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 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
  28                                 struct bpf_prog *old, u32 which);
  29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
  30
  31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  32 {
  33         struct bpf_stab *stab;
  34
  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 = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
  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         stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
  50                                        sizeof(struct sock *),
  51                                        stab->map.numa_node);
  52         if (!stab->sks) {
  53                 bpf_map_area_free(stab);
  54                 return ERR_PTR(-ENOMEM);
  55         }
  56
  57         return &stab->map;
  58 }
  59
  60 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  61 {
  62         u32 ufd = attr->target_fd;
  63         struct bpf_map *map;
  64         struct fd f;
  65         int ret;
  66
  67         if (attr->attach_flags || attr->replace_bpf_fd)
  68                 return -EINVAL;
  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, NULL, attr->attach_type);
  75         fdput(f);
  76         return ret;
  77 }
  78
  79 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  80 {
  81         u32 ufd = attr->target_fd;
  82         struct bpf_prog *prog;
  83         struct bpf_map *map;
  84         struct fd f;
  85         int ret;
  86
  87         if (attr->attach_flags || attr->replace_bpf_fd)
  88                 return -EINVAL;
  89
  90         f = fdget(ufd);
  91         map = __bpf_map_get(f);
  92         if (IS_ERR(map))
  93                 return PTR_ERR(map);
  94
  95         prog = bpf_prog_get(attr->attach_bpf_fd);
  96         if (IS_ERR(prog)) {
  97                 ret = PTR_ERR(prog);
  98                 goto put_map;
  99         }
 100
 101         if (prog->type != ptype) {
 102                 ret = -EINVAL;
 103                 goto put_prog;
 104         }
 105
 106         ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
 107 put_prog:
 108         bpf_prog_put(prog);
 109 put_map:
 110         fdput(f);
 111         return ret;
 112 }
 113
 114 static void sock_map_sk_acquire(struct sock *sk)
 115         __acquires(&sk->sk_lock.slock)
 116 {
 117         lock_sock(sk);
 118         preempt_disable();
 119         rcu_read_lock();
 120 }
 121
 122 static void sock_map_sk_release(struct sock *sk)
 123         __releases(&sk->sk_lock.slock)
 124 {
 125         rcu_read_unlock();
 126         preempt_enable();
 127         release_sock(sk);
 128 }
 129
 130 static void sock_map_add_link(struct sk_psock *psock,
 131                               struct sk_psock_link *link,
 132                               struct bpf_map *map, void *link_raw)
 133 {
 134         link->link_raw = link_raw;
 135         link->map = map;
 136         spin_lock_bh(&psock->link_lock);
 137         list_add_tail(&link->list, &psock->link);
 138         spin_unlock_bh(&psock->link_lock);
 139 }
 140
 141 static void sock_map_del_link(struct sock *sk,
 142                               struct sk_psock *psock, void *link_raw)
 143 {
 144         bool strp_stop = false, verdict_stop = false;
 145         struct sk_psock_link *link, *tmp;
 146
 147         spin_lock_bh(&psock->link_lock);
 148         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 149                 if (link->link_raw == link_raw) {
 150                         struct bpf_map *map = link->map;
 151                         struct bpf_stab *stab = container_of(map, struct bpf_stab,
 152                                                              map);
 153                         if (psock->saved_data_ready && stab->progs.stream_parser)
 154                                 strp_stop = true;
 155                         if (psock->saved_data_ready && stab->progs.stream_verdict)
 156                                 verdict_stop = true;
 157                         if (psock->saved_data_ready && stab->progs.skb_verdict)
 158                                 verdict_stop = true;
 159                         list_del(&link->list);
 160                         sk_psock_free_link(link);
 161                 }
 162         }
 163         spin_unlock_bh(&psock->link_lock);
 164         if (strp_stop || verdict_stop) {
 165                 write_lock_bh(&sk->sk_callback_lock);
 166                 if (strp_stop)
 167                         sk_psock_stop_strp(sk, psock);
 168                 if (verdict_stop)
 169                         sk_psock_stop_verdict(sk, psock);
 170
 171                 if (psock->psock_update_sk_prot)
 172                         psock->psock_update_sk_prot(sk, psock, false);
 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         if (!sk->sk_prot->psock_update_sk_prot)
 190                 return -EINVAL;
 191         psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
 192         return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
 193 }
 194
 195 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 196 {
 197         struct sk_psock *psock;
 198
 199         rcu_read_lock();
 200         psock = sk_psock(sk);
 201         if (psock) {
 202                 if (sk->sk_prot->close != sock_map_close) {
 203                         psock = ERR_PTR(-EBUSY);
 204                         goto out;
 205                 }
 206
 207                 if (!refcount_inc_not_zero(&psock->refcnt))
 208                         psock = ERR_PTR(-EBUSY);
 209         }
 210 out:
 211         rcu_read_unlock();
 212         return psock;
 213 }
 214
 215 static int sock_map_link(struct bpf_map *map, struct sock *sk)
 216 {
 217         struct sk_psock_progs *progs = sock_map_progs(map);
 218         struct bpf_prog *stream_verdict = NULL;
 219         struct bpf_prog *stream_parser = NULL;
 220         struct bpf_prog *skb_verdict = NULL;
 221         struct bpf_prog *msg_parser = NULL;
 222         struct sk_psock *psock;
 223         int ret;
 224
 225         stream_verdict = READ_ONCE(progs->stream_verdict);
 226         if (stream_verdict) {
 227                 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
 228                 if (IS_ERR(stream_verdict))
 229                         return PTR_ERR(stream_verdict);
 230         }
 231
 232         stream_parser = READ_ONCE(progs->stream_parser);
 233         if (stream_parser) {
 234                 stream_parser = bpf_prog_inc_not_zero(stream_parser);
 235                 if (IS_ERR(stream_parser)) {
 236                         ret = PTR_ERR(stream_parser);
 237                         goto out_put_stream_verdict;
 238                 }
 239         }
 240
 241         msg_parser = READ_ONCE(progs->msg_parser);
 242         if (msg_parser) {
 243                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 244                 if (IS_ERR(msg_parser)) {
 245                         ret = PTR_ERR(msg_parser);
 246                         goto out_put_stream_parser;
 247                 }
 248         }
 249
 250         skb_verdict = READ_ONCE(progs->skb_verdict);
 251         if (skb_verdict) {
 252                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 253                 if (IS_ERR(skb_verdict)) {
 254                         ret = PTR_ERR(skb_verdict);
