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 /* This mutex is used to
  28  *  - protect race between prog/link attach/detach and link prog update, and
  29  *  - protect race between releasing and accessing map in bpf_link.
  30  * A single global mutex lock is used since it is expected contention is low.
  31  */
  32 static DEFINE_MUTEX(sockmap_mutex);
  33
  34 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
  35                                 struct bpf_prog *old, struct bpf_link *link,
  36                                 u32 which);
  37 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
  38
  39 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  40 {
  41         struct bpf_stab *stab;
  42
  43         if (attr->max_entries == 0 ||
  44             attr->key_size    != 4 ||
  45             (attr->value_size != sizeof(u32) &&
  46              attr->value_size != sizeof(u64)) ||
  47             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  48                 return ERR_PTR(-EINVAL);
  49
  50         stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
  51         if (!stab)
  52                 return ERR_PTR(-ENOMEM);
  53
  54         bpf_map_init_from_attr(&stab->map, attr);
  55         spin_lock_init(&stab->lock);
  56
  57         stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
  58                                        sizeof(struct sock *),
  59                                        stab->map.numa_node);
  60         if (!stab->sks) {
  61                 bpf_map_area_free(stab);
  62                 return ERR_PTR(-ENOMEM);
  63         }
  64
  65         return &stab->map;
  66 }
  67
  68 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  69 {
  70         struct bpf_map *map;
  71         int ret;
  72
  73         if (attr->attach_flags || attr->replace_bpf_fd)
  74                 return -EINVAL;
  75
  76         CLASS(fd, f)(attr->target_fd);
  77         map = __bpf_map_get(f);
  78         if (IS_ERR(map))
  79                 return PTR_ERR(map);
  80         mutex_lock(&sockmap_mutex);
  81         ret = sock_map_prog_update(map, prog, NULL, NULL, attr->attach_type);
  82         mutex_unlock(&sockmap_mutex);
  83         return ret;
  84 }
  85
  86 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  87 {
  88         struct bpf_prog *prog;
  89         struct bpf_map *map;
  90         int ret;
  91
  92         if (attr->attach_flags || attr->replace_bpf_fd)
  93                 return -EINVAL;
  94
  95         CLASS(fd, f)(attr->target_fd);
  96         map = __bpf_map_get(f);
  97         if (IS_ERR(map))
  98                 return PTR_ERR(map);
  99
 100         prog = bpf_prog_get(attr->attach_bpf_fd);
 101         if (IS_ERR(prog))
 102                 return PTR_ERR(prog);
 103
 104         if (prog->type != ptype) {
 105                 ret = -EINVAL;
 106                 goto put_prog;
 107         }
 108
 109         mutex_lock(&sockmap_mutex);
 110         ret = sock_map_prog_update(map, NULL, prog, NULL, attr->attach_type);
 111         mutex_unlock(&sockmap_mutex);
 112 put_prog:
 113         bpf_prog_put(prog);
 114         return ret;
 115 }
 116
 117 static void sock_map_sk_acquire(struct sock *sk)
 118         __acquires(&sk->sk_lock.slock)
 119 {
 120         lock_sock(sk);
 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         release_sock(sk);
 129 }
 130
 131 static void sock_map_add_link(struct sk_psock *psock,
 132                               struct sk_psock_link *link,
 133                               struct bpf_map *map, void *link_raw)
 134 {
 135         link->link_raw = link_raw;
 136         link->map = map;
 137         spin_lock_bh(&psock->link_lock);
 138         list_add_tail(&link->list, &psock->link);
 139         spin_unlock_bh(&psock->link_lock);
 140 }
 141
 142 static void sock_map_del_link(struct sock *sk,
 143                               struct sk_psock *psock, void *link_raw)
 144 {
 145         bool strp_stop = false, verdict_stop = false;
 146         struct sk_psock_link *link, *tmp;
 147
 148         spin_lock_bh(&psock->link_lock);
 149         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 150                 if (link->link_raw == link_raw) {
 151                         struct bpf_map *map = link->map;
 152                         struct sk_psock_progs *progs = sock_map_progs(map);
 153
 154                         if (psock->saved_data_ready && progs->stream_parser)
 155                                 strp_stop = true;
 156                         if (psock->saved_data_ready && progs->stream_verdict)
 157                                 verdict_stop = true;
 158                         if (psock->saved_data_ready && progs->skb_verdict)
 159                                 verdict_stop = true;
 160                         list_del(&link->list);
 161                         sk_psock_free_link(link);
 162                         break;
 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 = NULL;
 416         int err = 0;
 417
 418         spin_lock_bh(&stab->lock);
 419         if (!sk_test || sk_test == *psk)
 420                 sk = xchg(psk, NULL);
 421
 422         if (likely(sk))
 423                 sock_map_unref(sk, psk);
 424         else
 425                 err = -EINVAL;
 426
 427         spin_unlock_bh(&stab->lock);
 428         return err;
 429 }
 430
 431 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 432                                       void *link_raw)
 433 {
 434         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 435
 436         __sock_map_delete(stab, sk, link_raw);
 437 }
 438
 439 static long sock_map_delete_elem(struct bpf_map *map, void *key)
 440 {
 441         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 442         u32 i = *(u32 *)key;
 443         struct sock **psk;
 444
 445         if (unlikely(i >= map->max_entries))
 446                 return -EINVAL;
 447
 448         psk = &stab->sks[i];
 449         return __sock_map_delete(stab, NULL, psk);
 450 }
 451
 452 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 453 {
 454         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 455         u32 i = key ? *(u32 *)key : U32_MAX;
 456         u32 *key_next = next;
 457
 458         if (i == stab->map.max_entries - 1)
 459                 return -ENOENT;
 460         if (i >= stab->map.max_entries)
 461                 *key_next = 0;
 462         else
 463                 *key_next = i + 1;
 464         return 0;
 465 }
 466
 467 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 468                                   struct sock *sk, u64 flags)
 469 {
 470         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 471         struct sk_psock_link *link;
 472         struct sk_psock *psock;
 473         struct sock *osk;
 474         int ret;
 475
 476         WARN_ON_ONCE(!rcu_read_lock_held());
 477         if (unlikely(flags > BPF_EXIST))
 478                 return -EINVAL;
 479         if (unlikely(idx >= map->max_entries))
 480                 return -E2BIG;
 481
 482         link = sk_psock_init_link();
 483         if (!link)
 484                 return -ENOMEM;
 485
 486         ret = sock_map_link(map, sk);
 487         if (ret < 0)
 488                 goto out_free;
 489
 490         psock = sk_psock(sk);
 491         WARN_ON_ONCE(!psock);
 492
 493         spin_lock_bh(&stab->lock);
 494         osk = stab->sks[idx];
 495         if (osk && flags == BPF_NOEXIST) {
