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                         sock_hold(sk);
 353                         lock_sock(sk);
 354                         rcu_read_lock();
 355                         sock_map_unref(sk, psk);
 356                         rcu_read_unlock();
 357                         release_sock(sk);
 358                         sock_put(sk);
 359                 }
 360         }
 361
 362         /* wait for psock readers accessing its map link */
 363         synchronize_rcu();
 364
 365         bpf_map_area_free(stab->sks);
 366         bpf_map_area_free(stab);
 367 }
 368
 369 static void sock_map_release_progs(struct bpf_map *map)
 370 {
 371         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 372 }
 373
 374 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 375 {
 376         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 377
 378         WARN_ON_ONCE(!rcu_read_lock_held());
 379
 380         if (unlikely(key >= map->max_entries))
 381                 return NULL;
 382         return READ_ONCE(stab->sks[key]);
 383 }
 384
 385 static void *sock_map_lookup(struct bpf_map *map, void *key)
 386 {
 387         struct sock *sk;
 388
 389         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 390         if (!sk)
 391                 return NULL;
 392         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 393                 return NULL;
 394         return sk;
 395 }
 396
 397 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 398 {
 399         struct sock *sk;
 400
 401         if (map->value_size != sizeof(u64))
 402                 return ERR_PTR(-ENOSPC);
 403
 404         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 405         if (!sk)
 406                 return ERR_PTR(-ENOENT);
 407
 408         __sock_gen_cookie(sk);
 409         return &sk->sk_cookie;
 410 }
 411
 412 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 413                              struct sock **psk)
 414 {
 415         struct sock *sk;
 416         int err = 0;
 417         unsigned long flags;
 418
 419         raw_spin_lock_irqsave(&stab->lock, flags);
 420         sk = *psk;
 421         if (!sk_test || sk_test == sk)
 422                 sk = xchg(psk, NULL);
 423
 424         if (likely(sk))
 425                 sock_map_unref(sk, psk);
 426         else
 427                 err = -EINVAL;
 428
 429         raw_spin_unlock_irqrestore(&stab->lock, flags);
 430         return err;
 431 }
 432
 433 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 434                                       void *link_raw)
 435 {
 436         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 437
 438         __sock_map_delete(stab, sk, link_raw);
 439 }
 440
 441 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 442 {
 443         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 444         u32 i = *(u32 *)key;
 445         struct sock **psk;
 446
 447         if (unlikely(i >= map->max_entries))
 448                 return -EINVAL;
 449
 450         psk = &stab->sks[i];
 451         return __sock_map_delete(stab, NULL, psk);
 452 }
 453
 454 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 455 {
 456         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 457         u32 i = key ? *(u32 *)key : U32_MAX;
 458         u32 *key_next = next;
 459
 460         if (i == stab->map.max_entries - 1)
 461                 return -ENOENT;
 462         if (i >= stab->map.max_entries)
 463                 *key_next = 0;
 464         else
 465                 *key_next = i + 1;
 466         return 0;
 467 }
 468
 469 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 470                                   struct sock *sk, u64 flags)
 471 {
 472         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 473         struct sk_psock_link *link;
 474         struct sk_psock *psock;
 475         struct sock *osk;
 476         int ret;
 477
 478         WARN_ON_ONCE(!rcu_read_lock_held());
 479         if (unlikely(flags > BPF_EXIST))
 480                 return -EINVAL;
 481         if (unlikely(idx >= map->max_entries))
 482                 return -E2BIG;
 483
 484         link = sk_psock_init_link();
 485         if (!link)
 486                 return -ENOMEM;
 487
 488         ret = sock_map_link(map, sk);
 489         if (ret < 0)
 490                 goto out_free;
 491
 492         psock = sk_psock(sk);
 493         WARN_ON_ONCE(!psock);
 494
 495         raw_spin_lock_bh(&stab->lock);
 496         osk = stab->sks[idx];
 497         if (osk && flags == BPF_NOEXIST) {
 498                 ret = -EEXIST;
 499                 goto out_unlock;
 500         } else if (!osk && flags == BPF_EXIST) {
 501                 ret = -ENOENT;
 502                 goto out_unlock;
 503         }
 504
 505         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 506         stab->sks[idx] = sk;
 507         if (osk)
 508                 sock_map_unref(osk, &stab->sks[idx]);
 509         raw_spin_unlock_bh(&stab->lock);
 510         return 0;
 511 out_unlock:
 512         raw_spin_unlock_bh(&stab->lock);
 513         if (psock)
 514                 sk_psock_put(sk, psock);
 515 out_free:
 516         sk_psock_free_link(link);
 517         return ret;
 518 }
 519
 520 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 521 {
 522         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 523                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 524                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 525 }
 526
 527 static bool sock_map_redirect_allowed(const struct sock *sk)
 528 {
 529         if (sk_is_tcp(sk))
 530                 return sk->sk_state != TCP_LISTEN;
 531         else
 532                 return sk->sk_state == TCP_ESTABLISHED;
 533 }
 534
 535 static bool sock_map_sk_is_suitable(const struct sock *sk)
 536 {
 537         return !!sk->sk_prot->psock_update_sk_prot;
 538 }
 539
 540 static bool sock_map_sk_state_allowed(const struct sock *sk)
 541 {
 542         if (sk_is_tcp(sk))
