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(sysctl_local_ports.lock),
  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 int seq;
  46
  47         do {
  48                 seq = read_seqbegin(&sysctl_local_ports.lock);
  49
  50                 *low = sysctl_local_ports.range[0];
  51                 *high = sysctl_local_ports.range[1];
  52         } while (read_seqretry(&sysctl_local_ports.lock, seq));
  53 }
  54 EXPORT_SYMBOL(inet_get_local_port_range);
  55
  56 int inet_csk_bind_conflict(const struct sock *sk,
  57                            const struct inet_bind_bucket *tb, bool relax)
  58 {
  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 = sk_rcv_saddr(sk2);
  79                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  80                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  81                                         break;
  82                         }
  83                         if (!relax && reuse && sk2->sk_reuse &&
  84                             sk2->sk_state != TCP_LISTEN) {
  85                                 const __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2);
  86
  87                                 if (!sk2_rcv_saddr || !sk_rcv_saddr(sk) ||
  88                                     sk2_rcv_saddr == sk_rcv_saddr(sk))
  89                                         break;
  90                         }
  91                 }
  92         }
  93         return node != NULL;
  94 }
  95 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
  96
  97 /* Obtain a reference to a local port for the given sock,
  98  * if snum is zero it means select any available local port.
  99  */
 100 int inet_csk_get_port(struct sock *sk, unsigned short snum)
 101 {
 102         struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
 103         struct inet_bind_hashbucket *head;
 104         struct hlist_node *node;
 105         struct inet_bind_bucket *tb;
 106         int ret, attempts = 5;
 107         struct net *net = sock_net(sk);
 108         int smallest_size = -1, smallest_rover;
 109
 110         local_bh_disable();
 111         if (!snum) {
 112                 int remaining, rover, low, high;
 113
 114 again:
 115                 inet_get_local_port_range(&low, &high);
 116                 remaining = (high - low) + 1;
 117                 smallest_rover = rover = net_random() % remaining + low;
 118
 119                 smallest_size = -1;
 120                 do {
 121                         if (inet_is_reserved_local_port(rover))
 122                                 goto next_nolock;
 123                         head = &hashinfo->bhash[inet_bhashfn(net, rover,
 124                                         hashinfo->bhash_size)];
 125                         spin_lock(&head->lock);
 126                         inet_bind_bucket_for_each(tb, node, &head->chain)
 127                                 if (net_eq(ib_net(tb), net) && tb->port == rover) {
 128                                         if (tb->fastreuse > 0 &&
 129                                             sk->sk_reuse &&
 130                                             sk->sk_state != TCP_LISTEN &&
 131                                             (tb->num_owners < smallest_size || smallest_size == -1)) {
 132                                                 smallest_size = tb->num_owners;
 133                                                 smallest_rover = rover;
 134                                                 if (atomic_read(&hashinfo->bsockets) > (high - low) + 1 &&
 135                                                     !inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 136                                                         snum = smallest_rover;
 137                                                         goto tb_found;
 138                                                 }
 139                                         }
 140                                         if (!inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, false)) {
 141                                                 snum = rover;
 142                                                 goto tb_found;
 143                                         }
 144                                         goto next;
 145                                 }
 146                         break;
 147                 next:
 148                         spin_unlock(&head->lock);
 149                 next_nolock:
 150                         if (++rover > high)
 151                                 rover = low;
 152                 } while (--remaining > 0);
 153
 154                 /* Exhausted local port range during search?  It is not
 155                  * possible for us to be holding one of the bind hash
 156                  * locks if this test triggers, because if 'remaining'
 157                  * drops to zero, we broke out of the do/while loop at
 158                  * the top level, not from the 'break;' statement.
 159                  */
 160                 ret = 1;
 161                 if (remaining <= 0) {
 162                         if (smallest_size != -1) {
 163                                 snum = smallest_rover;
 164                                 goto have_snum;
 165                         }
 166                         goto fail;
 167                 }
 168                 /* OK, here is the one we will use.  HEAD is
 169                  * non-NULL and we hold it's mutex.
