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 = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
  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(stab->map.max_entries *
  52                                        sizeof(struct sock *),
  53                                        stab->map.numa_node);
  54         if (!stab->sks) {
  55                 kfree(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                 else
 171                         sk_psock_stop_verdict(sk, psock);
 172                 write_unlock_bh(&sk->sk_callback_lock);
 173         }
 174 }
 175
 176 static void sock_map_unref(struct sock *sk, void *link_raw)
 177 {
 178         struct sk_psock *psock = sk_psock(sk);
 179
 180         if (likely(psock)) {
 181                 sock_map_del_link(sk, psock, link_raw);
 182                 sk_psock_put(sk, psock);
 183         }
 184 }
 185
 186 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
 187 {
 188         struct proto *prot;
 189
 190         switch (sk->sk_type) {
 191         case SOCK_STREAM:
 192                 prot = tcp_bpf_get_proto(sk, psock);
 193                 break;
 194
 195         case SOCK_DGRAM:
 196                 prot = udp_bpf_get_proto(sk, psock);
 197                 break;
 198
 199         default:
 200                 return -EINVAL;
 201         }
 202
 203         if (IS_ERR(prot))
 204                 return PTR_ERR(prot);
 205
 206         sk_psock_update_proto(sk, psock, prot);
 207         return 0;
 208 }
 209
 210 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 211 {
 212         struct sk_psock *psock;
 213
 214         rcu_read_lock();
 215         psock = sk_psock(sk);
 216         if (psock) {
 217                 if (sk->sk_prot->close != sock_map_close) {
 218                         psock = ERR_PTR(-EBUSY);
 219                         goto out;
 220                 }
 221
 222                 if (!refcount_inc_not_zero(&psock->refcnt))
 223                         psock = ERR_PTR(-EBUSY);
 224         }
 225 out:
 226         rcu_read_unlock();
 227         return psock;
 228 }
 229
 230 static bool sock_map_redirect_allowed(const struct sock *sk);
 231
 232 static int sock_map_link(struct bpf_map *map, struct sock *sk)
 233 {
 234         struct sk_psock_progs *progs = sock_map_progs(map);
 235         struct bpf_prog *stream_verdict = NULL;
 236         struct bpf_prog *stream_parser = NULL;
 237         struct bpf_prog *skb_verdict = NULL;
 238         struct bpf_prog *msg_parser = NULL;
 239         struct sk_psock *psock;
 240         int ret;
 241
 242         /* Only sockets we can redirect into/from in BPF need to hold
 243          * refs to parser/verdict progs and have their sk_data_ready
 244          * and sk_write_space callbacks overridden.
 245          */
 246         if (!sock_map_redirect_allowed(sk))
 247                 goto no_progs;
 248
 249         stream_verdict = READ_ONCE(progs->stream_verdict);
 250         if (stream_verdict) {
 251                 stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
 252                 if (IS_ERR(stream_verdict))
 253                         return PTR_ERR(stream_verdict);
 254         }
 255
 256         stream_parser = READ_ONCE(progs->stream_parser);
 257         if (stream_parser) {
 258                 stream_parser = bpf_prog_inc_not_zero(stream_parser);
 259                 if (IS_ERR(stream_parser)) {
 260                         ret = PTR_ERR(stream_parser);
 261                         goto out_put_stream_verdict;
 262                 }
 263         }
 264
 265         msg_parser = READ_ONCE(progs->msg_parser);
 266         if (msg_parser) {
 267                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 268                 if (IS_ERR(msg_parser)) {
 269                         ret = PTR_ERR(msg_parser);
 270                         goto out_put_stream_parser;
 271                 }
 272         }
 273
 274         skb_verdict = READ_ONCE(progs->skb_verdict);
 275         if (skb_verdict) {
 276                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 277                 if (IS_ERR(skb_verdict)) {
 278                         ret = PTR_ERR(skb_verdict);
 279                         goto out_put_msg_parser;
 280                 }
 281         }
 282
 283 no_progs:
 284         psock = sock_map_psock_get_checked(sk);
 285         if (IS_ERR(psock)) {
 286                 ret = PTR_ERR(psock);
 287                 goto out_progs;
 288         }
 289
 290         if (psock) {
 291                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 292                     (stream_parser  && READ_ONCE(psock->progs.stream_parser)) ||
 293                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 294                     (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
 295                     (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
 296                     (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
 297                         sk_psock_put(sk, psock);
 298                         ret = -EBUSY;
 299                         goto out_progs;
 300                 }
 301         } else {
 302                 psock = sk_psock_init(sk, map->numa_node);
 303                 if (IS_ERR(psock)) {
 304                         ret = PTR_ERR(psock);
 305                         goto out_progs;
 306                 }
 307         }
 308
 309         if (msg_parser)
 310                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 311
 312         ret = sock_map_init_proto(sk, psock);
 313         if (ret < 0)
 314                 goto out_drop;
 315
 316         write_lock_bh(&sk->sk_callback_lock);
 317         if (stream_parser && stream_verdict && !psock->saved_data_ready) {
 318                 ret = sk_psock_init_strp(sk, psock);
 319                 if (ret)
 320                         goto out_unlock_drop;
 321                 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
 322                 psock_set_prog(&psock->progs.stream_parser, stream_parser);
 323                 sk_psock_start_strp(sk, psock);
 324         } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
 325                 psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
 326                 sk_psock_start_verdict(sk,psock);
 327         } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
 328                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 329                 sk_psock_start_verdict(sk, psock);
 330         }
 331         write_unlock_bh(&sk->sk_callback_lock);
 332         return 0;
 333 out_unlock_drop:
 334         write_unlock_bh(&sk->sk_callback_lock);
 335 out_drop:
 336         sk_psock_put(sk, psock);
 337 out_progs:
 338         if (skb_verdict)
 339                 bpf_prog_put(skb_verdict);
 340 out_put_msg_parser:
 341         if (msg_parser)
 342                 bpf_prog_put(msg_parser);
 343 out_put_stream_parser:
 344         if (stream_parser)
 345                 bpf_prog_put(stream_parser);
 346 out_put_stream_verdict:
 347         if (stream_verdict)
 348                 bpf_prog_put(stream_verdict);
 349         return ret;
 350 }
 351
 352 static void sock_map_free(struct bpf_map *map)
 353 {
 354         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 355         int i;
 356
 357         /* After the sync no updates or deletes will be in-flight so it
 358          * is safe to walk map and remove entries without risking a race
 359          * in EEXIST update case.
