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 (!capable(CAP_NET_ADMIN))
  36                 return ERR_PTR(-EPERM);
  37         if (attr->max_entries == 0 ||
  38             attr->key_size    != 4 ||
  39             (attr->value_size != sizeof(u32) &&
  40              attr->value_size != sizeof(u64)) ||
  41             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  42                 return ERR_PTR(-EINVAL);
  43
  44         stab = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
  45         if (!stab)
  46                 return ERR_PTR(-ENOMEM);
  47
  48         bpf_map_init_from_attr(&stab->map, attr);
  49         raw_spin_lock_init(&stab->lock);
  50
  51         stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
  52                                        sizeof(struct sock *),
  53                                        stab->map.numa_node);
  54         if (!stab->sks) {
  55                 kfree(stab);
  56                 return ERR_PTR(-ENOMEM);
  57         }
  58
  59         return &stab->map;
  60 }
  61
  62 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  63 {
  64         u32 ufd = attr->target_fd;
  65         struct bpf_map *map;
  66         struct fd f;
  67         int ret;
  68
  69         if (attr->attach_flags || attr->replace_bpf_fd)
  70                 return -EINVAL;
  71
  72         f = fdget(ufd);
  73         map = __bpf_map_get(f);
  74         if (IS_ERR(map))
  75                 return PTR_ERR(map);
  76         ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
  77         fdput(f);
  78         return ret;
  79 }
  80
  81 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  82 {
  83         u32 ufd = attr->target_fd;
  84         struct bpf_prog *prog;
  85         struct bpf_map *map;
  86         struct fd f;
  87         int ret;
  88
  89         if (attr->attach_flags || attr->replace_bpf_fd)
  90                 return -EINVAL;
  91
  92         f = fdget(ufd);
  93         map = __bpf_map_get(f);
  94         if (IS_ERR(map))
  95                 return PTR_ERR(map);
  96
  97         prog = bpf_prog_get(attr->attach_bpf_fd);
  98         if (IS_ERR(prog)) {
  99                 ret = PTR_ERR(prog);
 100                 goto put_map;
 101         }
 102
 103         if (prog->type != ptype) {
 104                 ret = -EINVAL;
 105                 goto put_prog;
 106         }
 107
 108         ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
 109 put_prog:
 110         bpf_prog_put(prog);
 111 put_map:
 112         fdput(f);
 113         return ret;
 114 }
 115
 116 static void sock_map_sk_acquire(struct sock *sk)
 117         __acquires(&sk->sk_lock.slock)
 118 {
 119         lock_sock(sk);
 120         preempt_disable();
 121         rcu_read_lock();
 122 }
 123
 124 static void sock_map_sk_release(struct sock *sk)
 125         __releases(&sk->sk_lock.slock)
 126 {
 127         rcu_read_unlock();
 128         preempt_enable();
 129         release_sock(sk);
 130 }
 131
 132 static void sock_map_add_link(struct sk_psock *psock,
 133                               struct sk_psock_link *link,
 134                               struct bpf_map *map, void *link_raw)
 135 {
 136         link->link_raw = link_raw;
 137         link->map = map;
 138         spin_lock_bh(&psock->link_lock);
 139         list_add_tail(&link->list, &psock->link);
 140         spin_unlock_bh(&psock->link_lock);
 141 }
 142
 143 static void sock_map_del_link(struct sock *sk,
 144                               struct sk_psock *psock, void *link_raw)
 145 {
 146         bool strp_stop = false, verdict_stop = false;
 147         struct sk_psock_link *link, *tmp;
 148
 149         spin_lock_bh(&psock->link_lock);
 150         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 151                 if (link->link_raw == link_raw) {
 152                         struct bpf_map *map = link->map;
 153                         struct bpf_stab *stab = container_of(map, struct bpf_stab,
 154                                                              map);
 155                         if (psock->saved_data_ready && stab->progs.stream_parser)
 156                                 strp_stop = true;
 157                         if (psock->saved_data_ready && stab->progs.stream_verdict)
 158                                 verdict_stop = true;
 159                         if (psock->saved_data_ready && stab->progs.skb_verdict)
 160                                 verdict_stop = true;
 161                         list_del(&link->list);
 162                         sk_psock_free_link(link);
 163                 }
 164         }
 165         spin_unlock_bh(&psock->link_lock);
 166         if (strp_stop || verdict_stop) {
 167                 write_lock_bh(&sk->sk_callback_lock);
 168                 if (strp_stop)
 169                         sk_psock_stop_strp(sk, psock);
 170                 if (verdict_stop)
 171                         sk_psock_stop_verdict(sk, psock);
 172
 173                 if (psock->psock_update_sk_prot)
 174                         psock->psock_update_sk_prot(sk, psock, false);
 175                 write_unlock_bh(&sk->sk_callback_lock);
 176         }
 177 }
 178
 179 static void sock_map_unref(struct sock *sk, void *link_raw)
 180 {
 181         struct sk_psock *psock = sk_psock(sk);
 182
 183         if (likely(psock)) {
 184                 sock_map_del_link(sk, psock, link_raw);
 185                 sk_psock_put(sk, psock);
 186         }
 187 }
 188
 189 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
 190 {
 191         if (!sk->sk_prot->psock_update_sk_prot)
 192                 return -EINVAL;
 193         psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
 194         return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
 195 }
 196
 197 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 198 {
