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 struct holds the first and last local port number.
35 struct local_ports sysctl_local_ports __read_mostly = {
36 .lock = SEQLOCK_UNLOCKED,
37 .range = { 32768, 61000 },
40 void inet_get_local_port_range(int *low, int *high)
44 seq = read_seqbegin(&sysctl_local_ports.lock);
46 *low = sysctl_local_ports.range[0];
47 *high = sysctl_local_ports.range[1];
48 } while (read_seqretry(&sysctl_local_ports.lock, seq));
50 EXPORT_SYMBOL(inet_get_local_port_range);
52 int inet_csk_bind_conflict(const struct sock *sk,
53 const struct inet_bind_bucket *tb)
55 const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
57 struct hlist_node *node;
58 int reuse = sk->sk_reuse;
61 * Unlike other sk lookup places we do not check
62 * for sk_net here, since _all_ the socks listed
63 * in tb->owners list belong to the same net - the
64 * one this bucket belongs to.
67 sk_for_each_bound(sk2, node, &tb->owners) {
69 !inet_v6_ipv6only(sk2) &&
70 (!sk->sk_bound_dev_if ||
71 !sk2->sk_bound_dev_if ||
72 sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
73 if (!reuse || !sk2->sk_reuse ||
74 sk2->sk_state == TCP_LISTEN) {
75 const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
76 if (!sk2_rcv_saddr || !sk_rcv_saddr ||
77 sk2_rcv_saddr == sk_rcv_saddr)
85 EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
87 /* Obtain a reference to a local port for the given sock,
88 * if snum is zero it means select any available local port.
90 int inet_csk_get_port(struct sock *sk, unsigned short snum)
92 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
93 struct inet_bind_hashbucket *head;
94 struct hlist_node *node;
95 struct inet_bind_bucket *tb;
96 int ret, attempts = 5;
97 struct net *net = sock_net(sk);
98 int smallest_size = -1, smallest_rover;
102 int remaining, rover, low, high;
105 inet_get_local_port_range(&low, &high);
106 remaining = (high - low) + 1;
107 smallest_rover = rover = net_random() % remaining + low;
111 head = &hashinfo->bhash[inet_bhashfn(net, rover,
112 hashinfo->bhash_size)];
113 spin_lock(&head->lock);
114 inet_bind_bucket_for_each(tb, node, &head->chain)
115 if (ib_net(tb) == net && tb->port == rover) {
116 if (tb->fastreuse > 0 &&
118 sk->sk_state != TCP_LISTEN &&
119 (tb->num_owners < smallest_size || smallest_size == -1)) {
120 smallest_size = tb->num_owners;
121 smallest_rover = rover;
122 if (hashinfo->bsockets > (high - low) + 1) {
123 spin_unlock(&head->lock);
124 snum = smallest_rover;
132 spin_unlock(&head->lock);
135 } while (--remaining > 0);
137 /* Exhausted local port range during search? It is not
138 * possible for us to be holding one of the bind hash
139 * locks if this test triggers, because if 'remaining'
140 * drops to zero, we broke out of the do/while loop at
141 * the top level, not from the 'break;' statement.
144 if (remaining <= 0) {
145 if (smallest_size != -1) {
146 snum = smallest_rover;
151 /* OK, here is the one we will use. HEAD is
152 * non-NULL and we hold it's mutex.
157 head = &hashinfo->bhash[inet_bhashfn(net, snum,
158 hashinfo->bhash_size)];
159 spin_lock(&head->lock);
160 inet_bind_bucket_for_each(tb, node, &head->chain)
161 if (ib_net(tb) == net && tb->port == snum)
167 if (!hlist_empty(&tb->owners)) {
168 if (tb->fastreuse > 0 &&
169 sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
170 smallest_size == -1) {
174 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
175 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && --attempts >= 0) {
176 spin_unlock(&head->lock);
185 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
186 net, head, snum)) == NULL)
188 if (hlist_empty(&tb->owners)) {
189 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
193 } else if (tb->fastreuse &&
194 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
197 if (!inet_csk(sk)->icsk_bind_hash)
198 inet_bind_hash(sk, tb, snum);
199 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
203 spin_unlock(&head->lock);
209 EXPORT_SYMBOL_GPL(inet_csk_get_port);
212 * Wait for an incoming connection, avoid race conditions. This must be called
213 * with the socket locked.
