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 struct bpf_map *sock_map_alloc(union bpf_attr *attr)
  28 {
  29         struct bpf_stab *stab;
  30         u64 cost;
  31         int err;
  32
  33         if (!capable(CAP_NET_ADMIN))
  34                 return ERR_PTR(-EPERM);
  35         if (attr->max_entries == 0 ||
  36             attr->key_size    != 4 ||
  37             (attr->value_size != sizeof(u32) &&
  38              attr->value_size != sizeof(u64)) ||
  39             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
  40                 return ERR_PTR(-EINVAL);
  41
  42         stab = kzalloc(sizeof(*stab), GFP_USER);
  43         if (!stab)
  44                 return ERR_PTR(-ENOMEM);
  45
  46         bpf_map_init_from_attr(&stab->map, attr);
  47         raw_spin_lock_init(&stab->lock);
  48
  49         /* Make sure page count doesn't overflow. */
  50         cost = (u64) stab->map.max_entries * sizeof(struct sock *);
  51         err = bpf_map_charge_init(&stab->map.memory, cost);
  52         if (err)
  53                 goto free_stab;
  54
  55         stab->sks = bpf_map_area_alloc(stab->map.max_entries *
  56                                        sizeof(struct sock *),
  57                                        stab->map.numa_node);
  58         if (stab->sks)
  59                 return &stab->map;
  60         err = -ENOMEM;
  61         bpf_map_charge_finish(&stab->map.memory);
  62 free_stab:
  63         kfree(stab);
  64         return ERR_PTR(err);
  65 }
  66
  67 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
  68 {
  69         u32 ufd = attr->target_fd;
  70         struct bpf_map *map;
  71         struct fd f;
  72         int ret;
  73
  74         if (attr->attach_flags || attr->replace_bpf_fd)
  75                 return -EINVAL;
  76
  77         f = fdget(ufd);
  78         map = __bpf_map_get(f);
  79         if (IS_ERR(map))
  80                 return PTR_ERR(map);
  81         ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
  82         fdput(f);
  83         return ret;
  84 }
  85
  86 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
  87 {
  88         u32 ufd = attr->target_fd;
  89         struct bpf_prog *prog;
  90         struct bpf_map *map;
  91         struct fd f;
  92         int ret;
  93
  94         if (attr->attach_flags || attr->replace_bpf_fd)
  95                 return -EINVAL;
  96
  97         f = fdget(ufd);
  98         map = __bpf_map_get(f);
  99         if (IS_ERR(map))
 100                 return PTR_ERR(map);
 101
 102         prog = bpf_prog_get(attr->attach_bpf_fd);
 103         if (IS_ERR(prog)) {
 104                 ret = PTR_ERR(prog);
 105                 goto put_map;
 106         }
 107
 108         if (prog->type != ptype) {
 109                 ret = -EINVAL;
 110                 goto put_prog;
 111         }
 112
 113         ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
 114 put_prog:
 115         bpf_prog_put(prog);
 116 put_map:
 117         fdput(f);
 118         return ret;
 119 }
 120
 121 static void sock_map_sk_acquire(struct sock *sk)
 122         __acquires(&sk->sk_lock.slock)
 123 {
 124         lock_sock(sk);
 125         preempt_disable();
 126         rcu_read_lock();
 127 }
 128
 129 static void sock_map_sk_release(struct sock *sk)
 130         __releases(&sk->sk_lock.slock)
 131 {
 132         rcu_read_unlock();
 133         preempt_enable();
 134         release_sock(sk);
 135 }
 136
 137 static void sock_map_add_link(struct sk_psock *psock,
 138                               struct sk_psock_link *link,
 139                               struct bpf_map *map, void *link_raw)
 140 {
 141         link->link_raw = link_raw;
 142         link->map = map;
 143         spin_lock_bh(&psock->link_lock);
 144         list_add_tail(&link->list, &psock->link);
 145         spin_unlock_bh(&psock->link_lock);
 146 }
 147
 148 static void sock_map_del_link(struct sock *sk,
 149                               struct sk_psock *psock, void *link_raw)
 150 {
 151         bool strp_stop = false, verdict_stop = false;
 152         struct sk_psock_link *link, *tmp;
 153
 154         spin_lock_bh(&psock->link_lock);
 155         list_for_each_entry_safe(link, tmp, &psock->link, list) {
 156                 if (link->link_raw == link_raw) {
 157                         struct bpf_map *map = link->map;
 158                         struct bpf_stab *stab = container_of(map, struct bpf_stab,
 159                                                              map);
 160                         if (psock->parser.enabled && stab->progs.skb_parser)
 161                                 strp_stop = true;
 162                         if (psock->parser.enabled && stab->progs.skb_verdict)
 163                                 verdict_stop = true;
 164                         list_del(&link->list);
 165                         sk_psock_free_link(link);
 166                 }
 167         }
 168         spin_unlock_bh(&psock->link_lock);
 169         if (strp_stop || verdict_stop) {
 170                 write_lock_bh(&sk->sk_callback_lock);
 171                 if (strp_stop)
 172                         sk_psock_stop_strp(sk, psock);
 173                 else
 174                         sk_psock_stop_verdict(sk, psock);
 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         struct proto *prot;
 192
 193         switch (sk->sk_type) {
 194         case SOCK_STREAM:
 195                 prot = tcp_bpf_get_proto(sk, psock);
 196                 break;
 197
 198         case SOCK_DGRAM:
 199                 prot = udp_bpf_get_proto(sk, psock);
 200                 break;
 201
 202         default:
 203                 return -EINVAL;
 204         }
 205
 206         if (IS_ERR(prot))
 207                 return PTR_ERR(prot);
 208
 209         sk_psock_update_proto(sk, psock, prot);
 210         return 0;
 211 }
 212
 213 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
 214 {
 215         struct sk_psock *psock;
 216
 217         rcu_read_lock();
 218         psock = sk_psock(sk);
 219         if (psock) {
 220                 if (sk->sk_prot->close != sock_map_close) {
 221                         psock = ERR_PTR(-EBUSY);
 222                         goto out;
 223                 }
 224
 225                 if (!refcount_inc_not_zero(&psock->refcnt))
 226                         psock = ERR_PTR(-EBUSY);
 227         }
 228 out:
 229         rcu_read_unlock();
 230         return psock;
 231 }
 232
 233 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
 234                          struct sock *sk)
 235 {
 236         struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
 237         struct sk_psock *psock;
 238         int ret;
 239
 240         skb_verdict = READ_ONCE(progs->skb_verdict);
 241         if (skb_verdict) {
 242                 skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
 243                 if (IS_ERR(skb_verdict))
 244                         return PTR_ERR(skb_verdict);
 245         }
 246
 247         skb_parser = READ_ONCE(progs->skb_parser);
 248         if (skb_parser) {
 249                 skb_parser = bpf_prog_inc_not_zero(skb_parser);
 250                 if (IS_ERR(skb_parser)) {
 251                         ret = PTR_ERR(skb_parser);
 252                         goto out_put_skb_verdict;
 253                 }
 254         }
 255
 256         msg_parser = READ_ONCE(progs->msg_parser);
 257         if (msg_parser) {
 258                 msg_parser = bpf_prog_inc_not_zero(msg_parser);
 259                 if (IS_ERR(msg_parser)) {
 260                         ret = PTR_ERR(msg_parser);
 261                         goto out_put_skb_parser;
 262                 }
 263         }
 264
 265         psock = sock_map_psock_get_checked(sk);
 266         if (IS_ERR(psock)) {