 199         struct sk_psock *psock;
 200
 201         rcu_read_lock();
 202         psock = sk_psock(sk);
 203         if (psock) {
 204                 if (sk->sk_prot->close != sock_map_close) {
 205                         psock = ERR_PTR(-EBUSY);
 206                         goto out;
 207                 }
 208
 209                 if (!refcount_inc_not_zero(&psock->refcnt))
 210                         psock = ERR_PTR(-EBUSY);
 211         }
 212 out:
 213         rcu_read_unlock();
 214         return psock;
 215 }
 216
 217 static int sock_map_link(struct bpf_map *map, struct sock *sk)
 218 {
 219         struct sk_psock_progs *progs = sock_map_progs(map);
 220         struct bpf_prog *stream_verdict = NULL;
 221         struct bpf_prog *stream_parser = NULL;
 222         struct bpf_prog *skb_verdict = NULL;
 223         struct bpf_prog *msg_parser = NULL;
 224         struct sk_psock *psock;
 225         int ret;
 226
 227         stream_verdict = READ_ONCE(progs->stream_verdict);
 228         if (stream_verdict) {
 229                 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
 230                 if (IS_ERR(stream_verdict))
 231                         return PTR_ERR(stream_verdict);
 232         }
 233
 234         stream_parser = READ_ONCE(progs->stream_parser);
 235         if (stream_parser) {
 236                 stream_parser = bpf_prog_inc_not_zero(stream_parser);
 237                 if (IS_ERR(stream_parser)) {
 238                         ret = PTR_ERR(stream_parser);
 239                         goto out_put_stream_verdict;
 240                 }
 241         }
 242
 243         msg_parser = READ_ONCE(progs->msg_parser);
 244         if (msg_parser) {
 245                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 246                 if (IS_ERR(msg_parser)) {
 247                         ret = PTR_ERR(msg_parser);
 248                         goto out_put_stream_parser;
 249                 }
 250         }
 251
 252         skb_verdict = READ_ONCE(progs->skb_verdict);
 253         if (skb_verdict) {
 254                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 255                 if (IS_ERR(skb_verdict)) {
 256                         ret = PTR_ERR(skb_verdict);
 257                         goto out_put_msg_parser;
 258                 }
 259         }
 260
 261         psock = sock_map_psock_get_checked(sk);
 262         if (IS_ERR(psock)) {
 263                 ret = PTR_ERR(psock);
 264                 goto out_progs;
 265         }
 266
 267         if (psock) {
 268                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 269                     (stream_parser  && READ_ONCE(psock->progs.stream_parser)) ||
 270                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 271                     (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
 272                     (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 273                     (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
 274                         sk_psock_put(sk, psock);
 275                         ret = -EBUSY;
 276                         goto out_progs;
 277                 }
 278         } else {
 279                 psock = sk_psock_init(sk, map->numa_node);
 280                 if (IS_ERR(psock)) {
 281                         ret = PTR_ERR(psock);
 282                         goto out_progs;
 283                 }
 284         }
 285
 286         if (msg_parser)
 287                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 288         if (stream_parser)
 289                 psock_set_prog(&psock->progs.stream_parser, stream_parser);
 290         if (stream_verdict)
 291                 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
 292         if (skb_verdict)
 293                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 294
 295         /* msg_* and stream_* programs references tracked in psock after this
 296          * point. Reference dec and cleanup will occur through psock destructor
 297          */
 298         ret = sock_map_init_proto(sk, psock);
 299         if (ret < 0) {
 300                 sk_psock_put(sk, psock);
 301                 goto out;
 302         }
 303
 304         write_lock_bh(&sk->sk_callback_lock);
 305         if (stream_parser && stream_verdict && !psock->saved_data_ready) {
 306                 ret = sk_psock_init_strp(sk, psock);
 307                 if (ret) {
 308                         write_unlock_bh(&sk->sk_callback_lock);
 309                         sk_psock_put(sk, psock);
 310                         goto out;
 311                 }
 312                 sk_psock_start_strp(sk, psock);
 313         } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
 314                 sk_psock_start_verdict(sk,psock);
 315         } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
 316                 sk_psock_start_verdict(sk, psock);
 317         }
 318         write_unlock_bh(&sk->sk_callback_lock);
 319         return 0;
 320 out_progs:
 321         if (skb_verdict)
 322                 bpf_prog_put(skb_verdict);
 323 out_put_msg_parser:
 324         if (msg_parser)
 325                 bpf_prog_put(msg_parser);
 326 out_put_stream_parser:
 327         if (stream_parser)
 328                 bpf_prog_put(stream_parser);
 329 out_put_stream_verdict:
 330         if (stream_verdict)
 331                 bpf_prog_put(stream_verdict);
 332 out:
 333         return ret;
 334 }
 335
 336 static void sock_map_free(struct bpf_map *map)
 337 {
 338         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 339         int i;
 340
 341         /* After the sync no updates or deletes will be in-flight so it
 342          * is safe to walk map and remove entries without risking a race
 343          * in EEXIST update case.