 496                 ret = -EEXIST;
 497                 goto out_unlock;
 498         } else if (!osk && flags == BPF_EXIST) {
 499                 ret = -ENOENT;
 500                 goto out_unlock;
 501         }
 502
 503         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 504         stab->sks[idx] = sk;
 505         if (osk)
 506                 sock_map_unref(osk, &stab->sks[idx]);
 507         spin_unlock_bh(&stab->lock);
 508         return 0;
 509 out_unlock:
 510         spin_unlock_bh(&stab->lock);
 511         if (psock)
 512                 sk_psock_put(sk, psock);
 513 out_free:
 514         sk_psock_free_link(link);
 515         return ret;
 516 }
 517
 518 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 519 {
 520         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 521                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 522                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 523 }
 524
 525 static bool sock_map_redirect_allowed(const struct sock *sk)
 526 {
 527         if (sk_is_tcp(sk))
 528                 return sk->sk_state != TCP_LISTEN;
 529         else
 530                 return sk->sk_state == TCP_ESTABLISHED;
 531 }
 532
 533 static bool sock_map_sk_is_suitable(const struct sock *sk)
 534 {
 535         return !!sk->sk_prot->psock_update_sk_prot;
 536 }
 537
 538 static bool sock_map_sk_state_allowed(const struct sock *sk)
 539 {
 540         if (sk_is_tcp(sk))
 541                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 542         if (sk_is_stream_unix(sk))
 543                 return (1 << sk->sk_state) & TCPF_ESTABLISHED;
 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 long 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         if ((flags & BPF_F_INGRESS) && sk_is_vsock(sk))
 651                 return SK_DROP;
 652
 653         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 654         return SK_PASS;
 655 }
 656
 657 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 658         .func           = bpf_sk_redirect_map,
 659         .gpl_only       = false,
 660         .ret_type       = RET_INTEGER,
 661         .arg1_type      = ARG_PTR_TO_CTX,
 662         .arg2_type      = ARG_CONST_MAP_PTR,
 663         .arg3_type      = ARG_ANYTHING,
 664         .arg4_type      = ARG_ANYTHING,
 665 };
 666
 667 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 668            struct bpf_map *, map, u32, key, u64, flags)
 669 {
 670         struct sock *sk;
 671
 672         if (unlikely(flags & ~(BPF_F_INGRESS)))
 673                 return SK_DROP;
 674
 675         sk = __sock_map_lookup_elem(map, key);
 676         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 677                 return SK_DROP;
 678         if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
 679                 return SK_DROP;
 680         if (sk_is_vsock(sk))
 681                 return SK_DROP;
 682
 683         msg->flags = flags;
 684         msg->sk_redir = sk;
 685         return SK_PASS;
 686 }
 687
 688 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 689         .func           = bpf_msg_redirect_map,
 690         .gpl_only       = false,
 691         .ret_type       = RET_INTEGER,
 692         .arg1_type      = ARG_PTR_TO_CTX,
 693         .arg2_type      = ARG_CONST_MAP_PTR,
 694         .arg3_type      = ARG_ANYTHING,
 695         .arg4_type      = ARG_ANYTHING,
 696 };
 697
 698 struct sock_map_seq_info {
 699         struct bpf_map *map;
 700         struct sock *sk;
 701         u32 index;
 702 };
 703
 704 struct bpf_iter__sockmap {
 705         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 706         __bpf_md_ptr(struct bpf_map *, map);
 707         __bpf_md_ptr(void *, key);
 708         __bpf_md_ptr(struct sock *, sk);
 709 };
 710
 711 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 712                      struct bpf_map *map, void *key,
 713                      struct sock *sk)
 714
 715 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 716 {
 717         if (unlikely(info->index >= info->map->max_entries))
 718                 return NULL;
 719
 720         info->sk = __sock_map_lookup_elem(info->map, info->index);
 721
 722         /* can't return sk directly, since that might be NULL */
 723         return info;
 724 }
 725
 726 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 727         __acquires(rcu)
 728 {
 729         struct sock_map_seq_info *info = seq->private;
 730
 731         if (*pos == 0)
 732                 ++*pos;
 733
 734         /* pairs with sock_map_seq_stop */
 735         rcu_read_lock();
 736         return sock_map_seq_lookup_elem(info);
 737 }
 738
 739 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 740         __must_hold(rcu)
 741 {
 742         struct sock_map_seq_info *info = seq->private;
 743
 744         ++*pos;
 745         ++info->index;
 746
 747         return sock_map_seq_lookup_elem(info);
 748 }
 749
 750 static int sock_map_seq_show(struct seq_file *seq, void *v)
 751         __must_hold(rcu)
 752 {
 753         struct sock_map_seq_info *info = seq->private;
 754         struct bpf_iter__sockmap ctx = {};
 755         struct bpf_iter_meta meta;
 756         struct bpf_prog *prog;
 757
 758         meta.seq = seq;
 759         prog = bpf_iter_get_info(&meta, !v);
 760         if (!prog)
 761                 return 0;
 762
 763         ctx.meta = &meta;
 764         ctx.map = info->map;
 765         if (v) {
 766                 ctx.key = &info->index;
 767                 ctx.sk = info->sk;
 768         }
 769
 770         return bpf_iter_run_prog(prog, &ctx);
 771 }
 772
 773 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 774         __releases(rcu)
 775 {
 776         if (!v)
 777                 (void)sock_map_seq_show(seq, NULL);
 778
 779         /* pairs with sock_map_seq_start */
 780         rcu_read_unlock();
 781 }
 782
 783 static const struct seq_operations sock_map_seq_ops = {
 784         .start  = sock_map_seq_start,
 785         .next   = sock_map_seq_next,
 786         .stop   = sock_map_seq_stop,
 787         .show   = sock_map_seq_show,
 788 };
 789
 790 static int sock_map_init_seq_private(void *priv_data,
 791                                      struct bpf_iter_aux_info *aux)
 792 {
 793         struct sock_map_seq_info *info = priv_data;
 794
 795         bpf_map_inc_with_uref(aux->map);
 796         info->map = aux->map;
 797         return 0;
 798 }
 799
 800 static void sock_map_fini_seq_private(void *priv_data)
 801 {
 802         struct sock_map_seq_info *info = priv_data;
 803
 804         bpf_map_put_with_uref(info->map);
 805 }
 806
 807 static u64 sock_map_mem_usage(const struct bpf_map *map)
 808 {
 809         u64 usage = sizeof(struct bpf_stab);
 810
 811         usage += (u64)map->max_entries * sizeof(struct sock *);
 812         return usage;
 813 }
 814
 815 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 816         .seq_ops                = &sock_map_seq_ops,
 817         .init_seq_private       = sock_map_init_seq_private,
 818         .fini_seq_private       = sock_map_fini_seq_private,
 819         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 820 };
 821
 822 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 823 const struct bpf_map_ops sock_map_ops = {
 824         .map_meta_equal         = bpf_map_meta_equal,
 825         .map_alloc              = sock_map_alloc,
 826         .map_free               = sock_map_free,
 827         .map_get_next_key       = sock_map_get_next_key,
