[mirror_ubuntu-artful-kernel.git] / net / rds / tcp_listen.c

/*
 * Copyright (c) 2006 Oracle.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/gfp.h>
#include <linux/in.h>
#include <net/tcp.h>

#include "rds.h"
#include "tcp.h"

int rds_tcp_keepalive(struct socket *sock)
{
	/* values below based on xs_udp_default_timeout */
	int keepidle = 5; /* send a probe 'keepidle' secs after last data */
	int keepcnt = 5; /* number of unack'ed probes before declaring dead */
	int keepalive = 1;
	int ret = 0;

	ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
				(char *)&keepalive, sizeof(keepalive));
	if (ret < 0)
		goto bail;

	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
				(char *)&keepcnt, sizeof(keepcnt));
	if (ret < 0)
		goto bail;

	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
				(char *)&keepidle, sizeof(keepidle));
	if (ret < 0)
		goto bail;

	/* KEEPINTVL is the interval between successive probes. We follow
	 * the model in xs_tcp_finish_connecting() and re-use keepidle.
	 */
	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
				(char *)&keepidle, sizeof(keepidle));
bail:
	return ret;
}

/* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the
 * client's ipaddr < server's ipaddr. Otherwise, close the accepted
 * socket and force a reconneect from smaller -> larger ip addr. The reason
 * we special case cp_index 0 is to allow the rds probe ping itself to itself
 * get through efficiently.
 * Since reconnects are only initiated from the node with the numerically
 * smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side
 * by moving them to CONNECTING in this function.
 */
struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn)
{
	int i;
	bool peer_is_smaller = (conn->c_faddr < conn->c_laddr);
	int npaths = conn->c_npaths;

	if (npaths <= 1) {
		struct rds_conn_path *cp = &conn->c_path[0];
		int ret;

		ret = rds_conn_path_transition(cp, RDS_CONN_DOWN,
					       RDS_CONN_CONNECTING);
		if (!ret)
			rds_conn_path_transition(cp, RDS_CONN_ERROR,
						 RDS_CONN_CONNECTING);
		return cp->cp_transport_data;
	}

	/* for mprds, paths with cp_index > 0 MUST be initiated by the peer
	 * with the smaller address.
	 */
	if (!peer_is_smaller)
		return NULL;

	for (i = 1; i < npaths; i++) {
		struct rds_conn_path *cp = &conn->c_path[i];

		if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
					     RDS_CONN_CONNECTING) ||
		    rds_conn_path_transition(cp, RDS_CONN_ERROR,
					     RDS_CONN_CONNECTING)) {
			return cp->cp_transport_data;
		}
	}
	return NULL;
}

int rds_tcp_accept_one(struct socket *sock)
{
	struct socket *new_sock = NULL;
	struct rds_connection *conn;
	int ret;
	struct inet_sock *inet;
	struct rds_tcp_connection *rs_tcp = NULL;
	int conn_state;
	struct rds_conn_path *cp;

	if (!sock) /* module unload or netns delete in progress */
		return -ENETUNREACH;

	ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
			       sock->sk->sk_type, sock->sk->sk_protocol,
			       &new_sock);
	if (ret)
		goto out;

	new_sock->type = sock->type;
	new_sock->ops = sock->ops;
	ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
	if (ret < 0)
		goto out;

	ret = rds_tcp_keepalive(new_sock);
	if (ret < 0)
		goto out;

	rds_tcp_tune(new_sock);

	inet = inet_sk(new_sock->sk);

	rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
		 &inet->inet_saddr, ntohs(inet->inet_sport),
		 &inet->inet_daddr, ntohs(inet->inet_dport));

	conn = rds_conn_create(sock_net(sock->sk),
			       inet->inet_saddr, inet->inet_daddr,
			       &rds_tcp_transport, GFP_KERNEL);
	if (IS_ERR(conn)) {
		ret = PTR_ERR(conn);
		goto out;
	}
	/* An incoming SYN request came in, and TCP just accepted it.
	 *
	 * If the client reboots, this conn will need to be cleaned up.
	 * rds_tcp_state_change() will do that cleanup
	 */
	rs_tcp = rds_tcp_accept_one_path(conn);
	if (!rs_tcp)
		goto rst_nsk;
	mutex_lock(&rs_tcp->t_conn_path_lock);
	cp = rs_tcp->t_cpath;
	conn_state = rds_conn_path_state(cp);
	if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_UP &&
	    conn_state != RDS_CONN_ERROR)
		goto rst_nsk;
	if (rs_tcp->t_sock) {
		/* Need to resolve a duelling SYN between peers.
		 * We have an outstanding SYN to this peer, which may
		 * potentially have transitioned to the RDS_CONN_UP state,
		 * so we must quiesce any send threads before resetting
		 * c_transport_data.
		 */
		if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) ||
		    !cp->cp_outgoing) {
			goto rst_nsk;
		} else {
			rds_tcp_reset_callbacks(new_sock, cp);
			cp->cp_outgoing = 0;
			/* rds_connect_path_complete() marks RDS_CONN_UP */
			rds_connect_path_complete(cp, RDS_CONN_RESETTING);
		}
	} else {
		rds_tcp_set_callbacks(new_sock, cp);
		rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
	}
	new_sock = NULL;
	ret = 0;
	goto out;
rst_nsk:
	/* reset the newly returned accept sock and bail */
	kernel_sock_shutdown(new_sock, SHUT_RDWR);
	ret = 0;
out:
	if (rs_tcp)
		mutex_unlock(&rs_tcp->t_conn_path_lock);
	if (new_sock)
		sock_release(new_sock);
	return ret;
}