 543                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 544         return true;
 545 }
 546
 547 static int sock_hash_update_common(struct bpf_map *map, void *key,
 548                                    struct sock *sk, u64 flags);
 549
 550 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 551                              u64 flags)
 552 {
 553         struct socket *sock;
 554         struct sock *sk;
 555         int ret;
 556         u64 ufd;
 557
 558         if (map->value_size == sizeof(u64))
 559                 ufd = *(u64 *)value;
 560         else
 561                 ufd = *(u32 *)value;
 562         if (ufd > S32_MAX)
 563                 return -EINVAL;
 564
 565         sock = sockfd_lookup(ufd, &ret);
 566         if (!sock)
 567                 return ret;
 568         sk = sock->sk;
 569         if (!sk) {
 570                 ret = -EINVAL;
 571                 goto out;
 572         }
 573         if (!sock_map_sk_is_suitable(sk)) {
 574                 ret = -EOPNOTSUPP;
 575                 goto out;
 576         }
 577
 578         sock_map_sk_acquire(sk);
 579         if (!sock_map_sk_state_allowed(sk))
 580                 ret = -EOPNOTSUPP;
 581         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 582                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 583         else
 584                 ret = sock_hash_update_common(map, key, sk, flags);
 585         sock_map_sk_release(sk);
 586 out:
 587         sockfd_put(sock);
 588         return ret;
 589 }
 590
 591 static int sock_map_update_elem(struct bpf_map *map, void *key,
 592                                 void *value, u64 flags)
 593 {
 594         struct sock *sk = (struct sock *)value;
 595         int ret;
 596
 597         if (unlikely(!sk || !sk_fullsock(sk)))
 598                 return -EINVAL;
 599
 600         if (!sock_map_sk_is_suitable(sk))
 601                 return -EOPNOTSUPP;
 602
 603         local_bh_disable();
 604         bh_lock_sock(sk);
 605         if (!sock_map_sk_state_allowed(sk))
 606                 ret = -EOPNOTSUPP;
 607         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 608                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 609         else
 610                 ret = sock_hash_update_common(map, key, sk, flags);
 611         bh_unlock_sock(sk);
 612         local_bh_enable();
 613         return ret;
 614 }
 615
 616 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 617            struct bpf_map *, map, void *, key, u64, flags)
 618 {
 619         WARN_ON_ONCE(!rcu_read_lock_held());
 620
 621         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 622                    sock_map_op_okay(sops)))
 623                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 624                                               flags);
 625         return -EOPNOTSUPP;
 626 }
 627
 628 const struct bpf_func_proto bpf_sock_map_update_proto = {
 629         .func           = bpf_sock_map_update,
 630         .gpl_only       = false,
 631         .pkt_access     = true,
 632         .ret_type       = RET_INTEGER,
 633         .arg1_type      = ARG_PTR_TO_CTX,
 634         .arg2_type      = ARG_CONST_MAP_PTR,
 635         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 636         .arg4_type      = ARG_ANYTHING,
 637 };
 638
 639 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 640            struct bpf_map *, map, u32, key, u64, flags)
 641 {
 642         struct sock *sk;
 643
 644         if (unlikely(flags & ~(BPF_F_INGRESS)))
 645                 return SK_DROP;
 646
 647         sk = __sock_map_lookup_elem(map, key);
 648         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 649                 return SK_DROP;
 650
 651         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 652         return SK_PASS;
 653 }
 654
 655 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 656         .func           = bpf_sk_redirect_map,
 657         .gpl_only       = false,
 658         .ret_type       = RET_INTEGER,
 659         .arg1_type      = ARG_PTR_TO_CTX,
 660         .arg2_type      = ARG_CONST_MAP_PTR,
 661         .arg3_type      = ARG_ANYTHING,
 662         .arg4_type      = ARG_ANYTHING,
 663 };
 664
 665 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 666            struct bpf_map *, map, u32, key, u64, flags)
 667 {
 668         struct sock *sk;
 669
 670         if (unlikely(flags & ~(BPF_F_INGRESS)))
 671                 return SK_DROP;
 672
 673         sk = __sock_map_lookup_elem(map, key);
 674         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 675                 return SK_DROP;
 676
 677         msg->flags = flags;
 678         msg->sk_redir = sk;
 679         return SK_PASS;
 680 }
 681
 682 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 683         .func           = bpf_msg_redirect_map,
 684         .gpl_only       = false,
 685         .ret_type       = RET_INTEGER,
 686         .arg1_type      = ARG_PTR_TO_CTX,
 687         .arg2_type      = ARG_CONST_MAP_PTR,
 688         .arg3_type      = ARG_ANYTHING,
 689         .arg4_type      = ARG_ANYTHING,
 690 };
 691
 692 struct sock_map_seq_info {
 693         struct bpf_map *map;
 694         struct sock *sk;
 695         u32 index;
 696 };
 697
 698 struct bpf_iter__sockmap {
 699         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 700         __bpf_md_ptr(struct bpf_map *, map);
 701         __bpf_md_ptr(void *, key);
 702         __bpf_md_ptr(struct sock *, sk);
 703 };
 704
 705 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 706                      struct bpf_map *map, void *key,
 707                      struct sock *sk)
 708
 709 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 710 {
 711         if (unlikely(info->index >= info->map->max_entries))
 712                 return NULL;
 713
 714         info->sk = __sock_map_lookup_elem(info->map, info->index);
 715
 716         /* can't return sk directly, since that might be NULL */
 717         return info;
 718 }
 719
 720 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 721         __acquires(rcu)
