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         struct sock *sk2;
  59         struct hlist_node *node;
  60         int reuse = sk->sk_reuse;
  61
  62         /*
  63          * Unlike other sk lookup places we do not check
  64          * for sk_net here, since _all_ the socks listed
  65          * in tb->owners list belong to the same net - the
  66          * one this bucket belongs to.
  67          */
  68
  69         sk_for_each_bound(sk2, node, &tb->owners) {
  70                 if (sk != sk2 &&
  71                     !inet_v6_ipv6only(sk2) &&
  72                     (!sk->sk_bound_dev_if ||
  73                      !sk2->sk_bound_dev_if ||
  74                      sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
  75                         if (!reuse || !sk2->sk_reuse ||
  76                             ((1 << sk2->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
  77                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
  78                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  79                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  80                                         break;
  81                         }
  82                 }
  83         }
  84         return node != NULL;
  85 }
  86 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  87
  88 /* Obtain a reference to a local port for the given sock,
  89  * if snum is zero it means select any available local port.
  90  */
  91 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  92 {
  93         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  94         struct inet_bind_hashbucket *head;
  95         struct hlist_node *node;
  96         struct inet_bind_bucket *tb;
  97         int ret, attempts = 5;
  98         struct net *net = sock_net(sk);
  99         int smallest_size = -1, smallest_rover;
 100
 101         local_bh_disable();
 102         if (!snum) {
 103                 int remaining, rover, low, high;
 104
 105 again:
 106                 inet_get_local_port_range(&low, &high);
 107                 remaining = (high - low) + 1;
 108                 smallest_rover = rover = net_random() % remaining + low;
 109
 110                 smallest_size = -1;
 111                 do {
 112                         if (inet_is_reserved_local_port(rover))
 113                                 goto next_nolock;
 114                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 115                                         hashinfo->bhash_size)];
 116                         spin_lock(&head->lock);
 117                         inet_bind_bucket_for_each(tb, node, &head->chain)
 118                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 119                                         if (tb->fastreuse > 0 &&
 120                                             sk->sk_reuse &&
 121                                             sk->sk_state != TCP_LISTEN &&
 122                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 123                                                 smallest_size = tb->num_owners;
 124                                                 smallest_rover = rover;
 125                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
 126                                                     !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 127                                                         spin_unlock(&head->lock);
 128                                                         snum = smallest_rover;
 129                                                         goto have_snum;
 130                                                 }
 131                                         }
 132                                         goto next;
 133                                 }
 134                         break;
 135                 next:
 136                         spin_unlock(&head->lock);
 137                 next_nolock:
 138                         if (++rover > high)
 139                                 rover = low;
 140                 } while (--remaining > 0);
 141
 142                 /* Exhausted local port range during search?  It is not
 143                  * possible for us to be holding one of the bind hash
 144                  * locks if this test triggers, because if 'remaining'
 145                  * drops to zero, we broke out of the do/while loop at
 146                  * the top level, not from the 'break;' statement.
 147                  */
 148                 ret = 1;
 149                 if (remaining <= 0) {
 150                         if (smallest_size != -1) {
 151                                 snum = smallest_rover;
 152                                 goto have_snum;
 153                         }
 154                         goto fail;
 155                 }
 156                 /* OK, here is the one we will use.  HEAD is
 157                  * non-NULL and we hold it's mutex.
 158                  */
 159                 snum = rover;
 160         } else {
 161 have_snum:
 162                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 163                                 hashinfo->bhash_size)];
 164                 spin_lock(&head->lock);
 165                 inet_bind_bucket_for_each(tb, node, &head->chain)
 166                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 167                                 goto tb_found;
 168         }
 169         tb = NULL;
 170         goto tb_not_found;
 171 tb_found:
 172         if (!hlist_empty(&tb->owners)) {
 173                 if (tb->fastreuse > 0 &&
 174                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 175                     smallest_size == -1) {
 176                         goto success;
 177                 } else {
 178                         ret = 1;
 179                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
 180                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 181                                     smallest_size != -1 && --attempts >= 0) {
 182                                         spin_unlock(&head->lock);
 183                                         goto again;
 184                                 }
 185                                 goto fail_unlock;
 186                         }
 187                 }
 188         }
 189 tb_not_found:
 190         ret = 1;
 191         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 192                                         net, head, snum)) == NULL)
 193                 goto fail_unlock;
 194         if (hlist_empty(&tb->owners)) {
 195                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 196                         tb->fastreuse = 1;
 197                 else
 198                         tb->fastreuse = 0;
 199         } else if (tb->fastreuse &&
 200                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 201                 tb->fastreuse = 0;
 202 success:
 203         if (!inet_csk(sk)->icsk_bind_hash)
 204                 inet_bind_hash(sk, tb, snum);
 205         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 206         ret = 0;
 207
 208 fail_unlock:
 209         spin_unlock(&head->lock);
 210 fail:
 211         local_bh_enable();
 212         return ret;
 213 }
 214 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 215
 216 /*
 217  * Wait for an incoming connection, avoid race conditions. This must be called
 218  * with the socket locked.
