xref: /linux/net/rds/connection.c (revision 44f57d78)
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/list.h>
35 #include <linux/slab.h>
36 #include <linux/export.h>
37 #include <net/ipv6.h>
38 #include <net/inet6_hashtables.h>
39 #include <net/addrconf.h>
40 
41 #include "rds.h"
42 #include "loop.h"
43 
44 #define RDS_CONNECTION_HASH_BITS 12
45 #define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS)
46 #define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1)
47 
48 /* converting this to RCU is a chore for another day.. */
49 static DEFINE_SPINLOCK(rds_conn_lock);
50 static unsigned long rds_conn_count;
51 static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
52 static struct kmem_cache *rds_conn_slab;
53 
54 static struct hlist_head *rds_conn_bucket(const struct in6_addr *laddr,
55 					  const struct in6_addr *faddr)
56 {
57 	static u32 rds6_hash_secret __read_mostly;
58 	static u32 rds_hash_secret __read_mostly;
59 
60 	u32 lhash, fhash, hash;
61 
62 	net_get_random_once(&rds_hash_secret, sizeof(rds_hash_secret));
63 	net_get_random_once(&rds6_hash_secret, sizeof(rds6_hash_secret));
64 
65 	lhash = (__force u32)laddr->s6_addr32[3];
66 #if IS_ENABLED(CONFIG_IPV6)
67 	fhash = __ipv6_addr_jhash(faddr, rds6_hash_secret);
68 #else
69 	fhash = (__force u32)faddr->s6_addr32[3];
70 #endif
71 	hash = __inet_ehashfn(lhash, 0, fhash, 0, rds_hash_secret);
72 
73 	return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
74 }
75 
76 #define rds_conn_info_set(var, test, suffix) do {		\
77 	if (test)						\
78 		var |= RDS_INFO_CONNECTION_FLAG_##suffix;	\
79 } while (0)
80 
81 /* rcu read lock must be held or the connection spinlock */
82 static struct rds_connection *rds_conn_lookup(struct net *net,
83 					      struct hlist_head *head,
84 					      const struct in6_addr *laddr,
85 					      const struct in6_addr *faddr,
86 					      struct rds_transport *trans,
87 					      u8 tos, int dev_if)
88 {
89 	struct rds_connection *conn, *ret = NULL;
90 
91 	hlist_for_each_entry_rcu(conn, head, c_hash_node) {
92 		if (ipv6_addr_equal(&conn->c_faddr, faddr) &&
93 		    ipv6_addr_equal(&conn->c_laddr, laddr) &&
94 		    conn->c_trans == trans &&
95 		    conn->c_tos == tos &&
96 		    net == rds_conn_net(conn) &&
97 		    conn->c_dev_if == dev_if) {
98 			ret = conn;
99 			break;
100 		}
101 	}
102 	rdsdebug("returning conn %p for %pI6c -> %pI6c\n", ret,
103 		 laddr, faddr);
104 	return ret;
105 }
106 
107 /*
108  * This is called by transports as they're bringing down a connection.
109  * It clears partial message state so that the transport can start sending
110  * and receiving over this connection again in the future.  It is up to
111  * the transport to have serialized this call with its send and recv.
112  */
113 static void rds_conn_path_reset(struct rds_conn_path *cp)
114 {
115 	struct rds_connection *conn = cp->cp_conn;
116 
117 	rdsdebug("connection %pI6c to %pI6c reset\n",
118 		 &conn->c_laddr, &conn->c_faddr);
119 
120 	rds_stats_inc(s_conn_reset);
121 	rds_send_path_reset(cp);
122 	cp->cp_flags = 0;
123 
124 	/* Do not clear next_rx_seq here, else we cannot distinguish
125 	 * retransmitted packets from new packets, and will hand all
126 	 * of them to the application. That is not consistent with the
127 	 * reliability guarantees of RDS. */
128 }
129 
130 static void __rds_conn_path_init(struct rds_connection *conn,
131 				 struct rds_conn_path *cp, bool is_outgoing)
132 {
133 	spin_lock_init(&cp->cp_lock);
134 	cp->cp_next_tx_seq = 1;
135 	init_waitqueue_head(&cp->cp_waitq);
136 	INIT_LIST_HEAD(&cp->cp_send_queue);
137 	INIT_LIST_HEAD(&cp->cp_retrans);
138 
139 	cp->cp_conn = conn;
140 	atomic_set(&cp->cp_state, RDS_CONN_DOWN);
141 	cp->cp_send_gen = 0;
142 	cp->cp_reconnect_jiffies = 0;
143 	cp->cp_conn->c_proposed_version = RDS_PROTOCOL_VERSION;
144 	INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker);
145 	INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker);
146 	INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker);
147 	INIT_WORK(&cp->cp_down_w, rds_shutdown_worker);
148 	mutex_init(&cp->cp_cm_lock);
149 	cp->cp_flags = 0;
150 }
151 
152 /*
153  * There is only every one 'conn' for a given pair of addresses in the
154  * system at a time.  They contain messages to be retransmitted and so
155  * span the lifetime of the actual underlying transport connections.