 255                         goto out_put_msg_parser;
 256                 }
 257         }
 258
 259         psock = sock_map_psock_get_checked(sk);
 260         if (IS_ERR(psock)) {
 261                 ret = PTR_ERR(psock);
 262                 goto out_progs;
 263         }
 264
 265         if (psock) {
 266                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 267                     (stream_parser  && READ_ONCE(psock->progs.stream_parser)) ||
 268                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 269                     (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
 270                     (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 271                     (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
 272                         sk_psock_put(sk, psock);
 273                         ret = -EBUSY;
 274                         goto out_progs;
 275                 }
 276         } else {
 277                 psock = sk_psock_init(sk, map->numa_node);
 278                 if (IS_ERR(psock)) {
 279                         ret = PTR_ERR(psock);
 280                         goto out_progs;
 281                 }
 282         }
 283
 284         if (msg_parser)
 285                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 286         if (stream_parser)
 287                 psock_set_prog(&psock->progs.stream_parser, stream_parser);
 288         if (stream_verdict)
 289                 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
 290         if (skb_verdict)
 291                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 292
 293         /* msg_* and stream_* programs references tracked in psock after this
 294          * point. Reference dec and cleanup will occur through psock destructor
 295          */
 296         ret = sock_map_init_proto(sk, psock);
 297         if (ret < 0) {
 298                 sk_psock_put(sk, psock);
 299                 goto out;
 300         }
 301
 302         write_lock_bh(&sk->sk_callback_lock);
 303         if (stream_parser && stream_verdict && !psock->saved_data_ready) {
 304                 ret = sk_psock_init_strp(sk, psock);
 305                 if (ret) {
 306                         write_unlock_bh(&sk->sk_callback_lock);
 307                         sk_psock_put(sk, psock);
 308                         goto out;
 309                 }
 310                 sk_psock_start_strp(sk, psock);
 311         } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
 312                 sk_psock_start_verdict(sk,psock);
 313         } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
 314                 sk_psock_start_verdict(sk, psock);
 315         }
 316         write_unlock_bh(&sk->sk_callback_lock);
 317         return 0;
 318 out_progs:
 319         if (skb_verdict)
 320                 bpf_prog_put(skb_verdict);
 321 out_put_msg_parser:
 322         if (msg_parser)
 323                 bpf_prog_put(msg_parser);
 324 out_put_stream_parser:
 325         if (stream_parser)
 326                 bpf_prog_put(stream_parser);
 327 out_put_stream_verdict:
 328         if (stream_verdict)
 329                 bpf_prog_put(stream_verdict);
 330 out:
 331         return ret;
 332 }
 333
 334 static void sock_map_free(struct bpf_map *map)
 335 {
 336         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 337         int i;
 338
 339         /* After the sync no updates or deletes will be in-flight so it
 340          * is safe to walk map and remove entries without risking a race
 341          * in EEXIST update case.
 342          */
 343         synchronize_rcu();
 344         for (i = 0; i < stab->map.max_entries; i++) {
 345                 struct sock **psk = &stab->sks[i];
 346                 struct sock *sk;
 347
 348                 sk = xchg(psk, NULL);
 349                 if (sk) {
 350                         sock_hold(sk);
 351                         lock_sock(sk);
 352                         rcu_read_lock();
 353                         sock_map_unref(sk, psk);
 354                         rcu_read_unlock();
 355                         release_sock(sk);
 356                         sock_put(sk);
 357                 }
 358         }
 359
 360         /* wait for psock readers accessing its map link */
 361         synchronize_rcu();
 362
 363         bpf_map_area_free(stab->sks);
 364         bpf_map_area_free(stab);
 365 }
 366
 367 static void sock_map_release_progs(struct bpf_map *map)
 368 {
 369         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 370 }
 371
 372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 373 {
 374         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 375
 376         WARN_ON_ONCE(!rcu_read_lock_held());
 377
 378         if (unlikely(key >= map->max_entries))
 379                 return NULL;
 380         return READ_ONCE(stab->sks[key]);
 381 }
 382
 383 static void *sock_map_lookup(struct bpf_map *map, void *key)
 384 {
 385         struct sock *sk;
 386
 387         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 388         if (!sk)
 389                 return NULL;
 390         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 391                 return NULL;
 392         return sk;
 393 }
 394
 395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 396 {
 397         struct sock *sk;
 398
 399         if (map->value_size != sizeof(u64))
 400                 return ERR_PTR(-ENOSPC);
 401
 402         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 403         if (!sk)
 404                 return ERR_PTR(-ENOENT);
 405
 406         __sock_gen_cookie(sk);
 407         return &sk->sk_cookie;
 408 }
 409
 410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 411                              struct sock **psk)
 412 {
 413         struct sock *sk;
 414         int err = 0;
 415
 416         raw_spin_lock_bh(&stab->lock);
 417         sk = *psk;
 418         if (!sk_test || sk_test == sk)
 419                 sk = xchg(psk, NULL);
 420
 421         if (likely(sk))
 422                 sock_map_unref(sk, psk);
 423         else
 424                 err = -EINVAL;
 425
 426         raw_spin_unlock_bh(&stab->lock);
 427         return err;
 428 }
 429
 430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 431                                       void *link_raw)
 432 {
 433         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 434
 435         __sock_map_delete(stab, sk, link_raw);
 436 }
 437
 438 static long sock_map_delete_elem(struct bpf_map *map, void *key)
 439 {
 440         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 441         u32 i = *(u32 *)key;
 442         struct sock **psk;
 443
 444         if (unlikely(i >= map->max_entries))
 445                 return -EINVAL;
 446
 447         psk = &stab->sks[i];
 448         return __sock_map_delete(stab, NULL, psk);
 449 }
 450
 451 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 452 {
 453         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 454         u32 i = key ? *(u32 *)key : U32_MAX;
 455         u32 *key_next = next;
 456
 457         if (i == stab->map.max_entries - 1)
 458                 return -ENOENT;
 459         if (i >= stab->map.max_entries)
 460                 *key_next = 0;
 461         else
 462                 *key_next = i + 1;
 463         return 0;
 464 }
 465
 466 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 467                                   struct sock *sk, u64 flags)