 267                 ret = PTR_ERR(psock);
 268                 goto out_progs;
 269         }
 270
 271         if (psock) {
 272                 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 273                     (skb_parser  && READ_ONCE(psock->progs.skb_parser)) ||
 274                     (skb_verdict && READ_ONCE(psock->progs.skb_verdict))) {
 275                         sk_psock_put(sk, psock);
 276                         ret = -EBUSY;
 277                         goto out_progs;
 278                 }
 279         } else {
 280                 psock = sk_psock_init(sk, map->numa_node);
 281                 if (IS_ERR(psock)) {
 282                         ret = PTR_ERR(psock);
 283                         goto out_progs;
 284                 }
 285         }
 286
 287         if (msg_parser)
 288                 psock_set_prog(&psock->progs.msg_parser, msg_parser);
 289
 290         ret = sock_map_init_proto(sk, psock);
 291         if (ret < 0)
 292                 goto out_drop;
 293
 294         write_lock_bh(&sk->sk_callback_lock);
 295         if (skb_parser && skb_verdict && !psock->parser.enabled) {
 296                 ret = sk_psock_init_strp(sk, psock);
 297                 if (ret)
 298                         goto out_unlock_drop;
 299                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 300                 psock_set_prog(&psock->progs.skb_parser, skb_parser);
 301                 sk_psock_start_strp(sk, psock);
 302         } else if (!skb_parser && skb_verdict && !psock->parser.enabled) {
 303                 psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
 304                 sk_psock_start_verdict(sk,psock);
 305         }
 306         write_unlock_bh(&sk->sk_callback_lock);
 307         return 0;
 308 out_unlock_drop:
 309         write_unlock_bh(&sk->sk_callback_lock);
 310 out_drop:
 311         sk_psock_put(sk, psock);
 312 out_progs:
 313         if (msg_parser)
 314                 bpf_prog_put(msg_parser);
 315 out_put_skb_parser:
 316         if (skb_parser)
 317                 bpf_prog_put(skb_parser);
 318 out_put_skb_verdict:
 319         if (skb_verdict)
 320                 bpf_prog_put(skb_verdict);
 321         return ret;
 322 }
 323
 324 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
 325 {
 326         struct sk_psock *psock;
 327         int ret;
 328
 329         psock = sock_map_psock_get_checked(sk);
 330         if (IS_ERR(psock))
 331                 return PTR_ERR(psock);
 332
 333         if (!psock) {
 334                 psock = sk_psock_init(sk, map->numa_node);
 335                 if (IS_ERR(psock))
 336                         return PTR_ERR(psock);
 337         }
 338
 339         ret = sock_map_init_proto(sk, psock);
 340         if (ret < 0)
 341                 sk_psock_put(sk, psock);
 342         return ret;
 343 }
 344
 345 static void sock_map_free(struct bpf_map *map)
 346 {
 347         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 348         int i;
 349
 350         /* After the sync no updates or deletes will be in-flight so it
 351          * is safe to walk map and remove entries without risking a race
 352          * in EEXIST update case.
 353          */
 354         synchronize_rcu();
 355         for (i = 0; i < stab->map.max_entries; i++) {
 356                 struct sock **psk = &stab->sks[i];
 357                 struct sock *sk;
 358
 359                 sk = xchg(psk, NULL);
 360                 if (sk) {
 361                         lock_sock(sk);
 362                         rcu_read_lock();
 363                         sock_map_unref(sk, psk);
 364                         rcu_read_unlock();
 365                         release_sock(sk);
 366                 }
 367         }
 368
 369         /* wait for psock readers accessing its map link */
 370         synchronize_rcu();
 371
 372         bpf_map_area_free(stab->sks);
 373         kfree(stab);
 374 }
 375
 376 static void sock_map_release_progs(struct bpf_map *map)
 377 {
 378         psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
 379 }
 380
 381 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
 382 {
 383         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 384
 385         WARN_ON_ONCE(!rcu_read_lock_held());
 386
 387         if (unlikely(key >= map->max_entries))
 388                 return NULL;
 389         return READ_ONCE(stab->sks[key]);
 390 }
 391
 392 static void *sock_map_lookup(struct bpf_map *map, void *key)
 393 {
 394         struct sock *sk;
 395
 396         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 397         if (!sk)
 398                 return NULL;
 399         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
 400                 return NULL;
 401         return sk;
 402 }
 403
 404 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
 405 {
 406         struct sock *sk;
 407
 408         if (map->value_size != sizeof(u64))
 409                 return ERR_PTR(-ENOSPC);
 410
 411         sk = __sock_map_lookup_elem(map, *(u32 *)key);
 412         if (!sk)
 413                 return ERR_PTR(-ENOENT);
 414
 415         __sock_gen_cookie(sk);
 416         return &sk->sk_cookie;
 417 }
 418
 419 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
 420                              struct sock **psk)
 421 {
 422         struct sock *sk;
 423         int err = 0;
 424
 425         raw_spin_lock_bh(&stab->lock);
 426         sk = *psk;
 427         if (!sk_test || sk_test == sk)
 428                 sk = xchg(psk, NULL);
 429
 430         if (likely(sk))
 431                 sock_map_unref(sk, psk);
 432         else
 433                 err = -EINVAL;
 434
 435         raw_spin_unlock_bh(&stab->lock);
 436         return err;
 437 }
 438
 439 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
 440                                       void *link_raw)
 441 {
 442         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 443
 444         __sock_map_delete(stab, sk, link_raw);
 445 }
 446
 447 static int sock_map_delete_elem(struct bpf_map *map, void *key)
 448 {
 449         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 450         u32 i = *(u32 *)key;
 451         struct sock **psk;
 452
 453         if (unlikely(i >= map->max_entries))
 454                 return -EINVAL;
 455
 456         psk = &stab->sks[i];
 457         return __sock_map_delete(stab, NULL, psk);
 458 }
 459
 460 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
 461 {
 462         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 463         u32 i = key ? *(u32 *)key : U32_MAX;
 464         u32 *key_next = next;
 465
 466         if (i == stab->map.max_entries - 1)
 467                 return -ENOENT;
 468         if (i >= stab->map.max_entries)
 469                 *key_next = 0;
 470         else
 471                 *key_next = i + 1;
 472         return 0;
 473 }
 474
 475 static bool sock_map_redirect_allowed(const struct sock *sk);
 476
 477 static int sock_map_update_common(struct bpf_map *map, u32 idx,
 478                                   struct sock *sk, u64 flags)
 479 {
 480         struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
 481         struct sk_psock_link *link;
 482         struct sk_psock *psock;
 483         struct sock *osk;
 484         int ret;
 485
 486         WARN_ON_ONCE(!rcu_read_lock_held());
 487         if (unlikely(flags > BPF_EXIST))
 488                 return -EINVAL;
 489         if (unlikely(idx >= map->max_entries))
 490                 return -E2BIG;
 491
 492         link = sk_psock_init_link();
 493         if (!link)
 494                 return -ENOMEM;
 495
 496         /* Only sockets we can redirect into/from in BPF need to hold
 497          * refs to parser/verdict progs and have their sk_data_ready
 498          * and sk_write_space callbacks overridden.
