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
  27 #ifdef INET_CSK_DEBUG
  28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
  29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
  30 #endif
  31
  32 /*
  33  * This struct holds the first and last local port number.
  34  */
  35 struct local_ports sysctl_local_ports __read_mostly = {
  36         .lock = SEQLOCK_UNLOCKED,
  37         .range = { 32768, 61000 },
  38 };
  39
  40 unsigned long *sysctl_local_reserved_ports;
  41 EXPORT_SYMBOL(sysctl_local_reserved_ports);
  42
  43 void inet_get_local_port_range(int *low, int *high)
  44 {
  45         unsigned seq;
  46         do {
  47                 seq = read_seqbegin(&sysctl_local_ports.lock);
  48
  49                 *low = sysctl_local_ports.range[0];
  50                 *high = sysctl_local_ports.range[1];
  51         } while (read_seqretry(&sysctl_local_ports.lock, seq));
  52 }
  53 EXPORT_SYMBOL(inet_get_local_port_range);
  54
  55 int inet_csk_bind_conflict(const struct sock *sk,
  56                            const struct inet_bind_bucket *tb)
  57 {
  58         const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
  59         struct sock *sk2;
  60         struct hlist_node *node;
  61         int reuse = sk->sk_reuse;
  62
  63         /*
  64          * Unlike other sk lookup places we do not check
  65          * for sk_net here, since _all_ the socks listed
  66          * in tb->owners list belong to the same net - the
  67          * one this bucket belongs to.
  68          */
  69
  70         sk_for_each_bound(sk2, node, &tb->owners) {
  71                 if (sk != sk2 &&
  72                     !inet_v6_ipv6only(sk2) &&
  73                     (!sk->sk_bound_dev_if ||
  74                      !sk2->sk_bound_dev_if ||
  75                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  76                         if (!reuse || !sk2->sk_reuse ||
  77                             sk2->sk_state == TCP_LISTEN) {
  78                                 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
  79                                 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
  80                                     sk2_rcv_saddr == sk_rcv_saddr)
  81                                         break;
  82                         }
  83                 }
  84         }
  85         return node != NULL;
  86 }
  87
  88 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  89
  90 /* Obtain a reference to a local port for the given sock,
  91  * if snum is zero it means select any available local port.
  92  */
  93 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  94 {
  95         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  96         struct inet_bind_hashbucket *head;
  97         struct hlist_node *node;
  98         struct inet_bind_bucket *tb;
  99         int ret, attempts = 5;
 100         struct net *net = sock_net(sk);
 101         int smallest_size = -1, smallest_rover;
 102
 103         local_bh_disable();
 104         if (!snum) {
 105                 int remaining, rover, low, high;
 106
 107 again:
 108                 inet_get_local_port_range(&low, &high);
 109                 remaining = (high - low) + 1;
 110                 smallest_rover = rover = net_random() % remaining + low;
 111
 112                 smallest_size = -1;
 113                 do {
 114                         if (inet_is_reserved_local_port(rover))
 115                                 goto next_nolock;
 116                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 117                                         hashinfo->bhash_size)];
 118                         spin_lock(&head->lock);
 119                         inet_bind_bucket_for_each(tb, node, &head->chain)
 120                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 121                                         if (tb->fastreuse > 0 &&
 122                                             sk->sk_reuse &&
 123                                             sk->sk_state != TCP_LISTEN &&
 124                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 125                                                 smallest_size = tb->num_owners;
 126                                                 smallest_rover = rover;
 127                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
 128                                                         spin_unlock(&head->lock);
 129                                                         snum = smallest_rover;
 130                                                         goto have_snum;
 131                                                 }
 132                                         }
 133                                         goto next;
 134                                 }
 135                         break;
 136                 next:
 137                         spin_unlock(&head->lock);
 138                 next_nolock:
 139                         if (++rover > high)
 140                                 rover = low;
 141                 } while (--remaining > 0);
 142
 143                 /* Exhausted local port range during search?  It is not
 144                  * possible for us to be holding one of the bind hash
 145                  * locks if this test triggers, because if 'remaining'
 146                  * drops to zero, we broke out of the do/while loop at
 147                  * the top level, not from the 'break;' statement.
 148                  */
 149                 ret = 1;
 150                 if (remaining <= 0) {
 151                         if (smallest_size != -1) {
 152                                 snum = smallest_rover;
 153                                 goto have_snum;
 154                         }
 155                         goto fail;
 156                 }
 157                 /* OK, here is the one we will use.  HEAD is
 158                  * non-NULL and we hold it's mutex.