 722 {
 723         struct sock_map_seq_info *info = seq->private;
 724
 725         if (*pos == 0)
 726                 ++*pos;
 727
 728         /* pairs with sock_map_seq_stop */
 729         rcu_read_lock();
 730         return sock_map_seq_lookup_elem(info);
 731 }
 732
 733 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 734         __must_hold(rcu)
 735 {
 736         struct sock_map_seq_info *info = seq->private;
 737
 738         ++*pos;
 739         ++info->index;
 740
 741         return sock_map_seq_lookup_elem(info);
 742 }
 743
 744 static int sock_map_seq_show(struct seq_file *seq, void *v)
 745         __must_hold(rcu)
 746 {
 747         struct sock_map_seq_info *info = seq->private;
 748         struct bpf_iter__sockmap ctx = {};
 749         struct bpf_iter_meta meta;
 750         struct bpf_prog *prog;
 751
 752         meta.seq = seq;
 753         prog = bpf_iter_get_info(&meta, !v);
 754         if (!prog)
 755                 return 0;
 756
 757         ctx.meta = &meta;
 758         ctx.map = info->map;
 759         if (v) {
 760                 ctx.key = &info->index;
 761                 ctx.sk = info->sk;
 762         }
 763
 764         return bpf_iter_run_prog(prog, &ctx);
 765 }
 766
 767 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 768         __releases(rcu)
 769 {
 770         if (!v)
 771                 (void)sock_map_seq_show(seq, NULL);
 772
 773         /* pairs with sock_map_seq_start */
 774         rcu_read_unlock();
 775 }
 776
 777 static const struct seq_operations sock_map_seq_ops = {
 778         .start  = sock_map_seq_start,
 779         .next   = sock_map_seq_next,
 780         .stop   = sock_map_seq_stop,
 781         .show   = sock_map_seq_show,
 782 };
 783
 784 static int sock_map_init_seq_private(void *priv_data,
 785                                      struct bpf_iter_aux_info *aux)
 786 {
 787         struct sock_map_seq_info *info = priv_data;
 788
 789         bpf_map_inc_with_uref(aux->map);
 790         info->map = aux->map;
 791         return 0;
 792 }
 793
 794 static void sock_map_fini_seq_private(void *priv_data)
 795 {
 796         struct sock_map_seq_info *info = priv_data;
 797
 798         bpf_map_put_with_uref(info->map);
 799 }
 800
 801 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 802         .seq_ops                = &sock_map_seq_ops,
 803         .init_seq_private       = sock_map_init_seq_private,
 804         .fini_seq_private       = sock_map_fini_seq_private,
 805         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 806 };
 807
 808 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 809 const struct bpf_map_ops sock_map_ops = {
 810         .map_meta_equal         = bpf_map_meta_equal,
 811         .map_alloc              = sock_map_alloc,
 812         .map_free               = sock_map_free,
 813         .map_get_next_key       = sock_map_get_next_key,
 814         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 815         .map_update_elem        = sock_map_update_elem,
 816         .map_delete_elem        = sock_map_delete_elem,
 817         .map_lookup_elem        = sock_map_lookup,
 818         .map_release_uref       = sock_map_release_progs,
 819         .map_check_btf          = map_check_no_btf,
 820         .map_btf_id             = &sock_map_btf_ids[0],
 821         .iter_seq_info          = &sock_map_iter_seq_info,
 822 };
 823
 824 struct bpf_shtab_elem {
 825         struct rcu_head rcu;
 826         u32 hash;
 827         struct sock *sk;
 828         struct hlist_node node;
 829         u8 key[];
 830 };
 831
 832 struct bpf_shtab_bucket {
 833         struct hlist_head head;
 834         raw_spinlock_t lock;
 835 };
 836
 837 struct bpf_shtab {
 838         struct bpf_map map;
 839         struct bpf_shtab_bucket *buckets;
 840         u32 buckets_num;
 841         u32 elem_size;
 842         struct sk_psock_progs progs;
 843         atomic_t count;
 844 };
 845
 846 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 847 {
 848         return jhash(key, len, 0);
 849 }
 850
 851 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 852                                                         u32 hash)
 853 {
 854         return &htab->buckets[hash & (htab->buckets_num - 1)];
 855 }
 856
 857 static struct bpf_shtab_elem *
 858 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 859                           u32 key_size)
 860 {
 861         struct bpf_shtab_elem *elem;
 862
 863         hlist_for_each_entry_rcu(elem, head, node) {
 864                 if (elem->hash == hash &&
 865                     !memcmp(&elem->key, key, key_size))
 866                         return elem;
 867         }
 868
 869         return NULL;
 870 }
 871
 872 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 873 {
 874         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 875         u32 key_size = map->key_size, hash;
 876         struct bpf_shtab_bucket *bucket;
 877         struct bpf_shtab_elem *elem;
 878
 879         WARN_ON_ONCE(!rcu_read_lock_held());
 880
 881         hash = sock_hash_bucket_hash(key, key_size);
 882         bucket = sock_hash_select_bucket(htab, hash);
 883         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 884
 885         return elem ? elem->sk : NULL;
 886 }
 887
 888 static void sock_hash_free_elem(struct bpf_shtab *htab,
 889                                 struct bpf_shtab_elem *elem)
 890 {
 891         atomic_dec(&htab->count);
 892         kfree_rcu(elem, rcu);
 893 }
 894
 895 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 896                                        void *link_raw)
 897 {
 898         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 899         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 900         struct bpf_shtab_bucket *bucket;
 901
 902         WARN_ON_ONCE(!rcu_read_lock_held());
 903         bucket = sock_hash_select_bucket(htab, elem->hash);
 904
 905         /* elem may be deleted in parallel from the map, but access here
 906          * is okay since it's going away only after RCU grace period.
