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 inet_hashinfo *hashinfo,
82 struct sock *sk, unsigned short snum,
83 int (*bind_conflict)(const struct sock *sk,
84 const struct inet_bind_bucket *tb))
86 struct inet_bind_hashbucket *head;
87 struct hlist_node *node;
88 struct inet_bind_bucket *tb;
93 int remaining, rover, low, high;
95 inet_get_local_port_range(&low, &high);
96 remaining = high - low;
97 rover = net_random() % remaining + low;
100 head = &hashinfo->bhash[inet_bhashfn(rover, hashinfo->bhash_size)];
101 spin_lock(&head->lock);
102 inet_bind_bucket_for_each(tb, node, &head->chain)
103 if (tb->port == rover)
107 spin_unlock(&head->lock);
110 } while (--remaining > 0);
112 /* Exhausted local port range during search? It is not
113 * possible for us to be holding one of the bind hash
114 * locks if this test triggers, because if 'remaining'
115 * drops to zero, we broke out of the do/while loop at
116 * the top level, not from the 'break;' statement.
122 /* OK, here is the one we will use. HEAD is
123 * non-NULL and we hold it's mutex.
127 head = &hashinfo->bhash[inet_bhashfn(snum, hashinfo->bhash_size)];
128 spin_lock(&head->lock);
129 inet_bind_bucket_for_each(tb, node, &head->chain)
130 if (tb->port == snum)
136 if (!hlist_empty(&tb->owners)) {
137 if (sk->sk_reuse > 1)
139 if (tb->fastreuse > 0 &&
140 sk->sk_reuse && sk->sk_state != TCP_LISTEN) {
144 if (bind_conflict(sk, tb))
150 if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep, head, snum)) == NULL)
152 if (hlist_empty(&tb->owners)) {
153 if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
157 } else if (tb->fastreuse &&
158 (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
161 if (!inet_csk(sk)->icsk_bind_hash)
162 inet_bind_hash(sk, tb, snum);
163 BUG_TRAP(inet_csk(sk)->icsk_bind_hash == tb);
167 spin_unlock(&head->lock);
173 EXPORT_SYMBOL_GPL(inet_csk_get_port);
176 * Wait for an incoming connection, avoid race conditions. This must be called
177 * with the socket locked.
179 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
181 struct inet_connection_sock *icsk = inet_csk(sk);
186 * True wake-one mechanism for incoming connections: only
187 * one process gets woken up, not the 'whole herd'.
188 * Since we do not 'race & poll' for established sockets
189 * anymore, the common case will execute the loop only once.
191 * Subtle issue: "add_wait_queue_exclusive()" will be added
192 * after any current non-exclusive waiters, and we know that
193 * it will always _stay_ after any new non-exclusive waiters
194 * because all non-exclusive waiters are added at the
195 * beginning of the wait-queue. As such, it's ok to "drop"
196 * our exclusiveness temporarily when we get woken up without
197 * having to remove and re-insert us on the wait queue.
200 prepare_to_wait_exclusive(sk->sk_sleep, &wait,
203 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
204 timeo = schedule_timeout(timeo);
207 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
210 if (sk->sk_state != TCP_LISTEN)
212 err = sock_intr_errno(timeo);
213 if (signal_pending(current))
219 finish_wait(sk->sk_sleep, &wait);
224 * This will accept the next outstanding connection.
226 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
228 struct inet_connection_sock *icsk = inet_csk(sk);
234 /* We need to make sure that this socket is listening,
235 * and that it has something pending.
238 if (sk->sk_state != TCP_LISTEN)
241 /* Find already established connection */
242 if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
243 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
245 /* If this is a non blocking socket don't sleep */
250 error = inet_csk_wait_for_connect(sk, timeo);
255 newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
256 BUG_TRAP(newsk->sk_state != TCP_SYN_RECV);
266 EXPORT_SYMBOL(inet_csk_accept);
269 * Using different timers for retransmit, delayed acks and probes
270 * We may wish use just one timer maintaining a list of expire jiffies
273 void inet_csk_init_xmit_timers(struct sock *sk,
274 void (*retransmit_handler)(unsigned long),
275 void (*delack_handler)(unsigned long),
276 void (*keepalive_handler)(unsigned long))
278 struct inet_connection_sock *icsk = inet_csk(sk);
280 init_timer(&icsk->icsk_retransmit_timer);
281 init_timer(&icsk->icsk_delack_timer);
282 init_timer(&sk->sk_timer);
284 icsk->icsk_retransmit_timer.function = retransmit_handler;
285 icsk->icsk_delack_timer.function = delack_handler;
286 sk->sk_timer.function = keepalive_handler;
288 icsk->icsk_retransmit_timer.data =
289 icsk->icsk_delack_timer.data =
290 sk->sk_timer.data = (unsigned long)sk;
292 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
295 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
297 void inet_csk_clear_xmit_timers(struct sock *sk)
299 struct inet_connection_sock *icsk = inet_csk(sk);
301 icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
303 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
304 sk_stop_timer(sk, &icsk->icsk_delack_timer);
305 sk_stop_timer(sk, &sk->sk_timer);
308 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
310 void inet_csk_delete_keepalive_timer(struct sock *sk)
312 sk_stop_timer(sk, &sk->sk_timer);
315 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
317 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
319 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
322 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
324 struct dst_entry* inet_csk_route_req(struct sock *sk,
325 const struct request_sock *req)
328 const struct inet_request_sock *ireq = inet_rsk(req);
329 struct ip_options *opt = inet_rsk(req)->opt;
330 struct flowi fl = { .oif = sk->sk_bound_dev_if,
332 { .daddr = ((opt && opt->srr) ?
