[INET]: Generalise the tcp_listen_ lock routines (f3f05f70) · Commits · EulixOS / Software / Kernel

include/net/inet_hashtables.h

+48 −0

Original line number	Diff line number	Diff line
		@@ -19,10 +19,14 @@
		#include <linux/list.h>
		#include <linux/slab.h>
		#include <linux/spinlock.h>
		#include <linux/tcp.h> /* only for TCP_LISTEN, damn :-( */
		#include <linux/types.h>
		#include <linux/wait.h>

		#include <net/sock.h>

		#include <asm/atomic.h>

		/* This is for all connections with a full identity, no wildcards.
		* New scheme, half the table is for TIME_WAIT, the other half is
		* for the rest. I'll experiment with dynamic table growth later.
		@@ -192,4 +196,48 @@ static inline void inet_inherit_port(struct inet_hashinfo *table,

		extern void inet_put_port(struct inet_hashinfo table, struct sock sk);

		extern void inet_listen_wlock(struct inet_hashinfo *hashinfo);

		/*
		* - We may sleep inside this lock.
		* - If sleeping is not required (or called from BH),
		* use plain read_(un)lock(&inet_hashinfo.lhash_lock).
		*/
		static inline void inet_listen_lock(struct inet_hashinfo *hashinfo)
		{
		/* read_lock synchronizes to candidates to writers */
		read_lock(&hashinfo->lhash_lock);
		atomic_inc(&hashinfo->lhash_users);
		read_unlock(&hashinfo->lhash_lock);
		}

		static inline void inet_listen_unlock(struct inet_hashinfo *hashinfo)
		{
		if (atomic_dec_and_test(&hashinfo->lhash_users))
		wake_up(&hashinfo->lhash_wait);
		}

		static inline void __inet_hash(struct inet_hashinfo *hashinfo,
		struct sock *sk, const int listen_possible)
		{
		struct hlist_head *list;
		rwlock_t *lock;

		BUG_TRAP(sk_unhashed(sk));
		if (listen_possible && sk->sk_state == TCP_LISTEN) {
		list = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
		lock = &hashinfo->lhash_lock;
		inet_listen_wlock(hashinfo);
		} else {
		sk->sk_hashent = inet_sk_ehashfn(sk, hashinfo->ehash_size);
		list = &hashinfo->ehash[sk->sk_hashent].chain;
		lock = &hashinfo->ehash[sk->sk_hashent].lock;
		write_lock(lock);
		}
		__sk_add_node(sk, list);
		sock_prot_inc_use(sk->sk_prot);
		write_unlock(lock);
		if (listen_possible && sk->sk_state == TCP_LISTEN)
		wake_up(&hashinfo->lhash_wait);
		}
		#endif /* _INET_HASHTABLES_H */

include/net/tcp.h

+0 −21

Original line number	Diff line number	Diff line
		@@ -1447,27 +1447,6 @@ static __inline__ void tcp_openreq_init(struct request_sock *req,

		extern void tcp_enter_memory_pressure(void);

		extern void tcp_listen_wlock(void);

		/* - We may sleep inside this lock.
		* - If sleeping is not required (or called from BH),
		* use plain read_(un)lock(&inet_hashinfo.lhash_lock).
		*/

		static inline void tcp_listen_lock(void)
		{
		/* read_lock synchronizes to candidates to writers */
		read_lock(&tcp_hashinfo.lhash_lock);
		atomic_inc(&tcp_hashinfo.lhash_users);
		read_unlock(&tcp_hashinfo.lhash_lock);
		}

		static inline void tcp_listen_unlock(void)
		{
		if (atomic_dec_and_test(&tcp_hashinfo.lhash_users))
		wake_up(&tcp_hashinfo.lhash_wait);
		}

		static inline int keepalive_intvl_when(const struct tcp_sock *tp)
		{
		return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;

net/ipv4/inet_hashtables.c

+32 −0

Original line number	Diff line number	Diff line
		@@ -15,7 +15,9 @@

		#include <linux/config.h>
		#include <linux/module.h>
		#include <linux/sched.h>
		#include <linux/slab.h>
		#include <linux/wait.h>

		#include <net/inet_hashtables.h>

		@@ -89,3 +91,33 @@ void inet_put_port(struct inet_hashinfo hashinfo, struct sock sk)
		}

		EXPORT_SYMBOL(inet_put_port);