 159                  */
 160                 snum = rover;
 161         } else {
 162 have_snum:
 163                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 164                                 hashinfo->bhash_size)];
 165                 spin_lock(&head->lock);
 166                 inet_bind_bucket_for_each(tb, node, &head->chain)
 167                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 168                                 goto tb_found;
 169         }
 170         tb = NULL;
 171         goto tb_not_found;
 172 tb_found:
 173         if (!hlist_empty(&tb->owners)) {
 174                 if (tb->fastreuse > 0 &&
 175                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 176                     smallest_size == -1) {
 177                         goto success;
 178                 } else {
 179                         ret = 1;
 180                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 181                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 182                                     smallest_size != -1 && --attempts >= 0) {
 183                                         spin_unlock(&head->lock);
 184                                         goto again;
 185                                 }
 186                                 goto fail_unlock;
 187                         }
 188                 }
 189         }
 190 tb_not_found:
 191         ret = 1;
 192         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 193                                         net, head, snum)) == NULL)
 194                 goto fail_unlock;
 195         if (hlist_empty(&tb->owners)) {
 196                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 197                         tb->fastreuse = 1;
 198                 else
 199                         tb->fastreuse = 0;
 200         } else if (tb->fastreuse &&
 201                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 202                 tb->fastreuse = 0;
 203 success:
 204         if (!inet_csk(sk)->icsk_bind_hash)
 205                 inet_bind_hash(sk, tb, snum);
 206         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 207         ret = 0;
 208
 209 fail_unlock:
 210         spin_unlock(&head->lock);
 211 fail:
 212         local_bh_enable();
 213         return ret;
 214 }
 215
 216 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 217
 218 /*
 219  * Wait for an incoming connection, avoid race conditions. This must be called
 220  * with the socket locked.
 221  */
 222 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 223 {
 224         struct inet_connection_sock *icsk = inet_csk(sk);
 225         DEFINE_WAIT(wait);
 226         int err;
 227
 228         /*
 229          * True wake-one mechanism for incoming connections: only
 230          * one process gets woken up, not the 'whole herd'.
 231          * Since we do not 'race & poll' for established sockets
 232          * anymore, the common case will execute the loop only once.
 233          *
 234          * Subtle issue: "add_wait_queue_exclusive()" will be added
 235          * after any current non-exclusive waiters, and we know that
 236          * it will always _stay_ after any new non-exclusive waiters
 237          * because all non-exclusive waiters are added at the
 238          * beginning of the wait-queue. As such, it's ok to "drop"
 239          * our exclusiveness temporarily when we get woken up without
 240          * having to remove and re-insert us on the wait queue.
 241          */
 242         for (;;) {
 243                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 244                                           TASK_INTERRUPTIBLE);
 245                 release_sock(sk);
 246                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 247                         timeo = schedule_timeout(timeo);
 248                 lock_sock(sk);
 249                 err = 0;
 250                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 251                         break;
 252                 err = -EINVAL;
 253                 if (sk->sk_state != TCP_LISTEN)
 254                         break;
 255                 err = sock_intr_errno(timeo);
 256                 if (signal_pending(current))
 257                         break;
 258                 err = -EAGAIN;
 259                 if (!timeo)
 260                         break;
 261         }
 262         finish_wait(sk_sleep(sk), &wait);
 263         return err;
 264 }
 265
 266 /*
 267  * This will accept the next outstanding connection.
 268  */
 269 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 270 {
 271         struct inet_connection_sock *icsk = inet_csk(sk);
 272         struct sock *newsk;
 273         int error;
 274
 275         lock_sock(sk);
 276
 277         /* We need to make sure that this socket is listening,
 278          * and that it has something pending.
 279          */
 280         error = -EINVAL;
 281         if (sk->sk_state != TCP_LISTEN)
 282                 goto out_err;
 283
 284         /* Find already established connection */
 285         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
 286                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 287
 288                 /* If this is a non blocking socket don't sleep */
 289                 error = -EAGAIN;
 290                 if (!timeo)
 291                         goto out_err;
 292
 293                 error = inet_csk_wait_for_connect(sk, timeo);
 294                 if (error)
 295                         goto out_err;
 296         }
 297
 298         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
 299         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
 300 out:
 301         release_sock(sk);
 302         return newsk;
 303 out_err:
 304         newsk = NULL;
 305         *err = error;
 306         goto out;
 307 }
 308
 309 EXPORT_SYMBOL(inet_csk_accept);
 310
 311 /*
 312  * Using different timers for retransmit, delayed acks and probes
 313  * We may wish use just one timer maintaining a list of expire jiffies
 314  * to optimize.