 499          */
 500         if (sock_map_redirect_allowed(sk))
 501                 ret = sock_map_link(map, &stab->progs, sk);
 502         else
 503                 ret = sock_map_link_no_progs(map, sk);
 504         if (ret < 0)
 505                 goto out_free;
 506
 507         psock = sk_psock(sk);
 508         WARN_ON_ONCE(!psock);
 509
 510         raw_spin_lock_bh(&stab->lock);
 511         osk = stab->sks[idx];
 512         if (osk && flags == BPF_NOEXIST) {
 513                 ret = -EEXIST;
 514                 goto out_unlock;
 515         } else if (!osk && flags == BPF_EXIST) {
 516                 ret = -ENOENT;
 517                 goto out_unlock;
 518         }
 519
 520         sock_map_add_link(psock, link, map, &stab->sks[idx]);
 521         stab->sks[idx] = sk;
 522         if (osk)
 523                 sock_map_unref(osk, &stab->sks[idx]);
 524         raw_spin_unlock_bh(&stab->lock);
 525         return 0;
 526 out_unlock:
 527         raw_spin_unlock_bh(&stab->lock);
 528         if (psock)
 529                 sk_psock_put(sk, psock);
 530 out_free:
 531         sk_psock_free_link(link);
 532         return ret;
 533 }
 534
 535 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
 536 {
 537         return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
 538                ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
 539                ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
 540 }
 541
 542 static bool sk_is_tcp(const struct sock *sk)
 543 {
 544         return sk->sk_type == SOCK_STREAM &&
 545                sk->sk_protocol == IPPROTO_TCP;
 546 }
 547
 548 static bool sk_is_udp(const struct sock *sk)
 549 {
 550         return sk->sk_type == SOCK_DGRAM &&
 551                sk->sk_protocol == IPPROTO_UDP;
 552 }
 553
 554 static bool sock_map_redirect_allowed(const struct sock *sk)
 555 {
 556         return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
 557 }
 558
 559 static bool sock_map_sk_is_suitable(const struct sock *sk)
 560 {
 561         return sk_is_tcp(sk) || sk_is_udp(sk);
 562 }
 563
 564 static bool sock_map_sk_state_allowed(const struct sock *sk)
 565 {
 566         if (sk_is_tcp(sk))
 567                 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
 568         else if (sk_is_udp(sk))
 569                 return sk_hashed(sk);
 570
 571         return false;
 572 }
 573
 574 static int sock_hash_update_common(struct bpf_map *map, void *key,
 575                                    struct sock *sk, u64 flags);
 576
 577 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
 578                              u64 flags)
 579 {
 580         struct socket *sock;
 581         struct sock *sk;
 582         int ret;
 583         u64 ufd;
 584
 585         if (map->value_size == sizeof(u64))
 586                 ufd = *(u64 *)value;
 587         else
 588                 ufd = *(u32 *)value;
 589         if (ufd > S32_MAX)
 590                 return -EINVAL;
 591
 592         sock = sockfd_lookup(ufd, &ret);
 593         if (!sock)
 594                 return ret;
 595         sk = sock->sk;
 596         if (!sk) {
 597                 ret = -EINVAL;
 598                 goto out;
 599         }
 600         if (!sock_map_sk_is_suitable(sk)) {
 601                 ret = -EOPNOTSUPP;
 602                 goto out;
 603         }
 604
 605         sock_map_sk_acquire(sk);
 606         if (!sock_map_sk_state_allowed(sk))
 607                 ret = -EOPNOTSUPP;
 608         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 609                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 610         else
 611                 ret = sock_hash_update_common(map, key, sk, flags);
 612         sock_map_sk_release(sk);
 613 out:
 614         fput(sock->file);
 615         return ret;
 616 }
 617
 618 static int sock_map_update_elem(struct bpf_map *map, void *key,
 619                                 void *value, u64 flags)
 620 {
 621         struct sock *sk = (struct sock *)value;
 622         int ret;
 623
 624         if (unlikely(!sk || !sk_fullsock(sk)))
 625                 return -EINVAL;
 626
 627         if (!sock_map_sk_is_suitable(sk))
 628                 return -EOPNOTSUPP;
 629
 630         local_bh_disable();
 631         bh_lock_sock(sk);
 632         if (!sock_map_sk_state_allowed(sk))
 633                 ret = -EOPNOTSUPP;
 634         else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
 635                 ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
 636         else
 637                 ret = sock_hash_update_common(map, key, sk, flags);
 638         bh_unlock_sock(sk);
 639         local_bh_enable();
 640         return ret;
 641 }
 642
 643 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
 644            struct bpf_map *, map, void *, key, u64, flags)
 645 {
 646         WARN_ON_ONCE(!rcu_read_lock_held());
 647
 648         if (likely(sock_map_sk_is_suitable(sops->sk) &&
 649                    sock_map_op_okay(sops)))
 650                 return sock_map_update_common(map, *(u32 *)key, sops->sk,
 651                                               flags);
 652         return -EOPNOTSUPP;
 653 }
 654
 655 const struct bpf_func_proto bpf_sock_map_update_proto = {
 656         .func           = bpf_sock_map_update,
 657         .gpl_only       = false,
 658         .pkt_access     = true,
 659         .ret_type       = RET_INTEGER,
 660         .arg1_type      = ARG_PTR_TO_CTX,
 661         .arg2_type      = ARG_CONST_MAP_PTR,
 662         .arg3_type      = ARG_PTR_TO_MAP_KEY,
 663         .arg4_type      = ARG_ANYTHING,
 664 };
 665
 666 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
 667            struct bpf_map *, map, u32, key, u64, flags)
 668 {
 669         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 670         struct sock *sk;
 671
 672         if (unlikely(flags & ~(BPF_F_INGRESS)))
 673                 return SK_DROP;
 674
 675         sk = __sock_map_lookup_elem(map, key);
 676         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 677                 return SK_DROP;
 678
 679         tcb->bpf.flags = flags;
 680         tcb->bpf.sk_redir = sk;
 681         return SK_PASS;
 682 }
 683
 684 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
