net/ipv4/inet_connection_sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Support for INET connection oriented protocols.
   7  *
   8  * Authors:     See the TCP sources
   9  *
  10  *              This program is free software; you can redistribute it and/or
  11  *              modify it under the terms of the GNU General Public License
  12  *              as published by the Free Software Foundation; either version
  13  *              2 of the License, or(at your option) any later version.
  14  */
  15
  16 #include <linux/module.h>
  17 #include <linux/jhash.h>
  18
  19 #include <net/inet_connection_sock.h>
  20 #include <net/inet_hashtables.h>
  21 #include <net/inet_timewait_sock.h>
  22 #include <net/ip.h>
  23 #include <net/route.h>
  24 #include <net/tcp_states.h>
  25 #include <net/xfrm.h>
  26 #include <net/tcp.h>
  27 #include <net/sock_reuseport.h>
  28 #include <net/addrconf.h>
  29
  30 #ifdef INET_CSK_DEBUG
  31 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  32 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  33 #endif
  34
  35 #if IS_ENABLED(CONFIG_IPV6)
  36 /* match_wildcard == true:  IPV6_ADDR_ANY equals to any IPv6 addresses if IPv6
  37  *                          only, and any IPv4 addresses if not IPv6 only
  38  * match_wildcard == false: addresses must be exactly the same, i.e.
  39  *                          IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
  40  *                          and 0.0.0.0 equals to 0.0.0.0 only
  41  */
  42 static int ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
  43                                 const struct in6_addr *sk2_rcv_saddr6,
  44                                 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  45                                 bool sk1_ipv6only, bool sk2_ipv6only,
  46                                 bool match_wildcard)
  47 {
  48         int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
  49         int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
  50
  51         /* if both are mapped, treat as IPv4 */
  52         if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
  53                 if (!sk2_ipv6only) {
  54                         if (sk1_rcv_saddr == sk2_rcv_saddr)
  55                                 return 1;
  56                         if (!sk1_rcv_saddr || !sk2_rcv_saddr)
  57                                 return match_wildcard;
  58                 }
  59                 return 0;
  60         }
  61
  62         if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
  63                 return 1;
  64
  65         if (addr_type2 == IPV6_ADDR_ANY && match_wildcard &&
  66             !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
  67                 return 1;
  68
  69         if (addr_type == IPV6_ADDR_ANY && match_wildcard &&
  70             !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
  71                 return 1;
  72
  73         if (sk2_rcv_saddr6 &&
  74             ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
  75                 return 1;
  76
  77         return 0;
  78 }
  79 #endif
  80
  81 /* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
  82  * match_wildcard == false: addresses must be exactly the same, i.e.
  83  *                          0.0.0.0 only equals to 0.0.0.0
  84  */
  85 static int ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
  86                                 bool sk2_ipv6only, bool match_wildcard)
  87 {
  88         if (!sk2_ipv6only) {
  89                 if (sk1_rcv_saddr == sk2_rcv_saddr)
  90                         return 1;
  91                 if (!sk1_rcv_saddr || !sk2_rcv_saddr)
  92                         return match_wildcard;
  93         }
  94         return 0;
  95 }
  96
  97 int inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
  98                          bool match_wildcard)
  99 {
 100 #if IS_ENABLED(CONFIG_IPV6)
 101         if (sk->sk_family == AF_INET6)
 102                 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
 103                                             inet6_rcv_saddr(sk2),
 104                                             sk->sk_rcv_saddr,
 105                                             sk2->sk_rcv_saddr,
 106                                             ipv6_only_sock(sk),
 107                                             ipv6_only_sock(sk2),
 108                                             match_wildcard);
 109 #endif
 110         return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
 111                                     ipv6_only_sock(sk2), match_wildcard);
 112 }
 113 EXPORT_SYMBOL(inet_rcv_saddr_equal);
 114
 115 void inet_get_local_port_range(struct net *net, int *low, int *high)
 116 {
 117         unsigned int seq;
 118
 119         do {
 120                 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
 121
 122                 *low = net->ipv4.ip_local_ports.range[0];
 123                 *high = net->ipv4.ip_local_ports.range[1];
 124         } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
 125 }
 126 EXPORT_SYMBOL(inet_get_local_port_range);
 127
 128 static int inet_csk_bind_conflict(const struct sock *sk,
 129                                   const struct inet_bind_bucket *tb,
 130                                   bool relax, bool reuseport_ok)
 131 {
 132         struct sock *sk2;
 133         bool reuse = sk->sk_reuse;
 134         bool reuseport = !!sk->sk_reuseport && reuseport_ok;
 135         kuid_t uid = sock_i_uid((struct sock *)sk);
 136
 137         /*
 138          * Unlike other sk lookup places we do not check
 139          * for sk_net here, since _all_ the socks listed
 140          * in tb->owners list belong to the same net - the
 141          * one this bucket belongs to.