 315  */
 316 void inet_csk_init_xmit_timers(struct sock *sk,
 317                                void (*retransmit_handler)(unsigned long),
 318                                void (*delack_handler)(unsigned long),
 319                                void (*keepalive_handler)(unsigned long))
 320 {
 321         struct inet_connection_sock *icsk = inet_csk(sk);
 322
 323         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 324                         (unsigned long)sk);
 325         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 326                         (unsigned long)sk);
 327         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 328         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 329 }
 330
 331 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 332
 333 void inet_csk_clear_xmit_timers(struct sock *sk)
 334 {
 335         struct inet_connection_sock *icsk = inet_csk(sk);
 336
 337         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 338
 339         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 340         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 341         sk_stop_timer(sk, &sk->sk_timer);
 342 }
 343
 344 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 345
 346 void inet_csk_delete_keepalive_timer(struct sock *sk)
 347 {
 348         sk_stop_timer(sk, &sk->sk_timer);
 349 }
 350
 351 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 352
 353 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 354 {
 355         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 356 }
 357
 358 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 359
 360 struct dst_entry *inet_csk_route_req(struct sock *sk,
 361                                      const struct request_sock *req)
 362 {
 363         struct rtable *rt;
 364         const struct inet_request_sock *ireq = inet_rsk(req);
 365         struct ip_options *opt = inet_rsk(req)->opt;
 366         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 367                             .mark = sk->sk_mark,
 368                             .nl_u = { .ip4_u =
 369                                       { .daddr = ((opt && opt->srr) ?
 370                                                   opt->faddr :
 371                                                   ireq->rmt_addr),
 372                                         .saddr = ireq->loc_addr,
 373                                         .tos = RT_CONN_FLAGS(sk) } },
 374                             .proto = sk->sk_protocol,
 375                             .flags = inet_sk_flowi_flags(sk),
 376                             .uli_u = { .ports =
 377                                        { .sport = inet_sk(sk)->inet_sport,
 378                                          .dport = ireq->rmt_port } } };
 379         struct net *net = sock_net(sk);
 380
 381         security_req_classify_flow(req, &fl);
 382         if (ip_route_output_flow(net, &rt, &fl, sk, 0))
 383                 goto no_route;
 384         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 385                 goto route_err;
 386         return &rt->u.dst;
 387
 388 route_err:
 389         ip_rt_put(rt);
 390 no_route:
 391         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 392         return NULL;
 393 }
 394
 395 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 396
 397 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 398                                  const u32 rnd, const u32 synq_hsize)
 399 {
 400         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 401 }
 402
 403 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 404 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 405 #else
 406 #define AF_INET_FAMILY(fam) 1
 407 #endif
 408
 409 struct request_sock *inet_csk_search_req(const struct sock *sk,
 410                                          struct request_sock ***prevp,
 411                                          const __be16 rport, const __be32 raddr,
 412                                          const __be32 laddr)
 413 {
 414         const struct inet_connection_sock *icsk = inet_csk(sk);
 415         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 416         struct request_sock *req, **prev;
 417
 418         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 419                                                     lopt->nr_table_entries)];
 420              (req = *prev) != NULL;
 421              prev = &req->dl_next) {
 422                 const struct inet_request_sock *ireq = inet_rsk(req);
 423
 424                 if (ireq->rmt_port == rport &&
 425                     ireq->rmt_addr == raddr &&
 426                     ireq->loc_addr == laddr &&
 427                     AF_INET_FAMILY(req->rsk_ops->family)) {
 428                         WARN_ON(req->sk);
 429                         *prevp = prev;
 430                         break;
 431                 }
 432         }
 433
 434         return req;
 435 }
 436
 437 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 438
 439 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 440                                    unsigned long timeout)
 441 {
 442         struct inet_connection_sock *icsk = inet_csk(sk);
 443         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 444         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 445                                      lopt->hash_rnd, lopt->nr_table_entries);
 446
 447         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 448         inet_csk_reqsk_queue_added(sk, timeout);
 449 }
 450
 451 /* Only thing we need from tcp.h */
 452 extern int sysctl_tcp_synack_retries;
 453
 454 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 455
 456 /* Decide when to expire the request and when to resend SYN-ACK */
 457 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 458                                   const int max_retries,
 459                                   const u8 rskq_defer_accept,
 460                                   int *expire, int *resend)
 461 {
 462         if (!rskq_defer_accept) {
 463                 *expire = req->retrans >= thresh;
 464                 *resend = 1;
 465                 return;
 466         }
 467         *expire = req->retrans >= thresh &&
 468                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 469         /*
 470          * Do not resend while waiting for data after ACK,
 471          * start to resend on end of deferring period to give
 472          * last chance for data or ACK to create established socket.