 828         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 829         .map_update_elem        = sock_map_update_elem,
 830         .map_delete_elem        = sock_map_delete_elem,
 831         .map_lookup_elem        = sock_map_lookup,
 832         .map_release_uref       = sock_map_release_progs,
 833         .map_check_btf          = map_check_no_btf,
 834         .map_mem_usage          = sock_map_mem_usage,
 835         .map_btf_id             = &sock_map_btf_ids[0],
 836         .iter_seq_info          = &sock_map_iter_seq_info,
 837 };
 838
 839 struct bpf_shtab_elem {
 840         struct rcu_head rcu;
 841         u32 hash;
 842         struct sock *sk;
 843         struct hlist_node node;
 844         u8 key[];
 845 };
 846
 847 struct bpf_shtab_bucket {
 848         struct hlist_head head;
 849         spinlock_t lock;
 850 };
 851
 852 struct bpf_shtab {
 853         struct bpf_map map;
 854         struct bpf_shtab_bucket *buckets;
 855         u32 buckets_num;
 856         u32 elem_size;
 857         struct sk_psock_progs progs;
 858         atomic_t count;
 859 };
 860
 861 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 862 {
 863         return jhash(key, len, 0);
 864 }
 865
 866 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 867                                                         u32 hash)
 868 {
 869         return &htab->buckets[hash & (htab->buckets_num - 1)];
 870 }
 871
 872 static struct bpf_shtab_elem *
 873 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 874                           u32 key_size)
 875 {
 876         struct bpf_shtab_elem *elem;
 877
 878         hlist_for_each_entry_rcu(elem, head, node) {
 879                 if (elem->hash == hash &&
 880                     !memcmp(&elem->key, key, key_size))
 881                         return elem;
 882         }
 883
 884         return NULL;
 885 }
 886
 887 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 888 {
 889         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 890         u32 key_size = map->key_size, hash;
 891         struct bpf_shtab_bucket *bucket;
 892         struct bpf_shtab_elem *elem;
 893
 894         WARN_ON_ONCE(!rcu_read_lock_held());
 895
 896         hash = sock_hash_bucket_hash(key, key_size);
 897         bucket = sock_hash_select_bucket(htab, hash);
 898         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 899
 900         return elem ? elem->sk : NULL;
 901 }
 902
 903 static void sock_hash_free_elem(struct bpf_shtab *htab,
 904                                 struct bpf_shtab_elem *elem)
 905 {
 906         atomic_dec(&htab->count);
 907         kfree_rcu(elem, rcu);
 908 }
 909
 910 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 911                                        void *link_raw)
 912 {
 913         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 914         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 915         struct bpf_shtab_bucket *bucket;
 916
 917         WARN_ON_ONCE(!rcu_read_lock_held());
 918         bucket = sock_hash_select_bucket(htab, elem->hash);
 919
 920         /* elem may be deleted in parallel from the map, but access here
 921          * is okay since it's going away only after RCU grace period.
 922          * However, we need to check whether it's still present.
 923          */
 924         spin_lock_bh(&bucket->lock);
 925         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 926                                                elem->key, map->key_size);
 927         if (elem_probe && elem_probe == elem) {
 928                 hlist_del_rcu(&elem->node);
 929                 sock_map_unref(elem->sk, elem);
 930                 sock_hash_free_elem(htab, elem);
 931         }
 932         spin_unlock_bh(&bucket->lock);
 933 }
 934
 935 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
 936 {
 937         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 938         u32 hash, key_size = map->key_size;
 939         struct bpf_shtab_bucket *bucket;
 940         struct bpf_shtab_elem *elem;
 941         int ret = -ENOENT;
 942
 943         hash = sock_hash_bucket_hash(key, key_size);
 944         bucket = sock_hash_select_bucket(htab, hash);
 945
 946         spin_lock_bh(&bucket->lock);
 947         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 948         if (elem) {
 949                 hlist_del_rcu(&elem->node);
 950                 sock_map_unref(elem->sk, elem);
 951                 sock_hash_free_elem(htab, elem);
 952                 ret = 0;
 953         }
 954         spin_unlock_bh(&bucket->lock);
 955         return ret;
 956 }
 957
 958 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 959                                                    void *key, u32 key_size,
 960                                                    u32 hash, struct sock *sk,
 961                                                    struct bpf_shtab_elem *old)
 962 {
 963         struct bpf_shtab_elem *new;
 964
 965         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 966                 if (!old) {
 967                         atomic_dec(&htab->count);
 968                         return ERR_PTR(-E2BIG);
 969                 }
 970         }
 971
 972         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 973                                    GFP_ATOMIC | __GFP_NOWARN,
 974                                    htab->map.numa_node);
 975         if (!new) {
 976                 atomic_dec(&htab->count);
 977                 return ERR_PTR(-ENOMEM);
 978         }
 979         memcpy(new->key, key, key_size);
 980         new->sk = sk;
 981         new->hash = hash;
 982         return new;
 983 }
 984
 985 static int sock_hash_update_common(struct bpf_map *map, void *key,
 986                                    struct sock *sk, u64 flags)
 987 {
 988         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 989         u32 key_size = map->key_size, hash;
 990         struct bpf_shtab_elem *elem, *elem_new;
 991         struct bpf_shtab_bucket *bucket;
 992         struct sk_psock_link *link;
 993         struct sk_psock *psock;
 994         int ret;
 995
 996         WARN_ON_ONCE(!rcu_read_lock_held());
 997         if (unlikely(flags > BPF_EXIST))
 998                 return -EINVAL;
 999
1000         link = sk_psock_init_link();
1001         if (!link)
1002                 return -ENOMEM;
1003
1004         ret = sock_map_link(map, sk);
1005         if (ret < 0)
1006                 goto out_free;
1007
1008         psock = sk_psock(sk);
1009         WARN_ON_ONCE(!psock);
1010
1011         hash = sock_hash_bucket_hash(key, key_size);
1012         bucket = sock_hash_select_bucket(htab, hash);
1013
1014         spin_lock_bh(&bucket->lock);
1015         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1016         if (elem && flags == BPF_NOEXIST) {
1017                 ret = -EEXIST;
1018                 goto out_unlock;
1019         } else if (!elem && flags == BPF_EXIST) {
1020                 ret = -ENOENT;
1021                 goto out_unlock;
1022         }
1023
1024         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1025         if (IS_ERR(elem_new)) {
1026                 ret = PTR_ERR(elem_new);
1027                 goto out_unlock;
1028         }
1029
1030         sock_map_add_link(psock, link, map, elem_new);
1031         /* Add new element to the head of the list, so that
1032          * concurrent search will find it before old elem.
