xref: /openbmc/linux/net/rds/connection.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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/list.h>
35 #include <net/inet_hashtables.h>
36 
37 #include "rds.h"
38 #include "loop.h"
39 #include "rdma.h"
40 
41 #define RDS_CONNECTION_HASH_BITS 12
42 #define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS)
43 #define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1)
44 
45 /* converting this to RCU is a chore for another day.. */
46 static DEFINE_SPINLOCK(rds_conn_lock);
47 static unsigned long rds_conn_count;
48 static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
49 static struct kmem_cache *rds_conn_slab;
50 
51 static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr)
52 {
53 	/* Pass NULL, don't need struct net for hash */
54 	unsigned long hash = inet_ehashfn(NULL,
55 					  be32_to_cpu(laddr), 0,
56 					  be32_to_cpu(faddr), 0);
57 	return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
58 }
59 
60 #define rds_conn_info_set(var, test, suffix) do {		\
61 	if (test)						\
62 		var |= RDS_INFO_CONNECTION_FLAG_##suffix;	\
63 } while (0)
64 
65 static inline int rds_conn_is_sending(struct rds_connection *conn)
66 {
67 	int ret = 0;
68 
69 	if (!mutex_trylock(&conn->c_send_lock))
70 		ret = 1;
71 	else
72 		mutex_unlock(&conn->c_send_lock);
73 
74 	return ret;
75 }
76 
77 static struct rds_connection *rds_conn_lookup(struct hlist_head *head,
78 					      __be32 laddr, __be32 faddr,
79 					      struct rds_transport *trans)
80 {
81 	struct rds_connection *conn, *ret = NULL;
82 	struct hlist_node *pos;
83 
84 	hlist_for_each_entry(conn, pos, head, c_hash_node) {
85 		if (conn->c_faddr == faddr && conn->c_laddr == laddr &&
86 				conn->c_trans == trans) {
87 			ret = conn;
88 			break;
89 		}
90 	}
91 	rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret,
92 		 &laddr, &faddr);
93 	return ret;
94 }
95 
96 /*
97  * This is called by transports as they're bringing down a connection.
98  * It clears partial message state so that the transport can start sending
99  * and receiving over this connection again in the future.  It is up to
100  * the transport to have serialized this call with its send and recv.
101  */
102 void rds_conn_reset(struct rds_connection *conn)
103 {
104 	rdsdebug("connection %pI4 to %pI4 reset\n",
105 	  &conn->c_laddr, &conn->c_faddr);
106 
107 	rds_stats_inc(s_conn_reset);
108 	rds_send_reset(conn);
109 	conn->c_flags = 0;
110 
111 	/* Do not clear next_rx_seq here, else we cannot distinguish
112 	 * retransmitted packets from new packets, and will hand all
113 	 * of them to the application. That is not consistent with the
114 	 * reliability guarantees of RDS. */
115 }
116 
117 /*
118  * There is only every one 'conn' for a given pair of addresses in the
119  * system at a time.  They contain messages to be retransmitted and so
120  * span the lifetime of the actual underlying transport connections.
121  *
122  * For now they are not garbage collected once they're created.  They
123  * are torn down as the module is removed, if ever.
124  */
125 static struct rds_connection *__rds_conn_create(__be32 laddr, __be32 faddr,
126 				       struct rds_transport *trans, gfp_t gfp,
127 				       int is_outgoing)
128 {
129 	struct rds_connection *conn, *tmp, *parent = NULL;
130 	struct hlist_head *head = rds_conn_bucket(laddr, faddr);
131 	unsigned long flags;
132 	int ret;
133 
134 	spin_lock_irqsave(&rds_conn_lock, flags);
135 	conn = rds_conn_lookup(head, laddr, faddr, trans);
136 	if (conn
137 	 && conn->c_loopback
138 	 && conn->c_trans != &rds_loop_transport
139 	 && !is_outgoing) {
140 		/* This is a looped back IB connection, and we're
141 		 * called by the code handling the incoming connect.