void rds_tcp_listen_data_ready(struct sock *sk)
{
	void (*ready)(struct sock *sk);

	rdsdebug("listen data ready sk %p\n", sk);

	read_lock_bh(&sk->sk_callback_lock);
	ready = sk->sk_user_data;
	if (!ready) { /* check for teardown race */
		ready = sk->sk_data_ready;
		goto out;
	}

	/*
	 * ->sk_data_ready is also called for a newly established child socket
	 * before it has been accepted and the accepter has set up their
	 * data_ready.. we only want to queue listen work for our listening
	 * socket
	 */
	if (sk->sk_state == TCP_LISTEN)
		rds_tcp_accept_work(sk);
	else
		ready = rds_tcp_listen_sock_def_readable(sock_net(sk));

out:
	read_unlock_bh(&sk->sk_callback_lock);
	ready(sk);
}

struct socket *rds_tcp_listen_init(struct net *net)
{
	struct sockaddr_in sin;
	struct socket *sock = NULL;
	int ret;

	ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
	if (ret < 0)
		goto out;

	sock->sk->sk_reuse = SK_CAN_REUSE;
	rds_tcp_nonagle(sock);

	write_lock_bh(&sock->sk->sk_callback_lock);
	sock->sk->sk_user_data = sock->sk->sk_data_ready;
	sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
	write_unlock_bh(&sock->sk->sk_callback_lock);

	sin.sin_family = PF_INET;
	sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
	sin.sin_port = (__force u16)htons(RDS_TCP_PORT);

	ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
	if (ret < 0)
		goto out;

	ret = sock->ops->listen(sock, 64);
	if (ret < 0)
		goto out;

	return sock;
out:
	if (sock)
		sock_release(sock);
	return NULL;
}

void rds_tcp_listen_stop(struct socket *sock)
{
	struct sock *sk;

	if (!sock)
		return;

	sk = sock->sk;

	/* serialize with and prevent further callbacks */
	lock_sock(sk);
	write_lock_bh(&sk->sk_callback_lock);
	if (sk->sk_user_data) {
		sk->sk_data_ready = sk->sk_user_data;
		sk->sk_user_data = NULL;
	}
	write_unlock_bh(&sk->sk_callback_lock);
	release_sock(sk);