 344          */
 345         synchronize_rcu();
 346         for (i = 0; i < stab->map.max_entries; i++) {
 347                 struct sock **psk = &stab->sks[i];
 348                 struct sock *sk;
 349
 350                 sk = xchg(psk, NULL);
 351                 if (sk) {
 352                         sock_hold(sk);
 353                         lock_sock(sk);
 354                         rcu_read_lock();
 355                         sock_map_unref(sk, psk);
 356                         rcu_read_unlock();
 357                         release_sock(sk);
 358                         sock_put(sk);
 359                 }
 360         }
 361
 362         /* wait for psock readers accessing its map link */
 363         synchronize_rcu();
 364
 365         bpf_map_area_free(stab->sks);
 366         kfree(stab);
 367 }
 368
 369 static void sock_map_release_progs(struct bpf_map *map)
 370 {
 371         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 372 }
 373
 374 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 375 {
 376         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 377
 378         WARN_ON_ONCE(!rcu_read_lock_held());
 379
 380         if (unlikely(key >= map->max_entries))
 381                 return NULL;
 382         return READ_ONCE(stab->sks[key]);
 383 }
 384
 385 static void *sock_map_lookup(struct bpf_map *map, void *key)
 386 {
 387         struct sock *sk;
 388
 389         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 390         if (!sk)
 391                 return NULL;
 392         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 393                 return NULL;
 394         return sk;
 395 }
 396
 397 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 398 {
 399         struct sock *sk;
 400
 401         if (map->value_size != sizeof(u64))
 402                 return ERR_PTR(-ENOSPC);
 403
 404         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 405         if (!sk)
 406                 return ERR_PTR(-ENOENT);
 407
 408         __sock_gen_cookie(sk);
 409         return &sk->sk_cookie;
 410 }
 411
 412 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 413                              struct sock **psk)
 414 {
 415         struct sock *sk;
 416         int err = 0;
 417
 418         raw_spin_lock_bh(&stab->lock);
 419         sk = *psk;
 420         if (!sk_test || sk_test == sk)
 421                 sk = xchg(psk, NULL);
 422
 423         if (likely(sk))
 424                 sock_map_unref(sk, psk);
 425         else
 426                 err = -EINVAL;
 427
 428         raw_spin_unlock_bh(&stab->lock);
 429         return err;
 430 }
 431
 432 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 433                                       void *link_raw)
 434 {
 435         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 436
 437         __sock_map_delete(stab, sk, link_raw);
 438 }
 439
 440 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 441 {
 442         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 443         u32 i = *(u32 *)key;
 444         struct sock **psk;
 445
 446         if (unlikely(i >= map->max_entries))
 447                 return -EINVAL;
 448
 449         psk = &stab->sks[i];
 450         return __sock_map_delete(stab, NULL, psk);
 451 }
 452
 453 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 454 {
 455         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 456         u32 i = key ? *(u32 *)key : U32_MAX;
 457         u32 *key_next = next;
 458
 459         if (i == stab->map.max_entries - 1)
 460                 return -ENOENT;
 461         if (i >= stab->map.max_entries)
 462                 *key_next = 0;
 463         else
 464                 *key_next = i + 1;
 465         return 0;
 466 }
 467
 468 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 469                                   struct sock *sk, u64 flags)
 470 {
 471         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 472         struct sk_psock_link *link;
 473         struct sk_psock *psock;
 474         struct sock *osk;
 475         int ret;
 476
 477         WARN_ON_ONCE(!rcu_read_lock_held());
 478         if (unlikely(flags > BPF_EXIST))
 479                 return -EINVAL;
 480         if (unlikely(idx >= map->max_entries))
 481                 return -E2BIG;
 482
 483         link = sk_psock_init_link();
 484         if (!link)
 485                 return -ENOMEM;
 486
 487         ret = sock_map_link(map, sk);
 488         if (ret < 0)
 489                 goto out_free;
 490
 491         psock = sk_psock(sk);
 492         WARN_ON_ONCE(!psock);
 493
 494         raw_spin_lock_bh(&stab->lock);
 495         osk = stab->sks[idx];
 496         if (osk && flags == BPF_NOEXIST) {
 497                 ret = -EEXIST;
 498                 goto out_unlock;
 499         } else if (!osk && flags == BPF_EXIST) {
 500                 ret = -ENOENT;
 501                 goto out_unlock;
 502         }
 503
 504         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 505         stab->sks[idx] = sk;
 506         if (osk)
 507                 sock_map_unref(osk, &stab->sks[idx]);
 508         raw_spin_unlock_bh(&stab->lock);
 509         return 0;
 510 out_unlock:
 511         raw_spin_unlock_bh(&stab->lock);
 512         if (psock)
 513                 sk_psock_put(sk, psock);
 514 out_free:
 515         sk_psock_free_link(link);
 516         return ret;
 517 }
 518
 519 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 520 {
 521         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 522                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 523                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 524 }
 525
 526 static bool sk_is_tcp(const struct sock *sk)
 527 {