142 		 * We need a second connection object into which we
143 		 * can stick the other QP. */
144 		parent = conn;
145 		conn = parent->c_passive;
146 	}
147 	spin_unlock_irqrestore(&rds_conn_lock, flags);
148 	if (conn)
149 		goto out;
150 
151 	conn = kmem_cache_alloc(rds_conn_slab, gfp);
152 	if (conn == NULL) {
153 		conn = ERR_PTR(-ENOMEM);
154 		goto out;
155 	}
156 
157 	memset(conn, 0, sizeof(*conn));
158 
159 	INIT_HLIST_NODE(&conn->c_hash_node);
160 	conn->c_version = RDS_PROTOCOL_3_0;
161 	conn->c_laddr = laddr;
162 	conn->c_faddr = faddr;
163 	spin_lock_init(&conn->c_lock);
164 	conn->c_next_tx_seq = 1;
165 
166 	mutex_init(&conn->c_send_lock);
167 	INIT_LIST_HEAD(&conn->c_send_queue);
168 	INIT_LIST_HEAD(&conn->c_retrans);
169 
170 	ret = rds_cong_get_maps(conn);
171 	if (ret) {
172 		kmem_cache_free(rds_conn_slab, conn);
173 		conn = ERR_PTR(ret);
174 		goto out;
175 	}
176 
177 	/*
178 	 * This is where a connection becomes loopback.  If *any* RDS sockets
179 	 * can bind to the destination address then we'd rather the messages
180 	 * flow through loopback rather than either transport.
181 	 */
182 	if (rds_trans_get_preferred(faddr)) {
183 		conn->c_loopback = 1;
184 		if (is_outgoing && trans->t_prefer_loopback) {
185 			/* "outgoing" connection - and the transport
186 			 * says it wants the connection handled by the
187 			 * loopback transport. This is what TCP does.
188 			 */
189 			trans = &rds_loop_transport;
190 		}
191 	}
192 
193 	conn->c_trans = trans;
194 
195 	ret = trans->conn_alloc(conn, gfp);
196 	if (ret) {
197 		kmem_cache_free(rds_conn_slab, conn);
198 		conn = ERR_PTR(ret);
199 		goto out;
200 	}
201 
202 	atomic_set(&conn->c_state, RDS_CONN_DOWN);
203 	conn->c_reconnect_jiffies = 0;
204 	INIT_DELAYED_WORK(&conn->c_send_w, rds_send_worker);
205 	INIT_DELAYED_WORK(&conn->c_recv_w, rds_recv_worker);
206 	INIT_DELAYED_WORK(&conn->c_conn_w, rds_connect_worker);
207 	INIT_WORK(&conn->c_down_w, rds_shutdown_worker);
208 	mutex_init(&conn->c_cm_lock);
209 	conn->c_flags = 0;
210 
211 	rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n",
212 	  conn, &laddr, &faddr,
213 	  trans->t_name ? trans->t_name : "[unknown]",
214 	  is_outgoing ? "(outgoing)" : "");
215 
216 	spin_lock_irqsave(&rds_conn_lock, flags);
217 	if (parent == NULL) {
218 		tmp = rds_conn_lookup(head, laddr, faddr, trans);
219 		if (tmp == NULL)
220 			hlist_add_head(&conn->c_hash_node, head);
221 	} else {
222 		tmp = parent->c_passive;
223 		if (!tmp)
224 			parent->c_passive = conn;
225 	}
226 
227 	if (tmp) {
228 		trans->conn_free(conn->c_transport_data);
229 		kmem_cache_free(rds_conn_slab, conn);
230 		conn = tmp;
231 	} else {
232 		rds_cong_add_conn(conn);
233 		rds_conn_count++;
234 	}
235 
236 	spin_unlock_irqrestore(&rds_conn_lock, flags);
237 
238 out:
239 	return conn;
240 }
241 
242 struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
243 				       struct rds_transport *trans, gfp_t gfp)
244 {
245 	return __rds_conn_create(laddr, faddr, trans, gfp, 0);
246 }
247 
248 struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
249 				       struct rds_transport *trans, gfp_t gfp)
250 {
251 	return __rds_conn_create(laddr, faddr, trans, gfp, 1);
252 }
253 
254 void rds_conn_destroy(struct rds_connection *conn)