 142          */
 143
 144         sk_for_each_bound(sk2, &tb->owners) {
 145                 if (sk != sk2 &&
 146                     (!sk->sk_bound_dev_if ||
 147                      !sk2->sk_bound_dev_if ||
 148                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
 149                         if ((!reuse || !sk2->sk_reuse ||
 150                             sk2->sk_state == TCP_LISTEN) &&
 151                             (!reuseport || !sk2->sk_reuseport ||
 152                              rcu_access_pointer(sk->sk_reuseport_cb) ||
 153                              (sk2->sk_state != TCP_TIME_WAIT &&
 154                              !uid_eq(uid, sock_i_uid(sk2))))) {
 155                                 if (inet_rcv_saddr_equal(sk, sk2, true))
 156                                         break;
 157                         }
 158                         if (!relax && reuse && sk2->sk_reuse &&
 159                             sk2->sk_state != TCP_LISTEN) {
 160                                 if (inet_rcv_saddr_equal(sk, sk2, true))
 161                                         break;
 162                         }
 163                 }
 164         }
 165         return sk2 != NULL;
 166 }
 167
 168 /*
 169  * Find an open port number for the socket.  Returns with the
 170  * inet_bind_hashbucket lock held.
 171  */
 172 static struct inet_bind_hashbucket *
 173 inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret)
 174 {
 175         struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
 176         int port = 0;
 177         struct inet_bind_hashbucket *head;
 178         struct net *net = sock_net(sk);
 179         int i, low, high, attempt_half;
 180         struct inet_bind_bucket *tb;
 181         u32 remaining, offset;
 182
 183         attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 184 other_half_scan:
 185         inet_get_local_port_range(net, &low, &high);
 186         high++; /* [32768, 60999] -> [32768, 61000[ */
 187         if (high - low < 4)
 188                 attempt_half = 0;
 189         if (attempt_half) {
 190                 int half = low + (((high - low) >> 2) << 1);
 191
 192                 if (attempt_half == 1)
 193                         high = half;
 194                 else
 195                         low = half;
 196         }
 197         remaining = high - low;
 198         if (likely(remaining > 1))
 199                 remaining &= ~1U;
 200
 201         offset = prandom_u32() % remaining;
 202         /* __inet_hash_connect() favors ports having @low parity
 203          * We do the opposite to not pollute connect() users.
 204          */
 205         offset |= 1U;
 206
 207 other_parity_scan:
 208         port = low + offset;
 209         for (i = 0; i < remaining; i += 2, port += 2) {
 210                 if (unlikely(port >= high))
 211                         port -= remaining;
 212                 if (inet_is_local_reserved_port(net, port))
 213                         continue;
 214                 head = &hinfo->bhash[inet_bhashfn(net, port,
 215                                                   hinfo->bhash_size)];
 216                 spin_lock_bh(&head->lock);
 217                 inet_bind_bucket_for_each(tb, &head->chain)
 218                         if (net_eq(ib_net(tb), net) && tb->port == port) {
 219                                 if (!inet_csk_bind_conflict(sk, tb, false, false))
 220                                         goto success;
 221                                 goto next_port;
 222                         }
 223                 tb = NULL;
 224                 goto success;
 225 next_port:
 226                 spin_unlock_bh(&head->lock);
 227                 cond_resched();
 228         }
 229
 230         offset--;
 231         if (!(offset & 1))
 232                 goto other_parity_scan;
 233
 234         if (attempt_half == 1) {
 235                 /* OK we now try the upper half of the range */
 236                 attempt_half = 2;
 237                 goto other_half_scan;
 238         }
 239         return NULL;
 240 success:
 241         *port_ret = port;
 242         *tb_ret = tb;
 243         return head;
 244 }
 245
 246 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
 247                                      struct sock *sk)
 248 {
 249         kuid_t uid = sock_i_uid(sk);
 250
 251         if (tb->fastreuseport <= 0)
 252                 return 0;
 253         if (!sk->sk_reuseport)
 254                 return 0;
 255         if (rcu_access_pointer(sk->sk_reuseport_cb))
 256                 return 0;
 257         if (!uid_eq(tb->fastuid, uid))
 258                 return 0;
 259         /* We only need to check the rcv_saddr if this tb was once marked
 260          * without fastreuseport and then was reset, as we can only know that
 261          * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
 262          * owners list.