 468 {
 469         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 470         struct sk_psock_link *link;
 471         struct sk_psock *psock;
 472         struct sock *osk;
 473         int ret;
 474
 475         WARN_ON_ONCE(!rcu_read_lock_held());
 476         if (unlikely(flags > BPF_EXIST))
 477                 return -EINVAL;
 478         if (unlikely(idx >= map->max_entries))
 479                 return -E2BIG;
 480
 481         link = sk_psock_init_link();
 482         if (!link)
 483                 return -ENOMEM;
 484
 485         ret = sock_map_link(map, sk);
 486         if (ret < 0)
 487                 goto out_free;
 488
 489         psock = sk_psock(sk);
 490         WARN_ON_ONCE(!psock);
 491
 492         raw_spin_lock_bh(&stab->lock);
 493         osk = stab->sks[idx];
 494         if (osk && flags == BPF_NOEXIST) {
 495                 ret = -EEXIST;
 496                 goto out_unlock;
 497         } else if (!osk && flags == BPF_EXIST) {
 498                 ret = -ENOENT;
 499                 goto out_unlock;
 500         }
 501
 502         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 503         stab->sks[idx] = sk;
 504         if (osk)
 505                 sock_map_unref(osk, &stab->sks[idx]);
 506         raw_spin_unlock_bh(&stab->lock);
 507         return 0;
 508 out_unlock:
 509         raw_spin_unlock_bh(&stab->lock);
 510         if (psock)
 511                 sk_psock_put(sk, psock);
 512 out_free:
 513         sk_psock_free_link(link);
 514         return ret;
 515 }
 516
 517 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 518 {
 519         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 520                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 521                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 522 }
 523
 524 static bool sock_map_redirect_allowed(const struct sock *sk)
 525 {
 526         if (sk_is_tcp(sk))
 527                 return sk->sk_state != TCP_LISTEN;
 528         else
 529                 return sk->sk_state == TCP_ESTABLISHED;
 530 }
 531
 532 static bool sock_map_sk_is_suitable(const struct sock *sk)
 533 {
 534         return !!sk->sk_prot->psock_update_sk_prot;
 535 }
 536
 537 static bool sock_map_sk_state_allowed(const struct sock *sk)
 538 {
 539         if (sk_is_tcp(sk))
 540                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 541         return true;
 542 }
 543
 544 static int sock_hash_update_common(struct bpf_map *map, void *key,
 545                                    struct sock *sk, u64 flags);
 546
 547 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 548                              u64 flags)
 549 {
 550         struct socket *sock;
 551         struct sock *sk;
 552         int ret;
 553         u64 ufd;
 554
 555         if (map->value_size == sizeof(u64))
 556                 ufd = *(u64 *)value;
 557         else
 558                 ufd = *(u32 *)value;
 559         if (ufd > S32_MAX)
 560                 return -EINVAL;
 561
 562         sock = sockfd_lookup(ufd, &ret);
 563         if (!sock)
 564                 return ret;
 565         sk = sock->sk;
 566         if (!sk) {
 567                 ret = -EINVAL;
 568                 goto out;
 569         }
 570         if (!sock_map_sk_is_suitable(sk)) {
 571                 ret = -EOPNOTSUPP;
 572                 goto out;
 573         }
 574
 575         sock_map_sk_acquire(sk);
 576         if (!sock_map_sk_state_allowed(sk))
 577                 ret = -EOPNOTSUPP;
 578         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 579                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 580         else
 581                 ret = sock_hash_update_common(map, key, sk, flags);
 582         sock_map_sk_release(sk);
 583 out:
 584         sockfd_put(sock);
 585         return ret;
 586 }
 587
 588 static long sock_map_update_elem(struct bpf_map *map, void *key,
 589                                  void *value, u64 flags)
 590 {
 591         struct sock *sk = (struct sock *)value;
 592         int ret;
 593
 594         if (unlikely(!sk || !sk_fullsock(sk)))
 595                 return -EINVAL;
 596
 597         if (!sock_map_sk_is_suitable(sk))
 598                 return -EOPNOTSUPP;
 599
 600         local_bh_disable();
 601         bh_lock_sock(sk);
 602         if (!sock_map_sk_state_allowed(sk))
 603                 ret = -EOPNOTSUPP;
 604         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 605                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 606         else
 607                 ret = sock_hash_update_common(map, key, sk, flags);
 608         bh_unlock_sock(sk);
 609         local_bh_enable();
 610         return ret;
 611 }
 612
 613 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 614            struct bpf_map *, map, void *, key, u64, flags)
 615 {
 616         WARN_ON_ONCE(!rcu_read_lock_held());
 617
 618         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 619                    sock_map_op_okay(sops)))
 620                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 621                                               flags);
 622         return -EOPNOTSUPP;
 623 }
 624
 625 const struct bpf_func_proto bpf_sock_map_update_proto = {
 626         .func           = bpf_sock_map_update,
 627         .gpl_only       = false,
 628         .pkt_access     = true,
 629         .ret_type       = RET_INTEGER,
 630         .arg1_type      = ARG_PTR_TO_CTX,
 631         .arg2_type      = ARG_CONST_MAP_PTR,
 632         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 633         .arg4_type      = ARG_ANYTHING,
 634 };
 635
 636 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 637            struct bpf_map *, map, u32, key, u64, flags)
 638 {
 639         struct sock *sk;
 640
 641         if (unlikely(flags & ~(BPF_F_INGRESS)))
 642                 return SK_DROP;
 643
 644         sk = __sock_map_lookup_elem(map, key);
 645         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 646                 return SK_DROP;
 647
 648         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 649         return SK_PASS;
 650 }
 651
 652 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 653         .func           = bpf_sk_redirect_map,
 654         .gpl_only       = false,
 655         .ret_type       = RET_INTEGER,
 656         .arg1_type      = ARG_PTR_TO_CTX,
 657         .arg2_type      = ARG_CONST_MAP_PTR,
 658         .arg3_type      = ARG_ANYTHING,
 659         .arg4_type      = ARG_ANYTHING,
 660 };
 661
 662 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 663            struct bpf_map *, map, u32, key, u64, flags)
 664 {
 665         struct sock *sk;
 666
 667         if (unlikely(flags & ~(BPF_F_INGRESS)))
 668                 return SK_DROP;
 669
 670         sk = __sock_map_lookup_elem(map, key);
 671         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 672                 return SK_DROP;
 673
 674         msg->flags = flags;
 675         msg->sk_redir = sk;
 676         return SK_PASS;
 677 }
 678
 679 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 680         .func           = bpf_msg_redirect_map,
 681         .gpl_only       = false,
 682         .ret_type       = RET_INTEGER,
 683         .arg1_type      = ARG_PTR_TO_CTX,
 684         .arg2_type      = ARG_CONST_MAP_PTR,