215 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
217 struct inet_connection_sock *icsk = inet_csk(sk);
222 * True wake-one mechanism for incoming connections: only
223 * one process gets woken up, not the 'whole herd'.
224 * Since we do not 'race & poll' for established sockets
225 * anymore, the common case will execute the loop only once.
227 * Subtle issue: "add_wait_queue_exclusive()" will be added
228 * after any current non-exclusive waiters, and we know that
229 * it will always _stay_ after any new non-exclusive waiters
230 * because all non-exclusive waiters are added at the
231 * beginning of the wait-queue. As such, it's ok to "drop"
232 * our exclusiveness temporarily when we get woken up without
233 * having to remove and re-insert us on the wait queue.
236 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
239 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
240 timeo = schedule_timeout(timeo);
243 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
246 if (sk->sk_state != TCP_LISTEN)
248 err = sock_intr_errno(timeo);
249 if (signal_pending(current))
255 finish_wait(sk->sk_sleep, &wait);
260 * This will accept the next outstanding connection.
262 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
264 struct inet_connection_sock *icsk = inet_csk(sk);
270 /* We need to make sure that this socket is listening,
271 * and that it has something pending.
274 if (sk->sk_state != TCP_LISTEN)
277 /* Find already established connection */
278 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
279 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
281 /* If this is a non blocking socket don't sleep */
286 error = inet_csk_wait_for_connect(sk, timeo);
291 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
292 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
302 EXPORT_SYMBOL(inet_csk_accept);
305 * Using different timers for retransmit, delayed acks and probes
306 * We may wish use just one timer maintaining a list of expire jiffies
309 void inet_csk_init_xmit_timers(struct sock *sk,
310 void (*retransmit_handler)(unsigned long),
311 void (*delack_handler)(unsigned long),
312 void (*keepalive_handler)(unsigned long))
314 struct inet_connection_sock *icsk = inet_csk(sk);
316 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
318 setup_timer(&icsk->icsk_delack_timer, delack_handler,
320 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
321 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
324 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
326 void inet_csk_clear_xmit_timers(struct sock *sk)
328 struct inet_connection_sock *icsk = inet_csk(sk);
330 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
332 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
333 sk_stop_timer(sk, &icsk->icsk_delack_timer);
334 sk_stop_timer(sk, &sk->sk_timer);
337 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
339 void inet_csk_delete_keepalive_timer(struct sock *sk)
341 sk_stop_timer(sk, &sk->sk_timer);
344 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
346 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
348 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
351 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
353 struct dst_entry *inet_csk_route_req(struct sock *sk,
354 const struct request_sock *req)
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,
361 { .daddr = ((opt && opt->srr) ?
