1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
|
/*
* 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/random.h>
#include <linux/export.h>
#include "rds.h"
/*
* All of connection management is simplified by serializing it through
* work queues that execute in a connection managing thread.
*
* TCP wants to send acks through sendpage() in response to data_ready(),
* but it needs a process context to do so.
*
* The receive paths need to allocate but can't drop packets (!) so we have
* a thread around to block allocating if the receive fast path sees an
* allocation failure.
*/
/* Grand Unified Theory of connection life cycle:
* At any point in time, the connection can be in one of these states:
* DOWN, CONNECTING, UP, DISCONNECTING, ERROR
*
* The following transitions are possible:
* ANY -> ERROR
* UP -> DISCONNECTING
* ERROR -> DISCONNECTING
* DISCONNECTING -> DOWN
* DOWN -> CONNECTING
* CONNECTING -> UP
*
* Transition to state DISCONNECTING/DOWN:
* - Inside the shutdown worker; synchronizes with xmit path
* through RDS_IN_XMIT, and with connection management callbacks
* via c_cm_lock.
*
* For receive callbacks, we rely on the underlying transport
* (TCP, IB/RDMA) to provide the necessary synchronisation.
*/
struct workqueue_struct *rds_wq;
EXPORT_SYMBOL_GPL(rds_wq);
void rds_connect_path_complete(struct rds_conn_path *cp, int curr)
{
if (!rds_conn_path_transition(cp, curr, RDS_CONN_UP)) {
printk(KERN_WARNING "%s: Cannot transition to state UP, "
"current state is %d\n",
__func__,
atomic_read(&cp->cp_state));
rds_conn_path_drop(cp);
return;
}
rdsdebug("conn %p for %pI4 to %pI4 complete\n",
cp->cp_conn, &cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr);
cp->cp_reconnect_jiffies = 0;
set_bit(0, &cp->cp_conn->c_map_queued);
queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
}
EXPORT_SYMBOL_GPL(rds_connect_path_complete);
void rds_connect_complete(struct rds_connection *conn)
{
rds_connect_path_complete(&conn->c_path[0], RDS_CONN_CONNECTING);
}
EXPORT_SYMBOL_GPL(rds_connect_complete);
/*
* This random exponential backoff is relied on to eventually resolve racing
* connects.
*
* If connect attempts race then both parties drop both connections and come
* here to wait for a random amount of time before trying again. Eventually
* the backoff range will be so much greater than the time it takes to
* establish a connection that one of the pair will establish the connection
* before the other's random delay fires.
*
* Connection attempts that arrive while a connection is already established
* are also considered to be racing connects. This lets a connection from
* a rebooted machine replace an existing stale connection before the transport
* notices that the connection has failed.
*
* We should *always* start with a random backoff; otherwise a broken connection
* will always take several iterations to be re-established.
*/
void rds_queue_reconnect(struct rds_conn_path *cp)
{
unsigned long rand;
struct rds_connection *conn = cp->cp_conn;
rdsdebug("conn %p for %pI4 to %pI4 reconnect jiffies %lu\n",
conn, &conn->c_laddr, &conn->c_faddr,
cp->cp_reconnect_jiffies);
/* let peer with smaller addr initiate reconnect, to avoid duels */
if (conn->c_trans->t_type == RDS_TRANS_TCP &&
conn->c_laddr > conn->c_faddr)
return;
set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
if (cp->cp_reconnect_jiffies == 0) {
cp->cp_reconnect_jiffies = rds_sysctl_reconnect_min_jiffies;
queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
return;
}
get_random_bytes(&rand, sizeof(rand));
rdsdebug("%lu delay %lu ceil conn %p for %pI4 -> %pI4\n",
rand % cp->cp_reconnect_jiffies, cp->cp_reconnect_jiffies,
conn, &conn->c_laddr, &conn->c_faddr);
queue_delayed_work(rds_wq, &cp->cp_conn_w,
rand % cp->cp_reconnect_jiffies);
cp->cp_reconnect_jiffies = min(cp->cp_reconnect_jiffies * 2,
rds_sysctl_reconnect_max_jiffies);
}
void rds_connect_worker(struct work_struct *work)
{
struct rds_conn_path *cp = container_of(work,
struct rds_conn_path,
cp_conn_w.work);
struct rds_connection *conn = cp->cp_conn;
int ret;
if (cp->cp_index > 1 && cp->cp_conn->c_laddr > cp->cp_conn->c_faddr)
return;
clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
ret = rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_CONNECTING);
if (ret) {
ret = conn->c_trans->conn_path_connect(cp);
rdsdebug("conn %p for %pI4 to %pI4 dispatched, ret %d\n",
conn, &conn->c_laddr, &conn->c_faddr, ret);
if (ret) {
if (rds_conn_path_transition(cp,
RDS_CONN_CONNECTING,
RDS_CONN_DOWN))
rds_queue_reconnect(cp);
else
rds_conn_path_error(cp, "connect failed\n");
}
}
}
void rds_send_worker(struct work_struct *work)
{
struct rds_conn_path *cp = container_of(work,
struct rds_conn_path,
cp_send_w.work);
int ret;
if (rds_conn_path_state(cp) == RDS_CONN_UP) {
clear_bit(RDS_LL_SEND_FULL, &cp->cp_flags);
ret = rds_send_xmit(cp);
cond_resched();
rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
switch (ret) {
case -EAGAIN:
rds_stats_inc(s_send_immediate_retry);
queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
break;
case -ENOMEM:
rds_stats_inc(s_send_delayed_retry);
queue_delayed_work(rds_wq, &cp->cp_send_w, 2);
default:
break;
}
}
}
void rds_recv_worker(struct work_struct *work)
{
struct rds_conn_path *cp = container_of(work,
struct rds_conn_path,
cp_recv_w.work);
int ret;
if (rds_conn_path_state(cp) == RDS_CONN_UP) {
ret = cp->cp_conn->c_trans->recv_path(cp);
rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
switch (ret) {
case -EAGAIN:
rds_stats_inc(s_recv_immediate_retry);
queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
break;
case -ENOMEM:
rds_stats_inc(s_recv_delayed_retry);
queue_delayed_work(rds_wq, &cp->cp_recv_w, 2);
default:
break;
}
}
}
void rds_shutdown_worker(struct work_struct *work)
{
struct rds_conn_path *cp = container_of(work,
struct rds_conn_path,
cp_down_w);
rds_conn_shutdown(cp);
}
void rds_threads_exit(void)
{
destroy_workqueue(rds_wq);
}
int rds_threads_init(void)
{
rds_wq = create_singlethread_workqueue("krdsd");
if (!rds_wq)
return -ENOMEM;
return 0;
}
|