 685         .func           = bpf_sk_redirect_map,
 686         .gpl_only       = false,
 687         .ret_type       = RET_INTEGER,
 688         .arg1_type      = ARG_PTR_TO_CTX,
 689         .arg2_type      = ARG_CONST_MAP_PTR,
 690         .arg3_type      = ARG_ANYTHING,
 691         .arg4_type      = ARG_ANYTHING,
 692 };
 693
 694 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
 695            struct bpf_map *, map, u32, key, u64, flags)
 696 {
 697         struct sock *sk;
 698
 699         if (unlikely(flags & ~(BPF_F_INGRESS)))
 700                 return SK_DROP;
 701
 702         sk = __sock_map_lookup_elem(map, key);
 703         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
 704                 return SK_DROP;
 705
 706         msg->flags = flags;
 707         msg->sk_redir = sk;
 708         return SK_PASS;
 709 }
 710
 711 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
 712         .func           = bpf_msg_redirect_map,
 713         .gpl_only       = false,
 714         .ret_type       = RET_INTEGER,
 715         .arg1_type      = ARG_PTR_TO_CTX,
 716         .arg2_type      = ARG_CONST_MAP_PTR,
 717         .arg3_type      = ARG_ANYTHING,
 718         .arg4_type      = ARG_ANYTHING,
 719 };
 720
 721 struct sock_map_seq_info {
 722         struct bpf_map *map;
 723         struct sock *sk;
 724         u32 index;
 725 };
 726
 727 struct bpf_iter__sockmap {
 728         __bpf_md_ptr(struct bpf_iter_meta *, meta);
 729         __bpf_md_ptr(struct bpf_map *, map);
 730         __bpf_md_ptr(void *, key);
 731         __bpf_md_ptr(struct sock *, sk);
 732 };
 733
 734 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
 735                      struct bpf_map *map, void *key,
 736                      struct sock *sk)
 737
 738 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
 739 {
 740         if (unlikely(info->index >= info->map->max_entries))
 741                 return NULL;
 742
 743         info->sk = __sock_map_lookup_elem(info->map, info->index);
 744
 745         /* can't return sk directly, since that might be NULL */
 746         return info;
 747 }
 748
 749 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
 750         __acquires(rcu)
 751 {
 752         struct sock_map_seq_info *info = seq->private;
 753
 754         if (*pos == 0)
 755                 ++*pos;
 756
 757         /* pairs with sock_map_seq_stop */
 758         rcu_read_lock();
 759         return sock_map_seq_lookup_elem(info);
 760 }
 761
 762 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 763         __must_hold(rcu)
 764 {
 765         struct sock_map_seq_info *info = seq->private;
 766
 767         ++*pos;
 768         ++info->index;
 769
 770         return sock_map_seq_lookup_elem(info);
 771 }
 772
 773 static int sock_map_seq_show(struct seq_file *seq, void *v)
 774         __must_hold(rcu)
 775 {
 776         struct sock_map_seq_info *info = seq->private;
 777         struct bpf_iter__sockmap ctx = {};
 778         struct bpf_iter_meta meta;
 779         struct bpf_prog *prog;
 780
 781         meta.seq = seq;
 782         prog = bpf_iter_get_info(&meta, !v);
 783         if (!prog)
 784                 return 0;
 785
 786         ctx.meta = &meta;
 787         ctx.map = info->map;
 788         if (v) {
 789                 ctx.key = &info->index;
 790                 ctx.sk = info->sk;
 791         }
 792
 793         return bpf_iter_run_prog(prog, &ctx);
 794 }
 795
 796 static void sock_map_seq_stop(struct seq_file *seq, void *v)
 797         __releases(rcu)
 798 {
 799         if (!v)
 800                 (void)sock_map_seq_show(seq, NULL);
 801
 802         /* pairs with sock_map_seq_start */
 803         rcu_read_unlock();
 804 }
 805
 806 static const struct seq_operations sock_map_seq_ops = {
 807         .start  = sock_map_seq_start,
 808         .next   = sock_map_seq_next,
 809         .stop   = sock_map_seq_stop,
 810         .show   = sock_map_seq_show,
 811 };
 812
 813 static int sock_map_init_seq_private(void *priv_data,
 814                                      struct bpf_iter_aux_info *aux)
 815 {
 816         struct sock_map_seq_info *info = priv_data;
 817
 818         info->map = aux->map;
 819         return 0;
 820 }
 821
 822 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
 823         .seq_ops                = &sock_map_seq_ops,
 824         .init_seq_private       = sock_map_init_seq_private,
 825         .seq_priv_size          = sizeof(struct sock_map_seq_info),
 826 };
 827
 828 static int sock_map_btf_id;
 829 const struct bpf_map_ops sock_map_ops = {
 830         .map_meta_equal         = bpf_map_meta_equal,
 831         .map_alloc              = sock_map_alloc,
 832         .map_free               = sock_map_free,
 833         .map_get_next_key       = sock_map_get_next_key,
 834         .map_lookup_elem_sys_only = sock_map_lookup_sys,
 835         .map_update_elem        = sock_map_update_elem,
 836         .map_delete_elem        = sock_map_delete_elem,
 837         .map_lookup_elem        = sock_map_lookup,
 838         .map_release_uref       = sock_map_release_progs,
 839         .map_check_btf          = map_check_no_btf,
 840         .map_btf_name           = "bpf_stab",
 841         .map_btf_id             = &sock_map_btf_id,
 842         .iter_seq_info          = &sock_map_iter_seq_info,
 843 };
 844
 845 struct bpf_shtab_elem {
 846         struct rcu_head rcu;
 847         u32 hash;
 848         struct sock *sk;
 849         struct hlist_node node;
 850         u8 key[];
 851 };
 852
 853 struct bpf_shtab_bucket {
 854         struct hlist_head head;
 855         raw_spinlock_t lock;
 856 };
 857
 858 struct bpf_shtab {
 859         struct bpf_map map;
 860         struct bpf_shtab_bucket *buckets;
 861         u32 buckets_num;
 862         u32 elem_size;
 863         struct sk_psock_progs progs;
 864         atomic_t count;
 865 };
 866
 867 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
 868 {
 869         return jhash(key, len, 0);
 870 }
 871
 872 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
 873                                                         u32 hash)
 874 {
 875         return &htab->buckets[hash & (htab->buckets_num - 1)];
 876 }
 877
 878 static struct bpf_shtab_elem *
 879 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
