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         spinlock_t lock;
  22 };
  23
  24 #define SOCK_CREATE_FLAG_MASK                           \
  25         (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
  26
  27 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
  28                                 struct bpf_prog *old, u32 which);
  29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
  30
  31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  32 {
  33         struct bpf_stab *stab;
  34
  35         if (attr->max_entries == 0 ||
  36             attr->key_size    != 4 ||
  37             (attr->value_size != sizeof(u32) &&
  38              attr->value_size != sizeof(u64)) ||
  39             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  40                 return ERR_PTR(-EINVAL);
  41
  42         stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
  43         if (!stab)
  44                 return ERR_PTR(-ENOMEM);
  45
  46         bpf_map_init_from_attr(&stab->map, attr);
  47         spin_lock_init(&stab->lock);
  48
  49         stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
  50                                        sizeof(struct sock *),
  51                                        stab->map.numa_node);
  52         if (!stab->sks) {
  53                 bpf_map_area_free(stab);
  54                 return ERR_PTR(-ENOMEM);
  55         }
  56
  57         return &stab->map;
  58 }
  59
  60 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  61 {
  62         u32 ufd = attr->target_fd;
  63         struct bpf_map *map;
  64         struct fd f;
  65         int ret;
  66
  67         if (attr->attach_flags || attr->replace_bpf_fd)
  68                 return -EINVAL;
  69
  70         f = fdget(ufd);
  71         map = __bpf_map_get(f);
  72         if (IS_ERR(map))
  73                 return PTR_ERR(map);
  74         ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
  75         fdput(f);
  76         return ret;
  77 }
  78
  79 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  80 {
  81         u32 ufd = attr->target_fd;
  82         struct bpf_prog *prog;
  83         struct bpf_map *map;
  84         struct fd f;
  85         int ret;
  86
  87         if (attr->attach_flags || attr->replace_bpf_fd)
  88                 return -EINVAL;
  89
  90         f = fdget(ufd);
  91         map = __bpf_map_get(f);
  92         if (IS_ERR(map))
  93                 return PTR_ERR(map);
  94
  95         prog = bpf_prog_get(attr->attach_bpf_fd);
  96         if (IS_ERR(prog)) {
  97                 ret = PTR_ERR(prog);
  98                 goto put_map;
  99         }
 100
 101         if (prog->type != ptype) {
 102                 ret = -EINVAL;
 103                 goto put_prog;
 104         }
 105
 106         ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
 107 put_prog:
 108         bpf_prog_put(prog);
 109 put_map:
 110         fdput(f);
 111         return ret;
 112 }
 113
 114 static void sock_map_sk_acquire(struct sock *sk)
 115         __acquires(&sk->sk_lock.slock)
 116 {
 117         lock_sock(sk);
 118         rcu_read_lock();
 119 }
 120
 121 static void sock_map_sk_release(struct sock *sk)
 122         __releases(&sk->sk_lock.slock)
 123 {
 124         rcu_read_unlock();
 125         release_sock(sk);
 126 }
 127
 128 static void sock_map_add_link(struct sk_psock *psock,
 129                               struct sk_psock_link *link,
 130                               struct bpf_map *map, void *link_raw)
 131 {
 132         link->link_raw = link_raw;
 133         link->map = map;
 134         spin_lock_bh(&psock->link_lock);
 135         list_add_tail(&link->list, &psock->link);
 136         spin_unlock_bh(&psock->link_lock);
 137 }
 138
 139 static void sock_map_del_link(struct sock *sk,
 140                               struct sk_psock *psock, void *link_raw)
 141 {
 142         bool strp_stop = false, verdict_stop = false;
 143         struct sk_psock_link *link, *tmp;
 144
 145         spin_lock_bh(&psock->link_lock);
 146         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 147                 if (link->link_raw == link_raw) {
 148                         struct bpf_map *map = link->map;
 149                         struct sk_psock_progs *progs = sock_map_progs(map);
 150
 151                         if (psock->saved_data_ready && progs->stream_parser)
 152                                 strp_stop = true;
 153                         if (psock->saved_data_ready && progs->stream_verdict)
 154                                 verdict_stop = true;
 155                         if (psock->saved_data_ready && progs->skb_verdict)
 156                                 verdict_stop = true;
 157                         list_del(&link->list);
 158                         sk_psock_free_link(link);
 159                 }
 160         }
 161         spin_unlock_bh(&psock->link_lock);
 162         if (strp_stop || verdict_stop) {
 163                 write_lock_bh(&sk->sk_callback_lock);
 164                 if (strp_stop)
 165                         sk_psock_stop_strp(sk, psock);
 166                 if (verdict_stop)
 167                         sk_psock_stop_verdict(sk, psock);
 168
 169                 if (psock->psock_update_sk_prot)
 170                         psock->psock_update_sk_prot(sk, psock, false);
 171                 write_unlock_bh(&sk->sk_callback_lock);
 172         }
 173 }
 174
 175 static void sock_map_unref(struct sock *sk, void *link_raw)
 176 {
 177         struct sk_psock *psock = sk_psock(sk);
 178
 179         if (likely(psock)) {
 180                 sock_map_del_link(sk, psock, link_raw);
 181                 sk_psock_put(sk, psock);
 182         }
 183 }
 184
 185 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
 186 {
 187         if (!sk->sk_prot->psock_update_sk_prot)
 188                 return -EINVAL;
 189         psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
