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.
6 * Support for INET connection oriented protocols.
8 * Authors: See the TCP sources
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.
16 #include <linux/module.h>
17 #include <linux/jhash.h>
19 #include <net/inet_connection_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <net/inet_timewait_sock.h>
23 #include <net/route.h>
24 #include <net/tcp_states.h>
28 const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
29 EXPORT_SYMBOL(inet_csk_timer_bug_msg);
33 * This array holds the first and last local port number.
35 int sysctl_local_port_range[2] = { 32768, 61000 };
36 DEFINE_SEQLOCK(sysctl_port_range_lock);
38 void inet_get_local_port_range(int *low, int *high)
42 seq = read_seqbegin(&sysctl_port_range_lock);
44 *low = sysctl_local_port_range[0];
45 *high = sysctl_local_port_range[1];
46 } while (read_seqretry(&sysctl_port_range_lock, seq));
48 EXPORT_SYMBOL(inet_get_local_port_range);
50 int inet_csk_bind_conflict(const struct sock *sk,
51 const struct inet_bind_bucket *tb)
53 const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
55 struct hlist_node *node;
56 int reuse = sk->sk_reuse;
58 sk_for_each_bound(sk2, node, &tb->owners) {
60 !inet_v6_ipv6only(sk2) &&
61 (!sk->sk_bound_dev_if ||
62 !sk2->sk_bound_dev_if ||
63 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
64 if (!reuse || !sk2->sk_reuse ||
65 sk2->sk_state == TCP_LISTEN) {
66 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
67 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
68 sk2_rcv_saddr == sk_rcv_saddr)
76 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
78 /* Obtain a reference to a local port for the given sock,
79 * if snum is zero it means select any available local port.
81 int inet_csk_get_port(struct sock *sk, unsigned short snum)
83 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
84 struct inet_bind_hashbucket *head;
85 struct hlist_node *node;
86 struct inet_bind_bucket *tb;
88 struct net *net = sock_net(sk);
92 int remaining, rover, low, high;
94 inet_get_local_port_range(&low, &high);
95 remaining = (high - low) + 1;
96 rover = net_random() % remaining + low;
99 head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)];
100 spin_lock(&head->lock);
101 inet_bind_bucket_for_each(tb, node, &head->chain)
102 if (tb->ib_net == net && tb->port == rover)
106 spin_unlock(&head->lock);
109 } while (--remaining > 0);
111 /* Exhausted local port range during search? It is not
112 * possible for us to be holding one of the bind hash
113 * locks if this test triggers, because if 'remaining'
114 * drops to zero, we broke out of the do/while loop at
115 * the top level, not from the 'break;' statement.
121 /* OK, here is the one we will use. HEAD is
122 * non-NULL and we hold it's mutex.
126 head = &hashinfo->bhash[inet_bhashfn(snum, hashinfo->bhash_size)];
127 spin_lock(&head->lock);
128 inet_bind_bucket_for_each(tb, node, &head->chain)
129 if (tb->ib_net == net && tb->port == snum)
135 if (!hlist_empty(&tb->owners)) {
136 if (tb->fastreuse > 0 &&
137 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
141 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))
147 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
148 net, head, snum)) == NULL)
150 if (hlist_empty(&tb->owners)) {
151 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
155 } else if (tb->fastreuse &&
156 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
159 if (!inet_csk(sk)->icsk_bind_hash)
160 inet_bind_hash(sk, tb, snum);
161 BUG_TRAP(inet_csk(sk)->icsk_bind_hash == tb);
165 spin_unlock(&head->lock);
171 EXPORT_SYMBOL_GPL(inet_csk_get_port);
174 * Wait for an incoming connection, avoid race conditions. This must be called
175 * with the socket locked.
177 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
179 struct inet_connection_sock *icsk = inet_csk(sk);
184 * True wake-one mechanism for incoming connections: only
185 * one process gets woken up, not the 'whole herd'.
186 * Since we do not 'race & poll' for established sockets
187 * anymore, the common case will execute the loop only once.
189 * Subtle issue: "add_wait_queue_exclusive()" will be added
190 * after any current non-exclusive waiters, and we know that
191 * it will always _stay_ after any new non-exclusive waiters
192 * because all non-exclusive waiters are added at the
193 * beginning of the wait-queue. As such, it's ok to "drop"
194 * our exclusiveness temporarily when we get woken up without
195 * having to remove and re-insert us on the wait queue.
198 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
201 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
202 timeo = schedule_timeout(timeo);
205 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
208 if (sk->sk_state != TCP_LISTEN)
210 err = sock_intr_errno(timeo);
211 if (signal_pending(current))
217 finish_wait(sk->sk_sleep, &wait);
222 * This will accept the next outstanding connection.
224 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
226 struct inet_connection_sock *icsk = inet_csk(sk);
232 /* We need to make sure that this socket is listening,
233 * and that it has something pending.
