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 int 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 int 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 const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 801         .seq_ops                = &sock_map_seq_ops,
 802         .init_seq_private       = sock_map_init_seq_private,
 803         .fini_seq_private       = sock_map_fini_seq_private,
 804         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 805 };
 806
 807 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
 808 const struct bpf_map_ops sock_map_ops = {
 809         .map_meta_equal         = bpf_map_meta_equal,
 810         .map_alloc              = sock_map_alloc,
 811         .map_free               = sock_map_free,
 812         .map_get_next_key       = sock_map_get_next_key,
 813         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 814         .map_update_elem        = sock_map_update_elem,
 815         .map_delete_elem        = sock_map_delete_elem,
 816         .map_lookup_elem        = sock_map_lookup,
 817         .map_release_uref       = sock_map_release_progs,
 818         .map_check_btf          = map_check_no_btf,
 819         .map_btf_id             = &sock_map_btf_ids[0],
 820         .iter_seq_info          = &sock_map_iter_seq_info,
 821 };
 822
 823 struct bpf_shtab_elem {
 824         struct rcu_head rcu;
 825         u32 hash;
 826         struct sock *sk;
 827         struct hlist_node node;
 828         u8 key[];
 829 };
 830
 831 struct bpf_shtab_bucket {
 832         struct hlist_head head;
 833         raw_spinlock_t lock;
 834 };
 835
 836 struct bpf_shtab {
 837         struct bpf_map map;
 838         struct bpf_shtab_bucket *buckets;
 839         u32 buckets_num;
 840         u32 elem_size;
 841         struct sk_psock_progs progs;
 842         atomic_t count;
 843 };
 844
 845 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 846 {
 847         return jhash(key, len, 0);
 848 }
 849
 850 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 851                                                         u32 hash)
 852 {
 853         return &htab->buckets[hash & (htab->buckets_num - 1)];
 854 }
 855
 856 static struct bpf_shtab_elem *
 857 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 858                           u32 key_size)
 859 {
 860         struct bpf_shtab_elem *elem;
 861
 862         hlist_for_each_entry_rcu(elem, head, node) {
 863                 if (elem->hash == hash &&
 864                     !memcmp(&elem->key, key, key_size))
 865                         return elem;
 866         }
 867
 868         return NULL;
 869 }
 870
 871 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 872 {
 873         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 874         u32 key_size = map->key_size, hash;
 875         struct bpf_shtab_bucket *bucket;
 876         struct bpf_shtab_elem *elem;
 877
 878         WARN_ON_ONCE(!rcu_read_lock_held());
 879
 880         hash = sock_hash_bucket_hash(key, key_size);
 881         bucket = sock_hash_select_bucket(htab, hash);
 882         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 883
 884         return elem ? elem->sk : NULL;
 885 }
 886
 887 static void sock_hash_free_elem(struct bpf_shtab *htab,
 888                                 struct bpf_shtab_elem *elem)
 889 {
 890         atomic_dec(&htab->count);
 891         kfree_rcu(elem, rcu);
 892 }
 893
 894 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 895                                        void *link_raw)
 896 {
 897         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 898         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 899         struct bpf_shtab_bucket *bucket;
 900
 901         WARN_ON_ONCE(!rcu_read_lock_held());
 902         bucket = sock_hash_select_bucket(htab, elem->hash);
 903
 904         /* elem may be deleted in parallel from the map, but access here
 905          * is okay since it's going away only after RCU grace period.
 906          * However, we need to check whether it's still present.
