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
 418         raw_spin_lock_bh(&stab->lock);
 419         sk = *psk;
 420         if (!sk_test || sk_test == sk)
 421                 sk = xchg(psk, NULL);
 422
 423         if (likely(sk))
 424                 sock_map_unref(sk, psk);
 425         else
 426                 err = -EINVAL;
 427
 428         raw_spin_unlock_bh(&stab->lock);
 429         return err;
 430 }
 431
 432 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 433                                       void *link_raw)
 434 {
 435         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 436
 437         __sock_map_delete(stab, sk, link_raw);
 438 }
 439
 440 static long sock_map_delete_elem(struct bpf_map *map, void *key)
 441 {
 442         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 443         u32 i = *(u32 *)key;
 444         struct sock **psk;
 445
 446         if (unlikely(i >= map->max_entries))
 447                 return -EINVAL;
 448
 449         psk = &stab->sks[i];
 450         return __sock_map_delete(stab, NULL, psk);
 451 }
 452
 453 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 454 {
 455         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 456         u32 i = key ? *(u32 *)key : U32_MAX;
 457         u32 *key_next = next;
 458
 459         if (i == stab->map.max_entries - 1)
 460                 return -ENOENT;
 461         if (i >= stab->map.max_entries)
 462                 *key_next = 0;
 463         else
 464                 *key_next = i + 1;
 465         return 0;
 466 }
 467
 468 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 469                                   struct sock *sk, u64 flags)
 470 {
 471         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 472         struct sk_psock_link *link;
 473         struct sk_psock *psock;
 474         struct sock *osk;
 475         int ret;
 476
 477         WARN_ON_ONCE(!rcu_read_lock_held());
 478         if (unlikely(flags > BPF_EXIST))
 479                 return -EINVAL;
 480         if (unlikely(idx >= map->max_entries))
 481                 return -E2BIG;
 482
 483         link = sk_psock_init_link();
 484         if (!link)
 485                 return -ENOMEM;
 486
 487         ret = sock_map_link(map, sk);
 488         if (ret < 0)
 489                 goto out_free;
 490
 491         psock = sk_psock(sk);
 492         WARN_ON_ONCE(!psock);
 493
 494         raw_spin_lock_bh(&stab->lock);
 495         osk = stab->sks[idx];
 496         if (osk && flags == BPF_NOEXIST) {
 497                 ret = -EEXIST;
 498                 goto out_unlock;
 499         } else if (!osk && flags == BPF_EXIST) {
 500                 ret = -ENOENT;
 501                 goto out_unlock;
 502         }
 503
 504         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 505         stab->sks[idx] = sk;
 506         if (osk)
 507                 sock_map_unref(osk, &stab->sks[idx]);
 508         raw_spin_unlock_bh(&stab->lock);
 509         return 0;
 510 out_unlock:
 511         raw_spin_unlock_bh(&stab->lock);
 512         if (psock)
 513                 sk_psock_put(sk, psock);
 514 out_free:
 515         sk_psock_free_link(link);
 516         return ret;
 517 }
 518
 519 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 520 {
 521         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 522                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 523                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 524 }
 525
 526 static bool sock_map_redirect_allowed(const struct sock *sk)
 527 {
 528         if (sk_is_tcp(sk))
 529                 return sk->sk_state != TCP_LISTEN;
 530         else
 531                 return sk->sk_state == TCP_ESTABLISHED;
 532 }
 533
 534 static bool sock_map_sk_is_suitable(const struct sock *sk)
 535 {
 536         return !!sk->sk_prot->psock_update_sk_prot;
 537 }
 538
 539 static bool sock_map_sk_state_allowed(const struct sock *sk)
 540 {
 541         if (sk_is_tcp(sk))
 542                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 543         return true;
 544 }
 545
 546 static int sock_hash_update_common(struct bpf_map *map, void *key,
 547                                    struct sock *sk, u64 flags);
 548
 549 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 550                              u64 flags)
 551 {
 552         struct socket *sock;
 553         struct sock *sk;
 554         int ret;
 555         u64 ufd;
 556
 557         if (map->value_size == sizeof(u64))
 558                 ufd = *(u64 *)value;
 559         else
 560                 ufd = *(u32 *)value;
 561         if (ufd > S32_MAX)
 562                 return -EINVAL;
 563
 564         sock = sockfd_lookup(ufd, &ret);
 565         if (!sock)
 566                 return ret;
 567         sk = sock->sk;
 568         if (!sk) {
 569                 ret = -EINVAL;
 570                 goto out;
 571         }
 572         if (!sock_map_sk_is_suitable(sk)) {
 573                 ret = -EOPNOTSUPP;
 574                 goto out;
 575         }
 576
 577         sock_map_sk_acquire(sk);
 578         if (!sock_map_sk_state_allowed(sk))
 579                 ret = -EOPNOTSUPP;
 580         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 581                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 582         else
 583                 ret = sock_hash_update_common(map, key, sk, flags);
 584         sock_map_sk_release(sk);
 585 out:
 586         sockfd_put(sock);
