2 * Copyright (c) 2006 Oracle. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
40 static void rds_tcp_cork(struct socket
*sock
, int val
)
46 sock
->ops
->setsockopt(sock
, SOL_TCP
, TCP_CORK
, (char __user
*)&val
,
51 void rds_tcp_xmit_prepare(struct rds_connection
*conn
)
53 struct rds_tcp_connection
*tc
= conn
->c_transport_data
;
55 rds_tcp_cork(tc
->t_sock
, 1);
58 void rds_tcp_xmit_complete(struct rds_connection
*conn
)
60 struct rds_tcp_connection
*tc
= conn
->c_transport_data
;
62 rds_tcp_cork(tc
->t_sock
, 0);
65 /* the core send_sem serializes this with other xmit and shutdown */
66 static int rds_tcp_sendmsg(struct socket
*sock
, void *data
, unsigned int len
)
73 .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
,
76 return kernel_sendmsg(sock
, &msg
, &vec
, 1, vec
.iov_len
);
79 /* the core send_sem serializes this with other xmit and shutdown */
80 int rds_tcp_xmit(struct rds_connection
*conn
, struct rds_message
*rm
,
81 unsigned int hdr_off
, unsigned int sg
, unsigned int off
)
83 struct rds_tcp_connection
*tc
= conn
->c_transport_data
;
89 * m_ack_seq is set to the sequence number of the last byte of
90 * header and data. see rds_tcp_is_acked().
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_atomic();
97 set_bit(RDS_MSG_HAS_ACK_SEQ
, &rm
->m_flags
);
98 tc
->t_last_expected_una
= rm
->m_ack_seq
+ 1;
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
);
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
);
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
);
115 if (hdr_off
+ done
!= sizeof(struct rds_header
))
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
,
133 if (off
== rm
->data
.op_sg
[sg
].length
) {
141 /* write_space will hit after EAGAIN, all else fatal */
142 if (ret
== -EAGAIN
) {
143 rds_tcp_stats_inc(s_tcp_sndbuf_full
);
146 printk(KERN_WARNING
"RDS/tcp: send to %pI4 "
147 "returned %d, disconnecting and reconnecting\n",
148 &conn
->c_faddr
, ret
);
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.
164 static int rds_tcp_is_acked(struct rds_message
*rm
, uint64_t ack
)
166 if (!test_bit(RDS_MSG_HAS_ACK_SEQ
, &rm
->m_flags
))
168 return (__s32
)((u32
)rm
->m_ack_seq
- (u32
)ack
) < 0;
171 void rds_tcp_write_space(struct sock
*sk
)
173 void (*write_space
)(struct sock
*sk
);
174 struct rds_connection
*conn
;
175 struct rds_tcp_connection
*tc
;
177 read_lock(&sk
->sk_callback_lock
);
178 conn
= sk
->sk_user_data
;
180 write_space
= sk
->sk_write_space
;
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
);
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
);
193 if ((atomic_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
)
194 queue_delayed_work(rds_wq
, &conn
->c_send_w
, 0);
197 read_unlock(&sk
->sk_callback_lock
);
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
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.
214 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);