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>
37 #include "rds_single_path.h"
41 static void rds_tcp_cork(struct socket
*sock
, int val
)
43 kernel_setsockopt(sock
, SOL_TCP
, TCP_CORK
, (void *)&val
, sizeof(val
));
46 void rds_tcp_xmit_path_prepare(struct rds_conn_path
*cp
)
48 struct rds_tcp_connection
*tc
= cp
->cp_transport_data
;
50 rds_tcp_cork(tc
->t_sock
, 1);
53 void rds_tcp_xmit_path_complete(struct rds_conn_path
*cp
)
55 struct rds_tcp_connection
*tc
= cp
->cp_transport_data
;
57 rds_tcp_cork(tc
->t_sock
, 0);
60 /* the core send_sem serializes this with other xmit and shutdown */
61 static int rds_tcp_sendmsg(struct socket
*sock
, void *data
, unsigned int len
)
68 .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
,
71 return kernel_sendmsg(sock
, &msg
, &vec
, 1, vec
.iov_len
);
74 /* the core send_sem serializes this with other xmit and shutdown */
75 int rds_tcp_xmit(struct rds_connection
*conn
, struct rds_message
*rm
,
76 unsigned int hdr_off
, unsigned int sg
, unsigned int off
)
78 struct rds_conn_path
*cp
= rm
->m_inc
.i_conn_path
;
79 struct rds_tcp_connection
*tc
= cp
->cp_transport_data
;
86 * m_ack_seq is set to the sequence number of the last byte of
87 * header and data. see rds_tcp_is_acked().
89 tc
->t_last_sent_nxt
= rds_tcp_write_seq(tc
);
90 rm
->m_ack_seq
= tc
->t_last_sent_nxt
+
91 sizeof(struct rds_header
) +
92 be32_to_cpu(rm
->m_inc
.i_hdr
.h_len
) - 1;
93 smp_mb__before_atomic();
94 set_bit(RDS_MSG_HAS_ACK_SEQ
, &rm
->m_flags
);
95 tc
->t_last_expected_una
= rm
->m_ack_seq
+ 1;
97 if (test_bit(RDS_MSG_RETRANSMITTED
, &rm
->m_flags
))
98 rm
->m_inc
.i_hdr
.h_flags
|= RDS_FLAG_RETRANSMITTED
;
100 rdsdebug("rm %p tcp nxt %u ack_seq %llu\n",
101 rm
, rds_tcp_write_seq(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 more
= rm
->data
.op_nents
> 1 ? (MSG_MORE
| MSG_SENDPAGE_NOTLAST
) : 0;
120 while (sg
< rm
->data
.op_nents
) {
121 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| more
;
123 ret
= tc
->t_sock
->ops
->sendpage(tc
->t_sock
,
124 sg_page(&rm
->data
.op_sg
[sg
]),
125 rm
->data
.op_sg
[sg
].offset
+ off
,
126 rm
->data
.op_sg
[sg
].length
- off
,
128 rdsdebug("tcp sendpage %p:%u:%u ret %d\n", (void *)sg_page(&rm
->data
.op_sg
[sg
]),
129 rm
->data
.op_sg
[sg
].offset
+ off
, rm
->data
.op_sg
[sg
].length
- off
,
136 if (off
== rm
->data
.op_sg
[sg
].length
) {
140 if (sg
== rm
->data
.op_nents
- 1)
146 /* write_space will hit after EAGAIN, all else fatal */
147 if (ret
== -EAGAIN
) {
148 rds_tcp_stats_inc(s_tcp_sndbuf_full
);
151 /* No need to disconnect/reconnect if path_drop
152 * has already been triggered, because, e.g., of
155 if (rds_conn_path_up(cp
)) {
156 pr_warn("RDS/tcp: send to %pI4 on cp [%d]"
158 "disconnecting and reconnecting\n",
159 &conn
->c_faddr
, cp
->cp_index
, ret
);
160 rds_conn_path_drop(cp
, false);
170 * rm->m_ack_seq is set to the tcp sequence number that corresponds to the
171 * last byte of the message, including the header. This means that the
172 * entire message has been received if rm->m_ack_seq is "before" the next
173 * unacked byte of the TCP sequence space. We have to do very careful
174 * wrapping 32bit comparisons here.
176 static int rds_tcp_is_acked(struct rds_message
*rm
, uint64_t ack
)
178 if (!test_bit(RDS_MSG_HAS_ACK_SEQ
, &rm
->m_flags
))
180 return (__s32
)((u32
)rm
->m_ack_seq
- (u32
)ack
) < 0;
183 void rds_tcp_write_space(struct sock
*sk
)
185 void (*write_space
)(struct sock
*sk
);
186 struct rds_conn_path
*cp
;
187 struct rds_tcp_connection
*tc
;
189 read_lock_bh(&sk
->sk_callback_lock
);
190 cp
= sk
->sk_user_data
;
192 write_space
= sk
->sk_write_space
;
196 tc
= cp
->cp_transport_data
;
197 rdsdebug("write_space for tc %p\n", tc
);
198 write_space
= tc
->t_orig_write_space
;
199 rds_tcp_stats_inc(s_tcp_write_space_calls
);
201 rdsdebug("tcp una %u\n", rds_tcp_snd_una(tc
));
202 tc
->t_last_seen_una
= rds_tcp_snd_una(tc
);
203 rds_send_path_drop_acked(cp
, rds_tcp_snd_una(tc
), rds_tcp_is_acked
);
206 if ((refcount_read(&sk
->sk_wmem_alloc
) << 1) <= sk
->sk_sndbuf
&&
207 !rds_destroy_pending(cp
->cp_conn
))
208 queue_delayed_work(rds_wq
, &cp
->cp_send_w
, 0);
212 read_unlock_bh(&sk
->sk_callback_lock
);
215 * write_space is only called when data leaves tcp's send queue if
216 * SOCK_NOSPACE is set. We set SOCK_NOSPACE every time we put
217 * data in tcp's send queue because we use write_space to parse the
218 * sequence numbers and notice that rds messages have been fully
221 * tcp's write_space clears SOCK_NOSPACE if the send queue has more
222 * than a certain amount of space. So we need to set it again *after*
223 * we call tcp's write_space or else we might only get called on the
224 * first of a series of incoming tcp acks.
229 set_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);