 170                  */
 171                 snum = rover;
 172         } else {
 173 have_snum:
 174                 head = &hashinfo->bhash[inet_bhashfn(net, snum,
 175                                 hashinfo->bhash_size)];
 176                 spin_lock(&head->lock);
 177                 inet_bind_bucket_for_each(tb, node, &head->chain)
 178                         if (net_eq(ib_net(tb), net) && tb->port == snum)
 179                                 goto tb_found;
 180         }
 181         tb = NULL;
 182         goto tb_not_found;
 183 tb_found:
 184         if (!hlist_empty(&tb->owners)) {
 185                 if (sk->sk_reuse == SK_FORCE_REUSE)
 186                         goto success;
 187
 188                 if (tb->fastreuse > 0 &&
 189                     sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 190                     smallest_size == -1) {
 191                         goto success;
 192                 } else {
 193                         ret = 1;
 194                         if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb, true)) {
 195                                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
 196                                     smallest_size != -1 && --attempts >= 0) {
 197                                         spin_unlock(&head->lock);
 198                                         goto again;
 199                                 }
 200
 201                                 goto fail_unlock;
 202                         }
 203                 }
 204         }
 205 tb_not_found:
 206         ret = 1;
 207         if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
 208                                         net, head, snum)) == NULL)
 209                 goto fail_unlock;
 210         if (hlist_empty(&tb->owners)) {
 211                 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
 212                         tb->fastreuse = 1;
 213                 else
 214                         tb->fastreuse = 0;
 215         } else if (tb->fastreuse &&
 216                    (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
 217                 tb->fastreuse = 0;
 218 success:
 219         if (!inet_csk(sk)->icsk_bind_hash)
 220                 inet_bind_hash(sk, tb, snum);
 221         WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
 222         ret = 0;
 223
 224 fail_unlock:
 225         spin_unlock(&head->lock);
 226 fail:
 227         local_bh_enable();
 228         return ret;
 229 }
 230 EXPORT_SYMBOL_GPL(inet_csk_get_port);
 231
 232 /*
 233  * Wait for an incoming connection, avoid race conditions. This must be called
 234  * with the socket locked.
 235  */
 236 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
 237 {
 238         struct inet_connection_sock *icsk = inet_csk(sk);
 239         DEFINE_WAIT(wait);
 240         int err;
 241
 242         /*
 243          * True wake-one mechanism for incoming connections: only
 244          * one process gets woken up, not the 'whole herd'.
 245          * Since we do not 'race & poll' for established sockets
 246          * anymore, the common case will execute the loop only once.
 247          *
 248          * Subtle issue: "add_wait_queue_exclusive()" will be added
 249          * after any current non-exclusive waiters, and we know that
 250          * it will always _stay_ after any new non-exclusive waiters
 251          * because all non-exclusive waiters are added at the
 252          * beginning of the wait-queue. As such, it's ok to "drop"
 253          * our exclusiveness temporarily when we get woken up without
 254          * having to remove and re-insert us on the wait queue.
 255          */
 256         for (;;) {
 257                 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
 258                                           TASK_INTERRUPTIBLE);
 259                 release_sock(sk);
 260                 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
 261                         timeo = schedule_timeout(timeo);
 262                 lock_sock(sk);
 263                 err = 0;
 264                 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
 265                         break;
 266                 err = -EINVAL;
 267                 if (sk->sk_state != TCP_LISTEN)
 268                         break;
 269                 err = sock_intr_errno(timeo);
 270                 if (signal_pending(current))
 271                         break;
 272                 err = -EAGAIN;
 273                 if (!timeo)
 274                         break;
 275         }
 276         finish_wait(sk_sleep(sk), &wait);
 277         return err;
 278 }
 279
 280 /*
 281  * This will accept the next outstanding connection.
 282  */
 283 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
 284 {
 285         struct inet_connection_sock *icsk = inet_csk(sk);
 286         struct sock *newsk;
 287         int error;
 288
 289         lock_sock(sk);
 290
 291         /* We need to make sure that this socket is listening,
 292          * and that it has something pending.