 190         return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
 191 }
 192
 193 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 194 {
 195         struct sk_psock *psock;
 196
 197         rcu_read_lock();
 198         psock = sk_psock(sk);
 199         if (psock) {
 200                 if (sk->sk_prot->close != sock_map_close) {
 201                         psock = ERR_PTR(-EBUSY);
 202                         goto out;
 203                 }
 204
 205                 if (!refcount_inc_not_zero(&psock->refcnt))
 206                         psock = ERR_PTR(-EBUSY);
 207         }
 208 out:
 209         rcu_read_unlock();
 210         return psock;
 211 }
 212
 213 static int sock_map_link(struct bpf_map *map, struct sock *sk)
 214 {
 215         struct sk_psock_progs *progs = sock_map_progs(map);
 216         struct bpf_prog *stream_verdict = NULL;
 217         struct bpf_prog *stream_parser = NULL;
 218         struct bpf_prog *skb_verdict = NULL;
 219         struct bpf_prog *msg_parser = NULL;
 220         struct sk_psock *psock;
 221         int ret;
 222
 223         stream_verdict = READ_ONCE(progs->stream_verdict);
 224         if (stream_verdict) {
 225                 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
 226                 if (IS_ERR(stream_verdict))
 227                         return PTR_ERR(stream_verdict);
 228         }
 229
 230         stream_parser = READ_ONCE(progs->stream_parser);
 231         if (stream_parser) {
 232                 stream_parser = bpf_prog_inc_not_zero(stream_parser);
 233                 if (IS_ERR(stream_parser)) {
 234                         ret = PTR_ERR(stream_parser);
 235                         goto out_put_stream_verdict;
 236                 }
 237         }
 238
 239         msg_parser = READ_ONCE(progs->msg_parser);
 240         if (msg_parser) {
 241                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 242                 if (IS_ERR(msg_parser)) {
 243                         ret = PTR_ERR(msg_parser);
 244                         goto out_put_stream_parser;
 245                 }
 246         }
 247
 248         skb_verdict = READ_ONCE(progs->skb_verdict);
 249         if (skb_verdict) {
 250                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 251                 if (IS_ERR(skb_verdict)) {
 252                         ret = PTR_ERR(skb_verdict);
 253                         goto out_put_msg_parser;
 254                 }
 255         }
 256
 257         psock = sock_map_psock_get_checked(sk);
 258         if (IS_ERR(psock)) {
 259                 ret = PTR_ERR(psock);
 260                 goto out_progs;
 261         }
 262
 263         if (psock) {
 264                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 265                     (stream_parser  && READ_ONCE(psock->progs.stream_parser)) ||
 266                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 267                     (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
 268                     (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 269                     (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
 270                         sk_psock_put(sk, psock);
 271                         ret = -EBUSY;
 272                         goto out_progs;
 273                 }
 274         } else {
 275                 psock = sk_psock_init(sk, map->numa_node);
 276                 if (IS_ERR(psock)) {
 277                         ret = PTR_ERR(psock);
 278                         goto out_progs;
 279                 }
 280         }
 281
 282         if (msg_parser)
 283                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 284         if (stream_parser)
 285                 psock_set_prog(&psock->progs.stream_parser, stream_parser);
 286         if (stream_verdict)
 287                 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
 288         if (skb_verdict)
 289                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 290
 291         /* msg_* and stream_* programs references tracked in psock after this
 292          * point. Reference dec and cleanup will occur through psock destructor
 293          */
 294         ret = sock_map_init_proto(sk, psock);
 295         if (ret < 0) {
 296                 sk_psock_put(sk, psock);
 297                 goto out;
 298         }
 299
 300         write_lock_bh(&sk->sk_callback_lock);
 301         if (stream_parser && stream_verdict && !psock->saved_data_ready) {
 302                 ret = sk_psock_init_strp(sk, psock);
 303                 if (ret) {
 304                         write_unlock_bh(&sk->sk_callback_lock);
 305                         sk_psock_put(sk, psock);
 306                         goto out;
 307                 }
 308                 sk_psock_start_strp(sk, psock);
 309         } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
 310                 sk_psock_start_verdict(sk,psock);
 311         } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
 312                 sk_psock_start_verdict(sk, psock);
 313         }
 314         write_unlock_bh(&sk->sk_callback_lock);
 315         return 0;
 316 out_progs:
 317         if (skb_verdict)
 318                 bpf_prog_put(skb_verdict);
 319 out_put_msg_parser:
 320         if (msg_parser)
 321                 bpf_prog_put(msg_parser);
 322 out_put_stream_parser:
 323         if (stream_parser)
 324                 bpf_prog_put(stream_parser);
 325 out_put_stream_verdict:
 326         if (stream_verdict)
 327                 bpf_prog_put(stream_verdict);
 328 out:
 329         return ret;
 330 }
 331
 332 static void sock_map_free(struct bpf_map *map)
 333 {
 334         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 335         int i;
 336
 337         /* After the sync no updates or deletes will be in-flight so it
 338          * is safe to walk map and remove entries without risking a race
 339          * in EEXIST update case.