236 if (sk->sk_state != TCP_LISTEN)
239 /* Find already established connection */
240 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
241 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
243 /* If this is a non blocking socket don't sleep */
248 error = inet_csk_wait_for_connect(sk, timeo);
253 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
254 BUG_TRAP(newsk->sk_state != TCP_SYN_RECV);
264 EXPORT_SYMBOL(inet_csk_accept);
267 * Using different timers for retransmit, delayed acks and probes
268 * We may wish use just one timer maintaining a list of expire jiffies
271 void inet_csk_init_xmit_timers(struct sock *sk,
272 void (*retransmit_handler)(unsigned long),
273 void (*delack_handler)(unsigned long),
274 void (*keepalive_handler)(unsigned long))
276 struct inet_connection_sock *icsk = inet_csk(sk);
278 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
280 setup_timer(&icsk->icsk_delack_timer, delack_handler,
282 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
283 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
286 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
288 void inet_csk_clear_xmit_timers(struct sock *sk)
290 struct inet_connection_sock *icsk = inet_csk(sk);
292 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
294 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
295 sk_stop_timer(sk, &icsk->icsk_delack_timer);
296 sk_stop_timer(sk, &sk->sk_timer);
299 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
301 void inet_csk_delete_keepalive_timer(struct sock *sk)
303 sk_stop_timer(sk, &sk->sk_timer);
306 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
308 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
310 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
313 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
315 struct dst_entry* inet_csk_route_req(struct sock *sk,
316 const struct request_sock *req)
319 const struct inet_request_sock *ireq = inet_rsk(req);
320 struct ip_options *opt = inet_rsk(req)->opt;
321 struct flowi fl = { .oif = sk->sk_bound_dev_if,
323 { .daddr = ((opt && opt->srr) ?
326 .saddr = ireq->loc_addr,
327 .tos = RT_CONN_FLAGS(sk) } },
328 .proto = sk->sk_protocol,
330 { .sport = inet_sk(sk)->sport,
331 .dport = ireq->rmt_port } } };
333 security_req_classify_flow(req, &fl);
334 if (ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0)) {
335 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
338 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
340 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
346 EXPORT_SYMBOL_GPL(inet_csk_route_req);
348 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
349 const u32 rnd, const u32 synq_hsize)
351 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
354 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
355 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
357 #define AF_INET_FAMILY(fam) 1
360 struct request_sock *inet_csk_search_req(const struct sock *sk,
361 struct request_sock ***prevp,
362 const __be16 rport, const __be32 raddr,
365 const struct inet_connection_sock *icsk = inet_csk(sk);
366 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
367 struct request_sock *req, **prev;
369 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
370 lopt->nr_table_entries)];
371 (req = *prev) != NULL;
372 prev = &req->dl_next) {
373 const struct inet_request_sock *ireq = inet_rsk(req);
375 if (ireq->rmt_port == rport &&
376 ireq->rmt_addr == raddr &&
377 ireq->loc_addr == laddr &&
378 AF_INET_FAMILY(req->rsk_ops->family)) {
388 EXPORT_SYMBOL_GPL(inet_csk_search_req);
390 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
391 unsigned long timeout)
393 struct inet_connection_sock *icsk = inet_csk(sk);
394 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
395 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
396 lopt->hash_rnd, lopt->nr_table_entries);
398 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
399 inet_csk_reqsk_queue_added(sk, timeout);
402 /* Only thing we need from tcp.h */
403 extern int sysctl_tcp_synack_retries;
405 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
407 void inet_csk_reqsk_queue_prune(struct sock *parent,
408 const unsigned long interval,
409 const unsigned long timeout,
410 const unsigned long max_rto)
412 struct inet_connection_sock *icsk = inet_csk(parent);
413 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
414 struct listen_sock *lopt = queue->listen_opt;
415 int thresh = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
416 unsigned long now = jiffies;
417 struct request_sock **reqp, *req;
420 if (lopt == NULL || lopt->qlen == 0)
423 /* Normally all the openreqs are young and become mature
424 * (i.e. converted to established socket) for first timeout.
425 * If synack was not acknowledged for 3 seconds, it means
426 * one of the following things: synack was lost, ack was lost,
427 * rtt is high or nobody planned to ack (i.e. synflood).
428 * When server is a bit loaded, queue is populated with old
429 * open requests, reducing effective size of queue.
430 * When server is well loaded, queue size reduces to zero
431 * after several minutes of work. It is not synflood,
432 * it is normal operation. The solution is pruning
433 * too old entries overriding normal timeout, when
434 * situation becomes dangerous.
436 * Essentially, we reserve half of room for young
437 * embrions; and abort old ones without pity, if old
438 * ones are about to clog our table.