 880                           u32 key_size)
 881 {
 882         struct bpf_shtab_elem *elem;
 883
 884         hlist_for_each_entry_rcu(elem, head, node) {
 885                 if (elem->hash == hash &&
 886                     !memcmp(&elem->key, key, key_size))
 887                         return elem;
 888         }
 889
 890         return NULL;
 891 }
 892
 893 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
 894 {
 895         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 896         u32 key_size = map->key_size, hash;
 897         struct bpf_shtab_bucket *bucket;
 898         struct bpf_shtab_elem *elem;
 899
 900         WARN_ON_ONCE(!rcu_read_lock_held());
 901
 902         hash = sock_hash_bucket_hash(key, key_size);
 903         bucket = sock_hash_select_bucket(htab, hash);
 904         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 905
 906         return elem ? elem->sk : NULL;
 907 }
 908
 909 static void sock_hash_free_elem(struct bpf_shtab *htab,
 910                                 struct bpf_shtab_elem *elem)
 911 {
 912         atomic_dec(&htab->count);
 913         kfree_rcu(elem, rcu);
 914 }
 915
 916 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
 917                                        void *link_raw)
 918 {
 919         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 920         struct bpf_shtab_elem *elem_probe, *elem = link_raw;
 921         struct bpf_shtab_bucket *bucket;
 922
 923         WARN_ON_ONCE(!rcu_read_lock_held());
 924         bucket = sock_hash_select_bucket(htab, elem->hash);
 925
 926         /* elem may be deleted in parallel from the map, but access here
 927          * is okay since it's going away only after RCU grace period.
 928          * However, we need to check whether it's still present.
 929          */
 930         raw_spin_lock_bh(&bucket->lock);
 931         elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
 932                                                elem->key, map->key_size);
 933         if (elem_probe && elem_probe == elem) {
 934                 hlist_del_rcu(&elem->node);
 935                 sock_map_unref(elem->sk, elem);
 936                 sock_hash_free_elem(htab, elem);
 937         }
 938         raw_spin_unlock_bh(&bucket->lock);
 939 }
 940
 941 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
 942 {
 943         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 944         u32 hash, key_size = map->key_size;
 945         struct bpf_shtab_bucket *bucket;
 946         struct bpf_shtab_elem *elem;
 947         int ret = -ENOENT;
 948
 949         hash = sock_hash_bucket_hash(key, key_size);
 950         bucket = sock_hash_select_bucket(htab, hash);
 951
 952         raw_spin_lock_bh(&bucket->lock);
 953         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
 954         if (elem) {
 955                 hlist_del_rcu(&elem->node);
 956                 sock_map_unref(elem->sk, elem);
 957                 sock_hash_free_elem(htab, elem);
 958                 ret = 0;
 959         }
 960         raw_spin_unlock_bh(&bucket->lock);
 961         return ret;
 962 }
 963
 964 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
 965                                                    void *key, u32 key_size,
 966                                                    u32 hash, struct sock *sk,
 967                                                    struct bpf_shtab_elem *old)
 968 {
 969         struct bpf_shtab_elem *new;
 970
 971         if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
 972                 if (!old) {
 973                         atomic_dec(&htab->count);
 974                         return ERR_PTR(-E2BIG);
 975                 }
 976         }
 977
 978         new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
 979                            htab->map.numa_node);
 980         if (!new) {
 981                 atomic_dec(&htab->count);
 982                 return ERR_PTR(-ENOMEM);
 983         }
 984         memcpy(new->key, key, key_size);
 985         new->sk = sk;
 986         new->hash = hash;
 987         return new;
 988 }
 989
 990 static int sock_hash_update_common(struct bpf_map *map, void *key,
 991                                    struct sock *sk, u64 flags)
 992 {
 993         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
 994         u32 key_size = map->key_size, hash;
 995         struct bpf_shtab_elem *elem, *elem_new;
 996         struct bpf_shtab_bucket *bucket;
 997         struct sk_psock_link *link;
 998         struct sk_psock *psock;
 999         int ret;
1000
1001         WARN_ON_ONCE(!rcu_read_lock_held());
1002         if (unlikely(flags > BPF_EXIST))
1003                 return -EINVAL;
1004
1005         link = sk_psock_init_link();
1006         if (!link)
1007                 return -ENOMEM;
1008
1009         /* Only sockets we can redirect into/from in BPF need to hold
1010          * refs to parser/verdict progs and have their sk_data_ready
1011          * and sk_write_space callbacks overridden.
1012          */
1013         if (sock_map_redirect_allowed(sk))
1014                 ret = sock_map_link(map, &htab->progs, sk);
1015         else
1016                 ret = sock_map_link_no_progs(map, sk);
1017         if (ret < 0)
1018                 goto out_free;
1019
1020         psock = sk_psock(sk);
1021         WARN_ON_ONCE(!psock);
1022
1023         hash = sock_hash_bucket_hash(key, key_size);
1024         bucket = sock_hash_select_bucket(htab, hash);
1025
1026         raw_spin_lock_bh(&bucket->lock);
1027         elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1028         if (elem && flags == BPF_NOEXIST) {
1029                 ret = -EEXIST;
1030                 goto out_unlock;
1031         } else if (!elem && flags == BPF_EXIST) {
1032                 ret = -ENOENT;
1033                 goto out_unlock;
1034         }
1035
1036         elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1037         if (IS_ERR(elem_new)) {
1038                 ret = PTR_ERR(elem_new);
1039                 goto out_unlock;
1040         }
1041
1042         sock_map_add_link(psock, link, map, elem_new);
1043         /* Add new element to the head of the list, so that
1044          * concurrent search will find it before old elem.