1033          */
1034         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1035         if (elem) {
1036                 hlist_del_rcu(&elem->node);
1037                 sock_map_unref(elem->sk, elem);
1038                 sock_hash_free_elem(htab, elem);
1039         }
1040         spin_unlock_bh(&bucket->lock);
1041         return 0;
1042 out_unlock:
1043         spin_unlock_bh(&bucket->lock);
1044         sk_psock_put(sk, psock);
1045 out_free:
1046         sk_psock_free_link(link);
1047         return ret;
1048 }
1049
1050 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1051                                   void *key_next)
1052 {
1053         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1054         struct bpf_shtab_elem *elem, *elem_next;
1055         u32 hash, key_size = map->key_size;
1056         struct hlist_head *head;
1057         int i = 0;
1058
1059         if (!key)
1060                 goto find_first_elem;
1061         hash = sock_hash_bucket_hash(key, key_size);
1062         head = &sock_hash_select_bucket(htab, hash)->head;
1063         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1064         if (!elem)
1065                 goto find_first_elem;
1066
1067         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1068                                      struct bpf_shtab_elem, node);
1069         if (elem_next) {
1070                 memcpy(key_next, elem_next->key, key_size);
1071                 return 0;
1072         }
1073
1074         i = hash & (htab->buckets_num - 1);
1075         i++;
1076 find_first_elem:
1077         for (; i < htab->buckets_num; i++) {
1078                 head = &sock_hash_select_bucket(htab, i)->head;
1079                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1080                                              struct bpf_shtab_elem, node);
1081                 if (elem_next) {
1082                         memcpy(key_next, elem_next->key, key_size);
1083                         return 0;
1084                 }
1085         }
1086
1087         return -ENOENT;
1088 }
1089
1090 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1091 {
1092         struct bpf_shtab *htab;
1093         int i, err;
1094
1095         if (attr->max_entries == 0 ||
1096             attr->key_size    == 0 ||
1097             (attr->value_size != sizeof(u32) &&
1098              attr->value_size != sizeof(u64)) ||
1099             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1100                 return ERR_PTR(-EINVAL);
1101         if (attr->key_size > MAX_BPF_STACK)
1102                 return ERR_PTR(-E2BIG);
1103
1104         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1105         if (!htab)
1106                 return ERR_PTR(-ENOMEM);
1107
1108         bpf_map_init_from_attr(&htab->map, attr);
1109
1110         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1111         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1112                           round_up(htab->map.key_size, 8);
1113         if (htab->buckets_num == 0 ||
1114             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1115                 err = -EINVAL;
1116                 goto free_htab;
1117         }
1118
1119         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1120                                            sizeof(struct bpf_shtab_bucket),
1121                                            htab->map.numa_node);
1122         if (!htab->buckets) {
1123                 err = -ENOMEM;
1124                 goto free_htab;
1125         }
1126
1127         for (i = 0; i < htab->buckets_num; i++) {
1128                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1129                 spin_lock_init(&htab->buckets[i].lock);
1130         }
1131
1132         return &htab->map;
1133 free_htab:
1134         bpf_map_area_free(htab);
1135         return ERR_PTR(err);
1136 }
1137
1138 static void sock_hash_free(struct bpf_map *map)
1139 {
1140         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1141         struct bpf_shtab_bucket *bucket;
1142         struct hlist_head unlink_list;
1143         struct bpf_shtab_elem *elem;
1144         struct hlist_node *node;
1145         int i;
1146
1147         /* After the sync no updates or deletes will be in-flight so it
1148          * is safe to walk map and remove entries without risking a race
1149          * in EEXIST update case.
1150          */
1151         synchronize_rcu();
1152         for (i = 0; i < htab->buckets_num; i++) {
1153                 bucket = sock_hash_select_bucket(htab, i);
1154
1155                 /* We are racing with sock_hash_delete_from_link to
1156                  * enter the spin-lock critical section. Every socket on
1157                  * the list is still linked to sockhash. Since link
1158                  * exists, psock exists and holds a ref to socket. That
1159                  * lets us to grab a socket ref too.
1160                  */
1161                 spin_lock_bh(&bucket->lock);
1162                 hlist_for_each_entry(elem, &bucket->head, node)
1163                         sock_hold(elem->sk);
1164                 hlist_move_list(&bucket->head, &unlink_list);
1165                 spin_unlock_bh(&bucket->lock);
1166
1167                 /* Process removed entries out of atomic context to
1168                  * block for socket lock before deleting the psock's
1169                  * link to sockhash.