 219  */
 220 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 221 {
 222         struct inet_connection_sock *icsk = inet_csk(sk);
 223         DEFINE_WAIT(wait);
 224         int err;
 225
 226         /*
 227          * True wake-one mechanism for incoming connections: only
 228          * one process gets woken up, not the 'whole herd'.
 229          * Since we do not 'race & poll' for established sockets
 230          * anymore, the common case will execute the loop only once.
 231          *
 232          * Subtle issue: "add_wait_queue_exclusive()" will be added
 233          * after any current non-exclusive waiters, and we know that
 234          * it will always _stay_ after any new non-exclusive waiters
 235          * because all non-exclusive waiters are added at the
 236          * beginning of the wait-queue. As such, it's ok to "drop"
 237          * our exclusiveness temporarily when we get woken up without
 238          * having to remove and re-insert us on the wait queue.
 239          */
 240         for (;;) {
 241                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 242                                           TASK_INTERRUPTIBLE);
 243                 release_sock(sk);
 244                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 245                         timeo = schedule_timeout(timeo);
 246                 lock_sock(sk);
 247                 err = 0;
 248                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 249                         break;
 250                 err = -EINVAL;
 251                 if (sk->sk_state != TCP_LISTEN)
 252                         break;
 253                 err = sock_intr_errno(timeo);
 254                 if (signal_pending(current))
 255                         break;
 256                 err = -EAGAIN;
 257                 if (!timeo)
 258                         break;
 259         }
 260         finish_wait(sk_sleep(sk), &wait);
 261         return err;
 262 }
 263
 264 /*
 265  * This will accept the next outstanding connection.
 266  */
 267 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 268 {
 269         struct inet_connection_sock *icsk = inet_csk(sk);
 270         struct sock *newsk;
 271         int error;
 272
 273         lock_sock(sk);
 274
 275         /* We need to make sure that this socket is listening,
 276          * and that it has something pending.
 277          */
 278         error = -EINVAL;
 279         if (sk->sk_state != TCP_LISTEN)
 280                 goto out_err;
 281
 282         /* Find already established connection */
 283         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
 284                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 285
 286                 /* If this is a non blocking socket don't sleep */
 287                 error = -EAGAIN;
 288                 if (!timeo)
 289                         goto out_err;
 290
 291                 error = inet_csk_wait_for_connect(sk, timeo);
 292                 if (error)
 293                         goto out_err;
 294         }
 295
 296         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
 297         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
 298 out:
 299         release_sock(sk);
 300         return newsk;
 301 out_err:
 302         newsk = NULL;
 303         *err = error;
 304         goto out;
 305 }
 306 EXPORT_SYMBOL(inet_csk_accept);
 307
 308 /*
 309  * Using different timers for retransmit, delayed acks and probes
 310  * We may wish use just one timer maintaining a list of expire jiffies
 311  * to optimize.