		/*
		* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
		* Look, when several writers sleep and reader wakes them up, all but one
		* immediately hit write lock and grab all the cpus. Exclusive sleep solves
		* this, _but_ remember, it adds useless work on UP machines (wake up each
		* exclusive lock release). It should be ifdefed really.
		*/
		void inet_listen_wlock(struct inet_hashinfo *hashinfo)
		{
		write_lock(&hashinfo->lhash_lock);

		if (atomic_read(&hashinfo->lhash_users)) {
		DEFINE_WAIT(wait);

		for (;;) {
		prepare_to_wait_exclusive(&hashinfo->lhash_wait,
		&wait, TASK_UNINTERRUPTIBLE);
		if (!atomic_read(&hashinfo->lhash_users))
		break;
		write_unlock_bh(&hashinfo->lhash_lock);
		schedule();
		write_lock_bh(&hashinfo->lhash_lock);
		}

		finish_wait(&hashinfo->lhash_wait, &wait);
		}
		}

		EXPORT_SYMBOL(inet_listen_wlock);

net/ipv4/tcp_diag.c

+4 −4

Original line number	Diff line number	Diff line
		@@ -589,7 +589,7 @@ static int tcpdiag_dump(struct sk_buff skb, struct netlink_callback cb)
		if (cb->args[0] == 0) {
		if (!(r->tcpdiag_states&(TCPF_LISTEN\|TCPF_SYN_RECV)))
		goto skip_listen_ht;
		tcp_listen_lock();
		inet_listen_lock(&tcp_hashinfo);
		for (i = s_i; i < INET_LHTABLE_SIZE; i++) {
		struct sock *sk;
		struct hlist_node *node;
		@@ -613,7 +613,7 @@ static int tcpdiag_dump(struct sk_buff skb, struct netlink_callback cb)
		goto syn_recv;

		if (tcpdiag_dump_sock(skb, sk, cb) < 0) {
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		goto done;
		}

		@@ -622,7 +622,7 @@ static int tcpdiag_dump(struct sk_buff skb, struct netlink_callback cb)
		goto next_listen;

		if (tcpdiag_dump_reqs(skb, sk, cb) < 0) {
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		goto done;
		}

		@@ -636,7 +636,7 @@ static int tcpdiag_dump(struct sk_buff skb, struct netlink_callback cb)
		cb->args[3] = 0;
		cb->args[4] = 0;
		}
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		skip_listen_ht:
		cb->args[0] = 1;
		s_i = num = s_num = 0;

net/ipv4/tcp_ipv4.c

+9 −61

Original line number	Diff line number	Diff line
		@@ -228,62 +228,11 @@ static int tcp_v4_get_port(struct sock *sk, unsigned short snum)
		return ret;
		}

		/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
		* Look, when several writers sleep and reader wakes them up, all but one
		* immediately hit write lock and grab all the cpus. Exclusive sleep solves
		* this, _but_ remember, it adds useless work on UP machines (wake up each
		* exclusive lock release). It should be ifdefed really.
		*/

		void tcp_listen_wlock(void)
		{
		write_lock(&tcp_hashinfo.lhash_lock);

		if (atomic_read(&tcp_hashinfo.lhash_users)) {
		DEFINE_WAIT(wait);

		for (;;) {
		prepare_to_wait_exclusive(&tcp_hashinfo.lhash_wait,
		&wait, TASK_UNINTERRUPTIBLE);
		if (!atomic_read(&tcp_hashinfo.lhash_users))
		break;
		write_unlock_bh(&tcp_hashinfo.lhash_lock);
		schedule();
		write_lock_bh(&tcp_hashinfo.lhash_lock);
		}

		finish_wait(&tcp_hashinfo.lhash_wait, &wait);
		}
		}

		static __inline__ void __tcp_v4_hash(struct sock *sk, const int listen_possible)
		{
		struct hlist_head *list;
		rwlock_t *lock;