 263          */
 264         if (tb->fastreuseport == FASTREUSEPORT_ANY)
 265                 return 1;
 266 #if IS_ENABLED(CONFIG_IPV6)
 267         if (tb->fast_sk_family == AF_INET6)
 268                 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
 269                                             &sk->sk_v6_rcv_saddr,
 270                                             tb->fast_rcv_saddr,
 271                                             sk->sk_rcv_saddr,
 272                                             tb->fast_ipv6_only,
 273                                             ipv6_only_sock(sk), true);
 274 #endif
 275         return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
 276                                     ipv6_only_sock(sk), true);
 277 }
 278
 279 /* Obtain a reference to a local port for the given sock,
 280  * if snum is zero it means select any available local port.
 281  * We try to allocate an odd port (and leave even ports for connect())
 282  */
 283 int inet_csk_get_port(struct sock *sk, unsigned short snum)
 284 {
 285         bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
 286         struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
 287         int ret = 1, port = snum;
 288         struct inet_bind_hashbucket *head;
 289         struct net *net = sock_net(sk);
 290         struct inet_bind_bucket *tb = NULL;
 291         kuid_t uid = sock_i_uid(sk);
 292
 293         if (!port) {
 294                 head = inet_csk_find_open_port(sk, &tb, &port);
 295                 if (!head)
 296                         return ret;
 297                 if (!tb)
 298                         goto tb_not_found;
 299                 goto success;
 300         }
 301         head = &hinfo->bhash[inet_bhashfn(net, port,
 302                                           hinfo->bhash_size)];
 303         spin_lock_bh(&head->lock);
 304         inet_bind_bucket_for_each(tb, &head->chain)
 305                 if (net_eq(ib_net(tb), net) && tb->port == port)
 306                         goto tb_found;
 307 tb_not_found:
 308         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 309                                      net, head, port);
 310         if (!tb)
 311                 goto fail_unlock;
 312 tb_found:
 313         if (!hlist_empty(&tb->owners)) {
 314                 if (sk->sk_reuse == SK_FORCE_REUSE)
 315                         goto success;
 316
 317                 if ((tb->fastreuse > 0 && reuse) ||
 318                     sk_reuseport_match(tb, sk))
 319                         goto success;
 320                 if (inet_csk_bind_conflict(sk, tb, true, true))
 321                         goto fail_unlock;
 322         }
 323 success:
 324         if (!hlist_empty(&tb->owners)) {
 325                 tb->fastreuse = reuse;
 326                 if (sk->sk_reuseport) {
 327                         tb->fastreuseport = FASTREUSEPORT_ANY;
 328                         tb->fastuid = uid;
 329                         tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 330                         tb->fast_ipv6_only = ipv6_only_sock(sk);
 331 #if IS_ENABLED(CONFIG_IPV6)
 332                         tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 333 #endif
 334                 } else {
 335                         tb->fastreuseport = 0;
 336                 }
 337         } else {
 338                 if (!reuse)
 339                         tb->fastreuse = 0;
 340                 if (sk->sk_reuseport) {
 341                         /* We didn't match or we don't have fastreuseport set on
 342                          * the tb, but we have sk_reuseport set on this socket
 343                          * and we know that there are no bind conflicts with
 344                          * this socket in this tb, so reset our tb's reuseport
 345                          * settings so that any subsequent sockets that match
 346                          * our current socket will be put on the fast path.
 347                          *
 348                          * If we reset we need to set FASTREUSEPORT_STRICT so we
 349                          * do extra checking for all subsequent sk_reuseport
 350                          * socks.
 351                          */
 352                         if (!sk_reuseport_match(tb, sk)) {
 353                                 tb->fastreuseport = FASTREUSEPORT_STRICT;
 354                                 tb->fastuid = uid;
 355                                 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
 356                                 tb->fast_ipv6_only = ipv6_only_sock(sk);
 357 #if IS_ENABLED(CONFIG_IPV6)
 358                                 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
 359 #endif
 360                         }
 361                 } else {
 362                         tb->fastreuseport = 0;
 363                 }
 364         }
 365         if (!inet_csk(sk)->icsk_bind_hash)
 366                 inet_bind_hash(sk, tb, port);
 367         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 368         ret = 0;
 369
 370 fail_unlock:
 371         spin_unlock_bh(&head->lock);
 372         return ret;
 373 }
 374 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 375
 376 /*
 377  * Wait for an incoming connection, avoid race conditions. This must be called
 378  * with the socket locked.