 312  */
 313 void inet_csk_init_xmit_timers(struct sock *sk,
 314                                void (*retransmit_handler)(unsigned long),
 315                                void (*delack_handler)(unsigned long),
 316                                void (*keepalive_handler)(unsigned long))
 317 {
 318         struct inet_connection_sock *icsk = inet_csk(sk);
 319
 320         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 321                         (unsigned long)sk);
 322         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 323                         (unsigned long)sk);
 324         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 325         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 326 }
 327 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 328
 329 void inet_csk_clear_xmit_timers(struct sock *sk)
 330 {
 331         struct inet_connection_sock *icsk = inet_csk(sk);
 332
 333         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 334
 335         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 336         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 337         sk_stop_timer(sk, &sk->sk_timer);
 338 }
 339 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 340
 341 void inet_csk_delete_keepalive_timer(struct sock *sk)
 342 {
 343         sk_stop_timer(sk, &sk->sk_timer);
 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 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 352
 353 struct dst_entry *inet_csk_route_req(struct sock *sk,
 354                                      const struct request_sock *req)
 355 {
 356         struct rtable *rt;
 357         const struct inet_request_sock *ireq = inet_rsk(req);
 358         struct ip_options *opt = inet_rsk(req)->opt;
 359         struct flowi fl = { .oif = sk->sk_bound_dev_if,
 360                             .mark = sk->sk_mark,
 361                             .fl4_dst = ((opt && opt->srr) ?
 362                                           opt->faddr : ireq->rmt_addr),
 363                             .fl4_src = ireq->loc_addr,
 364                             .fl4_tos = RT_CONN_FLAGS(sk),
 365                             .proto = sk->sk_protocol,
 366                             .flags = inet_sk_flowi_flags(sk),
 367                             .fl_ip_sport = inet_sk(sk)->inet_sport,
 368                             .fl_ip_dport = ireq->rmt_port };
 369         struct net *net = sock_net(sk);
 370
 371         security_req_classify_flow(req, &fl);
 372         rt = ip_route_output_flow(net, &fl, sk);
 373         if (IS_ERR(rt))
 374                 goto no_route;
 375         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 376                 goto route_err;
 377         return &rt->dst;
 378
 379 route_err:
 380         ip_rt_put(rt);
 381 no_route:
 382         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 383         return NULL;
 384 }
 385 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 386
 387 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 388                                  const u32 rnd, const u32 synq_hsize)
 389 {
 390         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 391 }
 392
 393 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 394 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 395 #else
 396 #define AF_INET_FAMILY(fam) 1
 397 #endif
 398
 399 struct request_sock *inet_csk_search_req(const struct sock *sk,
 400                                          struct request_sock ***prevp,
 401                                          const __be16 rport, const __be32 raddr,
 402                                          const __be32 laddr)
 403 {
 404         const struct inet_connection_sock *icsk = inet_csk(sk);
 405         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 406         struct request_sock *req, **prev;
 407
 408         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 409                                                     lopt->nr_table_entries)];
 410              (req = *prev) != NULL;
 411              prev = &req->dl_next) {
 412                 const struct inet_request_sock *ireq = inet_rsk(req);
 413
 414                 if (ireq->rmt_port == rport &&
 415                     ireq->rmt_addr == raddr &&
 416                     ireq->loc_addr == laddr &&
 417                     AF_INET_FAMILY(req->rsk_ops->family)) {
 418                         WARN_ON(req->sk);
 419                         *prevp = prev;
 420                         break;
 421                 }
 422         }
 423
 424         return req;
 425 }
 426 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 427
 428 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 429                                    unsigned long timeout)
 430 {
 431         struct inet_connection_sock *icsk = inet_csk(sk);
 432         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 433         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 434                                      lopt->hash_rnd, lopt->nr_table_entries);
 435
 436         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 437         inet_csk_reqsk_queue_added(sk, timeout);
 438 }
 439 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 440
 441 /* Only thing we need from tcp.h */
 442 extern int sysctl_tcp_synack_retries;
 443
 444
 445 /* Decide when to expire the request and when to resend SYN-ACK */
 446 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 447                                   const int max_retries,
 448                                   const u8 rskq_defer_accept,
 449                                   int *expire, int *resend)
 450 {
 451         if (!rskq_defer_accept) {
 452                 *expire = req->retrans >= thresh;
 453                 *resend = 1;
 454                 return;
 455         }
 456         *expire = req->retrans >= thresh &&
 457                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 458         /*
 459          * Do not resend while waiting for data after ACK,
 460          * start to resend on end of deferring period to give
 461          * last chance for data or ACK to create established socket.