 293          */
 294         error = -EINVAL;
 295         if (sk->sk_state != TCP_LISTEN)
 296                 goto out_err;
 297
 298         /* Find already established connection */
 299         if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
 300                 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
 301
 302                 /* If this is a non blocking socket don't sleep */
 303                 error = -EAGAIN;
 304                 if (!timeo)
 305                         goto out_err;
 306
 307                 error = inet_csk_wait_for_connect(sk, timeo);
 308                 if (error)
 309                         goto out_err;
 310         }
 311
 312         newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
 313         WARN_ON(newsk->sk_state == TCP_SYN_RECV);
 314 out:
 315         release_sock(sk);
 316         return newsk;
 317 out_err:
 318         newsk = NULL;
 319         *err = error;
 320         goto out;
 321 }
 322 EXPORT_SYMBOL(inet_csk_accept);
 323
 324 /*
 325  * Using different timers for retransmit, delayed acks and probes
 326  * We may wish use just one timer maintaining a list of expire jiffies
 327  * to optimize.
 328  */
 329 void inet_csk_init_xmit_timers(struct sock *sk,
 330                                void (*retransmit_handler)(unsigned long),
 331                                void (*delack_handler)(unsigned long),
 332                                void (*keepalive_handler)(unsigned long))
 333 {
 334         struct inet_connection_sock *icsk = inet_csk(sk);
 335
 336         setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
 337                         (unsigned long)sk);
 338         setup_timer(&icsk->icsk_delack_timer, delack_handler,
 339                         (unsigned long)sk);
 340         setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
 341         icsk->icsk_pending = icsk->icsk_ack.pending = 0;
 342 }
 343 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
 344
 345 void inet_csk_clear_xmit_timers(struct sock *sk)
 346 {
 347         struct inet_connection_sock *icsk = inet_csk(sk);
 348
 349         icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
 350
 351         sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
 352         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 353         sk_stop_timer(sk, &sk->sk_timer);
 354 }
 355 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
 356
 357 void inet_csk_delete_keepalive_timer(struct sock *sk)
 358 {
 359         sk_stop_timer(sk, &sk->sk_timer);
 360 }
 361 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
 362
 363 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
 364 {
 365         sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
 366 }
 367 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
 368
 369 struct dst_entry *inet_csk_route_req(struct sock *sk,
 370                                      struct flowi4 *fl4,
 371                                      const struct request_sock *req)
 372 {
 373         struct rtable *rt;
 374         const struct inet_request_sock *ireq = inet_rsk(req);
 375         struct ip_options_rcu *opt = inet_rsk(req)->opt;
 376         struct net *net = sock_net(sk);
 377
 378         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 379                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 380                            sk->sk_protocol,
 381                            inet_sk_flowi_flags(sk) & ~FLOWI_FLAG_PRECOW_METRICS,
 382                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 383                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 384         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 385         rt = ip_route_output_flow(net, fl4, sk);
 386         if (IS_ERR(rt))
 387                 goto no_route;
 388         if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
 389                 goto route_err;
 390         return &rt->dst;
 391
 392 route_err:
 393         ip_rt_put(rt);
 394 no_route:
 395         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 396         return NULL;
 397 }
 398 EXPORT_SYMBOL_GPL(inet_csk_route_req);
 399
 400 struct dst_entry *inet_csk_route_child_sock(struct sock *sk,
 401                                             struct sock *newsk,
 402                                             const struct request_sock *req)
 403 {
 404         const struct inet_request_sock *ireq = inet_rsk(req);
 405         struct inet_sock *newinet = inet_sk(newsk);
 406         struct ip_options_rcu *opt = ireq->opt;
 407         struct net *net = sock_net(sk);
 408         struct flowi4 *fl4;
 409         struct rtable *rt;
 410
 411         fl4 = &newinet->cork.fl.u.ip4;
 412         flowi4_init_output(fl4, sk->sk_bound_dev_if, sk->sk_mark,
 413                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE,
 414                            sk->sk_protocol, inet_sk_flowi_flags(sk),
 415                            (opt && opt->opt.srr) ? opt->opt.faddr : ireq->rmt_addr,
 416                            ireq->loc_addr, ireq->rmt_port, inet_sk(sk)->inet_sport);
 417         security_req_classify_flow(req, flowi4_to_flowi(fl4));
 418         rt = ip_route_output_flow(net, fl4, sk);
 419         if (IS_ERR(rt))
 420                 goto no_route;
 421         if (opt && opt->opt.is_strictroute && fl4->daddr != rt->rt_gateway)