 685         .arg3_type      = ARG_ANYTHING,
 686         .arg4_type      = ARG_ANYTHING,
 687 };
 688
 689 struct sock_map_seq_info {
 690         struct bpf_map *map;
 691         struct sock *sk;
 692         u32 index;
 693 };
 694
 695 struct bpf_iter__sockmap {
 696         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 697         __bpf_md_ptr(struct bpf_map *, map);
 698         __bpf_md_ptr(void *, key);
 699         __bpf_md_ptr(struct sock *, sk);
 700 };
 701
 702 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 703                      struct bpf_map *map, void *key,
 704                      struct sock *sk)
 705
 706 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 707 {
 708         if (unlikely(info->index >= info->map->max_entries))
 709                 return NULL;
 710
 711         info->sk = __sock_map_lookup_elem(info->map, info->index);
 712
 713         /* can't return sk directly, since that might be NULL */
 714         return info;
 715 }
 716
 717 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 718         __acquires(rcu)
 719 {
 720         struct sock_map_seq_info *info = seq->private;
 721
 722         if (*pos == 0)
 723                 ++*pos;
 724
 725         /* pairs with sock_map_seq_stop */
 726         rcu_read_lock();
 727         return sock_map_seq_lookup_elem(info);
 728 }
 729
 730 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 731         __must_hold(rcu)
 732 {
 733         struct sock_map_seq_info *info = seq->private;
 734
 735         ++*pos;
 736         ++info->index;
 737
 738         return sock_map_seq_lookup_elem(info);
 739 }
 740
 741 static int sock_map_seq_show(struct seq_file *seq, void *v)
 742         __must_hold(rcu)
 743 {
 744         struct sock_map_seq_info *info = seq->private;
 745         struct bpf_iter__sockmap ctx = {};
 746         struct bpf_iter_meta meta;
 747         struct bpf_prog *prog;
 748
 749         meta.seq = seq;
 750         prog = bpf_iter_get_info(&meta, !v);
 751         if (!prog)
 752                 return 0;
 753
 754         ctx.meta = &meta;
 755         ctx.map = info->map;
 756         if (v) {
 757                 ctx.key = &info->index;
 758                 ctx.sk = info->sk;
 759         }
 760
 761         return bpf_iter_run_prog(prog, &ctx);
 762 }
 763
 764 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 765         __releases(rcu)
 766 {
 767         if (!v)
 768                 (void)sock_map_seq_show(seq, NULL);
 769
 770         /* pairs with sock_map_seq_start */
 771         rcu_read_unlock();
 772 }
 773
 774 static const struct seq_operations sock_map_seq_ops = {
 775         .start  = sock_map_seq_start,
 776         .next   = sock_map_seq_next,
 777         .stop   = sock_map_seq_stop,
 778         .show   = sock_map_seq_show,
 779 };
 780
 781 static int sock_map_init_seq_private(void *priv_data,
 782                                      struct bpf_iter_aux_info *aux)
 783 {
 784         struct sock_map_seq_info *info = priv_data;
 785
 786         bpf_map_inc_with_uref(aux->map);
 787         info->map = aux->map;
 788         return 0;
 789 }
 790
 791 static void sock_map_fini_seq_private(void *priv_data)
 792 {
 793         struct sock_map_seq_info *info = priv_data;
 794
 795         bpf_map_put_with_uref(info->map);
 796 }
 797
 798 static u64 sock_map_mem_usage(const struct bpf_map *map)
 799 {
 800         u64 usage = sizeof(struct bpf_stab);
 801
 802         usage += (u64)map->max_entries * sizeof(struct sock *);
 803         return usage;
 804 }
 805
 806 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 807         .seq_ops                = &sock_map_seq_ops,
 808         .init_seq_private       = sock_map_init_seq_private,
 809         .fini_seq_private       = sock_map_fini_seq_private,
 810         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 811 };
 812
 813 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 814 const struct bpf_map_ops sock_map_ops = {
 815         .map_meta_equal         = bpf_map_meta_equal,
 816         .map_alloc              = sock_map_alloc,
 817         .map_free               = sock_map_free,
 818         .map_get_next_key       = sock_map_get_next_key,
 819         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 820         .map_update_elem        = sock_map_update_elem,
 821         .map_delete_elem        = sock_map_delete_elem,
 822         .map_lookup_elem        = sock_map_lookup,
 823         .map_release_uref       = sock_map_release_progs,
 824         .map_check_btf          = map_check_no_btf,
 825         .map_mem_usage          = sock_map_mem_usage,
 826         .map_btf_id             = &sock_map_btf_ids[0],
 827         .iter_seq_info          = &sock_map_iter_seq_info,
 828 };
 829
 830 struct bpf_shtab_elem {
 831         struct rcu_head rcu;
 832         u32 hash;
 833         struct sock *sk;
 834         struct hlist_node node;
 835         u8 key[];
 836 };
 837
 838 struct bpf_shtab_bucket {
 839         struct hlist_head head;
 840         raw_spinlock_t lock;
 841 };
 842
 843 struct bpf_shtab {
 844         struct bpf_map map;
 845         struct bpf_shtab_bucket *buckets;
 846         u32 buckets_num;
 847         u32 elem_size;
 848         struct sk_psock_progs progs;
 849         atomic_t count;
 850 };
 851
 852 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 853 {
 854         return jhash(key, len, 0);
 855 }
 856
 857 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 858                                                         u32 hash)
 859 {
 860         return &htab->buckets[hash & (htab->buckets_num - 1)];
 861 }
 862
 863 static struct bpf_shtab_elem *
 864 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 865                           u32 key_size)
 866 {
 867         struct bpf_shtab_elem *elem;
 868
 869         hlist_for_each_entry_rcu(elem, head, node) {
 870                 if (elem->hash == hash &&
 871                     !memcmp(&elem->key, key, key_size))
 872                         return elem;
 873         }
 874
 875         return NULL;
 876 }
 877
 878 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 879 {
 880         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 881         u32 key_size = map->key_size, hash;
 882         struct bpf_shtab_bucket *bucket;
 883         struct bpf_shtab_elem *elem;
 884
 885         WARN_ON_ONCE(!rcu_read_lock_held());
 886
 887         hash = sock_hash_bucket_hash(key, key_size);
 888         bucket = sock_hash_select_bucket(htab, hash);
 889         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 890
 891         return elem ? elem->sk : NULL;
 892 }
 893
 894 static void sock_hash_free_elem(struct bpf_shtab *htab,
 895                                 struct bpf_shtab_elem *elem)
 896 {
 897         atomic_dec(&htab->count);
 898         kfree_rcu(elem, rcu);
 899 }
 900
 901 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 902                                        void *link_raw)
 903 {
 904         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 905         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 906         struct bpf_shtab_bucket *bucket;
 907
 908         WARN_ON_ONCE(!rcu_read_lock_held());
 909         bucket = sock_hash_select_bucket(htab, elem->hash);
 910
 911         /* elem may be deleted in parallel from the map, but access here
 912          * is okay since it's going away only after RCU grace period.