364 .saddr = ireq->loc_addr,
365 .tos = RT_CONN_FLAGS(sk) } },
366 .proto = sk->sk_protocol,
367 .flags = inet_sk_flowi_flags(sk),
369 { .sport = inet_sk(sk)->sport,
370 .dport = ireq->rmt_port } } };
371 struct net *net = sock_net(sk);
373 security_req_classify_flow(req, &fl);
374 if (ip_route_output_flow(net, &rt, &fl, sk, 0))
376 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
383 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
387 EXPORT_SYMBOL_GPL(inet_csk_route_req);
389 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
390 const u32 rnd, const u32 synq_hsize)
392 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
395 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
396 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
398 #define AF_INET_FAMILY(fam) 1
401 struct request_sock *inet_csk_search_req(const struct sock *sk,
402 struct request_sock ***prevp,
403 const __be16 rport, const __be32 raddr,
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;
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);
416 if (ireq->rmt_port == rport &&
417 ireq->rmt_addr == raddr &&
418 ireq->loc_addr == laddr &&
419 AF_INET_FAMILY(req->rsk_ops->family)) {
429 EXPORT_SYMBOL_GPL(inet_csk_search_req);
431 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
432 unsigned long timeout)
434 struct inet_connection_sock *icsk = inet_csk(sk);
435 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
436 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
437 lopt->hash_rnd, lopt->nr_table_entries);
439 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
440 inet_csk_reqsk_queue_added(sk, timeout);
443 /* Only thing we need from tcp.h */
444 extern int sysctl_tcp_synack_retries;
446 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
448 void inet_csk_reqsk_queue_prune(struct sock *parent,
449 const unsigned long interval,
450 const unsigned long timeout,
451 const unsigned long max_rto)
453 struct inet_connection_sock *icsk = inet_csk(parent);
454 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
455 struct listen_sock *lopt = queue->listen_opt;
456 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
457 int thresh = max_retries;
458 unsigned long now = jiffies;
459 struct request_sock **reqp, *req;
462 if (lopt == NULL || lopt->qlen == 0)
465 /* Normally all the openreqs are young and become mature
466 * (i.e. converted to established socket) for first timeout.
467 * If synack was not acknowledged for 3 seconds, it means
468 * one of the following things: synack was lost, ack was lost,
469 * rtt is high or nobody planned to ack (i.e. synflood).
470 * When server is a bit loaded, queue is populated with old
471 * open requests, reducing effective size of queue.
472 * When server is well loaded, queue size reduces to zero
473 * after several minutes of work. It is not synflood,
474 * it is normal operation. The solution is pruning
475 * too old entries overriding normal timeout, when
476 * situation becomes dangerous.
478 * Essentially, we reserve half of room for young
479 * embrions; and abort old ones without pity, if old
480 * ones are about to clog our table.
482 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
483 int young = (lopt->qlen_young<<1);
486 if (lopt->qlen < young)
493 if (queue->rskq_defer_accept)
494 max_retries = queue->rskq_defer_accept;
496 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
497 i = lopt->clock_hand;
500 reqp=&lopt->syn_table[i];
501 while ((req = *reqp) != NULL) {
502 if (time_after_eq(now, req->expires)) {
503 if ((req->retrans < thresh ||
504 (inet_rsk(req)->acked && req->retrans < max_retries))
505 && !req->rsk_ops->rtx_syn_ack(parent, req)) {
508 if (req->retrans++ == 0)
510 timeo = min((timeout << req->retrans), max_rto);
511 req->expires = now + timeo;
512 reqp = &req->dl_next;
516 /* Drop this request */
517 inet_csk_reqsk_queue_unlink(parent, req, reqp);
518 reqsk_queue_removed(queue, req);
522 reqp = &req->dl_next;
525 i = (i + 1) & (lopt->nr_table_entries - 1);
527 } while (--budget > 0);
529 lopt->clock_hand = i;
532 inet_csk_reset_keepalive_timer(parent, interval);
535 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
537 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
538 const gfp_t priority)
540 struct sock *newsk = sk_clone(sk, priority);
543 struct inet_connection_sock *newicsk = inet_csk(newsk);
545 newsk->sk_state = TCP_SYN_RECV;
546 newicsk->icsk_bind_hash = NULL;
548 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
549 inet_sk(newsk)->num = ntohs(inet_rsk(req)->loc_port);
550 inet_sk(newsk)->sport = inet_rsk(req)->loc_port;
551 newsk->sk_write_space = sk_stream_write_space;
553 newicsk->icsk_retransmits = 0;
554 newicsk->icsk_backoff = 0;
555 newicsk->icsk_probes_out = 0;
557 /* Deinitialize accept_queue to trap illegal accesses. */
558 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
560 security_inet_csk_clone(newsk, req);
565 EXPORT_SYMBOL_GPL(inet_csk_clone);
568 * At this point, there should be no process reference to this
569 * socket, and thus no user references at all. Therefore we
570 * can assume the socket waitqueue is inactive and nobody will
571 * try to jump onto it.