	/* wait for accepts to stop and close the socket */
	flush_workqueue(rds_wq);
	sock_release(sock);
}
Commit	Line	Data
70041088 AG	1	/*
	2	* Copyright (c) 2006 Oracle. All rights reserved.
	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*
	32	*/
	33	#include <linux/kernel.h>
5a0e3ad6	34	#include <linux/gfp.h>
70041088 AG	35	#include <linux/in.h>
	36	#include <net/tcp.h>
	37
	38	#include "rds.h"
	39	#include "tcp.h"
	40
467fa153	41	int rds_tcp_keepalive(struct socket *sock)
f711a6ae SV	42	{
	43	/* values below based on xs_udp_default_timeout */
	44	int keepidle = 5; /* send a probe 'keepidle' secs after last data */
	45	int keepcnt = 5; /* number of unack'ed probes before declaring dead */
	46	int keepalive = 1;
	47	int ret = 0;
	48
	49	ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
	50	(char *)&keepalive, sizeof(keepalive));
	51	if (ret < 0)
	52	goto bail;
	53
	54	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
	55	(char *)&keepcnt, sizeof(keepcnt));
	56	if (ret < 0)
	57	goto bail;
	58
	59	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
	60	(char *)&keepidle, sizeof(keepidle));
	61	if (ret < 0)
	62	goto bail;
	63
	64	/* KEEPINTVL is the interval between successive probes. We follow
	65	* the model in xs_tcp_finish_connecting() and re-use keepidle.
	66	*/
	67	ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
	68	(char *)&keepidle, sizeof(keepidle));
	69	bail:
	70	return ret;
	71	}
	72
5916e2c1 SV	73	/* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the
	74	* client's ipaddr < server's ipaddr. Otherwise, close the accepted
	75	* socket and force a reconneect from smaller -> larger ip addr. The reason
	76	* we special case cp_index 0 is to allow the rds probe ping itself to itself
	77	* get through efficiently.
	78	* Since reconnects are only initiated from the node with the numerically
	79	* smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side
	80	* by moving them to CONNECTING in this function.
	81	*/
	82	struct rds_tcp_connection rds_tcp_accept_one_path(struct rds_connection conn)
	83	{
	84	int i;
	85	bool peer_is_smaller = (conn->c_faddr < conn->c_laddr);
	86	int npaths = conn->c_npaths;
	87
	88	if (npaths <= 1) {
	89	struct rds_conn_path *cp = &conn->c_path[0];
	90	int ret;
	91
	92	ret = rds_conn_path_transition(cp, RDS_CONN_DOWN,
	93	RDS_CONN_CONNECTING);
	94	if (!ret)
	95	rds_conn_path_transition(cp, RDS_CONN_ERROR,
	96	RDS_CONN_CONNECTING);
	97	return cp->cp_transport_data;
	98	}
	99
	100	/* for mprds, paths with cp_index > 0 MUST be initiated by the peer
	101	* with the smaller address.
	102	*/
	103	if (!peer_is_smaller)
	104	return NULL;
	105
	106	for (i = 1; i < npaths; i++) {
	107	struct rds_conn_path *cp = &conn->c_path[i];
	108
	109	if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
	110	RDS_CONN_CONNECTING) \|\|
	111	rds_conn_path_transition(cp, RDS_CONN_ERROR,
	112	RDS_CONN_CONNECTING)) {
	113	return cp->cp_transport_data;
	114	}
	115	}
	116	return NULL;
	117	}
	118
467fa153	119	int rds_tcp_accept_one(struct socket *sock)
70041088 AG	120	{
	121	struct socket *new_sock = NULL;
	122	struct rds_connection *conn;
	123	int ret;
	124	struct inet_sock *inet;
bd7c5f98 SV	125	struct rds_tcp_connection *rs_tcp = NULL;
bd7c5f98 SV	126	int conn_state;
ea3b1ea5	127	struct rds_conn_path *cp;
70041088	128
37e14f4f SV	129	if (!sock) /* module unload or netns delete in progress */
	130	return -ENETUNREACH;
	131
d5a8ac28 SV	132	ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
	133	sock->sk->sk_type, sock->sk->sk_protocol,
	134	&new_sock);
70041088 AG	135	if (ret)
	136	goto out;
	137
	138	new_sock->type = sock->type;
	139	new_sock->ops = sock->ops;
	140	ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
	141	if (ret < 0)
	142	goto out;
	143
f711a6ae SV	144	ret = rds_tcp_keepalive(new_sock);
	145	if (ret < 0)
	146	goto out;
	147
70041088 AG	148	rds_tcp_tune(new_sock);
	149
	150	inet = inet_sk(new_sock->sk);
	151
6884b348 JP	152	rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
	153	&inet->inet_saddr, ntohs(inet->inet_sport),
	154	&inet->inet_daddr, ntohs(inet->inet_dport));
70041088	155
d5a8ac28 SV	156	conn = rds_conn_create(sock_net(sock->sk),
d5a8ac28 SV	157	inet->inet_saddr, inet->inet_daddr,
c720c7e8	158	&rds_tcp_transport, GFP_KERNEL);
70041088 AG	159	if (IS_ERR(conn)) {
	160	ret = PTR_ERR(conn);
	161	goto out;
	162	}
f711a6ae	163	/* An incoming SYN request came in, and TCP just accepted it.
f711a6ae SV	164	*
