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                         lock_sock(sk);
 353                         rcu_read_lock();
 354                         sock_map_unref(sk, psk);
 355                         rcu_read_unlock();
 356                         release_sock(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         kfree(stab);
 365 }
 366
 367 static void sock_map_release_progs(struct bpf_map *map)
 368 {
 369         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 370 }
 371
 372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 373 {
 374         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 375
 376         WARN_ON_ONCE(!rcu_read_lock_held());
 377
 378         if (unlikely(key >= map->max_entries))
 379                 return NULL;
 380         return READ_ONCE(stab->sks[key]);
 381 }
 382
 383 static void *sock_map_lookup(struct bpf_map *map, void *key)
 384 {
 385         struct sock *sk;
 386
 387         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 388         if (!sk)
 389                 return NULL;
 390         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 391                 return NULL;
 392         return sk;
 393 }
 394
 395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 396 {
 397         struct sock *sk;
 398
 399         if (map->value_size != sizeof(u64))
 400                 return ERR_PTR(-ENOSPC);
 401
 402         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 403         if (!sk)
 404                 return ERR_PTR(-ENOENT);
 405
 406         __sock_gen_cookie(sk);
 407         return &sk->sk_cookie;
 408 }
 409
 410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 411                              struct sock **psk)
 412 {
 413         struct sock *sk;
 414         int err = 0;
 415
 416         raw_spin_lock_bh(&stab->lock);
 417         sk = *psk;
 418         if (!sk_test || sk_test == sk)
 419                 sk = xchg(psk, NULL);
 420
 421         if (likely(sk))
 422                 sock_map_unref(sk, psk);
 423         else
 424                 err = -EINVAL;
 425
 426         raw_spin_unlock_bh(&stab->lock);
 427         return err;
 428 }
 429
 430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 431                                       void *link_raw)
 432 {
 433         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 434
 435         __sock_map_delete(stab, sk, link_raw);
 436 }
 437
 438 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 439 {
 440         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 441         u32 i = *(u32 *)key;
 442         struct sock **psk;
 443
 444         if (unlikely(i >= map->max_entries))
 445                 return -EINVAL;
 446
 447         psk = &stab->sks[i];
 448         return __sock_map_delete(stab, NULL, psk);
 449 }
 450
 451 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 452 {
 453         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 454         u32 i = key ? *(u32 *)key : U32_MAX;
 455         u32 *key_next = next;
 456
 457         if (i == stab->map.max_entries - 1)
 458                 return -ENOENT;
 459         if (i >= stab->map.max_entries)
 460                 *key_next = 0;
 461         else
 462                 *key_next = i + 1;
 463         return 0;
 464 }
 465
 466 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 467                                   struct sock *sk, u64 flags)
 468 {
 469         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 470         struct sk_psock_link *link;
 471         struct sk_psock *psock;
 472         struct sock *osk;
 473         int ret;
 474
 475         WARN_ON_ONCE(!rcu_read_lock_held());
 476         if (unlikely(flags > BPF_EXIST))
 477                 return -EINVAL;
 478         if (unlikely(idx >= map->max_entries))
 479                 return -E2BIG;
 480
 481         link = sk_psock_init_link();
 482         if (!link)
 483                 return -ENOMEM;
 484
 485         ret = sock_map_link(map, sk);
 486         if (ret < 0)
 487                 goto out_free;
 488
 489         psock = sk_psock(sk);
 490         WARN_ON_ONCE(!psock);
 491
 492         raw_spin_lock_bh(&stab->lock);
 493         osk = stab->sks[idx];
 494         if (osk && flags == BPF_NOEXIST) {
 495                 ret = -EEXIST;
 496                 goto out_unlock;
 497         } else if (!osk && flags == BPF_EXIST) {
 498                 ret = -ENOENT;
 499                 goto out_unlock;
 500         }
 501
 502         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 503         stab->sks[idx] = sk;
 504         if (osk)
 505                 sock_map_unref(osk, &stab->sks[idx]);
 506         raw_spin_unlock_bh(&stab->lock);