255 {
256 	struct rds_message *rm, *rtmp;
257 
258 	rdsdebug("freeing conn %p for %pI4 -> "
259 		 "%pI4\n", conn, &conn->c_laddr,
260 		 &conn->c_faddr);
261 
262 	hlist_del_init(&conn->c_hash_node);
263 
264 	/* wait for the rds thread to shut it down */
265 	atomic_set(&conn->c_state, RDS_CONN_ERROR);
266 	cancel_delayed_work(&conn->c_conn_w);
267 	queue_work(rds_wq, &conn->c_down_w);
268 	flush_workqueue(rds_wq);
269 
270 	/* tear down queued messages */
271 	list_for_each_entry_safe(rm, rtmp,
272 				 &conn->c_send_queue,
273 				 m_conn_item) {
274 		list_del_init(&rm->m_conn_item);
275 		BUG_ON(!list_empty(&rm->m_sock_item));
276 		rds_message_put(rm);
277 	}
278 	if (conn->c_xmit_rm)
279 		rds_message_put(conn->c_xmit_rm);
280 
281 	conn->c_trans->conn_free(conn->c_transport_data);
282 
283 	/*
284 	 * The congestion maps aren't freed up here.  They're
285 	 * freed by rds_cong_exit() after all the connections
286 	 * have been freed.
287 	 */
288 	rds_cong_remove_conn(conn);
289 
290 	BUG_ON(!list_empty(&conn->c_retrans));
291 	kmem_cache_free(rds_conn_slab, conn);
292 
293 	rds_conn_count--;
294 }
295 
296 static void rds_conn_message_info(struct socket *sock, unsigned int len,
297 				  struct rds_info_iterator *iter,
298 				  struct rds_info_lengths *lens,
299 				  int want_send)
300 {
301 	struct hlist_head *head;
302 	struct hlist_node *pos;
303 	struct list_head *list;
304 	struct rds_connection *conn;
305 	struct rds_message *rm;
306 	unsigned long flags;
307 	unsigned int total = 0;
308 	size_t i;
309 
310 	len /= sizeof(struct rds_info_message);
311 
312 	spin_lock_irqsave(&rds_conn_lock, flags);
313 
314 	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
315 	     i++, head++) {
316 		hlist_for_each_entry(conn, pos, head, c_hash_node) {
317 			if (want_send)
318 				list = &conn->c_send_queue;
319 			else
320 				list = &conn->c_retrans;
321 
322 			spin_lock(&conn->c_lock);
323 
324 			/* XXX too lazy to maintain counts.. */
325 			list_for_each_entry(rm, list, m_conn_item) {
326 				total++;
327 				if (total <= len)
328 					rds_inc_info_copy(&rm->m_inc, iter,
329 							  conn->c_laddr,
330 							  conn->c_faddr, 0);
331 			}
332 
333 			spin_unlock(&conn->c_lock);
334 		}
335 	}
336 
337 	spin_unlock_irqrestore(&rds_conn_lock, flags);
338 
339 	lens->nr = total;
340 	lens->each = sizeof(struct rds_info_message);
341 }
342 
343 static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
344 				       struct rds_info_iterator *iter,
345 				       struct rds_info_lengths *lens)
346 {
347 	rds_conn_message_info(sock, len, iter, lens, 1);
348 }
349 
350 static void rds_conn_message_info_retrans(struct socket *sock,
351 					  unsigned int len,
352 					  struct rds_info_iterator *iter,
353 					  struct rds_info_lengths *lens)
354 {
355 	rds_conn_message_info(sock, len, iter, lens, 0);
356 }
357 
358 void rds_for_each_conn_info(struct socket *sock, unsigned int len,
359 			  struct rds_info_iterator *iter,
360 			  struct rds_info_lengths *lens,
361 			  int (*visitor)(struct rds_connection *, void *),
362 			  size_t item_len)
363 {
364 	uint64_t buffer[(item_len + 7) / 8];
365 	struct hlist_head *head;
366 	struct hlist_node *pos;
367 	struct hlist_node *tmp;
368 	struct rds_connection *conn;