 360          */
 361         synchronize_rcu();
 362         for (i = 0; i < stab->map.max_entries; i++) {
 363                 struct sock **psk = &stab->sks[i];
 364                 struct sock *sk;
 365
 366                 sk = xchg(psk, NULL);
 367                 if (sk) {
 368                         lock_sock(sk);
 369                         rcu_read_lock();
 370                         sock_map_unref(sk, psk);
 371                         rcu_read_unlock();
 372                         release_sock(sk);
 373                 }
 374         }
 375
 376         /* wait for psock readers accessing its map link */
 377         synchronize_rcu();
 378
 379         bpf_map_area_free(stab->sks);
 380         kfree(stab);
 381 }
 382
 383 static void sock_map_release_progs(struct bpf_map *map)
 384 {
 385         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 386 }
 387
 388 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 389 {
 390         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 391
 392         WARN_ON_ONCE(!rcu_read_lock_held());
 393
 394         if (unlikely(key >= map->max_entries))
 395                 return NULL;
 396         return READ_ONCE(stab->sks[key]);
 397 }
 398
 399 static void *sock_map_lookup(struct bpf_map *map, void *key)
 400 {
 401         struct sock *sk;
 402
 403         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 404         if (!sk)
 405                 return NULL;
 406         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 407                 return NULL;
 408         return sk;
 409 }
 410
 411 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 412 {
 413         struct sock *sk;
 414
 415         if (map->value_size != sizeof(u64))
 416                 return ERR_PTR(-ENOSPC);
 417
 418         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 419         if (!sk)
 420                 return ERR_PTR(-ENOENT);
 421
 422         __sock_gen_cookie(sk);
 423         return &sk->sk_cookie;
 424 }
 425
 426 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 427                              struct sock **psk)
 428 {
 429         struct sock *sk;
 430         int err = 0;
 431
 432         raw_spin_lock_bh(&stab->lock);
 433         sk = *psk;
 434         if (!sk_test || sk_test == sk)
 435                 sk = xchg(psk, NULL);
 436
 437         if (likely(sk))
 438                 sock_map_unref(sk, psk);
 439         else
 440                 err = -EINVAL;
 441
 442         raw_spin_unlock_bh(&stab->lock);
 443         return err;
 444 }
 445
 446 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 447                                       void *link_raw)
 448 {
 449         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 450
 451         __sock_map_delete(stab, sk, link_raw);
 452 }
 453
 454 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 455 {
 456         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 457         u32 i = *(u32 *)key;
 458         struct sock **psk;
 459
 460         if (unlikely(i >= map->max_entries))
 461                 return -EINVAL;
 462
 463         psk = &stab->sks[i];
 464         return __sock_map_delete(stab, NULL, psk);
 465 }
 466
 467 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 468 {
 469         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 470         u32 i = key ? *(u32 *)key : U32_MAX;
 471         u32 *key_next = next;
 472
 473         if (i == stab->map.max_entries - 1)
 474                 return -ENOENT;
 475         if (i >= stab->map.max_entries)
 476                 *key_next = 0;
 477         else
 478                 *key_next = i + 1;
 479         return 0;
 480 }
 481
 482 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 483                                   struct sock *sk, u64 flags)
 484 {
 485         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 486         struct sk_psock_link *link;
 487         struct sk_psock *psock;
 488         struct sock *osk;
 489         int ret;
 490
 491         WARN_ON_ONCE(!rcu_read_lock_held());
 492         if (unlikely(flags > BPF_EXIST))
 493                 return -EINVAL;
 494         if (unlikely(idx >= map->max_entries))
 495                 return -E2BIG;
 496
 497         link = sk_psock_init_link();
 498         if (!link)
 499                 return -ENOMEM;
 500
 501         ret = sock_map_link(map, sk);
 502         if (ret < 0)
 503                 goto out_free;
 504
 505         psock = sk_psock(sk);
 506         WARN_ON_ONCE(!psock);
 507
 508         raw_spin_lock_bh(&stab->lock);
 509         osk = stab->sks[idx];
 510         if (osk && flags == BPF_NOEXIST) {
 511                 ret = -EEXIST;
 512                 goto out_unlock;
 513         } else if (!osk && flags == BPF_EXIST) {
 514                 ret = -ENOENT;
 515                 goto out_unlock;
 516         }
 517
 518         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 519         stab->sks[idx] = sk;
 520         if (osk)
 521                 sock_map_unref(osk, &stab->sks[idx]);
 522         raw_spin_unlock_bh(&stab->lock);
 523         return 0;
 524 out_unlock:
 525         raw_spin_unlock_bh(&stab->lock);
 526         if (psock)
 527                 sk_psock_put(sk, psock);
 528 out_free:
 529         sk_psock_free_link(link);
 530         return ret;
 531 }
 532
 533 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 534 {
 535         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 536                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 537                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 538 }
 539
 540 static bool sk_is_tcp(const struct sock *sk)
 541 {
 542         return sk->sk_type == SOCK_STREAM &&
 543                sk->sk_protocol == IPPROTO_TCP;
 544 }
 545
 546 static bool sk_is_udp(const struct sock *sk)
 547 {
 548         return sk->sk_type == SOCK_DGRAM &&
 549                sk->sk_protocol == IPPROTO_UDP;
