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>
34#include <linux/in.h>
35#include <net/tcp.h>
36
37#include "rds.h"
38#include "tcp.h"
39
40static void rds_tcp_cork(struct socket *sock, int val)
41{
42	mm_segment_t oldfs;
43
44	oldfs = get_fs();
45	set_fs(KERNEL_DS);
46	sock->ops->setsockopt(sock, SOL_TCP, TCP_CORK, (char __user *)&val,
47			      sizeof(val));
48	set_fs(oldfs);
49}
50
51void rds_tcp_xmit_prepare(struct rds_connection *conn)
52{
53	struct rds_tcp_connection *tc = conn->c_transport_data;
54
55	rds_tcp_cork(tc->t_sock, 1);
56}
57
58void rds_tcp_xmit_complete(struct rds_connection *conn)
59{
60	struct rds_tcp_connection *tc = conn->c_transport_data;
61
62	rds_tcp_cork(tc->t_sock, 0);
63}
64
65/* the core send_sem serializes this with other xmit and shutdown */
66static int rds_tcp_sendmsg(struct socket *sock, void *data, unsigned int len)
67{
68	struct kvec vec = {
69                .iov_base = data,
70                .iov_len = len,
71	};
72        struct msghdr msg = {
73                .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL,
74        };
75
76	return kernel_sendmsg(sock, &msg, &vec, 1, vec.iov_len);
77}
78
79/* the core send_sem serializes this with other xmit and shutdown */
80int rds_tcp_xmit(struct rds_connection *conn, struct rds_message *rm,
81	         unsigned int hdr_off, unsigned int sg, unsigned int off)
82{
83	struct rds_tcp_connection *tc = conn->c_transport_data;
84	int done = 0;
85	int ret = 0;
86	int more;
87
88	if (hdr_off == 0) {
89		/*
90		 * m_ack_seq is set to the sequence number of the last byte of
91		 * header and data.  see rds_tcp_is_acked().
92		 */
93		tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc);
94		rm->m_ack_seq = tc->t_last_sent_nxt +
95				sizeof(struct rds_header) +
96				be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1;
97		smp_mb__before_atomic();
98		set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags);
99		tc->t_last_expected_una = rm->m_ack_seq + 1;
100
101		rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
102			 rm, rds_tcp_snd_nxt(tc),
103			 (unsigned long long)rm->m_ack_seq);
104	}
105
106	if (hdr_off < sizeof(struct rds_header)) {
107		/* see rds_tcp_write_space() */
108		set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags);
109
110		ret = rds_tcp_sendmsg(tc->t_sock,
111				      (void *)&rm->m_inc.i_hdr + hdr_off,
112				      sizeof(rm->m_inc.i_hdr) - hdr_off);
113		if (ret < 0)
114			goto out;
115		done += ret;
116		if (hdr_off + done != sizeof(struct rds_header))
117			goto out;
118	}
119
120	more = rm->data.op_nents > 1 ? (MSG_MORE | MSG_SENDPAGE_NOTLAST) : 0;
121	while (sg < rm->data.op_nents) {
122		int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
123
124		ret = tc->t_sock->ops->sendpage(tc->t_sock,
125						sg_page(&rm->data.op_sg[sg]),
126						rm->data.op_sg[sg].offset + off,
127						rm->data.op_sg[sg].length - off,
128						flags);
129		rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]),
130			 rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off,
131			 ret);
132		if (ret <= 0)
133			break;
134
135		off += ret;
136		done += ret;
137		if (off == rm->data.op_sg[sg].length) {
138			off = 0;
139			sg++;
140		}
141		if (sg == rm->data.op_nents - 1)
142			more = 0;
143	}
144
145out:
146	if (ret <= 0) {
147		/* write_space will hit after EAGAIN, all else fatal */
148		if (ret == -EAGAIN) {
149			rds_tcp_stats_inc(s_tcp_sndbuf_full);
150			ret = 0;
151		} else {
152			printk(KERN_WARNING "RDS/tcp: send to %pI4 "
153			       "returned %d, disconnecting and reconnecting\n",
154			       &conn->c_faddr, ret);
155			rds_conn_drop(conn);
156		}
157	}
158	if (done == 0)
159		done = ret;
160	return done;
161}
162
163/*
164 * rm->m_ack_seq is set to the tcp sequence number that corresponds to the
165 * last byte of the message, including the header.  This means that the
166 * entire message has been received if rm->m_ack_seq is "before" the next
167 * unacked byte of the TCP sequence space.  We have to do very careful
168 * wrapping 32bit comparisons here.
169 */
170static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack)
171{
172	if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags))
173		return 0;
174	return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0;
175}
176
177void rds_tcp_write_space(struct sock *sk)
178{
179	void (*write_space)(struct sock *sk);
180	struct rds_connection *conn;
181	struct rds_tcp_connection *tc;
182
183	read_lock(&sk->sk_callback_lock);
184	conn = sk->sk_user_data;
185	if (!conn) {
186		write_space = sk->sk_write_space;
187		goto out;
188	}
189
190	tc = conn->c_transport_data;
191	rdsdebug("write_space for tc %p\n", tc);
192	write_space = tc->t_orig_write_space;
193	rds_tcp_stats_inc(s_tcp_write_space_calls);
194
195	rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc));
196	tc->t_last_seen_una = rds_tcp_snd_una(tc);
197	rds_send_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked);
198
199        if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf)
200		queue_delayed_work(rds_wq, &conn->c_send_w, 0);
201
202out:
203	read_unlock(&sk->sk_callback_lock);
204
205	/*
206	 * write_space is only called when data leaves tcp's send queue if
207	 * SOCK_NOSPACE is set.  We set SOCK_NOSPACE every time we put
208	 * data in tcp's send queue because we use write_space to parse the
209	 * sequence numbers and notice that rds messages have been fully
210	 * received.
211	 *
212	 * tcp's write_space clears SOCK_NOSPACE if the send queue has more
213	 * than a certain amount of space. So we need to set it again *after*
214	 * we call tcp's write_space or else we might only get called on the
215	 * first of a series of incoming tcp acks.
216	 */
217	write_space(sk);
218
219	if (sk->sk_socket)
220		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
221}
222