 422                 goto route_err;
 423         return &rt->dst;
 424
 425 route_err:
 426         ip_rt_put(rt);
 427 no_route:
 428         IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 429         return NULL;
 430 }
 431 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
 432
 433 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
 434                                  const u32 rnd, const u32 synq_hsize)
 435 {
 436         return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
 437 }
 438
 439 #if IS_ENABLED(CONFIG_IPV6)
 440 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
 441 #else
 442 #define AF_INET_FAMILY(fam) 1
 443 #endif
 444
 445 struct request_sock *inet_csk_search_req(const struct sock *sk,
 446                                          struct request_sock ***prevp,
 447                                          const __be16 rport, const __be32 raddr,
 448                                          const __be32 laddr)
 449 {
 450         const struct inet_connection_sock *icsk = inet_csk(sk);
 451         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 452         struct request_sock *req, **prev;
 453
 454         for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
 455                                                     lopt->nr_table_entries)];
 456              (req = *prev) != NULL;
 457              prev = &req->dl_next) {
 458                 const struct inet_request_sock *ireq = inet_rsk(req);
 459
 460                 if (ireq->rmt_port == rport &&
 461                     ireq->rmt_addr == raddr &&
 462                     ireq->loc_addr == laddr &&
 463                     AF_INET_FAMILY(req->rsk_ops->family)) {
 464                         WARN_ON(req->sk);
 465                         *prevp = prev;
 466                         break;
 467                 }
 468         }
 469
 470         return req;
 471 }
 472 EXPORT_SYMBOL_GPL(inet_csk_search_req);
 473
 474 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
 475                                    unsigned long timeout)
 476 {
 477         struct inet_connection_sock *icsk = inet_csk(sk);
 478         struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
 479         const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
 480                                      lopt->hash_rnd, lopt->nr_table_entries);
 481
 482         reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
 483         inet_csk_reqsk_queue_added(sk, timeout);
 484 }
 485 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
 486
 487 /* Only thing we need from tcp.h */
 488 extern int sysctl_tcp_synack_retries;
 489
 490
 491 /* Decide when to expire the request and when to resend SYN-ACK */
 492 static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
 493                                   const int max_retries,
 494                                   const u8 rskq_defer_accept,
 495                                   int *expire, int *resend)
 496 {
 497         if (!rskq_defer_accept) {
 498                 *expire = req->retrans >= thresh;
 499                 *resend = 1;
 500                 return;
 501         }
 502         *expire = req->retrans >= thresh &&
 503                   (!inet_rsk(req)->acked || req->retrans >= max_retries);
 504         /*
 505          * Do not resend while waiting for data after ACK,
 506          * start to resend on end of deferring period to give
 507          * last chance for data or ACK to create established socket.
 508          */
 509         *resend = !inet_rsk(req)->acked ||
 510                   req->retrans >= rskq_defer_accept - 1;
 511 }
 512
 513 void inet_csk_reqsk_queue_prune(struct sock *parent,
 514                                 const unsigned long interval,
 515                                 const unsigned long timeout,
 516                                 const unsigned long max_rto)
 517 {
 518         struct inet_connection_sock *icsk = inet_csk(parent);
 519         struct request_sock_queue *queue = &icsk->icsk_accept_queue;
 520         struct listen_sock *lopt = queue->listen_opt;
 521         int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
 522         int thresh = max_retries;
 523         unsigned long now = jiffies;
 524         struct request_sock **reqp, *req;
 525         int i, budget;
 526
 527         if (lopt == NULL || lopt->qlen == 0)
 528                 return;
 529
 530         /* Normally all the openreqs are young and become mature
 531          * (i.e. converted to established socket) for first timeout.
 532          * If synack was not acknowledged for 1 second, it means
 533          * one of the following things: synack was lost, ack was lost,
 534          * rtt is high or nobody planned to ack (i.e. synflood).
 535          * When server is a bit loaded, queue is populated with old
 536          * open requests, reducing effective size of queue.
 537          * When server is well loaded, queue size reduces to zero
 538          * after several minutes of work. It is not synflood,
 539          * it is normal operation. The solution is pruning
 540          * too old entries overriding normal timeout, when
 541          * situation becomes dangerous.
 542          *
 543          * Essentially, we reserve half of room for young
 544          * embrions; and abort old ones without pity, if old
 545          * ones are about to clog our table.
