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
  29 #ifdef INET_CSK_DEBUG
  30 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  31 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  32 #endif
  33
  34 void inet_get_local_port_range(struct net *net, int *low, int *high)
  35 {
  36         unsigned int seq;
  37
  38         do {
  39                 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
  40
  41                 *low = net->ipv4.ip_local_ports.range[0];
  42                 *high = net->ipv4.ip_local_ports.range[1];
  43         } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
  44 }
  45 EXPORT_SYMBOL(inet_get_local_port_range);
  46
  47 int inet_csk_bind_conflict(const struct sock *sk,
  48                            const struct inet_bind_bucket *tb, bool relax)
  49 {
  50         struct sock *sk2;
  51         int reuse = sk->sk_reuse;
  52         int reuseport = sk->sk_reuseport;
  53         kuid_t uid = sock_i_uid((struct sock *)sk);
  54
  55         /*
  56          * Unlike other sk lookup places we do not check
  57          * for sk_net here, since _all_ the socks listed
  58          * in tb->owners list belong to the same net - the
  59          * one this bucket belongs to.
  60          */
  61
  62         sk_for_each_bound(sk2, &tb->owners) {
  63                 if (sk != sk2 &&
  64                     !inet_v6_ipv6only(sk2) &&
  65                     (!sk->sk_bound_dev_if ||
  66                      !sk2->sk_bound_dev_if ||
  67                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  68                         if ((!reuse || !sk2->sk_reuse ||
  69                             sk2->sk_state == TCP_LISTEN) &&
  70                             (!reuseport || !sk2->sk_reuseport ||
  71                              rcu_access_pointer(sk->sk_reuseport_cb) ||
  72                              (sk2->sk_state != TCP_TIME_WAIT &&
  73                              !uid_eq(uid, sock_i_uid(sk2))))) {
  74
  75                                 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
  76                                     sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
  77                                         break;
  78                         }
  79                         if (!relax && reuse && sk2->sk_reuse &&
  80                             sk2->sk_state != TCP_LISTEN) {
  81
  82                                 if (!sk2->sk_rcv_saddr || !sk->sk_rcv_saddr ||
  83                                     sk2->sk_rcv_saddr == sk->sk_rcv_saddr)
  84                                         break;
  85                         }
  86                 }
  87         }
  88         return sk2 != NULL;
  89 }
  90 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  91
  92 /* Obtain a reference to a local port for the given sock,
  93  * if snum is zero it means select any available local port.
  94  * We try to allocate an odd port (and leave even ports for connect())
  95  */
  96 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  97 {
  98         bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
  99         struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo;
 100         int ret = 1, attempts = 5, port = snum;
 101         int smallest_size = -1, smallest_port;
 102         struct inet_bind_hashbucket *head;
 103         struct net *net = sock_net(sk);
 104         int i, low, high, attempt_half;
 105         struct inet_bind_bucket *tb;
 106         kuid_t uid = sock_i_uid(sk);
 107         u32 remaining, offset;
 108
 109         if (port) {
 110 have_port:
 111                 head = &hinfo->bhash[inet_bhashfn(net, port,
 112                                                   hinfo->bhash_size)];
 113                 spin_lock_bh(&head->lock);
 114                 inet_bind_bucket_for_each(tb, &head->chain)
 115                         if (net_eq(ib_net(tb), net) && tb->port == port)
 116                                 goto tb_found;
 117
 118                 goto tb_not_found;
 119         }
 120 again:
 121         attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
 122 other_half_scan:
 123         inet_get_local_port_range(net, &low, &high);
 124         high++; /* [32768, 60999] -> [32768, 61000[ */
 125         if (high - low < 4)
 126                 attempt_half = 0;
 127         if (attempt_half) {
 128                 int half = low + (((high - low) >> 2) << 1);
 129
 130                 if (attempt_half == 1)
 131                         high = half;
 132                 else
 133                         low = half;
 134         }
 135         remaining = high - low;
 136         if (likely(remaining > 1))
 137                 remaining &= ~1U;
 138
 139         offset = prandom_u32() % remaining;
 140         /* __inet_hash_connect() favors ports having @low parity
 141          * We do the opposite to not pollute connect() users.
