xref: /linux/net/ipv4/inet_connection_sock.c (revision 09ad9bc752519cc167d0a573e1acf69b5c707c67)
13f421baaSArnaldo Carvalho de Melo /*
23f421baaSArnaldo Carvalho de Melo  * INET		An implementation of the TCP/IP protocol suite for the LINUX
33f421baaSArnaldo Carvalho de Melo  *		operating system.  INET is implemented using the  BSD Socket
43f421baaSArnaldo Carvalho de Melo  *		interface as the means of communication with the user level.
53f421baaSArnaldo Carvalho de Melo  *
63f421baaSArnaldo Carvalho de Melo  *		Support for INET connection oriented protocols.
73f421baaSArnaldo Carvalho de Melo  *
83f421baaSArnaldo Carvalho de Melo  * Authors:	See the TCP sources
93f421baaSArnaldo Carvalho de Melo  *
103f421baaSArnaldo Carvalho de Melo  *		This program is free software; you can redistribute it and/or
113f421baaSArnaldo Carvalho de Melo  *		modify it under the terms of the GNU General Public License
123f421baaSArnaldo Carvalho de Melo  *		as published by the Free Software Foundation; either version
133f421baaSArnaldo Carvalho de Melo  *		2 of the License, or(at your option) any later version.
143f421baaSArnaldo Carvalho de Melo  */
153f421baaSArnaldo Carvalho de Melo 
163f421baaSArnaldo Carvalho de Melo #include <linux/module.h>
173f421baaSArnaldo Carvalho de Melo #include <linux/jhash.h>
183f421baaSArnaldo Carvalho de Melo 
193f421baaSArnaldo Carvalho de Melo #include <net/inet_connection_sock.h>
203f421baaSArnaldo Carvalho de Melo #include <net/inet_hashtables.h>
213f421baaSArnaldo Carvalho de Melo #include <net/inet_timewait_sock.h>
223f421baaSArnaldo Carvalho de Melo #include <net/ip.h>
233f421baaSArnaldo Carvalho de Melo #include <net/route.h>
243f421baaSArnaldo Carvalho de Melo #include <net/tcp_states.h>
25a019d6feSArnaldo Carvalho de Melo #include <net/xfrm.h>
263f421baaSArnaldo Carvalho de Melo 
273f421baaSArnaldo Carvalho de Melo #ifdef INET_CSK_DEBUG
283f421baaSArnaldo Carvalho de Melo const char inet_csk_timer_bug_msg[] = "inet_csk BUG: unknown timer value\n";
293f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_timer_bug_msg);
303f421baaSArnaldo Carvalho de Melo #endif
313f421baaSArnaldo Carvalho de Melo 
323f421baaSArnaldo Carvalho de Melo /*
333c689b73SEric Dumazet  * This struct holds the first and last local port number.
343f421baaSArnaldo Carvalho de Melo  */
353c689b73SEric Dumazet struct local_ports sysctl_local_ports __read_mostly = {
363c689b73SEric Dumazet 	.lock = SEQLOCK_UNLOCKED,
373c689b73SEric Dumazet 	.range = { 32768, 61000 },
383c689b73SEric Dumazet };
39227b60f5SStephen Hemminger 
40227b60f5SStephen Hemminger void inet_get_local_port_range(int *low, int *high)
41227b60f5SStephen Hemminger {
42227b60f5SStephen Hemminger 	unsigned seq;
43227b60f5SStephen Hemminger 	do {
443c689b73SEric Dumazet 		seq = read_seqbegin(&sysctl_local_ports.lock);
45227b60f5SStephen Hemminger 
463c689b73SEric Dumazet 		*low = sysctl_local_ports.range[0];
473c689b73SEric Dumazet 		*high = sysctl_local_ports.range[1];
483c689b73SEric Dumazet 	} while (read_seqretry(&sysctl_local_ports.lock, seq));
49227b60f5SStephen Hemminger }
50227b60f5SStephen Hemminger EXPORT_SYMBOL(inet_get_local_port_range);
513f421baaSArnaldo Carvalho de Melo 
52971af18bSArnaldo Carvalho de Melo int inet_csk_bind_conflict(const struct sock *sk,
53971af18bSArnaldo Carvalho de Melo 			   const struct inet_bind_bucket *tb)
543f421baaSArnaldo Carvalho de Melo {
5582103232SAl Viro 	const __be32 sk_rcv_saddr = inet_rcv_saddr(sk);
563f421baaSArnaldo Carvalho de Melo 	struct sock *sk2;
573f421baaSArnaldo Carvalho de Melo 	struct hlist_node *node;
583f421baaSArnaldo Carvalho de Melo 	int reuse = sk->sk_reuse;
593f421baaSArnaldo Carvalho de Melo 
607477fd2eSPavel Emelyanov 	/*
617477fd2eSPavel Emelyanov 	 * Unlike other sk lookup places we do not check
627477fd2eSPavel Emelyanov 	 * for sk_net here, since _all_ the socks listed
637477fd2eSPavel Emelyanov 	 * in tb->owners list belong to the same net - the
647477fd2eSPavel Emelyanov 	 * one this bucket belongs to.
657477fd2eSPavel Emelyanov 	 */
667477fd2eSPavel Emelyanov 
673f421baaSArnaldo Carvalho de Melo 	sk_for_each_bound(sk2, node, &tb->owners) {
683f421baaSArnaldo Carvalho de Melo 		if (sk != sk2 &&
693f421baaSArnaldo Carvalho de Melo 		    !inet_v6_ipv6only(sk2) &&
703f421baaSArnaldo Carvalho de Melo 		    (!sk->sk_bound_dev_if ||
713f421baaSArnaldo Carvalho de Melo 		     !sk2->sk_bound_dev_if ||
723f421baaSArnaldo Carvalho de Melo 		     sk->sk_bound_dev_if == sk2->sk_bound_dev_if)) {
733f421baaSArnaldo Carvalho de Melo 			if (!reuse || !sk2->sk_reuse ||
743f421baaSArnaldo Carvalho de Melo 			    sk2->sk_state == TCP_LISTEN) {
7582103232SAl Viro 				const __be32 sk2_rcv_saddr = inet_rcv_saddr(sk2);
763f421baaSArnaldo Carvalho de Melo 				if (!sk2_rcv_saddr || !sk_rcv_saddr ||
773f421baaSArnaldo Carvalho de Melo 				    sk2_rcv_saddr == sk_rcv_saddr)
783f421baaSArnaldo Carvalho de Melo 					break;
793f421baaSArnaldo Carvalho de Melo 			}
803f421baaSArnaldo Carvalho de Melo 		}
813f421baaSArnaldo Carvalho de Melo 	}
823f421baaSArnaldo Carvalho de Melo 	return node != NULL;
833f421baaSArnaldo Carvalho de Melo }
843f421baaSArnaldo Carvalho de Melo 
85971af18bSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_bind_conflict);
86971af18bSArnaldo Carvalho de Melo 
873f421baaSArnaldo Carvalho de Melo /* Obtain a reference to a local port for the given sock,
883f421baaSArnaldo Carvalho de Melo  * if snum is zero it means select any available local port.