335 .saddr = ireq->loc_addr,
336 .tos = RT_CONN_FLAGS(sk) } },
337 .proto = sk->sk_protocol,
339 { .sport = inet_sk(sk)->sport,
340 .dport = ireq->rmt_port } } };
342 security_req_classify_flow(req, &fl);
343 if (ip_route_output_flow(&rt, &fl, sk, 0)) {
344 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
347 if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
349 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
355 EXPORT_SYMBOL_GPL(inet_csk_route_req);
357 static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
358 const u32 rnd, const u32 synq_hsize)
360 return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
363 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
364 #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
366 #define AF_INET_FAMILY(fam) 1
369 struct request_sock *inet_csk_search_req(const struct sock *sk,
370 struct request_sock ***prevp,
371 const __be16 rport, const __be32 raddr,
374 const struct inet_connection_sock *icsk = inet_csk(sk);
375 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
376 struct request_sock *req, **prev;
378 for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
379 lopt->nr_table_entries)];
380 (req = *prev) != NULL;
381 prev = &req->dl_next) {
382 const struct inet_request_sock *ireq = inet_rsk(req);
384 if (ireq->rmt_port == rport &&
385 ireq->rmt_addr == raddr &&
386 ireq->loc_addr == laddr &&
387 AF_INET_FAMILY(req->rsk_ops->family)) {
397 EXPORT_SYMBOL_GPL(inet_csk_search_req);
399 void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
400 unsigned long timeout)
402 struct inet_connection_sock *icsk = inet_csk(sk);
403 struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
404 const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
405 lopt->hash_rnd, lopt->nr_table_entries);
407 reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
408 inet_csk_reqsk_queue_added(sk, timeout);
411 /* Only thing we need from tcp.h */
412 extern int sysctl_tcp_synack_retries;
414 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
416 void inet_csk_reqsk_queue_prune(struct sock *parent,
417 const unsigned long interval,
418 const unsigned long timeout,
419 const unsigned long max_rto)
421 struct inet_connection_sock *icsk = inet_csk(parent);
422 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
423 struct listen_sock *lopt = queue->listen_opt;
424 int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
425 int thresh = max_retries;
426 unsigned long now = jiffies;
427 struct request_sock **reqp, *req;
430 if (lopt == NULL || lopt->qlen == 0)
433 /* Normally all the openreqs are young and become mature
434 * (i.e. converted to established socket) for first timeout.
435 * If synack was not acknowledged for 3 seconds, it means
436 * one of the following things: synack was lost, ack was lost,
437 * rtt is high or nobody planned to ack (i.e. synflood).
438 * When server is a bit loaded, queue is populated with old
439 * open requests, reducing effective size of queue.
440 * When server is well loaded, queue size reduces to zero
441 * after several minutes of work. It is not synflood,
442 * it is normal operation. The solution is pruning
443 * too old entries overriding normal timeout, when
444 * situation becomes dangerous.
446 * Essentially, we reserve half of room for young
447 * embrions; and abort old ones without pity, if old
448 * ones are about to clog our table.
450 if (lopt->qlen>>(lopt->max_qlen_log-1)) {
451 int young = (lopt->qlen_young<<1);
454 if (lopt->qlen < young)
461 if (queue->rskq_defer_accept)
462 max_retries = queue->rskq_defer_accept;
464 budget = 2 * (lopt->nr_table_entries / (timeout / interval));
465 i = lopt->clock_hand;
468 reqp=&lopt->syn_table[i];
469 while ((req = *reqp) != NULL) {
470 if (time_after_eq(now, req->expires)) {
471 if ((req->retrans < thresh ||
472 (inet_rsk(req)->acked && req->retrans < max_retries))
473 && !req->rsk_ops->rtx_syn_ack(parent, req, NULL)) {
476 if (req->retrans++ == 0)
478 timeo = min((timeout << req->retrans), max_rto);
479 req->expires = now + timeo;
480 reqp = &req->dl_next;
484 /* Drop this request */
485 inet_csk_reqsk_queue_unlink(parent, req, reqp);
486 reqsk_queue_removed(queue, req);
490 reqp = &req->dl_next;
493 i = (i + 1) & (lopt->nr_table_entries - 1);
495 } while (--budget > 0);
497 lopt->clock_hand = i;
500 inet_csk_reset_keepalive_timer(parent, interval);
503 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
505 struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
506 const gfp_t priority)
508 struct sock *newsk = sk_clone(sk, priority);
511 struct inet_connection_sock *newicsk = inet_csk(newsk);
513 newsk->sk_state = TCP_SYN_RECV;
514 newicsk->icsk_bind_hash = NULL;
516 inet_sk(newsk)->dport = inet_rsk(req)->rmt_port;
517 newsk->sk_write_space = sk_stream_write_space;
519 newicsk->icsk_retransmits = 0;
520 newicsk->icsk_backoff = 0;
521 newicsk->icsk_probes_out = 0;
523 /* Deinitialize accept_queue to trap illegal accesses. */
524 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
526 security_inet_csk_clone(newsk, req);
531 EXPORT_SYMBOL_GPL(inet_csk_clone);
534 * At this point, there should be no process reference to this
535 * socket, and thus no user references at all. Therefore we
536 * can assume the socket waitqueue is inactive and nobody will
537 * try to jump onto it.
