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 = 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         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         bpf_map_area_free(stab);
 365 }
 366
 367 static void sock_map_release_progs(struct bpf_map *map)
 368 {
 369         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 370 }
 371
 372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 373 {
 374         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 375
 376         WARN_ON_ONCE(!rcu_read_lock_held());
 377
 378         if (unlikely(key >= map->max_entries))
 379                 return NULL;
 380         return READ_ONCE(stab->sks[key]);
 381 }
 382
 383 static void *sock_map_lookup(struct bpf_map *map, void *key)
 384 {
 385         struct sock *sk;
 386
 387         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 388         if (!sk)
 389                 return NULL;
 390         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 391                 return NULL;
 392         return sk;
 393 }
 394
 395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 396 {
 397         struct sock *sk;
 398
 399         if (map->value_size != sizeof(u64))
 400                 return ERR_PTR(-ENOSPC);
 401
 402         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 403         if (!sk)
 404                 return ERR_PTR(-ENOENT);
 405
 406         __sock_gen_cookie(sk);
 407         return &sk->sk_cookie;
 408 }
 409
 410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 411                              struct sock **psk)
 412 {
 413         struct sock *sk;
 414         int err = 0;
 415
 416         raw_spin_lock_bh(&stab->lock);
 417         sk = *psk;
 418         if (!sk_test || sk_test == sk)
 419                 sk = xchg(psk, NULL);
 420
 421         if (likely(sk))
 422                 sock_map_unref(sk, psk);
 423         else
 424                 err = -EINVAL;
 425
 426         raw_spin_unlock_bh(&stab->lock);
 427         return err;
 428 }
 429
 430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 431                                       void *link_raw)
 432 {
 433         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 434
 435         __sock_map_delete(stab, sk, link_raw);
 436 }
 437
 438 static 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 sock_map_redirect_allowed(const struct sock *sk)
 525 {
 526         if (sk_is_tcp(sk))
 527                 return sk->sk_state != TCP_LISTEN;
 528         else
 529                 return sk->sk_state == TCP_ESTABLISHED;
 530 }
 531
 532 static bool sock_map_sk_is_suitable(const struct sock *sk)
 533 {
 534         return !!sk->sk_prot->psock_update_sk_prot;
 535 }
 536
 537 static bool sock_map_sk_state_allowed(const struct sock *sk)
 538 {
 539         if (sk_is_tcp(sk))
 540                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 541         return true;
 542 }
 543
 544 static int sock_hash_update_common(struct bpf_map *map, void *key,
 545                                    struct sock *sk, u64 flags);
 546
 547 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 548                              u64 flags)
 549 {
 550         struct socket *sock;
 551         struct sock *sk;
 552         int ret;
 553         u64 ufd;
 554
 555         if (map->value_size == sizeof(u64))
 556                 ufd = *(u64 *)value;
 557         else
 558                 ufd = *(u32 *)value;
 559         if (ufd > S32_MAX)
 560                 return -EINVAL;
 561
 562         sock = sockfd_lookup(ufd, &ret);
 563         if (!sock)
 564                 return ret;
 565         sk = sock->sk;
 566         if (!sk) {
 567                 ret = -EINVAL;
 568                 goto out;
 569         }
 570         if (!sock_map_sk_is_suitable(sk)) {
 571                 ret = -EOPNOTSUPP;
 572                 goto out;
 573         }
 574
 575         sock_map_sk_acquire(sk);
 576         if (!sock_map_sk_state_allowed(sk))
 577                 ret = -EOPNOTSUPP;
 578         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 579                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 580         else
 581                 ret = sock_hash_update_common(map, key, sk, flags);
 582         sock_map_sk_release(sk);
 583 out:
 584         sockfd_put(sock);
 585         return ret;
 586 }
 587
 588 static int sock_map_update_elem(struct bpf_map *map, void *key,
 589                                 void *value, u64 flags)
 590 {
 591         struct sock *sk = (struct sock *)value;
 592         int ret;
 593
 594         if (unlikely(!sk || !sk_fullsock(sk)))
 595                 return -EINVAL;
 596
 597         if (!sock_map_sk_is_suitable(sk))
 598                 return -EOPNOTSUPP;
 599
 600         local_bh_disable();
 601         bh_lock_sock(sk);
 602         if (!sock_map_sk_state_allowed(sk))
 603                 ret = -EOPNOTSUPP;
 604         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 605                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 606         else