 550 }
 551
 552 static bool sock_map_redirect_allowed(const struct sock *sk)
 553 {
 554         return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
 555 }
 556
 557 static bool sock_map_sk_is_suitable(const struct sock *sk)
 558 {
 559         return sk_is_tcp(sk) || sk_is_udp(sk);
 560 }
 561
 562 static bool sock_map_sk_state_allowed(const struct sock *sk)
 563 {
 564         if (sk_is_tcp(sk))
 565                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 566         else if (sk_is_udp(sk))
 567                 return sk_hashed(sk);
 568
 569         return false;
 570 }
 571
 572 static int sock_hash_update_common(struct bpf_map *map, void *key,
 573                                    struct sock *sk, u64 flags);
 574
 575 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 576                              u64 flags)
 577 {
 578         struct socket *sock;
 579         struct sock *sk;
 580         int ret;
 581         u64 ufd;
 582
 583         if (map->value_size == sizeof(u64))
 584                 ufd = *(u64 *)value;
 585         else
 586                 ufd = *(u32 *)value;
 587         if (ufd > S32_MAX)
 588                 return -EINVAL;
 589
 590         sock = sockfd_lookup(ufd, &ret);
 591         if (!sock)
 592                 return ret;
 593         sk = sock->sk;
 594         if (!sk) {
 595                 ret = -EINVAL;
 596                 goto out;
 597         }
 598         if (!sock_map_sk_is_suitable(sk)) {
 599                 ret = -EOPNOTSUPP;
 600                 goto out;
 601         }
 602
 603         sock_map_sk_acquire(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         sock_map_sk_release(sk);
 611 out:
 612         sockfd_put(sock);
 613         return ret;
 614 }
 615
 616 static int sock_map_update_elem(struct bpf_map *map, void *key,
 617                                 void *value, u64 flags)
 618 {
 619         struct sock *sk = (struct sock *)value;
 620         int ret;
 621
 622         if (unlikely(!sk || !sk_fullsock(sk)))
 623                 return -EINVAL;
 624
 625         if (!sock_map_sk_is_suitable(sk))
 626                 return -EOPNOTSUPP;
 627
 628         local_bh_disable();
 629         bh_lock_sock(sk);
 630         if (!sock_map_sk_state_allowed(sk))
 631                 ret = -EOPNOTSUPP;
 632         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 633                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 634         else
 635                 ret = sock_hash_update_common(map, key, sk, flags);
 636         bh_unlock_sock(sk);
 637         local_bh_enable();
 638         return ret;
 639 }
 640
 641 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 642            struct bpf_map *, map, void *, key, u64, flags)
 643 {
 644         WARN_ON_ONCE(!rcu_read_lock_held());
 645
 646         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 647                    sock_map_op_okay(sops)))
 648                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 649                                               flags);
 650         return -EOPNOTSUPP;
 651 }
 652
 653 const struct bpf_func_proto bpf_sock_map_update_proto = {
 654         .func           = bpf_sock_map_update,
 655         .gpl_only       = false,
 656         .pkt_access     = true,
 657         .ret_type       = RET_INTEGER,
 658         .arg1_type      = ARG_PTR_TO_CTX,
 659         .arg2_type      = ARG_CONST_MAP_PTR,
 660         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 661         .arg4_type      = ARG_ANYTHING,
 662 };
 663
 664 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 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         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
 677         return SK_PASS;
 678 }
 679
 680 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 681         .func           = bpf_sk_redirect_map,
 682         .gpl_only       = false,
 683         .ret_type       = RET_INTEGER,
 684         .arg1_type      = ARG_PTR_TO_CTX,
 685         .arg2_type      = ARG_CONST_MAP_PTR,
 686         .arg3_type      = ARG_ANYTHING,
 687         .arg4_type      = ARG_ANYTHING,
 688 };
 689
 690 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 691            struct bpf_map *, map, u32, key, u64, flags)
 692 {
 693         struct sock *sk;
 694
 695         if (unlikely(flags & ~(BPF_F_INGRESS)))
 696                 return SK_DROP;
 697
 698         sk = __sock_map_lookup_elem(map, key);
 699         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 700                 return SK_DROP;
 701
 702         msg->flags = flags;
 703         msg->sk_redir = sk;
 704         return SK_PASS;
 705 }
 706
 707 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 708         .func           = bpf_msg_redirect_map,
 709         .gpl_only       = false,
 710         .ret_type       = RET_INTEGER,
 711         .arg1_type      = ARG_PTR_TO_CTX,
 712         .arg2_type      = ARG_CONST_MAP_PTR,
 713         .arg3_type      = ARG_ANYTHING,
 714         .arg4_type      = ARG_ANYTHING,
 715 };
 716
 717 struct sock_map_seq_info {
 718         struct bpf_map *map;
 719         struct sock *sk;
 720         u32 index;
 721 };
 722
 723 struct bpf_iter__sockmap {
 724         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 725         __bpf_md_ptr(struct bpf_map *, map);
 726         __bpf_md_ptr(void *, key);
 727         __bpf_md_ptr(struct sock *, sk);
 728 };
 729
 730 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 731                      struct bpf_map *map, void *key,
 732                      struct sock *sk)
 733
 734 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 735 {
 736         if (unlikely(info->index >= info->map->max_entries))
 737                 return NULL;
 738
 739         info->sk = __sock_map_lookup_elem(info->map, info->index);
 740
 741         /* can't return sk directly, since that might be NULL */