 507         return 0;
 508 out_unlock:
 509         raw_spin_unlock_bh(&stab->lock);
 510         if (psock)
 511                 sk_psock_put(sk, psock);
 512 out_free:
 513         sk_psock_free_link(link);
 514         return ret;
 515 }
 516
 517 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 518 {
 519         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 520                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 521                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 522 }
 523
 524 static bool sk_is_tcp(const struct sock *sk)
 525 {
 526         return sk->sk_type == SOCK_STREAM &&
 527                sk->sk_protocol == IPPROTO_TCP;
 528 }
 529
 530 static bool sock_map_redirect_allowed(const struct sock *sk)
 531 {
 532         if (sk_is_tcp(sk))
 533                 return sk->sk_state != TCP_LISTEN;
 534         else
 535                 return sk->sk_state == TCP_ESTABLISHED;
 536 }
 537
 538 static bool sock_map_sk_is_suitable(const struct sock *sk)
 539 {
 540         return !!sk->sk_prot->psock_update_sk_prot;
 541 }
 542
 543 static bool sock_map_sk_state_allowed(const struct sock *sk)
 544 {
 545         if (sk_is_tcp(sk))
 546                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 547         return true;
 548 }
 549
 550 static int sock_hash_update_common(struct bpf_map *map, void *key,
 551                                    struct sock *sk, u64 flags);
 552
 553 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 554                              u64 flags)
 555 {
 556         struct socket *sock;
 557         struct sock *sk;
 558         int ret;
 559         u64 ufd;
 560
 561         if (map->value_size == sizeof(u64))
 562                 ufd = *(u64 *)value;
 563         else
 564                 ufd = *(u32 *)value;
 565         if (ufd > S32_MAX)
 566                 return -EINVAL;
 567
 568         sock = sockfd_lookup(ufd, &ret);
 569         if (!sock)
 570                 return ret;
 571         sk = sock->sk;
 572         if (!sk) {
 573                 ret = -EINVAL;
 574                 goto out;
 575         }
 576         if (!sock_map_sk_is_suitable(sk)) {
 577                 ret = -EOPNOTSUPP;
 578                 goto out;
 579         }
 580
 581         sock_map_sk_acquire(sk);
 582         if (!sock_map_sk_state_allowed(sk))
 583                 ret = -EOPNOTSUPP;
 584         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 585                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 586         else
 587                 ret = sock_hash_update_common(map, key, sk, flags);
 588         sock_map_sk_release(sk);
 589 out:
 590         sockfd_put(sock);
 591         return ret;
 592 }
 593
 594 static int sock_map_update_elem(struct bpf_map *map, void *key,
 595                                 void *value, u64 flags)
 596 {
 597         struct sock *sk = (struct sock *)value;
 598         int ret;
 599
 600         if (unlikely(!sk || !sk_fullsock(sk)))
 601                 return -EINVAL;
 602
 603         if (!sock_map_sk_is_suitable(sk))
 604                 return -EOPNOTSUPP;
 605
 606         local_bh_disable();
 607         bh_lock_sock(sk);
 608         if (!sock_map_sk_state_allowed(sk))
 609                 ret = -EOPNOTSUPP;
 610         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 611                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 612         else
 613                 ret = sock_hash_update_common(map, key, sk, flags);
 614         bh_unlock_sock(sk);
 615         local_bh_enable();
 616         return ret;
 617 }
 618
 619 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 620            struct bpf_map *, map, void *, key, u64, flags)
 621 {
 622         WARN_ON_ONCE(!rcu_read_lock_held());
 623
 624         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 625                    sock_map_op_okay(sops)))
 626                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 627                                               flags);
 628         return -EOPNOTSUPP;
 629 }
 630
 631 const struct bpf_func_proto bpf_sock_map_update_proto = {
 632         .func           = bpf_sock_map_update,
 633         .gpl_only       = false,
 634         .pkt_access     = true,
 635         .ret_type       = RET_INTEGER,
 636         .arg1_type      = ARG_PTR_TO_CTX,
 637         .arg2_type      = ARG_CONST_MAP_PTR,
 638         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 639         .arg4_type      = ARG_ANYTHING,
 640 };
 641
 642 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 643            struct bpf_map *, map, u32, key, u64, flags)
 644 {
 645         struct sock *sk;
 646