 546          */
 547         if (lopt->qlen>>(lopt->max_qlen_log-1)) {
 548                 int young = (lopt->qlen_young<<1);
 549
 550                 while (thresh > 2) {
 551                         if (lopt->qlen < young)
 552                                 break;
 553                         thresh--;
 554                         young <<= 1;
 555                 }
 556         }
 557
 558         if (queue->rskq_defer_accept)
 559                 max_retries = queue->rskq_defer_accept;
 560
 561         budget = 2 * (lopt->nr_table_entries / (timeout / interval));
 562         i = lopt->clock_hand;
 563
 564         do {
 565                 reqp=&lopt->syn_table[i];
 566                 while ((req = *reqp) != NULL) {
 567                         if (time_after_eq(now, req->expires)) {
 568                                 int expire = 0, resend = 0;
 569
 570                                 syn_ack_recalc(req, thresh, max_retries,
 571                                                queue->rskq_defer_accept,
 572                                                &expire, &resend);
 573                                 req->rsk_ops->syn_ack_timeout(parent, req);
 574                                 if (!expire &&
 575                                     (!resend ||
 576                                      !req->rsk_ops->rtx_syn_ack(parent, req, NULL) ||
 577                                      inet_rsk(req)->acked)) {
 578                                         unsigned long timeo;
 579
 580                                         if (req->retrans++ == 0)
 581                                                 lopt->qlen_young--;
 582                                         timeo = min((timeout << req->retrans), max_rto);
 583                                         req->expires = now + timeo;
 584                                         reqp = &req->dl_next;
 585                                         continue;
 586                                 }
 587
 588                                 /* Drop this request */
 589                                 inet_csk_reqsk_queue_unlink(parent, req, reqp);
 590                                 reqsk_queue_removed(queue, req);
 591                                 reqsk_free(req);
 592                                 continue;
 593                         }
 594                         reqp = &req->dl_next;
 595                 }
 596
 597                 i = (i + 1) & (lopt->nr_table_entries - 1);
 598
 599         } while (--budget > 0);
 600
 601         lopt->clock_hand = i;
 602
 603         if (lopt->qlen)
 604                 inet_csk_reset_keepalive_timer(parent, interval);
 605 }
 606 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
 607
 608 /**
 609  *      inet_csk_clone_lock - clone an inet socket, and lock its clone
 610  *      @sk: the socket to clone
 611  *      @req: request_sock
 612  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 613  *
 614  *      Caller must unlock socket even in error path (bh_unlock_sock(newsk))
 615  */
 616 struct sock *inet_csk_clone_lock(const struct sock *sk,
 617                                  const struct request_sock *req,
 618                                  const gfp_t priority)
 619 {
 620         struct sock *newsk = sk_clone_lock(sk, priority);
 621
 622         if (newsk != NULL) {
 623                 struct inet_connection_sock *newicsk = inet_csk(newsk);
 624
 625                 newsk->sk_state = TCP_SYN_RECV;
 626                 newicsk->icsk_bind_hash = NULL;
 627
 628                 inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
 629                 inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
 630                 inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
 631                 newsk->sk_write_space = sk_stream_write_space;
 632
 633                 newicsk->icsk_retransmits = 0;
 634                 newicsk->icsk_backoff     = 0;
 635                 newicsk->icsk_probes_out  = 0;
 636
 637                 /* Deinitialize accept_queue to trap illegal accesses. */
 638                 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
 639
 640                 security_inet_csk_clone(newsk, req);
 641         }
 642         return newsk;
 643 }
 644 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
 645
 646 /*
 647  * At this point, there should be no process reference to this
 648  * socket, and thus no user references at all.  Therefore we
 649  * can assume the socket waitqueue is inactive and nobody will
 650  * try to jump onto it.
 651  */
 652 void inet_csk_destroy_sock(struct sock *sk)
 653 {
 654         WARN_ON(sk->sk_state != TCP_CLOSE);
 655         WARN_ON(!sock_flag(sk, SOCK_DEAD));
 656
 657         /* It cannot be in hash table! */
 658         WARN_ON(!sk_unhashed(sk));
 659
 660         /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
 661         WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
 662
 663         sk->sk_prot->destroy(sk);
 664
 665         sk_stream_kill_queues(sk);
 666
 667         xfrm_sk_free_policy(sk);
 668
 669         sk_refcnt_debug_release(sk);
 670
 671         percpu_counter_dec(sk->sk_prot->orphan_count);
 672         sock_put(sk);
 673 }
 674 EXPORT_SYMBOL(inet_csk_destroy_sock);
 675
 676 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
 677 {
 678         struct inet_sock *inet = inet_sk(sk);
 679         struct inet_connection_sock *icsk = inet_csk(sk);
 680         int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
 681
 682         if (rc != 0)
 683                 return rc;
 684
 685         sk->sk_max_ack_backlog = 0;
 686         sk->sk_ack_backlog = 0;
 687         inet_csk_delack_init(sk);
 688
 689         /* There is race window here: we announce ourselves listening,
 690          * but this transition is still not validated by get_port().