 742         return info;
 743 }
 744
 745 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 746         __acquires(rcu)
 747 {
 748         struct sock_map_seq_info *info = seq->private;
 749
 750         if (*pos == 0)
 751                 ++*pos;
 752
 753         /* pairs with sock_map_seq_stop */
 754         rcu_read_lock();
 755         return sock_map_seq_lookup_elem(info);
 756 }
 757
 758 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 759         __must_hold(rcu)
 760 {
 761         struct sock_map_seq_info *info = seq->private;
 762
 763         ++*pos;
 764         ++info->index;
 765
 766         return sock_map_seq_lookup_elem(info);
 767 }
 768
 769 static int sock_map_seq_show(struct seq_file *seq, void *v)
 770         __must_hold(rcu)
 771 {
 772         struct sock_map_seq_info *info = seq->private;
 773         struct bpf_iter__sockmap ctx = {};
 774         struct bpf_iter_meta meta;
 775         struct bpf_prog *prog;
 776
 777         meta.seq = seq;
 778         prog = bpf_iter_get_info(&meta, !v);
 779         if (!prog)
 780                 return 0;
 781
 782         ctx.meta = &meta;
 783         ctx.map = info->map;
 784         if (v) {
 785                 ctx.key = &info->index;
 786                 ctx.sk = info->sk;
 787         }
 788
 789         return bpf_iter_run_prog(prog, &ctx);
 790 }
 791
 792 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 793         __releases(rcu)
 794 {
 795         if (!v)
 796                 (void)sock_map_seq_show(seq, NULL);
 797
 798         /* pairs with sock_map_seq_start */
 799         rcu_read_unlock();
 800 }
 801
 802 static const struct seq_operations sock_map_seq_ops = {
 803         .start  = sock_map_seq_start,
 804         .next   = sock_map_seq_next,
 805         .stop   = sock_map_seq_stop,
 806         .show   = sock_map_seq_show,
 807 };
 808
 809 static int sock_map_init_seq_private(void *priv_data,
 810                                      struct bpf_iter_aux_info *aux)
 811 {
 812         struct sock_map_seq_info *info = priv_data;
 813
 814         info->map = aux->map;
 815         return 0;
 816 }
 817
 818 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 819         .seq_ops                = &sock_map_seq_ops,
 820         .init_seq_private       = sock_map_init_seq_private,
 821         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 822 };
 823
 824 static int sock_map_btf_id;
 825 const struct bpf_map_ops sock_map_ops = {
 826         .map_meta_equal         = bpf_map_meta_equal,
 827         .map_alloc              = sock_map_alloc,
 828         .map_free               = sock_map_free,
 829         .map_get_next_key       = sock_map_get_next_key,
 830         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 831         .map_update_elem        = sock_map_update_elem,
 832         .map_delete_elem        = sock_map_delete_elem,
 833         .map_lookup_elem        = sock_map_lookup,
 834         .map_release_uref       = sock_map_release_progs,
 835         .map_check_btf          = map_check_no_btf,
 836         .map_btf_name           = "bpf_stab",
 837         .map_btf_id             = &sock_map_btf_id,
 838         .iter_seq_info          = &sock_map_iter_seq_info,
 839 };
 840
 841 struct bpf_shtab_elem {
 842         struct rcu_head rcu;
 843         u32 hash;
 844         struct sock *sk;
 845         struct hlist_node node;
 846         u8 key[];
 847 };
 848
 849 struct bpf_shtab_bucket {
 850         struct hlist_head head;
 851         raw_spinlock_t lock;
 852 };
 853
 854 struct bpf_shtab {
 855         struct bpf_map map;
 856         struct bpf_shtab_bucket *buckets;
 857         u32 buckets_num;
 858         u32 elem_size;
 859         struct sk_psock_progs progs;
 860         atomic_t count;
 861 };
 862
 863 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 864 {
 865         return jhash(key, len, 0);
 866 }
 867
 868 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 869                                                         u32 hash)
 870 {
 871         return &htab->buckets[hash & (htab->buckets_num - 1)];
 872 }
 873
 874 static struct bpf_shtab_elem *
 875 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 876                           u32 key_size)
 877 {
 878         struct bpf_shtab_elem *elem;
 879
 880         hlist_for_each_entry_rcu(elem, head, node) {
 881                 if (elem->hash == hash &&
 882                     !memcmp(&elem->key, key, key_size))
 883                         return elem;
 884         }
 885
 886         return NULL;
 887 }
 888
 889 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 890 {
 891         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 892         u32 key_size = map->key_size, hash;
 893         struct bpf_shtab_bucket *bucket;
 894         struct bpf_shtab_elem *elem;
 895
 896         WARN_ON_ONCE(!rcu_read_lock_held());
 897
 898         hash = sock_hash_bucket_hash(key, key_size);
 899         bucket = sock_hash_select_bucket(htab, hash);
 900         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 901
 902         return elem ? elem->sk : NULL;
 903 }
 904
 905 static void sock_hash_free_elem(struct bpf_shtab *htab,
 906                                 struct bpf_shtab_elem *elem)
 907 {
 908         atomic_dec(&htab->count);
 909         kfree_rcu(elem, rcu);
 910 }
 911
 912 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 913                                        void *link_raw)
 914 {
 915         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 916         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 917         struct bpf_shtab_bucket *bucket;
 918
 919         WARN_ON_ONCE(!rcu_read_lock_held());
 920         bucket = sock_hash_select_bucket(htab, elem->hash);
 921
 922         /* elem may be deleted in parallel from the map, but access here
 923          * is okay since it's going away only after RCU grace period.
