connection.c

  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.. */
 49static DEFINE_SPINLOCK(rds_conn_lock);
 50static unsigned long rds_conn_count;
 51static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
 52static struct kmem_cache *rds_conn_slab;
 53
 54static 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	__be32 lhash, fhash;
 61	u32 hash;
 62
 63	net_get_random_once(&rds_hash_secret, sizeof(rds_hash_secret));
 64	net_get_random_once(&rds6_hash_secret, sizeof(rds6_hash_secret));
 65
 66	lhash = laddr->s6_addr32[3];
 67#if IS_ENABLED(CONFIG_IPV6)
 68	fhash = (__force __be32)__ipv6_addr_jhash(faddr, rds6_hash_secret);
 69#else
 70	fhash = faddr->s6_addr32[3];
 71#endif
 72	hash = __inet_ehashfn(lhash, 0, fhash, 0, rds_hash_secret);
 73
 74	return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
 75}
 76
 77#define rds_conn_info_set(var, test, suffix) do {		\
 78	if (test)						\
 79		var |= RDS_INFO_CONNECTION_FLAG_##suffix;	\
 80} while (0)
 81
 82/* rcu read lock must be held or the connection spinlock */
 83static struct rds_connection *rds_conn_lookup(struct net *net,
 84					      struct hlist_head *head,
 85					      const struct in6_addr *laddr,
 86					      const struct in6_addr *faddr,
 87					      struct rds_transport *trans,
 88					      u8 tos, int dev_if)
 89{
 90	struct rds_connection *conn, *ret = NULL;
 91
 92	hlist_for_each_entry_rcu(conn, head, c_hash_node) {
 93		if (ipv6_addr_equal(&conn->c_faddr, faddr) &&
 94		    ipv6_addr_equal(&conn->c_laddr, laddr) &&
 95		    conn->c_trans == trans &&
 96		    conn->c_tos == tos &&
 97		    net == rds_conn_net(conn) &&
 98		    conn->c_dev_if == dev_if) {
 99			ret = conn;
100			break;
101		}
102	}
103	rdsdebug("returning conn %p for %pI6c -> %pI6c\n", ret,
104		 laddr, faddr);
105	return ret;
106}
107
108/*
109 * This is called by transports as they're bringing down a connection.
110 * It clears partial message state so that the transport can start sending
111 * and receiving over this connection again in the future.  It is up to
112 * the transport to have serialized this call with its send and recv.
113 */
114static void rds_conn_path_reset(struct rds_conn_path *cp)
115{
116	struct rds_connection *conn = cp->cp_conn;
117
118	rdsdebug("connection %pI6c to %pI6c reset\n",
119		 &conn->c_laddr, &conn->c_faddr);
120
121	rds_stats_inc(s_conn_reset);
122	rds_send_path_reset(cp);
123	cp->cp_flags = 0;
124
125	/* Do not clear next_rx_seq here, else we cannot distinguish
126	 * retransmitted packets from new packets, and will hand all
127	 * of them to the application. That is not consistent with the
128	 * reliability guarantees of RDS. */
129}
130
131static void __rds_conn_path_init(struct rds_connection *conn,
132				 struct rds_conn_path *cp, bool is_outgoing)
133{
134	spin_lock_init(&cp->cp_lock);
135	cp->cp_next_tx_seq = 1;
136	init_waitqueue_head(&cp->cp_waitq);
137	INIT_LIST_HEAD(&cp->cp_send_queue);
138	INIT_LIST_HEAD(&cp->cp_retrans);
139
140	cp->cp_conn = conn;
141	atomic_set(&cp->cp_state, RDS_CONN_DOWN);
142	cp->cp_send_gen = 0;
143	cp->cp_reconnect_jiffies = 0;
144	cp->cp_conn->c_proposed_version = RDS_PROTOCOL_VERSION;
145	INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker);
146	INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker);
147	INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker);
148	INIT_WORK(&cp->cp_down_w, rds_shutdown_worker);
149	mutex_init(&cp->cp_cm_lock);
150	cp->cp_flags = 0;
151}
152
153/*
154 * There is only every one 'conn' for a given pair of addresses in the
155 * system at a time.  They contain messages to be retransmitted and so
156 * span the lifetime of the actual underlying transport connections.