893f421baaSArnaldo Carvalho de Melo  */
90ab1e0a13SArnaldo Carvalho de Melo int inet_csk_get_port(struct sock *sk, unsigned short snum)
913f421baaSArnaldo Carvalho de Melo {
9239d8cda7SPavel Emelyanov 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
933f421baaSArnaldo Carvalho de Melo 	struct inet_bind_hashbucket *head;
943f421baaSArnaldo Carvalho de Melo 	struct hlist_node *node;
953f421baaSArnaldo Carvalho de Melo 	struct inet_bind_bucket *tb;
96a9d8f911SEvgeniy Polyakov 	int ret, attempts = 5;
973b1e0a65SYOSHIFUJI Hideaki 	struct net *net = sock_net(sk);
98a9d8f911SEvgeniy Polyakov 	int smallest_size = -1, smallest_rover;
993f421baaSArnaldo Carvalho de Melo 
1003f421baaSArnaldo Carvalho de Melo 	local_bh_disable();
1013f421baaSArnaldo Carvalho de Melo 	if (!snum) {
102227b60f5SStephen Hemminger 		int remaining, rover, low, high;
103227b60f5SStephen Hemminger 
104a9d8f911SEvgeniy Polyakov again:
105227b60f5SStephen Hemminger 		inet_get_local_port_range(&low, &high);
106a25de534SAnton Arapov 		remaining = (high - low) + 1;
107a9d8f911SEvgeniy Polyakov 		smallest_rover = rover = net_random() % remaining + low;
1083f421baaSArnaldo Carvalho de Melo 
109a9d8f911SEvgeniy Polyakov 		smallest_size = -1;
1103f421baaSArnaldo Carvalho de Melo 		do {
1117f635ab7SPavel Emelyanov 			head = &hashinfo->bhash[inet_bhashfn(net, rover,
1127f635ab7SPavel Emelyanov 					hashinfo->bhash_size)];
1133f421baaSArnaldo Carvalho de Melo 			spin_lock(&head->lock);
1143f421baaSArnaldo Carvalho de Melo 			inet_bind_bucket_for_each(tb, node, &head->chain)
115*09ad9bc7SOctavian Purdila 				if (net_eq(ib_net(tb), net) && tb->port == rover) {
116a9d8f911SEvgeniy Polyakov 					if (tb->fastreuse > 0 &&
117a9d8f911SEvgeniy Polyakov 					    sk->sk_reuse &&
118a9d8f911SEvgeniy Polyakov 					    sk->sk_state != TCP_LISTEN &&
119a9d8f911SEvgeniy Polyakov 					    (tb->num_owners < smallest_size || smallest_size == -1)) {
120a9d8f911SEvgeniy Polyakov 						smallest_size = tb->num_owners;
121a9d8f911SEvgeniy Polyakov 						smallest_rover = rover;
12224dd1fa1SEric Dumazet 						if (atomic_read(&hashinfo->bsockets) > (high - low) + 1) {
123a9d8f911SEvgeniy Polyakov 							spin_unlock(&head->lock);
124a9d8f911SEvgeniy Polyakov 							snum = smallest_rover;
125a9d8f911SEvgeniy Polyakov 							goto have_snum;
126a9d8f911SEvgeniy Polyakov 						}
127a9d8f911SEvgeniy Polyakov 					}
1283f421baaSArnaldo Carvalho de Melo 					goto next;
129a9d8f911SEvgeniy Polyakov 				}
1303f421baaSArnaldo Carvalho de Melo 			break;
1313f421baaSArnaldo Carvalho de Melo 		next:
1323f421baaSArnaldo Carvalho de Melo 			spin_unlock(&head->lock);
1336df71634SStephen Hemminger 			if (++rover > high)
1346df71634SStephen Hemminger 				rover = low;
1353f421baaSArnaldo Carvalho de Melo 		} while (--remaining > 0);
1363f421baaSArnaldo Carvalho de Melo 
1373f421baaSArnaldo Carvalho de Melo 		/* Exhausted local port range during search?  It is not
1383f421baaSArnaldo Carvalho de Melo 		 * possible for us to be holding one of the bind hash
1393f421baaSArnaldo Carvalho de Melo 		 * locks if this test triggers, because if 'remaining'
1403f421baaSArnaldo Carvalho de Melo 		 * drops to zero, we broke out of the do/while loop at
1413f421baaSArnaldo Carvalho de Melo 		 * the top level, not from the 'break;' statement.
1423f421baaSArnaldo Carvalho de Melo 		 */
1433f421baaSArnaldo Carvalho de Melo 		ret = 1;
144a9d8f911SEvgeniy Polyakov 		if (remaining <= 0) {
145a9d8f911SEvgeniy Polyakov 			if (smallest_size != -1) {
146a9d8f911SEvgeniy Polyakov 				snum = smallest_rover;
147a9d8f911SEvgeniy Polyakov 				goto have_snum;
148a9d8f911SEvgeniy Polyakov 			}
1493f421baaSArnaldo Carvalho de Melo 			goto fail;
150a9d8f911SEvgeniy Polyakov 		}
1513f421baaSArnaldo Carvalho de Melo 		/* OK, here is the one we will use.  HEAD is
1523f421baaSArnaldo Carvalho de Melo 		 * non-NULL and we hold it's mutex.
1533f421baaSArnaldo Carvalho de Melo 		 */
1543f421baaSArnaldo Carvalho de Melo 		snum = rover;
1553f421baaSArnaldo Carvalho de Melo 	} else {
156a9d8f911SEvgeniy Polyakov have_snum:
1577f635ab7SPavel Emelyanov 		head = &hashinfo->bhash[inet_bhashfn(net, snum,
1587f635ab7SPavel Emelyanov 				hashinfo->bhash_size)];
1593f421baaSArnaldo Carvalho de Melo 		spin_lock(&head->lock);
1603f421baaSArnaldo Carvalho de Melo 		inet_bind_bucket_for_each(tb, node, &head->chain)
161*09ad9bc7SOctavian Purdila 			if (net_eq(ib_net(tb), net) && tb->port == snum)
1623f421baaSArnaldo Carvalho de Melo 				goto tb_found;
1633f421baaSArnaldo Carvalho de Melo 	}
1643f421baaSArnaldo Carvalho de Melo 	tb = NULL;
1653f421baaSArnaldo Carvalho de Melo 	goto tb_not_found;
1663f421baaSArnaldo Carvalho de Melo tb_found:
1673f421baaSArnaldo Carvalho de Melo 	if (!hlist_empty(&tb->owners)) {
1683f421baaSArnaldo Carvalho de Melo 		if (tb->fastreuse > 0 &&
169a9d8f911SEvgeniy Polyakov 		    sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
170a9d8f911SEvgeniy Polyakov 		    smallest_size == -1) {
1713f421baaSArnaldo Carvalho de Melo 			goto success;
1723f421baaSArnaldo Carvalho de Melo 		} else {
1733f421baaSArnaldo Carvalho de Melo 			ret = 1;
174a9d8f911SEvgeniy Polyakov 			if (inet_csk(sk)->icsk_af_ops->bind_conflict(sk, tb)) {
1755add3009SStephen Hemminger 				if (sk->sk_reuse && sk->sk_state != TCP_LISTEN &&
1765add3009SStephen Hemminger 				    smallest_size != -1 && --attempts >= 0) {
177a9d8f911SEvgeniy Polyakov 					spin_unlock(&head->lock);
178a9d8f911SEvgeniy Polyakov 					goto again;
179a9d8f911SEvgeniy Polyakov 				}
1803f421baaSArnaldo Carvalho de Melo 				goto fail_unlock;
1813f421baaSArnaldo Carvalho de Melo 			}
1823f421baaSArnaldo Carvalho de Melo 		}
183a9d8f911SEvgeniy Polyakov 	}
1843f421baaSArnaldo Carvalho de Melo tb_not_found:
1853f421baaSArnaldo Carvalho de Melo 	ret = 1;
186941b1d22SPavel Emelyanov 	if (!tb && (tb = inet_bind_bucket_create(hashinfo->bind_bucket_cachep,
187941b1d22SPavel Emelyanov 					net, head, snum)) == NULL)
1883f421baaSArnaldo Carvalho de Melo 		goto fail_unlock;
1893f421baaSArnaldo Carvalho de Melo 	if (hlist_empty(&tb->owners)) {
1903f421baaSArnaldo Carvalho de Melo 		if (sk->sk_reuse && sk->sk_state != TCP_LISTEN)
1913f421baaSArnaldo Carvalho de Melo 			tb->fastreuse = 1;
1923f421baaSArnaldo Carvalho de Melo 		else
1933f421baaSArnaldo Carvalho de Melo 			tb->fastreuse = 0;
1943f421baaSArnaldo Carvalho de Melo 	} else if (tb->fastreuse &&
1953f421baaSArnaldo Carvalho de Melo 		   (!sk->sk_reuse || sk->sk_state == TCP_LISTEN))
1963f421baaSArnaldo Carvalho de Melo 		tb->fastreuse = 0;
1973f421baaSArnaldo Carvalho de Melo success:
1983f421baaSArnaldo Carvalho de Melo 	if (!inet_csk(sk)->icsk_bind_hash)
1993f421baaSArnaldo Carvalho de Melo 		inet_bind_hash(sk, tb, snum);
200547b792cSIlpo Järvinen 	WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
2013f421baaSArnaldo Carvalho de Melo 	ret = 0;
2023f421baaSArnaldo Carvalho de Melo 
2033f421baaSArnaldo Carvalho de Melo fail_unlock:
2043f421baaSArnaldo Carvalho de Melo 	spin_unlock(&head->lock);
2053f421baaSArnaldo Carvalho de Melo fail:
2063f421baaSArnaldo Carvalho de Melo 	local_bh_enable();
2073f421baaSArnaldo Carvalho de Melo 	return ret;
2083f421baaSArnaldo Carvalho de Melo }
2093f421baaSArnaldo Carvalho de Melo 
2103f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_get_port);
2113f421baaSArnaldo Carvalho de Melo 
2123f421baaSArnaldo Carvalho de Melo /*
2133f421baaSArnaldo Carvalho de Melo  * Wait for an incoming connection, avoid race conditions. This must be called
2143f421baaSArnaldo Carvalho de Melo  * with the socket locked.
