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 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  88
  89 /* Obtain a reference to a local port for the given sock,
  90  * if snum is zero it means select any available local port.
  91  */
  92 int inet_csk_get_port(struct sock *sk, unsigned short snum)
  93 {
  94         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
  95         struct inet_bind_hashbucket *head;
  96         struct hlist_node *node;
  97         struct inet_bind_bucket *tb;
  98         int ret, attempts = 5;
  99         struct net *net = sock_net(sk);
 100         int smallest_size = -1, smallest_rover;
 101
 102         local_bh_disable();
 103         if (!snum) {
 104                 int remaining, rover, low, high;
 105
 106 again:
 107                 inet_get_local_port_range(&low, &high);
 108                 remaining = (high - low) + 1;
 109                 smallest_rover = rover = net_random() % remaining + low;
 110
 111                 smallest_size = -1;
 112                 do {
 113                         if (inet_is_reserved_local_port(rover))
 114                                 goto next_nolock;
 115                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 116                                         hashinfo->bhash_size)];
 117                         spin_lock(&head->lock);
 118                         inet_bind_bucket_for_each(tb, node, &head->chain)
 119                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 120                                         if (tb->fastreuse > 0 &&
 121                                             sk->sk_reuse &&
 122                                             sk->sk_state != TCP_LISTEN &&
 123                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 124                                                 smallest_size = tb->num_owners;
 125                                                 smallest_rover = rover;
 126                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
 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                             .nl_u = { .ip4_u =
 362                                       { .daddr = ((opt && opt->srr) ?
 363                                                   opt->faddr :
 364                                                   ireq->rmt_addr),
 365                                         .saddr = ireq->loc_addr,
 366                                         .tos = RT_CONN_FLAGS(sk) } },
 367                             .proto = sk->sk_protocol,
 368                             .flags = inet_sk_flowi_flags(sk),
 369                             .uli_u = { .ports =
 370                                        { .sport = inet_sk(sk)->inet_sport,
 371                                          .dport = ireq->rmt_port } } };
 372         struct net *net = sock_net(sk);
 373
 374         security_req_classify_flow(req, &fl);
 375         if (ip_route_output_flow(net, &rt, &fl, sk, 0))
 376                 goto no_route;
 377         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
 378                 goto route_err;
 379         return &rt->dst;
 380
 381 route_err:
 382         ip_rt_put(rt);
 383 no_route:
 384         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 385         return NULL;
 386 }
 387 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 388
 389 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 390                                  const u32 rnd, const u32 synq_hsize)
 391 {
 392         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 393 }
 394
 395 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 396 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 397 #else
 398 #define AF_INET_FAMILY(fam) 1
 399 #endif
 400
 401 struct request_sock *inet_csk_search_req(const struct sock *sk,
 402                                          struct request_sock ***prevp,
 403                                          const __be16 rport, const __be32 raddr,
 404                                          const __be32 laddr)
 405 {
 406         const struct inet_connection_sock *icsk = inet_csk(sk);
 407         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 408         struct request_sock *req, **prev;
 409
 410         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 411                                                     lopt->nr_table_entries)];
 412              (req = *prev) != NULL;
 413              prev = &req->dl_next) {
 414                 const struct inet_request_sock *ireq = inet_rsk(req);
 415
 416                 if (ireq->rmt_port == rport &&
 417                     ireq->rmt_addr == raddr &&
 418                     ireq->loc_addr == laddr &&
 419                     AF_INET_FAMILY(req->rsk_ops->family)) {
 420                         WARN_ON(req->sk);
 421                         *prevp = prev;
 422                         break;
 423                 }
 424         }
 425
 426         return req;
 427 }
 428 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 429
 430 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 431                                    unsigned long timeout)
 432 {
 433         struct inet_connection_sock *icsk = inet_csk(sk);
 434         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 435         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 436                                      lopt->hash_rnd, lopt->nr_table_entries);
 437
 438         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 439         inet_csk_reqsk_queue_added(sk, timeout);
 440 }
 441 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 442
 443 /* Only thing we need from tcp.h */
 444 extern int sysctl_tcp_synack_retries;
 445
 446
 447 /* Decide when to expire the request and when to resend SYN-ACK */
 448 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 449                                   const int max_retries,
 450                                   const u8 rskq_defer_accept,
 451                                   int *expire, int *resend)
 452 {
 453         if (!rskq_defer_accept) {
 454                 *expire = req->retrans >= thresh;
 455                 *resend = 1;
 456                 return;
 457         }
 458         *expire = req->retrans >= thresh &&
 459                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 460         /*
 461          * Do not resend while waiting for data after ACK,
 462          * start to resend on end of deferring period to give
 463          * last chance for data or ACK to create established socket.