 340          */
 341         synchronize_rcu();
 342         for (i = 0; i < stab->map.max_entries; i++) {
 343                 struct sock **psk = &stab->sks[i];
 344                 struct sock *sk;
 345
 346                 sk = xchg(psk, NULL);
 347                 if (sk) {
 348                         sock_hold(sk);
 349                         lock_sock(sk);
 350                         rcu_read_lock();
 351                         sock_map_unref(sk, psk);
 352                         rcu_read_unlock();
 353                         release_sock(sk);
 354                         sock_put(sk);
 355                 }
 356         }
 357
 358         /* wait for psock readers accessing its map link */
 359         synchronize_rcu();
 360
 361         bpf_map_area_free(stab->sks);
 362         bpf_map_area_free(stab);
 363 }
 364
 365 static void sock_map_release_progs(struct bpf_map *map)
 366 {
 367         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 368 }
 369
 370 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 371 {
 372         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 373
 374         WARN_ON_ONCE(!rcu_read_lock_held());
 375
 376         if (unlikely(key >= map->max_entries))
 377                 return NULL;
 378         return READ_ONCE(stab->sks[key]);
 379 }
 380
 381 static void *sock_map_lookup(struct bpf_map *map, void *key)
 382 {
 383         struct sock *sk;
 384
 385         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 386         if (!sk)
 387                 return NULL;
 388         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 389                 return NULL;
 390         return sk;
 391 }
 392
 393 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 394 {
 395         struct sock *sk;
 396
 397         if (map->value_size != sizeof(u64))
 398                 return ERR_PTR(-ENOSPC);
 399
 400         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 401         if (!sk)
 402                 return ERR_PTR(-ENOENT);
 403
 404         __sock_gen_cookie(sk);
 405         return &sk->sk_cookie;
 406 }
 407
 408 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 409                              struct sock **psk)
 410 {
 411         struct sock *sk;
 412         int err = 0;
 413
 414         spin_lock_bh(&stab->lock);
 415         sk = *psk;
 416         if (!sk_test || sk_test == sk)
 417                 sk = xchg(psk, NULL);
 418
 419         if (likely(sk))
 420                 sock_map_unref(sk, psk);
 421         else
 422                 err = -EINVAL;
 423
 424         spin_unlock_bh(&stab->lock);
 425         return err;
 426 }
 427
 428 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 429                                       void *link_raw)
 430 {
 431         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 432
 433         __sock_map_delete(stab, sk, link_raw);
 434 }
 435
 436 static long sock_map_delete_elem(struct bpf_map *map, void *key)
 437 {
 438         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 439         u32 i = *(u32 *)key;
 440         struct sock **psk;
 441
 442         if (unlikely(i >= map->max_entries))
 443                 return -EINVAL;
 444
 445         psk = &stab->sks[i];
 446         return __sock_map_delete(stab, NULL, psk);
 447 }
 448
 449 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 450 {
 451         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 452         u32 i = key ? *(u32 *)key : U32_MAX;
 453         u32 *key_next = next;
 454
 455         if (i == stab->map.max_entries - 1)
 456                 return -ENOENT;
 457         if (i >= stab->map.max_entries)
 458                 *key_next = 0;
 459         else
 460                 *key_next = i + 1;
 461         return 0;
 462 }
 463
 464 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 465                                   struct sock *sk, u64 flags)
 466 {
 467         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 468         struct sk_psock_link *link;
 469         struct sk_psock *psock;
 470         struct sock *osk;
 471         int ret;
 472
 473         WARN_ON_ONCE(!rcu_read_lock_held());
 474         if (unlikely(flags > BPF_EXIST))
 475                 return -EINVAL;
 476         if (unlikely(idx >= map->max_entries))
 477                 return -E2BIG;
 478
 479         link = sk_psock_init_link();
 480         if (!link)
 481                 return -ENOMEM;
 482
 483         ret = sock_map_link(map, sk);
 484         if (ret < 0)
 485                 goto out_free;
 486
 487         psock = sk_psock(sk);
 488         WARN_ON_ONCE(!psock);
 489
 490         spin_lock_bh(&stab->lock);
 491         osk = stab->sks[idx];
 492         if (osk && flags == BPF_NOEXIST) {
 493                 ret = -EEXIST;
 494                 goto out_unlock;
 495         } else if (!osk && flags == BPF_EXIST) {
 496                 ret = -ENOENT;
 497                 goto out_unlock;
 498         }
 499
 500         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 501         stab->sks[idx] = sk;
 502         if (osk)
 503                 sock_map_unref(osk, &stab->sks[idx]);
 504         spin_unlock_bh(&stab->lock);
 505         return 0;
 506 out_unlock:
 507         spin_unlock_bh(&stab->lock);
 508         if (psock)
 509                 sk_psock_put(sk, psock);
 510 out_free:
 511         sk_psock_free_link(link);
 512         return ret;
 513 }
 514
 515 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 516 {
 517         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 518                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 519                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 520 }
 521
 522 static bool sock_map_redirect_allowed(const struct sock *sk)
 523 {
 524         if (sk_is_tcp(sk))
 525                 return sk->sk_state != TCP_LISTEN;
 526         else
 527                 return sk->sk_state == TCP_ESTABLISHED;
 528 }
 529
 530 static bool sock_map_sk_is_suitable(const struct sock *sk)
 531 {
 532         return !!sk->sk_prot->psock_update_sk_prot;
 533 }
 534
 535 static bool sock_map_sk_state_allowed(const struct sock *sk)
 536 {
 537         if (sk_is_tcp(sk))
 538                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 539         return true;
 540 }
 541
 542 static int sock_hash_update_common(struct bpf_map *map, void *key,