 587         return ret;
 588 }
 589
 590 static long sock_map_update_elem(struct bpf_map *map, void *key,
 591                                  void *value, u64 flags)
 592 {
 593         struct sock *sk = (struct sock *)value;
 594         int ret;
 595
 596         if (unlikely(!sk || !sk_fullsock(sk)))
 597                 return -EINVAL;
 598
 599         if (!sock_map_sk_is_suitable(sk))
 600                 return -EOPNOTSUPP;
 601
 602         local_bh_disable();
 603         bh_lock_sock(sk);
 604         if (!sock_map_sk_state_allowed(sk))
 605                 ret = -EOPNOTSUPP;
 606         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 607                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 608         else
 609                 ret = sock_hash_update_common(map, key, sk, flags);
 610         bh_unlock_sock(sk);
 611         local_bh_enable();
 612         return ret;
 613 }
 614
 615 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 616            struct bpf_map *, map, void *, key, u64, flags)
 617 {
 618         WARN_ON_ONCE(!rcu_read_lock_held());
 619
 620         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 621                    sock_map_op_okay(sops)))
 622                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 623                                               flags);
 624         return -EOPNOTSUPP;
 625 }
 626
 627 const struct bpf_func_proto bpf_sock_map_update_proto = {
 628         .func           = bpf_sock_map_update,
 629         .gpl_only       = false,
 630         .pkt_access     = true,
 631         .ret_type       = RET_INTEGER,
 632         .arg1_type      = ARG_PTR_TO_CTX,
 633         .arg2_type      = ARG_CONST_MAP_PTR,
 634         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 635         .arg4_type      = ARG_ANYTHING,
 636 };
 637
 638 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 639            struct bpf_map *, map, u32, key, u64, flags)
 640 {
 641         struct sock *sk;
 642
 643         if (unlikely(flags & ~(BPF_F_INGRESS)))
 644                 return SK_DROP;
 645
 646         sk = __sock_map_lookup_elem(map, key);
 647         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 648                 return SK_DROP;
 649
 650         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 651         return SK_PASS;
 652 }
 653
 654 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 655         .func           = bpf_sk_redirect_map,
 656         .gpl_only       = false,
 657         .ret_type       = RET_INTEGER,
 658         .arg1_type      = ARG_PTR_TO_CTX,
 659         .arg2_type      = ARG_CONST_MAP_PTR,
 660         .arg3_type      = ARG_ANYTHING,
 661         .arg4_type      = ARG_ANYTHING,
 662 };
 663
 664 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 665            struct bpf_map *, map, u32, key, u64, flags)
 666 {
 667         struct sock *sk;
 668
 669         if (unlikely(flags & ~(BPF_F_INGRESS)))
 670                 return SK_DROP;
 671
 672         sk = __sock_map_lookup_elem(map, key);
 673         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 674                 return SK_DROP;
 675
 676         msg->flags = flags;
 677         msg->sk_redir = sk;
 678         return SK_PASS;
 679 }
 680
 681 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 682         .func           = bpf_msg_redirect_map,
 683         .gpl_only       = false,
 684         .ret_type       = RET_INTEGER,
 685         .arg1_type      = ARG_PTR_TO_CTX,
 686         .arg2_type      = ARG_CONST_MAP_PTR,
 687         .arg3_type      = ARG_ANYTHING,
 688         .arg4_type      = ARG_ANYTHING,
 689 };
 690
 691 struct sock_map_seq_info {
 692         struct bpf_map *map;
 693         struct sock *sk;
 694         u32 index;
 695 };
 696
 697 struct bpf_iter__sockmap {
 698         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 699         __bpf_md_ptr(struct bpf_map *, map);
 700         __bpf_md_ptr(void *, key);
 701         __bpf_md_ptr(struct sock *, sk);
 702 };
 703
 704 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 705                      struct bpf_map *map, void *key,
 706                      struct sock *sk)
 707
 708 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 709 {
 710         if (unlikely(info->index >= info->map->max_entries))
 711                 return NULL;
 712
 713         info->sk = __sock_map_lookup_elem(info->map, info->index);
 714
 715         /* can't return sk directly, since that might be NULL */
 716         return info;
 717 }
 718
 719 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 720         __acquires(rcu)
 721 {
 722         struct sock_map_seq_info *info = seq->private;
 723
 724         if (*pos == 0)
 725                 ++*pos;
 726
 727         /* pairs with sock_map_seq_stop */
 728         rcu_read_lock();
 729         return sock_map_seq_lookup_elem(info);
 730 }
 731
 732 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 733         __must_hold(rcu)
 734 {
 735         struct sock_map_seq_info *info = seq->private;
 736
 737         ++*pos;
 738         ++info->index;
 739
 740         return sock_map_seq_lookup_elem(info);
 741 }
 742
 743 static int sock_map_seq_show(struct seq_file *seq, void *v)
 744         __must_hold(rcu)
 745 {
 746         struct sock_map_seq_info *info = seq->private;
 747         struct bpf_iter__sockmap ctx = {};
 748         struct bpf_iter_meta meta;
 749         struct bpf_prog *prog;
 750
 751         meta.seq = seq;