 647         if (unlikely(flags & ~(BPF_F_INGRESS)))
 648                 return SK_DROP;
 649
 650         sk = __sock_map_lookup_elem(map, key);
 651         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 652                 return SK_DROP;
 653
 654         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 655         return SK_PASS;
 656 }
 657
 658 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 659         .func           = bpf_sk_redirect_map,
 660         .gpl_only       = false,
 661         .ret_type       = RET_INTEGER,
 662         .arg1_type      = ARG_PTR_TO_CTX,
 663         .arg2_type      = ARG_CONST_MAP_PTR,
 664         .arg3_type      = ARG_ANYTHING,
 665         .arg4_type      = ARG_ANYTHING,
 666 };
 667
 668 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 669            struct bpf_map *, map, u32, key, u64, flags)
 670 {
 671         struct sock *sk;
 672
 673         if (unlikely(flags & ~(BPF_F_INGRESS)))
 674                 return SK_DROP;
 675
 676         sk = __sock_map_lookup_elem(map, key);
 677         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 678                 return SK_DROP;
 679
 680         msg->flags = flags;
 681         msg->sk_redir = sk;
 682         return SK_PASS;
 683 }
 684
 685 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 686         .func           = bpf_msg_redirect_map,
 687         .gpl_only       = false,
 688         .ret_type       = RET_INTEGER,
 689         .arg1_type      = ARG_PTR_TO_CTX,
 690         .arg2_type      = ARG_CONST_MAP_PTR,
 691         .arg3_type      = ARG_ANYTHING,
 692         .arg4_type      = ARG_ANYTHING,
 693 };
 694
 695 struct sock_map_seq_info {
 696         struct bpf_map *map;
 697         struct sock *sk;
 698         u32 index;
 699 };
 700
 701 struct bpf_iter__sockmap {
 702         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 703         __bpf_md_ptr(struct bpf_map *, map);
 704         __bpf_md_ptr(void *, key);
 705         __bpf_md_ptr(struct sock *, sk);
 706 };
 707
 708 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 709                      struct bpf_map *map, void *key,
 710                      struct sock *sk)
 711
 712 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 713 {
 714         if (unlikely(info->index >= info->map->max_entries))
 715                 return NULL;
 716
 717         info->sk = __sock_map_lookup_elem(info->map, info->index);
 718
 719         /* can't return sk directly, since that might be NULL */
 720         return info;
 721 }
 722
 723 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 724         __acquires(rcu)
 725 {
 726         struct sock_map_seq_info *info = seq->private;
 727
 728         if (*pos == 0)
 729                 ++*pos;
 730
 731         /* pairs with sock_map_seq_stop */
 732         rcu_read_lock();
 733         return sock_map_seq_lookup_elem(info);
 734 }
 735
 736 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 737         __must_hold(rcu)
 738 {
 739         struct sock_map_seq_info *info = seq->private;
 740
 741         ++*pos;
 742         ++info->index;
 743
 744         return sock_map_seq_lookup_elem(info);
 745 }
 746
 747 static int sock_map_seq_show(struct seq_file *seq, void *v)
 748         __must_hold(rcu)
 749 {
 750         struct sock_map_seq_info *info = seq->private;
 751         struct bpf_iter__sockmap ctx = {};
 752         struct bpf_iter_meta meta;
 753         struct bpf_prog *prog;
 754
 755         meta.seq = seq;
 756         prog = bpf_iter_get_info(&meta, !v);
 757         if (!prog)
 758                 return 0;
 759
 760         ctx.meta = &meta;
 761         ctx.map = info->map;
 762         if (v) {
 763                 ctx.key = &info->index;
 764                 ctx.sk = info->sk;
 765         }
 766
 767         return bpf_iter_run_prog(prog, &ctx);
 768 }
 769
 770 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 771         __releases(rcu)
 772 {
 773         if (!v)
 774                 (void)sock_map_seq_show(seq, NULL);
 775
 776         /* pairs with sock_map_seq_start */
 777         rcu_read_unlock();
 778 }
 779
 780 static const struct seq_operations sock_map_seq_ops = {
 781         .start  = sock_map_seq_start,
 782         .next   = sock_map_seq_next,
 783         .stop   = sock_map_seq_stop,
 784         .show   = sock_map_seq_show,
 785 };
 786
 787 static int sock_map_init_seq_private(void *priv_data,
 788                                      struct bpf_iter_aux_info *aux)
 789 {
 790         struct sock_map_seq_info *info = priv_data;
 791
 792         bpf_map_inc_with_uref(aux->map);
 793         info->map = aux->map;
 794         return 0;
 795 }
 796