1045          */
1046         hlist_add_head_rcu(&elem_new->node, &bucket->head);
1047         if (elem) {
1048                 hlist_del_rcu(&elem->node);
1049                 sock_map_unref(elem->sk, elem);
1050                 sock_hash_free_elem(htab, elem);
1051         }
1052         raw_spin_unlock_bh(&bucket->lock);
1053         return 0;
1054 out_unlock:
1055         raw_spin_unlock_bh(&bucket->lock);
1056         sk_psock_put(sk, psock);
1057 out_free:
1058         sk_psock_free_link(link);
1059         return ret;
1060 }
1061
1062 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1063                                   void *key_next)
1064 {
1065         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1066         struct bpf_shtab_elem *elem, *elem_next;
1067         u32 hash, key_size = map->key_size;
1068         struct hlist_head *head;
1069         int i = 0;
1070
1071         if (!key)
1072                 goto find_first_elem;
1073         hash = sock_hash_bucket_hash(key, key_size);
1074         head = &sock_hash_select_bucket(htab, hash)->head;
1075         elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1076         if (!elem)
1077                 goto find_first_elem;
1078
1079         elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1080                                      struct bpf_shtab_elem, node);
1081         if (elem_next) {
1082                 memcpy(key_next, elem_next->key, key_size);
1083                 return 0;
1084         }
1085
1086         i = hash & (htab->buckets_num - 1);
1087         i++;
1088 find_first_elem:
1089         for (; i < htab->buckets_num; i++) {
1090                 head = &sock_hash_select_bucket(htab, i)->head;
1091                 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1092                                              struct bpf_shtab_elem, node);
1093                 if (elem_next) {
1094                         memcpy(key_next, elem_next->key, key_size);
1095                         return 0;
1096                 }
1097         }
1098
1099         return -ENOENT;
1100 }
1101
1102 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1103 {
1104         struct bpf_shtab *htab;
1105         int i, err;
1106         u64 cost;
1107
1108         if (!capable(CAP_NET_ADMIN))
1109                 return ERR_PTR(-EPERM);
1110         if (attr->max_entries == 0 ||
1111             attr->key_size    == 0 ||
1112             (attr->value_size != sizeof(u32) &&
1113              attr->value_size != sizeof(u64)) ||
1114             attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1115                 return ERR_PTR(-EINVAL);
1116         if (attr->key_size > MAX_BPF_STACK)
1117                 return ERR_PTR(-E2BIG);
1118
1119         htab = kzalloc(sizeof(*htab), GFP_USER);
1120         if (!htab)
1121                 return ERR_PTR(-ENOMEM);
1122
1123         bpf_map_init_from_attr(&htab->map, attr);
1124
1125         htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1126         htab->elem_size = sizeof(struct bpf_shtab_elem) +
1127                           round_up(htab->map.key_size, 8);
1128         if (htab->buckets_num == 0 ||
1129             htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1130                 err = -EINVAL;
1131                 goto free_htab;
1132         }
1133
1134         cost = (u64) htab->buckets_num * sizeof(struct bpf_shtab_bucket) +
1135                (u64) htab->elem_size * htab->map.max_entries;
1136         if (cost >= U32_MAX - PAGE_SIZE) {
1137                 err = -EINVAL;
1138                 goto free_htab;
1139         }
1140         err = bpf_map_charge_init(&htab->map.memory, cost);
1141         if (err)
1142                 goto free_htab;
1143
1144         htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1145                                            sizeof(struct bpf_shtab_bucket),
1146                                            htab->map.numa_node);
1147         if (!htab->buckets) {
1148                 bpf_map_charge_finish(&htab->map.memory);
1149                 err = -ENOMEM;
1150                 goto free_htab;
1151         }
1152
1153         for (i = 0; i < htab->buckets_num; i++) {
1154                 INIT_HLIST_HEAD(&htab->buckets[i].head);
1155                 raw_spin_lock_init(&htab->buckets[i].lock);
1156         }
1157
1158         return &htab->map;
1159 free_htab:
1160         kfree(htab);
1161         return ERR_PTR(err);
1162 }
1163
1164 static void sock_hash_free(struct bpf_map *map)
1165 {
1166         struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1167         struct bpf_shtab_bucket *bucket;
1168         struct hlist_head unlink_list;
1169         struct bpf_shtab_elem *elem;
1170         struct hlist_node *node;
1171         int i;
1172
1173         /* After the sync no updates or deletes will be in-flight so it
1174          * is safe to walk map and remove entries without risking a race
1175          * in EEXIST update case.
1176          */
1177         synchronize_rcu();
1178         for (i = 0; i < htab->buckets_num; i++) {
1179                 bucket = sock_hash_select_bucket(htab, i);
1180
1181                 /* We are racing with sock_hash_delete_from_link to
1182                  * enter the spin-lock critical section. Every socket on
1183                  * the list is still linked to sockhash. Since link
1184                  * exists, psock exists and holds a ref to socket. That
1185                  * lets us to grab a socket ref too.
1186                  */
1187                 raw_spin_lock_bh(&bucket->lock);
1188                 hlist_for_each_entry(elem, &bucket->head, node)
1189                         sock_hold(elem->sk);
1190                 hlist_move_list(&bucket->head, &unlink_list);
1191                 raw_spin_unlock_bh(&bucket->lock);
1192
1193                 /* Process removed entries out of atomic context to
1194                  * block for socket lock before deleting the psock's
1195                  * link to sockhash.
