SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 void inet_get_local_port_range(int *low, int *high)
  41 {
  42         unsigned seq;
  43         do {
  44                 seq = read_seqbegin(&sysctl_local_ports.lock);
  45
  46                 *low = sysctl_local_ports.range[0];
  47                 *high = sysctl_local_ports.range[1];
  48         } while (read_seqretry(&sysctl_local_ports.lock, seq));
  49 }
  50 EXPORT_SYMBOL(inet_get_local_port_range);
  51
  52 int inet_csk_bind_conflict(const struct sock *sk,
  53                            const struct inet_bind_bucket *tb)
  54 {
  55         const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
  56         struct sock *sk2;
  57         struct hlist_node *node;
  58         int reuse = sk->sk_reuse;
  59
  60         /*
  61          * Unlike other sk lookup places we do not check
  62          * for sk_net here, since _all_ the socks listed
  63          * in tb->owners list belong to the same net - the
  64          * one this bucket belongs to.
  65          */
  66
  67         sk_for_each_bound(sk2, node, &tb->owners) {
  68                 if (sk != sk2 &&
  69                     !inet_v6_ipv6only(sk2) &&
  70                     (!sk->sk_bound_dev_if ||
  71                      !sk2->sk_bound_dev_if ||
  72                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  73                         if (!reuse || !sk2->sk_reuse ||
  74                             sk2->sk_state == TCP_LISTEN) {
  75                                 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
  76                                 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
  77                                     sk2_rcv_saddr == sk_rcv_saddr)
  78                                         break;
  79                         }
  80                 }
  81         }
  82         return node != NULL;
  83 }
  84
  85 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  86
  87 /* Obtain a reference to a local port for the given sock,
  88  * if snum is zero it means select any available local port.
  89  */
  90 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  91 {
  92         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  93         struct inet_bind_hashbucket *head;
  94         struct hlist_node *node;
  95         struct inet_bind_bucket *tb;
  96         int ret, attempts = 5;
  97         struct net *net = sock_net(sk);
  98         int smallest_size = -1, smallest_rover;
  99
 100         local_bh_disable();
 101         if (!snum) {
 102                 int remaining, rover, low, high;
 103
 104 again:
 105                 inet_get_local_port_range(&low, &high);
 106                 remaining = (high - low) + 1;
 107                 smallest_rover = rover = net_random() % remaining + low;
 108
 109                 smallest_size = -1;
 110                 do {
 111                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 112                                         hashinfo->bhash_size)];
 113                         spin_lock(&head->lock);
 114                         inet_bind_bucket_for_each(tb, node, &head->chain)
 115                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 116                                         if (tb->fastreuse > 0 &&
 117                                             sk->sk_reuse &&
 118                                             sk->sk_state != TCP_LISTEN &&
 119                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 120                                                 smallest_size = tb->num_owners;
 121                                                 smallest_rover = rover;
 122                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
 123                                                         spin_unlock(&head->lock);
 124                                                         snum = smallest_rover;
 125                                                         goto have_snum;
 126                                                 }
 127                                         }
 128                                         goto next;
 129                                 }
 130                         break;
 131                 next:
 132                         spin_unlock(&head->lock);
 133                         if (++rover > high)
 134                                 rover = low;
 135                 } while (--remaining > 0);
 136
 137                 /* Exhausted local port range during search?  It is not
 138                  * possible for us to be holding one of the bind hash
 139                  * locks if this test triggers, because if 'remaining'
 140                  * drops to zero, we broke out of the do/while loop at
 141                  * the top level, not from the 'break;' statement.
 142                  */
 143                 ret = 1;
 144                 if (remaining <= 0) {
 145                         if (smallest_size != -1) {
 146                                 snum = smallest_rover;
 147                                 goto have_snum;
 148                         }
 149                         goto fail;
 150                 }
 151                 /* OK, here is the one we will use.  HEAD is
 152                  * non-NULL and we hold it's mutex.