 528         return sk->sk_type == SOCK_STREAM &&
 529                sk->sk_protocol == IPPROTO_TCP;
 530 }
 531
 532 static bool sock_map_redirect_allowed(const struct sock *sk)
 533 {
 534         if (sk_is_tcp(sk))
 535                 return sk->sk_state != TCP_LISTEN;
 536         else
 537                 return sk->sk_state == TCP_ESTABLISHED;
 538 }
 539
 540 static bool sock_map_sk_is_suitable(const struct sock *sk)
 541 {
 542         return !!sk->sk_prot->psock_update_sk_prot;
 543 }
 544
 545 static bool sock_map_sk_state_allowed(const struct sock *sk)
 546 {
 547         if (sk_is_tcp(sk))
 548                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 549         return true;
 550 }
 551
 552 static int sock_hash_update_common(struct bpf_map *map, void *key,
 553                                    struct sock *sk, u64 flags);
 554
 555 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 556                              u64 flags)
 557 {
 558         struct socket *sock;
 559         struct sock *sk;
 560         int ret;
 561         u64 ufd;
 562
 563         if (map->value_size == sizeof(u64))
 564                 ufd = *(u64 *)value;
 565         else
 566                 ufd = *(u32 *)value;
 567         if (ufd > S32_MAX)
 568                 return -EINVAL;
 569
 570         sock = sockfd_lookup(ufd, &ret);
 571         if (!sock)
 572                 return ret;
 573         sk = sock->sk;
 574         if (!sk) {
 575                 ret = -EINVAL;
 576                 goto out;
 577         }
 578         if (!sock_map_sk_is_suitable(sk)) {
 579                 ret = -EOPNOTSUPP;
 580                 goto out;
 581         }
 582
 583         sock_map_sk_acquire(sk);
 584         if (!sock_map_sk_state_allowed(sk))
 585                 ret = -EOPNOTSUPP;
 586         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 587                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 588         else
 589                 ret = sock_hash_update_common(map, key, sk, flags);
 590         sock_map_sk_release(sk);
 591 out:
 592         sockfd_put(sock);
 593         return ret;
 594 }
 595
 596 static int sock_map_update_elem(struct bpf_map *map, void *key,
 597                                 void *value, u64 flags)
 598 {
 599         struct sock *sk = (struct sock *)value;
 600         int ret;
 601
 602         if (unlikely(!sk || !sk_fullsock(sk)))
 603                 return -EINVAL;
 604
 605         if (!sock_map_sk_is_suitable(sk))
 606                 return -EOPNOTSUPP;
 607
 608         local_bh_disable();
 609         bh_lock_sock(sk);
 610         if (!sock_map_sk_state_allowed(sk))
 611                 ret = -EOPNOTSUPP;
 612         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 613                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 614         else
 615                 ret = sock_hash_update_common(map, key, sk, flags);
 616         bh_unlock_sock(sk);
 617         local_bh_enable();
 618         return ret;
 619 }
 620
 621 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 622            struct bpf_map *, map, void *, key, u64, flags)
 623 {
 624         WARN_ON_ONCE(!rcu_read_lock_held());
 625
 626         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 627                    sock_map_op_okay(sops)))
 628                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 629                                               flags);
 630         return -EOPNOTSUPP;
 631 }
 632
 633 const struct bpf_func_proto bpf_sock_map_update_proto = {
 634         .func           = bpf_sock_map_update,
 635         .gpl_only       = false,
 636         .pkt_access     = true,
 637         .ret_type       = RET_INTEGER,
 638         .arg1_type      = ARG_PTR_TO_CTX,
 639         .arg2_type      = ARG_CONST_MAP_PTR,
 640         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 641         .arg4_type      = ARG_ANYTHING,
 642 };
 643
 644 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 645            struct bpf_map *, map, u32, key, u64, flags)
 646 {
 647         struct sock *sk;
 648
 649         if (unlikely(flags & ~(BPF_F_INGRESS)))
 650                 return SK_DROP;
 651
 652         sk = __sock_map_lookup_elem(map, key);
 653         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 654                 return SK_DROP;
 655
 656         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 657         return SK_PASS;
 658 }
 659
 660 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 661         .func           = bpf_sk_redirect_map,
 662         .gpl_only       = false,
 663         .ret_type       = RET_INTEGER,
 664         .arg1_type      = ARG_PTR_TO_CTX,
 665         .arg2_type      = ARG_CONST_MAP_PTR,
 666         .arg3_type      = ARG_ANYTHING,
 667         .arg4_type      = ARG_ANYTHING,
 668 };
 669
 670 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 671            struct bpf_map *, map, u32, key, u64, flags)
 672 {
 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         msg->flags = flags;
 683         msg->sk_redir = sk;
 684         return SK_PASS;
 685 }
 686
 687 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 688         .func           = bpf_msg_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 struct sock_map_seq_info {
 698         struct bpf_map *map;
 699         struct sock *sk;
 700         u32 index;
 701 };
 702
 703 struct bpf_iter__sockmap {
 704         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 705         __bpf_md_ptr(struct bpf_map *, map);
 706         __bpf_md_ptr(void *, key);
 707         __bpf_md_ptr(struct sock *, sk);
 708 };
 709
 710 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 711                      struct bpf_map *map, void *key,
 712                      struct sock *sk)
