1582
1583         saved_close = psock->saved_close;
1584         sock_map_remove_links(sk, psock);
1585         rcu_read_unlock();
1586         release_sock(sk);
1587         saved_close(sk, timeout);
1588 }
1589
1590 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1591                                        union bpf_iter_link_info *linfo,
1592                                        struct bpf_iter_aux_info *aux)
1593 {
1594         struct bpf_map *map;
1595         int err = -EINVAL;
1596
1597         if (!linfo->map.map_fd)
1598                 return -EBADF;
1599
1600         map = bpf_map_get_with_uref(linfo->map.map_fd);
1601         if (IS_ERR(map))
1602                 return PTR_ERR(map);
1603
1604         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1605             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1606                 goto put_map;
1607
1608         if (prog->aux->max_rdonly_access > map->key_size) {
1609                 err = -EACCES;
1610                 goto put_map;
1611         }
1612
1613         aux->map = map;
1614         return 0;
1615
1616 put_map:
1617         bpf_map_put_with_uref(map);
1618         return err;
1619 }
1620
1621 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1622 {
1623         bpf_map_put_with_uref(aux->map);
1624 }
1625
1626 static struct bpf_iter_reg sock_map_iter_reg = {
1627         .target                 = "sockmap",
1628         .attach_target          = sock_map_iter_attach_target,
1629         .detach_target          = sock_map_iter_detach_target,
1630         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1631         .fill_link_info         = bpf_iter_map_fill_link_info,
1632         .ctx_arg_info_size      = 2,
1633         .ctx_arg_info           = {
1634                 { offsetof(struct bpf_iter__sockmap, key),
1635                   PTR_TO_RDONLY_BUF_OR_NULL },
1636                 { offsetof(struct bpf_iter__sockmap, sk),
1637                   PTR_TO_BTF_ID_OR_NULL },
1638         },
1639 };
1640
1641 static int __init bpf_sockmap_iter_init(void)
1642 {
1643         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1644                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1645         return bpf_iter_reg_target(&sock_map_iter_reg);
1646 }
1647 late_initcall(bpf_sockmap_iter_init);