 607                 ret = sock_hash_update_common(map, key, sk, flags);
 608         bh_unlock_sock(sk);
 609         local_bh_enable();
 610         return ret;
 611 }
 612
 613 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 614            struct bpf_map *, map, void *, key, u64, flags)
 615 {
 616         WARN_ON_ONCE(!rcu_read_lock_held());
 617
 618         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 619                    sock_map_op_okay(sops)))
 620                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 621                                               flags);
 622         return -EOPNOTSUPP;
 623 }
 624
 625 const struct bpf_func_proto bpf_sock_map_update_proto = {
 626         .func           = bpf_sock_map_update,
 627         .gpl_only       = false,
 628         .pkt_access     = true,
 629         .ret_type       = RET_INTEGER,
 630         .arg1_type      = ARG_PTR_TO_CTX,
 631         .arg2_type      = ARG_CONST_MAP_PTR,
 632         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 633         .arg4_type      = ARG_ANYTHING,
 634 };
 635
 636 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 637            struct bpf_map *, map, u32, key, u64, flags)
 638 {
 639         struct sock *sk;
 640
 641         if (unlikely(flags & ~(BPF_F_INGRESS)))
 642                 return SK_DROP;
 643
 644         sk = __sock_map_lookup_elem(map, key);
 645         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 646                 return SK_DROP;
 647
 648         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 649         return SK_PASS;
 650 }
 651
 652 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 653         .func           = bpf_sk_redirect_map,
 654         .gpl_only       = false,
 655         .ret_type       = RET_INTEGER,
 656         .arg1_type      = ARG_PTR_TO_CTX,
 657         .arg2_type      = ARG_CONST_MAP_PTR,
 658         .arg3_type      = ARG_ANYTHING,
 659         .arg4_type      = ARG_ANYTHING,
 660 };
 661
 662 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 663            struct bpf_map *, map, u32, key, u64, flags)
 664 {
 665         struct sock *sk;
 666
 667         if (unlikely(flags & ~(BPF_F_INGRESS)))
 668                 return SK_DROP;
 669
 670         sk = __sock_map_lookup_elem(map, key);
 671         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 672                 return SK_DROP;
 673
 674         msg->flags = flags;
 675         msg->sk_redir = sk;
 676         return SK_PASS;
 677 }
 678
 679 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 680         .func           = bpf_msg_redirect_map,
 681         .gpl_only       = false,
 682         .ret_type       = RET_INTEGER,
 683         .arg1_type      = ARG_PTR_TO_CTX,
 684         .arg2_type      = ARG_CONST_MAP_PTR,
 685         .arg3_type      = ARG_ANYTHING,
 686         .arg4_type      = ARG_ANYTHING,
 687 };
 688
 689 struct sock_map_seq_info {
 690         struct bpf_map *map;
 691         struct sock *sk;
 692         u32 index;
 693 };
 694
 695 struct bpf_iter__sockmap {
 696         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 697         __bpf_md_ptr(struct bpf_map *, map);
 698         __bpf_md_ptr(void *, key);
 699         __bpf_md_ptr(struct sock *, sk);
 700 };
 701
 702 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 703                      struct bpf_map *map, void *key,
 704                      struct sock *sk)
 705
 706 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 707 {
 708         if (unlikely(info->index >= info->map->max_entries))
 709                 return NULL;
 710
 711         info->sk = __sock_map_lookup_elem(info->map, info->index);
 712
 713         /* can't return sk directly, since that might be NULL */
 714         return info;
 715 }
 716
 717 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 718         __acquires(rcu)
 719 {
 720         struct sock_map_seq_info *info = seq->private;
 721
 722         if (*pos == 0)
 723                 ++*pos;
 724
 725         /* pairs with sock_map_seq_stop */
 726         rcu_read_lock();
 727         return sock_map_seq_lookup_elem(info);
 728 }
 729
 730 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 731         __must_hold(rcu)
 732 {
 733         struct sock_map_seq_info *info = seq->private;
 734
 735         ++*pos;
 736         ++info->index;
 737
 738         return sock_map_seq_lookup_elem(info);
 739 }
 740
 741 static int sock_map_seq_show(struct seq_file *seq, void *v)
 742         __must_hold(rcu)
 743 {
 744         struct sock_map_seq_info *info = seq->private;
 745         struct bpf_iter__sockmap ctx = {};
 746         struct bpf_iter_meta meta;
 747         struct bpf_prog *prog;
 748
 749         meta.seq = seq;
 750         prog = bpf_iter_get_info(&meta, !v);
 751         if (!prog)
 752                 return 0;
 753
 754         ctx.meta = &meta;
 755         ctx.map = info->map;
 756         if (v) {
 757                 ctx.key = &info->index;