156  *
157  * For now they are not garbage collected once they're created.  They
158  * are torn down as the module is removed, if ever.
159  */
160 static struct rds_connection *__rds_conn_create(struct net *net,
161 						const struct in6_addr *laddr,
162 						const struct in6_addr *faddr,
163 						struct rds_transport *trans,
164 						gfp_t gfp, u8 tos,
165 						int is_outgoing,
166 						int dev_if)
167 {
168 	struct rds_connection *conn, *parent = NULL;
169 	struct hlist_head *head = rds_conn_bucket(laddr, faddr);
170 	struct rds_transport *loop_trans;
171 	unsigned long flags;
172 	int ret, i;
173 	int npaths = (trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
174 
175 	rcu_read_lock();
176 	conn = rds_conn_lookup(net, head, laddr, faddr, trans, tos, dev_if);
177 	if (conn &&
178 	    conn->c_loopback &&
179 	    conn->c_trans != &rds_loop_transport &&
180 	    ipv6_addr_equal(laddr, faddr) &&
181 	    !is_outgoing) {
182 		/* This is a looped back IB connection, and we're
183 		 * called by the code handling the incoming connect.
184 		 * We need a second connection object into which we
185 		 * can stick the other QP. */
186 		parent = conn;
187 		conn = parent->c_passive;
188 	}
189 	rcu_read_unlock();
190 	if (conn)
191 		goto out;
192 
193 	conn = kmem_cache_zalloc(rds_conn_slab, gfp);
194 	if (!conn) {
195 		conn = ERR_PTR(-ENOMEM);
196 		goto out;
197 	}
198 	conn->c_path = kcalloc(npaths, sizeof(struct rds_conn_path), gfp);
199 	if (!conn->c_path) {
200 		kmem_cache_free(rds_conn_slab, conn);
201 		conn = ERR_PTR(-ENOMEM);
202 		goto out;
203 	}
204 
205 	INIT_HLIST_NODE(&conn->c_hash_node);
206 	conn->c_laddr = *laddr;
207 	conn->c_isv6 = !ipv6_addr_v4mapped(laddr);
208 	conn->c_faddr = *faddr;
209 	conn->c_dev_if = dev_if;
210 	conn->c_tos = tos;
211 
212 #if IS_ENABLED(CONFIG_IPV6)
213 	/* If the local address is link local, set c_bound_if to be the
214 	 * index used for this connection.  Otherwise, set it to 0 as
215 	 * the socket is not bound to an interface.  c_bound_if is used
216 	 * to look up a socket when a packet is received
217 	 */
218 	if (ipv6_addr_type(laddr) & IPV6_ADDR_LINKLOCAL)
219 		conn->c_bound_if = dev_if;
220 	else
221 #endif
222 		conn->c_bound_if = 0;
223 
224 	rds_conn_net_set(conn, net);
225 
226 	ret = rds_cong_get_maps(conn);
227 	if (ret) {
228 		kfree(conn->c_path);
229 		kmem_cache_free(rds_conn_slab, conn);
230 		conn = ERR_PTR(ret);
231 		goto out;
232 	}
233 
234 	/*
235 	 * This is where a connection becomes loopback.  If *any* RDS sockets
236 	 * can bind to the destination address then we'd rather the messages
237 	 * flow through loopback rather than either transport.