1170                  */
1171                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1172                         hlist_del(&elem->node);
1173                         lock_sock(elem->sk);
1174                         rcu_read_lock();
1175                         sock_map_unref(elem->sk, elem);
1176                         rcu_read_unlock();
1177                         release_sock(elem->sk);
1178                         sock_put(elem->sk);
1179                         sock_hash_free_elem(htab, elem);
1180                 }
1181                 cond_resched();
1182         }
1183
1184         /* wait for psock readers accessing its map link */
1185         synchronize_rcu();
1186
1187         bpf_map_area_free(htab->buckets);
1188         bpf_map_area_free(htab);
1189 }
1190
1191 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1192 {
1193         struct sock *sk;
1194
1195         if (map->value_size != sizeof(u64))
1196                 return ERR_PTR(-ENOSPC);
1197
1198         sk = __sock_hash_lookup_elem(map, key);
1199         if (!sk)
1200                 return ERR_PTR(-ENOENT);
1201
1202         __sock_gen_cookie(sk);
1203         return &sk->sk_cookie;
1204 }
1205
1206 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1207 {
1208         struct sock *sk;
1209
1210         sk = __sock_hash_lookup_elem(map, key);
1211         if (!sk)
1212                 return NULL;
1213         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1214                 return NULL;
1215         return sk;
1216 }
1217
1218 static void sock_hash_release_progs(struct bpf_map *map)
1219 {
1220         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1221 }
1222
1223 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1224            struct bpf_map *, map, void *, key, u64, flags)
1225 {
1226         WARN_ON_ONCE(!rcu_read_lock_held());
1227
1228         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1229                    sock_map_op_okay(sops)))
1230                 return sock_hash_update_common(map, key, sops->sk, flags);
1231         return -EOPNOTSUPP;
1232 }
1233
1234 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1235         .func           = bpf_sock_hash_update,
1236         .gpl_only       = false,
1237         .pkt_access     = true,
1238         .ret_type       = RET_INTEGER,
1239         .arg1_type      = ARG_PTR_TO_CTX,
1240         .arg2_type      = ARG_CONST_MAP_PTR,
1241         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1242         .arg4_type      = ARG_ANYTHING,
1243 };
1244
1245 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1246            struct bpf_map *, map, void *, key, u64, flags)
1247 {
1248         struct sock *sk;
1249
1250         if (unlikely(flags & ~(BPF_F_INGRESS)))
1251                 return SK_DROP;
1252
1253         sk = __sock_hash_lookup_elem(map, key);
1254         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1255                 return SK_DROP;
1256         if ((flags & BPF_F_INGRESS) && sk_is_vsock(sk))
1257                 return SK_DROP;
1258
1259         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1260         return SK_PASS;
1261 }
1262
1263 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1264         .func           = bpf_sk_redirect_hash,
1265         .gpl_only       = false,
1266         .ret_type       = RET_INTEGER,
1267         .arg1_type      = ARG_PTR_TO_CTX,
1268         .arg2_type      = ARG_CONST_MAP_PTR,
1269         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1270         .arg4_type      = ARG_ANYTHING,
1271 };
1272
1273 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1274            struct bpf_map *, map, void *, key, u64, flags)
1275 {
1276         struct sock *sk;
1277
1278         if (unlikely(flags & ~(BPF_F_INGRESS)))
1279                 return SK_DROP;
1280
1281         sk = __sock_hash_lookup_elem(map, key);
1282         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1283                 return SK_DROP;
1284         if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1285                 return SK_DROP;
1286         if (sk_is_vsock(sk))
1287                 return SK_DROP;
1288
1289         msg->flags = flags;
1290         msg->sk_redir = sk;
1291         return SK_PASS;
1292 }
1293
1294 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1295         .func           = bpf_msg_redirect_hash,
1296         .gpl_only       = false,
1297         .ret_type       = RET_INTEGER,
1298         .arg1_type      = ARG_PTR_TO_CTX,
1299         .arg2_type      = ARG_CONST_MAP_PTR,
1300         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1301         .arg4_type      = ARG_ANYTHING,
1302 };
1303
1304 struct sock_hash_seq_info {
1305         struct bpf_map *map;
1306         struct bpf_shtab *htab;
1307         u32 bucket_id;
1308 };
1309
1310 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1311                                      struct bpf_shtab_elem *prev_elem)
1312 {
1313         const struct bpf_shtab *htab = info->htab;
1314         struct bpf_shtab_bucket *bucket;
1315         struct bpf_shtab_elem *elem;
1316         struct hlist_node *node;
1317
1318         /* try to find next elem in the same bucket */
1319         if (prev_elem) {
1320                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1321                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1322                 if (elem)
1323                         return elem;
1324
1325                 /* no more elements, continue in the next bucket */
1326                 info->bucket_id++;
1327         }
1328
1329         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1330                 bucket = &htab->buckets[info->bucket_id];
1331                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1332                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1333                 if (elem)
1334                         return elem;
1335         }
1336
1337         return NULL;
1338 }
1339
1340 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1341         __acquires(rcu)
1342 {
1343         struct sock_hash_seq_info *info = seq->private;
1344
1345         if (*pos == 0)
1346                 ++*pos;
1347
1348         /* pairs with sock_hash_seq_stop */
1349         rcu_read_lock();
1350         return sock_hash_seq_find_next(info, NULL);
1351 }
1352
1353 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1354         __must_hold(rcu)
1355 {
1356         struct sock_hash_seq_info *info = seq->private;
1357
1358         ++*pos;
1359         return sock_hash_seq_find_next(info, v);
1360 }
1361
1362 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1363         __must_hold(rcu)
1364 {
1365         struct sock_hash_seq_info *info = seq->private;
1366         struct bpf_iter__sockmap ctx = {};
1367         struct bpf_shtab_elem *elem = v;
1368         struct bpf_iter_meta meta;
1369         struct bpf_prog *prog;
1370
1371         meta.seq = seq;
1372         prog = bpf_iter_get_info(&meta, !elem);
1373         if (!prog)
1374                 return 0;
1375
1376         ctx.meta = &meta;
1377         ctx.map = info->map;
1378         if (elem) {
1379                 ctx.key = elem->key;
1380                 ctx.sk = elem->sk;
