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;
97 struct net *net = sock_net(sk);
101 int remaining, rover, low, high;
103 inet_get_local_port_range(&low, &high);
104 remaining = (high - low) + 1;
105 rover = net_random() % remaining + low;
108 head = &hashinfo->bhash[inet_bhashfn(net, rover,
109 hashinfo->bhash_size)];
110 spin_lock(&head->lock);
111 inet_bind_bucket_for_each(tb, node, &head->chain)
112 if (tb->ib_net == net && tb->port == rover)
116 spin_unlock(&head->lock);
119 } while (--remaining > 0);
121 /* Exhausted local port range during search? It is not
122 * possible for us to be holding one of the bind hash
123 * locks if this test triggers, because if 'remaining'
124 * drops to zero, we broke out of the do/while loop at
125 * the top level, not from the 'break;' statement.
131 /* OK, here is the one we will use. HEAD is
132 * non-NULL and we hold it's mutex.
136 head = &hashinfo->bhash[inet_bhashfn(net, snum,
137 hashinfo->bhash_size)];
138 spin_lock(&head->lock);
139 inet_bind_bucket_for_each(tb, node, &head->chain)
140 if (tb->ib_net == net && tb->port == snum)
146 if (!hlist_empty(&tb->owners)) {
147 if (tb->fastreuse > 0 &&
148 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
152 if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb))
158 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
159 net, head, snum)) == NULL)
161 if (hlist_empty(&tb->owners)) {
162 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
166 } else if (tb->fastreuse &&
167 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
170 if (!inet_csk(sk)->icsk_bind_hash)
171 inet_bind_hash(sk, tb, snum);
172 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
176 spin_unlock(&head->lock);
182 EXPORT_SYMBOL_GPL(inet_csk_get_port);
185 * Wait for an incoming connection, avoid race conditions. This must be called
186 * with the socket locked.
188 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
190 struct inet_connection_sock *icsk = inet_csk(sk);
195 * True wake-one mechanism for incoming connections: only
196 * one process gets woken up, not the 'whole herd'.
197 * Since we do not 'race & poll' for established sockets
198 * anymore, the common case will execute the loop only once.
200 * Subtle issue: "add_wait_queue_exclusive()" will be added
201 * after any current non-exclusive waiters, and we know that
202 * it will always _stay_ after any new non-exclusive waiters
203 * because all non-exclusive waiters are added at the
204 * beginning of the wait-queue. As such, it's ok to "drop"
205 * our exclusiveness temporarily when we get woken up without
206 * having to remove and re-insert us on the wait queue.
209 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
212 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
213 timeo = schedule_timeout(timeo);
216 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
219 if (sk->sk_state != TCP_LISTEN)
221 err = sock_intr_errno(timeo);
222 if (signal_pending(current))
228 finish_wait(sk->sk_sleep, &wait);
233 * This will accept the next outstanding connection.
235 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
237 struct inet_connection_sock *icsk = inet_csk(sk);
243 /* We need to make sure that this socket is listening,
244 * and that it has something pending.
247 if (sk->sk_state != TCP_LISTEN)
250 /* Find already established connection */
251 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
252 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
254 /* If this is a non blocking socket don't sleep */
259 error = inet_csk_wait_for_connect(sk, timeo);
264 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
265 WARN_ON(newsk->sk_state == TCP_SYN_RECV);
275 EXPORT_SYMBOL(inet_csk_accept);
278 * Using different timers for retransmit, delayed acks and probes
279 * We may wish use just one timer maintaining a list of expire jiffies
282 void inet_csk_init_xmit_timers(struct sock *sk,
283 void (*retransmit_handler)(unsigned long),
284 void (*delack_handler)(unsigned long),
285 void (*keepalive_handler)(unsigned long))
287 struct inet_connection_sock *icsk = inet_csk(sk);
289 setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
291 setup_timer(&icsk->icsk_delack_timer, delack_handler,
293 setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
294 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
297 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
299 void inet_csk_clear_xmit_timers(struct sock *sk)
301 struct inet_connection_sock *icsk = inet_csk(sk);
303 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
305 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
306 sk_stop_timer(sk, &icsk->icsk_delack_timer);
307 sk_stop_timer(sk, &sk->sk_timer);
310 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
312 void inet_csk_delete_keepalive_timer(struct sock *sk)
314 sk_stop_timer(sk, &sk->sk_timer);
317 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
319 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
321 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
324 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
326 struct dst_entry *inet_csk_route_req(struct sock *sk,
327 const struct request_sock *req)
330 const struct inet_request_sock *ireq = inet_rsk(req);
331 struct ip_options *opt = inet_rsk(req)->opt;
332 struct flowi fl = { .oif = sk->sk_bound_dev_if,
334 { .daddr = ((opt && opt->srr) ?