 913          * However, we need to check whether it's still present.
 914          */
 915         raw_spin_lock_bh(&bucket->lock);
 916         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 917                                                elem->key, map->key_size);
 918         if (elem_probe && elem_probe == elem) {
 919                 hlist_del_rcu(&elem->node);
 920                 sock_map_unref(elem->sk, elem);
 921                 sock_hash_free_elem(htab, elem);
 922         }
 923         raw_spin_unlock_bh(&bucket->lock);
 924 }
 925
 926 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
 927 {
 928         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 929         u32 hash, key_size = map->key_size;
 930         struct bpf_shtab_bucket *bucket;
 931         struct bpf_shtab_elem *elem;
 932         int ret = -ENOENT;
 933
 934         hash = sock_hash_bucket_hash(key, key_size);
 935         bucket = sock_hash_select_bucket(htab, hash);
 936
 937         raw_spin_lock_bh(&bucket->lock);
 938         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 939         if (elem) {
 940                 hlist_del_rcu(&elem->node);
 941                 sock_map_unref(elem->sk, elem);
 942                 sock_hash_free_elem(htab, elem);
 943                 ret = 0;
 944         }
 945         raw_spin_unlock_bh(&bucket->lock);
 946         return ret;
 947 }
 948
 949 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 950                                                    void *key, u32 key_size,
 951                                                    u32 hash, struct sock *sk,
 952                                                    struct bpf_shtab_elem *old)
 953 {
 954         struct bpf_shtab_elem *new;
 955
 956         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 957                 if (!old) {
 958                         atomic_dec(&htab->count);
 959                         return ERR_PTR(-E2BIG);
 960                 }
 961         }
 962
 963         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 964                                    GFP_ATOMIC | __GFP_NOWARN,
 965                                    htab->map.numa_node);
 966         if (!new) {
 967                 atomic_dec(&htab->count);
 968                 return ERR_PTR(-ENOMEM);
 969         }
 970         memcpy(new->key, key, key_size);
 971         new->sk = sk;
 972         new->hash = hash;
 973         return new;
 974 }
 975
 976 static int sock_hash_update_common(struct bpf_map *map, void *key,
 977                                    struct sock *sk, u64 flags)
 978 {
 979         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 980         u32 key_size = map->key_size, hash;
 981         struct bpf_shtab_elem *elem, *elem_new;
 982         struct bpf_shtab_bucket *bucket;
 983         struct sk_psock_link *link;
 984         struct sk_psock *psock;
 985         int ret;
 986
 987         WARN_ON_ONCE(!rcu_read_lock_held());
 988         if (unlikely(flags > BPF_EXIST))
 989                 return -EINVAL;
 990
 991         link = sk_psock_init_link();
 992         if (!link)
 993                 return -ENOMEM;
 994
 995         ret = sock_map_link(map, sk);
 996         if (ret < 0)
 997                 goto out_free;
 998
 999         psock = sk_psock(sk);
1000         WARN_ON_ONCE(!psock);
1001
1002         hash = sock_hash_bucket_hash(key, key_size);
1003         bucket = sock_hash_select_bucket(htab, hash);
1004
1005         raw_spin_lock_bh(&bucket->lock);
1006         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1007         if (elem && flags == BPF_NOEXIST) {
1008                 ret = -EEXIST;
1009                 goto out_unlock;
1010         } else if (!elem && flags == BPF_EXIST) {
1011                 ret = -ENOENT;
1012                 goto out_unlock;
1013         }
1014
1015         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1016         if (IS_ERR(elem_new)) {
1017                 ret = PTR_ERR(elem_new);
1018                 goto out_unlock;
1019         }
1020
1021         sock_map_add_link(psock, link, map, elem_new);
1022         /* Add new element to the head of the list, so that
1023          * concurrent search will find it before old elem.
1024          */
1025         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1026         if (elem) {
1027                 hlist_del_rcu(&elem->node);
1028                 sock_map_unref(elem->sk, elem);
1029                 sock_hash_free_elem(htab, elem);
1030         }
1031         raw_spin_unlock_bh(&bucket->lock);
1032         return 0;
1033 out_unlock:
1034         raw_spin_unlock_bh(&bucket->lock);
1035         sk_psock_put(sk, psock);
1036 out_free:
1037         sk_psock_free_link(link);
1038         return ret;
1039 }
1040
1041 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1042                                   void *key_next)
1043 {
1044         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1045         struct bpf_shtab_elem *elem, *elem_next;
1046         u32 hash, key_size = map->key_size;
1047         struct hlist_head *head;
1048         int i = 0;
1049
1050         if (!key)
1051                 goto find_first_elem;
1052         hash = sock_hash_bucket_hash(key, key_size);
1053         head = &sock_hash_select_bucket(htab, hash)->head;
1054         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1055         if (!elem)
1056                 goto find_first_elem;
1057
1058         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1059                                      struct bpf_shtab_elem, node);
1060         if (elem_next) {
1061                 memcpy(key_next, elem_next->key, key_size);
1062                 return 0;
1063         }
1064
1065         i = hash & (htab->buckets_num - 1);
1066         i++;
1067 find_first_elem:
1068         for (; i < htab->buckets_num; i++) {
1069                 head = &sock_hash_select_bucket(htab, i)->head;
1070                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1071                                              struct bpf_shtab_elem, node);
1072                 if (elem_next) {
1073                         memcpy(key_next, elem_next->key, key_size);
1074                         return 0;
1075                 }
1076         }
1077
1078         return -ENOENT;
1079 }
1080
1081 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1082 {
1083         struct bpf_shtab *htab;
1084         int i, err;
1085
1086         if (attr->max_entries == 0 ||
1087             attr->key_size    == 0 ||
1088             (attr->value_size != sizeof(u32) &&
1089              attr->value_size != sizeof(u64)) ||
1090             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1091                 return ERR_PTR(-EINVAL);
1092         if (attr->key_size > MAX_BPF_STACK)
1093                 return ERR_PTR(-E2BIG);
1094
1095         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1096         if (!htab)
1097                 return ERR_PTR(-ENOMEM);
1098
1099         bpf_map_init_from_attr(&htab->map, attr);
1100
1101         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1102         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1103                           round_up(htab->map.key_size, 8);
1104         if (htab->buckets_num == 0 ||
1105             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1106                 err = -EINVAL;
1107                 goto free_htab;
1108         }
1109
1110         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1111                                            sizeof(struct bpf_shtab_bucket),
1112                                            htab->map.numa_node);
1113         if (!htab->buckets) {
1114                 err = -ENOMEM;
1115                 goto free_htab;
1116         }
1117
1118         for (i = 0; i < htab->buckets_num; i++) {
1119                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1120                 raw_spin_lock_init(&htab->buckets[i].lock);
1121         }
1122
1123         return &htab->map;
1124 free_htab:
1125         bpf_map_area_free(htab);
1126         return ERR_PTR(err);
1127 }
1128
1129 static void sock_hash_free(struct bpf_map *map)
1130 {
1131         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1132         struct bpf_shtab_bucket *bucket;
1133         struct hlist_head unlink_list;
1134         struct bpf_shtab_elem *elem;
1135         struct hlist_node *node;
1136         int i;
1137
1138         /* After the sync no updates or deletes will be in-flight so it
1139          * is safe to walk map and remove entries without risking a race
1140          * in EEXIST update case.