 924          * However, we need to check whether it's still present.
 925          */
 926         raw_spin_lock_bh(&bucket->lock);
 927         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 928                                                elem->key, map->key_size);
 929         if (elem_probe && elem_probe == elem) {
 930                 hlist_del_rcu(&elem->node);
 931                 sock_map_unref(elem->sk, elem);
 932                 sock_hash_free_elem(htab, elem);
 933         }
 934         raw_spin_unlock_bh(&bucket->lock);
 935 }
 936
 937 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 938 {
 939         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 940         u32 hash, key_size = map->key_size;
 941         struct bpf_shtab_bucket *bucket;
 942         struct bpf_shtab_elem *elem;
 943         int ret = -ENOENT;
 944
 945         hash = sock_hash_bucket_hash(key, key_size);
 946         bucket = sock_hash_select_bucket(htab, hash);
 947
 948         raw_spin_lock_bh(&bucket->lock);
 949         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 950         if (elem) {
 951                 hlist_del_rcu(&elem->node);
 952                 sock_map_unref(elem->sk, elem);
 953                 sock_hash_free_elem(htab, elem);
 954                 ret = 0;
 955         }
 956         raw_spin_unlock_bh(&bucket->lock);
 957         return ret;
 958 }
 959
 960 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 961                                                    void *key, u32 key_size,
 962                                                    u32 hash, struct sock *sk,
 963                                                    struct bpf_shtab_elem *old)
 964 {
 965         struct bpf_shtab_elem *new;
 966
 967         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 968                 if (!old) {
 969                         atomic_dec(&htab->count);
 970                         return ERR_PTR(-E2BIG);
 971                 }
 972         }
 973
 974         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 975                                    GFP_ATOMIC | __GFP_NOWARN,
 976                                    htab->map.numa_node);
 977         if (!new) {
 978                 atomic_dec(&htab->count);
 979                 return ERR_PTR(-ENOMEM);
 980         }
 981         memcpy(new->key, key, key_size);
 982         new->sk = sk;
 983         new->hash = hash;
 984         return new;
 985 }
 986
 987 static int sock_hash_update_common(struct bpf_map *map, void *key,
 988                                    struct sock *sk, u64 flags)
 989 {
 990         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 991         u32 key_size = map->key_size, hash;
 992         struct bpf_shtab_elem *elem, *elem_new;
 993         struct bpf_shtab_bucket *bucket;
 994         struct sk_psock_link *link;
 995         struct sk_psock *psock;
 996         int ret;
 997
 998         WARN_ON_ONCE(!rcu_read_lock_held());
 999         if (unlikely(flags > BPF_EXIST))
1000                 return -EINVAL;
1001
1002         link = sk_psock_init_link();
1003         if (!link)
1004                 return -ENOMEM;
1005
1006         ret = sock_map_link(map, sk);
1007         if (ret < 0)
1008                 goto out_free;
1009
1010         psock = sk_psock(sk);
1011         WARN_ON_ONCE(!psock);
1012
1013         hash = sock_hash_bucket_hash(key, key_size);
1014         bucket = sock_hash_select_bucket(htab, hash);
1015
1016         raw_spin_lock_bh(&bucket->lock);
1017         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1018         if (elem && flags == BPF_NOEXIST) {
1019                 ret = -EEXIST;
1020                 goto out_unlock;
1021         } else if (!elem && flags == BPF_EXIST) {
1022                 ret = -ENOENT;
1023                 goto out_unlock;
1024         }
1025
1026         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1027         if (IS_ERR(elem_new)) {
1028                 ret = PTR_ERR(elem_new);
1029                 goto out_unlock;
1030         }
1031
1032         sock_map_add_link(psock, link, map, elem_new);
1033         /* Add new element to the head of the list, so that
1034          * concurrent search will find it before old elem.