 543                                    struct sock *sk, u64 flags);
 544
 545 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 546                              u64 flags)
 547 {
 548         struct socket *sock;
 549         struct sock *sk;
 550         int ret;
 551         u64 ufd;
 552
 553         if (map->value_size == sizeof(u64))
 554                 ufd = *(u64 *)value;
 555         else
 556                 ufd = *(u32 *)value;
 557         if (ufd > S32_MAX)
 558                 return -EINVAL;
 559
 560         sock = sockfd_lookup(ufd, &ret);
 561         if (!sock)
 562                 return ret;
 563         sk = sock->sk;
 564         if (!sk) {
 565                 ret = -EINVAL;
 566                 goto out;
 567         }
 568         if (!sock_map_sk_is_suitable(sk)) {
 569                 ret = -EOPNOTSUPP;
 570                 goto out;
 571         }
 572
 573         sock_map_sk_acquire(sk);
 574         if (!sock_map_sk_state_allowed(sk))
 575                 ret = -EOPNOTSUPP;
 576         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 577                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 578         else
 579                 ret = sock_hash_update_common(map, key, sk, flags);
 580         sock_map_sk_release(sk);
 581 out:
 582         sockfd_put(sock);
 583         return ret;
 584 }
 585
 586 static long sock_map_update_elem(struct bpf_map *map, void *key,
 587                                  void *value, u64 flags)
 588 {
 589         struct sock *sk = (struct sock *)value;
 590         int ret;
 591
 592         if (unlikely(!sk || !sk_fullsock(sk)))
 593                 return -EINVAL;
 594
 595         if (!sock_map_sk_is_suitable(sk))
 596                 return -EOPNOTSUPP;
 597
 598         local_bh_disable();
 599         bh_lock_sock(sk);
 600         if (!sock_map_sk_state_allowed(sk))
 601                 ret = -EOPNOTSUPP;
 602         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 603                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 604         else
 605                 ret = sock_hash_update_common(map, key, sk, flags);
 606         bh_unlock_sock(sk);
 607         local_bh_enable();
 608         return ret;
 609 }
 610
 611 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 612            struct bpf_map *, map, void *, key, u64, flags)
 613 {
 614         WARN_ON_ONCE(!rcu_read_lock_held());
 615
 616         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 617                    sock_map_op_okay(sops)))
 618                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 619                                               flags);
 620         return -EOPNOTSUPP;
 621 }
 622
 623 const struct bpf_func_proto bpf_sock_map_update_proto = {
 624         .func           = bpf_sock_map_update,
 625         .gpl_only       = false,
 626         .pkt_access     = true,
 627         .ret_type       = RET_INTEGER,
 628         .arg1_type      = ARG_PTR_TO_CTX,
 629         .arg2_type      = ARG_CONST_MAP_PTR,
 630         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 631         .arg4_type      = ARG_ANYTHING,
 632 };
 633
 634 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 635            struct bpf_map *, map, u32, key, u64, flags)
 636 {
 637         struct sock *sk;
 638
 639         if (unlikely(flags & ~(BPF_F_INGRESS)))
 640                 return SK_DROP;
 641
 642         sk = __sock_map_lookup_elem(map, key);
 643         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 644                 return SK_DROP;
 645
 646         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 647         return SK_PASS;
 648 }
 649
 650 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 651         .func           = bpf_sk_redirect_map,
 652         .gpl_only       = false,
 653         .ret_type       = RET_INTEGER,
 654         .arg1_type      = ARG_PTR_TO_CTX,
 655         .arg2_type      = ARG_CONST_MAP_PTR,
 656         .arg3_type      = ARG_ANYTHING,
 657         .arg4_type      = ARG_ANYTHING,
 658 };
 659
 660 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 661            struct bpf_map *, map, u32, key, u64, flags)
 662 {
 663         struct sock *sk;
 664
 665         if (unlikely(flags & ~(BPF_F_INGRESS)))
 666                 return SK_DROP;
 667
 668         sk = __sock_map_lookup_elem(map, key);
 669         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 670                 return SK_DROP;
 671
 672         msg->flags = flags;
 673         msg->sk_redir = sk;
 674         return SK_PASS;
 675 }
 676
 677 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 678         .func           = bpf_msg_redirect_map,
 679         .gpl_only       = false,
 680         .ret_type       = RET_INTEGER,
 681         .arg1_type      = ARG_PTR_TO_CTX,
 682         .arg2_type      = ARG_CONST_MAP_PTR,
 683         .arg3_type      = ARG_ANYTHING,
 684         .arg4_type      = ARG_ANYTHING,
 685 };
 686
 687 struct sock_map_seq_info {
 688         struct bpf_map *map;
 689         struct sock *sk;
 690         u32 index;
 691 };
 692
 693 struct bpf_iter__sockmap {
 694         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 695         __bpf_md_ptr(struct bpf_map *, map);
 696         __bpf_md_ptr(void *, key);
 697         __bpf_md_ptr(struct sock *, sk);
 698 };
 699
 700 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 701                      struct bpf_map *map, void *key,
 702                      struct sock *sk)
 703
 704 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 705 {
 706         if (unlikely(info->index >= info->map->max_entries))
 707                 return NULL;
 708
 709         info->sk = __sock_map_lookup_elem(info->map, info->index);
 710
 711         /* can't return sk directly, since that might be NULL */
 712         return info;
 713 }
 714
 715 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 716         __acquires(rcu)
 717 {
 718         struct sock_map_seq_info *info = seq->private;
 719
 720         if (*pos == 0)
 721                 ++*pos;
 722
 723         /* pairs with sock_map_seq_stop */
 724         rcu_read_lock();
 725         return sock_map_seq_lookup_elem(info);
 726 }
 727
 728 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 729         __must_hold(rcu)
 730 {
 731         struct sock_map_seq_info *info = seq->private;
 732