 142          */
 143         offset |= 1U;
 144         smallest_size = -1;
 145         smallest_port = low; /* avoid compiler warning */
 146
 147 other_parity_scan:
 148         port = low + offset;
 149         for (i = 0; i < remaining; i += 2, port += 2) {
 150                 if (unlikely(port >= high))
 151                         port -= remaining;
 152                 if (inet_is_local_reserved_port(net, port))
 153                         continue;
 154                 head = &hinfo->bhash[inet_bhashfn(net, port,
 155                                                   hinfo->bhash_size)];
 156                 spin_lock_bh(&head->lock);
 157                 inet_bind_bucket_for_each(tb, &head->chain)
 158                         if (net_eq(ib_net(tb), net) && tb->port == port) {
 159                                 if (((tb->fastreuse > 0 && reuse) ||
 160                                      (tb->fastreuseport > 0 &&
 161                                       sk->sk_reuseport &&
 162                                       !rcu_access_pointer(sk->sk_reuseport_cb) &&
 163                                       uid_eq(tb->fastuid, uid))) &&
 164                                     (tb->num_owners < smallest_size || smallest_size == -1)) {
 165                                         smallest_size = tb->num_owners;
 166                                         smallest_port = port;
 167                                 }
 168                                 if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false))
 169                                         goto tb_found;
 170                                 goto next_port;
 171                         }
 172                 goto tb_not_found;
 173 next_port:
 174                 spin_unlock_bh(&head->lock);
 175                 cond_resched();
 176         }
 177
 178         if (smallest_size != -1) {
 179                 port = smallest_port;
 180                 goto have_port;
 181         }
 182         offset--;
 183         if (!(offset & 1))
 184                 goto other_parity_scan;
 185
 186         if (attempt_half == 1) {
 187                 /* OK we now try the upper half of the range */
 188                 attempt_half = 2;
 189                 goto other_half_scan;
 190         }
 191         return ret;
 192
 193 tb_not_found:
 194         tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
 195                                      net, head, port);
 196         if (!tb)
 197                 goto fail_unlock;
 198 tb_found:
 199         if (!hlist_empty(&tb->owners)) {
 200                 if (sk->sk_reuse == SK_FORCE_REUSE)
 201                         goto success;
 202
 203                 if (((tb->fastreuse > 0 && reuse) ||
 204                      (tb->fastreuseport > 0 &&
 205                       sk->sk_reuseport && uid_eq(tb->fastuid, uid))) &&
 206                     smallest_size == -1)
 207                         goto success;
 208                 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 209                         if ((reuse ||
 210                              (tb->fastreuseport > 0 &&
 211                               sk->sk_reuseport &&
 212                               !rcu_access_pointer(sk->sk_reuseport_cb) &&
 213                               uid_eq(tb->fastuid, uid))) &&
 214                             smallest_size != -1 && --attempts >= 0) {
 215                                 spin_unlock_bh(&head->lock);
 216                                 goto again;
 217                         }
 218                         goto fail_unlock;
 219                 }
 220                 if (!reuse)
 221                         tb->fastreuse = 0;
 222                 if (!sk->sk_reuseport || !uid_eq(tb->fastuid, uid))
 223                         tb->fastreuseport = 0;
 224         } else {
 225                 tb->fastreuse = reuse;
 226                 if (sk->sk_reuseport) {
 227                         tb->fastreuseport = 1;
 228                         tb->fastuid = uid;
 229                 } else {
 230                         tb->fastreuseport = 0;
 231                 }
 232         }
 233 success:
 234         if (!inet_csk(sk)->icsk_bind_hash)
 235                 inet_bind_hash(sk, tb, port);
 236         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 237         ret = 0;
 238
 239 fail_unlock:
 240         spin_unlock_bh(&head->lock);
 241         return ret;
 242 }
 243 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 244
 245 /*
 246  * Wait for an incoming connection, avoid race conditions. This must be called
 247  * with the socket locked.