 153                  */
 154                 snum = rover;
 155         } else {
 156 have_snum:
 157                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 158                                 hashinfo->bhash_size)];
 159                 spin_lock(&head->lock);
 160                 inet_bind_bucket_for_each(tb, node, &head->chain)
 161                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 162                                 goto tb_found;
 163         }
 164         tb = NULL;
 165         goto tb_not_found;
 166 tb_found:
 167         if (!hlist_empty(&tb->owners)) {
 168                 if (tb->fastreuse > 0 &&
 169                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 170                     smallest_size == -1) {
 171                         goto success;
 172                 } else {
 173                         ret = 1;
 174                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 175                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 176                                     smallest_size != -1 && --attempts >= 0) {
 177                                         spin_unlock(&head->lock);
 178                                         goto again;
 179                                 }
 180                                 goto fail_unlock;
 181                         }
 182                 }
 183         }
 184 tb_not_found:
 185         ret = 1;
 186         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 187                                         net, head, snum)) == NULL)
 188                 goto fail_unlock;
 189         if (hlist_empty(&tb->owners)) {
 190                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 191                         tb->fastreuse = 1;
 192                 else
 193                         tb->fastreuse = 0;
 194         } else if (tb->fastreuse &&
 195                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 196                 tb->fastreuse = 0;
 197 success:
 198         if (!inet_csk(sk)->icsk_bind_hash)
 199                 inet_bind_hash(sk, tb, snum);
 200         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 201         ret = 0;
 202
 203 fail_unlock:
 204         spin_unlock(&head->lock);
 205 fail:
 206         local_bh_enable();
 207         return ret;
 208 }
 209
 210 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 211
 212 /*
 213  * Wait for an incoming connection, avoid race conditions. This must be called
 214  * with the socket locked.
 215  */
 216 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 217 {
 218         struct inet_connection_sock *icsk = inet_csk(sk);
 219         DEFINE_WAIT(wait);
 220         int err;
 221
 222         /*
 223          * True wake-one mechanism for incoming connections: only
 224          * one process gets woken up, not the 'whole herd'.
 225          * Since we do not 'race & poll' for established sockets
 226          * anymore, the common case will execute the loop only once.
 227          *
 228          * Subtle issue: "add_wait_queue_exclusive()" will be added
 229          * after any current non-exclusive waiters, and we know that
 230          * it will always _stay_ after any new non-exclusive waiters
 231          * because all non-exclusive waiters are added at the
 232          * beginning of the wait-queue. As such, it's ok to "drop"
 233          * our exclusiveness temporarily when we get woken up without
 234          * having to remove and re-insert us on the wait queue.
 235          */
 236         for (;;) {
 237                 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
 238                                           TASK_INTERRUPTIBLE);
 239                 release_sock(sk);
 240                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 241                         timeo = schedule_timeout(timeo);
 242                 lock_sock(sk);
 243                 err = 0;
 244                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 245                         break;
 246                 err = -EINVAL;
 247                 if (sk->sk_state != TCP_LISTEN)
 248                         break;
 249                 err = sock_intr_errno(timeo);
 250                 if (signal_pending(current))
 251                         break;
 252                 err = -EAGAIN;
 253                 if (!timeo)
 254                         break;
 255         }
 256         finish_wait(sk->sk_sleep, &wait);
 257         return err;
 258 }
 259
 260 /*
 261  * This will accept the next outstanding connection.
 262  */
 263 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 264 {
 265         struct inet_connection_sock *icsk = inet_csk(sk);
 266         struct sock *newsk;
 267         int error;
 268
 269         lock_sock(sk);
 270
 271         /* We need to make sure that this socket is listening,
 272          * and that it has something pending.
 273          */
 274         error = -EINVAL;
 275         if (sk->sk_state != TCP_LISTEN)
 276                 goto out_err;
 277
 278         /* Find already established connection */
 279         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
 280                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 281
 282                 /* If this is a non blocking socket don't sleep */
 283                 error = -EAGAIN;
 284                 if (!timeo)
 285                         goto out_err;
 286
 287                 error = inet_csk_wait_for_connect(sk, timeo);
 288                 if (error)
 289                         goto out_err;
 290         }
 291
 292         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
 293         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
 294 out:
 295         release_sock(sk);
 296         return newsk;
 297 out_err:
 298         newsk = NULL;
 299         *err = error;
 300         goto out;
 301 }
 302
 303 EXPORT_SYMBOL(inet_csk_accept);
 304
 305 /*
 306  * Using different timers for retransmit, delayed acks and probes
 307  * We may wish use just one timer maintaining a list of expire jiffies
 308  * to optimize.