238 	 */
239 	loop_trans = rds_trans_get_preferred(net, faddr, conn->c_dev_if);
240 	if (loop_trans) {
241 		rds_trans_put(loop_trans);
242 		conn->c_loopback = 1;
243 		if (is_outgoing && trans->t_prefer_loopback) {
244 			/* "outgoing" connection - and the transport
245 			 * says it wants the connection handled by the
246 			 * loopback transport. This is what TCP does.
247 			 */
248 			trans = &rds_loop_transport;
249 		}
250 	}
251 
252 	conn->c_trans = trans;
253 
254 	init_waitqueue_head(&conn->c_hs_waitq);
255 	for (i = 0; i < npaths; i++) {
256 		__rds_conn_path_init(conn, &conn->c_path[i],
257 				     is_outgoing);
258 		conn->c_path[i].cp_index = i;
259 	}
260 	rcu_read_lock();
261 	if (rds_destroy_pending(conn))
262 		ret = -ENETDOWN;
263 	else
264 		ret = trans->conn_alloc(conn, GFP_ATOMIC);
265 	if (ret) {
266 		rcu_read_unlock();
267 		kfree(conn->c_path);
268 		kmem_cache_free(rds_conn_slab, conn);
269 		conn = ERR_PTR(ret);
270 		goto out;
271 	}
272 
273 	rdsdebug("allocated conn %p for %pI6c -> %pI6c over %s %s\n",
274 		 conn, laddr, faddr,
275 		 strnlen(trans->t_name, sizeof(trans->t_name)) ?
276 		 trans->t_name : "[unknown]", is_outgoing ? "(outgoing)" : "");
277 
278 	/*
279 	 * Since we ran without holding the conn lock, someone could
280 	 * have created the same conn (either normal or passive) in the
281 	 * interim. We check while holding the lock. If we won, we complete
282 	 * init and return our conn. If we lost, we rollback and return the
283 	 * other one.
284 	 */
285 	spin_lock_irqsave(&rds_conn_lock, flags);
286 	if (parent) {
287 		/* Creating passive conn */
288 		if (parent->c_passive) {
289 			trans->conn_free(conn->c_path[0].cp_transport_data);
290 			kfree(conn->c_path);
291 			kmem_cache_free(rds_conn_slab, conn);
292 			conn = parent->c_passive;
293 		} else {
294 			parent->c_passive = conn;
295 			rds_cong_add_conn(conn);
296 			rds_conn_count++;
297 		}
298 	} else {
299 		/* Creating normal conn */
300 		struct rds_connection *found;
301 
302 		found = rds_conn_lookup(net, head, laddr, faddr, trans,
303 					tos, dev_if);
304 		if (found) {
305 			struct rds_conn_path *cp;
306 			int i;
307 
308 			for (i = 0; i < npaths; i++) {
309 				cp = &conn->c_path[i];
310 				/* The ->conn_alloc invocation may have
311 				 * allocated resource for all paths, so all
312 				 * of them may have to be freed here.
313 				 */
314 				if (cp->cp_transport_data)
315 					trans->conn_free(cp->cp_transport_data);
316 			}
317 			kfree(conn->c_path);
318 			kmem_cache_free(rds_conn_slab, conn);
319 			conn = found;
320 		} else {
321 			conn->c_my_gen_num = rds_gen_num;
322 			conn->c_peer_gen_num = 0;
323 			hlist_add_head_rcu(&conn->c_hash_node, head);
324 			rds_cong_add_conn(conn);
325 			rds_conn_count++;
326 		}
327 	}
328 	spin_unlock_irqrestore(&rds_conn_lock, flags);
329 	rcu_read_unlock();
330 
331 out:
332 	return conn;
333 }
334 
335 struct rds_connection *rds_conn_create(struct net *net,
336 				       const struct in6_addr *laddr,
337 				       const struct in6_addr *faddr,
338 				       struct rds_transport *trans, u8 tos,
339 				       gfp_t gfp, int dev_if)
340 {
341 	return __rds_conn_create(net, laddr, faddr, trans, gfp, tos, 0, dev_if);
342 }
343 EXPORT_SYMBOL_GPL(rds_conn_create);
344 
345 struct rds_connection *rds_conn_create_outgoing(struct net *net,
346 						const struct in6_addr *laddr,
347 						const struct in6_addr *faddr,
348 						struct rds_transport *trans,
349 						u8 tos, gfp_t gfp, int dev_if)
350 {
351 	return __rds_conn_create(net, laddr, faddr, trans, gfp, tos, 1, dev_if);
352 }
353 EXPORT_SYMBOL_GPL(rds_conn_create_outgoing);
354 
355 void rds_conn_shutdown(struct rds_conn_path *cp)
356 {