 248  */
 249 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 250 {
 251         struct inet_connection_sock *icsk = inet_csk(sk);
 252         DEFINE_WAIT(wait);
 253         int err;
 254
 255         /*
 256          * True wake-one mechanism for incoming connections: only
 257          * one process gets woken up, not the 'whole herd'.
 258          * Since we do not 'race & poll' for established sockets
 259          * anymore, the common case will execute the loop only once.
 260          *
 261          * Subtle issue: "add_wait_queue_exclusive()" will be added
 262          * after any current non-exclusive waiters, and we know that
 263          * it will always _stay_ after any new non-exclusive waiters
 264          * because all non-exclusive waiters are added at the
 265          * beginning of the wait-queue. As such, it's ok to "drop"
 266          * our exclusiveness temporarily when we get woken up without
 267          * having to remove and re-insert us on the wait queue.
 268          */
 269         for (;;) {
 270                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 271                                           TASK_INTERRUPTIBLE);
 272                 release_sock(sk);
 273                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 274                         timeo = schedule_timeout(timeo);
 275                 sched_annotate_sleep();
 276                 lock_sock(sk);
 277                 err = 0;
 278                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 279                         break;
 280                 err = -EINVAL;
 281                 if (sk->sk_state != TCP_LISTEN)
 282                         break;
 283                 err = sock_intr_errno(timeo);
 284                 if (signal_pending(current))
 285                         break;
 286                 err = -EAGAIN;
 287                 if (!timeo)
 288                         break;
 289         }
 290         finish_wait(sk_sleep(sk), &wait);
 291         return err;
 292 }
 293
 294 /*
 295  * This will accept the next outstanding connection.
 296  */
 297 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 298 {
 299         struct inet_connection_sock *icsk = inet_csk(sk);
 300         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 301         struct request_sock *req;
 302         struct sock *newsk;
 303         int error;
 304
 305         lock_sock(sk);
 306
 307         /* We need to make sure that this socket is listening,
 308          * and that it has something pending.
 309          */
 310         error = -EINVAL;
 311         if (sk->sk_state != TCP_LISTEN)
 312                 goto out_err;
 313
 314         /* Find already established connection */
 315         if (reqsk_queue_empty(queue)) {
 316                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 317
 318                 /* If this is a non blocking socket don't sleep */
 319                 error = -EAGAIN;
 320                 if (!timeo)
 321                         goto out_err;
 322
 323                 error = inet_csk_wait_for_connect(sk, timeo);
 324                 if (error)
 325                         goto out_err;
 326         }
 327         req = reqsk_queue_remove(queue, sk);
 328         newsk = req->sk;
 329
 330         if (sk->sk_protocol == IPPROTO_TCP &&
 331             tcp_rsk(req)->tfo_listener) {
 332                 spin_lock_bh(&queue->fastopenq.lock);
 333                 if (tcp_rsk(req)->tfo_listener) {
 334                         /* We are still waiting for the final ACK from 3WHS
 335                          * so can't free req now. Instead, we set req->sk to
 336                          * NULL to signify that the child socket is taken
 337                          * so reqsk_fastopen_remove() will free the req
 338                          * when 3WHS finishes (or is aborted).
 339                          */
 340                         req->sk = NULL;
 341                         req = NULL;
 342                 }
 343                 spin_unlock_bh(&queue->fastopenq.lock);
 344         }
 345 out:
 346         release_sock(sk);
 347         if (req)
 348                 reqsk_put(req);
 349         return newsk;
 350 out_err:
 351         newsk = NULL;
 352         req = NULL;
 353         *err = error;
 354         goto out;
 355 }
 356 EXPORT_SYMBOL(inet_csk_accept);
 357
 358 /*
 359  * Using different timers for retransmit, delayed acks and probes
 360  * We may wish use just one timer maintaining a list of expire jiffies
 361  * to optimize.