337 .saddr = ireq->loc_addr,
338 .tos = RT_CONN_FLAGS(sk) } },
339 .proto = sk->sk_protocol,
340 .flags = inet_sk_flowi_flags(sk),
342 { .sport = inet_sk(sk)->sport,
343 .dport = ireq->rmt_port } } };
344 struct net *net = sock_net(sk);
346 security_req_classify_flow(req, &fl);
347 if (ip_route_output_flow(net, &rt, &fl, sk, 0)) {
348 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
351 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
353 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
359 EXPORT_SYMBOL_GPL(inet_csk_route_req);
361 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
362 const u32 rnd, const u32 synq_hsize)
364 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
367 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
368 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
370 #define AF_INET_FAMILY(fam) 1
373 struct request_sock *inet_csk_search_req(const struct sock *sk,
374 struct request_sock ***prevp,
375 const __be16 rport, const __be32 raddr,
378 const struct inet_connection_sock *icsk = inet_csk(sk);
379 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
380 struct request_sock *req, **prev;
382 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
383 lopt->nr_table_entries)];
384 (req = *prev) != NULL;
385 prev = &req->dl_next) {
386 const struct inet_request_sock *ireq = inet_rsk(req);
388 if (ireq->rmt_port == rport &&
389 ireq->rmt_addr == raddr &&
390 ireq->loc_addr == laddr &&
391 AF_INET_FAMILY(req->rsk_ops->family)) {
401 EXPORT_SYMBOL_GPL(inet_csk_search_req);
403 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
404 unsigned long timeout)
406 struct inet_connection_sock *icsk = inet_csk(sk);
407 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
408 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
409 lopt->hash_rnd, lopt->nr_table_entries);
411 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
412 inet_csk_reqsk_queue_added(sk, timeout);
415 /* Only thing we need from tcp.h */
416 extern int sysctl_tcp_synack_retries;
418 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
420 void inet_csk_reqsk_queue_prune(struct sock *parent,
421 const unsigned long interval,
422 const unsigned long timeout,
423 const unsigned long max_rto)
425 struct inet_connection_sock *icsk = inet_csk(parent);
426 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
427 struct listen_sock *lopt = queue->listen_opt;
428 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
429 int thresh = max_retries;
430 unsigned long now = jiffies;
431 struct request_sock **reqp, *req;
434 if (lopt == NULL || lopt->qlen == 0)
437 /* Normally all the openreqs are young and become mature
438 * (i.e. converted to established socket) for first timeout.
439 * If synack was not acknowledged for 3 seconds, it means
440 * one of the following things: synack was lost, ack was lost,
441 * rtt is high or nobody planned to ack (i.e. synflood).
442 * When server is a bit loaded, queue is populated with old
443 * open requests, reducing effective size of queue.
444 * When server is well loaded, queue size reduces to zero
445 * after several minutes of work. It is not synflood,
446 * it is normal operation. The solution is pruning
447 * too old entries overriding normal timeout, when
448 * situation becomes dangerous.
450 * Essentially, we reserve half of room for young
451 * embrions; and abort old ones without pity, if old
452 * ones are about to clog our table.
454 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
455 int young = (lopt->qlen_young<<1);
458 if (lopt->qlen < young)
465 if (queue->rskq_defer_accept)
466 max_retries = queue->rskq_defer_accept;
468 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
469 i = lopt->clock_hand;
472 reqp=&lopt->syn_table[i];
473 while ((req = *reqp) != NULL) {
474 if (time_after_eq(now, req->expires)) {
475 if ((req->retrans < thresh ||
476 (inet_rsk(req)->acked && req->retrans < max_retries))
477 && !req->rsk_ops->rtx_syn_ack(parent, req)) {
480 if (req->retrans++ == 0)
482 timeo = min((timeout << req->retrans), max_rto);
483 req->expires = now + timeo;
484 reqp = &req->dl_next;
488 /* Drop this request */
489 inet_csk_reqsk_queue_unlink(parent, req, reqp);
490 reqsk_queue_removed(queue, req);
494 reqp = &req->dl_next;
497 i = (i + 1) & (lopt->nr_table_entries - 1);
499 } while (--budget > 0);
501 lopt->clock_hand = i;
504 inet_csk_reset_keepalive_timer(parent, interval);
507 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
509 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
510 const gfp_t priority)
512 struct sock *newsk = sk_clone(sk, priority);
515 struct inet_connection_sock *newicsk = inet_csk(newsk);
517 newsk->sk_state = TCP_SYN_RECV;
518 newicsk->icsk_bind_hash = NULL;
520 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
521 inet_sk(newsk)->num = ntohs(inet_rsk(req)->loc_port);
522 inet_sk(newsk)->sport = inet_rsk(req)->loc_port;
523 newsk->sk_write_space = sk_stream_write_space;
525 newicsk->icsk_retransmits = 0;
526 newicsk->icsk_backoff = 0;
527 newicsk->icsk_probes_out = 0;
529 /* Deinitialize accept_queue to trap illegal accesses. */
530 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
532 security_inet_csk_clone(newsk, req);
537 EXPORT_SYMBOL_GPL(inet_csk_clone);
540 * At this point, there should be no process reference to this
541 * socket, and thus no user references at all. Therefore we
542 * can assume the socket waitqueue is inactive and nobody will
543 * try to jump onto it.