1035          */
1036         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1037         if (elem) {
1038                 hlist_del_rcu(&elem->node);
1039                 sock_map_unref(elem->sk, elem);
1040                 sock_hash_free_elem(htab, elem);
1041         }
1042         raw_spin_unlock_bh(&bucket->lock);
1043         return 0;
1044 out_unlock:
1045         raw_spin_unlock_bh(&bucket->lock);
1046         sk_psock_put(sk, psock);
1047 out_free:
1048         sk_psock_free_link(link);
1049         return ret;
1050 }
1051
1052 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1053                                   void *key_next)
1054 {
1055         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1056         struct bpf_shtab_elem *elem, *elem_next;
1057         u32 hash, key_size = map->key_size;
1058         struct hlist_head *head;
1059         int i = 0;
1060
1061         if (!key)
1062                 goto find_first_elem;
1063         hash = sock_hash_bucket_hash(key, key_size);
1064         head = &sock_hash_select_bucket(htab, hash)->head;
1065         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1066         if (!elem)
1067                 goto find_first_elem;
1068
1069         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1070                                      struct bpf_shtab_elem, node);
1071         if (elem_next) {
1072                 memcpy(key_next, elem_next->key, key_size);
1073                 return 0;
1074         }
1075
1076         i = hash & (htab->buckets_num - 1);
1077         i++;
1078 find_first_elem:
1079         for (; i < htab->buckets_num; i++) {
1080                 head = &sock_hash_select_bucket(htab, i)->head;
1081                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1082                                              struct bpf_shtab_elem, node);
1083                 if (elem_next) {
1084                         memcpy(key_next, elem_next->key, key_size);
1085                         return 0;
1086                 }
1087         }
1088
1089         return -ENOENT;
1090 }
1091
1092 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1093 {
1094         struct bpf_shtab *htab;
1095         int i, err;
1096
1097         if (!capable(CAP_NET_ADMIN))
1098                 return ERR_PTR(-EPERM);
1099         if (attr->max_entries == 0 ||
1100             attr->key_size    == 0 ||
1101             (attr->value_size != sizeof(u32) &&
1102              attr->value_size != sizeof(u64)) ||
1103             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1104                 return ERR_PTR(-EINVAL);
1105         if (attr->key_size > MAX_BPF_STACK)
1106                 return ERR_PTR(-E2BIG);
1107
1108         htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1109         if (!htab)
1110                 return ERR_PTR(-ENOMEM);
1111
1112         bpf_map_init_from_attr(&htab->map, attr);
1113
1114         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1115         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1116                           round_up(htab->map.key_size, 8);
1117         if (htab->buckets_num == 0 ||
1118             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1119                 err = -EINVAL;
1120                 goto free_htab;
1121         }
1122
1123         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1124                                            sizeof(struct bpf_shtab_bucket),
1125                                            htab->map.numa_node);
1126         if (!htab->buckets) {
1127                 err = -ENOMEM;
1128                 goto free_htab;
1129         }
1130
1131         for (i = 0; i < htab->buckets_num; i++) {
1132                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1133                 raw_spin_lock_init(&htab->buckets[i].lock);
1134         }
1135
1136         return &htab->map;
1137 free_htab:
1138         kfree(htab);
1139         return ERR_PTR(err);
1140 }
1141
1142 static void sock_hash_free(struct bpf_map *map)
1143 {
1144         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1145         struct bpf_shtab_bucket *bucket;
1146         struct hlist_head unlink_list;
1147         struct bpf_shtab_elem *elem;
1148         struct hlist_node *node;
1149         int i;
1150
1151         /* After the sync no updates or deletes will be in-flight so it
1152          * is safe to walk map and remove entries without risking a race
1153          * in EEXIST update case.
1154          */
1155         synchronize_rcu();
1156         for (i = 0; i < htab->buckets_num; i++) {
1157                 bucket = sock_hash_select_bucket(htab, i);
1158
1159                 /* We are racing with sock_hash_delete_from_link to
1160                  * enter the spin-lock critical section. Every socket on
1161                  * the list is still linked to sockhash. Since link
1162                  * exists, psock exists and holds a ref to socket. That
1163                  * lets us to grab a socket ref too.
1164                  */
1165                 raw_spin_lock_bh(&bucket->lock);
1166                 hlist_for_each_entry(elem, &bucket->head, node)
1167                         sock_hold(elem->sk);
1168                 hlist_move_list(&bucket->head, &unlink_list);
1169                 raw_spin_unlock_bh(&bucket->lock);
1170
1171                 /* Process removed entries out of atomic context to
1172                  * block for socket lock before deleting the psock's
1173                  * link to sockhash.
1174                  */
1175                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1176                         hlist_del(&elem->node);
1177                         lock_sock(elem->sk);
1178                         rcu_read_lock();
1179                         sock_map_unref(elem->sk, elem);
1180                         rcu_read_unlock();
1181                         release_sock(elem->sk);
1182                         sock_put(elem->sk);
1183                         sock_hash_free_elem(htab, elem);
1184                 }
1185         }
1186
1187         /* wait for psock readers accessing its map link */
1188         synchronize_rcu();
1189
1190         bpf_map_area_free(htab->buckets);
1191         kfree(htab);
1192 }
1193
1194 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1195 {
1196         struct sock *sk;
1197
1198         if (map->value_size != sizeof(u64))
1199                 return ERR_PTR(-ENOSPC);
1200
1201         sk = __sock_hash_lookup_elem(map, key);
1202         if (!sk)
1203                 return ERR_PTR(-ENOENT);
1204
1205         __sock_gen_cookie(sk);
1206         return &sk->sk_cookie;
1207 }
1208
1209 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1210 {
1211         struct sock *sk;
1212
1213         sk = __sock_hash_lookup_elem(map, key);
1214         if (!sk)