 362  */
 363 void inet_csk_init_xmit_timers(struct sock *sk,
 364                                void (*retransmit_handler)(unsigned long),
 365                                void (*delack_handler)(unsigned long),
 366                                void (*keepalive_handler)(unsigned long))
 367 {
 368         struct inet_connection_sock *icsk = inet_csk(sk);
 369
 370         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 371                         (unsigned long)sk);
 372         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 373                         (unsigned long)sk);
 374         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 375         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 376 }
 377 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 378
 379 void inet_csk_clear_xmit_timers(struct sock *sk)
 380 {
 381         struct inet_connection_sock *icsk = inet_csk(sk);
 382
 383         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 384
 385         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 386         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 387         sk_stop_timer(sk, &sk->sk_timer);
 388 }
 389 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 390
 391 void inet_csk_delete_keepalive_timer(struct sock *sk)
 392 {
 393         sk_stop_timer(sk, &sk->sk_timer);
 394 }
 395 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 396
 397 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 398 {
 399         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 400 }
 401 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 402
 403 struct dst_entry *inet_csk_route_req(const struct sock *sk,
 404                                      struct flowi4 *fl4,
 405                                      const struct request_sock *req)
 406 {
 407         const struct inet_request_sock *ireq = inet_rsk(req);
 408         struct net *net = read_pnet(&ireq->ireq_net);
 409         struct ip_options_rcu *opt = ireq->opt;
 410         struct rtable *rt;
 411
 412         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 413                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 414                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 415                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 416                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 417                            htons(ireq->ir_num));
 418         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 419         rt = ip_route_output_flow(net, fl4, sk);
 420         if (IS_ERR(rt))
 421                 goto no_route;
 422         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 423                 goto route_err;
 424         return &rt->dst;
 425
 426 route_err:
 427         ip_rt_put(rt);
 428 no_route:
 429         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 430         return NULL;
 431 }
 432 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 433
 434 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
 435                                             struct sock *newsk,
 436                                             const struct request_sock *req)
 437 {
 438         const struct inet_request_sock *ireq = inet_rsk(req);
 439         struct net *net = read_pnet(&ireq->ireq_net);
 440         struct inet_sock *newinet = inet_sk(newsk);
 441         struct ip_options_rcu *opt;
 442         struct flowi4 *fl4;
 443         struct rtable *rt;
 444
 445         fl4 = &newinet->cork.fl.u.ip4;
 446
 447         rcu_read_lock();
 448         opt = rcu_dereference(newinet->inet_opt);
 449         flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
 450                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 451                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 452                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
 453                            ireq->ir_loc_addr, ireq->ir_rmt_port,
 454                            htons(ireq->ir_num));
 455         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 456         rt = ip_route_output_flow(net, fl4, sk);
 457         if (IS_ERR(rt))
 458                 goto no_route;
 459         if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
 460                 goto route_err;
 461         rcu_read_unlock();
 462         return &rt->dst;
 463
 464 route_err:
 465         ip_rt_put(rt);
 466 no_route:
 467         rcu_read_unlock();
 468         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 469         return NULL;
 470 }
 471 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 472
 473 #if IS_ENABLED(CONFIG_IPV6)
 474 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 475 #else
 476 #define AF_INET_FAMILY(fam) true
 477 #endif
 478
 479 /* Decide when to expire the request and when to resend SYN-ACK */
 480 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 481                                   const int max_retries,
 482                                   const u8 rskq_defer_accept,
 483                                   int *expire, int *resend)
 484 {
 485         if (!rskq_defer_accept) {
 486                 *expire = req->num_timeout >= thresh;
 487                 *resend = 1;
 488                 return;
 489         }
 490         *expire = req->num_timeout >= thresh &&
 491                   (!inet_rsk(req)->acked || req->num_timeout >= max_retries);
 492         /*
 493          * Do not resend while waiting for data after ACK,
 494          * start to resend on end of deferring period to give
 495          * last chance for data or ACK to create established socket.