 733         ++*pos;
 734         ++info->index;
 735
 736         return sock_map_seq_lookup_elem(info);
 737 }
 738
 739 static int sock_map_seq_show(struct seq_file *seq, void *v)
 740         __must_hold(rcu)
 741 {
 742         struct sock_map_seq_info *info = seq->private;
 743         struct bpf_iter__sockmap ctx = {};
 744         struct bpf_iter_meta meta;
 745         struct bpf_prog *prog;
 746
 747         meta.seq = seq;
 748         prog = bpf_iter_get_info(&meta, !v);
 749         if (!prog)
 750                 return 0;
 751
 752         ctx.meta = &meta;
 753         ctx.map = info->map;
 754         if (v) {
 755                 ctx.key = &info->index;
 756                 ctx.sk = info->sk;
 757         }
 758
 759         return bpf_iter_run_prog(prog, &ctx);
 760 }
 761
 762 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 763         __releases(rcu)
 764 {
 765         if (!v)
 766                 (void)sock_map_seq_show(seq, NULL);
 767
 768         /* pairs with sock_map_seq_start */
 769         rcu_read_unlock();
 770 }
 771
 772 static const struct seq_operations sock_map_seq_ops = {
 773         .start  = sock_map_seq_start,
 774         .next   = sock_map_seq_next,
 775         .stop   = sock_map_seq_stop,
 776         .show   = sock_map_seq_show,
 777 };
 778
 779 static int sock_map_init_seq_private(void *priv_data,
 780                                      struct bpf_iter_aux_info *aux)
 781 {
 782         struct sock_map_seq_info *info = priv_data;
 783
 784         bpf_map_inc_with_uref(aux->map);
 785         info->map = aux->map;
 786         return 0;
 787 }
 788
 789 static void sock_map_fini_seq_private(void *priv_data)
 790 {
 791         struct sock_map_seq_info *info = priv_data;
 792
 793         bpf_map_put_with_uref(info->map);
 794 }
 795
 796 static u64 sock_map_mem_usage(const struct bpf_map *map)
 797 {
 798         u64 usage = sizeof(struct bpf_stab);
 799
 800         usage += (u64)map->max_entries * sizeof(struct sock *);
 801         return usage;
 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 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 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_mem_usage          = sock_map_mem_usage,
 824         .map_btf_id             = &sock_map_btf_ids[0],
 825         .iter_seq_info          = &sock_map_iter_seq_info,
 826 };
 827
 828 struct bpf_shtab_elem {
 829         struct rcu_head rcu;
 830         u32 hash;
 831         struct sock *sk;
 832         struct hlist_node node;
 833         u8 key[];
 834 };
 835
 836 struct bpf_shtab_bucket {
 837         struct hlist_head head;
 838         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         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         spin_unlock_bh(&bucket->lock);
 922 }
 923
 924 static long 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         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         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         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         spin_unlock_bh(&bucket->lock);
1030         return 0;
1031 out_unlock:
1032         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 (attr->max_entries == 0 ||
1085             attr->key_size    == 0 ||
1086             (attr->value_size != sizeof(u32) &&
1087              attr->value_size != sizeof(u64)) ||
1088             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1089                 return ERR_PTR(-EINVAL);
1090         if (attr->key_size > MAX_BPF_STACK)
1091                 return ERR_PTR(-E2BIG);
1092
1093         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1094         if (!htab)
1095                 return ERR_PTR(-ENOMEM);
1096
1097         bpf_map_init_from_attr(&htab->map, attr);
1098
1099         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1100         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1101                           round_up(htab->map.key_size, 8);
1102         if (htab->buckets_num == 0 ||
1103             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1104                 err = -EINVAL;
1105                 goto free_htab;
1106         }
1107
1108         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1109                                            sizeof(struct bpf_shtab_bucket),
1110                                            htab->map.numa_node);
1111         if (!htab->buckets) {
1112                 err = -ENOMEM;
1113                 goto free_htab;
1114         }
1115
1116         for (i = 0; i < htab->buckets_num; i++) {
1117                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1118                 spin_lock_init(&htab->buckets[i].lock);
1119         }
1120
1121         return &htab->map;
1122 free_htab:
1123         bpf_map_area_free(htab);
1124         return ERR_PTR(err);
1125 }
1126
1127 static void sock_hash_free(struct bpf_map *map)
1128 {
1129         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1130         struct bpf_shtab_bucket *bucket;
1131         struct hlist_head unlink_list;
1132         struct bpf_shtab_elem *elem;
1133         struct hlist_node *node;
1134         int i;
1135
1136         /* After the sync no updates or deletes will be in-flight so it
1137          * is safe to walk map and remove entries without risking a race
1138          * in EEXIST update case.
1139          */
1140         synchronize_rcu();
1141         for (i = 0; i < htab->buckets_num; i++) {
1142                 bucket = sock_hash_select_bucket(htab, i);
1143
1144                 /* We are racing with sock_hash_delete_from_link to
1145                  * enter the spin-lock critical section. Every socket on
1146                  * the list is still linked to sockhash. Since link
1147                  * exists, psock exists and holds a ref to socket. That
1148                  * lets us to grab a socket ref too.
1149                  */
1150                 spin_lock_bh(&bucket->lock);
1151                 hlist_for_each_entry(elem, &bucket->head, node)
1152                         sock_hold(elem->sk);
1153                 hlist_move_list(&bucket->head, &unlink_list);
1154                 spin_unlock_bh(&bucket->lock);
1155
1156                 /* Process removed entries out of atomic context to
1157                  * block for socket lock before deleting the psock's
1158                  * link to sockhash.