 309  */
 310 void inet_csk_init_xmit_timers(struct sock *sk,
 311                                void (*retransmit_handler)(unsigned long),
 312                                void (*delack_handler)(unsigned long),
 313                                void (*keepalive_handler)(unsigned long))
 314 {
 315         struct inet_connection_sock *icsk = inet_csk(sk);
 316
 317         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 318                         (unsigned long)sk);
 319         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 320                         (unsigned long)sk);
 321         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 322         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 323 }
 324
 325 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 326
 327 void inet_csk_clear_xmit_timers(struct sock *sk)
 328 {
 329         struct inet_connection_sock *icsk = inet_csk(sk);
 330
 331         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 332
 333         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 334         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 335         sk_stop_timer(sk, &sk->sk_timer);
 336 }
 337
 338 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 339
 340 void inet_csk_delete_keepalive_timer(struct sock *sk)
 341 {
 342         sk_stop_timer(sk, &sk->sk_timer);
 343 }
 344
 345 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 346
 347 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 348 {
 349         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 350 }
 351
 352 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 353
 354 struct dst_entry *inet_csk_route_req(struct sock *sk,
 355                                      const struct request_sock *req)
 356 {
 357         struct rtable *rt;
 358         const struct inet_request_sock *ireq = inet_rsk(req);
 359         struct ip_options *opt = inet_rsk(req)->opt;
 360         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 361                             .mark = sk->sk_mark,
 362                             .nl_u = { .ip4_u =
 363                                       { .daddr = ((opt && opt->srr) ?
 364                                                   opt->faddr :
 365                                                   ireq->rmt_addr),
 366                                         .saddr = ireq->loc_addr,
 367                                         .tos = RT_CONN_FLAGS(sk) } },
 368                             .proto = sk->sk_protocol,
 369                             .flags = inet_sk_flowi_flags(sk),
 370                             .uli_u = { .ports =
 371                                        { .sport = inet_sk(sk)->inet_sport,
 372                                          .dport = ireq->rmt_port } } };
 373         struct net *net = sock_net(sk);
 374
 375         security_req_classify_flow(req, &fl);
 376         if (ip_route_output_flow(net, &rt, &fl, sk, 0))
 377                 goto no_route;
 378         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 379                 goto route_err;
 380         return &rt->u.dst;
 381
 382 route_err:
 383         ip_rt_put(rt);
 384 no_route:
 385         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 386         return NULL;
 387 }
 388
 389 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 390
 391 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 392                                  const u32 rnd, const u32 synq_hsize)
 393 {
 394         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 395 }
 396
 397 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 398 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 399 #else
 400 #define AF_INET_FAMILY(fam) 1
 401 #endif
 402
 403 struct request_sock *inet_csk_search_req(const struct sock *sk,
 404                                          struct request_sock ***prevp,
 405                                          const __be16 rport, const __be32 raddr,
 406                                          const __be32 laddr)
 407 {
 408         const struct inet_connection_sock *icsk = inet_csk(sk);
 409         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 410         struct request_sock *req, **prev;
 411
 412         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 413                                                     lopt->nr_table_entries)];
 414              (req = *prev) != NULL;
 415              prev = &req->dl_next) {
 416                 const struct inet_request_sock *ireq = inet_rsk(req);
 417
 418                 if (ireq->rmt_port == rport &&
 419                     ireq->rmt_addr == raddr &&
 420                     ireq->loc_addr == laddr &&
 421                     AF_INET_FAMILY(req->rsk_ops->family)) {
 422                         WARN_ON(req->sk);
 423                         *prevp = prev;
 424                         break;
 425                 }
 426         }
 427
 428         return req;
 429 }
 430
 431 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 432
 433 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 434                                    unsigned long timeout)
 435 {
 436         struct inet_connection_sock *icsk = inet_csk(sk);
 437         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 438         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 439                                      lopt->hash_rnd, lopt->nr_table_entries);
 440
 441         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 442         inet_csk_reqsk_queue_added(sk, timeout);
 443 }
 444
 445 /* Only thing we need from tcp.h */
 446 extern int sysctl_tcp_synack_retries;
 447
 448 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 449
 450 /* Decide when to expire the request and when to resend SYN-ACK */
 451 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 452                                   const int max_retries,
 453                                   const u8 rskq_defer_accept,
 454                                   int *expire, int *resend)
 455 {
 456         if (!rskq_defer_accept) {
 457                 *expire = req->retrans >= thresh;
 458                 *resend = 1;
 459                 return;
 460         }
 461         *expire = req->retrans >= thresh &&
 462                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 463         /*
 464          * Do not resend while waiting for data after ACK,
 465          * start to resend on end of deferring period to give
 466          * last chance for data or ACK to create established socket.