 907          * However, we need to check whether it's still present.
 908          */
 909         raw_spin_lock_bh(&bucket->lock);
 910         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 911                                                elem->key, map->key_size);
 912         if (elem_probe && elem_probe == elem) {
 913                 hlist_del_rcu(&elem->node);
 914                 sock_map_unref(elem->sk, elem);
 915                 sock_hash_free_elem(htab, elem);
 916         }
 917         raw_spin_unlock_bh(&bucket->lock);
 918 }
 919
 920 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 921 {
 922         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 923         u32 hash, key_size = map->key_size;
 924         struct bpf_shtab_bucket *bucket;
 925         struct bpf_shtab_elem *elem;
 926         int ret = -ENOENT;
 927         unsigned long flags;
 928
 929         hash = sock_hash_bucket_hash(key, key_size);
 930         bucket = sock_hash_select_bucket(htab, hash);
 931
 932         raw_spin_lock_irqsave(&bucket->lock, flags);
 933         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 934         if (elem) {
 935                 hlist_del_rcu(&elem->node);
 936                 sock_map_unref(elem->sk, elem);
 937                 sock_hash_free_elem(htab, elem);
 938                 ret = 0;
 939         }
 940         raw_spin_unlock_irqrestore(&bucket->lock, flags);
 941         return ret;
 942 }
 943
 944 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 945                                                    void *key, u32 key_size,
 946                                                    u32 hash, struct sock *sk,
 947                                                    struct bpf_shtab_elem *old)
 948 {
 949         struct bpf_shtab_elem *new;
 950
 951         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 952                 if (!old) {
 953                         atomic_dec(&htab->count);
 954                         return ERR_PTR(-E2BIG);
 955                 }
 956         }
 957
 958         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 959                                    GFP_ATOMIC | __GFP_NOWARN,
 960                                    htab->map.numa_node);
 961         if (!new) {
 962                 atomic_dec(&htab->count);
 963                 return ERR_PTR(-ENOMEM);
 964         }
 965         memcpy(new->key, key, key_size);
 966         new->sk = sk;
 967         new->hash = hash;
 968         return new;
 969 }
 970
 971 static int sock_hash_update_common(struct bpf_map *map, void *key,
 972                                    struct sock *sk, u64 flags)
 973 {
 974         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 975         u32 key_size = map->key_size, hash;
 976         struct bpf_shtab_elem *elem, *elem_new;
 977         struct bpf_shtab_bucket *bucket;
 978         struct sk_psock_link *link;
 979         struct sk_psock *psock;
 980         int ret;
 981
 982         WARN_ON_ONCE(!rcu_read_lock_held());
 983         if (unlikely(flags > BPF_EXIST))
 984                 return -EINVAL;
 985
 986         link = sk_psock_init_link();
 987         if (!link)
 988                 return -ENOMEM;
 989
 990         ret = sock_map_link(map, sk);
 991         if (ret < 0)
 992                 goto out_free;
 993
 994         psock = sk_psock(sk);
 995         WARN_ON_ONCE(!psock);
 996
 997         hash = sock_hash_bucket_hash(key, key_size);
 998         bucket = sock_hash_select_bucket(htab, hash);
 999
1000         raw_spin_lock_bh(&bucket->lock);
1001         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1002         if (elem && flags == BPF_NOEXIST) {
1003                 ret = -EEXIST;
1004                 goto out_unlock;
1005         } else if (!elem && flags == BPF_EXIST) {
1006                 ret = -ENOENT;
1007                 goto out_unlock;
1008         }
1009
1010         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1011         if (IS_ERR(elem_new)) {
1012                 ret = PTR_ERR(elem_new);
1013                 goto out_unlock;
1014         }
1015
1016         sock_map_add_link(psock, link, map, elem_new);
1017         /* Add new element to the head of the list, so that
1018          * concurrent search will find it before old elem.