1141          */
1142         synchronize_rcu();
1143         for (i = 0; i < htab->buckets_num; i++) {
1144                 bucket = sock_hash_select_bucket(htab, i);
1145
1146                 /* We are racing with sock_hash_delete_from_link to
1147                  * enter the spin-lock critical section. Every socket on
1148                  * the list is still linked to sockhash. Since link
1149                  * exists, psock exists and holds a ref to socket. That
1150                  * lets us to grab a socket ref too.
1151                  */
1152                 raw_spin_lock_bh(&bucket->lock);
1153                 hlist_for_each_entry(elem, &bucket->head, node)
1154                         sock_hold(elem->sk);
1155                 hlist_move_list(&bucket->head, &unlink_list);
1156                 raw_spin_unlock_bh(&bucket->lock);
1157
1158                 /* Process removed entries out of atomic context to
1159                  * block for socket lock before deleting the psock's
1160                  * link to sockhash.
1161                  */
1162                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1163                         hlist_del(&elem->node);
1164                         lock_sock(elem->sk);
1165                         rcu_read_lock();
1166                         sock_map_unref(elem->sk, elem);
1167                         rcu_read_unlock();
1168                         release_sock(elem->sk);
1169                         sock_put(elem->sk);
1170                         sock_hash_free_elem(htab, elem);
1171                 }
1172         }
1173
1174         /* wait for psock readers accessing its map link */
1175         synchronize_rcu();
1176
1177         bpf_map_area_free(htab->buckets);
1178         bpf_map_area_free(htab);
1179 }
1180
1181 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1182 {
1183         struct sock *sk;
1184
1185         if (map->value_size != sizeof(u64))
1186                 return ERR_PTR(-ENOSPC);
1187
1188         sk = __sock_hash_lookup_elem(map, key);
1189         if (!sk)
1190                 return ERR_PTR(-ENOENT);
1191
1192         __sock_gen_cookie(sk);
1193         return &sk->sk_cookie;
1194 }
1195
1196 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1197 {
1198         struct sock *sk;
1199
1200         sk = __sock_hash_lookup_elem(map, key);
1201         if (!sk)
1202                 return NULL;
1203         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1204                 return NULL;
1205         return sk;
1206 }
1207
1208 static void sock_hash_release_progs(struct bpf_map *map)
1209 {
1210         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1211 }
1212
1213 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1214            struct bpf_map *, map, void *, key, u64, flags)
1215 {
1216         WARN_ON_ONCE(!rcu_read_lock_held());
1217
1218         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1219                    sock_map_op_okay(sops)))
1220                 return sock_hash_update_common(map, key, sops->sk, flags);
1221         return -EOPNOTSUPP;
1222 }
1223
1224 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1225         .func           = bpf_sock_hash_update,
1226         .gpl_only       = false,
1227         .pkt_access     = true,
1228         .ret_type       = RET_INTEGER,
1229         .arg1_type      = ARG_PTR_TO_CTX,
1230         .arg2_type      = ARG_CONST_MAP_PTR,
1231         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1232         .arg4_type      = ARG_ANYTHING,
1233 };
1234
1235 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1236            struct bpf_map *, map, void *, key, u64, flags)
1237 {
1238         struct sock *sk;
1239
1240         if (unlikely(flags & ~(BPF_F_INGRESS)))
1241                 return SK_DROP;
1242
1243         sk = __sock_hash_lookup_elem(map, key);
1244         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1245                 return SK_DROP;
1246
1247         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1248         return SK_PASS;
1249 }
1250
1251 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1252         .func           = bpf_sk_redirect_hash,
1253         .gpl_only       = false,
1254         .ret_type       = RET_INTEGER,
1255         .arg1_type      = ARG_PTR_TO_CTX,
1256         .arg2_type      = ARG_CONST_MAP_PTR,
1257         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1258         .arg4_type      = ARG_ANYTHING,
1259 };
1260
1261 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1262            struct bpf_map *, map, void *, key, u64, flags)
1263 {
1264         struct sock *sk;
1265
1266         if (unlikely(flags & ~(BPF_F_INGRESS)))
1267                 return SK_DROP;
1268
1269         sk = __sock_hash_lookup_elem(map, key);
1270         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1271                 return SK_DROP;
1272
1273         msg->flags = flags;
1274         msg->sk_redir = sk;
1275         return SK_PASS;
1276 }
1277
1278 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1279         .func           = bpf_msg_redirect_hash,
1280         .gpl_only       = false,
1281         .ret_type       = RET_INTEGER,
1282         .arg1_type      = ARG_PTR_TO_CTX,
1283         .arg2_type      = ARG_CONST_MAP_PTR,
1284         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1285         .arg4_type      = ARG_ANYTHING,
1286 };
1287
1288 struct sock_hash_seq_info {
1289         struct bpf_map *map;
1290         struct bpf_shtab *htab;
1291         u32 bucket_id;
1292 };
1293
1294 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1295                                      struct bpf_shtab_elem *prev_elem)
1296 {
1297         const struct bpf_shtab *htab = info->htab;
1298         struct bpf_shtab_bucket *bucket;
1299         struct bpf_shtab_elem *elem;
1300         struct hlist_node *node;
1301
1302         /* try to find next elem in the same bucket */
1303         if (prev_elem) {
1304                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1305                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1306                 if (elem)
1307                         return elem;
1308
1309                 /* no more elements, continue in the next bucket */
1310                 info->bucket_id++;
1311         }
1312
1313         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1314                 bucket = &htab->buckets[info->bucket_id];
1315                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1316                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1317                 if (elem)
1318                         return elem;
1319         }
1320
1321         return NULL;
1322 }
1323
1324 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1325         __acquires(rcu)
1326 {
1327         struct sock_hash_seq_info *info = seq->private;
1328
1329         if (*pos == 0)
1330                 ++*pos;
1331