1215                 return NULL;
1216         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1217                 return NULL;
1218         return sk;
1219 }
1220
1221 static void sock_hash_release_progs(struct bpf_map *map)
1222 {
1223         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1224 }
1225
1226 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1227            struct bpf_map *, map, void *, key, u64, flags)
1228 {
1229         WARN_ON_ONCE(!rcu_read_lock_held());
1230
1231         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1232                    sock_map_op_okay(sops)))
1233                 return sock_hash_update_common(map, key, sops->sk, flags);
1234         return -EOPNOTSUPP;
1235 }
1236
1237 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1238         .func           = bpf_sock_hash_update,
1239         .gpl_only       = false,
1240         .pkt_access     = true,
1241         .ret_type       = RET_INTEGER,
1242         .arg1_type      = ARG_PTR_TO_CTX,
1243         .arg2_type      = ARG_CONST_MAP_PTR,
1244         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1245         .arg4_type      = ARG_ANYTHING,
1246 };
1247
1248 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1249            struct bpf_map *, map, void *, key, u64, flags)
1250 {
1251         struct sock *sk;
1252
1253         if (unlikely(flags & ~(BPF_F_INGRESS)))
1254                 return SK_DROP;
1255
1256         sk = __sock_hash_lookup_elem(map, key);
1257         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1258                 return SK_DROP;
1259
1260         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1261         return SK_PASS;
1262 }
1263
1264 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1265         .func           = bpf_sk_redirect_hash,
1266         .gpl_only       = false,
1267         .ret_type       = RET_INTEGER,
1268         .arg1_type      = ARG_PTR_TO_CTX,
1269         .arg2_type      = ARG_CONST_MAP_PTR,
1270         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1271         .arg4_type      = ARG_ANYTHING,
1272 };
1273
1274 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1275            struct bpf_map *, map, void *, key, u64, flags)
1276 {
1277         struct sock *sk;
1278
1279         if (unlikely(flags & ~(BPF_F_INGRESS)))
1280                 return SK_DROP;
1281
1282         sk = __sock_hash_lookup_elem(map, key);
1283         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1284                 return SK_DROP;
1285
1286         msg->flags = flags;
1287         msg->sk_redir = sk;
1288         return SK_PASS;
1289 }
1290
1291 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1292         .func           = bpf_msg_redirect_hash,
1293         .gpl_only       = false,
1294         .ret_type       = RET_INTEGER,
1295         .arg1_type      = ARG_PTR_TO_CTX,
1296         .arg2_type      = ARG_CONST_MAP_PTR,
1297         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1298         .arg4_type      = ARG_ANYTHING,
1299 };
1300
1301 struct sock_hash_seq_info {
1302         struct bpf_map *map;
1303         struct bpf_shtab *htab;
1304         u32 bucket_id;
1305 };
1306
1307 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1308                                      struct bpf_shtab_elem *prev_elem)
1309 {
1310         const struct bpf_shtab *htab = info->htab;
1311         struct bpf_shtab_bucket *bucket;
1312         struct bpf_shtab_elem *elem;
1313         struct hlist_node *node;
1314
1315         /* try to find next elem in the same bucket */
1316         if (prev_elem) {
1317                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1318                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1319                 if (elem)
1320                         return elem;
1321
1322                 /* no more elements, continue in the next bucket */
1323                 info->bucket_id++;
1324         }
1325
1326         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1327                 bucket = &htab->buckets[info->bucket_id];
1328                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1329                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1330                 if (elem)
1331                         return elem;
1332         }
1333
1334         return NULL;
1335 }
1336
1337 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1338         __acquires(rcu)
1339 {
1340         struct sock_hash_seq_info *info = seq->private;
1341
1342         if (*pos == 0)
1343                 ++*pos;
1344
1345         /* pairs with sock_hash_seq_stop */
1346         rcu_read_lock();
1347         return sock_hash_seq_find_next(info, NULL);
1348 }
1349
1350 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1351         __must_hold(rcu)
1352 {
1353         struct sock_hash_seq_info *info = seq->private;
1354
1355         ++*pos;
1356         return sock_hash_seq_find_next(info, v);
1357 }
1358
1359 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1360         __must_hold(rcu)
1361 {
1362         struct sock_hash_seq_info *info = seq->private;
1363         struct bpf_iter__sockmap ctx = {};
1364         struct bpf_shtab_elem *elem = v;
1365         struct bpf_iter_meta meta;
1366         struct bpf_prog *prog;
1367
1368         meta.seq = seq;
1369         prog = bpf_iter_get_info(&meta, !elem);
1370         if (!prog)
1371                 return 0;
1372
1373         ctx.meta = &meta;
1374         ctx.map = info->map;
1375         if (elem) {
1376                 ctx.key = elem->key;
1377                 ctx.sk = elem->sk;
1378         }
1379
1380         return bpf_iter_run_prog(prog, &ctx);
1381 }
1382
1383 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1384         __releases(rcu)
1385 {
1386         if (!v)
1387                 (void)sock_hash_seq_show(seq, NULL);
1388
1389         /* pairs with sock_hash_seq_start */
1390         rcu_read_unlock();
1391 }
1392
1393 static const struct seq_operations sock_hash_seq_ops = {
1394         .start  = sock_hash_seq_start,
1395         .next   = sock_hash_seq_next,
1396         .stop   = sock_hash_seq_stop,
1397         .show   = sock_hash_seq_show,
1398 };
1399
1400 static int sock_hash_init_seq_private(void *priv_data,
1401                                      struct bpf_iter_aux_info *aux)
1402 {
1403         struct sock_hash_seq_info *info = priv_data;
1404
1405         info->map = aux->map;
1406         info->htab = container_of(aux->map, struct bpf_shtab, map);
1407         return 0;
1408 }
1409
1410 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1411         .seq_ops                = &sock_hash_seq_ops,
1412         .init_seq_private       = sock_hash_init_seq_private,
1413         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1414 };
1415
1416 static int sock_hash_map_btf_id;