 379  */
 380 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 381 {
 382         struct inet_connection_sock *icsk = inet_csk(sk);
 383         DEFINE_WAIT(wait);
 384         int err;
 385
 386         /*
 387          * True wake-one mechanism for incoming connections: only
 388          * one process gets woken up, not the 'whole herd'.
 389          * Since we do not 'race & poll' for established sockets
 390          * anymore, the common case will execute the loop only once.
 391          *
 392          * Subtle issue: "add_wait_queue_exclusive()" will be added
 393          * after any current non-exclusive waiters, and we know that
 394          * it will always _stay_ after any new non-exclusive waiters
 395          * because all non-exclusive waiters are added at the
 396          * beginning of the wait-queue. As such, it's ok to "drop"
 397          * our exclusiveness temporarily when we get woken up without
 398          * having to remove and re-insert us on the wait queue.
 399          */
 400         for (;;) {
 401                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 402                                           TASK_INTERRUPTIBLE);
 403                 release_sock(sk);
 404                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 405                         timeo = schedule_timeout(timeo);
 406                 sched_annotate_sleep();
 407                 lock_sock(sk);
 408                 err = 0;
 409                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 410                         break;
 411                 err = -EINVAL;
 412                 if (sk->sk_state != TCP_LISTEN)
 413                         break;
 414                 err = sock_intr_errno(timeo);
 415                 if (signal_pending(current))
 416                         break;
 417                 err = -EAGAIN;
 418                 if (!timeo)
 419                         break;
 420         }
 421         finish_wait(sk_sleep(sk), &wait);
 422         return err;
 423 }
 424
 425 /*
 426  * This will accept the next outstanding connection.
 427  */
 428 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
 429 {
 430         struct inet_connection_sock *icsk = inet_csk(sk);
 431         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 432         struct request_sock *req;
 433         struct sock *newsk;
 434         int error;
 435
 436         lock_sock(sk);
 437
 438         /* We need to make sure that this socket is listening,
 439          * and that it has something pending.
 440          */
 441         error = -EINVAL;
 442         if (sk->sk_state != TCP_LISTEN)
 443                 goto out_err;
 444
 445         /* Find already established connection */
 446         if (reqsk_queue_empty(queue)) {
 447                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 448
 449                 /* If this is a non blocking socket don't sleep */
 450                 error = -EAGAIN;
 451                 if (!timeo)
 452                         goto out_err;
 453
 454                 error = inet_csk_wait_for_connect(sk, timeo);
 455                 if (error)
 456                         goto out_err;
 457         }
 458         req = reqsk_queue_remove(queue, sk);
 459         newsk = req->sk;
 460
 461         if (sk->sk_protocol == IPPROTO_TCP &&
 462             tcp_rsk(req)->tfo_listener) {
 463                 spin_lock_bh(&queue->fastopenq.lock);
 464                 if (tcp_rsk(req)->tfo_listener) {
 465                         /* We are still waiting for the final ACK from 3WHS
 466                          * so can't free req now. Instead, we set req->sk to
 467                          * NULL to signify that the child socket is taken
 468                          * so reqsk_fastopen_remove() will free the req
 469                          * when 3WHS finishes (or is aborted).
 470                          */
 471                         req->sk = NULL;
 472                         req = NULL;
 473                 }
 474                 spin_unlock_bh(&queue->fastopenq.lock);
 475         }
 476 out:
 477         release_sock(sk);
 478         if (req)
 479                 reqsk_put(req);
 480         return newsk;
 481 out_err:
 482         newsk = NULL;
 483         req = NULL;
 484         *err = error;
 485         goto out;
 486 }
 487 EXPORT_SYMBOL(inet_csk_accept);
 488
 489 /*
 490  * Using different timers for retransmit, delayed acks and probes
 491  * We may wish use just one timer maintaining a list of expire jiffies
 492  * to optimize.