1019          */
1020         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1021         if (elem) {
1022                 hlist_del_rcu(&elem->node);
1023                 sock_map_unref(elem->sk, elem);
1024                 sock_hash_free_elem(htab, elem);
1025         }
1026         raw_spin_unlock_bh(&bucket->lock);
1027         return 0;
1028 out_unlock:
1029         raw_spin_unlock_bh(&bucket->lock);
1030         sk_psock_put(sk, psock);
1031 out_free:
1032         sk_psock_free_link(link);
1033         return ret;
1034 }
1035
1036 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1037                                   void *key_next)
1038 {
1039         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1040         struct bpf_shtab_elem *elem, *elem_next;
1041         u32 hash, key_size = map->key_size;
1042         struct hlist_head *head;
1043         int i = 0;
1044
1045         if (!key)
1046                 goto find_first_elem;
1047         hash = sock_hash_bucket_hash(key, key_size);
1048         head = &sock_hash_select_bucket(htab, hash)->head;
1049         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1050         if (!elem)
1051                 goto find_first_elem;
1052
1053         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1054                                      struct bpf_shtab_elem, node);
1055         if (elem_next) {
1056                 memcpy(key_next, elem_next->key, key_size);
1057                 return 0;
1058         }
1059
1060         i = hash & (htab->buckets_num - 1);
1061         i++;
1062 find_first_elem:
1063         for (; i < htab->buckets_num; i++) {
1064                 head = &sock_hash_select_bucket(htab, i)->head;
1065                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1066                                              struct bpf_shtab_elem, node);
1067                 if (elem_next) {
1068                         memcpy(key_next, elem_next->key, key_size);
1069                         return 0;
1070                 }
1071         }
1072
1073         return -ENOENT;
1074 }
1075
1076 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1077 {
1078         struct bpf_shtab *htab;
1079         int i, err;
1080
1081         if (!capable(CAP_NET_ADMIN))
1082                 return ERR_PTR(-EPERM);
1083         if (attr->max_entries == 0 ||
1084             attr->key_size    == 0 ||
1085             (attr->value_size != sizeof(u32) &&
1086              attr->value_size != sizeof(u64)) ||
1087             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1088                 return ERR_PTR(-EINVAL);
1089         if (attr->key_size > MAX_BPF_STACK)
1090                 return ERR_PTR(-E2BIG);
1091
1092         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1093         if (!htab)
1094                 return ERR_PTR(-ENOMEM);
1095
1096         bpf_map_init_from_attr(&htab->map, attr);
1097
1098         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1099         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1100                           round_up(htab->map.key_size, 8);
1101         if (htab->buckets_num == 0 ||
1102             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1103                 err = -EINVAL;
1104                 goto free_htab;
1105         }
1106
1107         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1108                                            sizeof(struct bpf_shtab_bucket),
1109                                            htab->map.numa_node);
1110         if (!htab->buckets) {
1111                 err = -ENOMEM;
1112                 goto free_htab;
1113         }
1114
1115         for (i = 0; i < htab->buckets_num; i++) {
1116                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1117                 raw_spin_lock_init(&htab->buckets[i].lock);
1118         }
1119
1120         return &htab->map;
1121 free_htab:
1122         bpf_map_area_free(htab);
1123         return ERR_PTR(err);
1124 }
1125
1126 static void sock_hash_free(struct bpf_map *map)
1127 {
1128         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1129         struct bpf_shtab_bucket *bucket;
1130         struct hlist_head unlink_list;
1131         struct bpf_shtab_elem *elem;
1132         struct hlist_node *node;
1133         int i;
1134
1135         /* After the sync no updates or deletes will be in-flight so it
1136          * is safe to walk map and remove entries without risking a race
1137          * in EEXIST update case.
1138          */
1139         synchronize_rcu();
1140         for (i = 0; i < htab->buckets_num; i++) {
1141                 bucket = sock_hash_select_bucket(htab, i);
1142
1143                 /* We are racing with sock_hash_delete_from_link to
1144                  * enter the spin-lock critical section. Every socket on
1145                  * the list is still linked to sockhash. Since link
1146                  * exists, psock exists and holds a ref to socket. That
1147                  * lets us to grab a socket ref too.
1148                  */
1149                 raw_spin_lock_bh(&bucket->lock);
1150                 hlist_for_each_entry(elem, &bucket->head, node)
1151                         sock_hold(elem->sk);
1152                 hlist_move_list(&bucket->head, &unlink_list);
1153                 raw_spin_unlock_bh(&bucket->lock);
1154
1155                 /* Process removed entries out of atomic context to
1156                  * block for socket lock before deleting the psock's
1157                  * link to sockhash.