 797 static void sock_map_fini_seq_private(void *priv_data)
 798 {
 799         struct sock_map_seq_info *info = priv_data;
 800
 801         bpf_map_put_with_uref(info->map);
 802 }
 803
 804 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 805         .seq_ops                = &sock_map_seq_ops,
 806         .init_seq_private       = sock_map_init_seq_private,
 807         .fini_seq_private       = sock_map_fini_seq_private,
 808         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 809 };
 810
 811 static int sock_map_btf_id;
 812 const struct bpf_map_ops sock_map_ops = {
 813         .map_meta_equal         = bpf_map_meta_equal,
 814         .map_alloc              = sock_map_alloc,
 815         .map_free               = sock_map_free,
 816         .map_get_next_key       = sock_map_get_next_key,
 817         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 818         .map_update_elem        = sock_map_update_elem,
 819         .map_delete_elem        = sock_map_delete_elem,
 820         .map_lookup_elem        = sock_map_lookup,
 821         .map_release_uref       = sock_map_release_progs,
 822         .map_check_btf          = map_check_no_btf,
 823         .map_btf_name           = "bpf_stab",
 824         .map_btf_id             = &sock_map_btf_id,
 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         hash = sock_hash_bucket_hash(key, key_size);
 933         bucket = sock_hash_select_bucket(htab, hash);
 934
 935         raw_spin_lock_bh(&bucket->lock);
 936         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 937         if (elem) {
 938                 hlist_del_rcu(&elem->node);
 939                 sock_map_unref(elem->sk, elem);
 940                 sock_hash_free_elem(htab, elem);
 941                 ret = 0;
 942         }
 943         raw_spin_unlock_bh(&bucket->lock);
 944         return ret;
 945 }
 946
 947 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 948                                                    void *key, u32 key_size,
 949                                                    u32 hash, struct sock *sk,
 950                                                    struct bpf_shtab_elem *old)
 951 {
 952         struct bpf_shtab_elem *new;
 953
 954         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 955                 if (!old) {
 956                         atomic_dec(&htab->count);
 957                         return ERR_PTR(-E2BIG);
 958                 }
 959         }
 960
 961         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 962                                    GFP_ATOMIC | __GFP_NOWARN,
 963                                    htab->map.numa_node);
 964         if (!new) {
 965                 atomic_dec(&htab->count);
 966                 return ERR_PTR(-ENOMEM);
 967         }
 968         memcpy(new->key, key, key_size);
 969         new->sk = sk;
 970         new->hash = hash;
 971         return new;
 972 }
 973
 974 static int sock_hash_update_common(struct bpf_map *map, void *key,
 975                                    struct sock *sk, u64 flags)
 976 {
 977         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 978         u32 key_size = map->key_size, hash;
 979         struct bpf_shtab_elem *elem, *elem_new;
 980         struct bpf_shtab_bucket *bucket;
 981         struct sk_psock_link *link;
 982         struct sk_psock *psock;
 983         int ret;
 984
 985         WARN_ON_ONCE(!rcu_read_lock_held());
 986         if (unlikely(flags > BPF_EXIST))
 987                 return -EINVAL;
 988
 989         link = sk_psock_init_link();
 990         if (!link)
 991                 return -ENOMEM;
 992
 993         ret = sock_map_link(map, sk);
 994         if (ret < 0)
 995                 goto out_free;
 996
 997         psock = sk_psock(sk);
 998         WARN_ON_ONCE(!psock);
 999
1000         hash = sock_hash_bucket_hash(key, key_size);
1001         bucket = sock_hash_select_bucket(htab, hash);
1002
1003         raw_spin_lock_bh(&bucket->lock);
1004         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1005         if (elem && flags == BPF_NOEXIST) {
1006                 ret = -EEXIST;
1007                 goto out_unlock;
1008         } else if (!elem && flags == BPF_EXIST) {
1009                 ret = -ENOENT;
1010                 goto out_unlock;
1011         }
1012
1013         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1014         if (IS_ERR(elem_new)) {
1015                 ret = PTR_ERR(elem_new);
1016                 goto out_unlock;
1017         }
1018
1019         sock_map_add_link(psock, link, map, elem_new);
1020         /* Add new element to the head of the list, so that
1021          * concurrent search will find it before old elem.