 907          */
 908         raw_spin_lock_bh(&bucket->lock);
 909         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 910                                                elem->key, map->key_size);
 911         if (elem_probe && elem_probe == elem) {
 912                 hlist_del_rcu(&elem->node);
 913                 sock_map_unref(elem->sk, elem);
 914                 sock_hash_free_elem(htab, elem);
 915         }
 916         raw_spin_unlock_bh(&bucket->lock);
 917 }
 918
 919 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 920 {
 921         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 922         u32 hash, key_size = map->key_size;
 923         struct bpf_shtab_bucket *bucket;
 924         struct bpf_shtab_elem *elem;
 925         int ret = -ENOENT;
 926
 927         hash = sock_hash_bucket_hash(key, key_size);
 928         bucket = sock_hash_select_bucket(htab, hash);
 929
 930         raw_spin_lock_bh(&bucket->lock);
 931         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 932         if (elem) {
 933                 hlist_del_rcu(&elem->node);
 934                 sock_map_unref(elem->sk, elem);
 935                 sock_hash_free_elem(htab, elem);
 936                 ret = 0;
 937         }
 938         raw_spin_unlock_bh(&bucket->lock);
 939         return ret;
 940 }
 941
 942 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 943                                                    void *key, u32 key_size,
 944                                                    u32 hash, struct sock *sk,
 945                                                    struct bpf_shtab_elem *old)
 946 {
 947         struct bpf_shtab_elem *new;
 948
 949         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 950                 if (!old) {
 951                         atomic_dec(&htab->count);
 952                         return ERR_PTR(-E2BIG);
 953                 }
 954         }
 955
 956         new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
 957                                    GFP_ATOMIC | __GFP_NOWARN,
 958                                    htab->map.numa_node);
 959         if (!new) {
 960                 atomic_dec(&htab->count);
 961                 return ERR_PTR(-ENOMEM);
 962         }
 963         memcpy(new->key, key, key_size);
 964         new->sk = sk;
 965         new->hash = hash;
 966         return new;
 967 }
 968
 969 static int sock_hash_update_common(struct bpf_map *map, void *key,
 970                                    struct sock *sk, u64 flags)
 971 {
 972         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 973         u32 key_size = map->key_size, hash;
 974         struct bpf_shtab_elem *elem, *elem_new;
 975         struct bpf_shtab_bucket *bucket;
 976         struct sk_psock_link *link;
 977         struct sk_psock *psock;
 978         int ret;
 979
 980         WARN_ON_ONCE(!rcu_read_lock_held());
 981         if (unlikely(flags > BPF_EXIST))
 982                 return -EINVAL;
 983
 984         link = sk_psock_init_link();
 985         if (!link)
 986                 return -ENOMEM;
 987
 988         ret = sock_map_link(map, sk);
 989         if (ret < 0)
 990                 goto out_free;
 991
 992         psock = sk_psock(sk);
 993         WARN_ON_ONCE(!psock);
 994
 995         hash = sock_hash_bucket_hash(key, key_size);
 996         bucket = sock_hash_select_bucket(htab, hash);
 997
 998         raw_spin_lock_bh(&bucket->lock);
 999         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1000         if (elem && flags == BPF_NOEXIST) {
1001                 ret = -EEXIST;
1002                 goto out_unlock;
1003         } else if (!elem && flags == BPF_EXIST) {
1004                 ret = -ENOENT;
1005                 goto out_unlock;
1006         }
1007
1008         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1009         if (IS_ERR(elem_new)) {
1010                 ret = PTR_ERR(elem_new);
1011                 goto out_unlock;
1012         }
1013
1014         sock_map_add_link(psock, link, map, elem_new);
1015         /* Add new element to the head of the list, so that
1016          * concurrent search will find it before old elem.