2153f421baaSArnaldo Carvalho de Melo  */
2163f421baaSArnaldo Carvalho de Melo static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
2173f421baaSArnaldo Carvalho de Melo {
2183f421baaSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
2193f421baaSArnaldo Carvalho de Melo 	DEFINE_WAIT(wait);
2203f421baaSArnaldo Carvalho de Melo 	int err;
2213f421baaSArnaldo Carvalho de Melo 
2223f421baaSArnaldo Carvalho de Melo 	/*
2233f421baaSArnaldo Carvalho de Melo 	 * True wake-one mechanism for incoming connections: only
2243f421baaSArnaldo Carvalho de Melo 	 * one process gets woken up, not the 'whole herd'.
2253f421baaSArnaldo Carvalho de Melo 	 * Since we do not 'race & poll' for established sockets
2263f421baaSArnaldo Carvalho de Melo 	 * anymore, the common case will execute the loop only once.
2273f421baaSArnaldo Carvalho de Melo 	 *
2283f421baaSArnaldo Carvalho de Melo 	 * Subtle issue: "add_wait_queue_exclusive()" will be added
2293f421baaSArnaldo Carvalho de Melo 	 * after any current non-exclusive waiters, and we know that
2303f421baaSArnaldo Carvalho de Melo 	 * it will always _stay_ after any new non-exclusive waiters
2313f421baaSArnaldo Carvalho de Melo 	 * because all non-exclusive waiters are added at the
2323f421baaSArnaldo Carvalho de Melo 	 * beginning of the wait-queue. As such, it's ok to "drop"
2333f421baaSArnaldo Carvalho de Melo 	 * our exclusiveness temporarily when we get woken up without
2343f421baaSArnaldo Carvalho de Melo 	 * having to remove and re-insert us on the wait queue.
2353f421baaSArnaldo Carvalho de Melo 	 */
2363f421baaSArnaldo Carvalho de Melo 	for (;;) {
2373f421baaSArnaldo Carvalho de Melo 		prepare_to_wait_exclusive(sk->sk_sleep, &wait,
2383f421baaSArnaldo Carvalho de Melo 					  TASK_INTERRUPTIBLE);
2393f421baaSArnaldo Carvalho de Melo 		release_sock(sk);
2403f421baaSArnaldo Carvalho de Melo 		if (reqsk_queue_empty(&icsk->icsk_accept_queue))
2413f421baaSArnaldo Carvalho de Melo 			timeo = schedule_timeout(timeo);
2423f421baaSArnaldo Carvalho de Melo 		lock_sock(sk);
2433f421baaSArnaldo Carvalho de Melo 		err = 0;
2443f421baaSArnaldo Carvalho de Melo 		if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
2453f421baaSArnaldo Carvalho de Melo 			break;
2463f421baaSArnaldo Carvalho de Melo 		err = -EINVAL;
2473f421baaSArnaldo Carvalho de Melo 		if (sk->sk_state != TCP_LISTEN)
2483f421baaSArnaldo Carvalho de Melo 			break;
2493f421baaSArnaldo Carvalho de Melo 		err = sock_intr_errno(timeo);
2503f421baaSArnaldo Carvalho de Melo 		if (signal_pending(current))
2513f421baaSArnaldo Carvalho de Melo 			break;
2523f421baaSArnaldo Carvalho de Melo 		err = -EAGAIN;
2533f421baaSArnaldo Carvalho de Melo 		if (!timeo)
2543f421baaSArnaldo Carvalho de Melo 			break;
2553f421baaSArnaldo Carvalho de Melo 	}
2563f421baaSArnaldo Carvalho de Melo 	finish_wait(sk->sk_sleep, &wait);
2573f421baaSArnaldo Carvalho de Melo 	return err;
2583f421baaSArnaldo Carvalho de Melo }
2593f421baaSArnaldo Carvalho de Melo 
2603f421baaSArnaldo Carvalho de Melo /*
2613f421baaSArnaldo Carvalho de Melo  * This will accept the next outstanding connection.
2623f421baaSArnaldo Carvalho de Melo  */
2633f421baaSArnaldo Carvalho de Melo struct sock *inet_csk_accept(struct sock *sk, int flags, int *err)
2643f421baaSArnaldo Carvalho de Melo {
2653f421baaSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
2663f421baaSArnaldo Carvalho de Melo 	struct sock *newsk;
2673f421baaSArnaldo Carvalho de Melo 	int error;
2683f421baaSArnaldo Carvalho de Melo 
2693f421baaSArnaldo Carvalho de Melo 	lock_sock(sk);
2703f421baaSArnaldo Carvalho de Melo 
2713f421baaSArnaldo Carvalho de Melo 	/* We need to make sure that this socket is listening,
2723f421baaSArnaldo Carvalho de Melo 	 * and that it has something pending.