357 	struct rds_connection *conn = cp->cp_conn;
358 
359 	/* shut it down unless it's down already */
360 	if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) {
361 		/*
362 		 * Quiesce the connection mgmt handlers before we start tearing
363 		 * things down. We don't hold the mutex for the entire
364 		 * duration of the shutdown operation, else we may be
365 		 * deadlocking with the CM handler. Instead, the CM event
366 		 * handler is supposed to check for state DISCONNECTING
367 		 */
368 		mutex_lock(&cp->cp_cm_lock);
369 		if (!rds_conn_path_transition(cp, RDS_CONN_UP,
370 					      RDS_CONN_DISCONNECTING) &&
371 		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
372 					      RDS_CONN_DISCONNECTING)) {
373 			rds_conn_path_error(cp,
374 					    "shutdown called in state %d\n",
375 					    atomic_read(&cp->cp_state));
376 			mutex_unlock(&cp->cp_cm_lock);
377 			return;
378 		}
379 		mutex_unlock(&cp->cp_cm_lock);
380 
381 		wait_event(cp->cp_waitq,
382 			   !test_bit(RDS_IN_XMIT, &cp->cp_flags));
383 		wait_event(cp->cp_waitq,
384 			   !test_bit(RDS_RECV_REFILL, &cp->cp_flags));
385 
386 		conn->c_trans->conn_path_shutdown(cp);
387 		rds_conn_path_reset(cp);
388 
389 		if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING,
390 					      RDS_CONN_DOWN) &&
391 		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
392 					      RDS_CONN_DOWN)) {
393 			/* This can happen - eg when we're in the middle of tearing
394 			 * down the connection, and someone unloads the rds module.
395 			 * Quite reproducible with loopback connections.
396 			 * Mostly harmless.
397 			 *
398 			 * Note that this also happens with rds-tcp because
399 			 * we could have triggered rds_conn_path_drop in irq
400 			 * mode from rds_tcp_state change on the receipt of
401 			 * a FIN, thus we need to recheck for RDS_CONN_ERROR
402 			 * here.
403 			 */
404 			rds_conn_path_error(cp, "%s: failed to transition "
405 					    "to state DOWN, current state "
406 					    "is %d\n", __func__,
407 					    atomic_read(&cp->cp_state));
408 			return;
409 		}
410 	}
411 
412 	/* Then reconnect if it's still live.
413 	 * The passive side of an IB loopback connection is never added
414 	 * to the conn hash, so we never trigger a reconnect on this
415 	 * conn - the reconnect is always triggered by the active peer. */
416 	cancel_delayed_work_sync(&cp->cp_conn_w);
417 	rcu_read_lock();
418 	if (!hlist_unhashed(&conn->c_hash_node)) {
419 		rcu_read_unlock();
420 		rds_queue_reconnect(cp);
421 	} else {
422 		rcu_read_unlock();
423 	}
424 }
425 
426 /* destroy a single rds_conn_path. rds_conn_destroy() iterates over
427  * all paths using rds_conn_path_destroy()
428  */
429 static void rds_conn_path_destroy(struct rds_conn_path *cp)
430 {
431 	struct rds_message *rm, *rtmp;
432 
433 	if (!cp->cp_transport_data)
434 		return;
435 
436 	/* make sure lingering queued work won't try to ref the conn */
437 	cancel_delayed_work_sync(&cp->cp_send_w);
438 	cancel_delayed_work_sync(&cp->cp_recv_w);
439 
440 	rds_conn_path_drop(cp, true);
441 	flush_work(&cp->cp_down_w);
442 
443 	/* tear down queued messages */
444 	list_for_each_entry_safe(rm, rtmp,
445 				 &cp->cp_send_queue,
446 				 m_conn_item) {
447 		list_del_init(&rm->m_conn_item);
448 		BUG_ON(!list_empty(&rm->m_sock_item));
449 		rds_message_put(rm);
450 	}
451 	if (cp->cp_xmit_rm)
452 		rds_message_put(cp->cp_xmit_rm);
453 
454 	WARN_ON(delayed_work_pending(&cp->cp_send_w));
455 	WARN_ON(delayed_work_pending(&cp->cp_recv_w));
456 	WARN_ON(delayed_work_pending(&cp->cp_conn_w));
457 	WARN_ON(work_pending(&cp->cp_down_w));
458 
459 	cp->cp_conn->c_trans->conn_free(cp->cp_transport_data);
460 }
461 
462 /*
463  * Stop and free a connection.