 691          * It is OK, because this socket enters to hash table only
 692          * after validation is complete.
 693          */
 694         sk->sk_state = TCP_LISTEN;
 695         if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
 696                 inet->inet_sport = htons(inet->inet_num);
 697
 698                 sk_dst_reset(sk);
 699                 sk->sk_prot->hash(sk);
 700
 701                 return 0;
 702         }
 703
 704         sk->sk_state = TCP_CLOSE;
 705         __reqsk_queue_destroy(&icsk->icsk_accept_queue);
 706         return -EADDRINUSE;
 707 }
 708 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
 709
 710 /*
 711  *      This routine closes sockets which have been at least partially
 712  *      opened, but not yet accepted.
 713  */
 714 void inet_csk_listen_stop(struct sock *sk)
 715 {
 716         struct inet_connection_sock *icsk = inet_csk(sk);
 717         struct request_sock *acc_req;
 718         struct request_sock *req;
 719
 720         inet_csk_delete_keepalive_timer(sk);
 721
 722         /* make all the listen_opt local to us */
 723         acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
 724
 725         /* Following specs, it would be better either to send FIN
 726          * (and enter FIN-WAIT-1, it is normal close)
 727          * or to send active reset (abort).
 728          * Certainly, it is pretty dangerous while synflood, but it is
 729          * bad justification for our negligence 8)
 730          * To be honest, we are not able to make either
 731          * of the variants now.                 --ANK
 732          */
 733         reqsk_queue_destroy(&icsk->icsk_accept_queue);
 734
 735         while ((req = acc_req) != NULL) {
 736                 struct sock *child = req->sk;
 737
 738                 acc_req = req->dl_next;
 739
 740                 local_bh_disable();
 741                 bh_lock_sock(child);
 742                 WARN_ON(sock_owned_by_user(child));
 743                 sock_hold(child);
 744
 745                 sk->sk_prot->disconnect(child, O_NONBLOCK);
 746
 747                 sock_orphan(child);
 748
 749                 percpu_counter_inc(sk->sk_prot->orphan_count);
 750
 751                 inet_csk_destroy_sock(child);
 752
 753                 bh_unlock_sock(child);
 754                 local_bh_enable();
 755                 sock_put(child);
 756
 757                 sk_acceptq_removed(sk);
 758                 __reqsk_free(req);
 759         }
 760         WARN_ON(sk->sk_ack_backlog);
 761 }
 762 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
 763
 764 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
 765 {
 766         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
 767         const struct inet_sock *inet = inet_sk(sk);
 768
 769         sin->sin_family         = AF_INET;
 770         sin->sin_addr.s_addr    = inet->inet_daddr;
 771         sin->sin_port           = inet->inet_dport;
 772 }
 773 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
 774
 775 #ifdef CONFIG_COMPAT
 776 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
 777                                char __user *optval, int __user *optlen)
 778 {
 779         const struct inet_connection_sock *icsk = inet_csk(sk);
 780
 781         if (icsk->icsk_af_ops->compat_getsockopt != NULL)
 782                 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
 783                                                             optval, optlen);
 784         return icsk->icsk_af_ops->getsockopt(sk, level, optname,
 785                                              optval, optlen);
 786 }
 787 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
 788
 789 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
 790                                char __user *optval, unsigned int optlen)
 791 {
 792         const struct inet_connection_sock *icsk = inet_csk(sk);
 793
 794         if (icsk->icsk_af_ops->compat_setsockopt != NULL)
 795                 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
 796                                                             optval, optlen);
 797         return icsk->icsk_af_ops->setsockopt(sk, level, optname,
 798                                              optval, optlen);
 799 }
 800 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
 801 #endif