 496          */
 497         *resend = !inet_rsk(req)->acked ||
 498                   req->num_timeout >= rskq_defer_accept - 1;
 499 }
 500
 501 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
 502 {
 503         int err = req->rsk_ops->rtx_syn_ack(parent, req);
 504
 505         if (!err)
 506                 req->num_retrans++;
 507         return err;
 508 }
 509 EXPORT_SYMBOL(inet_rtx_syn_ack);
 510
 511 /* return true if req was found in the ehash table */
 512 static bool reqsk_queue_unlink(struct request_sock_queue *queue,
 513                                struct request_sock *req)
 514 {
 515         struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo;
 516         bool found = false;
 517
 518         if (sk_hashed(req_to_sk(req))) {
 519                 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
 520
 521                 spin_lock(lock);
 522                 found = __sk_nulls_del_node_init_rcu(req_to_sk(req));
 523                 spin_unlock(lock);
 524         }
 525         if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
 526                 reqsk_put(req);
 527         return found;
 528 }
 529
 530 void inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
 531 {
 532         if (reqsk_queue_unlink(&inet_csk(sk)->icsk_accept_queue, req)) {
 533                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 534                 reqsk_put(req);
 535         }
 536 }
 537 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
 538
 539 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
 540 {
 541         inet_csk_reqsk_queue_drop(sk, req);
 542         reqsk_put(req);
 543 }
 544 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
 545
 546 static void reqsk_timer_handler(unsigned long data)
 547 {
 548         struct request_sock *req = (struct request_sock *)data;
 549         struct sock *sk_listener = req->rsk_listener;
 550         struct net *net = sock_net(sk_listener);
 551         struct inet_connection_sock *icsk = inet_csk(sk_listener);
 552         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 553         int qlen, expire = 0, resend = 0;
 554         int max_retries, thresh;
 555         u8 defer_accept;
 556
 557         if (sk_state_load(sk_listener) != TCP_LISTEN)
 558                 goto drop;
 559
 560         max_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries;
 561         thresh = max_retries;
 562         /* Normally all the openreqs are young and become mature
 563          * (i.e. converted to established socket) for first timeout.
 564          * If synack was not acknowledged for 1 second, it means
 565          * one of the following things: synack was lost, ack was lost,
 566          * rtt is high or nobody planned to ack (i.e. synflood).
 567          * When server is a bit loaded, queue is populated with old
 568          * open requests, reducing effective size of queue.
 569          * When server is well loaded, queue size reduces to zero
 570          * after several minutes of work. It is not synflood,
 571          * it is normal operation. The solution is pruning
 572          * too old entries overriding normal timeout, when
 573          * situation becomes dangerous.
 574          *
 575          * Essentially, we reserve half of room for young
 576          * embrions; and abort old ones without pity, if old
 577          * ones are about to clog our table.
 578          */
 579         qlen = reqsk_queue_len(queue);
 580         if ((qlen << 1) > max(8U, sk_listener->sk_max_ack_backlog)) {
 581                 int young = reqsk_queue_len_young(queue) << 1;
 582
 583                 while (thresh > 2) {
 584                         if (qlen < young)
 585                                 break;
 586                         thresh--;
 587                         young <<= 1;
 588                 }
 589         }
 590         defer_accept = READ_ONCE(queue->rskq_defer_accept);
 591         if (defer_accept)
 592                 max_retries = defer_accept;
 593         syn_ack_recalc(req, thresh, max_retries, defer_accept,
 594                        &expire, &resend);
 595         req->rsk_ops->syn_ack_timeout(req);
 596         if (!expire &&
 597             (!resend ||
 598              !inet_rtx_syn_ack(sk_listener, req) ||
 599              inet_rsk(req)->acked)) {
 600                 unsigned long timeo;
 601
 602                 if (req->num_timeout++ == 0)
 603                         atomic_dec(&queue->young);
 604                 timeo = min(TCP_TIMEOUT_INIT << req->num_timeout, TCP_RTO_MAX);
 605                 mod_timer_pinned(&req->rsk_timer, jiffies + timeo);
 606                 return;
 607         }
 608 drop:
 609         inet_csk_reqsk_queue_drop_and_put(sk_listener, req);
 610 }
 611
 612 static void reqsk_queue_hash_req(struct request_sock *req,
 613                                  unsigned long timeout)
 614 {
 615         req->num_retrans = 0;
 616         req->num_timeout = 0;
 617         req->sk = NULL;
 618
 619         setup_timer(&req->rsk_timer, reqsk_timer_handler, (unsigned long)req);
 620         mod_timer_pinned(&req->rsk_timer, jiffies + timeout);
 621
 622         inet_ehash_insert(req_to_sk(req), NULL);
 623         /* before letting lookups find us, make sure all req fields
 624          * are committed to memory and refcnt initialized.