2733f421baaSArnaldo Carvalho de Melo 	 */
2743f421baaSArnaldo Carvalho de Melo 	error = -EINVAL;
2753f421baaSArnaldo Carvalho de Melo 	if (sk->sk_state != TCP_LISTEN)
2763f421baaSArnaldo Carvalho de Melo 		goto out_err;
2773f421baaSArnaldo Carvalho de Melo 
2783f421baaSArnaldo Carvalho de Melo 	/* Find already established connection */
2793f421baaSArnaldo Carvalho de Melo 	if (reqsk_queue_empty(&icsk->icsk_accept_queue)) {
2803f421baaSArnaldo Carvalho de Melo 		long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
2813f421baaSArnaldo Carvalho de Melo 
2823f421baaSArnaldo Carvalho de Melo 		/* If this is a non blocking socket don't sleep */
2833f421baaSArnaldo Carvalho de Melo 		error = -EAGAIN;
2843f421baaSArnaldo Carvalho de Melo 		if (!timeo)
2853f421baaSArnaldo Carvalho de Melo 			goto out_err;
2863f421baaSArnaldo Carvalho de Melo 
2873f421baaSArnaldo Carvalho de Melo 		error = inet_csk_wait_for_connect(sk, timeo);
2883f421baaSArnaldo Carvalho de Melo 		if (error)
2893f421baaSArnaldo Carvalho de Melo 			goto out_err;
2903f421baaSArnaldo Carvalho de Melo 	}
2913f421baaSArnaldo Carvalho de Melo 
2923f421baaSArnaldo Carvalho de Melo 	newsk = reqsk_queue_get_child(&icsk->icsk_accept_queue, sk);
293547b792cSIlpo Järvinen 	WARN_ON(newsk->sk_state == TCP_SYN_RECV);
2943f421baaSArnaldo Carvalho de Melo out:
2953f421baaSArnaldo Carvalho de Melo 	release_sock(sk);
2963f421baaSArnaldo Carvalho de Melo 	return newsk;
2973f421baaSArnaldo Carvalho de Melo out_err:
2983f421baaSArnaldo Carvalho de Melo 	newsk = NULL;
2993f421baaSArnaldo Carvalho de Melo 	*err = error;
3003f421baaSArnaldo Carvalho de Melo 	goto out;
3013f421baaSArnaldo Carvalho de Melo }
3023f421baaSArnaldo Carvalho de Melo 
3033f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_accept);
3043f421baaSArnaldo Carvalho de Melo 
3053f421baaSArnaldo Carvalho de Melo /*
3063f421baaSArnaldo Carvalho de Melo  * Using different timers for retransmit, delayed acks and probes
3073f421baaSArnaldo Carvalho de Melo  * We may wish use just one timer maintaining a list of expire jiffies
3083f421baaSArnaldo Carvalho de Melo  * to optimize.
3093f421baaSArnaldo Carvalho de Melo  */
3103f421baaSArnaldo Carvalho de Melo void inet_csk_init_xmit_timers(struct sock *sk,
3113f421baaSArnaldo Carvalho de Melo 			       void (*retransmit_handler)(unsigned long),
3123f421baaSArnaldo Carvalho de Melo 			       void (*delack_handler)(unsigned long),
3133f421baaSArnaldo Carvalho de Melo 			       void (*keepalive_handler)(unsigned long))
3143f421baaSArnaldo Carvalho de Melo {
3153f421baaSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
3163f421baaSArnaldo Carvalho de Melo 
317b24b8a24SPavel Emelyanov 	setup_timer(&icsk->icsk_retransmit_timer, retransmit_handler,
318b24b8a24SPavel Emelyanov 			(unsigned long)sk);
319b24b8a24SPavel Emelyanov 	setup_timer(&icsk->icsk_delack_timer, delack_handler,
320b24b8a24SPavel Emelyanov 			(unsigned long)sk);
321b24b8a24SPavel Emelyanov 	setup_timer(&sk->sk_timer, keepalive_handler, (unsigned long)sk);
3223f421baaSArnaldo Carvalho de Melo 	icsk->icsk_pending = icsk->icsk_ack.pending = 0;
3233f421baaSArnaldo Carvalho de Melo }
3243f421baaSArnaldo Carvalho de Melo 
3253f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_init_xmit_timers);
3263f421baaSArnaldo Carvalho de Melo 
3273f421baaSArnaldo Carvalho de Melo void inet_csk_clear_xmit_timers(struct sock *sk)
3283f421baaSArnaldo Carvalho de Melo {
3293f421baaSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
3303f421baaSArnaldo Carvalho de Melo 
3313f421baaSArnaldo Carvalho de Melo 	icsk->icsk_pending = icsk->icsk_ack.pending = icsk->icsk_ack.blocked = 0;
3323f421baaSArnaldo Carvalho de Melo 
3333f421baaSArnaldo Carvalho de Melo 	sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
3343f421baaSArnaldo Carvalho de Melo 	sk_stop_timer(sk, &icsk->icsk_delack_timer);
3353f421baaSArnaldo Carvalho de Melo 	sk_stop_timer(sk, &sk->sk_timer);
3363f421baaSArnaldo Carvalho de Melo }
3373f421baaSArnaldo Carvalho de Melo 
3383f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
3393f421baaSArnaldo Carvalho de Melo 
3403f421baaSArnaldo Carvalho de Melo void inet_csk_delete_keepalive_timer(struct sock *sk)
3413f421baaSArnaldo Carvalho de Melo {
3423f421baaSArnaldo Carvalho de Melo 	sk_stop_timer(sk, &sk->sk_timer);
3433f421baaSArnaldo Carvalho de Melo }
3443f421baaSArnaldo Carvalho de Melo 
3453f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
3463f421baaSArnaldo Carvalho de Melo 
3473f421baaSArnaldo Carvalho de Melo void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
3483f421baaSArnaldo Carvalho de Melo {
3493f421baaSArnaldo Carvalho de Melo 	sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
3503f421baaSArnaldo Carvalho de Melo }
3513f421baaSArnaldo Carvalho de Melo 
3523f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
3533f421baaSArnaldo Carvalho de Melo 
3543f421baaSArnaldo Carvalho de Melo struct dst_entry *inet_csk_route_req(struct sock *sk,
3553f421baaSArnaldo Carvalho de Melo 				     const struct request_sock *req)
3563f421baaSArnaldo Carvalho de Melo {
3573f421baaSArnaldo Carvalho de Melo 	struct rtable *rt;
3583f421baaSArnaldo Carvalho de Melo 	const struct inet_request_sock *ireq = inet_rsk(req);
3593f421baaSArnaldo Carvalho de Melo 	struct ip_options *opt = inet_rsk(req)->opt;
3603f421baaSArnaldo Carvalho de Melo 	struct flowi fl = { .oif = sk->sk_bound_dev_if,
361ffce9082SAtis Elsts 			    .mark = sk->sk_mark,
3623f421baaSArnaldo Carvalho de Melo 			    .nl_u = { .ip4_u =
3633f421baaSArnaldo Carvalho de Melo 				      { .daddr = ((opt && opt->srr) ?