 467          */
 468         *resend = !inet_rsk(req)->acked ||
 469                   req->retrans >= rskq_defer_accept - 1;
 470 }
 471
 472 void inet_csk_reqsk_queue_prune(struct sock *parent,
 473                                 const unsigned long interval,
 474                                 const unsigned long timeout,
 475                                 const unsigned long max_rto)
 476 {
 477         struct inet_connection_sock *icsk = inet_csk(parent);
 478         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 479         struct listen_sock *lopt = queue->listen_opt;
 480         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 481         int thresh = max_retries;
 482         unsigned long now = jiffies;
 483         struct request_sock **reqp, *req;
 484         int i, budget;
 485
 486         if (lopt == NULL || lopt->qlen == 0)
 487                 return;
 488
 489         /* Normally all the openreqs are young and become mature
 490          * (i.e. converted to established socket) for first timeout.
 491          * If synack was not acknowledged for 3 seconds, it means
 492          * one of the following things: synack was lost, ack was lost,
 493          * rtt is high or nobody planned to ack (i.e. synflood).
 494          * When server is a bit loaded, queue is populated with old
 495          * open requests, reducing effective size of queue.
 496          * When server is well loaded, queue size reduces to zero
 497          * after several minutes of work. It is not synflood,
 498          * it is normal operation. The solution is pruning
 499          * too old entries overriding normal timeout, when
 500          * situation becomes dangerous.
 501          *
 502          * Essentially, we reserve half of room for young
 503          * embrions; and abort old ones without pity, if old
 504          * ones are about to clog our table.
 505          */
 506         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 507                 int young = (lopt->qlen_young<<1);
 508
 509                 while (thresh > 2) {
 510                         if (lopt->qlen < young)
 511                                 break;
 512                         thresh--;
 513                         young <<= 1;
 514                 }
 515         }
 516
 517         if (queue->rskq_defer_accept)
 518                 max_retries = queue->rskq_defer_accept;
 519
 520         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 521         i = lopt->clock_hand;
 522
 523         do {
 524                 reqp=&lopt->syn_table[i];
 525                 while ((req = *reqp) != NULL) {
 526                         if (time_after_eq(now, req->expires)) {
 527                                 int expire = 0, resend = 0;
 528
 529                                 syn_ack_recalc(req, thresh, max_retries,
 530                                                queue->rskq_defer_accept,
 531                                                &expire, &resend);
 532                                 if (req->rsk_ops->syn_ack_timeout)
 533                                         req->rsk_ops->syn_ack_timeout(parent, req);
 534                                 if (!expire &&
 535                                     (!resend ||
 536                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 537                                      inet_rsk(req)->acked)) {
 538                                         unsigned long timeo;
 539
 540                                         if (req->retrans++ == 0)
 541                                                 lopt->qlen_young--;
 542                                         timeo = min((timeout << req->retrans), max_rto);
 543                                         req->expires = now + timeo;
 544                                         reqp = &req->dl_next;
 545                                         continue;
 546                                 }
 547
 548                                 /* Drop this request */
 549                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 550                                 reqsk_queue_removed(queue, req);
 551                                 reqsk_free(req);
 552                                 continue;
 553                         }
 554                         reqp = &req->dl_next;
 555                 }
 556
 557                 i = (i + 1) & (lopt->nr_table_entries - 1);
 558
 559         } while (--budget > 0);
 560
 561         lopt->clock_hand = i;
 562
 563         if (lopt->qlen)
 564                 inet_csk_reset_keepalive_timer(parent, interval);
 565 }
 566
 567 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 568
 569 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
 570                             const gfp_t priority)
 571 {
 572         struct sock *newsk = sk_clone(sk, priority);
 573
 574         if (newsk != NULL) {
 575                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 576
 577                 newsk->sk_state = TCP_SYN_RECV;
 578                 newicsk->icsk_bind_hash = NULL;
 579
 580                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 581                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 582                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 583                 newsk->sk_write_space = sk_stream_write_space;
 584
 585                 newicsk->icsk_retransmits = 0;
 586                 newicsk->icsk_backoff     = 0;
 587                 newicsk->icsk_probes_out  = 0;
 588
 589                 /* Deinitialize accept_queue to trap illegal accesses. */
 590                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 591
 592                 security_inet_csk_clone(newsk, req);
 593         }
 594         return newsk;
 595 }
 596
 597 EXPORT_SYMBOL_GPL(inet_csk_clone);
 598
 599 /*
 600  * At this point, there should be no process reference to this
 601  * socket, and thus no user references at all.  Therefore we
 602  * can assume the socket waitqueue is inactive and nobody will
 603  * try to jump onto it.