1417 const struct bpf_map_ops sock_hash_ops = {
1418         .map_meta_equal         = bpf_map_meta_equal,
1419         .map_alloc              = sock_hash_alloc,
1420         .map_free               = sock_hash_free,
1421         .map_get_next_key       = sock_hash_get_next_key,
1422         .map_update_elem        = sock_map_update_elem,
1423         .map_delete_elem        = sock_hash_delete_elem,
1424         .map_lookup_elem        = sock_hash_lookup,
1425         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1426         .map_release_uref       = sock_hash_release_progs,
1427         .map_check_btf          = map_check_no_btf,
1428         .map_btf_name           = "bpf_shtab",
1429         .map_btf_id             = &sock_hash_map_btf_id,
1430         .iter_seq_info          = &sock_hash_iter_seq_info,
1431 };
1432
1433 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1434 {
1435         switch (map->map_type) {
1436         case BPF_MAP_TYPE_SOCKMAP:
1437                 return &container_of(map, struct bpf_stab, map)->progs;
1438         case BPF_MAP_TYPE_SOCKHASH:
1439                 return &container_of(map, struct bpf_shtab, map)->progs;
1440         default:
1441                 break;
1442         }
1443
1444         return NULL;
1445 }
1446
1447 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1448                                 struct bpf_prog *old, u32 which)
1449 {
1450         struct sk_psock_progs *progs = sock_map_progs(map);
1451         struct bpf_prog **pprog;
1452
1453         if (!progs)
1454                 return -EOPNOTSUPP;
1455
1456         switch (which) {
1457         case BPF_SK_MSG_VERDICT:
1458                 pprog = &progs->msg_parser;
1459                 break;
1460 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1461         case BPF_SK_SKB_STREAM_PARSER:
1462                 pprog = &progs->stream_parser;
1463                 break;
1464 #endif
1465         case BPF_SK_SKB_STREAM_VERDICT:
1466                 if (progs->skb_verdict)
1467                         return -EBUSY;
1468                 pprog = &progs->stream_verdict;
1469                 break;
1470         case BPF_SK_SKB_VERDICT:
1471                 if (progs->stream_verdict)
1472                         return -EBUSY;
1473                 pprog = &progs->skb_verdict;
1474                 break;
1475         default:
1476                 return -EOPNOTSUPP;
1477         }
1478
1479         if (old)
1480                 return psock_replace_prog(pprog, prog, old);
1481
1482         psock_set_prog(pprog, prog);
1483         return 0;
1484 }
1485
1486 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1487 {
1488         switch (link->map->map_type) {
1489         case BPF_MAP_TYPE_SOCKMAP:
1490                 return sock_map_delete_from_link(link->map, sk,
1491                                                  link->link_raw);
1492         case BPF_MAP_TYPE_SOCKHASH:
1493                 return sock_hash_delete_from_link(link->map, sk,
1494                                                   link->link_raw);
1495         default:
1496                 break;
1497         }
1498 }
1499
1500 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1501 {
1502         struct sk_psock_link *link;
1503
1504         while ((link = sk_psock_link_pop(psock))) {
1505                 sock_map_unlink(sk, link);
1506                 sk_psock_free_link(link);
1507         }
1508 }
1509
1510 void sock_map_unhash(struct sock *sk)
1511 {
1512         void (*saved_unhash)(struct sock *sk);
1513         struct sk_psock *psock;
1514
1515         rcu_read_lock();
1516         psock = sk_psock(sk);
1517         if (unlikely(!psock)) {
1518                 rcu_read_unlock();
1519                 if (sk->sk_prot->unhash)
1520                         sk->sk_prot->unhash(sk);
1521                 return;
1522         }
1523
1524         saved_unhash = psock->saved_unhash;
1525         sock_map_remove_links(sk, psock);
1526         rcu_read_unlock();
1527         saved_unhash(sk);
1528 }
1529
1530 void sock_map_close(struct sock *sk, long timeout)
1531 {
1532         void (*saved_close)(struct sock *sk, long timeout);
1533         struct sk_psock *psock;
1534
1535         lock_sock(sk);
1536         rcu_read_lock();
1537         psock = sk_psock(sk);
1538         if (unlikely(!psock)) {
1539                 rcu_read_unlock();
1540                 release_sock(sk);
1541                 return sk->sk_prot->close(sk, timeout);
1542         }
1543
1544         saved_close = psock->saved_close;
1545         sock_map_remove_links(sk, psock);
1546         rcu_read_unlock();
1547         sk_psock_stop(psock, true);
1548         release_sock(sk);
1549         saved_close(sk, timeout);
1550 }
1551
1552 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1553                                        union bpf_iter_link_info *linfo,
1554                                        struct bpf_iter_aux_info *aux)
1555 {
1556         struct bpf_map *map;
1557         int err = -EINVAL;
1558
1559         if (!linfo->map.map_fd)
1560                 return -EBADF;
1561
1562         map = bpf_map_get_with_uref(linfo->map.map_fd);
1563         if (IS_ERR(map))
1564                 return PTR_ERR(map);
1565
1566         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1567             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1568                 goto put_map;
1569
1570         if (prog->aux->max_rdonly_access > map->key_size) {
1571                 err = -EACCES;
1572                 goto put_map;
1573         }
1574
1575         aux->map = map;
1576         return 0;
1577
1578 put_map:
1579         bpf_map_put_with_uref(map);
1580         return err;
1581 }
1582
1583 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1584 {
1585         bpf_map_put_with_uref(aux->map);
1586 }
1587
1588 static struct bpf_iter_reg sock_map_iter_reg = {
1589         .target                 = "sockmap",
1590         .attach_target          = sock_map_iter_attach_target,
1591         .detach_target          = sock_map_iter_detach_target,
1592         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1593         .fill_link_info         = bpf_iter_map_fill_link_info,
1594         .ctx_arg_info_size      = 2,
1595         .ctx_arg_info           = {
1596                 { offsetof(struct bpf_iter__sockmap, key),
1597                   PTR_TO_RDONLY_BUF_OR_NULL },
1598                 { offsetof(struct bpf_iter__sockmap, sk),
1599                   PTR_TO_BTF_ID_OR_NULL },
1600         },
1601 };
1602
1603 static int __init bpf_sockmap_iter_init(void)
1604 {
1605         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1606                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1607         return bpf_iter_reg_target(&sock_map_iter_reg);
1608 }
1609 late_initcall(bpf_sockmap_iter_init);