1381         }
1382
1383         return bpf_iter_run_prog(prog, &ctx);
1384 }
1385
1386 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1387         __releases(rcu)
1388 {
1389         if (!v)
1390                 (void)sock_hash_seq_show(seq, NULL);
1391
1392         /* pairs with sock_hash_seq_start */
1393         rcu_read_unlock();
1394 }
1395
1396 static const struct seq_operations sock_hash_seq_ops = {
1397         .start  = sock_hash_seq_start,
1398         .next   = sock_hash_seq_next,
1399         .stop   = sock_hash_seq_stop,
1400         .show   = sock_hash_seq_show,
1401 };
1402
1403 static int sock_hash_init_seq_private(void *priv_data,
1404                                       struct bpf_iter_aux_info *aux)
1405 {
1406         struct sock_hash_seq_info *info = priv_data;
1407
1408         bpf_map_inc_with_uref(aux->map);
1409         info->map = aux->map;
1410         info->htab = container_of(aux->map, struct bpf_shtab, map);
1411         return 0;
1412 }
1413
1414 static void sock_hash_fini_seq_private(void *priv_data)
1415 {
1416         struct sock_hash_seq_info *info = priv_data;
1417
1418         bpf_map_put_with_uref(info->map);
1419 }
1420
1421 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1422 {
1423         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1424         u64 usage = sizeof(*htab);
1425
1426         usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1427         usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1428         return usage;
1429 }
1430
1431 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1432         .seq_ops                = &sock_hash_seq_ops,
1433         .init_seq_private       = sock_hash_init_seq_private,
1434         .fini_seq_private       = sock_hash_fini_seq_private,
1435         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1436 };
1437
1438 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1439 const struct bpf_map_ops sock_hash_ops = {
1440         .map_meta_equal         = bpf_map_meta_equal,
1441         .map_alloc              = sock_hash_alloc,
1442         .map_free               = sock_hash_free,
1443         .map_get_next_key       = sock_hash_get_next_key,
1444         .map_update_elem        = sock_map_update_elem,
1445         .map_delete_elem        = sock_hash_delete_elem,
1446         .map_lookup_elem        = sock_hash_lookup,
1447         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1448         .map_release_uref       = sock_hash_release_progs,
1449         .map_check_btf          = map_check_no_btf,
1450         .map_mem_usage          = sock_hash_mem_usage,
1451         .map_btf_id             = &sock_hash_map_btf_ids[0],
1452         .iter_seq_info          = &sock_hash_iter_seq_info,
1453 };
1454
1455 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1456 {
1457         switch (map->map_type) {
1458         case BPF_MAP_TYPE_SOCKMAP:
1459                 return &container_of(map, struct bpf_stab, map)->progs;
1460         case BPF_MAP_TYPE_SOCKHASH:
1461                 return &container_of(map, struct bpf_shtab, map)->progs;
1462         default:
1463                 break;
1464         }
1465
1466         return NULL;
1467 }
1468
1469 static int sock_map_prog_link_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1470                                      struct bpf_link ***plink, u32 which)
1471 {
1472         struct sk_psock_progs *progs = sock_map_progs(map);
1473         struct bpf_prog **cur_pprog;
1474         struct bpf_link **cur_plink;
1475
1476         if (!progs)
1477                 return -EOPNOTSUPP;
1478
1479         switch (which) {
1480         case BPF_SK_MSG_VERDICT:
1481                 cur_pprog = &progs->msg_parser;
1482                 cur_plink = &progs->msg_parser_link;
1483                 break;
1484 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1485         case BPF_SK_SKB_STREAM_PARSER:
1486                 cur_pprog = &progs->stream_parser;
1487                 cur_plink = &progs->stream_parser_link;
1488                 break;
1489 #endif
1490         case BPF_SK_SKB_STREAM_VERDICT:
1491                 if (progs->skb_verdict)
1492                         return -EBUSY;
1493                 cur_pprog = &progs->stream_verdict;
1494                 cur_plink = &progs->stream_verdict_link;
1495                 break;
1496         case BPF_SK_SKB_VERDICT:
1497                 if (progs->stream_verdict)
1498                         return -EBUSY;
1499                 cur_pprog = &progs->skb_verdict;
1500                 cur_plink = &progs->skb_verdict_link;
1501                 break;
1502         default:
1503                 return -EOPNOTSUPP;
1504         }
1505
1506         *pprog = cur_pprog;
1507         if (plink)
1508                 *plink = cur_plink;
1509         return 0;
1510 }
1511
1512 /* Handle the following four cases:
1513  * prog_attach: prog != NULL, old == NULL, link == NULL
1514  * prog_detach: prog == NULL, old != NULL, link == NULL
1515  * link_attach: prog != NULL, old == NULL, link != NULL
1516  * link_detach: prog == NULL, old != NULL, link != NULL
1517  */
1518 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1519                                 struct bpf_prog *old, struct bpf_link *link,
1520                                 u32 which)
1521 {
1522         struct bpf_prog **pprog;
1523         struct bpf_link **plink;
1524         int ret;
1525
1526         ret = sock_map_prog_link_lookup(map, &pprog, &plink, which);
1527         if (ret)
1528                 return ret;
1529
1530         /* for prog_attach/prog_detach/link_attach, return error if a bpf_link
1531          * exists for that prog.
1532          */
1533         if ((!link || prog) && *plink)
1534                 return -EBUSY;
1535
1536         if (old) {
1537                 ret = psock_replace_prog(pprog, prog, old);
1538                 if (!ret)
1539                         *plink = NULL;
1540         } else {
1541                 psock_set_prog(pprog, prog);
1542                 if (link)
1543                         *plink = link;
1544         }
1545
1546         return ret;
1547 }
1548
1549 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1550                             union bpf_attr __user *uattr)
1551 {
1552         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1553         u32 prog_cnt = 0, flags = 0;
1554         struct bpf_prog **pprog;
1555         struct bpf_prog *prog;
1556         struct bpf_map *map;
1557         u32 id = 0;
1558         int ret;
1559
1560         if (attr->query.query_flags)
1561                 return -EINVAL;
1562
1563         CLASS(fd, f)(attr->target_fd);
1564         map = __bpf_map_get(f);
1565         if (IS_ERR(map))
1566                 return PTR_ERR(map);
1567
1568         rcu_read_lock();
1569
1570         ret = sock_map_prog_link_lookup(map, &pprog, NULL, attr->query.attach_type);
1571         if (ret)
1572                 goto end;
1573
1574         prog = *pprog;
1575         prog_cnt = !prog ? 0 : 1;
1576
1577         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1578                 goto end;
1579
1580         /* we do not hold the refcnt, the bpf prog may be released
1581          * asynchronously and the id would be set to 0.