1159                  */
1160                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1161                         hlist_del(&elem->node);
1162                         lock_sock(elem->sk);
1163                         rcu_read_lock();
1164                         sock_map_unref(elem->sk, elem);
1165                         rcu_read_unlock();
1166                         release_sock(elem->sk);
1167                         sock_put(elem->sk);
1168                         sock_hash_free_elem(htab, elem);
1169                 }
1170         }
1171
1172         /* wait for psock readers accessing its map link */
1173         synchronize_rcu();
1174
1175         bpf_map_area_free(htab->buckets);
1176         bpf_map_area_free(htab);
1177 }
1178
1179 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1180 {
1181         struct sock *sk;
1182
1183         if (map->value_size != sizeof(u64))
1184                 return ERR_PTR(-ENOSPC);
1185
1186         sk = __sock_hash_lookup_elem(map, key);
1187         if (!sk)
1188                 return ERR_PTR(-ENOENT);
1189
1190         __sock_gen_cookie(sk);
1191         return &sk->sk_cookie;
1192 }
1193
1194 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1195 {
1196         struct sock *sk;
1197
1198         sk = __sock_hash_lookup_elem(map, key);
1199         if (!sk)
1200                 return NULL;
1201         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1202                 return NULL;
1203         return sk;
1204 }
1205
1206 static void sock_hash_release_progs(struct bpf_map *map)
1207 {
1208         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1209 }
1210
1211 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1212            struct bpf_map *, map, void *, key, u64, flags)
1213 {
1214         WARN_ON_ONCE(!rcu_read_lock_held());
1215
1216         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1217                    sock_map_op_okay(sops)))
1218                 return sock_hash_update_common(map, key, sops->sk, flags);
1219         return -EOPNOTSUPP;
1220 }
1221
1222 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1223         .func           = bpf_sock_hash_update,
1224         .gpl_only       = false,
1225         .pkt_access     = true,
1226         .ret_type       = RET_INTEGER,
1227         .arg1_type      = ARG_PTR_TO_CTX,
1228         .arg2_type      = ARG_CONST_MAP_PTR,
1229         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1230         .arg4_type      = ARG_ANYTHING,
1231 };
1232
1233 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1234            struct bpf_map *, map, void *, key, u64, flags)
1235 {
1236         struct sock *sk;
1237
1238         if (unlikely(flags & ~(BPF_F_INGRESS)))
1239                 return SK_DROP;
1240
1241         sk = __sock_hash_lookup_elem(map, key);
1242         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1243                 return SK_DROP;
1244
1245         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1246         return SK_PASS;
1247 }
1248
1249 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1250         .func           = bpf_sk_redirect_hash,
1251         .gpl_only       = false,
1252         .ret_type       = RET_INTEGER,
1253         .arg1_type      = ARG_PTR_TO_CTX,
1254         .arg2_type      = ARG_CONST_MAP_PTR,
1255         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1256         .arg4_type      = ARG_ANYTHING,
1257 };
1258
1259 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1260            struct bpf_map *, map, void *, key, u64, flags)
1261 {
1262         struct sock *sk;
1263
1264         if (unlikely(flags & ~(BPF_F_INGRESS)))
1265                 return SK_DROP;
1266
1267         sk = __sock_hash_lookup_elem(map, key);
1268         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1269                 return SK_DROP;
1270
1271         msg->flags = flags;
1272         msg->sk_redir = sk;
1273         return SK_PASS;
1274 }
1275
1276 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1277         .func           = bpf_msg_redirect_hash,
1278         .gpl_only       = false,
1279         .ret_type       = RET_INTEGER,
1280         .arg1_type      = ARG_PTR_TO_CTX,
1281         .arg2_type      = ARG_CONST_MAP_PTR,
1282         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1283         .arg4_type      = ARG_ANYTHING,
1284 };
1285
1286 struct sock_hash_seq_info {
1287         struct bpf_map *map;
1288         struct bpf_shtab *htab;
1289         u32 bucket_id;
1290 };
1291
1292 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1293                                      struct bpf_shtab_elem *prev_elem)
1294 {
1295         const struct bpf_shtab *htab = info->htab;
1296         struct bpf_shtab_bucket *bucket;
1297         struct bpf_shtab_elem *elem;
1298         struct hlist_node *node;
1299
1300         /* try to find next elem in the same bucket */
1301         if (prev_elem) {
1302                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1303                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1304                 if (elem)
1305                         return elem;
1306
1307                 /* no more elements, continue in the next bucket */
1308                 info->bucket_id++;
1309         }
1310
1311         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1312                 bucket = &htab->buckets[info->bucket_id];
1313                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1314                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1315                 if (elem)
1316                         return elem;
1317         }
1318
1319         return NULL;
1320 }
1321
1322 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1323         __acquires(rcu)
1324 {
1325         struct sock_hash_seq_info *info = seq->private;
1326
1327         if (*pos == 0)
1328                 ++*pos;
1329
1330         /* pairs with sock_hash_seq_stop */
1331         rcu_read_lock();
1332         return sock_hash_seq_find_next(info, NULL);
1333 }
1334
1335 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1336         __must_hold(rcu)
1337 {
1338         struct sock_hash_seq_info *info = seq->private;
1339
1340         ++*pos;
1341         return sock_hash_seq_find_next(info, v);
1342 }
1343
1344 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1345         __must_hold(rcu)
1346 {
1347         struct sock_hash_seq_info *info = seq->private;