 493  */
 494 void inet_csk_init_xmit_timers(struct sock *sk,
 495                                void (*retransmit_handler)(unsigned long),
 496                                void (*delack_handler)(unsigned long),
 497                                void (*keepalive_handler)(unsigned long))
 498 {
 499         struct inet_connection_sock *icsk = inet_csk(sk);
 500
 501         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 502                         (unsigned long)sk);
 503         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 504                         (unsigned long)sk);
 505         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 506         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 507 }
 508 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 509
 510 void inet_csk_clear_xmit_timers(struct sock *sk)
 511 {
 512         struct inet_connection_sock *icsk = inet_csk(sk);
 513
 514         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 515
 516         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 517         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 518         sk_stop_timer(sk, &sk->sk_timer);
 519 }
 520 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 521
 522 void inet_csk_delete_keepalive_timer(struct sock *sk)
 523 {
 524         sk_stop_timer(sk, &sk->sk_timer);
 525 }
 526 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 527
 528 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 529 {
 530         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 531 }
 532 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 533
 534 struct dst_entry *inet_csk_route_req(const struct sock *sk,
 535                                      struct flowi4 *fl4,
 536                                      const struct request_sock *req)
 537 {
 538         const struct inet_request_sock *ireq = inet_rsk(req);
 539         struct net *net = read_pnet(&ireq->ireq_net);
 540         struct ip_options_rcu *opt = ireq->opt;
 541         struct rtable *rt;
 542
 543         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 544                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 545                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 546                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 547                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 548                            htons(ireq->ir_num), sk->sk_uid);
 549         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 550         rt = ip_route_output_flow(net, fl4, sk);
 551         if (IS_ERR(rt))
 552                 goto no_route;
 553         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 554                 goto route_err;
 555         return &rt->dst;
 556
 557 route_err:
 558         ip_rt_put(rt);
 559 no_route:
 560         __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 561         return NULL;
 562 }
 563 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 564
 565 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 566                                             struct sock *newsk,
 567                                             const struct request_sock *req)
 568 {
 569         const struct inet_request_sock *ireq = inet_rsk(req);
 570         struct net *net = read_pnet(&ireq->ireq_net);
 571         struct inet_sock *newinet = inet_sk(newsk);
 572         struct ip_options_rcu *opt;
 573         struct flowi4 *fl4;
 574         struct rtable *rt;
 575
 576         fl4 = &newinet->cork.fl.u.ip4;
 577
 578         rcu_read_lock();
 579         opt = rcu_dereference(newinet->inet_opt);
 580         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 581                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 582                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 583                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 584                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 585                            htons(ireq->ir_num), sk->sk_uid);
 586         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 587         rt = ip_route_output_flow(net, fl4, sk);
 588         if (IS_ERR(rt))
 589                 goto no_route;
 590         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 591                 goto route_err;
 592         rcu_read_unlock();
 593         return &rt->dst;
 594
 595 route_err:
 596         ip_rt_put(rt);
 597 no_route:
 598         rcu_read_unlock();
 599         __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 600         return NULL;
 601 }
 602 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 603
 604 #if IS_ENABLED(CONFIG_IPV6)
 605 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 606 #else
 607 #define AF_INET_FAMILY(fam) true
 608 #endif
 609
 610 /* Decide when to expire the request and when to resend SYN-ACK */
 611 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 612                                   const int max_retries,
 613                                   const u8 rskq_defer_accept,
 614                                   int *expire, int *resend)
 615 {
 616         if (!rskq_defer_accept) {
 617                 *expire = req->num_timeout >= thresh;
 618                 *resend = 1;
 619                 return;
 620         }
 621         *expire = req->num_timeout >= thresh &&
 622                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 623         /*
 624          * Do not resend while waiting for data after ACK,
 625          * start to resend on end of deferring period to give
 626          * last chance for data or ACK to create established socket.
 627          */
 628         *resend = !inet_rsk(req)->acked ||
 629                   req->num_timeout >= rskq_defer_accept - 1;
 630 }
 631
 632 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 633 {
 634         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 635
 636         if (!err)
 637                 req->num_retrans++;
 638         return err;
 639 }
 640 EXPORT_SYMBOL(inet_rtx_syn_ack);
 641
 642 /* return true if req was found in the ehash table */
 643 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 644                                struct request_sock *req)
 645 {
 646         struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
 647         bool found = false;
 648
 649         if (sk_hashed(req_to_sk(req))) {
 650                 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 651
 652                 spin_lock(lock);
 653                 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
 654                 spin_unlock(lock);
 655         }
 656         if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 657                 reqsk_put(req);
 658         return found;
 659 }
 660
 661 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 662 {
 663         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 664                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 665                 reqsk_put(req);
 666         }
 667 }
 668 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 669
 670 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 671 {
 672         inet_csk_reqsk_queue_drop(sk, req);
 673         reqsk_put(req);
 674 }
 675 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 676
 677 static void reqsk_timer_handler(unsigned long data)
 678 {
 679         struct request_sock *req = (struct request_sock *)data;
 680         struct sock *sk_listener = req->rsk_listener;
 681         struct net *net = sock_net(sk_listener);
 682         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 683         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 684         int qlen, expire = 0, resend = 0;
 685         int max_retries, thresh;
 686         u8 defer_accept;
 687
 688         if (sk_state_load(sk_listener) != TCP_LISTEN)
 689                 goto drop;
 690
 691         max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
 692         thresh = max_retries;
 693         /* Normally all the openreqs are young and become mature
 694          * (i.e. converted to established socket) for first timeout.