464  *
465  * This can only be used in very limited circumstances.  It assumes that once
466  * the conn has been shutdown that no one else is referencing the connection.
467  * We can only ensure this in the rmmod path in the current code.
468  */
469 void rds_conn_destroy(struct rds_connection *conn)
470 {
471 	unsigned long flags;
472 	int i;
473 	struct rds_conn_path *cp;
474 	int npaths = (conn->c_trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
475 
476 	rdsdebug("freeing conn %p for %pI4 -> "
477 		 "%pI4\n", conn, &conn->c_laddr,
478 		 &conn->c_faddr);
479 
480 	/* Ensure conn will not be scheduled for reconnect */
481 	spin_lock_irq(&rds_conn_lock);
482 	hlist_del_init_rcu(&conn->c_hash_node);
483 	spin_unlock_irq(&rds_conn_lock);
484 	synchronize_rcu();
485 
486 	/* shut the connection down */
487 	for (i = 0; i < npaths; i++) {
488 		cp = &conn->c_path[i];
489 		rds_conn_path_destroy(cp);
490 		BUG_ON(!list_empty(&cp->cp_retrans));
491 	}
492 
493 	/*
494 	 * The congestion maps aren't freed up here.  They're
495 	 * freed by rds_cong_exit() after all the connections
496 	 * have been freed.
497 	 */
498 	rds_cong_remove_conn(conn);
499 
500 	kfree(conn->c_path);
501 	kmem_cache_free(rds_conn_slab, conn);
502 
503 	spin_lock_irqsave(&rds_conn_lock, flags);
504 	rds_conn_count--;
505 	spin_unlock_irqrestore(&rds_conn_lock, flags);
506 }
507 EXPORT_SYMBOL_GPL(rds_conn_destroy);
508 
509 static void __rds_inc_msg_cp(struct rds_incoming *inc,
510 			     struct rds_info_iterator *iter,
511 			     void *saddr, void *daddr, int flip, bool isv6)
512 {
513 #if IS_ENABLED(CONFIG_IPV6)
514 	if (isv6)
515 		rds6_inc_info_copy(inc, iter, saddr, daddr, flip);
516 	else
517 #endif
518 		rds_inc_info_copy(inc, iter, *(__be32 *)saddr,
519 				  *(__be32 *)daddr, flip);
520 }
521 
522 static void rds_conn_message_info_cmn(struct socket *sock, unsigned int len,
523 				      struct rds_info_iterator *iter,
524 				      struct rds_info_lengths *lens,
525 				      int want_send, bool isv6)
526 {
527 	struct hlist_head *head;
528 	struct list_head *list;
529 	struct rds_connection *conn;
530 	struct rds_message *rm;
531 	unsigned int total = 0;
532 	unsigned long flags;
533 	size_t i;
534 	int j;
535 
536 	if (isv6)
537 		len /= sizeof(struct rds6_info_message);
538 	else
539 		len /= sizeof(struct rds_info_message);
540 
541 	rcu_read_lock();
542 
543 	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
544 	     i++, head++) {
545 		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
546 			struct rds_conn_path *cp;
547 			int npaths;
548 
549 			if (!isv6 && conn->c_isv6)
550 				continue;
551 
552 			npaths = (conn->c_trans->t_mp_capable ?