1158                  */
1159                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1160                         hlist_del(&elem->node);
1161                         lock_sock(elem->sk);
1162                         rcu_read_lock();
1163                         sock_map_unref(elem->sk, elem);
1164                         rcu_read_unlock();
1165                         release_sock(elem->sk);
1166                         sock_put(elem->sk);
1167                         sock_hash_free_elem(htab, elem);
1168                 }
1169         }
1170
1171         /* wait for psock readers accessing its map link */
1172         synchronize_rcu();
1173
1174         bpf_map_area_free(htab->buckets);
1175         bpf_map_area_free(htab);
1176 }
1177
1178 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1179 {
1180         struct sock *sk;
1181
1182         if (map->value_size != sizeof(u64))
1183                 return ERR_PTR(-ENOSPC);
1184
1185         sk = __sock_hash_lookup_elem(map, key);
1186         if (!sk)
1187                 return ERR_PTR(-ENOENT);
1188
1189         __sock_gen_cookie(sk);
1190         return &sk->sk_cookie;
1191 }
1192
1193 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1194 {
1195         struct sock *sk;
1196
1197         sk = __sock_hash_lookup_elem(map, key);
1198         if (!sk)
1199                 return NULL;
1200         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1201                 return NULL;
1202         return sk;
1203 }
1204
1205 static void sock_hash_release_progs(struct bpf_map *map)
1206 {
1207         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1208 }
1209
1210 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1211            struct bpf_map *, map, void *, key, u64, flags)
1212 {
1213         WARN_ON_ONCE(!rcu_read_lock_held());
1214
1215         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1216                    sock_map_op_okay(sops)))
1217                 return sock_hash_update_common(map, key, sops->sk, flags);
1218         return -EOPNOTSUPP;
1219 }
1220
1221 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1222         .func           = bpf_sock_hash_update,
1223         .gpl_only       = false,
1224         .pkt_access     = true,
1225         .ret_type       = RET_INTEGER,
1226         .arg1_type      = ARG_PTR_TO_CTX,
1227         .arg2_type      = ARG_CONST_MAP_PTR,
1228         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1229         .arg4_type      = ARG_ANYTHING,
1230 };
1231
1232 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1233            struct bpf_map *, map, void *, key, u64, flags)
1234 {
1235         struct sock *sk;
1236
1237         if (unlikely(flags & ~(BPF_F_INGRESS)))
1238                 return SK_DROP;
1239
1240         sk = __sock_hash_lookup_elem(map, key);
1241         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1242                 return SK_DROP;
1243
1244         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1245         return SK_PASS;
1246 }
1247
1248 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1249         .func           = bpf_sk_redirect_hash,
1250         .gpl_only       = false,
1251         .ret_type       = RET_INTEGER,
1252         .arg1_type      = ARG_PTR_TO_CTX,
1253         .arg2_type      = ARG_CONST_MAP_PTR,
1254         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1255         .arg4_type      = ARG_ANYTHING,
1256 };
1257
1258 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1259            struct bpf_map *, map, void *, key, u64, flags)
1260 {
1261         struct sock *sk;
1262
1263         if (unlikely(flags & ~(BPF_F_INGRESS)))
1264                 return SK_DROP;
1265
1266         sk = __sock_hash_lookup_elem(map, key);
1267         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1268                 return SK_DROP;
1269
1270         msg->flags = flags;
1271         msg->sk_redir = sk;
1272         return SK_PASS;
1273 }
1274
1275 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1276         .func           = bpf_msg_redirect_hash,
1277         .gpl_only       = false,
1278         .ret_type       = RET_INTEGER,
1279         .arg1_type      = ARG_PTR_TO_CTX,
1280         .arg2_type      = ARG_CONST_MAP_PTR,
1281         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1282         .arg4_type      = ARG_ANYTHING,
1283 };
1284
1285 struct sock_hash_seq_info {
1286         struct bpf_map *map;
1287         struct bpf_shtab *htab;
1288         u32 bucket_id;
1289 };
1290
1291 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1292                                      struct bpf_shtab_elem *prev_elem)
1293 {
1294         const struct bpf_shtab *htab = info->htab;
1295         struct bpf_shtab_bucket *bucket;
1296         struct bpf_shtab_elem *elem;
1297         struct hlist_node *node;
1298
1299         /* try to find next elem in the same bucket */
1300         if (prev_elem) {
1301                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1302                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1303                 if (elem)
1304                         return elem;
1305
1306                 /* no more elements, continue in the next bucket */
1307                 info->bucket_id++;
1308         }
1309
1310         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1311                 bucket = &htab->buckets[info->bucket_id];
1312                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1313                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1314                 if (elem)
1315                         return elem;
1316         }
1317
1318         return NULL;
1319 }
1320
1321 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1322         __acquires(rcu)
1323 {
1324         struct sock_hash_seq_info *info = seq->private;
1325
1326         if (*pos == 0)
1327                 ++*pos;
1328
1329         /* pairs with sock_hash_seq_stop */
1330         rcu_read_lock();
1331         return sock_hash_seq_find_next(info, NULL);
1332 }
1333
1334 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1335         __must_hold(rcu)
1336 {
1337         struct sock_hash_seq_info *info = seq->private;
1338
1339         ++*pos;