3643f421baaSArnaldo Carvalho de Melo 						  opt->faddr :
3653f421baaSArnaldo Carvalho de Melo 						  ireq->rmt_addr),
3663f421baaSArnaldo Carvalho de Melo 					.saddr = ireq->loc_addr,
3673f421baaSArnaldo Carvalho de Melo 					.tos = RT_CONN_FLAGS(sk) } },
3683f421baaSArnaldo Carvalho de Melo 			    .proto = sk->sk_protocol,
36986b08d86SKOVACS Krisztian 			    .flags = inet_sk_flowi_flags(sk),
3703f421baaSArnaldo Carvalho de Melo 			    .uli_u = { .ports =
371c720c7e8SEric Dumazet 				       { .sport = inet_sk(sk)->inet_sport,
3723f421baaSArnaldo Carvalho de Melo 					 .dport = ireq->rmt_port } } };
37384a3aa00SPavel Emelyanov 	struct net *net = sock_net(sk);
3743f421baaSArnaldo Carvalho de Melo 
3754237c75cSVenkat Yekkirala 	security_req_classify_flow(req, &fl);
376857a6e0aSIlpo Järvinen 	if (ip_route_output_flow(net, &rt, &fl, sk, 0))
377857a6e0aSIlpo Järvinen 		goto no_route;
378857a6e0aSIlpo Järvinen 	if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
379857a6e0aSIlpo Järvinen 		goto route_err;
3803f421baaSArnaldo Carvalho de Melo 	return &rt->u.dst;
381857a6e0aSIlpo Järvinen 
382857a6e0aSIlpo Järvinen route_err:
383857a6e0aSIlpo Järvinen 	ip_rt_put(rt);
384857a6e0aSIlpo Järvinen no_route:
385857a6e0aSIlpo Järvinen 	IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
386857a6e0aSIlpo Järvinen 	return NULL;
3873f421baaSArnaldo Carvalho de Melo }
3883f421baaSArnaldo Carvalho de Melo 
3893f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_route_req);
3903f421baaSArnaldo Carvalho de Melo 
3916b72977bSAl Viro static inline u32 inet_synq_hash(const __be32 raddr, const __be16 rport,
39272a3effaSEric Dumazet 				 const u32 rnd, const u32 synq_hsize)
3933f421baaSArnaldo Carvalho de Melo {
3946b72977bSAl Viro 	return jhash_2words((__force u32)raddr, (__force u32)rport, rnd) & (synq_hsize - 1);
3953f421baaSArnaldo Carvalho de Melo }
3963f421baaSArnaldo Carvalho de Melo 
3973f421baaSArnaldo Carvalho de Melo #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
3983f421baaSArnaldo Carvalho de Melo #define AF_INET_FAMILY(fam) ((fam) == AF_INET)
3993f421baaSArnaldo Carvalho de Melo #else
4003f421baaSArnaldo Carvalho de Melo #define AF_INET_FAMILY(fam) 1
4013f421baaSArnaldo Carvalho de Melo #endif
4023f421baaSArnaldo Carvalho de Melo 
4033f421baaSArnaldo Carvalho de Melo struct request_sock *inet_csk_search_req(const struct sock *sk,
4043f421baaSArnaldo Carvalho de Melo 					 struct request_sock ***prevp,
4056b72977bSAl Viro 					 const __be16 rport, const __be32 raddr,
4067f25afbbSAl Viro 					 const __be32 laddr)
4073f421baaSArnaldo Carvalho de Melo {
4083f421baaSArnaldo Carvalho de Melo 	const struct inet_connection_sock *icsk = inet_csk(sk);
4093f421baaSArnaldo Carvalho de Melo 	struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
4103f421baaSArnaldo Carvalho de Melo 	struct request_sock *req, **prev;
4113f421baaSArnaldo Carvalho de Melo 
4123f421baaSArnaldo Carvalho de Melo 	for (prev = &lopt->syn_table[inet_synq_hash(raddr, rport, lopt->hash_rnd,
4133f421baaSArnaldo Carvalho de Melo 						    lopt->nr_table_entries)];
4143f421baaSArnaldo Carvalho de Melo 	     (req = *prev) != NULL;
4153f421baaSArnaldo Carvalho de Melo 	     prev = &req->dl_next) {
4163f421baaSArnaldo Carvalho de Melo 		const struct inet_request_sock *ireq = inet_rsk(req);
4173f421baaSArnaldo Carvalho de Melo 
4183f421baaSArnaldo Carvalho de Melo 		if (ireq->rmt_port == rport &&
4193f421baaSArnaldo Carvalho de Melo 		    ireq->rmt_addr == raddr &&
4203f421baaSArnaldo Carvalho de Melo 		    ireq->loc_addr == laddr &&
4213f421baaSArnaldo Carvalho de Melo 		    AF_INET_FAMILY(req->rsk_ops->family)) {
422547b792cSIlpo Järvinen 			WARN_ON(req->sk);
4233f421baaSArnaldo Carvalho de Melo 			*prevp = prev;
4243f421baaSArnaldo Carvalho de Melo 			break;
4253f421baaSArnaldo Carvalho de Melo 		}
4263f421baaSArnaldo Carvalho de Melo 	}
4273f421baaSArnaldo Carvalho de Melo 
4283f421baaSArnaldo Carvalho de Melo 	return req;
4293f421baaSArnaldo Carvalho de Melo }
4303f421baaSArnaldo Carvalho de Melo 
4313f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_search_req);
4323f421baaSArnaldo Carvalho de Melo 
4333f421baaSArnaldo Carvalho de Melo void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
434c2977c22SArnaldo Carvalho de Melo 				   unsigned long timeout)
4353f421baaSArnaldo Carvalho de Melo {
4363f421baaSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
4373f421baaSArnaldo Carvalho de Melo 	struct listen_sock *lopt = icsk->icsk_accept_queue.listen_opt;
4383f421baaSArnaldo Carvalho de Melo 	const u32 h = inet_synq_hash(inet_rsk(req)->rmt_addr, inet_rsk(req)->rmt_port,
4393f421baaSArnaldo Carvalho de Melo 				     lopt->hash_rnd, lopt->nr_table_entries);
4403f421baaSArnaldo Carvalho de Melo 
4413f421baaSArnaldo Carvalho de Melo 	reqsk_queue_hash_req(&icsk->icsk_accept_queue, h, req, timeout);
4423f421baaSArnaldo Carvalho de Melo 	inet_csk_reqsk_queue_added(sk, timeout);
4433f421baaSArnaldo Carvalho de Melo }
4443f421baaSArnaldo Carvalho de Melo 
445a019d6feSArnaldo Carvalho de Melo /* Only thing we need from tcp.h */
446a019d6feSArnaldo Carvalho de Melo extern int sysctl_tcp_synack_retries;
447a019d6feSArnaldo Carvalho de Melo 
4483f421baaSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
4499f1d2604SArnaldo Carvalho de Melo 
4500c3d79bcSJulian Anastasov /* Decide when to expire the request and when to resend SYN-ACK */
4510c3d79bcSJulian Anastasov static inline void syn_ack_recalc(struct request_sock *req, const int thresh,
4520c3d79bcSJulian Anastasov 				  const int max_retries,
4530c3d79bcSJulian Anastasov 				  const u8 rskq_defer_accept,
4540c3d79bcSJulian Anastasov 				  int *expire, int *resend)
4550c3d79bcSJulian Anastasov {
4560c3d79bcSJulian Anastasov 	if (!rskq_defer_accept) {
4570c3d79bcSJulian Anastasov 		*expire = req->retrans >= thresh;
4580c3d79bcSJulian Anastasov 		*resend = 1;
4590c3d79bcSJulian Anastasov 		return;
4600c3d79bcSJulian Anastasov 	}
4610c3d79bcSJulian Anastasov 	*expire = req->retrans >= thresh &&
4620c3d79bcSJulian Anastasov 		  (!inet_rsk(req)->acked || req->retrans >= max_retries);
4630c3d79bcSJulian Anastasov 	/*
4640c3d79bcSJulian Anastasov 	 * Do not resend while waiting for data after ACK,
4650c3d79bcSJulian Anastasov 	 * start to resend on end of deferring period to give
4660c3d79bcSJulian Anastasov 	 * last chance for data or ACK to create established socket.
4670c3d79bcSJulian Anastasov 	 */
4680c3d79bcSJulian Anastasov 	*resend = !inet_rsk(req)->acked ||
4690c3d79bcSJulian Anastasov 		  req->retrans >= rskq_defer_accept - 1;
4700c3d79bcSJulian Anastasov }
4710c3d79bcSJulian Anastasov 
472a019d6feSArnaldo Carvalho de Melo void inet_csk_reqsk_queue_prune(struct sock *parent,
473a019d6feSArnaldo Carvalho de Melo 				const unsigned long interval,
474a019d6feSArnaldo Carvalho de Melo 				const unsigned long timeout,
475a019d6feSArnaldo Carvalho de Melo 				const unsigned long max_rto)
476a019d6feSArnaldo Carvalho de Melo {
477a019d6feSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(parent);
478a019d6feSArnaldo Carvalho de Melo 	struct request_sock_queue *queue = &icsk->icsk_accept_queue;
479a019d6feSArnaldo Carvalho de Melo 	struct listen_sock *lopt = queue->listen_opt;
480ec0a1966SDavid S. Miller 	int max_retries = icsk->icsk_syn_retries ? : sysctl_tcp_synack_retries;
481ec0a1966SDavid S. Miller 	int thresh = max_retries;
482a019d6feSArnaldo Carvalho de Melo 	unsigned long now = jiffies;
483a019d6feSArnaldo Carvalho de Melo 	struct request_sock **reqp, *req;
484a019d6feSArnaldo Carvalho de Melo 	int i, budget;
485a019d6feSArnaldo Carvalho de Melo 
486a019d6feSArnaldo Carvalho de Melo 	if (lopt == NULL || lopt->qlen == 0)
487a019d6feSArnaldo Carvalho de Melo 		return;
488a019d6feSArnaldo Carvalho de Melo 
489a019d6feSArnaldo Carvalho de Melo 	/* Normally all the openreqs are young and become mature
490a019d6feSArnaldo Carvalho de Melo 	 * (i.e. converted to established socket) for first timeout.