1582          */
1583         id = data_race(prog->aux->id);
1584         if (id == 0)
1585                 prog_cnt = 0;
1586
1587 end:
1588         rcu_read_unlock();
1589
1590         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1591             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1592             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1593                 ret = -EFAULT;
1594
1595         return ret;
1596 }
1597
1598 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1599 {
1600         switch (link->map->map_type) {
1601         case BPF_MAP_TYPE_SOCKMAP:
1602                 return sock_map_delete_from_link(link->map, sk,
1603                                                  link->link_raw);
1604         case BPF_MAP_TYPE_SOCKHASH:
1605                 return sock_hash_delete_from_link(link->map, sk,
1606                                                   link->link_raw);
1607         default:
1608                 break;
1609         }
1610 }
1611
1612 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1613 {
1614         struct sk_psock_link *link;
1615
1616         while ((link = sk_psock_link_pop(psock))) {
1617                 sock_map_unlink(sk, link);
1618                 sk_psock_free_link(link);
1619         }
1620 }
1621
1622 void sock_map_unhash(struct sock *sk)
1623 {
1624         void (*saved_unhash)(struct sock *sk);
1625         struct sk_psock *psock;
1626
1627         rcu_read_lock();
1628         psock = sk_psock(sk);
1629         if (unlikely(!psock)) {
1630                 rcu_read_unlock();
1631                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1632         } else {
1633                 saved_unhash = psock->saved_unhash;
1634                 sock_map_remove_links(sk, psock);
1635                 rcu_read_unlock();
1636         }
1637         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1638                 return;
1639         if (saved_unhash)
1640                 saved_unhash(sk);
1641 }
1642 EXPORT_SYMBOL_GPL(sock_map_unhash);
1643
1644 void sock_map_destroy(struct sock *sk)
1645 {
1646         void (*saved_destroy)(struct sock *sk);
1647         struct sk_psock *psock;
1648
1649         rcu_read_lock();
1650         psock = sk_psock_get(sk);
1651         if (unlikely(!psock)) {
1652                 rcu_read_unlock();
1653                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1654         } else {
1655                 saved_destroy = psock->saved_destroy;
1656                 sock_map_remove_links(sk, psock);
1657                 rcu_read_unlock();
1658                 sk_psock_stop(psock);
1659                 sk_psock_put(sk, psock);
1660         }
1661         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1662                 return;
1663         if (saved_destroy)
1664                 saved_destroy(sk);
1665 }
1666 EXPORT_SYMBOL_GPL(sock_map_destroy);
1667
1668 void sock_map_close(struct sock *sk, long timeout)
1669 {
1670         void (*saved_close)(struct sock *sk, long timeout);
1671         struct sk_psock *psock;
1672
1673         lock_sock(sk);
1674         rcu_read_lock();
1675         psock = sk_psock(sk);
1676         if (likely(psock)) {
1677                 saved_close = psock->saved_close;
1678                 sock_map_remove_links(sk, psock);
1679                 psock = sk_psock_get(sk);
1680                 if (unlikely(!psock))
1681                         goto no_psock;
1682                 rcu_read_unlock();
1683                 sk_psock_stop(psock);
1684                 release_sock(sk);
1685                 cancel_delayed_work_sync(&psock->work);
1686                 sk_psock_put(sk, psock);
1687         } else {
1688                 saved_close = READ_ONCE(sk->sk_prot)->close;
1689 no_psock:
1690                 rcu_read_unlock();
1691                 release_sock(sk);
1692         }
1693
1694         /* Make sure we do not recurse. This is a bug.
1695          * Leak the socket instead of crashing on a stack overflow.
1696          */
1697         if (WARN_ON_ONCE(saved_close == sock_map_close))
1698                 return;
1699         saved_close(sk, timeout);
1700 }
1701 EXPORT_SYMBOL_GPL(sock_map_close);
1702
1703 struct sockmap_link {
1704         struct bpf_link link;
1705         struct bpf_map *map;
1706         enum bpf_attach_type attach_type;
1707 };
1708
1709 static void sock_map_link_release(struct bpf_link *link)
1710 {
1711         struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1712
1713         mutex_lock(&sockmap_mutex);
1714         if (!sockmap_link->map)
1715                 goto out;
1716
1717         WARN_ON_ONCE(sock_map_prog_update(sockmap_link->map, NULL, link->prog, link,
1718                                           sockmap_link->attach_type));
1719
1720         bpf_map_put_with_uref(sockmap_link->map);
1721         sockmap_link->map = NULL;
1722 out:
1723         mutex_unlock(&sockmap_mutex);
1724 }
1725
1726 static int sock_map_link_detach(struct bpf_link *link)
1727 {
1728         sock_map_link_release(link);
1729         return 0;
1730 }
1731
1732 static void sock_map_link_dealloc(struct bpf_link *link)
1733 {
1734         kfree(link);
1735 }
1736
1737 /* Handle the following two cases:
1738  * case 1: link != NULL, prog != NULL, old != NULL
1739  * case 2: link != NULL, prog != NULL, old == NULL
1740  */
1741 static int sock_map_link_update_prog(struct bpf_link *link,
1742                                      struct bpf_prog *prog,
1743                                      struct bpf_prog *old)
1744 {
1745         const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1746         struct bpf_prog **pprog, *old_link_prog;
1747         struct bpf_link **plink;
1748         int ret = 0;
1749
1750         mutex_lock(&sockmap_mutex);
1751
1752         /* If old prog is not NULL, ensure old prog is the same as link->prog. */
1753         if (old && link->prog != old) {
1754                 ret = -EPERM;
1755                 goto out;
1756         }
1757         /* Ensure link->prog has the same type/attach_type as the new prog. */
1758         if (link->prog->type != prog->type ||
1759             link->prog->expected_attach_type != prog->expected_attach_type) {
1760                 ret = -EINVAL;
1761                 goto out;
1762         }
1763         if (!sockmap_link->map) {
1764                 ret = -ENOLINK;
1765                 goto out;
1766         }
1767
1768         ret = sock_map_prog_link_lookup(sockmap_link->map, &pprog, &plink,
1769                                         sockmap_link->attach_type);
1770         if (ret)
1771                 goto out;
1772