1348         struct bpf_iter__sockmap ctx = {};
1349         struct bpf_shtab_elem *elem = v;
1350         struct bpf_iter_meta meta;
1351         struct bpf_prog *prog;
1352
1353         meta.seq = seq;
1354         prog = bpf_iter_get_info(&meta, !elem);
1355         if (!prog)
1356                 return 0;
1357
1358         ctx.meta = &meta;
1359         ctx.map = info->map;
1360         if (elem) {
1361                 ctx.key = elem->key;
1362                 ctx.sk = elem->sk;
1363         }
1364
1365         return bpf_iter_run_prog(prog, &ctx);
1366 }
1367
1368 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1369         __releases(rcu)
1370 {
1371         if (!v)
1372                 (void)sock_hash_seq_show(seq, NULL);
1373
1374         /* pairs with sock_hash_seq_start */
1375         rcu_read_unlock();
1376 }
1377
1378 static const struct seq_operations sock_hash_seq_ops = {
1379         .start  = sock_hash_seq_start,
1380         .next   = sock_hash_seq_next,
1381         .stop   = sock_hash_seq_stop,
1382         .show   = sock_hash_seq_show,
1383 };
1384
1385 static int sock_hash_init_seq_private(void *priv_data,
1386                                       struct bpf_iter_aux_info *aux)
1387 {
1388         struct sock_hash_seq_info *info = priv_data;
1389
1390         bpf_map_inc_with_uref(aux->map);
1391         info->map = aux->map;
1392         info->htab = container_of(aux->map, struct bpf_shtab, map);
1393         return 0;
1394 }
1395
1396 static void sock_hash_fini_seq_private(void *priv_data)
1397 {
1398         struct sock_hash_seq_info *info = priv_data;
1399
1400         bpf_map_put_with_uref(info->map);
1401 }
1402
1403 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1404 {
1405         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1406         u64 usage = sizeof(*htab);
1407
1408         usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1409         usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1410         return usage;
1411 }
1412
1413 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1414         .seq_ops                = &sock_hash_seq_ops,
1415         .init_seq_private       = sock_hash_init_seq_private,
1416         .fini_seq_private       = sock_hash_fini_seq_private,
1417         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1418 };
1419
1420 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1421 const struct bpf_map_ops sock_hash_ops = {
1422         .map_meta_equal         = bpf_map_meta_equal,
1423         .map_alloc              = sock_hash_alloc,
1424         .map_free               = sock_hash_free,
1425         .map_get_next_key       = sock_hash_get_next_key,
1426         .map_update_elem        = sock_map_update_elem,
1427         .map_delete_elem        = sock_hash_delete_elem,
1428         .map_lookup_elem        = sock_hash_lookup,
1429         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1430         .map_release_uref       = sock_hash_release_progs,
1431         .map_check_btf          = map_check_no_btf,
1432         .map_mem_usage          = sock_hash_mem_usage,
1433         .map_btf_id             = &sock_hash_map_btf_ids[0],
1434         .iter_seq_info          = &sock_hash_iter_seq_info,
1435 };
1436
1437 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1438 {
1439         switch (map->map_type) {
1440         case BPF_MAP_TYPE_SOCKMAP:
1441                 return &container_of(map, struct bpf_stab, map)->progs;
1442         case BPF_MAP_TYPE_SOCKHASH:
1443                 return &container_of(map, struct bpf_shtab, map)->progs;
1444         default:
1445                 break;
1446         }
1447
1448         return NULL;
1449 }
1450
1451 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1452                                 u32 which)
1453 {
1454         struct sk_psock_progs *progs = sock_map_progs(map);
1455
1456         if (!progs)
1457                 return -EOPNOTSUPP;
1458
1459         switch (which) {
1460         case BPF_SK_MSG_VERDICT:
1461                 *pprog = &progs->msg_parser;
1462                 break;
1463 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1464         case BPF_SK_SKB_STREAM_PARSER:
1465                 *pprog = &progs->stream_parser;
1466                 break;
1467 #endif
1468         case BPF_SK_SKB_STREAM_VERDICT:
1469                 if (progs->skb_verdict)
1470                         return -EBUSY;
1471                 *pprog = &progs->stream_verdict;
1472                 break;
1473         case BPF_SK_SKB_VERDICT:
1474                 if (progs->stream_verdict)
1475                         return -EBUSY;
1476                 *pprog = &progs->skb_verdict;
1477                 break;
1478         default:
1479                 return -EOPNOTSUPP;
1480         }
1481
1482         return 0;
1483 }
1484
1485 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1486                                 struct bpf_prog *old, u32 which)
1487 {
1488         struct bpf_prog **pprog;
1489         int ret;
1490
1491         ret = sock_map_prog_lookup(map, &pprog, which);
1492         if (ret)
1493                 return ret;
1494
1495         if (old)
1496                 return psock_replace_prog(pprog, prog, old);
1497
1498         psock_set_prog(pprog, prog);
1499         return 0;
1500 }
1501
1502 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1503                             union bpf_attr __user *uattr)
1504 {
1505         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1506         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1507         struct bpf_prog **pprog;
1508         struct bpf_prog *prog;
1509         struct bpf_map *map;
1510         struct fd f;
1511         u32 id = 0;
1512         int ret;
1513
1514         if (attr->query.query_flags)
1515                 return -EINVAL;
1516
1517         f = fdget(ufd);
1518         map = __bpf_map_get(f);
1519         if (IS_ERR(map))
1520                 return PTR_ERR(map);
1521
1522         rcu_read_lock();
1523
1524         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1525         if (ret)
1526                 goto end;
1527
1528         prog = *pprog;
1529         prog_cnt = !prog ? 0 : 1;
1530
1531         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1532                 goto end;
1533
1534         /* we do not hold the refcnt, the bpf prog may be released
1535          * asynchronously and the id would be set to 0.