 464          */
 465         *resend = !inet_rsk(req)->acked ||
 466                   req->retrans >= rskq_defer_accept - 1;
 467 }
 468
 469 void inet_csk_reqsk_queue_prune(struct sock *parent,
 470                                 const unsigned long interval,
 471                                 const unsigned long timeout,
 472                                 const unsigned long max_rto)
 473 {
 474         struct inet_connection_sock *icsk = inet_csk(parent);
 475         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 476         struct listen_sock *lopt = queue->listen_opt;
 477         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 478         int thresh = max_retries;
 479         unsigned long now = jiffies;
 480         struct request_sock **reqp, *req;
 481         int i, budget;
 482
 483         if (lopt == NULL || lopt->qlen == 0)
 484                 return;
 485
 486         /* Normally all the openreqs are young and become mature
 487          * (i.e. converted to established socket) for first timeout.
 488          * If synack was not acknowledged for 3 seconds, it means
 489          * one of the following things: synack was lost, ack was lost,
 490          * rtt is high or nobody planned to ack (i.e. synflood).
 491          * When server is a bit loaded, queue is populated with old
 492          * open requests, reducing effective size of queue.
 493          * When server is well loaded, queue size reduces to zero
 494          * after several minutes of work. It is not synflood,
 495          * it is normal operation. The solution is pruning
 496          * too old entries overriding normal timeout, when
 497          * situation becomes dangerous.
 498          *
 499          * Essentially, we reserve half of room for young
 500          * embrions; and abort old ones without pity, if old
 501          * ones are about to clog our table.
 502          */
 503         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 504                 int young = (lopt->qlen_young<<1);
 505
 506                 while (thresh > 2) {
 507                         if (lopt->qlen < young)
 508                                 break;
 509                         thresh--;
 510                         young <<= 1;
 511                 }
 512         }
 513
 514         if (queue->rskq_defer_accept)
 515                 max_retries = queue->rskq_defer_accept;
 516
 517         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 518         i = lopt->clock_hand;
 519
 520         do {
 521                 reqp=&lopt->syn_table[i];
 522                 while ((req = *reqp) != NULL) {
 523                         if (time_after_eq(now, req->expires)) {
 524                                 int expire = 0, resend = 0;
 525
 526                                 syn_ack_recalc(req, thresh, max_retries,
 527                                                queue->rskq_defer_accept,
 528                                                &expire, &resend);
 529                                 if (req->rsk_ops->syn_ack_timeout)
 530                                         req->rsk_ops->syn_ack_timeout(parent, req);
 531                                 if (!expire &&
 532                                     (!resend ||
 533                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 534                                      inet_rsk(req)->acked)) {
 535                                         unsigned long timeo;
 536
 537                                         if (req->retrans++ == 0)
 538                                                 lopt->qlen_young--;
 539                                         timeo = min((timeout << req->retrans), max_rto);
 540                                         req->expires = now + timeo;
 541                                         reqp = &req->dl_next;
 542                                         continue;
 543                                 }
 544
 545                                 /* Drop this request */
 546                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 547                                 reqsk_queue_removed(queue, req);
 548                                 reqsk_free(req);
 549                                 continue;
 550                         }
 551                         reqp = &req->dl_next;
 552                 }
 553
 554                 i = (i + 1) & (lopt->nr_table_entries - 1);
 555
 556         } while (--budget > 0);
 557
 558         lopt->clock_hand = i;
 559
 560         if (lopt->qlen)
 561                 inet_csk_reset_keepalive_timer(parent, interval);
 562 }
 563 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 564
 565 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
 566                             const gfp_t priority)
 567 {
 568         struct sock *newsk = sk_clone(sk, priority);
 569
 570         if (newsk != NULL) {
 571                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 572
 573                 newsk->sk_state = TCP_SYN_RECV;
 574                 newicsk->icsk_bind_hash = NULL;
 575
 576                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 577                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 578                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 579                 newsk->sk_write_space = sk_stream_write_space;
 580
 581                 newicsk->icsk_retransmits = 0;
 582                 newicsk->icsk_backoff     = 0;
 583                 newicsk->icsk_probes_out  = 0;
 584
 585                 /* Deinitialize accept_queue to trap illegal accesses. */
 586                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 587
 588                 security_inet_csk_clone(newsk, req);
 589         }
 590         return newsk;
 591 }
 592 EXPORT_SYMBOL_GPL(inet_csk_clone);
 593
 594 /*
 595  * At this point, there should be no process reference to this
 596  * socket, and thus no user references at all.  Therefore we
 597  * can assume the socket waitqueue is inactive and nobody will
 598  * try to jump onto it.