 625          */
 626         smp_wmb();
 627         atomic_set(&req->rsk_refcnt, 2 + 1);
 628 }
 629
 630 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 631                                    unsigned long timeout)
 632 {
 633         reqsk_queue_hash_req(req, timeout);
 634         inet_csk_reqsk_queue_added(sk);
 635 }
 636 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 637
 638 /**
 639  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 640  *      @sk: the socket to clone
 641  *      @req: request_sock
 642  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 643  *
 644  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 645  */
 646 struct sock *inet_csk_clone_lock(const struct sock *sk,
 647                                  const struct request_sock *req,
 648                                  const gfp_t priority)
 649 {
 650         struct sock *newsk = sk_clone_lock(sk, priority);
 651
 652         if (newsk) {
 653                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 654
 655                 newsk->sk_state = TCP_SYN_RECV;
 656                 newicsk->icsk_bind_hash = NULL;
 657
 658                 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
 659                 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
 660                 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
 661                 newsk->sk_write_space = sk_stream_write_space;
 662
 663                 newsk->sk_mark = inet_rsk(req)->ir_mark;
 664                 atomic64_set(&newsk->sk_cookie,
 665                              atomic64_read(&inet_rsk(req)->ir_cookie));
 666
 667                 newicsk->icsk_retransmits = 0;
 668                 newicsk->icsk_backoff     = 0;
 669                 newicsk->icsk_probes_out  = 0;
 670
 671                 /* Deinitialize accept_queue to trap illegal accesses. */
 672                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 673
 674                 security_inet_csk_clone(newsk, req);
 675         }
 676         return newsk;
 677 }
 678 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 679
 680 /*
 681  * At this point, there should be no process reference to this
 682  * socket, and thus no user references at all.  Therefore we
 683  * can assume the socket waitqueue is inactive and nobody will
 684  * try to jump onto it.
 685  */
 686 void inet_csk_destroy_sock(struct sock *sk)
 687 {
 688         WARN_ON(sk->sk_state != TCP_CLOSE);
 689         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 690
 691         /* It cannot be in hash table! */
 692         WARN_ON(!sk_unhashed(sk));
 693
 694         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 695         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 696
 697         sk->sk_prot->destroy(sk);
 698
 699         sk_stream_kill_queues(sk);
 700
 701         xfrm_sk_free_policy(sk);
 702
 703         sk_refcnt_debug_release(sk);
 704
 705         percpu_counter_dec(sk->sk_prot->orphan_count);
 706         sock_put(sk);
 707 }
 708 EXPORT_SYMBOL(inet_csk_destroy_sock);
 709
 710 /* This function allows to force a closure of a socket after the call to
 711  * tcp/dccp_create_openreq_child().
