Allow to bind to an already in use local port

during connect when the connection will still have a unique
identity.  Fixes port space exhaustion, especially in web
caches.

Initial work done by Andi Kleen.

Documentation/networking/ip-sysctl.txt

@@ -126,7 +126,13 @@ tcp_max_tw_buckets - INTEGER
 	if network conditions require more than default value.
 
 tcp_tw_recycle - BOOLEAN
-	Enable fast recycling TIME-WAIT sockets. Default value is 1.
+	Enable fast recycling TIME-WAIT sockets. Default value is 0.
+	It should not be changed without advice/request of technical
+	experts.
+
+tcp_tw_reuse - BOOLEAN
+	Allow to reuse TIME-WAIT sockets for new connections when it is
+	safe from protocol viewpoint. Default value is 0.
 	It should not be changed without advice/request of technical
 	experts.
 

include/linux/sysctl.h

@@ -288,7 +288,8 @@ enum
 	NET_TCP_ADV_WIN_SCALE=87,
 	NET_IPV4_NONLOCAL_BIND=88,
 	NET_IPV4_ICMP_RATELIMIT=89,
-	NET_IPV4_ICMP_RATEMASK=90
+	NET_IPV4_ICMP_RATEMASK=90,
+	NET_TCP_TW_REUSE=91
 };
 
 enum {

include/net/tcp.h

@@ -78,7 +78,7 @@ struct tcp_ehash_bucket {
  */
 struct tcp_bind_bucket {
 	unsigned short		port;
-	unsigned short		fastreuse;
+	signed short		fastreuse;
 	struct tcp_bind_bucket	*next;
 	struct sock		*owners;
 	struct tcp_bind_bucket	**pprev;
@@ -469,6 +469,7 @@ extern int sysctl_tcp_wmem[3];
 extern int sysctl_tcp_rmem[3];
 extern int sysctl_tcp_app_win;
 extern int sysctl_tcp_adv_win_scale;
+extern int sysctl_tcp_tw_reuse;
 
 extern atomic_t tcp_memory_allocated;
 extern atomic_t tcp_sockets_allocated;
@@ -577,9 +578,7 @@ struct tcp_func {
 							 struct sk_buff *skb,
 							 struct open_request *req,
 							 struct dst_entry *dst);
-	
-	int			(*hash_connecting)	(struct sock *sk);
-
+    
 	int			(*remember_stamp)	(struct sock *sk);
 
 	__u16			net_header_len;
@@ -781,8 +780,7 @@ extern int			tcp_v4_connect(struct sock *sk,
 					       struct sockaddr *uaddr,
 					       int addr_len);
 
-extern int			tcp_connect(struct sock *sk,
-					    struct sk_buff *skb);
+extern int			tcp_connect(struct sock *sk);
 
 extern struct sk_buff *		tcp_make_synack(struct sock *sk,
 						struct dst_entry *dst,

net/ipv4/af_inet.c

@@ -655,13 +655,6 @@ int inet_stream_connect(struct socket *sock, struct sockaddr * uaddr,
 		if (sk->state != TCP_CLOSE) 
 			goto out;
 
-		err = -EAGAIN;
-		if (!inet->num) {
-			if (sk->prot->get_port(sk, 0) != 0)
-				goto out;
-			inet->sport = htons(inet->num);
-		}
-
 		err = sk->prot->connect(sk, uaddr, addr_len);
 		if (err < 0)
 			goto out;

net/ipv4/sysctl_net_ipv4.c

@@ -219,6 +219,8 @@ ctl_table ipv4_table[] = {
 	 &sysctl_icmp_ratelimit, sizeof(int), 0644, NULL, &proc_dointvec},
 	{NET_IPV4_ICMP_RATEMASK, "icmp_ratemask",
 	 &sysctl_icmp_ratemask, sizeof(int), 0644, NULL, &proc_dointvec},
+	{NET_TCP_TW_REUSE, "tcp_tw_reuse",
+	 &sysctl_tcp_tw_reuse, sizeof(int), 0644, NULL, &proc_dointvec},
 	{0}
 };
 

net/ipv4/tcp_output.c

@@ -38,6 +38,7 @@
 
 #include <net/tcp.h>
 
+#include <linux/compiler.h>
 #include <linux/smp_lock.h>
 
 /* People can turn this off for buggy TCP's found in printers etc. */
@@ -1156,14 +1157,14 @@ struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
 	return skb;
 }
 
-int tcp_connect(struct sock *sk, struct sk_buff *buff)
+/* 
+ * Do all connect socket setups that can be done AF independent.
+ */ 
+static inline void tcp_connect_init(struct sock *sk)
 {
 	struct dst_entry *dst = __sk_dst_get(sk);
 	struct tcp_opt *tp = tcp_sk(sk);
 
-	/* Reserve space for headers. */
-	skb_reserve(buff, MAX_TCP_HEADER);
-
 	/* We'll fix this up when we get a response from the other end.
 	 * See tcp_input.c:tcp_rcv_state_process case TCP_SYN_SENT.
 	 */
@@ -1190,14 +1191,6 @@ int tcp_connect(struct sock *sk, struct sk_buff *buff)
 
 	tp->rcv_ssthresh = tp->rcv_wnd;
 