1340         return sock_hash_seq_find_next(info, v);
1341 }
1342
1343 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1344         __must_hold(rcu)
1345 {
1346         struct sock_hash_seq_info *info = seq->private;
1347         struct bpf_iter__sockmap ctx = {};
1348         struct bpf_shtab_elem *elem = v;
1349         struct bpf_iter_meta meta;
1350         struct bpf_prog *prog;
1351
1352         meta.seq = seq;
1353         prog = bpf_iter_get_info(&meta, !elem);
1354         if (!prog)
1355                 return 0;
1356
1357         ctx.meta = &meta;
1358         ctx.map = info->map;
1359         if (elem) {
1360                 ctx.key = elem->key;
1361                 ctx.sk = elem->sk;
1362         }
1363
1364         return bpf_iter_run_prog(prog, &ctx);
1365 }
1366
1367 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1368         __releases(rcu)
1369 {
1370         if (!v)
1371                 (void)sock_hash_seq_show(seq, NULL);
1372
1373         /* pairs with sock_hash_seq_start */
1374         rcu_read_unlock();
1375 }
1376
1377 static const struct seq_operations sock_hash_seq_ops = {
1378         .start  = sock_hash_seq_start,
1379         .next   = sock_hash_seq_next,
1380         .stop   = sock_hash_seq_stop,
1381         .show   = sock_hash_seq_show,
1382 };
1383
1384 static int sock_hash_init_seq_private(void *priv_data,
1385                                       struct bpf_iter_aux_info *aux)
1386 {
1387         struct sock_hash_seq_info *info = priv_data;
1388
1389         bpf_map_inc_with_uref(aux->map);
1390         info->map = aux->map;
1391         info->htab = container_of(aux->map, struct bpf_shtab, map);
1392         return 0;
1393 }
1394
1395 static void sock_hash_fini_seq_private(void *priv_data)
1396 {
1397         struct sock_hash_seq_info *info = priv_data;
1398
1399         bpf_map_put_with_uref(info->map);
1400 }
1401
1402 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1403         .seq_ops                = &sock_hash_seq_ops,
1404         .init_seq_private       = sock_hash_init_seq_private,
1405         .fini_seq_private       = sock_hash_fini_seq_private,
1406         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1407 };
1408
1409 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1410 const struct bpf_map_ops sock_hash_ops = {
1411         .map_meta_equal         = bpf_map_meta_equal,
1412         .map_alloc              = sock_hash_alloc,
1413         .map_free               = sock_hash_free,
1414         .map_get_next_key       = sock_hash_get_next_key,
1415         .map_update_elem        = sock_map_update_elem,
1416         .map_delete_elem        = sock_hash_delete_elem,
1417         .map_lookup_elem        = sock_hash_lookup,
1418         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1419         .map_release_uref       = sock_hash_release_progs,
1420         .map_check_btf          = map_check_no_btf,
1421         .map_btf_id             = &sock_hash_map_btf_ids[0],
1422         .iter_seq_info          = &sock_hash_iter_seq_info,
1423 };
1424
1425 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1426 {
1427         switch (map->map_type) {
1428         case BPF_MAP_TYPE_SOCKMAP:
1429                 return &container_of(map, struct bpf_stab, map)->progs;
1430         case BPF_MAP_TYPE_SOCKHASH:
1431                 return &container_of(map, struct bpf_shtab, map)->progs;
1432         default:
1433                 break;
1434         }
1435
1436         return NULL;
1437 }
1438
1439 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1440                                 u32 which)
1441 {
1442         struct sk_psock_progs *progs = sock_map_progs(map);
1443
1444         if (!progs)
1445                 return -EOPNOTSUPP;
1446
1447         switch (which) {
1448         case BPF_SK_MSG_VERDICT:
1449                 *pprog = &progs->msg_parser;
1450                 break;
1451 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1452         case BPF_SK_SKB_STREAM_PARSER:
1453                 *pprog = &progs->stream_parser;
1454                 break;
1455 #endif
1456         case BPF_SK_SKB_STREAM_VERDICT:
1457                 if (progs->skb_verdict)
1458                         return -EBUSY;
1459                 *pprog = &progs->stream_verdict;
1460                 break;
1461         case BPF_SK_SKB_VERDICT:
1462                 if (progs->stream_verdict)
1463                         return -EBUSY;
1464                 *pprog = &progs->skb_verdict;
1465                 break;
1466         default:
1467                 return -EOPNOTSUPP;
1468         }
1469
1470         return 0;
1471 }
1472
1473 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1474                                 struct bpf_prog *old, u32 which)
1475 {
1476         struct bpf_prog **pprog;
1477         int ret;
1478
1479         ret = sock_map_prog_lookup(map, &pprog, which);
1480         if (ret)
1481                 return ret;
1482
1483         if (old)
1484                 return psock_replace_prog(pprog, prog, old);
1485
1486         psock_set_prog(pprog, prog);
1487         return 0;
1488 }
1489
1490 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1491                             union bpf_attr __user *uattr)
1492 {
1493         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1494         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1495         struct bpf_prog **pprog;
1496         struct bpf_prog *prog;
1497         struct bpf_map *map;
1498         struct fd f;
1499         u32 id = 0;
1500         int ret;
1501
1502         if (attr->query.query_flags)
1503                 return -EINVAL;
1504
1505         f = fdget(ufd);
1506         map = __bpf_map_get(f);
1507         if (IS_ERR(map))
1508                 return PTR_ERR(map);
1509
1510         rcu_read_lock();
1511
1512         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1513         if (ret)
1514                 goto end;
1515
1516         prog = *pprog;
1517         prog_cnt = !prog ? 0 : 1;
1518
1519         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1520                 goto end;
1521
1522         /* we do not hold the refcnt, the bpf prog may be released
1523          * asynchronously and the id would be set to 0.