1022          */
1023         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1024         if (elem) {
1025                 hlist_del_rcu(&elem->node);
1026                 sock_map_unref(elem->sk, elem);
1027                 sock_hash_free_elem(htab, elem);
1028         }
1029         raw_spin_unlock_bh(&bucket->lock);
1030         return 0;
1031 out_unlock:
1032         raw_spin_unlock_bh(&bucket->lock);
1033         sk_psock_put(sk, psock);
1034 out_free:
1035         sk_psock_free_link(link);
1036         return ret;
1037 }
1038
1039 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1040                                   void *key_next)
1041 {
1042         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1043         struct bpf_shtab_elem *elem, *elem_next;
1044         u32 hash, key_size = map->key_size;
1045         struct hlist_head *head;
1046         int i = 0;
1047
1048         if (!key)
1049                 goto find_first_elem;
1050         hash = sock_hash_bucket_hash(key, key_size);
1051         head = &sock_hash_select_bucket(htab, hash)->head;
1052         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1053         if (!elem)
1054                 goto find_first_elem;
1055
1056         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1057                                      struct bpf_shtab_elem, node);
1058         if (elem_next) {
1059                 memcpy(key_next, elem_next->key, key_size);
1060                 return 0;
1061         }
1062
1063         i = hash & (htab->buckets_num - 1);
1064         i++;
1065 find_first_elem:
1066         for (; i < htab->buckets_num; i++) {
1067                 head = &sock_hash_select_bucket(htab, i)->head;
1068                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1069                                              struct bpf_shtab_elem, node);
1070                 if (elem_next) {
1071                         memcpy(key_next, elem_next->key, key_size);
1072                         return 0;
1073                 }
1074         }
1075
1076         return -ENOENT;
1077 }
1078
1079 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1080 {
1081         struct bpf_shtab *htab;
1082         int i, err;
1083
1084         if (!capable(CAP_NET_ADMIN))
1085                 return ERR_PTR(-EPERM);
1086         if (attr->max_entries == 0 ||
1087             attr->key_size    == 0 ||
1088             (attr->value_size != sizeof(u32) &&
1089              attr->value_size != sizeof(u64)) ||
1090             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1091                 return ERR_PTR(-EINVAL);
1092         if (attr->key_size > MAX_BPF_STACK)
1093                 return ERR_PTR(-E2BIG);
1094
1095         htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1096         if (!htab)
1097                 return ERR_PTR(-ENOMEM);
1098
1099         bpf_map_init_from_attr(&htab->map, attr);
1100
1101         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1102         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1103                           round_up(htab->map.key_size, 8);
1104         if (htab->buckets_num == 0 ||
1105             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1106                 err = -EINVAL;
1107                 goto free_htab;
1108         }
1109
1110         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1111                                            sizeof(struct bpf_shtab_bucket),
1112                                            htab->map.numa_node);
1113         if (!htab->buckets) {
1114                 err = -ENOMEM;
1115                 goto free_htab;
1116         }
1117
1118         for (i = 0; i < htab->buckets_num; i++) {
1119                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1120                 raw_spin_lock_init(&htab->buckets[i].lock);
1121         }
1122
1123         return &htab->map;
1124 free_htab:
1125         kfree(htab);
1126         return ERR_PTR(err);
1127 }
1128
1129 static void sock_hash_free(struct bpf_map *map)
1130 {
1131         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1132         struct bpf_shtab_bucket *bucket;
1133         struct hlist_head unlink_list;
1134         struct bpf_shtab_elem *elem;
1135         struct hlist_node *node;
1136         int i;
1137
1138         /* After the sync no updates or deletes will be in-flight so it
1139          * is safe to walk map and remove entries without risking a race
1140          * in EEXIST update case.
1141          */
1142         synchronize_rcu();
1143         for (i = 0; i < htab->buckets_num; i++) {
1144                 bucket = sock_hash_select_bucket(htab, i);
1145
1146                 /* We are racing with sock_hash_delete_from_link to
1147                  * enter the spin-lock critical section. Every socket on
1148                  * the list is still linked to sockhash. Since link
1149                  * exists, psock exists and holds a ref to socket. That
1150                  * lets us to grab a socket ref too.
1151                  */
1152                 raw_spin_lock_bh(&bucket->lock);
1153                 hlist_for_each_entry(elem, &bucket->head, node)
1154                         sock_hold(elem->sk);
1155                 hlist_move_list(&bucket->head, &unlink_list);
1156                 raw_spin_unlock_bh(&bucket->lock);
1157
1158                 /* Process removed entries out of atomic context to
1159                  * block for socket lock before deleting the psock's
1160                  * link to sockhash.