	165	* If the client reboots, this conn will need to be cleaned up.
	166	* rds_tcp_state_change() will do that cleanup
	167	*/
5916e2c1 SV	168	rs_tcp = rds_tcp_accept_one_path(conn);
	169	if (!rs_tcp)
	170	goto rst_nsk;
02105b2c	171	mutex_lock(&rs_tcp->t_conn_path_lock);
5916e2c1 SV	172	cp = rs_tcp->t_cpath;
	173	conn_state = rds_conn_path_state(cp);
	174	if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_UP &&
	175	conn_state != RDS_CONN_ERROR)
bd7c5f98	176	goto rst_nsk;
eb192840 SV	177	if (rs_tcp->t_sock) {
	178	/* Need to resolve a duelling SYN between peers.
	179	* We have an outstanding SYN to this peer, which may
	180	* potentially have transitioned to the RDS_CONN_UP state,
	181	* so we must quiesce any send threads before resetting
	182	* c_transport_data.
	183	*/
c948bb5c	184	if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) \|\|
5916e2c1	185	!cp->cp_outgoing) {
bd7c5f98	186	goto rst_nsk;
c948bb5c	187	} else {
ea3b1ea5	188	rds_tcp_reset_callbacks(new_sock, cp);
5916e2c1	189	cp->cp_outgoing = 0;
9c79440e	190	/* rds_connect_path_complete() marks RDS_CONN_UP */
ea3b1ea5	191	rds_connect_path_complete(cp, RDS_CONN_RESETTING);
eb192840	192	}
335b48d9	193	} else {
ea3b1ea5 SV	194	rds_tcp_set_callbacks(new_sock, cp);
ea3b1ea5 SV	195	rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
eb192840	196	}
70041088 AG	197	new_sock = NULL;
70041088 AG	198	ret = 0;
bd7c5f98 SV	199	goto out;
	200	rst_nsk:
	201	/* reset the newly returned accept sock and bail */
335b48d9	202	kernel_sock_shutdown(new_sock, SHUT_RDWR);
bd7c5f98	203	ret = 0;
70041088	204	out:
bd7c5f98	205	if (rs_tcp)
02105b2c	206	mutex_unlock(&rs_tcp->t_conn_path_lock);
70041088 AG	207	if (new_sock)
	208	sock_release(new_sock);
	209	return ret;
	210	}
	211
676d2369	212	void rds_tcp_listen_data_ready(struct sock *sk)
70041088	213	{
676d2369	214	void (ready)(struct sock sk);
70041088 AG	215
	216	rdsdebug("listen data ready sk %p\n", sk);
	217
38036629	218	read_lock_bh(&sk->sk_callback_lock);
70041088	219	ready = sk->sk_user_data;
8690bfa1	220	if (!ready) { /* check for teardown race */
70041088 AG	221	ready = sk->sk_data_ready;
	222	goto out;
	223	}
	224
	225	/*
	226	* ->sk_data_ready is also called for a newly established child socket
	227	* before it has been accepted and the accepter has set up their
	228	* data_ready.. we only want to queue listen work for our listening
	229	* socket
	230	*/
	231	if (sk->sk_state == TCP_LISTEN)
467fa153	232	rds_tcp_accept_work(sk);
a93d01f5 SV	233	else
a93d01f5 SV	234	ready = rds_tcp_listen_sock_def_readable(sock_net(sk));
70041088 AG	235
70041088 AG	236	out:
38036629	237	read_unlock_bh(&sk->sk_callback_lock);
676d2369	238	ready(sk);
70041088 AG	239	}
70041088 AG	240
467fa153	241	struct socket rds_tcp_listen_init(struct net net)
70041088 AG	242	{
	243	struct sockaddr_in sin;
	244	struct socket *sock = NULL;
	245	int ret;
	246
467fa153	247	ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
70041088 AG	248	if (ret < 0)
	249	goto out;
	250
4a17fd52	251	sock->sk->sk_reuse = SK_CAN_REUSE;
70041088 AG	252	rds_tcp_nonagle(sock);
	253
	254	write_lock_bh(&sock->sk->sk_callback_lock);
	255	sock->sk->sk_user_data = sock->sk->sk_data_ready;
	256	sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
	257	write_unlock_bh(&sock->sk->sk_callback_lock);
	258
01728371	259	sin.sin_family = PF_INET;
70041088 AG	260	sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
	261	sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
	262
	263	ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
	264	if (ret < 0)
	265	goto out;
	266
	267	ret = sock->ops->listen(sock, 64);
	268	if (ret < 0)
	269	goto out;
	270
467fa153	271	return sock;
70041088 AG	272	out:
	273	if (sock)
	274	sock_release(sock);
467fa153	275	return NULL;
70041088 AG	276	}
70041088 AG	277
467fa153	278	void rds_tcp_listen_stop(struct socket *sock)
70041088	279	{
70041088 AG	280	struct sock *sk;
70041088 AG	281
8690bfa1	282	if (!sock)
70041088 AG	283	return;
	284
	285	sk = sock->sk;
	286
	287	/* serialize with and prevent further callbacks */
	288	lock_sock(sk);
	289	write_lock_bh(&sk->sk_callback_lock);
	290	if (sk->sk_user_data) {
	291	sk->sk_data_ready = sk->sk_user_data;
	292	sk->sk_user_data = NULL;
	293	}
	294	write_unlock_bh(&sk->sk_callback_lock);
	295	release_sock(sk);
	296
	297	/* wait for accepts to stop and close the socket */
	298	flush_workqueue(rds_wq);
	299	sock_release(sock);
70041088	300	}