553 				 RDS_MPATH_WORKERS : 1);
554 
555 			for (j = 0; j < npaths; j++) {
556 				cp = &conn->c_path[j];
557 				if (want_send)
558 					list = &cp->cp_send_queue;
559 				else
560 					list = &cp->cp_retrans;
561 
562 				spin_lock_irqsave(&cp->cp_lock, flags);
563 
564 				/* XXX too lazy to maintain counts.. */
565 				list_for_each_entry(rm, list, m_conn_item) {
566 					total++;
567 					if (total <= len)
568 						__rds_inc_msg_cp(&rm->m_inc,
569 								 iter,
570 								 &conn->c_laddr,
571 								 &conn->c_faddr,
572 								 0, isv6);
573 				}
574 
575 				spin_unlock_irqrestore(&cp->cp_lock, flags);
576 			}
577 		}
578 	}
579 	rcu_read_unlock();
580 
581 	lens->nr = total;
582 	if (isv6)
583 		lens->each = sizeof(struct rds6_info_message);
584 	else
585 		lens->each = sizeof(struct rds_info_message);
586 }
587 
588 static void rds_conn_message_info(struct socket *sock, unsigned int len,
589 				  struct rds_info_iterator *iter,
590 				  struct rds_info_lengths *lens,
591 				  int want_send)
592 {
593 	rds_conn_message_info_cmn(sock, len, iter, lens, want_send, false);
594 }
595 
596 #if IS_ENABLED(CONFIG_IPV6)
597 static void rds6_conn_message_info(struct socket *sock, unsigned int len,
598 				   struct rds_info_iterator *iter,
599 				   struct rds_info_lengths *lens,
600 				   int want_send)
601 {
602 	rds_conn_message_info_cmn(sock, len, iter, lens, want_send, true);
603 }
604 #endif
605 
606 static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
607 				       struct rds_info_iterator *iter,
608 				       struct rds_info_lengths *lens)
609 {
610 	rds_conn_message_info(sock, len, iter, lens, 1);
611 }
612 
613 #if IS_ENABLED(CONFIG_IPV6)
614 static void rds6_conn_message_info_send(struct socket *sock, unsigned int len,
615 					struct rds_info_iterator *iter,
616 					struct rds_info_lengths *lens)
617 {
618 	rds6_conn_message_info(sock, len, iter, lens, 1);
619 }
620 #endif
621 
622 static void rds_conn_message_info_retrans(struct socket *sock,
623 					  unsigned int len,
624 					  struct rds_info_iterator *iter,
625 					  struct rds_info_lengths *lens)
626 {
627 	rds_conn_message_info(sock, len, iter, lens, 0);
628 }
629 
630 #if IS_ENABLED(CONFIG_IPV6)
631 static void rds6_conn_message_info_retrans(struct socket *sock,
632 					   unsigned int len,
633 					   struct rds_info_iterator *iter,
634 					   struct rds_info_lengths *lens)
635 {
636 	rds6_conn_message_info(sock, len, iter, lens, 0);
637 }
638 #endif
639 
640 void rds_for_each_conn_info(struct socket *sock, unsigned int len,
641 			  struct rds_info_iterator *iter,
642 			  struct rds_info_lengths *lens,
643 			  int (*visitor)(struct rds_connection *, void *),
644 			  u64 *buffer,
645 			  size_t item_len)
646 {
647 	struct hlist_head *head;
648 	struct rds_connection *conn;
649 	size_t i;
650 
651 	rcu_read_lock();
652 
653 	lens->nr = 0;
654 	lens->each = item_len;
655 
656 	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
657 	     i++, head++) {
658 		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
659 
660 			/* XXX no c_lock usage.. */
661 			if (!visitor(conn, buffer))
662 				continue;
663 
664 			/* We copy as much as we can fit in the buffer,
665 			 * but we count all items so that the caller
666 			 * can resize the buffer. */
667 			if (len >= item_len) {
668 				rds_info_copy(iter, buffer, item_len);
669 				len -= item_len;
670 			}
671 			lens->nr++;
672 		}
673 	}
674 	rcu_read_unlock();
675 }
676 EXPORT_SYMBOL_GPL(rds_for_each_conn_info);
677 
678 static void rds_walk_conn_path_info(struct socket *sock, unsigned int len,
679 				    struct rds_info_iterator *iter,
680 				    struct rds_info_lengths *lens,
681 				    int (*visitor)(struct rds_conn_path *, void *),
682 				    u64 *buffer,
683 				    size_t item_len)
684 {
685 	struct hlist_head *head;
686 	struct rds_connection *conn;
687 	size_t i;
688 
689 	rcu_read_lock();
690 
691 	lens->nr = 0;
692 	lens->each = item_len;
693 
694 	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
695 	     i++, head++) {
696 		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
697 			struct rds_conn_path *cp;
698 
699 			/* XXX We only copy the information from the first
700 			 * path for now.  The problem is that if there are
701 			 * more than one underlying paths, we cannot report
702 			 * information of all of them using the existing
703 			 * API.  For example, there is only one next_tx_seq,
704 			 * which path's next_tx_seq should we report?  It is
705 			 * a bug in the design of MPRDS.