1332         /* pairs with sock_hash_seq_stop */
1333         rcu_read_lock();
1334         return sock_hash_seq_find_next(info, NULL);
1335 }
1336
1337 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1338         __must_hold(rcu)
1339 {
1340         struct sock_hash_seq_info *info = seq->private;
1341
1342         ++*pos;
1343         return sock_hash_seq_find_next(info, v);
1344 }
1345
1346 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1347         __must_hold(rcu)
1348 {
1349         struct sock_hash_seq_info *info = seq->private;
1350         struct bpf_iter__sockmap ctx = {};
1351         struct bpf_shtab_elem *elem = v;
1352         struct bpf_iter_meta meta;
1353         struct bpf_prog *prog;
1354
1355         meta.seq = seq;
1356         prog = bpf_iter_get_info(&meta, !elem);
1357         if (!prog)
1358                 return 0;
1359
1360         ctx.meta = &meta;
1361         ctx.map = info->map;
1362         if (elem) {
1363                 ctx.key = elem->key;
1364                 ctx.sk = elem->sk;
1365         }
1366
1367         return bpf_iter_run_prog(prog, &ctx);
1368 }
1369
1370 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1371         __releases(rcu)
1372 {
1373         if (!v)
1374                 (void)sock_hash_seq_show(seq, NULL);
1375
1376         /* pairs with sock_hash_seq_start */
1377         rcu_read_unlock();
1378 }
1379
1380 static const struct seq_operations sock_hash_seq_ops = {
1381         .start  = sock_hash_seq_start,
1382         .next   = sock_hash_seq_next,
1383         .stop   = sock_hash_seq_stop,
1384         .show   = sock_hash_seq_show,
1385 };
1386
1387 static int sock_hash_init_seq_private(void *priv_data,
1388                                       struct bpf_iter_aux_info *aux)
1389 {
1390         struct sock_hash_seq_info *info = priv_data;
1391
1392         bpf_map_inc_with_uref(aux->map);
1393         info->map = aux->map;
1394         info->htab = container_of(aux->map, struct bpf_shtab, map);
1395         return 0;
1396 }
1397
1398 static void sock_hash_fini_seq_private(void *priv_data)
1399 {
1400         struct sock_hash_seq_info *info = priv_data;
1401
1402         bpf_map_put_with_uref(info->map);
1403 }
1404
1405 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1406 {
1407         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1408         u64 usage = sizeof(*htab);
1409
1410         usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1411         usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1412         return usage;
1413 }
1414
1415 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1416         .seq_ops                = &sock_hash_seq_ops,
1417         .init_seq_private       = sock_hash_init_seq_private,
1418         .fini_seq_private       = sock_hash_fini_seq_private,
1419         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1420 };
1421
1422 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1423 const struct bpf_map_ops sock_hash_ops = {
1424         .map_meta_equal         = bpf_map_meta_equal,
1425         .map_alloc              = sock_hash_alloc,
1426         .map_free               = sock_hash_free,
1427         .map_get_next_key       = sock_hash_get_next_key,
1428         .map_update_elem        = sock_map_update_elem,
1429         .map_delete_elem        = sock_hash_delete_elem,
1430         .map_lookup_elem        = sock_hash_lookup,
1431         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1432         .map_release_uref       = sock_hash_release_progs,
1433         .map_check_btf          = map_check_no_btf,
1434         .map_mem_usage          = sock_hash_mem_usage,
1435         .map_btf_id             = &sock_hash_map_btf_ids[0],
1436         .iter_seq_info          = &sock_hash_iter_seq_info,
1437 };
1438
1439 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1440 {
1441         switch (map->map_type) {
1442         case BPF_MAP_TYPE_SOCKMAP:
1443                 return &container_of(map, struct bpf_stab, map)->progs;
1444         case BPF_MAP_TYPE_SOCKHASH:
1445                 return &container_of(map, struct bpf_shtab, map)->progs;
1446         default:
1447                 break;
1448         }
1449
1450         return NULL;
1451 }
1452
1453 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1454                                 u32 which)
1455 {
1456         struct sk_psock_progs *progs = sock_map_progs(map);
1457
1458         if (!progs)
1459                 return -EOPNOTSUPP;
1460
1461         switch (which) {
1462         case BPF_SK_MSG_VERDICT:
1463                 *pprog = &progs->msg_parser;
1464                 break;
1465 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1466         case BPF_SK_SKB_STREAM_PARSER:
1467                 *pprog = &progs->stream_parser;
1468                 break;
1469 #endif
1470         case BPF_SK_SKB_STREAM_VERDICT:
1471                 if (progs->skb_verdict)
1472                         return -EBUSY;
1473                 *pprog = &progs->stream_verdict;
1474                 break;
1475         case BPF_SK_SKB_VERDICT:
1476                 if (progs->stream_verdict)
1477                         return -EBUSY;
1478                 *pprog = &progs->skb_verdict;
1479                 break;
1480         default:
1481                 return -EOPNOTSUPP;
1482         }
1483
1484         return 0;
1485 }
1486
1487 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1488                                 struct bpf_prog *old, u32 which)
1489 {
1490         struct bpf_prog **pprog;
1491         int ret;
1492
1493         ret = sock_map_prog_lookup(map, &pprog, which);
1494         if (ret)
1495                 return ret;
1496
1497         if (old)
1498                 return psock_replace_prog(pprog, prog, old);
1499
1500         psock_set_prog(pprog, prog);
1501         return 0;
1502 }
1503
1504 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1505                             union bpf_attr __user *uattr)
1506 {
1507         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1508         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1509         struct bpf_prog **pprog;
1510         struct bpf_prog *prog;
1511         struct bpf_map *map;
1512         struct fd f;
1513         u32 id = 0;
1514         int ret;
1515
1516         if (attr->query.query_flags)
1517                 return -EINVAL;
1518
1519         f = fdget(ufd);
1520         map = __bpf_map_get(f);
1521         if (IS_ERR(map))
1522                 return PTR_ERR(map);
1523
1524         rcu_read_lock();
1525
1526         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1527         if (ret)
1528                 goto end;
1529
1530         prog = *pprog;
1531         prog_cnt = !prog ? 0 : 1;
1532
1533         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1534                 goto end;
1535
1536         /* we do not hold the refcnt, the bpf prog may be released
1537          * asynchronously and the id would be set to 0.