1524          */
1525         id = data_race(prog->aux->id);
1526         if (id == 0)
1527                 prog_cnt = 0;
1528
1529 end:
1530         rcu_read_unlock();
1531
1532         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1533             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1534             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1535                 ret = -EFAULT;
1536
1537         fdput(f);
1538         return ret;
1539 }
1540
1541 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1542 {
1543         switch (link->map->map_type) {
1544         case BPF_MAP_TYPE_SOCKMAP:
1545                 return sock_map_delete_from_link(link->map, sk,
1546                                                  link->link_raw);
1547         case BPF_MAP_TYPE_SOCKHASH:
1548                 return sock_hash_delete_from_link(link->map, sk,
1549                                                   link->link_raw);
1550         default:
1551                 break;
1552         }
1553 }
1554
1555 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1556 {
1557         struct sk_psock_link *link;
1558
1559         while ((link = sk_psock_link_pop(psock))) {
1560                 sock_map_unlink(sk, link);
1561                 sk_psock_free_link(link);
1562         }
1563 }
1564
1565 void sock_map_unhash(struct sock *sk)
1566 {
1567         void (*saved_unhash)(struct sock *sk);
1568         struct sk_psock *psock;
1569
1570         rcu_read_lock();
1571         psock = sk_psock(sk);
1572         if (unlikely(!psock)) {
1573                 rcu_read_unlock();
1574                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1575         } else {
1576                 saved_unhash = psock->saved_unhash;
1577                 sock_map_remove_links(sk, psock);
1578                 rcu_read_unlock();
1579         }
1580         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1581                 return;
1582         if (saved_unhash)
1583                 saved_unhash(sk);
1584 }
1585 EXPORT_SYMBOL_GPL(sock_map_unhash);
1586
1587 void sock_map_destroy(struct sock *sk)
1588 {
1589         void (*saved_destroy)(struct sock *sk);
1590         struct sk_psock *psock;
1591
1592         rcu_read_lock();
1593         psock = sk_psock_get(sk);
1594         if (unlikely(!psock)) {
1595                 rcu_read_unlock();
1596                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1597         } else {
1598                 saved_destroy = psock->saved_destroy;
1599                 sock_map_remove_links(sk, psock);
1600                 rcu_read_unlock();
1601                 sk_psock_stop(psock);
1602                 sk_psock_put(sk, psock);
1603         }
1604         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1605                 return;
1606         if (saved_destroy)
1607                 saved_destroy(sk);
1608 }
1609 EXPORT_SYMBOL_GPL(sock_map_destroy);
1610
1611 void sock_map_close(struct sock *sk, long timeout)
1612 {
1613         void (*saved_close)(struct sock *sk, long timeout);
1614         struct sk_psock *psock;
1615
1616         lock_sock(sk);
1617         rcu_read_lock();
1618         psock = sk_psock_get(sk);
1619         if (unlikely(!psock)) {
1620                 rcu_read_unlock();
1621                 release_sock(sk);
1622                 saved_close = READ_ONCE(sk->sk_prot)->close;
1623         } else {
1624                 saved_close = psock->saved_close;
1625                 sock_map_remove_links(sk, psock);
1626                 rcu_read_unlock();
1627                 sk_psock_stop(psock);
1628                 release_sock(sk);
1629                 cancel_work_sync(&psock->work);
1630                 sk_psock_put(sk, psock);
1631         }
1632         /* Make sure we do not recurse. This is a bug.
1633          * Leak the socket instead of crashing on a stack overflow.
1634          */
1635         if (WARN_ON_ONCE(saved_close == sock_map_close))
1636                 return;
1637         saved_close(sk, timeout);
1638 }
1639 EXPORT_SYMBOL_GPL(sock_map_close);
1640
1641 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1642                                        union bpf_iter_link_info *linfo,
1643                                        struct bpf_iter_aux_info *aux)
1644 {
1645         struct bpf_map *map;
1646         int err = -EINVAL;
1647
1648         if (!linfo->map.map_fd)
1649                 return -EBADF;
1650
1651         map = bpf_map_get_with_uref(linfo->map.map_fd);
1652         if (IS_ERR(map))
1653                 return PTR_ERR(map);
1654
1655         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1656             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1657                 goto put_map;
1658
1659         if (prog->aux->max_rdonly_access > map->key_size) {
1660                 err = -EACCES;
1661                 goto put_map;
1662         }
1663
1664         aux->map = map;
1665         return 0;
1666
1667 put_map:
1668         bpf_map_put_with_uref(map);
1669         return err;
1670 }
1671
1672 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1673 {
1674         bpf_map_put_with_uref(aux->map);
1675 }
1676
1677 static struct bpf_iter_reg sock_map_iter_reg = {
1678         .target                 = "sockmap",
1679         .attach_target          = sock_map_iter_attach_target,
1680         .detach_target          = sock_map_iter_detach_target,
1681         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1682         .fill_link_info         = bpf_iter_map_fill_link_info,
1683         .ctx_arg_info_size      = 2,
1684         .ctx_arg_info           = {
1685                 { offsetof(struct bpf_iter__sockmap, key),
1686                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1687                 { offsetof(struct bpf_iter__sockmap, sk),
1688                   PTR_TO_BTF_ID_OR_NULL },
1689         },
1690 };
1691
1692 static int __init bpf_sockmap_iter_init(void)
1693 {
1694         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1695                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1696         return bpf_iter_reg_target(&sock_map_iter_reg);
1697 }
1698 late_initcall(bpf_sockmap_iter_init);