 473          */
 474         *resend = !inet_rsk(req)->acked ||
 475                   req->retrans >= rskq_defer_accept - 1;
 476 }
 477
 478 void inet_csk_reqsk_queue_prune(struct sock *parent,
 479                                 const unsigned long interval,
 480                                 const unsigned long timeout,
 481                                 const unsigned long max_rto)
 482 {
 483         struct inet_connection_sock *icsk = inet_csk(parent);
 484         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 485         struct listen_sock *lopt = queue->listen_opt;
 486         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 487         int thresh = max_retries;
 488         unsigned long now = jiffies;
 489         struct request_sock **reqp, *req;
 490         int i, budget;
 491
 492         if (lopt == NULL || lopt->qlen == 0)
 493                 return;
 494
 495         /* Normally all the openreqs are young and become mature
 496          * (i.e. converted to established socket) for first timeout.
 497          * If synack was not acknowledged for 3 seconds, it means
 498          * one of the following things: synack was lost, ack was lost,
 499          * rtt is high or nobody planned to ack (i.e. synflood).
 500          * When server is a bit loaded, queue is populated with old
 501          * open requests, reducing effective size of queue.
 502          * When server is well loaded, queue size reduces to zero
 503          * after several minutes of work. It is not synflood,
 504          * it is normal operation. The solution is pruning
 505          * too old entries overriding normal timeout, when
 506          * situation becomes dangerous.
 507          *
 508          * Essentially, we reserve half of room for young
 509          * embrions; and abort old ones without pity, if old
 510          * ones are about to clog our table.
 511          */
 512         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 513                 int young = (lopt->qlen_young<<1);
 514
 515                 while (thresh > 2) {
 516                         if (lopt->qlen < young)
 517                                 break;
 518                         thresh--;
 519                         young <<= 1;
 520                 }
 521         }
 522
 523         if (queue->rskq_defer_accept)
 524                 max_retries = queue->rskq_defer_accept;
 525
 526         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 527         i = lopt->clock_hand;
 528
 529         do {
 530                 reqp=&lopt->syn_table[i];
 531                 while ((req = *reqp) != NULL) {
 532                         if (time_after_eq(now, req->expires)) {
 533                                 int expire = 0, resend = 0;
 534
 535                                 syn_ack_recalc(req, thresh, max_retries,
 536                                                queue->rskq_defer_accept,
 537                                                &expire, &resend);
 538                                 if (req->rsk_ops->syn_ack_timeout)
 539                                         req->rsk_ops->syn_ack_timeout(parent, req);
 540                                 if (!expire &&
 541                                     (!resend ||
 542                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 543                                      inet_rsk(req)->acked)) {
 544                                         unsigned long timeo;
 545
 546                                         if (req->retrans++ == 0)
 547                                                 lopt->qlen_young--;
 548                                         timeo = min((timeout << req->retrans), max_rto);
 549                                         req->expires = now + timeo;
 550                                         reqp = &req->dl_next;
 551                                         continue;
 552                                 }
 553
 554                                 /* Drop this request */
 555                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 556                                 reqsk_queue_removed(queue, req);
 557                                 reqsk_free(req);
 558                                 continue;
 559                         }
 560                         reqp = &req->dl_next;
 561                 }
 562
 563                 i = (i + 1) & (lopt->nr_table_entries - 1);
 564
 565         } while (--budget > 0);
 566
 567         lopt->clock_hand = i;
 568
 569         if (lopt->qlen)
 570                 inet_csk_reset_keepalive_timer(parent, interval);
 571 }
 572
 573 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 574
 575 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
 576                             const gfp_t priority)
 577 {
 578         struct sock *newsk = sk_clone(sk, priority);
 579
 580         if (newsk != NULL) {
 581                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 582
 583                 newsk->sk_state = TCP_SYN_RECV;
 584                 newicsk->icsk_bind_hash = NULL;
 585
 586                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 587                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 588                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 589                 newsk->sk_write_space = sk_stream_write_space;
 590
 591                 newicsk->icsk_retransmits = 0;
 592                 newicsk->icsk_backoff     = 0;
 593                 newicsk->icsk_probes_out  = 0;
 594
 595                 /* Deinitialize accept_queue to trap illegal accesses. */
 596                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 597
 598                 security_inet_csk_clone(newsk, req);
 599         }
 600         return newsk;
 601 }
 602
 603 EXPORT_SYMBOL_GPL(inet_csk_clone);
 604
 605 /*
 606  * At this point, there should be no process reference to this
 607  * socket, and thus no user references at all.  Therefore we
 608  * can assume the socket waitqueue is inactive and nobody will
 609  * try to jump onto it.