1538          */
1539         id = data_race(prog->aux->id);
1540         if (id == 0)
1541                 prog_cnt = 0;
1542
1543 end:
1544         rcu_read_unlock();
1545
1546         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1547             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1548             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1549                 ret = -EFAULT;
1550
1551         fdput(f);
1552         return ret;
1553 }
1554
1555 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1556 {
1557         switch (link->map->map_type) {
1558         case BPF_MAP_TYPE_SOCKMAP:
1559                 return sock_map_delete_from_link(link->map, sk,
1560                                                  link->link_raw);
1561         case BPF_MAP_TYPE_SOCKHASH:
1562                 return sock_hash_delete_from_link(link->map, sk,
1563                                                   link->link_raw);
1564         default:
1565                 break;
1566         }
1567 }
1568
1569 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1570 {
1571         struct sk_psock_link *link;
1572
1573         while ((link = sk_psock_link_pop(psock))) {
1574                 sock_map_unlink(sk, link);
1575                 sk_psock_free_link(link);
1576         }
1577 }
1578
1579 void sock_map_unhash(struct sock *sk)
1580 {
1581         void (*saved_unhash)(struct sock *sk);
1582         struct sk_psock *psock;
1583
1584         rcu_read_lock();
1585         psock = sk_psock(sk);
1586         if (unlikely(!psock)) {
1587                 rcu_read_unlock();
1588                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1589         } else {
1590                 saved_unhash = psock->saved_unhash;
1591                 sock_map_remove_links(sk, psock);
1592                 rcu_read_unlock();
1593         }
1594         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1595                 return;
1596         if (saved_unhash)
1597                 saved_unhash(sk);
1598 }
1599 EXPORT_SYMBOL_GPL(sock_map_unhash);
1600
1601 void sock_map_destroy(struct sock *sk)
1602 {
1603         void (*saved_destroy)(struct sock *sk);
1604         struct sk_psock *psock;
1605
1606         rcu_read_lock();
1607         psock = sk_psock_get(sk);
1608         if (unlikely(!psock)) {
1609                 rcu_read_unlock();
1610                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1611         } else {
1612                 saved_destroy = psock->saved_destroy;
1613                 sock_map_remove_links(sk, psock);
1614                 rcu_read_unlock();
1615                 sk_psock_stop(psock);
1616                 sk_psock_put(sk, psock);
1617         }
1618         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1619                 return;
1620         if (saved_destroy)
1621                 saved_destroy(sk);
1622 }
1623 EXPORT_SYMBOL_GPL(sock_map_destroy);
1624
1625 void sock_map_close(struct sock *sk, long timeout)
1626 {
1627         void (*saved_close)(struct sock *sk, long timeout);
1628         struct sk_psock *psock;
1629
1630         lock_sock(sk);
1631         rcu_read_lock();
1632         psock = sk_psock_get(sk);
1633         if (unlikely(!psock)) {
1634                 rcu_read_unlock();
1635                 release_sock(sk);
1636                 saved_close = READ_ONCE(sk->sk_prot)->close;
1637         } else {
1638                 saved_close = psock->saved_close;
1639                 sock_map_remove_links(sk, psock);
1640                 rcu_read_unlock();
1641                 sk_psock_stop(psock);
1642                 release_sock(sk);
1643                 cancel_delayed_work_sync(&psock->work);
1644                 sk_psock_put(sk, psock);
1645         }
1646
1647         /* Make sure we do not recurse. This is a bug.
1648          * Leak the socket instead of crashing on a stack overflow.
1649          */
1650         if (WARN_ON_ONCE(saved_close == sock_map_close))
1651                 return;
1652         saved_close(sk, timeout);
1653 }
1654 EXPORT_SYMBOL_GPL(sock_map_close);
1655
1656 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1657                                        union bpf_iter_link_info *linfo,
1658                                        struct bpf_iter_aux_info *aux)
1659 {
1660         struct bpf_map *map;
1661         int err = -EINVAL;
1662
1663         if (!linfo->map.map_fd)
1664                 return -EBADF;
1665
1666         map = bpf_map_get_with_uref(linfo->map.map_fd);
1667         if (IS_ERR(map))
1668                 return PTR_ERR(map);
1669
1670         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1671             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1672                 goto put_map;
1673
1674         if (prog->aux->max_rdonly_access > map->key_size) {
1675                 err = -EACCES;
1676                 goto put_map;
1677         }
1678
1679         aux->map = map;
1680         return 0;
1681
1682 put_map:
1683         bpf_map_put_with_uref(map);
1684         return err;
1685 }
1686
1687 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1688 {
1689         bpf_map_put_with_uref(aux->map);
1690 }
1691
1692 static struct bpf_iter_reg sock_map_iter_reg = {
1693         .target                 = "sockmap",
1694         .attach_target          = sock_map_iter_attach_target,
1695         .detach_target          = sock_map_iter_detach_target,
1696         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1697         .fill_link_info         = bpf_iter_map_fill_link_info,
1698         .ctx_arg_info_size      = 2,
1699         .ctx_arg_info           = {
1700                 { offsetof(struct bpf_iter__sockmap, key),
1701                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1702                 { offsetof(struct bpf_iter__sockmap, sk),
1703                   PTR_TO_BTF_ID_OR_NULL },
1704         },
1705 };
1706
1707 static int __init bpf_sockmap_iter_init(void)
1708 {
1709         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1710                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1711         return bpf_iter_reg_target(&sock_map_iter_reg);
1712 }
1713 late_initcall(bpf_sockmap_iter_init);