1017          */
1018         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1019         if (elem) {
1020                 hlist_del_rcu(&elem->node);
1021                 sock_map_unref(elem->sk, elem);
1022                 sock_hash_free_elem(htab, elem);
1023         }
1024         raw_spin_unlock_bh(&bucket->lock);
1025         return 0;
1026 out_unlock:
1027         raw_spin_unlock_bh(&bucket->lock);
1028         sk_psock_put(sk, psock);
1029 out_free:
1030         sk_psock_free_link(link);
1031         return ret;
1032 }
1033
1034 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1035                                   void *key_next)
1036 {
1037         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1038         struct bpf_shtab_elem *elem, *elem_next;
1039         u32 hash, key_size = map->key_size;
1040         struct hlist_head *head;
1041         int i = 0;
1042
1043         if (!key)
1044                 goto find_first_elem;
1045         hash = sock_hash_bucket_hash(key, key_size);
1046         head = &sock_hash_select_bucket(htab, hash)->head;
1047         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1048         if (!elem)
1049                 goto find_first_elem;
1050
1051         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1052                                      struct bpf_shtab_elem, node);
1053         if (elem_next) {
1054                 memcpy(key_next, elem_next->key, key_size);
1055                 return 0;
1056         }
1057
1058         i = hash & (htab->buckets_num - 1);
1059         i++;
1060 find_first_elem:
1061         for (; i < htab->buckets_num; i++) {
1062                 head = &sock_hash_select_bucket(htab, i)->head;
1063                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1064                                              struct bpf_shtab_elem, node);
1065                 if (elem_next) {
1066                         memcpy(key_next, elem_next->key, key_size);
1067                         return 0;
1068                 }
1069         }
1070
1071         return -ENOENT;
1072 }
1073
1074 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1075 {
1076         struct bpf_shtab *htab;
1077         int i, err;
1078
1079         if (!capable(CAP_NET_ADMIN))
1080                 return ERR_PTR(-EPERM);
1081         if (attr->max_entries == 0 ||
1082             attr->key_size    == 0 ||
1083             (attr->value_size != sizeof(u32) &&
1084              attr->value_size != sizeof(u64)) ||
1085             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1086                 return ERR_PTR(-EINVAL);
1087         if (attr->key_size > MAX_BPF_STACK)
1088                 return ERR_PTR(-E2BIG);
1089
1090         htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1091         if (!htab)
1092                 return ERR_PTR(-ENOMEM);
1093
1094         bpf_map_init_from_attr(&htab->map, attr);
1095
1096         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1097         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1098                           round_up(htab->map.key_size, 8);
1099         if (htab->buckets_num == 0 ||
1100             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1101                 err = -EINVAL;
1102                 goto free_htab;
1103         }
1104
1105         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1106                                            sizeof(struct bpf_shtab_bucket),
1107                                            htab->map.numa_node);
1108         if (!htab->buckets) {
1109                 err = -ENOMEM;
1110                 goto free_htab;
1111         }
1112
1113         for (i = 0; i < htab->buckets_num; i++) {
1114                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1115                 raw_spin_lock_init(&htab->buckets[i].lock);
1116         }
1117
1118         return &htab->map;
1119 free_htab:
1120         bpf_map_area_free(htab);
1121         return ERR_PTR(err);
1122 }
1123
1124 static void sock_hash_free(struct bpf_map *map)
1125 {
1126         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1127         struct bpf_shtab_bucket *bucket;
1128         struct hlist_head unlink_list;
1129         struct bpf_shtab_elem *elem;
1130         struct hlist_node *node;
1131         int i;
1132
1133         /* After the sync no updates or deletes will be in-flight so it
1134          * is safe to walk map and remove entries without risking a race
1135          * in EEXIST update case.
1136          */
1137         synchronize_rcu();
1138         for (i = 0; i < htab->buckets_num; i++) {
1139                 bucket = sock_hash_select_bucket(htab, i);
1140
1141                 /* We are racing with sock_hash_delete_from_link to
1142                  * enter the spin-lock critical section. Every socket on
1143                  * the list is still linked to sockhash. Since link
1144                  * exists, psock exists and holds a ref to socket. That
1145                  * lets us to grab a socket ref too.
1146                  */
1147                 raw_spin_lock_bh(&bucket->lock);
1148                 hlist_for_each_entry(elem, &bucket->head, node)
1149                         sock_hold(elem->sk);
1150                 hlist_move_list(&bucket->head, &unlink_list);
1151                 raw_spin_unlock_bh(&bucket->lock);
1152
1153                 /* Process removed entries out of atomic context to
1154                  * block for socket lock before deleting the psock's
1155                  * link to sockhash.