1161                  */
1162                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1163                         hlist_del(&elem->node);
1164                         lock_sock(elem->sk);
1165                         rcu_read_lock();
1166                         sock_map_unref(elem->sk, elem);
1167                         rcu_read_unlock();
1168                         release_sock(elem->sk);
1169                         sock_put(elem->sk);
1170                         sock_hash_free_elem(htab, elem);
1171                 }
1172         }
1173
1174         /* wait for psock readers accessing its map link */
1175         synchronize_rcu();
1176
1177         bpf_map_area_free(htab->buckets);
1178         kfree(htab);
1179 }
1180
1181 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1182 {
1183         struct sock *sk;
1184
1185         if (map->value_size != sizeof(u64))
1186                 return ERR_PTR(-ENOSPC);
1187
1188         sk = __sock_hash_lookup_elem(map, key);
1189         if (!sk)
1190                 return ERR_PTR(-ENOENT);
1191
1192         __sock_gen_cookie(sk);
1193         return &sk->sk_cookie;
1194 }
1195
1196 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1197 {
1198         struct sock *sk;
1199
1200         sk = __sock_hash_lookup_elem(map, key);
1201         if (!sk)
1202                 return NULL;
1203         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1204                 return NULL;
1205         return sk;
1206 }
1207
1208 static void sock_hash_release_progs(struct bpf_map *map)
1209 {
1210         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1211 }
1212
1213 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1214            struct bpf_map *, map, void *, key, u64, flags)
1215 {
1216         WARN_ON_ONCE(!rcu_read_lock_held());
1217
1218         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1219                    sock_map_op_okay(sops)))
1220                 return sock_hash_update_common(map, key, sops->sk, flags);
1221         return -EOPNOTSUPP;
1222 }
1223
1224 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1225         .func           = bpf_sock_hash_update,
1226         .gpl_only       = false,
1227         .pkt_access     = true,
1228         .ret_type       = RET_INTEGER,
1229         .arg1_type      = ARG_PTR_TO_CTX,
1230         .arg2_type      = ARG_CONST_MAP_PTR,
1231         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1232         .arg4_type      = ARG_ANYTHING,
1233 };
1234
1235 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1236            struct bpf_map *, map, void *, key, u64, flags)
1237 {
1238         struct sock *sk;
1239
1240         if (unlikely(flags & ~(BPF_F_INGRESS)))
1241                 return SK_DROP;
1242
1243         sk = __sock_hash_lookup_elem(map, key);
1244         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1245                 return SK_DROP;
1246
1247         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1248         return SK_PASS;
1249 }
1250
1251 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1252         .func           = bpf_sk_redirect_hash,
1253         .gpl_only       = false,
1254         .ret_type       = RET_INTEGER,
1255         .arg1_type      = ARG_PTR_TO_CTX,
1256         .arg2_type      = ARG_CONST_MAP_PTR,
1257         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1258         .arg4_type      = ARG_ANYTHING,
1259 };
1260
1261 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1262            struct bpf_map *, map, void *, key, u64, flags)
1263 {
1264         struct sock *sk;
1265
1266         if (unlikely(flags & ~(BPF_F_INGRESS)))
1267                 return SK_DROP;
1268
1269         sk = __sock_hash_lookup_elem(map, key);
1270         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1271                 return SK_DROP;
1272
1273         msg->flags = flags;
1274         msg->sk_redir = sk;
1275         return SK_PASS;
1276 }
1277
1278 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1279         .func           = bpf_msg_redirect_hash,
1280         .gpl_only       = false,
1281         .ret_type       = RET_INTEGER,
1282         .arg1_type      = ARG_PTR_TO_CTX,
1283         .arg2_type      = ARG_CONST_MAP_PTR,
1284         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1285         .arg4_type      = ARG_ANYTHING,
1286 };
1287
1288 struct sock_hash_seq_info {
1289         struct bpf_map *map;
1290         struct bpf_shtab *htab;
1291         u32 bucket_id;
1292 };
1293
1294 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1295                                      struct bpf_shtab_elem *prev_elem)
1296 {
1297         const struct bpf_shtab *htab = info->htab;
1298         struct bpf_shtab_bucket *bucket;
1299         struct bpf_shtab_elem *elem;
1300         struct hlist_node *node;
1301
1302         /* try to find next elem in the same bucket */
1303         if (prev_elem) {
1304                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1305                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1306                 if (elem)
1307                         return elem;
1308
1309                 /* no more elements, continue in the next bucket */
1310                 info->bucket_id++;
1311         }
1312
1313         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1314                 bucket = &htab->buckets[info->bucket_id];