 752         prog = bpf_iter_get_info(&meta, !v);
 753         if (!prog)
 754                 return 0;
 755
 756         ctx.meta = &meta;
 757         ctx.map = info->map;
 758         if (v) {
 759                 ctx.key = &info->index;
 760                 ctx.sk = info->sk;
 761         }
 762
 763         return bpf_iter_run_prog(prog, &ctx);
 764 }
 765
 766 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 767         __releases(rcu)
 768 {
 769         if (!v)
 770                 (void)sock_map_seq_show(seq, NULL);
 771
 772         /* pairs with sock_map_seq_start */
 773         rcu_read_unlock();
 774 }
 775
 776 static const struct seq_operations sock_map_seq_ops = {
 777         .start  = sock_map_seq_start,
 778         .next   = sock_map_seq_next,
 779         .stop   = sock_map_seq_stop,
 780         .show   = sock_map_seq_show,
 781 };
 782
 783 static int sock_map_init_seq_private(void *priv_data,
 784                                      struct bpf_iter_aux_info *aux)
 785 {
 786         struct sock_map_seq_info *info = priv_data;
 787
 788         bpf_map_inc_with_uref(aux->map);
 789         info->map = aux->map;
 790         return 0;
 791 }
 792
 793 static void sock_map_fini_seq_private(void *priv_data)
 794 {
 795         struct sock_map_seq_info *info = priv_data;
 796
 797         bpf_map_put_with_uref(info->map);
 798 }
 799
 800 static u64 sock_map_mem_usage(const struct bpf_map *map)
 801 {
 802         u64 usage = sizeof(struct bpf_stab);
 803
 804         usage += (u64)map->max_entries * sizeof(struct sock *);
 805         return usage;
 806 }
 807
 808 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 809         .seq_ops                = &sock_map_seq_ops,
 810         .init_seq_private       = sock_map_init_seq_private,
 811         .fini_seq_private       = sock_map_fini_seq_private,
 812         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 813 };
 814
 815 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 816 const struct bpf_map_ops sock_map_ops = {
 817         .map_meta_equal         = bpf_map_meta_equal,
 818         .map_alloc              = sock_map_alloc,
 819         .map_free               = sock_map_free,
 820         .map_get_next_key       = sock_map_get_next_key,
 821         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 822         .map_update_elem        = sock_map_update_elem,
 823         .map_delete_elem        = sock_map_delete_elem,
 824         .map_lookup_elem        = sock_map_lookup,
 825         .map_release_uref       = sock_map_release_progs,
 826         .map_check_btf          = map_check_no_btf,
 827         .map_mem_usage          = sock_map_mem_usage,
 828         .map_btf_id             = &sock_map_btf_ids[0],
 829         .iter_seq_info          = &sock_map_iter_seq_info,
 830 };
 831
 832 struct bpf_shtab_elem {
 833         struct rcu_head rcu;
 834         u32 hash;
 835         struct sock *sk;
 836         struct hlist_node node;
 837         u8 key[];
 838 };
 839
 840 struct bpf_shtab_bucket {
 841         struct hlist_head head;
 842         raw_spinlock_t lock;
 843 };
 844
 845 struct bpf_shtab {
 846         struct bpf_map map;
 847         struct bpf_shtab_bucket *buckets;
 848         u32 buckets_num;
 849         u32 elem_size;
 850         struct sk_psock_progs progs;
 851         atomic_t count;
 852 };
 853
 854 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 855 {
 856         return jhash(key, len, 0);
 857 }
 858
 859 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 860                                                         u32 hash)
 861 {
 862         return &htab->buckets[hash & (htab->buckets_num - 1)];
 863 }
 864
 865 static struct bpf_shtab_elem *
 866 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 867                           u32 key_size)
 868 {
 869         struct bpf_shtab_elem *elem;
 870
 871         hlist_for_each_entry_rcu(elem, head, node) {
 872                 if (elem->hash == hash &&
 873                     !memcmp(&elem->key, key, key_size))
 874                         return elem;
 875         }
 876
 877         return NULL;
 878 }
 879
 880 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 881 {
 882         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 883         u32 key_size = map->key_size, hash;
 884         struct bpf_shtab_bucket *bucket;
 885         struct bpf_shtab_elem *elem;
 886
 887         WARN_ON_ONCE(!rcu_read_lock_held());
 888
 889         hash = sock_hash_bucket_hash(key, key_size);
 890         bucket = sock_hash_select_bucket(htab, hash);
 891         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 892
 893         return elem ? elem->sk : NULL;
 894 }
 895
 896 static void sock_hash_free_elem(struct bpf_shtab *htab,
 897                                 struct bpf_shtab_elem *elem)
 898 {
 899         atomic_dec(&htab->count);
 900         kfree_rcu(elem, rcu);
 901 }
 902
 903 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 904                                        void *link_raw)
 905 {
 906         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 907         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 908         struct bpf_shtab_bucket *bucket;
 909
 910         WARN_ON_ONCE(!rcu_read_lock_held());
 911         bucket = sock_hash_select_bucket(htab, elem->hash);
 912
 913         /* elem may be deleted in parallel from the map, but access here
 914          * is okay since it's going away only after RCU grace period.