 462          */
 463         *resend = !inet_rsk(req)->acked ||
 464                   req->retrans >= rskq_defer_accept - 1;
 465 }
 466
 467 void inet_csk_reqsk_queue_prune(struct sock *parent,
 468                                 const unsigned long interval,
 469                                 const unsigned long timeout,
 470                                 const unsigned long max_rto)
 471 {
 472         struct inet_connection_sock *icsk = inet_csk(parent);
 473         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 474         struct listen_sock *lopt = queue->listen_opt;
 475         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 476         int thresh = max_retries;
 477         unsigned long now = jiffies;
 478         struct request_sock **reqp, *req;
 479         int i, budget;
 480
 481         if (lopt == NULL || lopt->qlen == 0)
 482                 return;
 483
 484         /* Normally all the openreqs are young and become mature
 485          * (i.e. converted to established socket) for first timeout.
 486          * If synack was not acknowledged for 3 seconds, it means
 487          * one of the following things: synack was lost, ack was lost,
 488          * rtt is high or nobody planned to ack (i.e. synflood).
 489          * When server is a bit loaded, queue is populated with old
 490          * open requests, reducing effective size of queue.
 491          * When server is well loaded, queue size reduces to zero
 492          * after several minutes of work. It is not synflood,
 493          * it is normal operation. The solution is pruning
 494          * too old entries overriding normal timeout, when
 495          * situation becomes dangerous.
 496          *
 497          * Essentially, we reserve half of room for young
 498          * embrions; and abort old ones without pity, if old
 499          * ones are about to clog our table.
 500          */
 501         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 502                 int young = (lopt->qlen_young<<1);
 503
 504                 while (thresh > 2) {
 505                         if (lopt->qlen < young)
 506                                 break;
 507                         thresh--;
 508                         young <<= 1;
 509                 }
 510         }
 511
 512         if (queue->rskq_defer_accept)
 513                 max_retries = queue->rskq_defer_accept;
 514
 515         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 516         i = lopt->clock_hand;
 517
 518         do {
 519                 reqp=&lopt->syn_table[i];
 520                 while ((req = *reqp) != NULL) {
 521                         if (time_after_eq(now, req->expires)) {
 522                                 int expire = 0, resend = 0;
 523
 524                                 syn_ack_recalc(req, thresh, max_retries,
 525                                                queue->rskq_defer_accept,
 526                                                &expire, &resend);
 527                                 if (req->rsk_ops->syn_ack_timeout)
 528                                         req->rsk_ops->syn_ack_timeout(parent, req);
 529                                 if (!expire &&
 530                                     (!resend ||
 531                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 532                                      inet_rsk(req)->acked)) {
 533                                         unsigned long timeo;
 534
 535                                         if (req->retrans++ == 0)
 536                                                 lopt->qlen_young--;
 537                                         timeo = min((timeout << req->retrans), max_rto);
 538                                         req->expires = now + timeo;
 539                                         reqp = &req->dl_next;
 540                                         continue;
 541                                 }
 542
 543                                 /* Drop this request */
 544                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 545                                 reqsk_queue_removed(queue, req);
 546                                 reqsk_free(req);
 547                                 continue;
 548                         }
 549                         reqp = &req->dl_next;
 550                 }
 551
 552                 i = (i + 1) & (lopt->nr_table_entries - 1);
 553
 554         } while (--budget > 0);
 555
 556         lopt->clock_hand = i;
 557
 558         if (lopt->qlen)
 559                 inet_csk_reset_keepalive_timer(parent, interval);
 560 }
 561 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 562
 563 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
 564                             const gfp_t priority)
 565 {
 566         struct sock *newsk = sk_clone(sk, priority);
 567
 568         if (newsk != NULL) {
 569                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 570
 571                 newsk->sk_state = TCP_SYN_RECV;
 572                 newicsk->icsk_bind_hash = NULL;
 573
 574                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 575                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 576                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 577                 newsk->sk_write_space = sk_stream_write_space;
 578
 579                 newicsk->icsk_retransmits = 0;
 580                 newicsk->icsk_backoff     = 0;
 581                 newicsk->icsk_probes_out  = 0;
 582
 583                 /* Deinitialize accept_queue to trap illegal accesses. */
 584                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 585
 586                 security_inet_csk_clone(newsk, req);
 587         }
 588         return newsk;
 589 }
 590 EXPORT_SYMBOL_GPL(inet_csk_clone);
 591
 592 /*
 593  * At this point, there should be no process reference to this
 594  * socket, and thus no user references at all.  Therefore we
 595  * can assume the socket waitqueue is inactive and nobody will
 596  * try to jump onto it.