1196                  */
1197                 hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1198                         hlist_del(&elem->node);
1199                         lock_sock(elem->sk);
1200                         rcu_read_lock();
1201                         sock_map_unref(elem->sk, elem);
1202                         rcu_read_unlock();
1203                         release_sock(elem->sk);
1204                         sock_put(elem->sk);
1205                         sock_hash_free_elem(htab, elem);
1206                 }
1207         }
1208
1209         /* wait for psock readers accessing its map link */
1210         synchronize_rcu();
1211
1212         bpf_map_area_free(htab->buckets);
1213         kfree(htab);
1214 }
1215
1216 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1217 {
1218         struct sock *sk;
1219
1220         if (map->value_size != sizeof(u64))
1221                 return ERR_PTR(-ENOSPC);
1222
1223         sk = __sock_hash_lookup_elem(map, key);
1224         if (!sk)
1225                 return ERR_PTR(-ENOENT);
1226
1227         __sock_gen_cookie(sk);
1228         return &sk->sk_cookie;
1229 }
1230
1231 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1232 {
1233         struct sock *sk;
1234
1235         sk = __sock_hash_lookup_elem(map, key);
1236         if (!sk)
1237                 return NULL;
1238         if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1239                 return NULL;
1240         return sk;
1241 }
1242
1243 static void sock_hash_release_progs(struct bpf_map *map)
1244 {
1245         psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1246 }
1247
1248 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1249            struct bpf_map *, map, void *, key, u64, flags)
1250 {
1251         WARN_ON_ONCE(!rcu_read_lock_held());
1252
1253         if (likely(sock_map_sk_is_suitable(sops->sk) &&
1254                    sock_map_op_okay(sops)))
1255                 return sock_hash_update_common(map, key, sops->sk, flags);
1256         return -EOPNOTSUPP;
1257 }
1258
1259 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1260         .func           = bpf_sock_hash_update,
1261         .gpl_only       = false,
1262         .pkt_access     = true,
1263         .ret_type       = RET_INTEGER,
1264         .arg1_type      = ARG_PTR_TO_CTX,
1265         .arg2_type      = ARG_CONST_MAP_PTR,
1266         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1267         .arg4_type      = ARG_ANYTHING,
1268 };
1269
1270 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1271            struct bpf_map *, map, void *, key, u64, flags)
1272 {
1273         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1274         struct sock *sk;
1275
1276         if (unlikely(flags & ~(BPF_F_INGRESS)))
1277                 return SK_DROP;
1278
1279         sk = __sock_hash_lookup_elem(map, key);
1280         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1281                 return SK_DROP;
1282
1283         tcb->bpf.flags = flags;
1284         tcb->bpf.sk_redir = sk;
1285         return SK_PASS;
1286 }
1287
1288 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1289         .func           = bpf_sk_redirect_hash,
1290         .gpl_only       = false,
1291         .ret_type       = RET_INTEGER,
1292         .arg1_type      = ARG_PTR_TO_CTX,
1293         .arg2_type      = ARG_CONST_MAP_PTR,
1294         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1295         .arg4_type      = ARG_ANYTHING,
1296 };
1297
1298 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1299            struct bpf_map *, map, void *, key, u64, flags)
1300 {
1301         struct sock *sk;
1302
1303         if (unlikely(flags & ~(BPF_F_INGRESS)))
1304                 return SK_DROP;
1305
1306         sk = __sock_hash_lookup_elem(map, key);
1307         if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1308                 return SK_DROP;
1309
1310         msg->flags = flags;
1311         msg->sk_redir = sk;
1312         return SK_PASS;
1313 }
1314
1315 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1316         .func           = bpf_msg_redirect_hash,
1317         .gpl_only       = false,
1318         .ret_type       = RET_INTEGER,
1319         .arg1_type      = ARG_PTR_TO_CTX,
1320         .arg2_type      = ARG_CONST_MAP_PTR,
1321         .arg3_type      = ARG_PTR_TO_MAP_KEY,
1322         .arg4_type      = ARG_ANYTHING,
1323 };
1324
1325 struct sock_hash_seq_info {
1326         struct bpf_map *map;
1327         struct bpf_shtab *htab;
1328         u32 bucket_id;
1329 };
1330
1331 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1332                                      struct bpf_shtab_elem *prev_elem)
1333 {
1334         const struct bpf_shtab *htab = info->htab;
1335         struct bpf_shtab_bucket *bucket;
1336         struct bpf_shtab_elem *elem;
1337         struct hlist_node *node;
1338
1339         /* try to find next elem in the same bucket */
1340         if (prev_elem) {
1341                 node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1342                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1343                 if (elem)
1344                         return elem;
1345
1346                 /* no more elements, continue in the next bucket */
1347                 info->bucket_id++;
1348         }
1349
1350         for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1351                 bucket = &htab->buckets[info->bucket_id];
1352                 node = rcu_dereference(hlist_first_rcu(&bucket->head));
1353                 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1354                 if (elem)
1355                         return elem;
1356         }
1357
1358         return NULL;
1359 }
1360
1361 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1362         __acquires(rcu)
1363 {
1364         struct sock_hash_seq_info *info = seq->private;
1365
1366         if (*pos == 0)
1367                 ++*pos;
1368
1369         /* pairs with sock_hash_seq_stop */
1370         rcu_read_lock();
1371         return sock_hash_seq_find_next(info, NULL);
1372 }
1373
1374 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1375         __must_hold(rcu)
1376 {
1377         struct sock_hash_seq_info *info = seq->private;
1378
1379         ++*pos;
1380         return sock_hash_seq_find_next(info, v);
1381 }
1382
1383 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1384         __must_hold(rcu)
1385 {
1386         struct sock_hash_seq_info *info = seq->private;
1387         struct bpf_iter__sockmap ctx = {};
1388         struct bpf_shtab_elem *elem = v;
1389         struct bpf_iter_meta meta;
1390         struct bpf_prog *prog;
1391
1392         meta.seq = seq;
1393         prog = bpf_iter_get_info(&meta, !elem);
1394         if (!prog)
1395                 return 0;
1396
1397         ctx.meta = &meta;
1398         ctx.map = info->map;
1399         if (elem) {
1400                 ctx.key = elem->key;
1401                 ctx.sk = elem->sk;
1402         }
1403
1404         return bpf_iter_run_prog(prog, &ctx);
1405 }
1406
1407 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1408         __releases(rcu)
1409 {
1410         if (!v)
1411                 (void)sock_hash_seq_show(seq, NULL);
1412
1413         /* pairs with sock_hash_seq_start */