 712  */
 713 void inet_csk_prepare_forced_close(struct sock *sk)
 714         __releases(&sk->sk_lock.slock)
 715 {
 716         /* sk_clone_lock locked the socket and set refcnt to 2 */
 717         bh_unlock_sock(sk);
 718         sock_put(sk);
 719
 720         /* The below has to be done to allow calling inet_csk_destroy_sock */
 721         sock_set_flag(sk, SOCK_DEAD);
 722         percpu_counter_inc(sk->sk_prot->orphan_count);
 723         inet_sk(sk)->inet_num = 0;
 724 }
 725 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
 726
 727 int inet_csk_listen_start(struct sock *sk, int backlog)
 728 {
 729         struct inet_connection_sock *icsk = inet_csk(sk);
 730         struct inet_sock *inet = inet_sk(sk);
 731         int err = -EADDRINUSE;
 732
 733         reqsk_queue_alloc(&icsk->icsk_accept_queue);
 734
 735         sk->sk_max_ack_backlog = backlog;
 736         sk->sk_ack_backlog = 0;
 737         inet_csk_delack_init(sk);
 738
 739         /* There is race window here: we announce ourselves listening,
 740          * but this transition is still not validated by get_port().
 741          * It is OK, because this socket enters to hash table only
 742          * after validation is complete.
 743          */
 744         sk_state_store(sk, TCP_LISTEN);
 745         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 746                 inet->inet_sport = htons(inet->inet_num);
 747
 748                 sk_dst_reset(sk);
 749                 err = sk->sk_prot->hash(sk);
 750
 751                 if (likely(!err))
 752                         return 0;
 753         }
 754
 755         sk->sk_state = TCP_CLOSE;
 756         return err;
 757 }
 758 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 759
 760 static void inet_child_forget(struct sock *sk, struct request_sock *req,
 761                               struct sock *child)
 762 {
 763         sk->sk_prot->disconnect(child, O_NONBLOCK);
 764
 765         sock_orphan(child);
 766
 767         percpu_counter_inc(sk->sk_prot->orphan_count);
 768
 769         if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
 770                 BUG_ON(tcp_sk(child)->fastopen_rsk != req);
 771                 BUG_ON(sk != req->rsk_listener);
 772
 773                 /* Paranoid, to prevent race condition if
 774                  * an inbound pkt destined for child is
 775                  * blocked by sock lock in tcp_v4_rcv().
 776                  * Also to satisfy an assertion in
 777                  * tcp_v4_destroy_sock().
 778                  */
 779                 tcp_sk(child)->fastopen_rsk = NULL;
 780         }
 781         inet_csk_destroy_sock(child);
 782         reqsk_put(req);
 783 }
 784
 785 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
 786                                       struct request_sock *req,
 787                                       struct sock *child)
 788 {
 789         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 790
 791         spin_lock(&queue->rskq_lock);
 792         if (unlikely(sk->sk_state != TCP_LISTEN)) {
 793                 inet_child_forget(sk, req, child);
 794                 child = NULL;
 795         } else {
 796                 req->sk = child;
 797                 req->dl_next = NULL;
 798                 if (queue->rskq_accept_head == NULL)
 799                         queue->rskq_accept_head = req;
 800                 else
 801                         queue->rskq_accept_tail->dl_next = req;
 802                 queue->rskq_accept_tail = req;
 803                 sk_acceptq_added(sk);
 804         }
 805         spin_unlock(&queue->rskq_lock);
 806         return child;
 807 }
 808 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
 809
 810 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
 811                                          struct request_sock *req, bool own_req)
 812 {
 813         if (own_req) {
 814                 inet_csk_reqsk_queue_drop(sk, req);
 815                 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
 816                 if (inet_csk_reqsk_queue_add(sk, req, child))
 817                         return child;
 818         }
 819         /* Too bad, another child took ownership of the request, undo. */
 820         bh_unlock_sock(child);
 821         sock_put(child);
 822         return NULL;
 823 }
 824 EXPORT_SYMBOL(inet_csk_complete_hashdance);
 825
 826 /*
 827  *      This routine closes sockets which have been at least partially
 828  *      opened, but not yet accepted.
 829  */
 830 void inet_csk_listen_stop(struct sock *sk)
 831 {
 832         struct inet_connection_sock *icsk = inet_csk(sk);
 833         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 834         struct request_sock *next, *req;
 835
 836         /* Following specs, it would be better either to send FIN
 837          * (and enter FIN-WAIT-1, it is normal close)
 838          * or to send active reset (abort).