		BUG_TRAP(sk_unhashed(sk));
		if (listen_possible && sk->sk_state == TCP_LISTEN) {
		list = &tcp_hashinfo.listening_hash[inet_sk_listen_hashfn(sk)];
		lock = &tcp_hashinfo.lhash_lock;
		tcp_listen_wlock();
		} else {
		sk->sk_hashent = inet_sk_ehashfn(sk, tcp_hashinfo.ehash_size);
		list = &tcp_hashinfo.ehash[sk->sk_hashent].chain;
		lock = &tcp_hashinfo.ehash[sk->sk_hashent].lock;
		write_lock(lock);
		}
		__sk_add_node(sk, list);
		sock_prot_inc_use(sk->sk_prot);
		write_unlock(lock);
		if (listen_possible && sk->sk_state == TCP_LISTEN)
		wake_up(&tcp_hashinfo.lhash_wait);
		}

		static void tcp_v4_hash(struct sock *sk)
		{
		if (sk->sk_state != TCP_CLOSE) {
		local_bh_disable();
		__tcp_v4_hash(sk, 1);
		__inet_hash(&tcp_hashinfo, sk, 1);
		local_bh_enable();
		}
		}
		@@ -297,7 +246,7 @@ void tcp_unhash(struct sock *sk)

		if (sk->sk_state == TCP_LISTEN) {
		local_bh_disable();
		tcp_listen_wlock();
		inet_listen_wlock(&tcp_hashinfo);
		lock = &tcp_hashinfo.lhash_lock;
		} else {
		struct inet_ehash_bucket *head = &tcp_hashinfo.ehash[sk->sk_hashent];
		@@ -624,7 +573,7 @@ static inline int tcp_v4_hash_connect(struct sock *sk)
		inet_bind_hash(sk, tb, port);
		if (sk_unhashed(sk)) {
		inet_sk(sk)->sport = htons(port);
		__tcp_v4_hash(sk, 0);
		__inet_hash(&tcp_hashinfo, sk, 0);
		}
		spin_unlock(&head->lock);

		@@ -641,7 +590,7 @@ static inline int tcp_v4_hash_connect(struct sock *sk)
		tb = inet_sk(sk)->bind_hash;
		spin_lock_bh(&head->lock);
		if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
		__tcp_v4_hash(sk, 0);
		__inet_hash(&tcp_hashinfo, sk, 0);
		spin_unlock_bh(&head->lock);
		return 0;
		} else {
		@@ -1479,7 +1428,7 @@ struct sock tcp_v4_syn_recv_sock(struct sock sk, struct sk_buff *skb,
		newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
		tcp_initialize_rcv_mss(newsk);

		__tcp_v4_hash(newsk, 0);
		__inet_hash(&tcp_hashinfo, newsk, 0);
		__inet_inherit_port(&tcp_hashinfo, sk, newsk);

		return newsk;
		@@ -2102,12 +2051,12 @@ static void tcp_get_idx(struct seq_file seq, loff_t pos)
		void *rc;
		struct tcp_iter_state* st = seq->private;

		tcp_listen_lock();
		inet_listen_lock(&tcp_hashinfo);
		st->state = TCP_SEQ_STATE_LISTENING;
		rc = listening_get_idx(seq, &pos);

		if (!rc) {
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		local_bh_disable();
		st->state = TCP_SEQ_STATE_ESTABLISHED;
		rc = established_get_idx(seq, pos);
		@@ -2140,7 +2089,7 @@ static void tcp_seq_next(struct seq_file seq, void v, loff_t pos)
		case TCP_SEQ_STATE_LISTENING:
		rc = listening_get_next(seq, v);
		if (!rc) {
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		local_bh_disable();
		st->state = TCP_SEQ_STATE_ESTABLISHED;
		rc = established_get_first(seq);
		@@ -2168,7 +2117,7 @@ static void tcp_seq_stop(struct seq_file seq, void v)
		}
		case TCP_SEQ_STATE_LISTENING:
		if (v != SEQ_START_TOKEN)
		tcp_listen_unlock();
		inet_listen_unlock(&tcp_hashinfo);
		break;
		case TCP_SEQ_STATE_TIME_WAIT:
		case TCP_SEQ_STATE_ESTABLISHED:
		@@ -2431,7 +2380,6 @@ void __init tcp_v4_init(struct net_proto_family *ops)
		EXPORT_SYMBOL(ipv4_specific);
		EXPORT_SYMBOL(inet_bind_bucket_create);
		EXPORT_SYMBOL(tcp_hashinfo);
		EXPORT_SYMBOL(tcp_listen_wlock);
		EXPORT_SYMBOL(tcp_prot);
		EXPORT_SYMBOL(tcp_unhash);
		EXPORT_SYMBOL(tcp_v4_conn_request);