1773         /* return error if the stored bpf_link does not match the incoming bpf_link. */
1774         if (link != *plink) {
1775                 ret = -EBUSY;
1776                 goto out;
1777         }
1778
1779         if (old) {
1780                 ret = psock_replace_prog(pprog, prog, old);
1781                 if (ret)
1782                         goto out;
1783         } else {
1784                 psock_set_prog(pprog, prog);
1785         }
1786
1787         bpf_prog_inc(prog);
1788         old_link_prog = xchg(&link->prog, prog);
1789         bpf_prog_put(old_link_prog);
1790
1791 out:
1792         mutex_unlock(&sockmap_mutex);
1793         return ret;
1794 }
1795
1796 static u32 sock_map_link_get_map_id(const struct sockmap_link *sockmap_link)
1797 {
1798         u32 map_id = 0;
1799
1800         mutex_lock(&sockmap_mutex);
1801         if (sockmap_link->map)
1802                 map_id = sockmap_link->map->id;
1803         mutex_unlock(&sockmap_mutex);
1804         return map_id;
1805 }
1806
1807 static int sock_map_link_fill_info(const struct bpf_link *link,
1808                                    struct bpf_link_info *info)
1809 {
1810         const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1811         u32 map_id = sock_map_link_get_map_id(sockmap_link);
1812
1813         info->sockmap.map_id = map_id;
1814         info->sockmap.attach_type = sockmap_link->attach_type;
1815         return 0;
1816 }
1817
1818 static void sock_map_link_show_fdinfo(const struct bpf_link *link,
1819                                       struct seq_file *seq)
1820 {
1821         const struct sockmap_link *sockmap_link = container_of(link, struct sockmap_link, link);
1822         u32 map_id = sock_map_link_get_map_id(sockmap_link);
1823
1824         seq_printf(seq, "map_id:\t%u\n", map_id);
1825         seq_printf(seq, "attach_type:\t%u\n", sockmap_link->attach_type);
1826 }
1827
1828 static const struct bpf_link_ops sock_map_link_ops = {
1829         .release = sock_map_link_release,
1830         .dealloc = sock_map_link_dealloc,
1831         .detach = sock_map_link_detach,
1832         .update_prog = sock_map_link_update_prog,
1833         .fill_link_info = sock_map_link_fill_info,
1834         .show_fdinfo = sock_map_link_show_fdinfo,
1835 };
1836
1837 int sock_map_link_create(const union bpf_attr *attr, struct bpf_prog *prog)
1838 {
1839         struct bpf_link_primer link_primer;
1840         struct sockmap_link *sockmap_link;
1841         enum bpf_attach_type attach_type;
1842         struct bpf_map *map;
1843         int ret;
1844
1845         if (attr->link_create.flags)
1846                 return -EINVAL;
1847
1848         map = bpf_map_get_with_uref(attr->link_create.target_fd);
1849         if (IS_ERR(map))
1850                 return PTR_ERR(map);
1851         if (map->map_type != BPF_MAP_TYPE_SOCKMAP && map->map_type != BPF_MAP_TYPE_SOCKHASH) {
1852                 ret = -EINVAL;
1853                 goto out;
1854         }
1855
1856         sockmap_link = kzalloc(sizeof(*sockmap_link), GFP_USER);
1857         if (!sockmap_link) {
1858                 ret = -ENOMEM;
1859                 goto out;
1860         }
1861
1862         attach_type = attr->link_create.attach_type;
1863         bpf_link_init(&sockmap_link->link, BPF_LINK_TYPE_SOCKMAP, &sock_map_link_ops, prog);
1864         sockmap_link->map = map;
1865         sockmap_link->attach_type = attach_type;
1866
1867         ret = bpf_link_prime(&sockmap_link->link, &link_primer);
1868         if (ret) {
1869                 kfree(sockmap_link);
1870                 goto out;
1871         }
1872
1873         mutex_lock(&sockmap_mutex);
1874         ret = sock_map_prog_update(map, prog, NULL, &sockmap_link->link, attach_type);
1875         mutex_unlock(&sockmap_mutex);
1876         if (ret) {
1877                 bpf_link_cleanup(&link_primer);
1878                 goto out;
1879         }
1880
1881         /* Increase refcnt for the prog since when old prog is replaced with
1882          * psock_replace_prog() and psock_set_prog() its refcnt will be decreased.
1883          *
1884          * Actually, we do not need to increase refcnt for the prog since bpf_link
1885          * will hold a reference. But in order to have less complexity w.r.t.
1886          * replacing/setting prog, let us increase the refcnt to make things simpler.
1887          */
1888         bpf_prog_inc(prog);
1889
1890         return bpf_link_settle(&link_primer);
1891
1892 out:
1893         bpf_map_put_with_uref(map);
1894         return ret;
1895 }
1896
1897 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1898                                        union bpf_iter_link_info *linfo,
1899                                        struct bpf_iter_aux_info *aux)
1900 {
1901         struct bpf_map *map;
1902         int err = -EINVAL;
1903
1904         if (!linfo->map.map_fd)
1905                 return -EBADF;
1906
1907         map = bpf_map_get_with_uref(linfo->map.map_fd);
1908         if (IS_ERR(map))
1909                 return PTR_ERR(map);
1910
1911         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1912             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1913                 goto put_map;
1914
1915         if (prog->aux->max_rdonly_access > map->key_size) {
1916                 err = -EACCES;
1917                 goto put_map;
1918         }
1919
1920         aux->map = map;
1921         return 0;
1922
1923 put_map:
1924         bpf_map_put_with_uref(map);
1925         return err;
1926 }
1927
1928 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1929 {
1930         bpf_map_put_with_uref(aux->map);
1931 }
1932
1933 static struct bpf_iter_reg sock_map_iter_reg = {
1934         .target                 = "sockmap",
1935         .attach_target          = sock_map_iter_attach_target,
1936         .detach_target          = sock_map_iter_detach_target,
1937         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1938         .fill_link_info         = bpf_iter_map_fill_link_info,
1939         .ctx_arg_info_size      = 2,
1940         .ctx_arg_info           = {
1941                 { offsetof(struct bpf_iter__sockmap, key),
1942                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1943                 { offsetof(struct bpf_iter__sockmap, sk),
1944                   PTR_TO_BTF_ID_OR_NULL },
1945         },
1946 };
1947
1948 static int __init bpf_sockmap_iter_init(void)
1949 {
1950         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1951                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1952         return bpf_iter_reg_target(&sock_map_iter_reg);
1953 }
1954 late_initcall(bpf_sockmap_iter_init);