440 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
441 int young = (lopt->qlen_young<<1);
444 if (lopt->qlen < young)
451 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
452 i = lopt->clock_hand;
455 reqp=&lopt->syn_table[i];
456 while ((req = *reqp) != NULL) {
457 if (time_after_eq(now, req->expires)) {
458 if (req->retrans < thresh &&
459 !req->rsk_ops->rtx_syn_ack(parent, req)) {
462 if (req->retrans++ == 0)
464 timeo = min((timeout << req->retrans), max_rto);
465 req->expires = now + timeo;
466 reqp = &req->dl_next;
470 /* Drop this request */
471 inet_csk_reqsk_queue_unlink(parent, req, reqp);
472 reqsk_queue_removed(queue, req);
476 reqp = &req->dl_next;
479 i = (i + 1) & (lopt->nr_table_entries - 1);
481 } while (--budget > 0);
483 lopt->clock_hand = i;
486 inet_csk_reset_keepalive_timer(parent, interval);
489 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
491 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
492 const gfp_t priority)
494 struct sock *newsk = sk_clone(sk, priority);
497 struct inet_connection_sock *newicsk = inet_csk(newsk);
499 newsk->sk_state = TCP_SYN_RECV;
500 newicsk->icsk_bind_hash = NULL;
502 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
503 newsk->sk_write_space = sk_stream_write_space;
505 newicsk->icsk_retransmits = 0;
506 newicsk->icsk_backoff = 0;
507 newicsk->icsk_probes_out = 0;
509 /* Deinitialize accept_queue to trap illegal accesses. */
510 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
512 security_inet_csk_clone(newsk, req);
517 EXPORT_SYMBOL_GPL(inet_csk_clone);
520 * At this point, there should be no process reference to this
521 * socket, and thus no user references at all. Therefore we
522 * can assume the socket waitqueue is inactive and nobody will
523 * try to jump onto it.
525 void inet_csk_destroy_sock(struct sock *sk)
527 BUG_TRAP(sk->sk_state == TCP_CLOSE);
528 BUG_TRAP(sock_flag(sk, SOCK_DEAD));
530 /* It cannot be in hash table! */
531 BUG_TRAP(sk_unhashed(sk));
533 /* If it has not 0 inet_sk(sk)->num, it must be bound */
534 BUG_TRAP(!inet_sk(sk)->num || inet_csk(sk)->icsk_bind_hash);
536 sk->sk_prot->destroy(sk);
538 sk_stream_kill_queues(sk);
540 xfrm_sk_free_policy(sk);
542 sk_refcnt_debug_release(sk);
544 atomic_dec(sk->sk_prot->orphan_count);
548 EXPORT_SYMBOL(inet_csk_destroy_sock);
550 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
552 struct inet_sock *inet = inet_sk(sk);
553 struct inet_connection_sock *icsk = inet_csk(sk);
554 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
559 sk->sk_max_ack_backlog = 0;
560 sk->sk_ack_backlog = 0;
561 inet_csk_delack_init(sk);
563 /* There is race window here: we announce ourselves listening,
564 * but this transition is still not validated by get_port().
565 * It is OK, because this socket enters to hash table only
566 * after validation is complete.
568 sk->sk_state = TCP_LISTEN;
569 if (!sk->sk_prot->get_port(sk, inet->num)) {
570 inet->sport = htons(inet->num);
573 sk->sk_prot->hash(sk);
578 sk->sk_state = TCP_CLOSE;
579 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
583 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
586 * This routine closes sockets which have been at least partially
587 * opened, but not yet accepted.
589 void inet_csk_listen_stop(struct sock *sk)
591 struct inet_connection_sock *icsk = inet_csk(sk);
592 struct request_sock *acc_req;
593 struct request_sock *req;
595 inet_csk_delete_keepalive_timer(sk);
597 /* make all the listen_opt local to us */
598 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
600 /* Following specs, it would be better either to send FIN
601 * (and enter FIN-WAIT-1, it is normal close)
602 * or to send active reset (abort).
603 * Certainly, it is pretty dangerous while synflood, but it is
604 * bad justification for our negligence 8)
605 * To be honest, we are not able to make either
606 * of the variants now. --ANK
608 reqsk_queue_destroy(&icsk->icsk_accept_queue);
610 while ((req = acc_req) != NULL) {
611 struct sock *child = req->sk;
613 acc_req = req->dl_next;
617 BUG_TRAP(!sock_owned_by_user(child));
620 sk->sk_prot->disconnect(child, O_NONBLOCK);
624 atomic_inc(sk->sk_prot->orphan_count);
626 inet_csk_destroy_sock(child);
628 bh_unlock_sock(child);
632 sk_acceptq_removed(sk);
635 BUG_TRAP(!sk->sk_ack_backlog);
638 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
640 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
642 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
643 const struct inet_sock *inet = inet_sk(sk);
645 sin->sin_family = AF_INET;
646 sin->sin_addr.s_addr = inet->daddr;
647 sin->sin_port = inet->dport;
650 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
653 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
654 char __user *optval, int __user *optlen)
656 const struct inet_connection_sock *icsk = inet_csk(sk);
658 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
659 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
661 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
665 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
667 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
668 char __user *optval, int optlen)
670 const struct inet_connection_sock *icsk = inet_csk(sk);
672 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
673 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
675 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
679 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);