 597  */
 598 void inet_csk_destroy_sock(struct sock *sk)
 599 {
 600         WARN_ON(sk->sk_state != TCP_CLOSE);
 601         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 602
 603         /* It cannot be in hash table! */
 604         WARN_ON(!sk_unhashed(sk));
 605
 606         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 607         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 608
 609         sk->sk_prot->destroy(sk);
 610
 611         sk_stream_kill_queues(sk);
 612
 613         xfrm_sk_free_policy(sk);
 614
 615         sk_refcnt_debug_release(sk);
 616
 617         percpu_counter_dec(sk->sk_prot->orphan_count);
 618         sock_put(sk);
 619 }
 620 EXPORT_SYMBOL(inet_csk_destroy_sock);
 621
 622 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 623 {
 624         struct inet_sock *inet = inet_sk(sk);
 625         struct inet_connection_sock *icsk = inet_csk(sk);
 626         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 627
 628         if (rc != 0)
 629                 return rc;
 630
 631         sk->sk_max_ack_backlog = 0;
 632         sk->sk_ack_backlog = 0;
 633         inet_csk_delack_init(sk);
 634
 635         /* There is race window here: we announce ourselves listening,
 636          * but this transition is still not validated by get_port().
 637          * It is OK, because this socket enters to hash table only
 638          * after validation is complete.
 639          */
 640         sk->sk_state = TCP_LISTEN;
 641         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 642                 inet->inet_sport = htons(inet->inet_num);
 643
 644                 sk_dst_reset(sk);
 645                 sk->sk_prot->hash(sk);
 646
 647                 return 0;
 648         }
 649
 650         sk->sk_state = TCP_CLOSE;
 651         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 652         return -EADDRINUSE;
 653 }
 654 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 655
 656 /*
 657  *      This routine closes sockets which have been at least partially
 658  *      opened, but not yet accepted.
 659  */
 660 void inet_csk_listen_stop(struct sock *sk)
 661 {
 662         struct inet_connection_sock *icsk = inet_csk(sk);
 663         struct request_sock *acc_req;
 664         struct request_sock *req;
 665
 666         inet_csk_delete_keepalive_timer(sk);
 667
 668         /* make all the listen_opt local to us */
 669         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 670
 671         /* Following specs, it would be better either to send FIN
 672          * (and enter FIN-WAIT-1, it is normal close)
 673          * or to send active reset (abort).
 674          * Certainly, it is pretty dangerous while synflood, but it is
 675          * bad justification for our negligence 8)
 676          * To be honest, we are not able to make either
 677          * of the variants now.                 --ANK
 678          */
 679         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 680
 681         while ((req = acc_req) != NULL) {
 682                 struct sock *child = req->sk;
 683
 684                 acc_req = req->dl_next;
 685
 686                 local_bh_disable();
 687                 bh_lock_sock(child);
 688                 WARN_ON(sock_owned_by_user(child));
 689                 sock_hold(child);
 690
 691                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 692
 693                 sock_orphan(child);
 694
 695                 percpu_counter_inc(sk->sk_prot->orphan_count);
 696
 697                 inet_csk_destroy_sock(child);
 698
 699                 bh_unlock_sock(child);
 700                 local_bh_enable();
 701                 sock_put(child);
 702
 703                 sk_acceptq_removed(sk);
 704                 __reqsk_free(req);
 705         }
 706         WARN_ON(sk->sk_ack_backlog);
 707 }
 708 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 709
 710 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 711 {
 712         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 713         const struct inet_sock *inet = inet_sk(sk);
 714
 715         sin->sin_family         = AF_INET;
 716         sin->sin_addr.s_addr    = inet->inet_daddr;
 717         sin->sin_port           = inet->inet_dport;
 718 }
 719 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 720
 721 #ifdef CONFIG_COMPAT
 722 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 723                                char __user *optval, int __user *optlen)
 724 {
 725         const struct inet_connection_sock *icsk = inet_csk(sk);
 726
 727         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 728                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 729                                                             optval, optlen);
 730         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 731                                              optval, optlen);
 732 }
 733 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 734
 735 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 736                                char __user *optval, unsigned int optlen)
 737 {
 738         const struct inet_connection_sock *icsk = inet_csk(sk);
 739
 740         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 741                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 742                                                             optval, optlen);
 743         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 744                                              optval, optlen);
 745 }
 746 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 747 #endif