369 	unsigned long flags;
370 	size_t i;
371 
372 	spin_lock_irqsave(&rds_conn_lock, flags);
373 
374 	lens->nr = 0;
375 	lens->each = item_len;
376 
377 	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
378 	     i++, head++) {
379 		hlist_for_each_entry_safe(conn, pos, tmp, head, c_hash_node) {
380 
381 			/* XXX no c_lock usage.. */
382 			if (!visitor(conn, buffer))
383 				continue;
384 
385 			/* We copy as much as we can fit in the buffer,
386 			 * but we count all items so that the caller
387 			 * can resize the buffer. */
388 			if (len >= item_len) {
389 				rds_info_copy(iter, buffer, item_len);
390 				len -= item_len;
391 			}
392 			lens->nr++;
393 		}
394 	}
395 
396 	spin_unlock_irqrestore(&rds_conn_lock, flags);
397 }
398 
399 static int rds_conn_info_visitor(struct rds_connection *conn,
400 				  void *buffer)
401 {
402 	struct rds_info_connection *cinfo = buffer;
403 
404 	cinfo->next_tx_seq = conn->c_next_tx_seq;
405 	cinfo->next_rx_seq = conn->c_next_rx_seq;
406 	cinfo->laddr = conn->c_laddr;
407 	cinfo->faddr = conn->c_faddr;
408 	strncpy(cinfo->transport, conn->c_trans->t_name,
409 		sizeof(cinfo->transport));
410 	cinfo->flags = 0;
411 
412 	rds_conn_info_set(cinfo->flags,
413 			  rds_conn_is_sending(conn), SENDING);
414 	/* XXX Future: return the state rather than these funky bits */
415 	rds_conn_info_set(cinfo->flags,
416 			  atomic_read(&conn->c_state) == RDS_CONN_CONNECTING,
417 			  CONNECTING);
418 	rds_conn_info_set(cinfo->flags,
419 			  atomic_read(&conn->c_state) == RDS_CONN_UP,
420 			  CONNECTED);
421 	return 1;
422 }
423 
424 static void rds_conn_info(struct socket *sock, unsigned int len,
425 			  struct rds_info_iterator *iter,
426 			  struct rds_info_lengths *lens)
427 {
428 	rds_for_each_conn_info(sock, len, iter, lens,
429 				rds_conn_info_visitor,
430 				sizeof(struct rds_info_connection));
431 }
432 
433 int __init rds_conn_init(void)
434 {
435 	rds_conn_slab = kmem_cache_create("rds_connection",
436 					  sizeof(struct rds_connection),
437 					  0, 0, NULL);
438 	if (rds_conn_slab == NULL)
439 		return -ENOMEM;
440 
441 	rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
442 	rds_info_register_func(RDS_INFO_SEND_MESSAGES,
443 			       rds_conn_message_info_send);
444 	rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
445 			       rds_conn_message_info_retrans);
446 
447 	return 0;
448 }
449 
450 void rds_conn_exit(void)
451 {
452 	rds_loop_exit();
453 
454 	WARN_ON(!hlist_empty(rds_conn_hash));
455 
456 	kmem_cache_destroy(rds_conn_slab);
457 
458 	rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
459 	rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
460 				 rds_conn_message_info_send);
461 	rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
462 				 rds_conn_message_info_retrans);
463 }
464 
465 /*
466  * Force a disconnect
467  */
468 void rds_conn_drop(struct rds_connection *conn)
469 {
470 	atomic_set(&conn->c_state, RDS_CONN_ERROR);
471 	queue_work(rds_wq, &conn->c_down_w);
472 }
473 
474 /*
475  * An error occurred on the connection
476  */
477 void
478 __rds_conn_error(struct rds_connection *conn, const char *fmt, ...)
479 {
480 	va_list ap;
481 
482 	va_start(ap, fmt);
483 	vprintk(fmt, ap);
484 	va_end(ap);
485 
486 	rds_conn_drop(conn);
487 }
488