706 			 */
707 			cp = conn->c_path;
708 
709 			/* XXX no cp_lock usage.. */
710 			if (!visitor(cp, buffer))
711 				continue;
712 
713 			/* We copy as much as we can fit in the buffer,
714 			 * but we count all items so that the caller
715 			 * can resize the buffer.
716 			 */
717 			if (len >= item_len) {
718 				rds_info_copy(iter, buffer, item_len);
719 				len -= item_len;
720 			}
721 			lens->nr++;
722 		}
723 	}
724 	rcu_read_unlock();
725 }
726 
727 static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
728 {
729 	struct rds_info_connection *cinfo = buffer;
730 	struct rds_connection *conn = cp->cp_conn;
731 
732 	if (conn->c_isv6)
733 		return 0;
734 
735 	cinfo->next_tx_seq = cp->cp_next_tx_seq;
736 	cinfo->next_rx_seq = cp->cp_next_rx_seq;
737 	cinfo->laddr = conn->c_laddr.s6_addr32[3];
738 	cinfo->faddr = conn->c_faddr.s6_addr32[3];
739 	strncpy(cinfo->transport, conn->c_trans->t_name,
740 		sizeof(cinfo->transport));
741 	cinfo->flags = 0;
742 
743 	rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
744 			  SENDING);
745 	/* XXX Future: return the state rather than these funky bits */
746 	rds_conn_info_set(cinfo->flags,
747 			  atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
748 			  CONNECTING);
749 	rds_conn_info_set(cinfo->flags,
750 			  atomic_read(&cp->cp_state) == RDS_CONN_UP,
751 			  CONNECTED);
752 	return 1;
753 }
754 
755 #if IS_ENABLED(CONFIG_IPV6)
756 static int rds6_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
757 {
758 	struct rds6_info_connection *cinfo6 = buffer;
759 	struct rds_connection *conn = cp->cp_conn;
760 
761 	cinfo6->next_tx_seq = cp->cp_next_tx_seq;
762 	cinfo6->next_rx_seq = cp->cp_next_rx_seq;
763 	cinfo6->laddr = conn->c_laddr;
764 	cinfo6->faddr = conn->c_faddr;
765 	strncpy(cinfo6->transport, conn->c_trans->t_name,
766 		sizeof(cinfo6->transport));
767 	cinfo6->flags = 0;
768 
769 	rds_conn_info_set(cinfo6->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
770 			  SENDING);
771 	/* XXX Future: return the state rather than these funky bits */
772 	rds_conn_info_set(cinfo6->flags,
773 			  atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
774 			  CONNECTING);
775 	rds_conn_info_set(cinfo6->flags,
776 			  atomic_read(&cp->cp_state) == RDS_CONN_UP,
777 			  CONNECTED);
778 	/* Just return 1 as there is no error case. This is a helper function
779 	 * for rds_walk_conn_path_info() and it wants a return value.