 758                 ctx.sk = info->sk;
 759         }
 760
 761         return bpf_iter_run_prog(prog, &ctx);
 762 }
 763
 764 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 765         __releases(rcu)
 766 {
 767         if (!v)
 768                 (void)sock_map_seq_show(seq, NULL);
 769
 770         /* pairs with sock_map_seq_start */
 771         rcu_read_unlock();
 772 }
 773
 774 static const struct seq_operations sock_map_seq_ops = {
 775         .start  = sock_map_seq_start,
 776         .next   = sock_map_seq_next,
 777         .stop   = sock_map_seq_stop,
 778         .show   = sock_map_seq_show,
 779 };
 780
 781 static int sock_map_init_seq_private(void *priv_data,
 782                                      struct bpf_iter_aux_info *aux)
 783 {
 784         struct sock_map_seq_info *info = priv_data;
 785
 786         bpf_map_inc_with_uref(aux->map);
 787         info->map = aux->map;
 788         return 0;
 789 }
 790
 791 static void sock_map_fini_seq_private(void *priv_data)
 792 {
 793         struct sock_map_seq_info *info = priv_data;
 794
 795         bpf_map_put_with_uref(info->map);
 796 }
 797
 798 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 799         .seq_ops                = &sock_map_seq_ops,
 800         .init_seq_private       = sock_map_init_seq_private,
 801         .fini_seq_private       = sock_map_fini_seq_private,
 802         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 803 };
 804
 805 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 806 const struct bpf_map_ops sock_map_ops = {
 807         .map_meta_equal         = bpf_map_meta_equal,
 808         .map_alloc              = sock_map_alloc,
 809         .map_free               = sock_map_free,
 810         .map_get_next_key       = sock_map_get_next_key,
 811         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 812         .map_update_elem        = sock_map_update_elem,
 813         .map_delete_elem        = sock_map_delete_elem,
 814         .map_lookup_elem        = sock_map_lookup,
 815         .map_release_uref       = sock_map_release_progs,
 816         .map_check_btf          = map_check_no_btf,
 817         .map_btf_id             = &sock_map_btf_ids[0],
 818         .iter_seq_info          = &sock_map_iter_seq_info,
 819 };
 820
 821 struct bpf_shtab_elem {
 822         struct rcu_head rcu;
 823         u32 hash;
 824         struct sock *sk;
 825         struct hlist_node node;
 826         u8 key[];
 827 };
 828
 829 struct bpf_shtab_bucket {
 830         struct hlist_head head;
 831         raw_spinlock_t lock;
 832 };
 833
 834 struct bpf_shtab {
 835         struct bpf_map map;
 836         struct bpf_shtab_bucket *buckets;
 837         u32 buckets_num;
 838         u32 elem_size;
 839         struct sk_psock_progs progs;
 840         atomic_t count;
 841 };
 842
 843 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 844 {
 845         return jhash(key, len, 0);
 846 }
 847
 848 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 849                                                         u32 hash)
 850 {
 851         return &htab->buckets[hash & (htab->buckets_num - 1)];
 852 }
 853
 854 static struct bpf_shtab_elem *
 855 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 856                           u32 key_size)
 857 {
 858         struct bpf_shtab_elem *elem;
 859
 860         hlist_for_each_entry_rcu(elem, head, node) {
 861                 if (elem->hash == hash &&
 862                     !memcmp(&elem->key, key, key_size))
 863                         return elem;
 864         }
 865
 866         return NULL;
 867 }
 868
 869 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 870 {
 871         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 872         u32 key_size = map->key_size, hash;
 873         struct bpf_shtab_bucket *bucket;
 874         struct bpf_shtab_elem *elem;
 875
 876         WARN_ON_ONCE(!rcu_read_lock_held());
 877
 878         hash = sock_hash_bucket_hash(key, key_size);
 879         bucket = sock_hash_select_bucket(htab, hash);
 880         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 881
 882         return elem ? elem->sk : NULL;
 883 }
 884
 885 static void sock_hash_free_elem(struct bpf_shtab *htab,
 886                                 struct bpf_shtab_elem *elem)
 887 {
 888         atomic_dec(&htab->count);
 889         kfree_rcu(elem, rcu);
 890 }
 891
 892 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 893                                        void *link_raw)
 894 {
 895         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 896         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 897         struct bpf_shtab_bucket *bucket;
 898
 899         WARN_ON_ONCE(!rcu_read_lock_held());
 900         bucket = sock_hash_select_bucket(htab, elem->hash);
 901
 902         /* elem may be deleted in parallel from the map, but access here
 903          * is okay since it's going away only after RCU grace period.
 904          * However, we need to check whether it's still present.