 915          * However, we need to check whether it's still present.
 916          */
 917         raw_spin_lock_bh(&bucket->lock);
 918         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 919                                                elem->key, map->key_size);
 920         if (elem_probe && elem_probe == elem) {
 921                 hlist_del_rcu(&elem->node);
 922                 sock_map_unref(elem->sk, elem);
 923                 sock_hash_free_elem(htab, elem);
 924         }
 925         raw_spin_unlock_bh(&bucket->lock);
 926 }
 927
 928 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
 929 {
 930         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 931         u32 hash, key_size = map->key_size;
 932         struct bpf_shtab_bucket *bucket;
 933         struct bpf_shtab_elem *elem;
 934         int ret = -ENOENT;
 935
 936         hash = sock_hash_bucket_hash(key, key_size);
 937         bucket = sock_hash_select_bucket(htab, hash);
 938
 939         raw_spin_lock_bh(&bucket->lock);
 940         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 941         if (elem) {
 942                 hlist_del_rcu(&elem->node);
 943                 sock_map_unref(elem->sk, elem);
 944                 sock_hash_free_elem(htab, elem);
 945                 ret = 0;
 946         }
 947         raw_spin_unlock_bh(&bucket->lock);
 948         return ret;
 949 }
 950
 951 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 952                                                    void *key, u32 key_size,
 953                                                    u32 hash, struct sock *sk,
 954                                                    struct bpf_shtab_elem *old)
 955 {
 956         struct bpf_shtab_elem *new;
 957
 958         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 959                 if (!old) {
 960                         atomic_dec(&htab->count);
 961                         return ERR_PTR(-E2BIG);
 962                 }
 963         }
 964
 965         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 966                                    GFP_ATOMIC | __GFP_NOWARN,
 967                                    htab->map.numa_node);
 968         if (!new) {
 969                 atomic_dec(&htab->count);
 970                 return ERR_PTR(-ENOMEM);
 971         }
 972         memcpy(new->key, key, key_size);
 973         new->sk = sk;
 974         new->hash = hash;
 975         return new;
 976 }
 977
 978 static int sock_hash_update_common(struct bpf_map *map, void *key,
 979                                    struct sock *sk, u64 flags)
 980 {
 981         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 982         u32 key_size = map->key_size, hash;
 983         struct bpf_shtab_elem *elem, *elem_new;
 984         struct bpf_shtab_bucket *bucket;
 985         struct sk_psock_link *link;
 986         struct sk_psock *psock;
 987         int ret;
 988
 989         WARN_ON_ONCE(!rcu_read_lock_held());
 990         if (unlikely(flags > BPF_EXIST))
 991                 return -EINVAL;
 992
 993         link = sk_psock_init_link();
 994         if (!link)
 995                 return -ENOMEM;
 996
 997         ret = sock_map_link(map, sk);
 998         if (ret < 0)
 999                 goto out_free;
1000
1001         psock = sk_psock(sk);
1002         WARN_ON_ONCE(!psock);
1003
1004         hash = sock_hash_bucket_hash(key, key_size);
1005         bucket = sock_hash_select_bucket(htab, hash);
1006
1007         raw_spin_lock_bh(&bucket->lock);
1008         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1009         if (elem && flags == BPF_NOEXIST) {
1010                 ret = -EEXIST;
1011                 goto out_unlock;
1012         } else if (!elem && flags == BPF_EXIST) {
1013                 ret = -ENOENT;
1014                 goto out_unlock;
1015         }
1016
1017         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1018         if (IS_ERR(elem_new)) {
1019                 ret = PTR_ERR(elem_new);
1020                 goto out_unlock;
1021         }
1022
1023         sock_map_add_link(psock, link, map, elem_new);
1024         /* Add new element to the head of the list, so that
1025          * concurrent search will find it before old elem.