573 void inet_csk_destroy_sock(struct sock *sk)
575 WARN_ON(sk->sk_state != TCP_CLOSE);
576 WARN_ON(!sock_flag(sk, SOCK_DEAD));
578 /* It cannot be in hash table! */
579 WARN_ON(!sk_unhashed(sk));
581 /* If it has not 0 inet_sk(sk)->num, it must be bound */
582 WARN_ON(inet_sk(sk)->num && !inet_csk(sk)->icsk_bind_hash);
584 sk->sk_prot->destroy(sk);
586 sk_stream_kill_queues(sk);
588 xfrm_sk_free_policy(sk);
590 sk_refcnt_debug_release(sk);
592 percpu_counter_dec(sk->sk_prot->orphan_count);
596 EXPORT_SYMBOL(inet_csk_destroy_sock);
598 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
600 struct inet_sock *inet = inet_sk(sk);
601 struct inet_connection_sock *icsk = inet_csk(sk);
602 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
607 sk->sk_max_ack_backlog = 0;
608 sk->sk_ack_backlog = 0;
609 inet_csk_delack_init(sk);
611 /* There is race window here: we announce ourselves listening,
612 * but this transition is still not validated by get_port().
613 * It is OK, because this socket enters to hash table only
614 * after validation is complete.
616 sk->sk_state = TCP_LISTEN;
617 if (!sk->sk_prot->get_port(sk, inet->num)) {
618 inet->sport = htons(inet->num);
621 sk->sk_prot->hash(sk);
626 sk->sk_state = TCP_CLOSE;
627 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
631 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
634 * This routine closes sockets which have been at least partially
635 * opened, but not yet accepted.
637 void inet_csk_listen_stop(struct sock *sk)
639 struct inet_connection_sock *icsk = inet_csk(sk);
640 struct request_sock *acc_req;
641 struct request_sock *req;
643 inet_csk_delete_keepalive_timer(sk);
645 /* make all the listen_opt local to us */
646 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
648 /* Following specs, it would be better either to send FIN
649 * (and enter FIN-WAIT-1, it is normal close)
650 * or to send active reset (abort).
651 * Certainly, it is pretty dangerous while synflood, but it is
652 * bad justification for our negligence 8)
653 * To be honest, we are not able to make either
654 * of the variants now. --ANK
656 reqsk_queue_destroy(&icsk->icsk_accept_queue);
658 while ((req = acc_req) != NULL) {
659 struct sock *child = req->sk;
661 acc_req = req->dl_next;
665 WARN_ON(sock_owned_by_user(child));
668 sk->sk_prot->disconnect(child, O_NONBLOCK);
672 percpu_counter_inc(sk->sk_prot->orphan_count);
674 inet_csk_destroy_sock(child);
676 bh_unlock_sock(child);
680 sk_acceptq_removed(sk);
683 WARN_ON(sk->sk_ack_backlog);
686 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
688 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
690 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
691 const struct inet_sock *inet = inet_sk(sk);
693 sin->sin_family = AF_INET;
694 sin->sin_addr.s_addr = inet->daddr;
695 sin->sin_port = inet->dport;
698 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
701 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
702 char __user *optval, int __user *optlen)
704 const struct inet_connection_sock *icsk = inet_csk(sk);
706 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
707 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
709 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
713 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
715 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
716 char __user *optval, int optlen)
718 const struct inet_connection_sock *icsk = inet_csk(sk);
720 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
721 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
723 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
727 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);