780 	 */
781 	return 1;
782 }
783 #endif
784 
785 static void rds_conn_info(struct socket *sock, unsigned int len,
786 			  struct rds_info_iterator *iter,
787 			  struct rds_info_lengths *lens)
788 {
789 	u64 buffer[(sizeof(struct rds_info_connection) + 7) / 8];
790 
791 	rds_walk_conn_path_info(sock, len, iter, lens,
792 				rds_conn_info_visitor,
793 				buffer,
794 				sizeof(struct rds_info_connection));
795 }
796 
797 #if IS_ENABLED(CONFIG_IPV6)
798 static void rds6_conn_info(struct socket *sock, unsigned int len,
799 			   struct rds_info_iterator *iter,
800 			   struct rds_info_lengths *lens)
801 {
802 	u64 buffer[(sizeof(struct rds6_info_connection) + 7) / 8];
803 
804 	rds_walk_conn_path_info(sock, len, iter, lens,
805 				rds6_conn_info_visitor,
806 				buffer,
807 				sizeof(struct rds6_info_connection));
808 }
809 #endif
810 
811 int rds_conn_init(void)
812 {
813 	int ret;
814 
815 	ret = rds_loop_net_init(); /* register pernet callback */
816 	if (ret)
817 		return ret;
818 
819 	rds_conn_slab = kmem_cache_create("rds_connection",
820 					  sizeof(struct rds_connection),
821 					  0, 0, NULL);
822 	if (!rds_conn_slab) {
823 		rds_loop_net_exit();
824 		return -ENOMEM;
825 	}
826 
827 	rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
828 	rds_info_register_func(RDS_INFO_SEND_MESSAGES,
829 			       rds_conn_message_info_send);
830 	rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
831 			       rds_conn_message_info_retrans);
832 #if IS_ENABLED(CONFIG_IPV6)
833 	rds_info_register_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
834 	rds_info_register_func(RDS6_INFO_SEND_MESSAGES,
835 			       rds6_conn_message_info_send);
836 	rds_info_register_func(RDS6_INFO_RETRANS_MESSAGES,
837 			       rds6_conn_message_info_retrans);
838 #endif
839 	return 0;
840 }
841 
842 void rds_conn_exit(void)
843 {
844 	rds_loop_net_exit(); /* unregister pernet callback */
845 	rds_loop_exit();
846 
847 	WARN_ON(!hlist_empty(rds_conn_hash));
848 
849 	kmem_cache_destroy(rds_conn_slab);
850 
851 	rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
852 	rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
853 				 rds_conn_message_info_send);
854 	rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
855 				 rds_conn_message_info_retrans);
856 #if IS_ENABLED(CONFIG_IPV6)
857 	rds_info_deregister_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
858 	rds_info_deregister_func(RDS6_INFO_SEND_MESSAGES,
859 				 rds6_conn_message_info_send);
860 	rds_info_deregister_func(RDS6_INFO_RETRANS_MESSAGES,
861 				 rds6_conn_message_info_retrans);
862 #endif
863 }
864 
865 /*
866  * Force a disconnect
867  */
868 void rds_conn_path_drop(struct rds_conn_path *cp, bool destroy)
869 {
870 	atomic_set(&cp->cp_state, RDS_CONN_ERROR);
871 
872 	rcu_read_lock();
873 	if (!destroy && rds_destroy_pending(cp->cp_conn)) {
874 		rcu_read_unlock();
875 		return;
876 	}
877 	queue_work(rds_wq, &cp->cp_down_w);
878 	rcu_read_unlock();
879 }
880 EXPORT_SYMBOL_GPL(rds_conn_path_drop);
881 
882 void rds_conn_drop(struct rds_connection *conn)
883 {
884 	WARN_ON(conn->c_trans->t_mp_capable);
885 	rds_conn_path_drop(&conn->c_path[0], false);
886 }
887 EXPORT_SYMBOL_GPL(rds_conn_drop);
888 
889 /*
890  * If the connection is down, trigger a connect. We may have scheduled a
891  * delayed reconnect however - in this case we should not interfere.
892  */
893 void rds_conn_path_connect_if_down(struct rds_conn_path *cp)
894 {
895 	rcu_read_lock();
896 	if (rds_destroy_pending(cp->cp_conn)) {
897 		rcu_read_unlock();
898 		return;
899 	}
900 	if (rds_conn_path_state(cp) == RDS_CONN_DOWN &&
901 	    !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
902 		queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
903 	rcu_read_unlock();
904 }
905 EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down);
906 
907 void rds_conn_connect_if_down(struct rds_connection *conn)
908 {
909 	WARN_ON(conn->c_trans->t_mp_capable);
910 	rds_conn_path_connect_if_down(&conn->c_path[0]);
911 }
912 EXPORT_SYMBOL_GPL(rds_conn_connect_if_down);
913 
914 void
915 __rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...)
916 {
917 	va_list ap;
918 
919 	va_start(ap, fmt);
920 	vprintk(fmt, ap);
921 	va_end(ap);
922 
923 	rds_conn_path_drop(cp, false);
924 }
925