1156                  */
1157                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1158                         hlist_del(&elem->node);
1159                         lock_sock(elem->sk);
1160                         rcu_read_lock();
1161                         sock_map_unref(elem->sk, elem);
1162                         rcu_read_unlock();
1163                         release_sock(elem->sk);
1164                         sock_put(elem->sk);
1165                         sock_hash_free_elem(htab, elem);
1166                 }
1167         }
1168
1169         /* wait for psock readers accessing its map link */
1170         synchronize_rcu();
1171
1172         bpf_map_area_free(htab->buckets);
1173         bpf_map_area_free(htab);
1174 }
1175
1176 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1177 {
1178         struct sock *sk;
1179
1180         if (map->value_size != sizeof(u64))
1181                 return ERR_PTR(-ENOSPC);
1182
1183         sk = __sock_hash_lookup_elem(map, key);
1184         if (!sk)
1185                 return ERR_PTR(-ENOENT);
1186
1187         __sock_gen_cookie(sk);
1188         return &sk->sk_cookie;
1189 }
1190
1191 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1192 {
1193         struct sock *sk;
1194
1195         sk = __sock_hash_lookup_elem(map, key);
1196         if (!sk)
1197                 return NULL;
1198         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1199                 return NULL;
1200         return sk;
1201 }
1202
1203 static void sock_hash_release_progs(struct bpf_map *map)
1204 {
1205         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1206 }
1207
1208 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1209            struct bpf_map *, map, void *, key, u64, flags)
1210 {
1211         WARN_ON_ONCE(!rcu_read_lock_held());
1212
1213         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1214                    sock_map_op_okay(sops)))
1215                 return sock_hash_update_common(map, key, sops->sk, flags);
1216         return -EOPNOTSUPP;
1217 }
1218
1219 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1220         .func           = bpf_sock_hash_update,
1221         .gpl_only       = false,
1222         .pkt_access     = true,
1223         .ret_type       = RET_INTEGER,
1224         .arg1_type      = ARG_PTR_TO_CTX,
1225         .arg2_type      = ARG_CONST_MAP_PTR,
1226         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1227         .arg4_type      = ARG_ANYTHING,
1228 };
1229
1230 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1231            struct bpf_map *, map, void *, key, u64, flags)
1232 {
1233         struct sock *sk;
1234
1235         if (unlikely(flags & ~(BPF_F_INGRESS)))
1236                 return SK_DROP;
1237
1238         sk = __sock_hash_lookup_elem(map, key);
1239         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1240                 return SK_DROP;
1241
1242         skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1243         return SK_PASS;
1244 }
1245
1246 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1247         .func           = bpf_sk_redirect_hash,
1248         .gpl_only       = false,
1249         .ret_type       = RET_INTEGER,
1250         .arg1_type      = ARG_PTR_TO_CTX,
1251         .arg2_type      = ARG_CONST_MAP_PTR,
1252         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1253         .arg4_type      = ARG_ANYTHING,
1254 };
1255
1256 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1257            struct bpf_map *, map, void *, key, u64, flags)
1258 {
1259         struct sock *sk;
1260
1261         if (unlikely(flags & ~(BPF_F_INGRESS)))
1262                 return SK_DROP;
1263
1264         sk = __sock_hash_lookup_elem(map, key);
1265         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1266                 return SK_DROP;
1267
1268         msg->flags = flags;
1269         msg->sk_redir = sk;
1270         return SK_PASS;
1271 }
1272
1273 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1274         .func           = bpf_msg_redirect_hash,
1275         .gpl_only       = false,
1276         .ret_type       = RET_INTEGER,
1277         .arg1_type      = ARG_PTR_TO_CTX,
1278         .arg2_type      = ARG_CONST_MAP_PTR,
1279         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1280         .arg4_type      = ARG_ANYTHING,
1281 };
1282
1283 struct sock_hash_seq_info {
1284         struct bpf_map *map;
1285         struct bpf_shtab *htab;
1286         u32 bucket_id;
1287 };
1288
1289 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1290                                      struct bpf_shtab_elem *prev_elem)
1291 {
