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