1026          */
1027         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1028         if (elem) {
1029                 hlist_del_rcu(&elem->node);
1030                 sock_map_unref(elem->sk, elem);
1031                 sock_hash_free_elem(htab, elem);
1032         }
1033         raw_spin_unlock_bh(&bucket->lock);
1034         return 0;
1035 out_unlock:
1036         raw_spin_unlock_bh(&bucket->lock);
1037         sk_psock_put(sk, psock);
1038 out_free:
1039         sk_psock_free_link(link);
1040         return ret;
1041 }
1042
1043 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1044                                   void *key_next)
1045 {
1046         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1047         struct bpf_shtab_elem *elem, *elem_next;
1048         u32 hash, key_size = map->key_size;
1049         struct hlist_head *head;
1050         int i = 0;
1051
1052         if (!key)
1053                 goto find_first_elem;
1054         hash = sock_hash_bucket_hash(key, key_size);
1055         head = &sock_hash_select_bucket(htab, hash)->head;
1056         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1057         if (!elem)
1058                 goto find_first_elem;
1059
1060         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1061                                      struct bpf_shtab_elem, node);
1062         if (elem_next) {
1063                 memcpy(key_next, elem_next->key, key_size);
1064                 return 0;
1065         }
1066
1067         i = hash & (htab->buckets_num - 1);
1068         i++;
1069 find_first_elem:
1070         for (; i < htab->buckets_num; i++) {
1071                 head = &sock_hash_select_bucket(htab, i)->head;
1072                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1073                                              struct bpf_shtab_elem, node);
1074                 if (elem_next) {
1075                         memcpy(key_next, elem_next->key, key_size);
1076                         return 0;
1077                 }
1078         }
1079
1080         return -ENOENT;
1081 }
1082
1083 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1084 {
1085         struct bpf_shtab *htab;
1086         int i, err;
1087
1088         if (!capable(CAP_NET_ADMIN))
1089                 return ERR_PTR(-EPERM);
1090         if (attr->max_entries == 0 ||
1091             attr->key_size    == 0 ||
1092             (attr->value_size != sizeof(u32) &&
1093              attr->value_size != sizeof(u64)) ||
1094             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1095                 return ERR_PTR(-EINVAL);
1096         if (attr->key_size > MAX_BPF_STACK)
1097                 return ERR_PTR(-E2BIG);
1098
1099         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1100         if (!htab)
1101                 return ERR_PTR(-ENOMEM);
1102
1103         bpf_map_init_from_attr(&htab->map, attr);
1104
1105         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1106         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1107                           round_up(htab->map.key_size, 8);
1108         if (htab->buckets_num == 0 ||
1109             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1110                 err = -EINVAL;
1111                 goto free_htab;
1112         }
1113
1114         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1115                                            sizeof(struct bpf_shtab_bucket),
1116                                            htab->map.numa_node);
1117         if (!htab->buckets) {
1118                 err = -ENOMEM;
1119                 goto free_htab;
1120         }
1121
1122         for (i = 0; i < htab->buckets_num; i++) {
1123                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1124                 raw_spin_lock_init(&htab->buckets[i].lock);
1125         }
1126
1127         return &htab->map;
1128 free_htab:
1129         bpf_map_area_free(htab);
1130         return ERR_PTR(err);
1131 }
1132
1133 static void sock_hash_free(struct bpf_map *map)
1134 {
1135         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1136         struct bpf_shtab_bucket *bucket;
1137         struct hlist_head unlink_list;
1138         struct bpf_shtab_elem *elem;
1139         struct hlist_node *node;
1140         int i;
1141
1142         /* After the sync no updates or deletes will be in-flight so it
1143          * is safe to walk map and remove entries without risking a race
1144          * in EEXIST update case.
1145          */
1146         synchronize_rcu();
1147         for (i = 0; i < htab->buckets_num; i++) {
1148                 bucket = sock_hash_select_bucket(htab, i);
1149
1150                 /* We are racing with sock_hash_delete_from_link to
1151                  * enter the spin-lock critical section. Every socket on
1152                  * the list is still linked to sockhash. Since link
1153                  * exists, psock exists and holds a ref to socket. That
1154                  * lets us to grab a socket ref too.
1155                  */
1156                 raw_spin_lock_bh(&bucket->lock);
1157                 hlist_for_each_entry(elem, &bucket->head, node)
1158                         sock_hold(elem->sk);
1159                 hlist_move_list(&bucket->head, &unlink_list);
1160                 raw_spin_unlock_bh(&bucket->lock);
1161
1162                 /* Process removed entries out of atomic context to
1163                  * block for socket lock before deleting the psock's
1164                  * link to sockhash.
1165                  */
1166                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1167                         hlist_del(&elem->node);
1168                         lock_sock(elem->sk);
1169                         rcu_read_lock();
1170                         sock_map_unref(elem->sk, elem);
1171                         rcu_read_unlock();
1172                         release_sock(elem->sk);
1173                         sock_put(elem->sk);
1174                         sock_hash_free_elem(htab, elem);
1175                 }
1176         }
1177
1178         /* wait for psock readers accessing its map link */