 695          * If synack was not acknowledged for 1 second, it means
 696          * one of the following things: synack was lost, ack was lost,
 697          * rtt is high or nobody planned to ack (i.e. synflood).
 698          * When server is a bit loaded, queue is populated with old
 699          * open requests, reducing effective size of queue.
 700          * When server is well loaded, queue size reduces to zero
 701          * after several minutes of work. It is not synflood,
 702          * it is normal operation. The solution is pruning
 703          * too old entries overriding normal timeout, when
 704          * situation becomes dangerous.
 705          *
 706          * Essentially, we reserve half of room for young
 707          * embrions; and abort old ones without pity, if old
 708          * ones are about to clog our table.
 709          */
 710         qlen = reqsk_queue_len(queue);
 711         if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 712                 int young = reqsk_queue_len_young(queue) << 1;
 713
 714                 while (thresh > 2) {
 715                         if (qlen < young)
 716                                 break;
 717                         thresh--;
 718                         young <<= 1;
 719                 }
 720         }
 721         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 722         if (defer_accept)
 723                 max_retries = defer_accept;
 724         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 725                        &expire, &resend);
 726         req->rsk_ops->syn_ack_timeout(req);
 727         if (!expire &&
 728             (!resend ||
 729              !inet_rtx_syn_ack(sk_listener, req) ||
 730              inet_rsk(req)->acked)) {
 731                 unsigned long timeo;
 732
 733                 if (req->num_timeout++ == 0)
 734                         atomic_dec(&queue->young);
 735                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 736                 mod_timer(&req->rsk_timer, jiffies + timeo);
 737                 return;
 738         }
 739 drop:
 740         inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
 741 }
 742
 743 static void reqsk_queue_hash_req(struct request_sock *req,
 744                                  unsigned long timeout)
 745 {
 746         req->num_retrans = 0;
 747         req->num_timeout = 0;
 748         req->sk = NULL;
 749
 750         setup_pinned_timer(&req->rsk_timer, reqsk_timer_handler,
 751                             (unsigned long)req);
 752         mod_timer(&req->rsk_timer, jiffies + timeout);
 753
 754         inet_ehash_insert(req_to_sk(req), NULL);
 755         /* before letting lookups find us, make sure all req fields
 756          * are committed to memory and refcnt initialized.
 757          */
 758         smp_wmb();
 759         atomic_set(&req->rsk_refcnt, 2 + 1);
 760 }
 761
 762 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 763                                    unsigned long timeout)
 764 {
 765         reqsk_queue_hash_req(req, timeout);
 766         inet_csk_reqsk_queue_added(sk);
 767 }
 768 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 769
 770 /**
 771  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 772  *      @sk: the socket to clone
 773  *      @req: request_sock
 774  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 775  *
 776  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 777  */
 778 struct sock *inet_csk_clone_lock(const struct sock *sk,
 779                                  const struct request_sock *req,
 780                                  const gfp_t priority)
 781 {
 782         struct sock *newsk = sk_clone_lock(sk, priority);
 783
 784         if (newsk) {
 785                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 786
 787                 newsk->sk_state = TCP_SYN_RECV;
 788                 newicsk->icsk_bind_hash = NULL;
 789
 790                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 791                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 792                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 793
 794                 /* listeners have SOCK_RCU_FREE, not the children */
 795                 sock_reset_flag(newsk, SOCK_RCU_FREE);
 796
 797                 inet_sk(newsk)->mc_list = NULL;
 798
 799                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 800                 atomic64_set(&newsk->sk_cookie,
 801                              atomic64_read(&inet_rsk(req)->ir_cookie));
 802
 803                 newicsk->icsk_retransmits = 0;
 804                 newicsk->icsk_backoff     = 0;
 805                 newicsk->icsk_probes_out  = 0;
 806
 807                 /* Deinitialize accept_queue to trap illegal accesses. */
 808                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 809
 810                 security_inet_csk_clone(newsk, req);
 811         }
 812         return newsk;
 813 }
 814 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 815
 816 /*
 817  * At this point, there should be no process reference to this
 818  * socket, and thus no user references at all.  Therefore we
 819  * can assume the socket waitqueue is inactive and nobody will
 820  * try to jump onto it.