 713
 714 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 715 {
 716         if (unlikely(info->index >= info->map->max_entries))
 717                 return NULL;
 718
 719         info->sk = __sock_map_lookup_elem(info->map, info->index);
 720
 721         /* can't return sk directly, since that might be NULL */
 722         return info;
 723 }
 724
 725 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 726         __acquires(rcu)
 727 {
 728         struct sock_map_seq_info *info = seq->private;
 729
 730         if (*pos == 0)
 731                 ++*pos;
 732
 733         /* pairs with sock_map_seq_stop */
 734         rcu_read_lock();
 735         return sock_map_seq_lookup_elem(info);
 736 }
 737
 738 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 739         __must_hold(rcu)
 740 {
 741         struct sock_map_seq_info *info = seq->private;
 742
 743         ++*pos;
 744         ++info->index;
 745
 746         return sock_map_seq_lookup_elem(info);
 747 }
 748
 749 static int sock_map_seq_show(struct seq_file *seq, void *v)
 750         __must_hold(rcu)
 751 {
 752         struct sock_map_seq_info *info = seq->private;
 753         struct bpf_iter__sockmap ctx = {};
 754         struct bpf_iter_meta meta;
 755         struct bpf_prog *prog;
 756
 757         meta.seq = seq;
 758         prog = bpf_iter_get_info(&meta, !v);
 759         if (!prog)
 760                 return 0;
 761
 762         ctx.meta = &meta;
 763         ctx.map = info->map;
 764         if (v) {
 765                 ctx.key = &info->index;
 766                 ctx.sk = info->sk;
 767         }
 768
 769         return bpf_iter_run_prog(prog, &ctx);
 770 }
 771
 772 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 773         __releases(rcu)
 774 {
 775         if (!v)
 776                 (void)sock_map_seq_show(seq, NULL);
 777
 778         /* pairs with sock_map_seq_start */
 779         rcu_read_unlock();
 780 }
 781
 782 static const struct seq_operations sock_map_seq_ops = {
 783         .start  = sock_map_seq_start,
 784         .next   = sock_map_seq_next,
 785         .stop   = sock_map_seq_stop,
 786         .show   = sock_map_seq_show,
 787 };
 788
 789 static int sock_map_init_seq_private(void *priv_data,
 790                                      struct bpf_iter_aux_info *aux)
 791 {
 792         struct sock_map_seq_info *info = priv_data;
 793
 794         bpf_map_inc_with_uref(aux->map);
 795         info->map = aux->map;
 796         return 0;
 797 }
 798
 799 static void sock_map_fini_seq_private(void *priv_data)
 800 {
 801         struct sock_map_seq_info *info = priv_data;
 802
 803         bpf_map_put_with_uref(info->map);
 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 static int sock_map_btf_id;
 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_btf_name           = "bpf_stab",
 826         .map_btf_id             = &sock_map_btf_id,
 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 int 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 (!capable(CAP_NET_ADMIN))
1087                 return ERR_PTR(-EPERM);
1088         if (attr->max_entries == 0 ||
1089             attr->key_size    == 0 ||
1090             (attr->value_size != sizeof(u32) &&
1091              attr->value_size != sizeof(u64)) ||
1092             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1093                 return ERR_PTR(-EINVAL);
1094         if (attr->key_size > MAX_BPF_STACK)
1095                 return ERR_PTR(-E2BIG);
1096
1097         htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1098         if (!htab)
1099                 return ERR_PTR(-ENOMEM);
1100
1101         bpf_map_init_from_attr(&htab->map, attr);
1102
1103         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1104         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1105                           round_up(htab->map.key_size, 8);
1106         if (htab->buckets_num == 0 ||
1107             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1108                 err = -EINVAL;
1109                 goto free_htab;
1110         }
1111
1112         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1113                                            sizeof(struct bpf_shtab_bucket),
1114                                            htab->map.numa_node);
1115         if (!htab->buckets) {
1116                 err = -ENOMEM;
1117                 goto free_htab;
1118         }
1119
1120         for (i = 0; i < htab->buckets_num; i++) {
1121                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1122                 raw_spin_lock_init(&htab->buckets[i].lock);
1123         }
1124
1125         return &htab->map;
1126 free_htab:
1127         kfree(htab);
1128         return ERR_PTR(err);
1129 }
1130
1131 static void sock_hash_free(struct bpf_map *map)
1132 {
1133         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1134         struct bpf_shtab_bucket *bucket;
1135         struct hlist_head unlink_list;
1136         struct bpf_shtab_elem *elem;
1137         struct hlist_node *node;
1138         int i;
1139
1140         /* After the sync no updates or deletes will be in-flight so it
1141          * is safe to walk map and remove entries without risking a race
1142          * in EEXIST update case.
1143          */
1144         synchronize_rcu();
1145         for (i = 0; i < htab->buckets_num; i++) {
1146                 bucket = sock_hash_select_bucket(htab, i);
1147
1148                 /* We are racing with sock_hash_delete_from_link to
1149                  * enter the spin-lock critical section. Every socket on
1150                  * the list is still linked to sockhash. Since link
1151                  * exists, psock exists and holds a ref to socket. That
1152                  * lets us to grab a socket ref too.