1179         synchronize_rcu();
1180
1181         bpf_map_area_free(htab->buckets);
1182         bpf_map_area_free(htab);
1183 }
1184
1185 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1186 {
1187         struct sock *sk;
1188
1189         if (map->value_size != sizeof(u64))
1190                 return ERR_PTR(-ENOSPC);
1191
1192         sk = __sock_hash_lookup_elem(map, key);
1193         if (!sk)
1194                 return ERR_PTR(-ENOENT);
1195
1196         __sock_gen_cookie(sk);
1197         return &sk->sk_cookie;
1198 }
1199
1200 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1201 {
1202         struct sock *sk;
1203
1204         sk = __sock_hash_lookup_elem(map, key);
1205         if (!sk)
1206                 return NULL;
1207         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1208                 return NULL;
1209         return sk;
1210 }
1211
1212 static void sock_hash_release_progs(struct bpf_map *map)
1213 {
1214         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1215 }
1216
1217 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1218            struct bpf_map *, map, void *, key, u64, flags)
1219 {
1220         WARN_ON_ONCE(!rcu_read_lock_held());
1221
1222         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1223                    sock_map_op_okay(sops)))
1224                 return sock_hash_update_common(map, key, sops->sk, flags);
1225         return -EOPNOTSUPP;
1226 }
1227
1228 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1229         .func           = bpf_sock_hash_update,
1230         .gpl_only       = false,
1231         .pkt_access     = true,
1232         .ret_type       = RET_INTEGER,
1233         .arg1_type      = ARG_PTR_TO_CTX,
1234         .arg2_type      = ARG_CONST_MAP_PTR,
1235         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1236         .arg4_type      = ARG_ANYTHING,
1237 };
1238
1239 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1240            struct bpf_map *, map, void *, key, u64, flags)
1241 {
1242         struct sock *sk;
1243
1244         if (unlikely(flags & ~(BPF_F_INGRESS)))
1245                 return SK_DROP;
1246
1247         sk = __sock_hash_lookup_elem(map, key);
1248         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1249                 return SK_DROP;
1250
1251         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1252         return SK_PASS;
1253 }
1254
1255 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1256         .func           = bpf_sk_redirect_hash,
1257         .gpl_only       = false,
1258         .ret_type       = RET_INTEGER,
1259         .arg1_type      = ARG_PTR_TO_CTX,
1260         .arg2_type      = ARG_CONST_MAP_PTR,
1261         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1262         .arg4_type      = ARG_ANYTHING,
1263 };
1264
1265 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1266            struct bpf_map *, map, void *, key, u64, flags)
1267 {
1268         struct sock *sk;
1269
1270         if (unlikely(flags & ~(BPF_F_INGRESS)))
1271                 return SK_DROP;
1272
1273         sk = __sock_hash_lookup_elem(map, key);
1274         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1275                 return SK_DROP;
1276
1277         msg->flags = flags;
1278         msg->sk_redir = sk;
1279         return SK_PASS;
1280 }
1281
1282 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1283         .func           = bpf_msg_redirect_hash,
1284         .gpl_only       = false,
1285         .ret_type       = RET_INTEGER,
1286         .arg1_type      = ARG_PTR_TO_CTX,
1287         .arg2_type      = ARG_CONST_MAP_PTR,
1288         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1289         .arg4_type      = ARG_ANYTHING,
1290 };
1291
1292 struct sock_hash_seq_info {
1293         struct bpf_map *map;
1294         struct bpf_shtab *htab;
1295         u32 bucket_id;
1296 };
1297
1298 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1299                                      struct bpf_shtab_elem *prev_elem)
1300 {
1301         const struct bpf_shtab *htab = info->htab;
1302         struct bpf_shtab_bucket *bucket;
1303         struct bpf_shtab_elem *elem;
1304         struct hlist_node *node;
1305
1306         /* try to find next elem in the same bucket */
1307         if (prev_elem) {
1308                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1309                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1310                 if (elem)
1311                         return elem;
1312
1313                 /* no more elements, continue in the next bucket */
1314                 info->bucket_id++;
1315         }
1316
1317         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1318                 bucket = &htab->buckets[info->bucket_id];
1319                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1320                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1321                 if (elem)
1322                         return elem;
1323         }
1324
1325         return NULL;
1326 }
1327
1328 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1329         __acquires(rcu)
1330 {
1331         struct sock_hash_seq_info *info = seq->private;
1332
1333         if (*pos == 0)
1334                 ++*pos;
1335
1336         /* pairs with sock_hash_seq_stop */
1337         rcu_read_lock();
1338         return sock_hash_seq_find_next(info, NULL);
1339 }
1340
1341 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1342         __must_hold(rcu)
1343 {
1344         struct sock_hash_seq_info *info = seq->private;
1345
1346         ++*pos;
1347         return sock_hash_seq_find_next(info, v);
1348 }
1349
1350 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1351         __must_hold(rcu)
1352 {
1353         struct sock_hash_seq_info *info = seq->private;
1354         struct bpf_iter__sockmap ctx = {};
1355         struct bpf_shtab_elem *elem = v;
1356         struct bpf_iter_meta meta;
1357         struct bpf_prog *prog;
1358
1359         meta.seq = seq;
1360         prog = bpf_iter_get_info(&meta, !elem);
1361         if (!prog)