 905          */
 906         raw_spin_lock_bh(&bucket->lock);
 907         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 908                                                elem->key, map->key_size);
 909         if (elem_probe && elem_probe == elem) {
 910                 hlist_del_rcu(&elem->node);
 911                 sock_map_unref(elem->sk, elem);
 912                 sock_hash_free_elem(htab, elem);
 913         }
 914         raw_spin_unlock_bh(&bucket->lock);
 915 }
 916
 917 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 918 {
 919         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 920         u32 hash, key_size = map->key_size;
 921         struct bpf_shtab_bucket *bucket;
 922         struct bpf_shtab_elem *elem;
 923         int ret = -ENOENT;
 924
 925         hash = sock_hash_bucket_hash(key, key_size);
 926         bucket = sock_hash_select_bucket(htab, hash);
 927
 928         raw_spin_lock_bh(&bucket->lock);
 929         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 930         if (elem) {
 931                 hlist_del_rcu(&elem->node);
 932                 sock_map_unref(elem->sk, elem);
 933                 sock_hash_free_elem(htab, elem);
 934                 ret = 0;
 935         }
 936         raw_spin_unlock_bh(&bucket->lock);
 937         return ret;
 938 }
 939
 940 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 941                                                    void *key, u32 key_size,
 942                                                    u32 hash, struct sock *sk,
 943                                                    struct bpf_shtab_elem *old)
 944 {
 945         struct bpf_shtab_elem *new;
 946
 947         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 948                 if (!old) {
 949                         atomic_dec(&htab->count);
 950                         return ERR_PTR(-E2BIG);
 951                 }
 952         }
 953
 954         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 955                                    GFP_ATOMIC | __GFP_NOWARN,
 956                                    htab->map.numa_node);
 957         if (!new) {
 958                 atomic_dec(&htab->count);
 959                 return ERR_PTR(-ENOMEM);
 960         }
 961         memcpy(new->key, key, key_size);
 962         new->sk = sk;
 963         new->hash = hash;
 964         return new;
 965 }
 966
 967 static int sock_hash_update_common(struct bpf_map *map, void *key,
 968                                    struct sock *sk, u64 flags)
 969 {
 970         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 971         u32 key_size = map->key_size, hash;
 972         struct bpf_shtab_elem *elem, *elem_new;
 973         struct bpf_shtab_bucket *bucket;
 974         struct sk_psock_link *link;
 975         struct sk_psock *psock;
 976         int ret;
 977
 978         WARN_ON_ONCE(!rcu_read_lock_held());
 979         if (unlikely(flags > BPF_EXIST))
 980                 return -EINVAL;
 981
 982         link = sk_psock_init_link();
 983         if (!link)
 984                 return -ENOMEM;
 985
 986         ret = sock_map_link(map, sk);
 987         if (ret < 0)
 988                 goto out_free;
 989
 990         psock = sk_psock(sk);
 991         WARN_ON_ONCE(!psock);
 992
 993         hash = sock_hash_bucket_hash(key, key_size);
 994         bucket = sock_hash_select_bucket(htab, hash);
 995
 996         raw_spin_lock_bh(&bucket->lock);
 997         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 998         if (elem && flags == BPF_NOEXIST) {
 999                 ret = -EEXIST;
1000                 goto out_unlock;
1001         } else if (!elem && flags == BPF_EXIST) {
1002                 ret = -ENOENT;
1003                 goto out_unlock;
1004         }
1005
1006         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1007         if (IS_ERR(elem_new)) {
1008                 ret = PTR_ERR(elem_new);
1009                 goto out_unlock;
1010         }
1011
1012         sock_map_add_link(psock, link, map, elem_new);
1013         /* Add new element to the head of the list, so that
1014          * concurrent search will find it before old elem.
1015          */
1016         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1017         if (elem) {
1018                 hlist_del_rcu(&elem->node);
1019                 sock_map_unref(elem->sk, elem);
1020                 sock_hash_free_elem(htab, elem);
1021         }
1022         raw_spin_unlock_bh(&bucket->lock);
1023         return 0;
1024 out_unlock:
1025         raw_spin_unlock_bh(&bucket->lock);
1026         sk_psock_put(sk, psock);
1027 out_free:
1028         sk_psock_free_link(link);
1029         return ret;
1030 }
1031
1032 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1033                                   void *key_next)
1034 {
1035         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1036         struct bpf_shtab_elem *elem, *elem_next;
1037         u32 hash, key_size = map->key_size;
1038         struct hlist_head *head;
1039         int i = 0;
1040
1041         if (!key)
1042                 goto find_first_elem;
1043         hash = sock_hash_bucket_hash(key, key_size);
1044         head = &sock_hash_select_bucket(htab, hash)->head;
1045         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1046         if (!elem)
1047                 goto find_first_elem;
1048
1049         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1050                                      struct bpf_shtab_elem, node);
1051         if (elem_next) {