1153                  */
1154                 raw_spin_lock_bh(&bucket->lock);
1155                 hlist_for_each_entry(elem, &bucket->head, node)
1156                         sock_hold(elem->sk);
1157                 hlist_move_list(&bucket->head, &unlink_list);
1158                 raw_spin_unlock_bh(&bucket->lock);
1159
1160                 /* Process removed entries out of atomic context to
1161                  * block for socket lock before deleting the psock's
1162                  * link to sockhash.
1163                  */
1164                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1165                         hlist_del(&elem->node);
1166                         lock_sock(elem->sk);
1167                         rcu_read_lock();
1168                         sock_map_unref(elem->sk, elem);
1169                         rcu_read_unlock();
1170                         release_sock(elem->sk);
1171                         sock_put(elem->sk);
1172                         sock_hash_free_elem(htab, elem);
1173                 }
1174         }
1175
1176         /* wait for psock readers accessing its map link */
1177         synchronize_rcu();
1178
1179         bpf_map_area_free(htab->buckets);
1180         kfree(htab);
1181 }
1182
1183 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1184 {
1185         struct sock *sk;
1186
1187         if (map->value_size != sizeof(u64))
1188                 return ERR_PTR(-ENOSPC);
1189
1190         sk = __sock_hash_lookup_elem(map, key);
1191         if (!sk)
1192                 return ERR_PTR(-ENOENT);
1193
1194         __sock_gen_cookie(sk);
1195         return &sk->sk_cookie;
1196 }
1197
1198 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1199 {
1200         struct sock *sk;
1201
1202         sk = __sock_hash_lookup_elem(map, key);
1203         if (!sk)
1204                 return NULL;
1205         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1206                 return NULL;
1207         return sk;
1208 }
1209
1210 static void sock_hash_release_progs(struct bpf_map *map)
1211 {
1212         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1213 }
1214
1215 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1216            struct bpf_map *, map, void *, key, u64, flags)
1217 {
1218         WARN_ON_ONCE(!rcu_read_lock_held());
1219
1220         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1221                    sock_map_op_okay(sops)))
1222                 return sock_hash_update_common(map, key, sops->sk, flags);
1223         return -EOPNOTSUPP;
1224 }
1225
1226 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1227         .func           = bpf_sock_hash_update,
1228         .gpl_only       = false,
1229         .pkt_access     = true,
1230         .ret_type       = RET_INTEGER,
1231         .arg1_type      = ARG_PTR_TO_CTX,
1232         .arg2_type      = ARG_CONST_MAP_PTR,
1233         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1234         .arg4_type      = ARG_ANYTHING,
1235 };
1236
1237 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1238            struct bpf_map *, map, void *, key, u64, flags)
1239 {
1240         struct sock *sk;
1241
1242         if (unlikely(flags & ~(BPF_F_INGRESS)))
1243                 return SK_DROP;
1244
1245         sk = __sock_hash_lookup_elem(map, key);
1246         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1247                 return SK_DROP;
1248
1249         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1250         return SK_PASS;
1251 }
1252
1253 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1254         .func           = bpf_sk_redirect_hash,
1255         .gpl_only       = false,
1256         .ret_type       = RET_INTEGER,
1257         .arg1_type      = ARG_PTR_TO_CTX,
1258         .arg2_type      = ARG_CONST_MAP_PTR,
1259         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1260         .arg4_type      = ARG_ANYTHING,
1261 };
1262
1263 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1264            struct bpf_map *, map, void *, key, u64, flags)
1265 {
1266         struct sock *sk;
1267
1268         if (unlikely(flags & ~(BPF_F_INGRESS)))
1269                 return SK_DROP;
1270
1271         sk = __sock_hash_lookup_elem(map, key);
1272         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1273                 return SK_DROP;
1274
1275         msg->flags = flags;
1276         msg->sk_redir = sk;
1277         return SK_PASS;
1278 }
1279
1280 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1281         .func           = bpf_msg_redirect_hash,
1282         .gpl_only       = false,
1283         .ret_type       = RET_INTEGER,
1284         .arg1_type      = ARG_PTR_TO_CTX,
1285         .arg2_type      = ARG_CONST_MAP_PTR,
1286         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1287         .arg4_type      = ARG_ANYTHING,
1288 };
1289
1290 struct sock_hash_seq_info {
1291         struct bpf_map *map;
1292         struct bpf_shtab *htab;
1293         u32 bucket_id;
1294 };
1295
1296 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1297                                      struct bpf_shtab_elem *prev_elem)
1298 {
1299         const struct bpf_shtab *htab = info->htab;
1300         struct bpf_shtab_bucket *bucket;
1301         struct bpf_shtab_elem *elem;
1302         struct hlist_node *node;
1303
1304         /* try to find next elem in the same bucket */
1305         if (prev_elem) {
1306                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1307                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1308                 if (elem)
