xref: /linux/net/rds/threads.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
1 /*
2  * Copyright (c) 2006, 2018 Oracle and/or its affiliates. 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/random.h>
35 #include <linux/export.h>
36 
37 #include "rds.h"
38 
39 /*
40  * All of connection management is simplified by serializing it through
41  * work queues that execute in a connection managing thread.
42  *
43  * TCP wants to send acks through sendpage() in response to data_ready(),
44  * but it needs a process context to do so.
45  *
46  * The receive paths need to allocate but can't drop packets (!) so we have
47  * a thread around to block allocating if the receive fast path sees an
48  * allocation failure.
49  */
50 
51 /* Grand Unified Theory of connection life cycle:
52  * At any point in time, the connection can be in one of these states:
53  * DOWN, CONNECTING, UP, DISCONNECTING, ERROR
54  *
55  * The following transitions are possible:
56  *  ANY		  -> ERROR
57  *  UP		  -> DISCONNECTING
58  *  ERROR	  -> DISCONNECTING
59  *  DISCONNECTING -> DOWN
60  *  DOWN	  -> CONNECTING
61  *  CONNECTING	  -> UP
62  *
63  * Transition to state DISCONNECTING/DOWN:
64  *  -	Inside the shutdown worker; synchronizes with xmit path
65  *	through RDS_IN_XMIT, and with connection management callbacks
66  *	via c_cm_lock.
67  *
68  *	For receive callbacks, we rely on the underlying transport
69  *	(TCP, IB/RDMA) to provide the necessary synchronisation.
70  */
71 struct workqueue_struct *rds_wq;
72 EXPORT_SYMBOL_GPL(rds_wq);
73 
74 void rds_connect_path_complete(struct rds_conn_path *cp, int curr)
75 {
76 	if (!rds_conn_path_transition(cp, curr, RDS_CONN_UP)) {
77 		printk(KERN_WARNING "%s: Cannot transition to state UP, "
78 				"current state is %d\n",
79 				__func__,
80 				atomic_read(&cp->cp_state));
81 		rds_conn_path_drop(cp, false);
82 		return;
83 	}
84 
85 	rdsdebug("conn %p for %pI6c to %pI6c complete\n",
86 		 cp->cp_conn, &cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr);
87 
88 	cp->cp_reconnect_jiffies = 0;
89 	set_bit(0, &cp->cp_conn->c_map_queued);
90 	rcu_read_lock();
91 	if (!rds_destroy_pending(cp->cp_conn)) {
92 		queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
93 		queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
94 	}
95 	rcu_read_unlock();
96 	cp->cp_conn->c_proposed_version = RDS_PROTOCOL_VERSION;
97 }
98 EXPORT_SYMBOL_GPL(rds_connect_path_complete);
99 
100 void rds_connect_complete(struct rds_connection *conn)
101 {
102 	rds_connect_path_complete(&conn->c_path[0], RDS_CONN_CONNECTING);
103 }
104 EXPORT_SYMBOL_GPL(rds_connect_complete);
105 
106 /*
107  * This random exponential backoff is relied on to eventually resolve racing
108  * connects.
109  *
110  * If connect attempts race then both parties drop both connections and come
111  * here to wait for a random amount of time before trying again.  Eventually
112  * the backoff range will be so much greater than the time it takes to
113  * establish a connection that one of the pair will establish the connection
114  * before the other's random delay fires.
115  *
116  * Connection attempts that arrive while a connection is already established
117  * are also considered to be racing connects.  This lets a connection from
118  * a rebooted machine replace an existing stale connection before the transport
119  * notices that the connection has failed.
120  *
121  * We should *always* start with a random backoff; otherwise a broken connection
122  * will always take several iterations to be re-established.
123  */
124 void rds_queue_reconnect(struct rds_conn_path *cp)
125 {
126 	unsigned long rand;
127 	struct rds_connection *conn = cp->cp_conn;
128 
129 	rdsdebug("conn %p for %pI6c to %pI6c reconnect jiffies %lu\n",
130 		 conn, &conn->c_laddr, &conn->c_faddr,
131 		 cp->cp_reconnect_jiffies);
132 
133 	/* let peer with smaller addr initiate reconnect, to avoid duels */
134 	if (conn->c_trans->t_type == RDS_TRANS_TCP &&
135 	    rds_addr_cmp(&conn->c_laddr, &conn->c_faddr) >= 0)
136 		return;
137 
138 	set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
139 	if (cp->cp_reconnect_jiffies == 0) {
140 		cp->cp_reconnect_jiffies = rds_sysctl_reconnect_min_jiffies;
141 		rcu_read_lock();
142 		if (!rds_destroy_pending(cp->cp_conn))
143 			queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
144 		rcu_read_unlock();
145 		return;
146 	}