545 void inet_csk_destroy_sock(struct sock *sk)
547 WARN_ON(sk->sk_state != TCP_CLOSE);
548 WARN_ON(!sock_flag(sk, SOCK_DEAD));
550 /* It cannot be in hash table! */
551 WARN_ON(!sk_unhashed(sk));
553 /* If it has not 0 inet_sk(sk)->num, it must be bound */
554 WARN_ON(inet_sk(sk)->num && !inet_csk(sk)->icsk_bind_hash);
556 sk->sk_prot->destroy(sk);
558 sk_stream_kill_queues(sk);
560 xfrm_sk_free_policy(sk);
562 sk_refcnt_debug_release(sk);
564 atomic_dec(sk->sk_prot->orphan_count);
568 EXPORT_SYMBOL(inet_csk_destroy_sock);
570 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
572 struct inet_sock *inet = inet_sk(sk);
573 struct inet_connection_sock *icsk = inet_csk(sk);
574 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
579 sk->sk_max_ack_backlog = 0;
580 sk->sk_ack_backlog = 0;
581 inet_csk_delack_init(sk);
583 /* There is race window here: we announce ourselves listening,
584 * but this transition is still not validated by get_port().
585 * It is OK, because this socket enters to hash table only
586 * after validation is complete.
588 sk->sk_state = TCP_LISTEN;
589 if (!sk->sk_prot->get_port(sk, inet->num)) {
590 inet->sport = htons(inet->num);
593 sk->sk_prot->hash(sk);
598 sk->sk_state = TCP_CLOSE;
599 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
603 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
606 * This routine closes sockets which have been at least partially
607 * opened, but not yet accepted.
609 void inet_csk_listen_stop(struct sock *sk)
611 struct inet_connection_sock *icsk = inet_csk(sk);
612 struct request_sock *acc_req;
613 struct request_sock *req;
615 inet_csk_delete_keepalive_timer(sk);
617 /* make all the listen_opt local to us */
618 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
620 /* Following specs, it would be better either to send FIN
621 * (and enter FIN-WAIT-1, it is normal close)
622 * or to send active reset (abort).
623 * Certainly, it is pretty dangerous while synflood, but it is
624 * bad justification for our negligence 8)
625 * To be honest, we are not able to make either
626 * of the variants now. --ANK
628 reqsk_queue_destroy(&icsk->icsk_accept_queue);
630 while ((req = acc_req) != NULL) {
631 struct sock *child = req->sk;
633 acc_req = req->dl_next;
637 WARN_ON(sock_owned_by_user(child));
640 sk->sk_prot->disconnect(child, O_NONBLOCK);
644 atomic_inc(sk->sk_prot->orphan_count);
646 inet_csk_destroy_sock(child);
648 bh_unlock_sock(child);
652 sk_acceptq_removed(sk);
655 WARN_ON(sk->sk_ack_backlog);
658 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
660 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
662 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
663 const struct inet_sock *inet = inet_sk(sk);
665 sin->sin_family = AF_INET;
666 sin->sin_addr.s_addr = inet->daddr;
667 sin->sin_port = inet->dport;
670 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
673 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
674 char __user *optval, int __user *optlen)
676 const struct inet_connection_sock *icsk = inet_csk(sk);
678 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
679 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
681 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
685 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
687 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
688 char __user *optval, int optlen)
690 const struct inet_connection_sock *icsk = inet_csk(sk);
692 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
693 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
695 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
699 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);