1315                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1316                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1317                 if (elem)
1318                         return elem;
1319         }
1320
1321         return NULL;
1322 }
1323
1324 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1325         __acquires(rcu)
1326 {
1327         struct sock_hash_seq_info *info = seq->private;
1328
1329         if (*pos == 0)
1330                 ++*pos;
1331
1332         /* pairs with sock_hash_seq_stop */
1333         rcu_read_lock();
1334         return sock_hash_seq_find_next(info, NULL);
1335 }
1336
1337 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1338         __must_hold(rcu)
1339 {
1340         struct sock_hash_seq_info *info = seq->private;
1341
1342         ++*pos;
1343         return sock_hash_seq_find_next(info, v);
1344 }
1345
1346 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1347         __must_hold(rcu)
1348 {
1349         struct sock_hash_seq_info *info = seq->private;
1350         struct bpf_iter__sockmap ctx = {};
1351         struct bpf_shtab_elem *elem = v;
1352         struct bpf_iter_meta meta;
1353         struct bpf_prog *prog;
1354
1355         meta.seq = seq;
1356         prog = bpf_iter_get_info(&meta, !elem);
1357         if (!prog)
1358                 return 0;
1359
1360         ctx.meta = &meta;
1361         ctx.map = info->map;
1362         if (elem) {
1363                 ctx.key = elem->key;
1364                 ctx.sk = elem->sk;
1365         }
1366
1367         return bpf_iter_run_prog(prog, &ctx);
1368 }
1369
1370 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1371         __releases(rcu)
1372 {
1373         if (!v)
1374                 (void)sock_hash_seq_show(seq, NULL);
1375
1376         /* pairs with sock_hash_seq_start */
1377         rcu_read_unlock();
1378 }
1379
1380 static const struct seq_operations sock_hash_seq_ops = {
1381         .start  = sock_hash_seq_start,
1382         .next   = sock_hash_seq_next,
1383         .stop   = sock_hash_seq_stop,
1384         .show   = sock_hash_seq_show,
1385 };
1386
1387 static int sock_hash_init_seq_private(void *priv_data,
1388                                       struct bpf_iter_aux_info *aux)
1389 {
1390         struct sock_hash_seq_info *info = priv_data;
1391
1392         bpf_map_inc_with_uref(aux->map);
1393         info->map = aux->map;
1394         info->htab = container_of(aux->map, struct bpf_shtab, map);
1395         return 0;
1396 }
1397
1398 static void sock_hash_fini_seq_private(void *priv_data)
1399 {
1400         struct sock_hash_seq_info *info = priv_data;
1401
1402         bpf_map_put_with_uref(info->map);
1403 }
1404
1405 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1406         .seq_ops                = &sock_hash_seq_ops,
1407         .init_seq_private       = sock_hash_init_seq_private,
1408         .fini_seq_private       = sock_hash_fini_seq_private,
1409         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1410 };
1411
1412 static int sock_hash_map_btf_id;
1413 const struct bpf_map_ops sock_hash_ops = {
1414         .map_meta_equal         = bpf_map_meta_equal,
1415         .map_alloc              = sock_hash_alloc,
1416         .map_free               = sock_hash_free,
1417         .map_get_next_key       = sock_hash_get_next_key,
1418         .map_update_elem        = sock_map_update_elem,
1419         .map_delete_elem        = sock_hash_delete_elem,
1420         .map_lookup_elem        = sock_hash_lookup,
1421         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1422         .map_release_uref       = sock_hash_release_progs,
1423         .map_check_btf          = map_check_no_btf,
1424         .map_btf_name           = "bpf_shtab",
1425         .map_btf_id             = &sock_hash_map_btf_id,
1426         .iter_seq_info          = &sock_hash_iter_seq_info,
1427 };
1428
1429 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1430 {
1431         switch (map->map_type) {
1432         case BPF_MAP_TYPE_SOCKMAP:
1433                 return &container_of(map, struct bpf_stab, map)->progs;
1434         case BPF_MAP_TYPE_SOCKHASH:
1435                 return &container_of(map, struct bpf_shtab, map)->progs;
1436         default:
1437                 break;
1438         }
1439
1440         return NULL;
1441 }
1442
1443 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1444                                 struct bpf_prog *old, u32 which)
1445 {
1446         struct sk_psock_progs *progs = sock_map_progs(map);
1447         struct bpf_prog **pprog;
1448
1449         if (!progs)
1450                 return -EOPNOTSUPP;
1451
1452         switch (which) {
1453         case BPF_SK_MSG_VERDICT:
1454                 pprog = &progs->msg_parser;
1455                 break;
1456 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1457         case BPF_SK_SKB_STREAM_PARSER:
1458                 pprog = &progs->stream_parser;
1459                 break;
1460 #endif
1461         case BPF_SK_SKB_STREAM_VERDICT:
1462                 if (progs->skb_verdict)
1463                         return -EBUSY;
1464                 pprog = &progs->stream_verdict;
1465                 break;
1466         case BPF_SK_SKB_VERDICT:
1467                 if (progs->stream_verdict)
1468                         return -EBUSY;
1469                 pprog = &progs->skb_verdict;