1292         const struct bpf_shtab *htab = info->htab;
1293         struct bpf_shtab_bucket *bucket;
1294         struct bpf_shtab_elem *elem;
1295         struct hlist_node *node;
1296
1297         /* try to find next elem in the same bucket */
1298         if (prev_elem) {
1299                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1300                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1301                 if (elem)
1302                         return elem;
1303
1304                 /* no more elements, continue in the next bucket */
1305                 info->bucket_id++;
1306         }
1307
1308         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1309                 bucket = &htab->buckets[info->bucket_id];
1310                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1311                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1312                 if (elem)
1313                         return elem;
1314         }
1315
1316         return NULL;
1317 }
1318
1319 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1320         __acquires(rcu)
1321 {
1322         struct sock_hash_seq_info *info = seq->private;
1323
1324         if (*pos == 0)
1325                 ++*pos;
1326
1327         /* pairs with sock_hash_seq_stop */
1328         rcu_read_lock();
1329         return sock_hash_seq_find_next(info, NULL);
1330 }
1331
1332 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1333         __must_hold(rcu)
1334 {
1335         struct sock_hash_seq_info *info = seq->private;
1336
1337         ++*pos;
1338         return sock_hash_seq_find_next(info, v);
1339 }
1340
1341 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1342         __must_hold(rcu)
1343 {
1344         struct sock_hash_seq_info *info = seq->private;
1345         struct bpf_iter__sockmap ctx = {};
1346         struct bpf_shtab_elem *elem = v;
1347         struct bpf_iter_meta meta;
1348         struct bpf_prog *prog;
1349
1350         meta.seq = seq;
1351         prog = bpf_iter_get_info(&meta, !elem);
1352         if (!prog)
1353                 return 0;
1354
1355         ctx.meta = &meta;
1356         ctx.map = info->map;
1357         if (elem) {
1358                 ctx.key = elem->key;
1359                 ctx.sk = elem->sk;
1360         }
1361
1362         return bpf_iter_run_prog(prog, &ctx);
1363 }
1364
1365 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1366         __releases(rcu)
1367 {
1368         if (!v)
1369                 (void)sock_hash_seq_show(seq, NULL);
1370
1371         /* pairs with sock_hash_seq_start */
1372         rcu_read_unlock();
1373 }
1374
1375 static const struct seq_operations sock_hash_seq_ops = {
1376         .start  = sock_hash_seq_start,
1377         .next   = sock_hash_seq_next,
1378         .stop   = sock_hash_seq_stop,
1379         .show   = sock_hash_seq_show,
1380 };
1381
1382 static int sock_hash_init_seq_private(void *priv_data,
1383                                       struct bpf_iter_aux_info *aux)
1384 {
1385         struct sock_hash_seq_info *info = priv_data;
1386
1387         bpf_map_inc_with_uref(aux->map);
1388         info->map = aux->map;
1389         info->htab = container_of(aux->map, struct bpf_shtab, map);
1390         return 0;
1391 }
1392
1393 static void sock_hash_fini_seq_private(void *priv_data)
1394 {
1395         struct sock_hash_seq_info *info = priv_data;
1396
1397         bpf_map_put_with_uref(info->map);
1398 }
1399
1400 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1401         .seq_ops                = &sock_hash_seq_ops,
1402         .init_seq_private       = sock_hash_init_seq_private,
1403         .fini_seq_private       = sock_hash_fini_seq_private,
1404         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1405 };
1406
1407 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1408 const struct bpf_map_ops sock_hash_ops = {
1409         .map_meta_equal         = bpf_map_meta_equal,
1410         .map_alloc              = sock_hash_alloc,
1411         .map_free               = sock_hash_free,
1412         .map_get_next_key       = sock_hash_get_next_key,
1413         .map_update_elem        = sock_map_update_elem,
1414         .map_delete_elem        = sock_hash_delete_elem,
1415         .map_lookup_elem        = sock_hash_lookup,
1416         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1417         .map_release_uref       = sock_hash_release_progs,
1418         .map_check_btf          = map_check_no_btf,
1419         .map_btf_id             = &sock_hash_map_btf_ids[0],
1420         .iter_seq_info          = &sock_hash_iter_seq_info,
1421 };
1422
1423 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1424 {