 839          * Certainly, it is pretty dangerous while synflood, but it is
 840          * bad justification for our negligence 8)
 841          * To be honest, we are not able to make either
 842          * of the variants now.                 --ANK
 843          */
 844         while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
 845                 struct sock *child = req->sk;
 846
 847                 local_bh_disable();
 848                 bh_lock_sock(child);
 849                 WARN_ON(sock_owned_by_user(child));
 850                 sock_hold(child);
 851
 852                 inet_child_forget(sk, req, child);
 853                 bh_unlock_sock(child);
 854                 local_bh_enable();
 855                 sock_put(child);
 856
 857                 cond_resched();
 858         }
 859         if (queue->fastopenq.rskq_rst_head) {
 860                 /* Free all the reqs queued in rskq_rst_head. */
 861                 spin_lock_bh(&queue->fastopenq.lock);
 862                 req = queue->fastopenq.rskq_rst_head;
 863                 queue->fastopenq.rskq_rst_head = NULL;
 864                 spin_unlock_bh(&queue->fastopenq.lock);
 865                 while (req != NULL) {
 866                         next = req->dl_next;
 867                         reqsk_put(req);
 868                         req = next;
 869                 }
 870         }
 871         WARN_ON_ONCE(sk->sk_ack_backlog);
 872 }
 873 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 874
 875 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 876 {
 877         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 878         const struct inet_sock *inet = inet_sk(sk);
 879
 880         sin->sin_family         = AF_INET;
 881         sin->sin_addr.s_addr    = inet->inet_daddr;
 882         sin->sin_port           = inet->inet_dport;
 883 }
 884 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 885
 886 #ifdef CONFIG_COMPAT
 887 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 888                                char __user *optval, int __user *optlen)
 889 {
 890         const struct inet_connection_sock *icsk = inet_csk(sk);
 891
 892         if (icsk->icsk_af_ops->compat_getsockopt)
 893                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 894                                                             optval, optlen);
 895         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 896                                              optval, optlen);
 897 }
 898 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 899
 900 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 901                                char __user *optval, unsigned int optlen)
 902 {
 903         const struct inet_connection_sock *icsk = inet_csk(sk);
 904
 905         if (icsk->icsk_af_ops->compat_setsockopt)
 906                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 907                                                             optval, optlen);
 908         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 909                                              optval, optlen);
 910 }
 911 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 912 #endif
 913
 914 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
 915 {
 916         const struct inet_sock *inet = inet_sk(sk);
 917         const struct ip_options_rcu *inet_opt;
 918         __be32 daddr = inet->inet_daddr;
 919         struct flowi4 *fl4;
 920         struct rtable *rt;
 921
 922         rcu_read_lock();
 923         inet_opt = rcu_dereference(inet->inet_opt);
 924         if (inet_opt && inet_opt->opt.srr)
 925                 daddr = inet_opt->opt.faddr;
 926         fl4 = &fl->u.ip4;
 927         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
 928                                    inet->inet_saddr, inet->inet_dport,
 929                                    inet->inet_sport, sk->sk_protocol,
 930                                    RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
 931         if (IS_ERR(rt))
 932                 rt = NULL;
 933         if (rt)
 934                 sk_setup_caps(sk, &rt->dst);
 935         rcu_read_unlock();
 936
 937         return &rt->dst;
 938 }
 939
 940 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
 941 {
 942         struct dst_entry *dst = __sk_dst_check(sk, 0);
 943         struct inet_sock *inet = inet_sk(sk);
 944
 945         if (!dst) {
 946                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 947                 if (!dst)
 948                         goto out;
 949         }
 950         dst->ops->update_pmtu(dst, sk, NULL, mtu);
 951
 952         dst = __sk_dst_check(sk, 0);
 953         if (!dst)
 954                 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
 955 out:
 956         return dst;
 957 }
 958 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);