1052                 memcpy(key_next, elem_next->key, key_size);
1053                 return 0;
1054         }
1055
1056         i = hash & (htab->buckets_num - 1);
1057         i++;
1058 find_first_elem:
1059         for (; i < htab->buckets_num; i++) {
1060                 head = &sock_hash_select_bucket(htab, i)->head;
1061                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1062                                              struct bpf_shtab_elem, node);
1063                 if (elem_next) {
1064                         memcpy(key_next, elem_next->key, key_size);
1065                         return 0;
1066                 }
1067         }
1068
1069         return -ENOENT;
1070 }
1071
1072 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1073 {
1074         struct bpf_shtab *htab;
1075         int i, err;
1076
1077         if (!capable(CAP_NET_ADMIN))
1078                 return ERR_PTR(-EPERM);
1079         if (attr->max_entries == 0 ||
1080             attr->key_size    == 0 ||
1081             (attr->value_size != sizeof(u32) &&
1082              attr->value_size != sizeof(u64)) ||
1083             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1084                 return ERR_PTR(-EINVAL);
1085         if (attr->key_size > MAX_BPF_STACK)
1086                 return ERR_PTR(-E2BIG);
1087
1088         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1089         if (!htab)
1090                 return ERR_PTR(-ENOMEM);
1091
1092         bpf_map_init_from_attr(&htab->map, attr);
1093
1094         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1095         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1096                           round_up(htab->map.key_size, 8);
1097         if (htab->buckets_num == 0 ||
1098             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1099                 err = -EINVAL;
1100                 goto free_htab;
1101         }
1102
1103         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1104                                            sizeof(struct bpf_shtab_bucket),
1105                                            htab->map.numa_node);
1106         if (!htab->buckets) {
1107                 err = -ENOMEM;
1108                 goto free_htab;
1109         }
1110
1111         for (i = 0; i < htab->buckets_num; i++) {
1112                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1113                 raw_spin_lock_init(&htab->buckets[i].lock);
1114         }
1115
1116         return &htab->map;
1117 free_htab:
1118         bpf_map_area_free(htab);
1119         return ERR_PTR(err);
1120 }
1121
1122 static void sock_hash_free(struct bpf_map *map)
1123 {
1124         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1125         struct bpf_shtab_bucket *bucket;
1126         struct hlist_head unlink_list;
1127         struct bpf_shtab_elem *elem;
1128         struct hlist_node *node;
1129         int i;
1130
1131         /* After the sync no updates or deletes will be in-flight so it
1132          * is safe to walk map and remove entries without risking a race
1133          * in EEXIST update case.
1134          */
1135         synchronize_rcu();
1136         for (i = 0; i < htab->buckets_num; i++) {
1137                 bucket = sock_hash_select_bucket(htab, i);
1138
1139                 /* We are racing with sock_hash_delete_from_link to
1140                  * enter the spin-lock critical section. Every socket on
1141                  * the list is still linked to sockhash. Since link
1142                  * exists, psock exists and holds a ref to socket. That
1143                  * lets us to grab a socket ref too.
1144                  */
1145                 raw_spin_lock_bh(&bucket->lock);
1146                 hlist_for_each_entry(elem, &bucket->head, node)
1147                         sock_hold(elem->sk);
1148                 hlist_move_list(&bucket->head, &unlink_list);
1149                 raw_spin_unlock_bh(&bucket->lock);
1150
1151                 /* Process removed entries out of atomic context to
1152                  * block for socket lock before deleting the psock's
1153                  * link to sockhash.
1154                  */
1155                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1156                         hlist_del(&elem->node);
1157                         lock_sock(elem->sk);
1158                         rcu_read_lock();
1159                         sock_map_unref(elem->sk, elem);
1160                         rcu_read_unlock();
1161                         release_sock(elem->sk);
1162                         sock_put(elem->sk);
1163                         sock_hash_free_elem(htab, elem);
1164                 }
1165         }
1166
1167         /* wait for psock readers accessing its map link */
1168         synchronize_rcu();
1169
1170         bpf_map_area_free(htab->buckets);
1171         bpf_map_area_free(htab);
1172 }
1173
1174 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1175 {
1176         struct sock *sk;
1177
1178         if (map->value_size != sizeof(u64))
1179                 return ERR_PTR(-ENOSPC);
1180
1181         sk = __sock_hash_lookup_elem(map, key);
1182         if (!sk)
1183                 return ERR_PTR(-ENOENT);
1184
1185         __sock_gen_cookie(sk);
1186         return &sk->sk_cookie;
1187 }
1188
1189 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1190 {
1191         struct sock *sk;
1192
1193         sk = __sock_hash_lookup_elem(map, key);
1194         if (!sk)
1195                 return NULL;