1470                 break;
1471         default:
1472                 return -EOPNOTSUPP;
1473         }
1474
1475         if (old)
1476                 return psock_replace_prog(pprog, prog, old);
1477
1478         psock_set_prog(pprog, prog);
1479         return 0;
1480 }
1481
1482 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1483 {
1484         switch (link->map->map_type) {
1485         case BPF_MAP_TYPE_SOCKMAP:
1486                 return sock_map_delete_from_link(link->map, sk,
1487                                                  link->link_raw);
1488         case BPF_MAP_TYPE_SOCKHASH:
1489                 return sock_hash_delete_from_link(link->map, sk,
1490                                                   link->link_raw);
1491         default:
1492                 break;
1493         }
1494 }
1495
1496 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1497 {
1498         struct sk_psock_link *link;
1499
1500         while ((link = sk_psock_link_pop(psock))) {
1501                 sock_map_unlink(sk, link);
1502                 sk_psock_free_link(link);
1503         }
1504 }
1505
1506 void sock_map_unhash(struct sock *sk)
1507 {
1508         void (*saved_unhash)(struct sock *sk);
1509         struct sk_psock *psock;
1510
1511         rcu_read_lock();
1512         psock = sk_psock(sk);
1513         if (unlikely(!psock)) {
1514                 rcu_read_unlock();
1515                 if (sk->sk_prot->unhash)
1516                         sk->sk_prot->unhash(sk);
1517                 return;
1518         }
1519
1520         saved_unhash = psock->saved_unhash;
1521         sock_map_remove_links(sk, psock);
1522         rcu_read_unlock();
1523         saved_unhash(sk);
1524 }
1525 EXPORT_SYMBOL_GPL(sock_map_unhash);
1526
1527 void sock_map_destroy(struct sock *sk)
1528 {
1529         void (*saved_destroy)(struct sock *sk);
1530         struct sk_psock *psock;
1531
1532         rcu_read_lock();
1533         psock = sk_psock_get(sk);
1534         if (unlikely(!psock)) {
1535                 rcu_read_unlock();
1536                 if (sk->sk_prot->destroy)
1537                         sk->sk_prot->destroy(sk);
1538                 return;
1539         }
1540
1541         saved_destroy = psock->saved_destroy;
1542         sock_map_remove_links(sk, psock);
1543         rcu_read_unlock();
1544         sk_psock_stop(psock, true);
1545         sk_psock_put(sk, psock);
1546         saved_destroy(sk);
1547 }
1548 EXPORT_SYMBOL_GPL(sock_map_destroy);
1549
1550 void sock_map_close(struct sock *sk, long timeout)
1551 {
1552         void (*saved_close)(struct sock *sk, long timeout);
1553         struct sk_psock *psock;
1554
1555         lock_sock(sk);
1556         rcu_read_lock();
1557         psock = sk_psock_get(sk);
1558         if (unlikely(!psock)) {
1559                 rcu_read_unlock();
1560                 release_sock(sk);
1561                 return sk->sk_prot->close(sk, timeout);
1562         }
1563
1564         saved_close = psock->saved_close;
1565         sock_map_remove_links(sk, psock);
1566         rcu_read_unlock();
1567         sk_psock_stop(psock, true);
1568         sk_psock_put(sk, psock);
1569         release_sock(sk);
1570         saved_close(sk, timeout);
1571 }
1572 EXPORT_SYMBOL_GPL(sock_map_close);
1573
1574 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1575                                        union bpf_iter_link_info *linfo,
1576                                        struct bpf_iter_aux_info *aux)
1577 {
1578         struct bpf_map *map;
1579         int err = -EINVAL;
1580
1581         if (!linfo->map.map_fd)
1582                 return -EBADF;
1583
1584         map = bpf_map_get_with_uref(linfo->map.map_fd);
1585         if (IS_ERR(map))
1586                 return PTR_ERR(map);
1587
1588         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1589             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1590                 goto put_map;
1591
1592         if (prog->aux->max_rdonly_access > map->key_size) {
1593                 err = -EACCES;
1594                 goto put_map;
1595         }
1596
1597         aux->map = map;
1598         return 0;
1599
1600 put_map:
1601         bpf_map_put_with_uref(map);
1602         return err;
1603 }
1604
1605 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1606 {
1607         bpf_map_put_with_uref(aux->map);
1608 }
1609
1610 static struct bpf_iter_reg sock_map_iter_reg = {
1611         .target                 = "sockmap",
1612         .attach_target          = sock_map_iter_attach_target,
1613         .detach_target          = sock_map_iter_detach_target,
1614         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1615         .fill_link_info         = bpf_iter_map_fill_link_info,
1616         .ctx_arg_info_size      = 2,
1617         .ctx_arg_info           = {
1618                 { offsetof(struct bpf_iter__sockmap, key),
1619                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1620                 { offsetof(struct bpf_iter__sockmap, sk),
1621                   PTR_TO_BTF_ID_OR_NULL },
1622         },
1623 };
1624
1625 static int __init bpf_sockmap_iter_init(void)
1626 {
1627         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1628                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1629         return bpf_iter_reg_target(&sock_map_iter_reg);
1630 }
1631 late_initcall(bpf_sockmap_iter_init);