-	/* Socket identity change complete, no longer
-	 * in TCP_CLOSE, so enter ourselves into the
-	 * hash tables.
-	 */
-	tcp_set_state(sk,TCP_SYN_SENT);
-	if (tp->af_specific->hash_connecting(sk))
-		goto err_out;
-
 	sk->err = 0;
 	sk->done = 0;
 	tp->snd_wnd = 0;
@@ -1211,6 +1204,24 @@ int tcp_connect(struct sock *sk, struct sk_buff *buff)
 	tp->rto = TCP_TIMEOUT_INIT;
 	tp->retransmits = 0;
 	tcp_clear_retrans(tp);
+}
+
+/*
+ * Build a SYN and send it off.
+ */ 
+int tcp_connect(struct sock *sk)
+{
+	struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+	struct sk_buff *buff;
+
+	tcp_connect_init(sk);
+
+	buff = alloc_skb(MAX_TCP_HEADER + 15, sk->allocation);
+	if (unlikely(buff == NULL))
+		return -ENOBUFS;
+
+	/* Reserve space for headers. */
+	skb_reserve(buff, MAX_TCP_HEADER);
 
 	TCP_SKB_CB(buff)->flags = TCPCB_FLAG_SYN;
 	TCP_ECN_send_syn(tp, buff);
@@ -1233,11 +1244,6 @@ int tcp_connect(struct sock *sk, struct sk_buff *buff)
 	/* Timer for repeating the SYN until an answer. */
 	tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
 	return 0;
-
-err_out:
-	tcp_set_state(sk,TCP_CLOSE);
-	kfree_skb(buff);
-	return -EADDRNOTAVAIL;
 }
 
 /* Send out a delayed ack, the caller does the policy checking

net/ipv6/tcp_ipv6.c

@@ -136,7 +136,7 @@ static int tcp_v6_get_port(struct sock *sk, unsigned short snum)
 				break;
 	}
 	if (tb != NULL && tb->owners != NULL) {
-		if (tb->fastreuse != 0 && sk->reuse != 0 && sk->state != TCP_LISTEN) {
+		if (tb->fastreuse > 0 && sk->reuse != 0 && sk->state != TCP_LISTEN) {
 			goto success;
 		} else {
 			struct ipv6_pinfo *np = inet6_sk(sk);
@@ -499,11 +499,21 @@ static int tcp_v6_check_established(struct sock *sk)
 	return -EADDRNOTAVAIL;
 }
 
-static int tcp_v6_hash_connecting(struct sock *sk)
+static int tcp_v6_hash_connect(struct sock *sk)
 {
-	unsigned short snum = inet_sk(sk)->num;
-	struct tcp_bind_hashbucket *head = &tcp_bhash[tcp_bhashfn(snum)];
-	struct tcp_bind_bucket *tb = head->chain;
+	struct tcp_bind_hashbucket *head;
+	struct tcp_bind_bucket *tb;
+
+	/* XXX */
+	if (inet_sk(sk)->num == 0) { 
+		int err = tcp_v6_get_port(sk, inet_sk(sk)->num);
+		if (err)
+			return err;
+		inet_sk(sk)->sport = htons(inet_sk(sk)->num);
+	}
+
+	head = &tcp_bhash[tcp_bhashfn(inet_sk(sk)->num)];
+	tb = head->chain;
 
 	spin_lock_bh(&head->lock);
 
@@ -534,7 +544,6 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 	struct in6_addr saddr_buf;
 	struct flowi fl;
 	struct dst_entry *dst;
-	struct sk_buff *buff;
 	int addr_type;
 	int err;
 
@@ -675,17 +684,12 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 		tp->ext_header_len = np->opt->opt_flen + np->opt->opt_nflen;
 	tp->mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
 
-	err = -ENOBUFS;
-	buff = alloc_skb(MAX_TCP_HEADER + 15, sk->allocation);
-
-	if (buff == NULL)
-		goto failure;
-
 	inet->dport = usin->sin6_port;
 
-	/*
-	 *	Init variables
-	 */
+	tcp_set_state(sk, TCP_SYN_SENT);
+	err = tcp_v6_hash_connect(sk);
+	if (err)
+		goto late_failure;
 
 	if (!tp->write_seq)
 		tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
@@ -693,10 +697,14 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 							     inet->sport,
 							     inet->dport);
 
-	err = tcp_connect(sk, buff);
-	if (err == 0)
-		return 0;
+	err = tcp_connect(sk);
+	if (err)
+		goto late_failure;
+
+	return 0;
 
+late_failure:
+	tcp_set_state(sk, TCP_CLOSE);
 failure:
 	__sk_dst_reset(sk);
 	inet->dport = 0;
@@ -1785,7 +1793,6 @@ static struct tcp_func ipv6_specific = {
 	tcp_v6_rebuild_header,
 	tcp_v6_conn_request,
 	tcp_v6_syn_recv_sock,
-	tcp_v6_hash_connecting,
 	tcp_v6_remember_stamp,
 	sizeof(struct ipv6hdr),
 
@@ -1805,7 +1812,6 @@ static struct tcp_func ipv6_mapped = {
 	tcp_v4_rebuild_header,
 	tcp_v6_conn_request,
 	tcp_v6_syn_recv_sock,
-	tcp_v4_hash_connecting,
 	tcp_v4_remember_stamp,
 	sizeof(struct iphdr),
 

net/netsyms.c

@@ -364,7 +364,6 @@ EXPORT_SYMBOL(tcp_inherit_port);
 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
 EXPORT_SYMBOL(tcp_v4_do_rcv);
 EXPORT_SYMBOL(tcp_v4_connect);
-EXPORT_SYMBOL(tcp_v4_hash_connecting);
 EXPORT_SYMBOL(tcp_unhash);
 EXPORT_SYMBOL(udp_prot);
 EXPORT_SYMBOL(tcp_prot);