1196         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1197                 return NULL;
1198         return sk;
1199 }
1200
1201 static void sock_hash_release_progs(struct bpf_map *map)
1202 {
1203         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1204 }
1205
1206 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1207            struct bpf_map *, map, void *, key, u64, flags)
1208 {
1209         WARN_ON_ONCE(!rcu_read_lock_held());
1210
1211         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1212                    sock_map_op_okay(sops)))
1213                 return sock_hash_update_common(map, key, sops->sk, flags);
1214         return -EOPNOTSUPP;
1215 }
1216
1217 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1218         .func           = bpf_sock_hash_update,
1219         .gpl_only       = false,
1220         .pkt_access     = true,
1221         .ret_type       = RET_INTEGER,
1222         .arg1_type      = ARG_PTR_TO_CTX,
1223         .arg2_type      = ARG_CONST_MAP_PTR,
1224         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1225         .arg4_type      = ARG_ANYTHING,
1226 };
1227
1228 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1229            struct bpf_map *, map, void *, key, u64, flags)
1230 {
1231         struct sock *sk;
1232
1233         if (unlikely(flags & ~(BPF_F_INGRESS)))
1234                 return SK_DROP;
1235
1236         sk = __sock_hash_lookup_elem(map, key);
1237         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1238                 return SK_DROP;
1239
1240         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1241         return SK_PASS;
1242 }
1243
1244 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1245         .func           = bpf_sk_redirect_hash,
1246         .gpl_only       = false,
1247         .ret_type       = RET_INTEGER,
1248         .arg1_type      = ARG_PTR_TO_CTX,
1249         .arg2_type      = ARG_CONST_MAP_PTR,
1250         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1251         .arg4_type      = ARG_ANYTHING,
1252 };
1253
1254 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1255            struct bpf_map *, map, void *, key, u64, flags)
1256 {
1257         struct sock *sk;
1258
1259         if (unlikely(flags & ~(BPF_F_INGRESS)))
1260                 return SK_DROP;
1261
1262         sk = __sock_hash_lookup_elem(map, key);
1263         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1264                 return SK_DROP;
1265
1266         msg->flags = flags;
1267         msg->sk_redir = sk;
1268         return SK_PASS;
1269 }
1270
1271 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1272         .func           = bpf_msg_redirect_hash,
1273         .gpl_only       = false,
1274         .ret_type       = RET_INTEGER,
1275         .arg1_type      = ARG_PTR_TO_CTX,
1276         .arg2_type      = ARG_CONST_MAP_PTR,
1277         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1278         .arg4_type      = ARG_ANYTHING,
1279 };
1280
1281 struct sock_hash_seq_info {
1282         struct bpf_map *map;
1283         struct bpf_shtab *htab;
1284         u32 bucket_id;
1285 };
1286
1287 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1288                                      struct bpf_shtab_elem *prev_elem)
1289 {
1290         const struct bpf_shtab *htab = info->htab;
1291         struct bpf_shtab_bucket *bucket;
1292         struct bpf_shtab_elem *elem;
1293         struct hlist_node *node;
1294
1295         /* try to find next elem in the same bucket */
1296         if (prev_elem) {
1297                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1298                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1299                 if (elem)
1300                         return elem;
1301
1302                 /* no more elements, continue in the next bucket */
1303                 info->bucket_id++;
1304         }
1305
1306         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1307                 bucket = &htab->buckets[info->bucket_id];
1308                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1309                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1310                 if (elem)
1311                         return elem;
1312         }
1313
1314         return NULL;
1315 }
1316
1317 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1318         __acquires(rcu)
1319 {
1320         struct sock_hash_seq_info *info = seq->private;
1321
1322         if (*pos == 0)
1323                 ++*pos;
1324
1325         /* pairs with sock_hash_seq_stop */
1326         rcu_read_lock();
1327         return sock_hash_seq_find_next(info, NULL);
1328 }
1329
1330 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1331         __must_hold(rcu)
1332 {
1333         struct sock_hash_seq_info *info = seq->private;
1334
1335         ++*pos;
1336         return sock_hash_seq_find_next(info, v);
1337 }
1338
1339 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1340         __must_hold(rcu)
1341 {
1342         struct sock_hash_seq_info *info = seq->private;
1343         struct bpf_iter__sockmap ctx = {};
1344         struct bpf_shtab_elem *elem = v;
1345         struct bpf_iter_meta meta;
1346         struct bpf_prog *prog;
1347
1348         meta.seq = seq;
1349         prog = bpf_iter_get_info(&meta, !elem);
1350         if (!prog)
1351                 return 0;
1352
1353         ctx.meta = &meta;
1354         ctx.map = info->map;
1355         if (elem) {
1356                 ctx.key = elem->key;
1357                 ctx.sk = elem->sk;