491a019d6feSArnaldo Carvalho de Melo 	 * If synack was not acknowledged for 3 seconds, it means
492a019d6feSArnaldo Carvalho de Melo 	 * one of the following things: synack was lost, ack was lost,
493a019d6feSArnaldo Carvalho de Melo 	 * rtt is high or nobody planned to ack (i.e. synflood).
494a019d6feSArnaldo Carvalho de Melo 	 * When server is a bit loaded, queue is populated with old
495a019d6feSArnaldo Carvalho de Melo 	 * open requests, reducing effective size of queue.
496a019d6feSArnaldo Carvalho de Melo 	 * When server is well loaded, queue size reduces to zero
497a019d6feSArnaldo Carvalho de Melo 	 * after several minutes of work. It is not synflood,
498a019d6feSArnaldo Carvalho de Melo 	 * it is normal operation. The solution is pruning
499a019d6feSArnaldo Carvalho de Melo 	 * too old entries overriding normal timeout, when
500a019d6feSArnaldo Carvalho de Melo 	 * situation becomes dangerous.
501a019d6feSArnaldo Carvalho de Melo 	 *
502a019d6feSArnaldo Carvalho de Melo 	 * Essentially, we reserve half of room for young
503a019d6feSArnaldo Carvalho de Melo 	 * embrions; and abort old ones without pity, if old
504a019d6feSArnaldo Carvalho de Melo 	 * ones are about to clog our table.
505a019d6feSArnaldo Carvalho de Melo 	 */
506a019d6feSArnaldo Carvalho de Melo 	if (lopt->qlen>>(lopt->max_qlen_log-1)) {
507a019d6feSArnaldo Carvalho de Melo 		int young = (lopt->qlen_young<<1);
508a019d6feSArnaldo Carvalho de Melo 
509a019d6feSArnaldo Carvalho de Melo 		while (thresh > 2) {
510a019d6feSArnaldo Carvalho de Melo 			if (lopt->qlen < young)
511a019d6feSArnaldo Carvalho de Melo 				break;
512a019d6feSArnaldo Carvalho de Melo 			thresh--;
513a019d6feSArnaldo Carvalho de Melo 			young <<= 1;
514a019d6feSArnaldo Carvalho de Melo 		}
515a019d6feSArnaldo Carvalho de Melo 	}
516a019d6feSArnaldo Carvalho de Melo 
517ec0a1966SDavid S. Miller 	if (queue->rskq_defer_accept)
518ec0a1966SDavid S. Miller 		max_retries = queue->rskq_defer_accept;
519ec0a1966SDavid S. Miller 
520a019d6feSArnaldo Carvalho de Melo 	budget = 2 * (lopt->nr_table_entries / (timeout / interval));
521a019d6feSArnaldo Carvalho de Melo 	i = lopt->clock_hand;
522a019d6feSArnaldo Carvalho de Melo 
523a019d6feSArnaldo Carvalho de Melo 	do {
524a019d6feSArnaldo Carvalho de Melo 		reqp=&lopt->syn_table[i];
525a019d6feSArnaldo Carvalho de Melo 		while ((req = *reqp) != NULL) {
526a019d6feSArnaldo Carvalho de Melo 			if (time_after_eq(now, req->expires)) {
5270c3d79bcSJulian Anastasov 				int expire = 0, resend = 0;
5280c3d79bcSJulian Anastasov 
5290c3d79bcSJulian Anastasov 				syn_ack_recalc(req, thresh, max_retries,
5300c3d79bcSJulian Anastasov 					       queue->rskq_defer_accept,
5310c3d79bcSJulian Anastasov 					       &expire, &resend);
5320c3d79bcSJulian Anastasov 				if (!expire &&
5330c3d79bcSJulian Anastasov 				    (!resend ||
5340c3d79bcSJulian Anastasov 				     !req->rsk_ops->rtx_syn_ack(parent, req) ||
5350c3d79bcSJulian Anastasov 				     inet_rsk(req)->acked)) {
536a019d6feSArnaldo Carvalho de Melo 					unsigned long timeo;
537a019d6feSArnaldo Carvalho de Melo 
538a019d6feSArnaldo Carvalho de Melo 					if (req->retrans++ == 0)
539a019d6feSArnaldo Carvalho de Melo 						lopt->qlen_young--;
540a019d6feSArnaldo Carvalho de Melo 					timeo = min((timeout << req->retrans), max_rto);
541a019d6feSArnaldo Carvalho de Melo 					req->expires = now + timeo;
542a019d6feSArnaldo Carvalho de Melo 					reqp = &req->dl_next;
543a019d6feSArnaldo Carvalho de Melo 					continue;
544a019d6feSArnaldo Carvalho de Melo 				}
545a019d6feSArnaldo Carvalho de Melo 
546a019d6feSArnaldo Carvalho de Melo 				/* Drop this request */
547a019d6feSArnaldo Carvalho de Melo 				inet_csk_reqsk_queue_unlink(parent, req, reqp);
548a019d6feSArnaldo Carvalho de Melo 				reqsk_queue_removed(queue, req);
549a019d6feSArnaldo Carvalho de Melo 				reqsk_free(req);
550a019d6feSArnaldo Carvalho de Melo 				continue;
551a019d6feSArnaldo Carvalho de Melo 			}
552a019d6feSArnaldo Carvalho de Melo 			reqp = &req->dl_next;
553a019d6feSArnaldo Carvalho de Melo 		}
554a019d6feSArnaldo Carvalho de Melo 
555a019d6feSArnaldo Carvalho de Melo 		i = (i + 1) & (lopt->nr_table_entries - 1);
556a019d6feSArnaldo Carvalho de Melo 
557a019d6feSArnaldo Carvalho de Melo 	} while (--budget > 0);
558a019d6feSArnaldo Carvalho de Melo 
559a019d6feSArnaldo Carvalho de Melo 	lopt->clock_hand = i;
560a019d6feSArnaldo Carvalho de Melo 
561a019d6feSArnaldo Carvalho de Melo 	if (lopt->qlen)
562a019d6feSArnaldo Carvalho de Melo 		inet_csk_reset_keepalive_timer(parent, interval);
563a019d6feSArnaldo Carvalho de Melo }
564a019d6feSArnaldo Carvalho de Melo 
565a019d6feSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_prune);
566a019d6feSArnaldo Carvalho de Melo 
5679f1d2604SArnaldo Carvalho de Melo struct sock *inet_csk_clone(struct sock *sk, const struct request_sock *req,
568dd0fc66fSAl Viro 			    const gfp_t priority)
5699f1d2604SArnaldo Carvalho de Melo {
5709f1d2604SArnaldo Carvalho de Melo 	struct sock *newsk = sk_clone(sk, priority);
5719f1d2604SArnaldo Carvalho de Melo 
5729f1d2604SArnaldo Carvalho de Melo 	if (newsk != NULL) {
5739f1d2604SArnaldo Carvalho de Melo 		struct inet_connection_sock *newicsk = inet_csk(newsk);
5749f1d2604SArnaldo Carvalho de Melo 
5759f1d2604SArnaldo Carvalho de Melo 		newsk->sk_state = TCP_SYN_RECV;
5769f1d2604SArnaldo Carvalho de Melo 		newicsk->icsk_bind_hash = NULL;