1362                 return 0;
1363
1364         ctx.meta = &meta;
1365         ctx.map = info->map;
1366         if (elem) {
1367                 ctx.key = elem->key;
1368                 ctx.sk = elem->sk;
1369         }
1370
1371         return bpf_iter_run_prog(prog, &ctx);
1372 }
1373
1374 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1375         __releases(rcu)
1376 {
1377         if (!v)
1378                 (void)sock_hash_seq_show(seq, NULL);
1379
1380         /* pairs with sock_hash_seq_start */
1381         rcu_read_unlock();
1382 }
1383
1384 static const struct seq_operations sock_hash_seq_ops = {
1385         .start  = sock_hash_seq_start,
1386         .next   = sock_hash_seq_next,
1387         .stop   = sock_hash_seq_stop,
1388         .show   = sock_hash_seq_show,
1389 };
1390
1391 static int sock_hash_init_seq_private(void *priv_data,
1392                                       struct bpf_iter_aux_info *aux)
1393 {
1394         struct sock_hash_seq_info *info = priv_data;
1395
1396         bpf_map_inc_with_uref(aux->map);
1397         info->map = aux->map;
1398         info->htab = container_of(aux->map, struct bpf_shtab, map);
1399         return 0;
1400 }
1401
1402 static void sock_hash_fini_seq_private(void *priv_data)
1403 {
1404         struct sock_hash_seq_info *info = priv_data;
1405
1406         bpf_map_put_with_uref(info->map);
1407 }
1408
1409 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1410 {
1411         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1412         u64 usage = sizeof(*htab);
1413
1414         usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1415         usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1416         return usage;
1417 }
1418
1419 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1420         .seq_ops                = &sock_hash_seq_ops,
1421         .init_seq_private       = sock_hash_init_seq_private,
1422         .fini_seq_private       = sock_hash_fini_seq_private,
1423         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1424 };
1425
1426 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1427 const struct bpf_map_ops sock_hash_ops = {
1428         .map_meta_equal         = bpf_map_meta_equal,
1429         .map_alloc              = sock_hash_alloc,
1430         .map_free               = sock_hash_free,
1431         .map_get_next_key       = sock_hash_get_next_key,
1432         .map_update_elem        = sock_map_update_elem,
1433         .map_delete_elem        = sock_hash_delete_elem,
1434         .map_lookup_elem        = sock_hash_lookup,
1435         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1436         .map_release_uref       = sock_hash_release_progs,
1437         .map_check_btf          = map_check_no_btf,
1438         .map_mem_usage          = sock_hash_mem_usage,
1439         .map_btf_id             = &sock_hash_map_btf_ids[0],
1440         .iter_seq_info          = &sock_hash_iter_seq_info,
1441 };
1442
1443 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1444 {
1445         switch (map->map_type) {
1446         case BPF_MAP_TYPE_SOCKMAP:
1447                 return &container_of(map, struct bpf_stab, map)->progs;
1448         case BPF_MAP_TYPE_SOCKHASH:
1449                 return &container_of(map, struct bpf_shtab, map)->progs;
1450         default:
1451                 break;
1452         }
1453
1454         return NULL;
1455 }
1456
1457 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1458                                 u32 which)
1459 {
1460         struct sk_psock_progs *progs = sock_map_progs(map);
1461
1462         if (!progs)
1463                 return -EOPNOTSUPP;
1464
1465         switch (which) {
1466         case BPF_SK_MSG_VERDICT:
1467                 *pprog = &progs->msg_parser;
1468                 break;
1469 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1470         case BPF_SK_SKB_STREAM_PARSER:
1471                 *pprog = &progs->stream_parser;
1472                 break;
1473 #endif
1474         case BPF_SK_SKB_STREAM_VERDICT:
1475                 if (progs->skb_verdict)
1476                         return -EBUSY;
1477                 *pprog = &progs->stream_verdict;
1478                 break;
1479         case BPF_SK_SKB_VERDICT:
1480                 if (progs->stream_verdict)
1481                         return -EBUSY;
1482                 *pprog = &progs->skb_verdict;
1483                 break;
1484         default:
1485                 return -EOPNOTSUPP;
1486         }
1487
1488         return 0;
1489 }
1490
1491 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1492                                 struct bpf_prog *old, u32 which)
1493 {
1494         struct bpf_prog **pprog;
1495         int ret;
1496
1497         ret = sock_map_prog_lookup(map, &pprog, which);
1498         if (ret)
1499                 return ret;
1500
1501         if (old)
1502                 return psock_replace_prog(pprog, prog, old);
1503
1504         psock_set_prog(pprog, prog);
1505         return 0;
1506 }
1507
1508 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1509                             union bpf_attr __user *uattr)
1510 {
1511         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1512         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1513         struct bpf_prog **pprog;
1514         struct bpf_prog *prog;
1515         struct bpf_map *map;
1516         struct fd f;
1517         u32 id = 0;
1518         int ret;
1519
1520         if (attr->query.query_flags)
1521                 return -EINVAL;
1522
1523         f = fdget(ufd);
1524         map = __bpf_map_get(f);
1525         if (IS_ERR(map))
1526                 return PTR_ERR(map);
1527
1528         rcu_read_lock();
1529
1530         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1531         if (ret)
1532                 goto end;
1533
1534         prog = *pprog;
1535         prog_cnt = !prog ? 0 : 1;
1536
1537         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1538                 goto end;
1539
1540         /* we do not hold the refcnt, the bpf prog may be released
1541          * asynchronously and the id would be set to 0.