1358         }
1359
1360         return bpf_iter_run_prog(prog, &ctx);
1361 }
1362
1363 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1364         __releases(rcu)
1365 {
1366         if (!v)
1367                 (void)sock_hash_seq_show(seq, NULL);
1368
1369         /* pairs with sock_hash_seq_start */
1370         rcu_read_unlock();
1371 }
1372
1373 static const struct seq_operations sock_hash_seq_ops = {
1374         .start  = sock_hash_seq_start,
1375         .next   = sock_hash_seq_next,
1376         .stop   = sock_hash_seq_stop,
1377         .show   = sock_hash_seq_show,
1378 };
1379
1380 static int sock_hash_init_seq_private(void *priv_data,
1381                                       struct bpf_iter_aux_info *aux)
1382 {
1383         struct sock_hash_seq_info *info = priv_data;
1384
1385         bpf_map_inc_with_uref(aux->map);
1386         info->map = aux->map;
1387         info->htab = container_of(aux->map, struct bpf_shtab, map);
1388         return 0;
1389 }
1390
1391 static void sock_hash_fini_seq_private(void *priv_data)
1392 {
1393         struct sock_hash_seq_info *info = priv_data;
1394
1395         bpf_map_put_with_uref(info->map);
1396 }
1397
1398 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1399         .seq_ops                = &sock_hash_seq_ops,
1400         .init_seq_private       = sock_hash_init_seq_private,
1401         .fini_seq_private       = sock_hash_fini_seq_private,
1402         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1403 };
1404
1405 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1406 const struct bpf_map_ops sock_hash_ops = {
1407         .map_meta_equal         = bpf_map_meta_equal,
1408         .map_alloc              = sock_hash_alloc,
1409         .map_free               = sock_hash_free,
1410         .map_get_next_key       = sock_hash_get_next_key,
1411         .map_update_elem        = sock_map_update_elem,
1412         .map_delete_elem        = sock_hash_delete_elem,
1413         .map_lookup_elem        = sock_hash_lookup,
1414         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1415         .map_release_uref       = sock_hash_release_progs,
1416         .map_check_btf          = map_check_no_btf,
1417         .map_btf_id             = &sock_hash_map_btf_ids[0],
1418         .iter_seq_info          = &sock_hash_iter_seq_info,
1419 };
1420
1421 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1422 {
1423         switch (map->map_type) {
1424         case BPF_MAP_TYPE_SOCKMAP:
1425                 return &container_of(map, struct bpf_stab, map)->progs;
1426         case BPF_MAP_TYPE_SOCKHASH:
1427                 return &container_of(map, struct bpf_shtab, map)->progs;
1428         default:
1429                 break;
1430         }
1431
1432         return NULL;
1433 }
1434
1435 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1436                                 u32 which)
1437 {
1438         struct sk_psock_progs *progs = sock_map_progs(map);
1439
1440         if (!progs)
1441                 return -EOPNOTSUPP;
1442
1443         switch (which) {
1444         case BPF_SK_MSG_VERDICT:
1445                 *pprog = &progs->msg_parser;
1446                 break;
1447 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1448         case BPF_SK_SKB_STREAM_PARSER:
1449                 *pprog = &progs->stream_parser;
1450                 break;
1451 #endif
1452         case BPF_SK_SKB_STREAM_VERDICT:
1453                 if (progs->skb_verdict)
1454                         return -EBUSY;
1455                 *pprog = &progs->stream_verdict;
1456                 break;
1457         case BPF_SK_SKB_VERDICT:
1458                 if (progs->stream_verdict)
1459                         return -EBUSY;
1460                 *pprog = &progs->skb_verdict;
1461                 break;
1462         default:
1463                 return -EOPNOTSUPP;
1464         }
1465
1466         return 0;
1467 }
1468
1469 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1470                                 struct bpf_prog *old, u32 which)
1471 {
1472         struct bpf_prog **pprog;
1473         int ret;
1474
1475         ret = sock_map_prog_lookup(map, &pprog, which);
1476         if (ret)
1477                 return ret;
1478
1479         if (old)
1480                 return psock_replace_prog(pprog, prog, old);
1481
1482         psock_set_prog(pprog, prog);
1483         return 0;
1484 }
1485
1486 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1487                             union bpf_attr __user *uattr)
1488 {
1489         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1490         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1491         struct bpf_prog **pprog;
1492         struct bpf_prog *prog;
1493         struct bpf_map *map;
1494         struct fd f;
1495         u32 id = 0;
1496         int ret;
1497
1498         if (attr->query.query_flags)
1499                 return -EINVAL;
1500
1501         f = fdget(ufd);
1502         map = __bpf_map_get(f);
1503         if (IS_ERR(map))
1504                 return PTR_ERR(map);
1505
1506         rcu_read_lock();
1507
1508         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1509         if (ret)
1510                 goto end;
1511
1512         prog = *pprog;
1513         prog_cnt = !prog ? 0 : 1;
1514
1515         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1516                 goto end;
1517
1518         /* we do not hold the refcnt, the bpf prog may be released
1519          * asynchronously and the id would be set to 0.