 610  */
 611 void inet_csk_destroy_sock(struct sock *sk)
 612 {
 613         WARN_ON(sk->sk_state != TCP_CLOSE);
 614         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 615
 616         /* It cannot be in hash table! */
 617         WARN_ON(!sk_unhashed(sk));
 618
 619         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 620         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 621
 622         sk->sk_prot->destroy(sk);
 623
 624         sk_stream_kill_queues(sk);
 625
 626         xfrm_sk_free_policy(sk);
 627
 628         sk_refcnt_debug_release(sk);
 629
 630         percpu_counter_dec(sk->sk_prot->orphan_count);
 631         sock_put(sk);
 632 }
 633
 634 EXPORT_SYMBOL(inet_csk_destroy_sock);
 635
 636 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 637 {
 638         struct inet_sock *inet = inet_sk(sk);
 639         struct inet_connection_sock *icsk = inet_csk(sk);
 640         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 641
 642         if (rc != 0)
 643                 return rc;
 644
 645         sk->sk_max_ack_backlog = 0;
 646         sk->sk_ack_backlog = 0;
 647         inet_csk_delack_init(sk);
 648
 649         /* There is race window here: we announce ourselves listening,
 650          * but this transition is still not validated by get_port().
 651          * It is OK, because this socket enters to hash table only
 652          * after validation is complete.
 653          */
 654         sk->sk_state = TCP_LISTEN;
 655         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 656                 inet->inet_sport = htons(inet->inet_num);
 657
 658                 sk_dst_reset(sk);
 659                 sk->sk_prot->hash(sk);
 660
 661                 return 0;
 662         }
 663
 664         sk->sk_state = TCP_CLOSE;
 665         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 666         return -EADDRINUSE;
 667 }
 668
 669 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 670
 671 /*
 672  *      This routine closes sockets which have been at least partially
 673  *      opened, but not yet accepted.
 674  */
 675 void inet_csk_listen_stop(struct sock *sk)
 676 {
 677         struct inet_connection_sock *icsk = inet_csk(sk);
 678         struct request_sock *acc_req;
 679         struct request_sock *req;
 680
 681         inet_csk_delete_keepalive_timer(sk);
 682
 683         /* make all the listen_opt local to us */
 684         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 685
 686         /* Following specs, it would be better either to send FIN
 687          * (and enter FIN-WAIT-1, it is normal close)
 688          * or to send active reset (abort).
 689          * Certainly, it is pretty dangerous while synflood, but it is
 690          * bad justification for our negligence 8)
 691          * To be honest, we are not able to make either
 692          * of the variants now.                 --ANK
 693          */
 694         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 695
 696         while ((req = acc_req) != NULL) {
 697                 struct sock *child = req->sk;
 698
 699                 acc_req = req->dl_next;
 700
 701                 local_bh_disable();
 702                 bh_lock_sock(child);
 703                 WARN_ON(sock_owned_by_user(child));
 704                 sock_hold(child);
 705
 706                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 707
 708                 sock_orphan(child);
 709
 710                 percpu_counter_inc(sk->sk_prot->orphan_count);
 711
 712                 inet_csk_destroy_sock(child);
 713
 714                 bh_unlock_sock(child);
 715                 local_bh_enable();
 716                 sock_put(child);
 717
 718                 sk_acceptq_removed(sk);
 719                 __reqsk_free(req);
 720         }
 721         WARN_ON(sk->sk_ack_backlog);
 722 }
 723
 724 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 725
 726 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 727 {
 728         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 729         const struct inet_sock *inet = inet_sk(sk);
 730
 731         sin->sin_family         = AF_INET;
 732         sin->sin_addr.s_addr    = inet->inet_daddr;
 733         sin->sin_port           = inet->inet_dport;
 734 }
 735
 736 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 737
 738 #ifdef CONFIG_COMPAT
 739 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 740                                char __user *optval, int __user *optlen)
 741 {
 742         const struct inet_connection_sock *icsk = inet_csk(sk);
 743
 744         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 745                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 746                                                             optval, optlen);
 747         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 748                                              optval, optlen);
 749 }
 750
 751 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 752
 753 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 754                                char __user *optval, unsigned int optlen)
 755 {
 756         const struct inet_connection_sock *icsk = inet_csk(sk);
 757
 758         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 759                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 760                                                             optval, optlen);
 761         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 762                                              optval, optlen);
 763 }
 764
 765 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 766 #endif