1542          */
1543         id = data_race(prog->aux->id);
1544         if (id == 0)
1545                 prog_cnt = 0;
1546
1547 end:
1548         rcu_read_unlock();
1549
1550         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1551             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1552             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1553                 ret = -EFAULT;
1554
1555         fdput(f);
1556         return ret;
1557 }
1558
1559 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1560 {
1561         switch (link->map->map_type) {
1562         case BPF_MAP_TYPE_SOCKMAP:
1563                 return sock_map_delete_from_link(link->map, sk,
1564                                                  link->link_raw);
1565         case BPF_MAP_TYPE_SOCKHASH:
1566                 return sock_hash_delete_from_link(link->map, sk,
1567                                                   link->link_raw);
1568         default:
1569                 break;
1570         }
1571 }
1572
1573 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1574 {
1575         struct sk_psock_link *link;
1576
1577         while ((link = sk_psock_link_pop(psock))) {
1578                 sock_map_unlink(sk, link);
1579                 sk_psock_free_link(link);
1580         }
1581 }
1582
1583 void sock_map_unhash(struct sock *sk)
1584 {
1585         void (*saved_unhash)(struct sock *sk);
1586         struct sk_psock *psock;
1587
1588         rcu_read_lock();
1589         psock = sk_psock(sk);
1590         if (unlikely(!psock)) {
1591                 rcu_read_unlock();
1592                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1593         } else {
1594                 saved_unhash = psock->saved_unhash;
1595                 sock_map_remove_links(sk, psock);
1596                 rcu_read_unlock();
1597         }
1598         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1599                 return;
1600         if (saved_unhash)
1601                 saved_unhash(sk);
1602 }
1603 EXPORT_SYMBOL_GPL(sock_map_unhash);
1604
1605 void sock_map_destroy(struct sock *sk)
1606 {
1607         void (*saved_destroy)(struct sock *sk);
1608         struct sk_psock *psock;
1609
1610         rcu_read_lock();
1611         psock = sk_psock_get(sk);
1612         if (unlikely(!psock)) {
1613                 rcu_read_unlock();
1614                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1615         } else {
1616                 saved_destroy = psock->saved_destroy;
1617                 sock_map_remove_links(sk, psock);
1618                 rcu_read_unlock();
1619                 sk_psock_stop(psock);
1620                 sk_psock_put(sk, psock);
1621         }
1622         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1623                 return;
1624         if (saved_destroy)
1625                 saved_destroy(sk);
1626 }
1627 EXPORT_SYMBOL_GPL(sock_map_destroy);
1628
1629 void sock_map_close(struct sock *sk, long timeout)
1630 {
1631         void (*saved_close)(struct sock *sk, long timeout);
1632         struct sk_psock *psock;
1633
1634         lock_sock(sk);
1635         rcu_read_lock();
1636         psock = sk_psock_get(sk);
1637         if (unlikely(!psock)) {
1638                 rcu_read_unlock();
1639                 release_sock(sk);
1640                 saved_close = READ_ONCE(sk->sk_prot)->close;
1641         } else {
1642                 saved_close = psock->saved_close;
1643                 sock_map_remove_links(sk, psock);
1644                 rcu_read_unlock();
1645                 sk_psock_stop(psock);
1646                 release_sock(sk);
1647                 cancel_work_sync(&psock->work);
1648                 sk_psock_put(sk, psock);
1649         }
1650         /* Make sure we do not recurse. This is a bug.
1651          * Leak the socket instead of crashing on a stack overflow.
1652          */
1653         if (WARN_ON_ONCE(saved_close == sock_map_close))
1654                 return;
1655         saved_close(sk, timeout);
1656 }
1657 EXPORT_SYMBOL_GPL(sock_map_close);
1658
1659 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1660                                        union bpf_iter_link_info *linfo,
1661                                        struct bpf_iter_aux_info *aux)
1662 {
1663         struct bpf_map *map;
1664         int err = -EINVAL;
1665
1666         if (!linfo->map.map_fd)
1667                 return -EBADF;
1668
1669         map = bpf_map_get_with_uref(linfo->map.map_fd);
1670         if (IS_ERR(map))
1671                 return PTR_ERR(map);
1672
1673         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1674             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1675                 goto put_map;
1676
1677         if (prog->aux->max_rdonly_access > map->key_size) {
1678                 err = -EACCES;
1679                 goto put_map;
1680         }
1681
1682         aux->map = map;
1683         return 0;
1684
1685 put_map:
1686         bpf_map_put_with_uref(map);
1687         return err;
1688 }
1689
1690 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1691 {
1692         bpf_map_put_with_uref(aux->map);
1693 }
1694
1695 static struct bpf_iter_reg sock_map_iter_reg = {
1696         .target                 = "sockmap",
1697         .attach_target          = sock_map_iter_attach_target,
1698         .detach_target          = sock_map_iter_detach_target,
1699         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1700         .fill_link_info         = bpf_iter_map_fill_link_info,
1701         .ctx_arg_info_size      = 2,
1702         .ctx_arg_info           = {
1703                 { offsetof(struct bpf_iter__sockmap, key),
1704                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1705                 { offsetof(struct bpf_iter__sockmap, sk),
1706                   PTR_TO_BTF_ID_OR_NULL },
1707         },
1708 };
1709
1710 static int __init bpf_sockmap_iter_init(void)
1711 {
1712         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1713                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1714         return bpf_iter_reg_target(&sock_map_iter_reg);
1715 }
1716 late_initcall(bpf_sockmap_iter_init);