1536          */
1537         id = data_race(prog->aux->id);
1538         if (id == 0)
1539                 prog_cnt = 0;
1540
1541 end:
1542         rcu_read_unlock();
1543
1544         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1545             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1546             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1547                 ret = -EFAULT;
1548
1549         fdput(f);
1550         return ret;
1551 }
1552
1553 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1554 {
1555         switch (link->map->map_type) {
1556         case BPF_MAP_TYPE_SOCKMAP:
1557                 return sock_map_delete_from_link(link->map, sk,
1558                                                  link->link_raw);
1559         case BPF_MAP_TYPE_SOCKHASH:
1560                 return sock_hash_delete_from_link(link->map, sk,
1561                                                   link->link_raw);
1562         default:
1563                 break;
1564         }
1565 }
1566
1567 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1568 {
1569         struct sk_psock_link *link;
1570
1571         while ((link = sk_psock_link_pop(psock))) {
1572                 sock_map_unlink(sk, link);
1573                 sk_psock_free_link(link);
1574         }
1575 }
1576
1577 void sock_map_unhash(struct sock *sk)
1578 {
1579         void (*saved_unhash)(struct sock *sk);
1580         struct sk_psock *psock;
1581
1582         rcu_read_lock();
1583         psock = sk_psock(sk);
1584         if (unlikely(!psock)) {
1585                 rcu_read_unlock();
1586                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1587         } else {
1588                 saved_unhash = psock->saved_unhash;
1589                 sock_map_remove_links(sk, psock);
1590                 rcu_read_unlock();
1591         }
1592         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1593                 return;
1594         if (saved_unhash)
1595                 saved_unhash(sk);
1596 }
1597 EXPORT_SYMBOL_GPL(sock_map_unhash);
1598
1599 void sock_map_destroy(struct sock *sk)
1600 {
1601         void (*saved_destroy)(struct sock *sk);
1602         struct sk_psock *psock;
1603
1604         rcu_read_lock();
1605         psock = sk_psock_get(sk);
1606         if (unlikely(!psock)) {
1607                 rcu_read_unlock();
1608                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1609         } else {
1610                 saved_destroy = psock->saved_destroy;
1611                 sock_map_remove_links(sk, psock);
1612                 rcu_read_unlock();
1613                 sk_psock_stop(psock);
1614                 sk_psock_put(sk, psock);
1615         }
1616         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1617                 return;
1618         if (saved_destroy)
1619                 saved_destroy(sk);
1620 }
1621 EXPORT_SYMBOL_GPL(sock_map_destroy);
1622
1623 void sock_map_close(struct sock *sk, long timeout)
1624 {
1625         void (*saved_close)(struct sock *sk, long timeout);
1626         struct sk_psock *psock;
1627
1628         lock_sock(sk);
1629         rcu_read_lock();
1630         psock = sk_psock_get(sk);
1631         if (unlikely(!psock)) {
1632                 rcu_read_unlock();
1633                 release_sock(sk);
1634                 saved_close = READ_ONCE(sk->sk_prot)->close;
1635         } else {
1636                 saved_close = psock->saved_close;
1637                 sock_map_remove_links(sk, psock);
1638                 rcu_read_unlock();
1639                 sk_psock_stop(psock);
1640                 release_sock(sk);
1641                 cancel_delayed_work_sync(&psock->work);
1642                 sk_psock_put(sk, psock);
1643         }
1644
1645         /* Make sure we do not recurse. This is a bug.
1646          * Leak the socket instead of crashing on a stack overflow.
1647          */
1648         if (WARN_ON_ONCE(saved_close == sock_map_close))
1649                 return;
1650         saved_close(sk, timeout);
1651 }
1652 EXPORT_SYMBOL_GPL(sock_map_close);
1653
1654 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1655                                        union bpf_iter_link_info *linfo,
1656                                        struct bpf_iter_aux_info *aux)
1657 {
1658         struct bpf_map *map;
1659         int err = -EINVAL;
1660
1661         if (!linfo->map.map_fd)
1662                 return -EBADF;
1663
1664         map = bpf_map_get_with_uref(linfo->map.map_fd);
1665         if (IS_ERR(map))
1666                 return PTR_ERR(map);
1667
1668         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1669             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1670                 goto put_map;
1671
1672         if (prog->aux->max_rdonly_access > map->key_size) {
1673                 err = -EACCES;
1674                 goto put_map;
1675         }
1676
1677         aux->map = map;
1678         return 0;
1679
1680 put_map:
1681         bpf_map_put_with_uref(map);
1682         return err;
1683 }
1684
1685 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1686 {
1687         bpf_map_put_with_uref(aux->map);
1688 }
1689
1690 static struct bpf_iter_reg sock_map_iter_reg = {
1691         .target                 = "sockmap",
1692         .attach_target          = sock_map_iter_attach_target,
1693         .detach_target          = sock_map_iter_detach_target,
1694         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1695         .fill_link_info         = bpf_iter_map_fill_link_info,
1696         .ctx_arg_info_size      = 2,
1697         .ctx_arg_info           = {
1698                 { offsetof(struct bpf_iter__sockmap, key),
1699                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1700                 { offsetof(struct bpf_iter__sockmap, sk),
1701                   PTR_TO_BTF_ID_OR_NULL },
1702         },
1703 };
1704
1705 static int __init bpf_sockmap_iter_init(void)
1706 {
1707         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1708                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1709         return bpf_iter_reg_target(&sock_map_iter_reg);
1710 }
1711 late_initcall(bpf_sockmap_iter_init);