 599  */
 600 void inet_csk_destroy_sock(struct sock *sk)
 601 {
 602         WARN_ON(sk->sk_state != TCP_CLOSE);
 603         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 604
 605         /* It cannot be in hash table! */
 606         WARN_ON(!sk_unhashed(sk));
 607
 608         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 609         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 610
 611         sk->sk_prot->destroy(sk);
 612
 613         sk_stream_kill_queues(sk);
 614
 615         xfrm_sk_free_policy(sk);
 616
 617         sk_refcnt_debug_release(sk);
 618
 619         percpu_counter_dec(sk->sk_prot->orphan_count);
 620         sock_put(sk);
 621 }
 622 EXPORT_SYMBOL(inet_csk_destroy_sock);
 623
 624 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 625 {
 626         struct inet_sock *inet = inet_sk(sk);
 627         struct inet_connection_sock *icsk = inet_csk(sk);
 628         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 629
 630         if (rc != 0)
 631                 return rc;
 632
 633         sk->sk_max_ack_backlog = 0;
 634         sk->sk_ack_backlog = 0;
 635         inet_csk_delack_init(sk);
 636
 637         /* There is race window here: we announce ourselves listening,
 638          * but this transition is still not validated by get_port().
 639          * It is OK, because this socket enters to hash table only
 640          * after validation is complete.
 641          */
 642         sk->sk_state = TCP_LISTEN;
 643         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 644                 inet->inet_sport = htons(inet->inet_num);
 645
 646                 sk_dst_reset(sk);
 647                 sk->sk_prot->hash(sk);
 648
 649                 return 0;
 650         }
 651
 652         sk->sk_state = TCP_CLOSE;
 653         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 654         return -EADDRINUSE;
 655 }
 656 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 657
 658 /*
 659  *      This routine closes sockets which have been at least partially
 660  *      opened, but not yet accepted.
 661  */
 662 void inet_csk_listen_stop(struct sock *sk)
 663 {
 664         struct inet_connection_sock *icsk = inet_csk(sk);
 665         struct request_sock *acc_req;
 666         struct request_sock *req;
 667
 668         inet_csk_delete_keepalive_timer(sk);
 669
 670         /* make all the listen_opt local to us */
 671         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 672
 673         /* Following specs, it would be better either to send FIN
 674          * (and enter FIN-WAIT-1, it is normal close)
 675          * or to send active reset (abort).
 676          * Certainly, it is pretty dangerous while synflood, but it is
 677          * bad justification for our negligence 8)
 678          * To be honest, we are not able to make either
 679          * of the variants now.                 --ANK
 680          */
 681         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 682
 683         while ((req = acc_req) != NULL) {
 684                 struct sock *child = req->sk;
 685
 686                 acc_req = req->dl_next;
 687
 688                 local_bh_disable();
 689                 bh_lock_sock(child);
 690                 WARN_ON(sock_owned_by_user(child));
 691                 sock_hold(child);
 692
 693                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 694
 695                 sock_orphan(child);
 696
 697                 percpu_counter_inc(sk->sk_prot->orphan_count);
 698
 699                 inet_csk_destroy_sock(child);
 700
 701                 bh_unlock_sock(child);
 702                 local_bh_enable();
 703                 sock_put(child);
 704
 705                 sk_acceptq_removed(sk);
 706                 __reqsk_free(req);
 707         }
 708         WARN_ON(sk->sk_ack_backlog);
 709 }
 710 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 711
 712 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 713 {
 714         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 715         const struct inet_sock *inet = inet_sk(sk);
 716
 717         sin->sin_family         = AF_INET;
 718         sin->sin_addr.s_addr    = inet->inet_daddr;
 719         sin->sin_port           = inet->inet_dport;
 720 }
 721 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 722
 723 #ifdef CONFIG_COMPAT
 724 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 725                                char __user *optval, int __user *optlen)
 726 {
 727         const struct inet_connection_sock *icsk = inet_csk(sk);
 728
 729         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 730                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 731                                                             optval, optlen);
 732         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 733                                              optval, optlen);
 734 }
 735 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 736
 737 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 738                                char __user *optval, unsigned int optlen)
 739 {
 740         const struct inet_connection_sock *icsk = inet_csk(sk);
 741
 742         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 743                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 744                                                             optval, optlen);
 745         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 746                                              optval, optlen);
 747 }
 748 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 749 #endif