1425         switch (map->map_type) {
1426         case BPF_MAP_TYPE_SOCKMAP:
1427                 return &container_of(map, struct bpf_stab, map)->progs;
1428         case BPF_MAP_TYPE_SOCKHASH:
1429                 return &container_of(map, struct bpf_shtab, map)->progs;
1430         default:
1431                 break;
1432         }
1433
1434         return NULL;
1435 }
1436
1437 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1438                                 u32 which)
1439 {
1440         struct sk_psock_progs *progs = sock_map_progs(map);
1441
1442         if (!progs)
1443                 return -EOPNOTSUPP;
1444
1445         switch (which) {
1446         case BPF_SK_MSG_VERDICT:
1447                 *pprog = &progs->msg_parser;
1448                 break;
1449 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1450         case BPF_SK_SKB_STREAM_PARSER:
1451                 *pprog = &progs->stream_parser;
1452                 break;
1453 #endif
1454         case BPF_SK_SKB_STREAM_VERDICT:
1455                 if (progs->skb_verdict)
1456                         return -EBUSY;
1457                 *pprog = &progs->stream_verdict;
1458                 break;
1459         case BPF_SK_SKB_VERDICT:
1460                 if (progs->stream_verdict)
1461                         return -EBUSY;
1462                 *pprog = &progs->skb_verdict;
1463                 break;
1464         default:
1465                 return -EOPNOTSUPP;
1466         }
1467
1468         return 0;
1469 }
1470
1471 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1472                                 struct bpf_prog *old, u32 which)
1473 {
1474         struct bpf_prog **pprog;
1475         int ret;
1476
1477         ret = sock_map_prog_lookup(map, &pprog, which);
1478         if (ret)
1479                 return ret;
1480
1481         if (old)
1482                 return psock_replace_prog(pprog, prog, old);
1483
1484         psock_set_prog(pprog, prog);
1485         return 0;
1486 }
1487
1488 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1489                             union bpf_attr __user *uattr)
1490 {
1491         __u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1492         u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1493         struct bpf_prog **pprog;
1494         struct bpf_prog *prog;
1495         struct bpf_map *map;
1496         struct fd f;
1497         u32 id = 0;
1498         int ret;
1499
1500         if (attr->query.query_flags)
1501                 return -EINVAL;
1502
1503         f = fdget(ufd);
1504         map = __bpf_map_get(f);
1505         if (IS_ERR(map))
1506                 return PTR_ERR(map);
1507
1508         rcu_read_lock();
1509
1510         ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1511         if (ret)
1512                 goto end;
1513
1514         prog = *pprog;
1515         prog_cnt = !prog ? 0 : 1;
1516
1517         if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1518                 goto end;
1519
1520         /* we do not hold the refcnt, the bpf prog may be released
1521          * asynchronously and the id would be set to 0.
1522          */
1523         id = data_race(prog->aux->id);
1524         if (id == 0)
1525                 prog_cnt = 0;
1526
1527 end:
1528         rcu_read_unlock();
1529
1530         if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1531             (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1532             copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1533                 ret = -EFAULT;
1534
1535         fdput(f);
1536         return ret;
1537 }
1538
1539 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1540 {
1541         switch (link->map->map_type) {
1542         case BPF_MAP_TYPE_SOCKMAP:
1543                 return sock_map_delete_from_link(link->map, sk,
1544                                                  link->link_raw);
1545         case BPF_MAP_TYPE_SOCKHASH:
1546                 return sock_hash_delete_from_link(link->map, sk,
1547                                                   link->link_raw);
1548         default:
1549                 break;
1550         }
1551 }
1552
1553 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1554 {
1555         struct sk_psock_link *link;
1556
1557         while ((link = sk_psock_link_pop(psock))) {
1558                 sock_map_unlink(sk, link);
1559                 sk_psock_free_link(link);
1560         }
1561 }
1562
1563 void sock_map_unhash(struct sock *sk)
1564 {
1565         void (*saved_unhash)(struct sock *sk);
1566         struct sk_psock *psock;
1567
1568         rcu_read_lock();
1569         psock = sk_psock(sk);
1570         if (unlikely(!psock)) {