1414         rcu_read_unlock();
1415 }
1416
1417 static const struct seq_operations sock_hash_seq_ops = {
1418         .start  = sock_hash_seq_start,
1419         .next   = sock_hash_seq_next,
1420         .stop   = sock_hash_seq_stop,
1421         .show   = sock_hash_seq_show,
1422 };
1423
1424 static int sock_hash_init_seq_private(void *priv_data,
1425                                      struct bpf_iter_aux_info *aux)
1426 {
1427         struct sock_hash_seq_info *info = priv_data;
1428
1429         info->map = aux->map;
1430         info->htab = container_of(aux->map, struct bpf_shtab, map);
1431         return 0;
1432 }
1433
1434 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1435         .seq_ops                = &sock_hash_seq_ops,
1436         .init_seq_private       = sock_hash_init_seq_private,
1437         .seq_priv_size          = sizeof(struct sock_hash_seq_info),
1438 };
1439
1440 static int sock_hash_map_btf_id;
1441 const struct bpf_map_ops sock_hash_ops = {
1442         .map_meta_equal         = bpf_map_meta_equal,
1443         .map_alloc              = sock_hash_alloc,
1444         .map_free               = sock_hash_free,
1445         .map_get_next_key       = sock_hash_get_next_key,
1446         .map_update_elem        = sock_map_update_elem,
1447         .map_delete_elem        = sock_hash_delete_elem,
1448         .map_lookup_elem        = sock_hash_lookup,
1449         .map_lookup_elem_sys_only = sock_hash_lookup_sys,
1450         .map_release_uref       = sock_hash_release_progs,
1451         .map_check_btf          = map_check_no_btf,
1452         .map_btf_name           = "bpf_shtab",
1453         .map_btf_id             = &sock_hash_map_btf_id,
1454         .iter_seq_info          = &sock_hash_iter_seq_info,
1455 };
1456
1457 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1458 {
1459         switch (map->map_type) {
1460         case BPF_MAP_TYPE_SOCKMAP:
1461                 return &container_of(map, struct bpf_stab, map)->progs;
1462         case BPF_MAP_TYPE_SOCKHASH:
1463                 return &container_of(map, struct bpf_shtab, map)->progs;
1464         default:
1465                 break;
1466         }
1467
1468         return NULL;
1469 }
1470
1471 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1472                          struct bpf_prog *old, u32 which)
1473 {
1474         struct sk_psock_progs *progs = sock_map_progs(map);
1475         struct bpf_prog **pprog;
1476
1477         if (!progs)
1478                 return -EOPNOTSUPP;
1479
1480         switch (which) {
1481         case BPF_SK_MSG_VERDICT:
1482                 pprog = &progs->msg_parser;
1483                 break;
1484         case BPF_SK_SKB_STREAM_PARSER:
1485                 pprog = &progs->skb_parser;
1486                 break;
1487         case BPF_SK_SKB_STREAM_VERDICT:
1488                 pprog = &progs->skb_verdict;
1489                 break;
1490         default:
1491                 return -EOPNOTSUPP;
1492         }
1493
1494         if (old)
1495                 return psock_replace_prog(pprog, prog, old);
1496
1497         psock_set_prog(pprog, prog);
1498         return 0;
1499 }
1500
1501 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1502 {
1503         switch (link->map->map_type) {
1504         case BPF_MAP_TYPE_SOCKMAP:
1505                 return sock_map_delete_from_link(link->map, sk,
1506                                                  link->link_raw);
1507         case BPF_MAP_TYPE_SOCKHASH:
1508                 return sock_hash_delete_from_link(link->map, sk,
1509                                                   link->link_raw);
1510         default:
1511                 break;
1512         }
1513 }
1514
1515 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1516 {
1517         struct sk_psock_link *link;
1518
1519         while ((link = sk_psock_link_pop(psock))) {
1520                 sock_map_unlink(sk, link);
1521                 sk_psock_free_link(link);
1522         }
1523 }
1524
1525 void sock_map_unhash(struct sock *sk)
1526 {
1527         void (*saved_unhash)(struct sock *sk);
1528         struct sk_psock *psock;
1529
1530         rcu_read_lock();
1531         psock = sk_psock(sk);
1532         if (unlikely(!psock)) {
1533                 rcu_read_unlock();
1534                 if (sk->sk_prot->unhash)
1535                         sk->sk_prot->unhash(sk);
1536                 return;
1537         }
1538
1539         saved_unhash = psock->saved_unhash;
1540         sock_map_remove_links(sk, psock);
1541         rcu_read_unlock();
1542         saved_unhash(sk);
1543 }
1544
1545 void sock_map_close(struct sock *sk, long timeout)
1546 {
1547         void (*saved_close)(struct sock *sk, long timeout);
1548         struct sk_psock *psock;
1549
1550         lock_sock(sk);
1551         rcu_read_lock();
1552         psock = sk_psock(sk);
1553         if (unlikely(!psock)) {
1554                 rcu_read_unlock();
1555                 release_sock(sk);
1556                 return sk->sk_prot->close(sk, timeout);
1557         }
1558
1559         saved_close = psock->saved_close;
1560         sock_map_remove_links(sk, psock);
1561         rcu_read_unlock();
1562         release_sock(sk);
1563         saved_close(sk, timeout);
1564 }
1565
1566 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1567                                        union bpf_iter_link_info *linfo,
1568                                        struct bpf_iter_aux_info *aux)
1569 {
1570         struct bpf_map *map;
1571         int err = -EINVAL;
1572
1573         if (!linfo->map.map_fd)
1574                 return -EBADF;
1575
1576         map = bpf_map_get_with_uref(linfo->map.map_fd);
1577         if (IS_ERR(map))
1578                 return PTR_ERR(map);
1579
1580         if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1581             map->map_type != BPF_MAP_TYPE_SOCKHASH)
1582                 goto put_map;
1583
1584         if (prog->aux->max_rdonly_access > map->key_size) {
1585                 err = -EACCES;
1586                 goto put_map;
1587         }
1588
1589         aux->map = map;
1590         return 0;
1591
1592 put_map:
1593         bpf_map_put_with_uref(map);
1594         return err;
1595 }
1596
1597 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1598 {
1599         bpf_map_put_with_uref(aux->map);
1600 }
1601
1602 static struct bpf_iter_reg sock_map_iter_reg = {
1603         .target                 = "sockmap",
1604         .attach_target          = sock_map_iter_attach_target,
1605         .detach_target          = sock_map_iter_detach_target,
1606         .show_fdinfo            = bpf_iter_map_show_fdinfo,
1607         .fill_link_info         = bpf_iter_map_fill_link_info,
1608         .ctx_arg_info_size      = 2,
1609         .ctx_arg_info           = {
1610                 { offsetof(struct bpf_iter__sockmap, key),
1611                   PTR_TO_RDONLY_BUF_OR_NULL },
1612                 { offsetof(struct bpf_iter__sockmap, sk),
1613                   PTR_TO_BTF_ID_OR_NULL },
1614         },
1615 };
1616
1617 static int __init bpf_sockmap_iter_init(void)
1618 {
1619         sock_map_iter_reg.ctx_arg_info[1].btf_id =
1620                 btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1621         return bpf_iter_reg_target(&sock_map_iter_reg);
1622 }
1623 late_initcall(bpf_sockmap_iter_init);