 821  */
 822 void inet_csk_destroy_sock(struct sock *sk)
 823 {
 824         WARN_ON(sk->sk_state != TCP_CLOSE);
 825         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 826
 827         /* It cannot be in hash table! */
 828         WARN_ON(!sk_unhashed(sk));
 829
 830         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 831         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 832
 833         sk->sk_prot->destroy(sk);
 834
 835         sk_stream_kill_queues(sk);
 836
 837         xfrm_sk_free_policy(sk);
 838
 839         sk_refcnt_debug_release(sk);
 840
 841         percpu_counter_dec(sk->sk_prot->orphan_count);
 842
 843         sock_put(sk);
 844 }
 845 EXPORT_SYMBOL(inet_csk_destroy_sock);
 846
 847 /* This function allows to force a closure of a socket after the call to
 848  * tcp/dccp_create_openreq_child().
 849  */
 850 void inet_csk_prepare_forced_close(struct sock *sk)
 851         __releases(&sk->sk_lock.slock)
 852 {
 853         /* sk_clone_lock locked the socket and set refcnt to 2 */
 854         bh_unlock_sock(sk);
 855         sock_put(sk);
 856
 857         /* The below has to be done to allow calling inet_csk_destroy_sock */
 858         sock_set_flag(sk, SOCK_DEAD);
 859         percpu_counter_inc(sk->sk_prot->orphan_count);
 860         inet_sk(sk)->inet_num = 0;
 861 }
 862 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 863
 864 int inet_csk_listen_start(struct sock *sk, int backlog)
 865 {
 866         struct inet_connection_sock *icsk = inet_csk(sk);
 867         struct inet_sock *inet = inet_sk(sk);
 868         int err = -EADDRINUSE;
 869
 870         reqsk_queue_alloc(&icsk->icsk_accept_queue);
 871
 872         sk->sk_max_ack_backlog = backlog;
 873         sk->sk_ack_backlog = 0;
 874         inet_csk_delack_init(sk);
 875
 876         /* There is race window here: we announce ourselves listening,
 877          * but this transition is still not validated by get_port().
 878          * It is OK, because this socket enters to hash table only
 879          * after validation is complete.
 880          */
 881         sk_state_store(sk, TCP_LISTEN);
 882         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 883                 inet->inet_sport = htons(inet->inet_num);
 884
 885                 sk_dst_reset(sk);
 886                 err = sk->sk_prot->hash(sk);
 887
 888                 if (likely(!err))
 889                         return 0;
 890         }
 891
 892         sk->sk_state = TCP_CLOSE;
 893         return err;
 894 }
 895 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 896
 897 static void inet_child_forget(struct sock *sk, struct request_sock *req,
 898                               struct sock *child)
 899 {
 900         sk->sk_prot->disconnect(child, O_NONBLOCK);
 901
 902         sock_orphan(child);
 903
 904         percpu_counter_inc(sk->sk_prot->orphan_count);
 905
 906         if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 907                 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 908                 BUG_ON(sk != req->rsk_listener);
 909
 910                 /* Paranoid, to prevent race condition if
 911                  * an inbound pkt destined for child is
 912                  * blocked by sock lock in tcp_v4_rcv().
 913                  * Also to satisfy an assertion in
 914                  * tcp_v4_destroy_sock().
 915                  */
 916                 tcp_sk(child)->fastopen_rsk = NULL;
 917         }
 918         inet_csk_destroy_sock(child);
 919         reqsk_put(req);
 920 }
 921
 922 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
 923                                       struct request_sock *req,
 924                                       struct sock *child)
 925 {
 926         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 927
 928         spin_lock(&queue->rskq_lock);
 929         if (unlikely(sk->sk_state != TCP_LISTEN)) {
 930                 inet_child_forget(sk, req, child);
 931                 child = NULL;
 932         } else {
 933                 req->sk = child;
 934                 req->dl_next = NULL;
 935                 if (queue->rskq_accept_head == NULL)
 936                         queue->rskq_accept_head = req;
 937                 else
 938                         queue->rskq_accept_tail->dl_next = req;
 939                 queue->rskq_accept_tail = req;
 940                 sk_acceptq_added(sk);
 941         }
 942         spin_unlock(&queue->rskq_lock);
 943         return child;
 944 }
 945 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 946
 947 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
 948                                          struct request_sock *req, bool own_req)
 949 {
 950         if (own_req) {
 951                 inet_csk_reqsk_queue_drop(sk, req);
 952                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 953                 if (inet_csk_reqsk_queue_add(sk, req, child))
 954                         return child;
 955         }
 956         /* Too bad, another child took ownership of the request, undo. */
 957         bh_unlock_sock(child);
 958         sock_put(child);
 959         return NULL;
 960 }
 961 EXPORT_SYMBOL(inet_csk_complete_hashdance);
 962
 963 /*
 964  *      This routine closes sockets which have been at least partially
 965  *      opened, but not yet accepted.