1571                 rcu_read_unlock();
1572                 saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1573         } else {
1574                 saved_unhash = psock->saved_unhash;
1575                 sock_map_remove_links(sk, psock);
1576                 rcu_read_unlock();
1577         }
1578         if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1579                 return;
1580         if (saved_unhash)
1581                 saved_unhash(sk);
1582 }
1583 EXPORT_SYMBOL_GPL(sock_map_unhash);
1584
1585 void sock_map_destroy(struct sock *sk)
1586 {
1587         void (*saved_destroy)(struct sock *sk);
1588         struct sk_psock *psock;
1589
1590         rcu_read_lock();
1591         psock = sk_psock_get(sk);
1592         if (unlikely(!psock)) {
1593                 rcu_read_unlock();
1594                 saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1595         } else {
1596                 saved_destroy = psock->saved_destroy;
1597                 sock_map_remove_links(sk, psock);
1598                 rcu_read_unlock();
1599                 sk_psock_stop(psock);
1600                 sk_psock_put(sk, psock);
1601         }
1602         if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1603                 return;
1604         if (saved_destroy)
1605                 saved_destroy(sk);
1606 }
1607 EXPORT_SYMBOL_GPL(sock_map_destroy);
1608
1609 void sock_map_close(struct sock *sk, long timeout)
1610 {
1611         void (*saved_close)(struct sock *sk, long timeout);
1612         struct sk_psock *psock;
1613
1614         lock_sock(sk);
1615         rcu_read_lock();
1616         psock = sk_psock_get(sk);
1617         if (unlikely(!psock)) {
1618                 rcu_read_unlock();
1619                 release_sock(sk);
1620                 saved_close = READ_ONCE(sk->sk_prot)->close;
1621         } else {
1622                 saved_close = psock->saved_close;
1623                 sock_map_remove_links(sk, psock);
1624                 rcu_read_unlock();
1625                 sk_psock_stop(psock);
1626                 release_sock(sk);
1627                 cancel_delayed_work_sync(&psock->work);
1628                 sk_psock_put(sk, psock);
1629         }
1630
1631         /* Make sure we do not recurse. This is a bug.
1632          * Leak the socket instead of crashing on a stack overflow.
1633          */
1634         if (WARN_ON_ONCE(saved_close == sock_map_close))
1635                 return;
1636         saved_close(sk, timeout);
1637 }
1638 EXPORT_SYMBOL_GPL(sock_map_close);
1639
1640 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1641                                        union bpf_iter_link_info *linfo,
1642                                        struct bpf_iter_aux_info *aux)
1643 {
1644         struct bpf_map *map;
1645         int err = -EINVAL;
1646
1647         if (!linfo->map.map_fd)
1648                 return -EBADF;
1649
1650         map = bpf_map_get_with_uref(linfo->map.map_fd);
1651         if (IS_ERR(map))
1652                 return PTR_ERR(map);
1653
1654         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1655             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1656                 goto put_map;
1657
1658         if (prog->aux->max_rdonly_access > map->key_size) {
1659                 err = -EACCES;
1660                 goto put_map;
1661         }
1662
1663         aux->map = map;
1664         return 0;
1665
1666 put_map:
1667         bpf_map_put_with_uref(map);
1668         return err;
1669 }
1670
1671 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1672 {
1673         bpf_map_put_with_uref(aux->map);
1674 }
1675
1676 static struct bpf_iter_reg sock_map_iter_reg = {
1677         .target                 = "sockmap",
1678         .attach_target          = sock_map_iter_attach_target,
1679         .detach_target          = sock_map_iter_detach_target,
1680         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1681         .fill_link_info         = bpf_iter_map_fill_link_info,
1682         .ctx_arg_info_size      = 2,
1683         .ctx_arg_info           = {
1684                 { offsetof(struct bpf_iter__sockmap, key),
1685                   PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1686                 { offsetof(struct bpf_iter__sockmap, sk),
1687                   PTR_TO_BTF_ID_OR_NULL },
1688         },
1689 };
1690
1691 static int __init bpf_sockmap_iter_init(void)
1692 {
1693         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1694                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1695         return bpf_iter_reg_target(&sock_map_iter_reg);
1696 }
1697 late_initcall(bpf_sockmap_iter_init);