539 void inet_csk_destroy_sock(struct sock *sk)
541 BUG_TRAP(sk->sk_state == TCP_CLOSE);
542 BUG_TRAP(sock_flag(sk, SOCK_DEAD));
544 /* It cannot be in hash table! */
545 BUG_TRAP(sk_unhashed(sk));
547 /* If it has not 0 inet_sk(sk)->num, it must be bound */
548 BUG_TRAP(!inet_sk(sk)->num || inet_csk(sk)->icsk_bind_hash);
550 sk->sk_prot->destroy(sk);
552 sk_stream_kill_queues(sk);
554 xfrm_sk_free_policy(sk);
556 sk_refcnt_debug_release(sk);
558 atomic_dec(sk->sk_prot->orphan_count);
562 EXPORT_SYMBOL(inet_csk_destroy_sock);
564 int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
566 struct inet_sock *inet = inet_sk(sk);
567 struct inet_connection_sock *icsk = inet_csk(sk);
568 int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
573 sk->sk_max_ack_backlog = 0;
574 sk->sk_ack_backlog = 0;
575 inet_csk_delack_init(sk);
577 /* There is race window here: we announce ourselves listening,
578 * but this transition is still not validated by get_port().
579 * It is OK, because this socket enters to hash table only
580 * after validation is complete.
582 sk->sk_state = TCP_LISTEN;
583 if (!sk->sk_prot->get_port(sk, inet->num)) {
584 inet->sport = htons(inet->num);
587 sk->sk_prot->hash(sk);
592 sk->sk_state = TCP_CLOSE;
593 __reqsk_queue_destroy(&icsk->icsk_accept_queue);
597 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
600 * This routine closes sockets which have been at least partially
601 * opened, but not yet accepted.
603 void inet_csk_listen_stop(struct sock *sk)
605 struct inet_connection_sock *icsk = inet_csk(sk);
606 struct request_sock *acc_req;
607 struct request_sock *req;
609 inet_csk_delete_keepalive_timer(sk);
611 /* make all the listen_opt local to us */
612 acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
614 /* Following specs, it would be better either to send FIN
615 * (and enter FIN-WAIT-1, it is normal close)
616 * or to send active reset (abort).
617 * Certainly, it is pretty dangerous while synflood, but it is
618 * bad justification for our negligence 8)
619 * To be honest, we are not able to make either
620 * of the variants now. --ANK
622 reqsk_queue_destroy(&icsk->icsk_accept_queue);
624 while ((req = acc_req) != NULL) {
625 struct sock *child = req->sk;
627 acc_req = req->dl_next;
631 BUG_TRAP(!sock_owned_by_user(child));
634 sk->sk_prot->disconnect(child, O_NONBLOCK);
638 atomic_inc(sk->sk_prot->orphan_count);
640 inet_csk_destroy_sock(child);
642 bh_unlock_sock(child);
646 sk_acceptq_removed(sk);
649 BUG_TRAP(!sk->sk_ack_backlog);
652 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
654 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
656 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
657 const struct inet_sock *inet = inet_sk(sk);
659 sin->sin_family = AF_INET;
660 sin->sin_addr.s_addr = inet->daddr;
661 sin->sin_port = inet->dport;
664 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
666 int inet_csk_ctl_sock_create(struct socket **sock, unsigned short family,
667 unsigned short type, unsigned char protocol)
669 int rc = sock_create_kern(family, type, protocol, sock);
672 (*sock)->sk->sk_allocation = GFP_ATOMIC;
673 inet_sk((*sock)->sk)->uc_ttl = -1;
675 * Unhash it so that IP input processing does not even see it,
676 * we do not wish this socket to see incoming packets.
678 (*sock)->sk->sk_prot->unhash((*sock)->sk);
683 EXPORT_SYMBOL_GPL(inet_csk_ctl_sock_create);
686 int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
687 char __user *optval, int __user *optlen)
689 const struct inet_connection_sock *icsk = inet_csk(sk);
691 if (icsk->icsk_af_ops->compat_getsockopt != NULL)
692 return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
694 return icsk->icsk_af_ops->getsockopt(sk, level, optname,
698 EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
700 int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
701 char __user *optval, int optlen)
703 const struct inet_connection_sock *icsk = inet_csk(sk);
705 if (icsk->icsk_af_ops->compat_setsockopt != NULL)
706 return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
708 return icsk->icsk_af_ops->setsockopt(sk, level, optname,
712 EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);