1309                         return elem;
1310
1311                 /* no more elements, continue in the next bucket */
1312                 info->bucket_id++;
1313         }
1314
1315         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1316                 bucket = &htab->buckets[info->bucket_id];
1317                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1318                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1319                 if (elem)
1320                         return elem;
1321         }
1322
1323         return NULL;
1324 }
1325
1326 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1327         __acquires(rcu)
1328 {
1329         struct sock_hash_seq_info *info = seq->private;
1330
1331         if (*pos == 0)
1332                 ++*pos;
1333
1334         /* pairs with sock_hash_seq_stop */
1335         rcu_read_lock();
1336         return sock_hash_seq_find_next(info, NULL);
1337 }
1338
1339 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1340         __must_hold(rcu)
1341 {
1342         struct sock_hash_seq_info *info = seq->private;
1343
1344         ++*pos;
1345         return sock_hash_seq_find_next(info, v);
1346 }
1347
1348 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1349         __must_hold(rcu)
1350 {
1351         struct sock_hash_seq_info *info = seq->private;
1352         struct bpf_iter__sockmap ctx = {};
1353         struct bpf_shtab_elem *elem = v;
1354         struct bpf_iter_meta meta;
1355         struct bpf_prog *prog;
1356
1357         meta.seq = seq;
1358         prog = bpf_iter_get_info(&meta, !elem);
1359         if (!prog)
1360                 return 0;
1361
1362         ctx.meta = &meta;
1363         ctx.map = info->map;
1364         if (elem) {
1365                 ctx.key = elem->key;
1366                 ctx.sk = elem->sk;
1367         }
1368
1369         return bpf_iter_run_prog(prog, &ctx);
1370 }
1371
1372 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1373         __releases(rcu)
1374 {
1375         if (!v)
1376                 (void)sock_hash_seq_show(seq, NULL);
1377
1378         /* pairs with sock_hash_seq_start */
1379         rcu_read_unlock();
1380 }
1381
1382 static const struct seq_operations sock_hash_seq_ops = {
1383         .start  = sock_hash_seq_start,
1384         .next   = sock_hash_seq_next,
1385         .stop   = sock_hash_seq_stop,
1386         .show   = sock_hash_seq_show,
1387 };
1388
1389 static int sock_hash_init_seq_private(void *priv_data,
1390                                       struct bpf_iter_aux_info *aux)
1391 {
1392         struct sock_hash_seq_info *info = priv_data;
1393
1394         bpf_map_inc_with_uref(aux->map);
1395         info->map = aux->map;
1396         info->htab = container_of(aux->map, struct bpf_shtab, map);
1397         return 0;
1398 }
1399
1400 static void sock_hash_fini_seq_private(void *priv_data)
1401 {
1402         struct sock_hash_seq_info *info = priv_data;
1403
1404         bpf_map_put_with_uref(info->map);
1405 }
1406
1407 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1408         .seq_ops                = &sock_hash_seq_ops,
1409         .init_seq_private       = sock_hash_init_seq_private,
1410         .fini_seq_private       = sock_hash_fini_seq_private,
1411         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1412 };
1413
1414 static int sock_hash_map_btf_id;
1415 const struct bpf_map_ops sock_hash_ops = {
1416         .map_meta_equal         = bpf_map_meta_equal,
1417         .map_alloc              = sock_hash_alloc,
1418         .map_free               = sock_hash_free,
1419         .map_get_next_key       = sock_hash_get_next_key,
1420         .map_update_elem        = sock_map_update_elem,
1421         .map_delete_elem        = sock_hash_delete_elem,
1422         .map_lookup_elem        = sock_hash_lookup,
1423         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1424         .map_release_uref       = sock_hash_release_progs,
1425         .map_check_btf          = map_check_no_btf,
1426         .map_btf_name           = "bpf_shtab",
1427         .map_btf_id             = &sock_hash_map_btf_id,
1428         .iter_seq_info          = &sock_hash_iter_seq_info,
1429 };
1430
1431 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1432 {
1433         switch (map->map_type) {
1434         case BPF_MAP_TYPE_SOCKMAP:
1435                 return &container_of(map, struct bpf_stab, map)->progs;
1436         case BPF_MAP_TYPE_SOCKHASH:
1437                 return &container_of(map, struct bpf_shtab, map)->progs;
1438         default:
1439                 break;
1440         }
1441
1442         return NULL;
1443 }
1444
1445 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1446                                 struct bpf_prog *old, u32 which)
1447 {
1448         struct sk_psock_progs *progs = sock_map_progs(map);
1449         struct bpf_prog **pprog;
1450
1451         if (!progs)
1452                 return -EOPNOTSUPP;
1453
1454         switch (which) {
1455         case BPF_SK_MSG_VERDICT:
1456                 pprog = &progs->msg_parser;
1457                 break;
1458 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1459         case BPF_SK_SKB_STREAM_PARSER:
1460                 pprog = &progs->stream_parser;
1461                 break;
1462 #endif
1463         case BPF_SK_SKB_STREAM_VERDICT:
1464                 if (progs->skb_verdict)
1465                         return -EBUSY;
1466                 pprog = &progs->stream_verdict;
1467                 break;
1468         case BPF_SK_SKB_VERDICT:
1469                 if (progs->stream_verdict)
1470                         return -EBUSY;
1471                 pprog = &progs->skb_verdict;
1472                 break;
1473         default:
1474                 return -EOPNOTSUPP;
1475         }
1476
1477         if (old)
1478                 return psock_replace_prog(pprog, prog, old);
1479
1480         psock_set_prog(pprog, prog);
1481         return 0;
1482 }
1483
1484 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1485 {
1486         switch (link->map->map_type) {
1487         case BPF_MAP_TYPE_SOCKMAP:
1488                 return sock_map_delete_from_link(link->map, sk,
1489                                                  link->link_raw);
1490         case BPF_MAP_TYPE_SOCKHASH:
1491                 return sock_hash_delete_from_link(link->map, sk,
1492                                                   link->link_raw);
1493         default:
1494                 break;
1495         }
1496 }
1497
1498 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1499 {
1500         struct sk_psock_link *link;
1501
1502         while ((link = sk_psock_link_pop(psock))) {
1503                 sock_map_unlink(sk, link);
1504                 sk_psock_free_link(link);
1505         }
1506 }
1507
1508 void sock_map_unhash(struct sock *sk)
1509 {
1510         void (*saved_unhash)(struct sock *sk);
1511         struct sk_psock *psock;
1512
1513         rcu_read_lock();
1514         psock = sk_psock(sk);
1515         if (unlikely(!psock)) {
1516                 rcu_read_unlock();
1517                 if (sk->sk_prot->unhash)
1518                         sk->sk_prot->unhash(sk);
1519                 return;
1520         }
1521
1522         saved_unhash = psock->saved_unhash;
1523         sock_map_remove_links(sk, psock);
1524         rcu_read_unlock();
1525         saved_unhash(sk);
1526 }
1527 EXPORT_SYMBOL_GPL(sock_map_unhash);
1528
1529 void sock_map_destroy(struct sock *sk)
1530 {
1531         void (*saved_destroy)(struct sock *sk);
1532         struct sk_psock *psock;
1533
1534         rcu_read_lock();
1535         psock = sk_psock_get(sk);
1536         if (unlikely(!psock)) {
1537                 rcu_read_unlock();
1538                 if (sk->sk_prot->destroy)
1539                         sk->sk_prot->destroy(sk);
1540                 return;
1541         }
1542
1543         saved_destroy = psock->saved_destroy;
1544         sock_map_remove_links(sk, psock);
1545         rcu_read_unlock();
1546         sk_psock_stop(psock);
1547         sk_psock_put(sk, psock);
1548         saved_destroy(sk);
1549 }
1550 EXPORT_SYMBOL_GPL(sock_map_destroy);
1551
1552 void sock_map_close(struct sock *sk, long timeout)
1553 {
1554         void (*saved_close)(struct sock *sk, long timeout);
1555         struct sk_psock *psock;
1556
1557         lock_sock(sk);
1558         rcu_read_lock();
1559         psock = sk_psock_get(sk);
1560         if (unlikely(!psock)) {
1561                 rcu_read_unlock();
1562                 release_sock(sk);
1563                 return sk->sk_prot->close(sk, timeout);
1564         }
1565
1566         saved_close = psock->saved_close;
1567         sock_map_remove_links(sk, psock);
1568         rcu_read_unlock();
1569         sk_psock_stop(psock);
1570         release_sock(sk);
1571         cancel_work_sync(&psock->work);
1572         sk_psock_put(sk, psock);
1573         saved_close(sk, timeout);
1574 }
1575 EXPORT_SYMBOL_GPL(sock_map_close);
1576
1577 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1578                                        union bpf_iter_link_info *linfo,
1579                                        struct bpf_iter_aux_info *aux)
1580 {
1581         struct bpf_map *map;
1582         int err = -EINVAL;
1583
1584         if (!linfo->map.map_fd)
1585                 return -EBADF;
1586
1587         map = bpf_map_get_with_uref(linfo->map.map_fd);
1588         if (IS_ERR(map))
1589                 return PTR_ERR(map);
1590
1591         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1592             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1593                 goto put_map;
1594
1595         if (prog->aux->max_rdonly_access > map->key_size) {
1596                 err = -EACCES;
1597                 goto put_map;
1598         }
1599
1600         aux->map = map;
1601         return 0;
1602
1603 put_map:
1604         bpf_map_put_with_uref(map);
1605         return err;
1606 }
1607
1608 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1609 {
1610         bpf_map_put_with_uref(aux->map);
1611 }
1612
1613 static struct bpf_iter_reg sock_map_iter_reg = {
1614         .target                 = "sockmap",
1615         .attach_target          = sock_map_iter_attach_target,
1616         .detach_target          = sock_map_iter_detach_target,
1617         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1618         .fill_link_info         = bpf_iter_map_fill_link_info,
1619         .ctx_arg_info_size      = 2,
1620         .ctx_arg_info           = {
1621                 { offsetof(struct bpf_iter__sockmap, key),
1622                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1623                 { offsetof(struct bpf_iter__sockmap, sk),
1624                   PTR_TO_BTF_ID_OR_NULL },
1625         },
1626 };
1627
1628 static int __init bpf_sockmap_iter_init(void)
1629 {
1630         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1631                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1632         return bpf_iter_reg_target(&sock_map_iter_reg);
1633 }
1634 late_initcall(bpf_sockmap_iter_init);