 966  */
 967 void inet_csk_listen_stop(struct sock *sk)
 968 {
 969         struct inet_connection_sock *icsk = inet_csk(sk);
 970         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 971         struct request_sock *next, *req;
 972
 973         /* Following specs, it would be better either to send FIN
 974          * (and enter FIN-WAIT-1, it is normal close)
 975          * or to send active reset (abort).
 976          * Certainly, it is pretty dangerous while synflood, but it is
 977          * bad justification for our negligence 8)
 978          * To be honest, we are not able to make either
 979          * of the variants now.                 --ANK
 980          */
 981         while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
 982                 struct sock *child = req->sk;
 983
 984                 local_bh_disable();
 985                 bh_lock_sock(child);
 986                 WARN_ON(sock_owned_by_user(child));
 987                 sock_hold(child);
 988
 989                 inet_child_forget(sk, req, child);
 990                 bh_unlock_sock(child);
 991                 local_bh_enable();
 992                 sock_put(child);
 993
 994                 cond_resched();
 995         }
 996         if (queue->fastopenq.rskq_rst_head) {
 997                 /* Free all the reqs queued in rskq_rst_head. */
 998                 spin_lock_bh(&queue->fastopenq.lock);
 999                 req = queue->fastopenq.rskq_rst_head;
1000                 queue->fastopenq.rskq_rst_head = NULL;
1001                 spin_unlock_bh(&queue->fastopenq.lock);
1002                 while (req != NULL) {
1003                         next = req->dl_next;
1004                         reqsk_put(req);
1005                         req = next;
1006                 }
1007         }
1008         WARN_ON_ONCE(sk->sk_ack_backlog);
1009 }
1010 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1011
1012 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1013 {
1014         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1015         const struct inet_sock *inet = inet_sk(sk);
1016
1017         sin->sin_family         = AF_INET;
1018         sin->sin_addr.s_addr    = inet->inet_daddr;
1019         sin->sin_port           = inet->inet_dport;
1020 }
1021 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1022
1023 #ifdef CONFIG_COMPAT
1024 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
1025                                char __user *optval, int __user *optlen)
1026 {
1027         const struct inet_connection_sock *icsk = inet_csk(sk);
1028
1029         if (icsk->icsk_af_ops->compat_getsockopt)
1030                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
1031                                                             optval, optlen);
1032         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
1033                                              optval, optlen);
1034 }
1035 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
1036
1037 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
1038                                char __user *optval, unsigned int optlen)
1039 {
1040         const struct inet_connection_sock *icsk = inet_csk(sk);
1041
1042         if (icsk->icsk_af_ops->compat_setsockopt)
1043                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
1044                                                             optval, optlen);
1045         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
1046                                              optval, optlen);
1047 }
1048 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
1049 #endif
1050
1051 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1052 {
1053         const struct inet_sock *inet = inet_sk(sk);
1054         const struct ip_options_rcu *inet_opt;
1055         __be32 daddr = inet->inet_daddr;
1056         struct flowi4 *fl4;
1057         struct rtable *rt;
1058
1059         rcu_read_lock();
1060         inet_opt = rcu_dereference(inet->inet_opt);
1061         if (inet_opt && inet_opt->opt.srr)
1062                 daddr = inet_opt->opt.faddr;
1063         fl4 = &fl->u.ip4;
1064         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1065                                    inet->inet_saddr, inet->inet_dport,
1066                                    inet->inet_sport, sk->sk_protocol,
1067                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1068         if (IS_ERR(rt))
1069                 rt = NULL;
1070         if (rt)
1071                 sk_setup_caps(sk, &rt->dst);
1072         rcu_read_unlock();
1073
1074         return &rt->dst;
1075 }
1076
1077 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1078 {
1079         struct dst_entry *dst = __sk_dst_check(sk, 0);
1080         struct inet_sock *inet = inet_sk(sk);
1081
1082         if (!dst) {
1083                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1084                 if (!dst)
1085                         goto out;
1086         }
1087         dst->ops->update_pmtu(dst, sk, NULL, mtu);
1088
1089         dst = __sk_dst_check(sk, 0);
1090         if (!dst)
1091                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1092 out:
1093         return dst;
1094 }
1095 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);