5779f1d2604SArnaldo Carvalho de Melo 
578c720c7e8SEric Dumazet 		inet_sk(newsk)->inet_dport = inet_rsk(req)->rmt_port;
579c720c7e8SEric Dumazet 		inet_sk(newsk)->inet_num = ntohs(inet_rsk(req)->loc_port);
580c720c7e8SEric Dumazet 		inet_sk(newsk)->inet_sport = inet_rsk(req)->loc_port;
5819f1d2604SArnaldo Carvalho de Melo 		newsk->sk_write_space = sk_stream_write_space;
5829f1d2604SArnaldo Carvalho de Melo 
5839f1d2604SArnaldo Carvalho de Melo 		newicsk->icsk_retransmits = 0;
5849f1d2604SArnaldo Carvalho de Melo 		newicsk->icsk_backoff	  = 0;
5856687e988SArnaldo Carvalho de Melo 		newicsk->icsk_probes_out  = 0;
5869f1d2604SArnaldo Carvalho de Melo 
5879f1d2604SArnaldo Carvalho de Melo 		/* Deinitialize accept_queue to trap illegal accesses. */
5889f1d2604SArnaldo Carvalho de Melo 		memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
5894237c75cSVenkat Yekkirala 
5904237c75cSVenkat Yekkirala 		security_inet_csk_clone(newsk, req);
5919f1d2604SArnaldo Carvalho de Melo 	}
5929f1d2604SArnaldo Carvalho de Melo 	return newsk;
5939f1d2604SArnaldo Carvalho de Melo }
5949f1d2604SArnaldo Carvalho de Melo 
5959f1d2604SArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_clone);
596a019d6feSArnaldo Carvalho de Melo 
597a019d6feSArnaldo Carvalho de Melo /*
598a019d6feSArnaldo Carvalho de Melo  * At this point, there should be no process reference to this
599a019d6feSArnaldo Carvalho de Melo  * socket, and thus no user references at all.  Therefore we
600a019d6feSArnaldo Carvalho de Melo  * can assume the socket waitqueue is inactive and nobody will
601a019d6feSArnaldo Carvalho de Melo  * try to jump onto it.
602a019d6feSArnaldo Carvalho de Melo  */
603a019d6feSArnaldo Carvalho de Melo void inet_csk_destroy_sock(struct sock *sk)
604a019d6feSArnaldo Carvalho de Melo {
605547b792cSIlpo Järvinen 	WARN_ON(sk->sk_state != TCP_CLOSE);
606547b792cSIlpo Järvinen 	WARN_ON(!sock_flag(sk, SOCK_DEAD));
607a019d6feSArnaldo Carvalho de Melo 
608a019d6feSArnaldo Carvalho de Melo 	/* It cannot be in hash table! */
609547b792cSIlpo Järvinen 	WARN_ON(!sk_unhashed(sk));
610a019d6feSArnaldo Carvalho de Melo 
611c720c7e8SEric Dumazet 	/* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
612c720c7e8SEric Dumazet 	WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
613a019d6feSArnaldo Carvalho de Melo 
614a019d6feSArnaldo Carvalho de Melo 	sk->sk_prot->destroy(sk);
615a019d6feSArnaldo Carvalho de Melo 
616a019d6feSArnaldo Carvalho de Melo 	sk_stream_kill_queues(sk);
617a019d6feSArnaldo Carvalho de Melo 
618a019d6feSArnaldo Carvalho de Melo 	xfrm_sk_free_policy(sk);
619a019d6feSArnaldo Carvalho de Melo 
620a019d6feSArnaldo Carvalho de Melo 	sk_refcnt_debug_release(sk);
621a019d6feSArnaldo Carvalho de Melo 
622dd24c001SEric Dumazet 	percpu_counter_dec(sk->sk_prot->orphan_count);
623a019d6feSArnaldo Carvalho de Melo 	sock_put(sk);
624a019d6feSArnaldo Carvalho de Melo }
625a019d6feSArnaldo Carvalho de Melo 
626a019d6feSArnaldo Carvalho de Melo EXPORT_SYMBOL(inet_csk_destroy_sock);
627a019d6feSArnaldo Carvalho de Melo 
628a019d6feSArnaldo Carvalho de Melo int inet_csk_listen_start(struct sock *sk, const int nr_table_entries)
629a019d6feSArnaldo Carvalho de Melo {
630a019d6feSArnaldo Carvalho de Melo 	struct inet_sock *inet = inet_sk(sk);
631a019d6feSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
632a019d6feSArnaldo Carvalho de Melo 	int rc = reqsk_queue_alloc(&icsk->icsk_accept_queue, nr_table_entries);
633a019d6feSArnaldo Carvalho de Melo 
634a019d6feSArnaldo Carvalho de Melo 	if (rc != 0)
635a019d6feSArnaldo Carvalho de Melo 		return rc;
636a019d6feSArnaldo Carvalho de Melo 
637a019d6feSArnaldo Carvalho de Melo 	sk->sk_max_ack_backlog = 0;
638a019d6feSArnaldo Carvalho de Melo 	sk->sk_ack_backlog = 0;
639a019d6feSArnaldo Carvalho de Melo 	inet_csk_delack_init(sk);
640a019d6feSArnaldo Carvalho de Melo 
641a019d6feSArnaldo Carvalho de Melo 	/* There is race window here: we announce ourselves listening,
642a019d6feSArnaldo Carvalho de Melo 	 * but this transition is still not validated by get_port().
643a019d6feSArnaldo Carvalho de Melo 	 * It is OK, because this socket enters to hash table only
644a019d6feSArnaldo Carvalho de Melo 	 * after validation is complete.
645a019d6feSArnaldo Carvalho de Melo 	 */
646a019d6feSArnaldo Carvalho de Melo 	sk->sk_state = TCP_LISTEN;
647c720c7e8SEric Dumazet 	if (!sk->sk_prot->get_port(sk, inet->inet_num)) {
648c720c7e8SEric Dumazet 		inet->inet_sport = htons(inet->inet_num);
649a019d6feSArnaldo Carvalho de Melo 
650a019d6feSArnaldo Carvalho de Melo 		sk_dst_reset(sk);
651a019d6feSArnaldo Carvalho de Melo 		sk->sk_prot->hash(sk);
652a019d6feSArnaldo Carvalho de Melo 
653a019d6feSArnaldo Carvalho de Melo 		return 0;
654a019d6feSArnaldo Carvalho de Melo 	}
655a019d6feSArnaldo Carvalho de Melo 
656a019d6feSArnaldo Carvalho de Melo 	sk->sk_state = TCP_CLOSE;
657a019d6feSArnaldo Carvalho de Melo 	__reqsk_queue_destroy(&icsk->icsk_accept_queue);
658a019d6feSArnaldo Carvalho de Melo 	return -EADDRINUSE;
659a019d6feSArnaldo Carvalho de Melo }
660a019d6feSArnaldo Carvalho de Melo 
661a019d6feSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_listen_start);
662a019d6feSArnaldo Carvalho de Melo 
663a019d6feSArnaldo Carvalho de Melo /*
664a019d6feSArnaldo Carvalho de Melo  *	This routine closes sockets which have been at least partially
665a019d6feSArnaldo Carvalho de Melo  *	opened, but not yet accepted.