147 
148 	get_random_bytes(&rand, sizeof(rand));
149 	rdsdebug("%lu delay %lu ceil conn %p for %pI6c -> %pI6c\n",
150 		 rand % cp->cp_reconnect_jiffies, cp->cp_reconnect_jiffies,
151 		 conn, &conn->c_laddr, &conn->c_faddr);
152 	rcu_read_lock();
153 	if (!rds_destroy_pending(cp->cp_conn))
154 		queue_delayed_work(rds_wq, &cp->cp_conn_w,
155 				   rand % cp->cp_reconnect_jiffies);
156 	rcu_read_unlock();
157 
158 	cp->cp_reconnect_jiffies = min(cp->cp_reconnect_jiffies * 2,
159 					rds_sysctl_reconnect_max_jiffies);
160 }
161 
162 void rds_connect_worker(struct work_struct *work)
163 {
164 	struct rds_conn_path *cp = container_of(work,
165 						struct rds_conn_path,
166 						cp_conn_w.work);
167 	struct rds_connection *conn = cp->cp_conn;
168 	int ret;
169 
170 	if (cp->cp_index > 0 &&
171 	    rds_addr_cmp(&cp->cp_conn->c_laddr, &cp->cp_conn->c_faddr) >= 0)
172 		return;
173 	clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
174 	ret = rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_CONNECTING);
175 	if (ret) {
176 		ret = conn->c_trans->conn_path_connect(cp);
177 		rdsdebug("conn %p for %pI6c to %pI6c dispatched, ret %d\n",
178 			 conn, &conn->c_laddr, &conn->c_faddr, ret);
179 
180 		if (ret) {
181 			if (rds_conn_path_transition(cp,
182 						     RDS_CONN_CONNECTING,
183 						     RDS_CONN_DOWN))
184 				rds_queue_reconnect(cp);
185 			else
186 				rds_conn_path_error(cp, "connect failed\n");
187 		}
188 	}
189 }
190 
191 void rds_send_worker(struct work_struct *work)
192 {
193 	struct rds_conn_path *cp = container_of(work,
194 						struct rds_conn_path,
195 						cp_send_w.work);
196 	int ret;
197 
198 	if (rds_conn_path_state(cp) == RDS_CONN_UP) {
199 		clear_bit(RDS_LL_SEND_FULL, &cp->cp_flags);
200 		ret = rds_send_xmit(cp);
201 		cond_resched();
202 		rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
203 		switch (ret) {
204 		case -EAGAIN:
205 			rds_stats_inc(s_send_immediate_retry);
206 			queue_delayed_work(rds_wq, &cp->cp_send_w, 0);
207 			break;
208 		case -ENOMEM:
209 			rds_stats_inc(s_send_delayed_retry);
210 			queue_delayed_work(rds_wq, &cp->cp_send_w, 2);
211 		default:
212 			break;
213 		}
214 	}
215 }
216 
217 void rds_recv_worker(struct work_struct *work)
218 {
219 	struct rds_conn_path *cp = container_of(work,
220 						struct rds_conn_path,
221 						cp_recv_w.work);
222 	int ret;
223 
224 	if (rds_conn_path_state(cp) == RDS_CONN_UP) {
225 		ret = cp->cp_conn->c_trans->recv_path(cp);
226 		rdsdebug("conn %p ret %d\n", cp->cp_conn, ret);
227 		switch (ret) {
228 		case -EAGAIN:
229 			rds_stats_inc(s_recv_immediate_retry);
230 			queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
231 			break;
232 		case -ENOMEM:
233 			rds_stats_inc(s_recv_delayed_retry);
234 			queue_delayed_work(rds_wq, &cp->cp_recv_w, 2);
235 		default:
236 			break;
237 		}
238 	}
239 }
240 
241 void rds_shutdown_worker(struct work_struct *work)
242 {
243 	struct rds_conn_path *cp = container_of(work,
244 						struct rds_conn_path,
245 						cp_down_w);
246 
247 	rds_conn_shutdown(cp);
248 }
249 
250 void rds_threads_exit(void)
251 {
252 	destroy_workqueue(rds_wq);
253 }
254 
255 int rds_threads_init(void)
256 {
257 	rds_wq = create_singlethread_workqueue("krdsd");
258 	if (!rds_wq)
259 		return -ENOMEM;
260 
261 	return 0;
262 }
263 
264 /* Compare two IPv6 addresses.  Return 0 if the two addresses are equal.
265  * Return 1 if the first is greater.  Return -1 if the second is greater.
266  */
267 int rds_addr_cmp(const struct in6_addr *addr1,
268 		 const struct in6_addr *addr2)
269 {
270 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
271 	const __be64 *a1, *a2;
272 	u64 x, y;
273 
274 	a1 = (__be64 *)addr1;
275 	a2 = (__be64 *)addr2;
276 
277 	if (*a1 != *a2) {
278 		if (be64_to_cpu(*a1) < be64_to_cpu(*a2))
279 			return -1;
280 		else
281 			return 1;
282 	} else {
283 		x = be64_to_cpu(*++a1);
284 		y = be64_to_cpu(*++a2);
285 		if (x < y)
286 			return -1;
287 		else if (x > y)
288 			return 1;
289 		else
290 			return 0;
291 	}
292 #else
293 	u32 a, b;
294 	int i;
295 
296 	for (i = 0; i < 4; i++) {
297 		if (addr1->s6_addr32[i] != addr2->s6_addr32[i]) {
298 			a = ntohl(addr1->s6_addr32[i]);
299 			b = ntohl(addr2->s6_addr32[i]);
300 			if (a < b)
301 				return -1;
302 			else if (a > b)
303 				return 1;
304 		}
305 	}
306 	return 0;
307 #endif
308 }
309 EXPORT_SYMBOL_GPL(rds_addr_cmp);
310