 604  */
 605 void inet_csk_destroy_sock(struct sock *sk)
 606 {
 607         WARN_ON(sk->sk_state != TCP_CLOSE);
 608         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 609
 610         /* It cannot be in hash table! */
 611         WARN_ON(!sk_unhashed(sk));
 612
 613         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 614         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 615
 616         sk->sk_prot->destroy(sk);
 617
 618         sk_stream_kill_queues(sk);
 619
 620         xfrm_sk_free_policy(sk);
 621
 622         sk_refcnt_debug_release(sk);
 623
 624         percpu_counter_dec(sk->sk_prot->orphan_count);
 625         sock_put(sk);
 626 }
 627
 628 EXPORT_SYMBOL(inet_csk_destroy_sock);
 629
 630 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 631 {
 632         struct inet_sock *inet = inet_sk(sk);
 633         struct inet_connection_sock *icsk = inet_csk(sk);
 634         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 635
 636         if (rc != 0)
 637                 return rc;
 638
 639         sk->sk_max_ack_backlog = 0;
 640         sk->sk_ack_backlog = 0;
 641         inet_csk_delack_init(sk);
 642
 643         /* There is race window here: we announce ourselves listening,
 644          * but this transition is still not validated by get_port().
 645          * It is OK, because this socket enters to hash table only
 646          * after validation is complete.
 647          */
 648         sk->sk_state = TCP_LISTEN;
 649         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 650                 inet->inet_sport = htons(inet->inet_num);
 651
 652                 sk_dst_reset(sk);
 653                 sk->sk_prot->hash(sk);
 654
 655                 return 0;
 656         }
 657
 658         sk->sk_state = TCP_CLOSE;
 659         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 660         return -EADDRINUSE;
 661 }
 662
 663 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 664
 665 /*
 666  *      This routine closes sockets which have been at least partially
 667  *      opened, but not yet accepted.
 668  */
 669 void inet_csk_listen_stop(struct sock *sk)
 670 {
 671         struct inet_connection_sock *icsk = inet_csk(sk);
 672         struct request_sock *acc_req;
 673         struct request_sock *req;
 674
 675         inet_csk_delete_keepalive_timer(sk);
 676
 677         /* make all the listen_opt local to us */
 678         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 679
 680         /* Following specs, it would be better either to send FIN
 681          * (and enter FIN-WAIT-1, it is normal close)
 682          * or to send active reset (abort).
 683          * Certainly, it is pretty dangerous while synflood, but it is
 684          * bad justification for our negligence 8)
 685          * To be honest, we are not able to make either
 686          * of the variants now.                 --ANK
 687          */
 688         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 689
 690         while ((req = acc_req) != NULL) {
 691                 struct sock *child = req->sk;
 692
 693                 acc_req = req->dl_next;
 694
 695                 local_bh_disable();
 696                 bh_lock_sock(child);
 697                 WARN_ON(sock_owned_by_user(child));
 698                 sock_hold(child);
 699
 700                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 701
 702                 sock_orphan(child);
 703
 704                 percpu_counter_inc(sk->sk_prot->orphan_count);
 705
 706                 inet_csk_destroy_sock(child);
 707
 708                 bh_unlock_sock(child);
 709                 local_bh_enable();
 710                 sock_put(child);
 711
 712                 sk_acceptq_removed(sk);
 713                 __reqsk_free(req);
 714         }
 715         WARN_ON(sk->sk_ack_backlog);
 716 }
 717
 718 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 719
 720 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 721 {
 722         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 723         const struct inet_sock *inet = inet_sk(sk);
 724
 725         sin->sin_family         = AF_INET;
 726         sin->sin_addr.s_addr    = inet->inet_daddr;
 727         sin->sin_port           = inet->inet_dport;
 728 }
 729
 730 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 731
 732 #ifdef CONFIG_COMPAT
 733 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 734                                char __user *optval, int __user *optlen)
 735 {
 736         const struct inet_connection_sock *icsk = inet_csk(sk);
 737
 738         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 739                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 740                                                             optval, optlen);
 741         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 742                                              optval, optlen);
 743 }
 744
 745 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 746
 747 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 748                                char __user *optval, unsigned int optlen)
 749 {
 750         const struct inet_connection_sock *icsk = inet_csk(sk);
 751
 752         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 753                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 754                                                             optval, optlen);
 755         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 756                                              optval, optlen);
 757 }
 758
 759 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 760 #endif