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
rds_tcp_cork(struct socket * sock,int val)40 static 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
rds_tcp_xmit_prepare(struct rds_connection * conn)51 void 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
rds_tcp_xmit_complete(struct rds_connection * conn)58 void 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 */
rds_tcp_sendmsg(struct socket * sock,void * data,unsigned int len)66 static 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 */
rds_tcp_xmit(struct rds_connection * conn,struct rds_message * rm,unsigned int hdr_off,unsigned int sg,unsigned int off)80 int 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
87 if (hdr_off == 0) {
88 /*
89 * m_ack_seq is set to the sequence number of the last byte of
90 * header and data. see rds_tcp_is_acked().
91 */
92 tc->t_last_sent_nxt = rds_tcp_snd_nxt(tc);
93 rm->m_ack_seq = tc->t_last_sent_nxt +
94 sizeof(struct rds_header) +
95 be32_to_cpu(rm->m_inc.i_hdr.h_len) - 1;
96 smp_mb__before_clear_bit();
97 set_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags);
98 tc->t_last_expected_una = rm->m_ack_seq + 1;
99
100 rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
101 rm, rds_tcp_snd_nxt(tc),
102 (unsigned long long)rm->m_ack_seq);
103 }
104
105 if (hdr_off < sizeof(struct rds_header)) {
106 /* see rds_tcp_write_space() */
107 set_bit(SOCK_NOSPACE, &tc->t_sock->sk->sk_socket->flags);
108
109 ret = rds_tcp_sendmsg(tc->t_sock,
110 (void *)&rm->m_inc.i_hdr + hdr_off,
111 sizeof(rm->m_inc.i_hdr) - hdr_off);
112 if (ret < 0)
113 goto out;
114 done += ret;
115 if (hdr_off + done != sizeof(struct rds_header))
116 goto out;
117 }
118
119 while (sg < rm->data.op_nents) {
120 ret = tc->t_sock->ops->sendpage(tc->t_sock,
121 sg_page(&rm->data.op_sg[sg]),
122 rm->data.op_sg[sg].offset + off,
123 rm->data.op_sg[sg].length - off,
124 MSG_DONTWAIT|MSG_NOSIGNAL);
125 rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm->data.op_sg[sg]),
126 rm->data.op_sg[sg].offset + off, rm->data.op_sg[sg].length - off,
127 ret);
128 if (ret <= 0)
129 break;
130
131 off += ret;
132 done += ret;
133 if (off == rm->data.op_sg[sg].length) {
134 off = 0;
135 sg++;
136 }
137 }
138
139 out:
140 if (ret <= 0) {
141 /* write_space will hit after EAGAIN, all else fatal */
142 if (ret == -EAGAIN) {
143 rds_tcp_stats_inc(s_tcp_sndbuf_full);
144 ret = 0;
145 } else {
146 printk(KERN_WARNING "RDS/tcp: send to %pI4 "
147 "returned %d, disconnecting and reconnecting\n",
148 &conn->c_faddr, ret);
149 rds_conn_drop(conn);
150 }
151 }
152 if (done == 0)
153 done = ret;
154 return done;
155 }
156
157 /*
158 * rm->m_ack_seq is set to the tcp sequence number that corresponds to the
159 * last byte of the message, including the header. This means that the
160 * entire message has been received if rm->m_ack_seq is "before" the next
161 * unacked byte of the TCP sequence space. We have to do very careful
162 * wrapping 32bit comparisons here.
163 */
rds_tcp_is_acked(struct rds_message * rm,uint64_t ack)164 static int rds_tcp_is_acked(struct rds_message *rm, uint64_t ack)
165 {
166 if (!test_bit(RDS_MSG_HAS_ACK_SEQ, &rm->m_flags))
167 return 0;
168 return (__s32)((u32)rm->m_ack_seq - (u32)ack) < 0;
169 }
170
rds_tcp_write_space(struct sock * sk)171 void rds_tcp_write_space(struct sock *sk)
172 {
173 void (*write_space)(struct sock *sk);
174 struct rds_connection *conn;
175 struct rds_tcp_connection *tc;
176
177 read_lock_bh(&sk->sk_callback_lock);
178 conn = sk->sk_user_data;
179 if (!conn) {
180 write_space = sk->sk_write_space;
181 goto out;
182 }
183
184 tc = conn->c_transport_data;
185 rdsdebug("write_space for tc %p\n", tc);
186 write_space = tc->t_orig_write_space;
187 rds_tcp_stats_inc(s_tcp_write_space_calls);
188
189 rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc));
190 tc->t_last_seen_una = rds_tcp_snd_una(tc);
191 rds_send_drop_acked(conn, rds_tcp_snd_una(tc), rds_tcp_is_acked);
192
193 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf)
194 queue_delayed_work(rds_wq, &conn->c_send_w, 0);
195
196 out:
197 read_unlock_bh(&sk->sk_callback_lock);
198
199 /*
200 * write_space is only called when data leaves tcp's send queue if
201 * SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put
202 * data in tcp's send queue because we use write_space to parse the
203 * sequence numbers and notice that rds messages have been fully
204 * received.
205 *
206 * tcp's write_space clears SOCK_NOSPACE if the send queue has more
207 * than a certain amount of space. So we need to set it again *after*
208 * we call tcp's write_space or else we might only get called on the
209 * first of a series of incoming tcp acks.
210 */
211 write_space(sk);
212
213 if (sk->sk_socket)
214 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
215 }
216