666a019d6feSArnaldo Carvalho de Melo  */
667a019d6feSArnaldo Carvalho de Melo void inet_csk_listen_stop(struct sock *sk)
668a019d6feSArnaldo Carvalho de Melo {
669a019d6feSArnaldo Carvalho de Melo 	struct inet_connection_sock *icsk = inet_csk(sk);
670a019d6feSArnaldo Carvalho de Melo 	struct request_sock *acc_req;
671a019d6feSArnaldo Carvalho de Melo 	struct request_sock *req;
672a019d6feSArnaldo Carvalho de Melo 
673a019d6feSArnaldo Carvalho de Melo 	inet_csk_delete_keepalive_timer(sk);
674a019d6feSArnaldo Carvalho de Melo 
675a019d6feSArnaldo Carvalho de Melo 	/* make all the listen_opt local to us */
676a019d6feSArnaldo Carvalho de Melo 	acc_req = reqsk_queue_yank_acceptq(&icsk->icsk_accept_queue);
677a019d6feSArnaldo Carvalho de Melo 
678a019d6feSArnaldo Carvalho de Melo 	/* Following specs, it would be better either to send FIN
679a019d6feSArnaldo Carvalho de Melo 	 * (and enter FIN-WAIT-1, it is normal close)
680a019d6feSArnaldo Carvalho de Melo 	 * or to send active reset (abort).
681a019d6feSArnaldo Carvalho de Melo 	 * Certainly, it is pretty dangerous while synflood, but it is
682a019d6feSArnaldo Carvalho de Melo 	 * bad justification for our negligence 8)
683a019d6feSArnaldo Carvalho de Melo 	 * To be honest, we are not able to make either
684a019d6feSArnaldo Carvalho de Melo 	 * of the variants now.			--ANK
685a019d6feSArnaldo Carvalho de Melo 	 */
686a019d6feSArnaldo Carvalho de Melo 	reqsk_queue_destroy(&icsk->icsk_accept_queue);
687a019d6feSArnaldo Carvalho de Melo 
688a019d6feSArnaldo Carvalho de Melo 	while ((req = acc_req) != NULL) {
689a019d6feSArnaldo Carvalho de Melo 		struct sock *child = req->sk;
690a019d6feSArnaldo Carvalho de Melo 
691a019d6feSArnaldo Carvalho de Melo 		acc_req = req->dl_next;
692a019d6feSArnaldo Carvalho de Melo 
693a019d6feSArnaldo Carvalho de Melo 		local_bh_disable();
694a019d6feSArnaldo Carvalho de Melo 		bh_lock_sock(child);
695547b792cSIlpo Järvinen 		WARN_ON(sock_owned_by_user(child));
696a019d6feSArnaldo Carvalho de Melo 		sock_hold(child);
697a019d6feSArnaldo Carvalho de Melo 
698a019d6feSArnaldo Carvalho de Melo 		sk->sk_prot->disconnect(child, O_NONBLOCK);
699a019d6feSArnaldo Carvalho de Melo 
700a019d6feSArnaldo Carvalho de Melo 		sock_orphan(child);
701a019d6feSArnaldo Carvalho de Melo 
702eb4dea58SHerbert Xu 		percpu_counter_inc(sk->sk_prot->orphan_count);
703eb4dea58SHerbert Xu 
704a019d6feSArnaldo Carvalho de Melo 		inet_csk_destroy_sock(child);
705a019d6feSArnaldo Carvalho de Melo 
706a019d6feSArnaldo Carvalho de Melo 		bh_unlock_sock(child);
707a019d6feSArnaldo Carvalho de Melo 		local_bh_enable();
708a019d6feSArnaldo Carvalho de Melo 		sock_put(child);
709a019d6feSArnaldo Carvalho de Melo 
710a019d6feSArnaldo Carvalho de Melo 		sk_acceptq_removed(sk);
711a019d6feSArnaldo Carvalho de Melo 		__reqsk_free(req);
712a019d6feSArnaldo Carvalho de Melo 	}
713547b792cSIlpo Järvinen 	WARN_ON(sk->sk_ack_backlog);
714a019d6feSArnaldo Carvalho de Melo }
715a019d6feSArnaldo Carvalho de Melo 
716a019d6feSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
717af05dc93SArnaldo Carvalho de Melo 
718af05dc93SArnaldo Carvalho de Melo void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
719af05dc93SArnaldo Carvalho de Melo {
720af05dc93SArnaldo Carvalho de Melo 	struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
721af05dc93SArnaldo Carvalho de Melo 	const struct inet_sock *inet = inet_sk(sk);
722af05dc93SArnaldo Carvalho de Melo 
723af05dc93SArnaldo Carvalho de Melo 	sin->sin_family		= AF_INET;
724c720c7e8SEric Dumazet 	sin->sin_addr.s_addr	= inet->inet_daddr;
725c720c7e8SEric Dumazet 	sin->sin_port		= inet->inet_dport;
726af05dc93SArnaldo Carvalho de Melo }
727af05dc93SArnaldo Carvalho de Melo 
728af05dc93SArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
729c4d93909SArnaldo Carvalho de Melo 
730dec73ff0SArnaldo Carvalho de Melo #ifdef CONFIG_COMPAT
731dec73ff0SArnaldo Carvalho de Melo int inet_csk_compat_getsockopt(struct sock *sk, int level, int optname,
732dec73ff0SArnaldo Carvalho de Melo 			       char __user *optval, int __user *optlen)
733dec73ff0SArnaldo Carvalho de Melo {
734dbeff12bSDavid S. Miller 	const struct inet_connection_sock *icsk = inet_csk(sk);
735dec73ff0SArnaldo Carvalho de Melo 
736dec73ff0SArnaldo Carvalho de Melo 	if (icsk->icsk_af_ops->compat_getsockopt != NULL)
737dec73ff0SArnaldo Carvalho de Melo 		return icsk->icsk_af_ops->compat_getsockopt(sk, level, optname,
738dec73ff0SArnaldo Carvalho de Melo 							    optval, optlen);
739dec73ff0SArnaldo Carvalho de Melo 	return icsk->icsk_af_ops->getsockopt(sk, level, optname,
740dec73ff0SArnaldo Carvalho de Melo 					     optval, optlen);
741dec73ff0SArnaldo Carvalho de Melo }
742dec73ff0SArnaldo Carvalho de Melo 
743dec73ff0SArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_compat_getsockopt);
744dec73ff0SArnaldo Carvalho de Melo 
745dec73ff0SArnaldo Carvalho de Melo int inet_csk_compat_setsockopt(struct sock *sk, int level, int optname,
746b7058842SDavid S. Miller 			       char __user *optval, unsigned int optlen)
747dec73ff0SArnaldo Carvalho de Melo {
748dbeff12bSDavid S. Miller 	const struct inet_connection_sock *icsk = inet_csk(sk);
749dec73ff0SArnaldo Carvalho de Melo 
750dec73ff0SArnaldo Carvalho de Melo 	if (icsk->icsk_af_ops->compat_setsockopt != NULL)
751dec73ff0SArnaldo Carvalho de Melo 		return icsk->icsk_af_ops->compat_setsockopt(sk, level, optname,
752dec73ff0SArnaldo Carvalho de Melo 							    optval, optlen);
753dec73ff0SArnaldo Carvalho de Melo 	return icsk->icsk_af_ops->setsockopt(sk, level, optname,
754dec73ff0SArnaldo Carvalho de Melo 					     optval, optlen);
755dec73ff0SArnaldo Carvalho de Melo }
756dec73ff0SArnaldo Carvalho de Melo 
757dec73ff0SArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(inet_csk_compat_setsockopt);
758dec73ff0SArnaldo Carvalho de Melo #endif
759