1520          */
1521         id = data_race(prog->aux->id);
1522         if (id == 0)
1523                 prog_cnt = 0;
1524
1525 end:
1526         rcu_read_unlock();
1527
1528         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1529             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1530             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1531                 ret = -EFAULT;
1532
1533         fdput(f);
1534         return ret;
1535 }
1536
1537 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1538 {
1539         switch (link->map->map_type) {
1540         case BPF_MAP_TYPE_SOCKMAP:
1541                 return sock_map_delete_from_link(link->map, sk,
1542                                                  link->link_raw);
1543         case BPF_MAP_TYPE_SOCKHASH:
1544                 return sock_hash_delete_from_link(link->map, sk,
1545                                                   link->link_raw);
1546         default:
1547                 break;
1548         }
1549 }
1550
1551 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1552 {
1553         struct sk_psock_link *link;
1554
1555         while ((link = sk_psock_link_pop(psock))) {
1556                 sock_map_unlink(sk, link);
1557                 sk_psock_free_link(link);
1558         }
1559 }
1560
1561 void sock_map_unhash(struct sock *sk)
1562 {
1563         void (*saved_unhash)(struct sock *sk);
1564         struct sk_psock *psock;
1565
1566         rcu_read_lock();
1567         psock = sk_psock(sk);
1568         if (unlikely(!psock)) {
1569                 rcu_read_unlock();
1570                 if (sk->sk_prot->unhash)
1571                         sk->sk_prot->unhash(sk);
1572                 return;
1573         }
1574
1575         saved_unhash = psock->saved_unhash;
1576         sock_map_remove_links(sk, psock);
1577         rcu_read_unlock();
1578         saved_unhash(sk);
1579 }
1580 EXPORT_SYMBOL_GPL(sock_map_unhash);
1581
1582 void sock_map_destroy(struct sock *sk)
1583 {
1584         void (*saved_destroy)(struct sock *sk);
1585         struct sk_psock *psock;
1586
1587         rcu_read_lock();
1588         psock = sk_psock_get(sk);
1589         if (unlikely(!psock)) {
1590                 rcu_read_unlock();
1591                 if (sk->sk_prot->destroy)
1592                         sk->sk_prot->destroy(sk);
1593                 return;
1594         }
1595
1596         saved_destroy = psock->saved_destroy;
1597         sock_map_remove_links(sk, psock);
1598         rcu_read_unlock();
1599         sk_psock_stop(psock, false);
1600         sk_psock_put(sk, psock);
1601         saved_destroy(sk);
1602 }
1603 EXPORT_SYMBOL_GPL(sock_map_destroy);
1604
1605 void sock_map_close(struct sock *sk, long timeout)
1606 {
1607         void (*saved_close)(struct sock *sk, long timeout);
1608         struct sk_psock *psock;
1609
1610         lock_sock(sk);
1611         rcu_read_lock();
1612         psock = sk_psock_get(sk);
1613         if (unlikely(!psock)) {
1614                 rcu_read_unlock();
1615                 release_sock(sk);
1616                 return sk->sk_prot->close(sk, timeout);
1617         }
1618
1619         saved_close = psock->saved_close;
1620         sock_map_remove_links(sk, psock);
1621         rcu_read_unlock();
1622         sk_psock_stop(psock, true);
1623         sk_psock_put(sk, psock);
1624         release_sock(sk);
1625         saved_close(sk, timeout);
1626 }
1627 EXPORT_SYMBOL_GPL(sock_map_close);
1628
1629 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1630                                        union bpf_iter_link_info *linfo,
1631                                        struct bpf_iter_aux_info *aux)
1632 {
1633         struct bpf_map *map;
1634         int err = -EINVAL;
1635
1636         if (!linfo->map.map_fd)
1637                 return -EBADF;
1638
1639         map = bpf_map_get_with_uref(linfo->map.map_fd);
1640         if (IS_ERR(map))
1641                 return PTR_ERR(map);
1642
1643         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1644             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1645                 goto put_map;
1646
1647         if (prog->aux->max_rdonly_access > map->key_size) {
1648                 err = -EACCES;
1649                 goto put_map;
1650         }
1651
1652         aux->map = map;
1653         return 0;
1654
1655 put_map:
1656         bpf_map_put_with_uref(map);
1657         return err;
1658 }
1659
1660 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1661 {
1662         bpf_map_put_with_uref(aux->map);
1663 }
1664
1665 static struct bpf_iter_reg sock_map_iter_reg = {
1666         .target                 = "sockmap",
1667         .attach_target          = sock_map_iter_attach_target,
1668         .detach_target          = sock_map_iter_detach_target,
1669         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1670         .fill_link_info         = bpf_iter_map_fill_link_info,
1671         .ctx_arg_info_size      = 2,
1672         .ctx_arg_info           = {
1673                 { offsetof(struct bpf_iter__sockmap, key),
1674                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1675                 { offsetof(struct bpf_iter__sockmap, sk),
1676                   PTR_TO_BTF_ID_OR_NULL },
1677         },
1678 };
1679
1680 static int __init bpf_sockmap_iter_init(void)
1681 {
1682         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1683                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1684         return bpf_iter_reg_target(&sock_map_iter_reg);
1685 }
1686 late_initcall(bpf_sockmap_iter_init);