157 *
158 * For now they are not garbage collected once they're created.  They
159 * are torn down as the module is removed, if ever.
160 */
161static struct rds_connection *__rds_conn_create(struct net *net,
162						const struct in6_addr *laddr,
163						const struct in6_addr *faddr,
164						struct rds_transport *trans,
165						gfp_t gfp, u8 tos,
166						int is_outgoing,
167						int dev_if)
168{
169	struct rds_connection *conn, *parent = NULL;
170	struct hlist_head *head = rds_conn_bucket(laddr, faddr);
171	struct rds_transport *loop_trans;
172	unsigned long flags;
173	int ret, i;
174	int npaths = (trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
175
176	rcu_read_lock();
177	conn = rds_conn_lookup(net, head, laddr, faddr, trans, tos, dev_if);
178	if (conn &&
179	    conn->c_loopback &&
180	    conn->c_trans != &rds_loop_transport &&
181	    ipv6_addr_equal(laddr, faddr) &&
182	    !is_outgoing) {
183		/* This is a looped back IB connection, and we're
184		 * called by the code handling the incoming connect.
185		 * We need a second connection object into which we
186		 * can stick the other QP. */
187		parent = conn;
188		conn = parent->c_passive;
189	}
190	rcu_read_unlock();
191	if (conn)
192		goto out;
193
194	conn = kmem_cache_zalloc(rds_conn_slab, gfp);
195	if (!conn) {
196		conn = ERR_PTR(-ENOMEM);
197		goto out;
198	}
199	conn->c_path = kcalloc(npaths, sizeof(struct rds_conn_path), gfp);
200	if (!conn->c_path) {
201		kmem_cache_free(rds_conn_slab, conn);
202		conn = ERR_PTR(-ENOMEM);
203		goto out;
204	}
205
206	INIT_HLIST_NODE(&conn->c_hash_node);
207	conn->c_laddr = *laddr;
208	conn->c_isv6 = !ipv6_addr_v4mapped(laddr);
209	conn->c_faddr = *faddr;
210	conn->c_dev_if = dev_if;
211	conn->c_tos = tos;
212
213#if IS_ENABLED(CONFIG_IPV6)
214	/* If the local address is link local, set c_bound_if to be the
215	 * index used for this connection.  Otherwise, set it to 0 as
216	 * the socket is not bound to an interface.  c_bound_if is used
217	 * to look up a socket when a packet is received
218	 */
219	if (ipv6_addr_type(laddr) & IPV6_ADDR_LINKLOCAL)
220		conn->c_bound_if = dev_if;
221	else
222#endif
223		conn->c_bound_if = 0;
224
225	rds_conn_net_set(conn, net);
226
227	ret = rds_cong_get_maps(conn);
228	if (ret) {
229		kfree(conn->c_path);
230		kmem_cache_free(rds_conn_slab, conn);
231		conn = ERR_PTR(ret);
232		goto out;
233	}
234
235	/*
236	 * This is where a connection becomes loopback.  If *any* RDS sockets
237	 * can bind to the destination address then we'd rather the messages
238	 * flow through loopback rather than either transport.
239	 */
240	loop_trans = rds_trans_get_preferred(net, faddr, conn->c_dev_if);
241	if (loop_trans) {
242		rds_trans_put(loop_trans);
243		conn->c_loopback = 1;
244		if (trans->t_prefer_loopback) {
245			if (likely(is_outgoing)) {
246				/* "outgoing" connection to local address.
247				 * Protocol says it wants the connection
248				 * handled by the loopback transport.
249				 * This is what TCP does.
250				 */
251				trans = &rds_loop_transport;
252			} else {
253				/* No transport currently in use
254				 * should end up here, but if it
255				 * does, reset/destroy the connection.
256				 */
257				kfree(conn->c_path);
258				kmem_cache_free(rds_conn_slab, conn);
259				conn = ERR_PTR(-EOPNOTSUPP);
260				goto out;
261			}
262		}
263	}
264
265	conn->c_trans = trans;
266
267	init_waitqueue_head(&conn->c_hs_waitq);
268	for (i = 0; i < npaths; i++) {
269		__rds_conn_path_init(conn, &conn->c_path[i],
270				     is_outgoing);
271		conn->c_path[i].cp_index = i;
272	}
273	rcu_read_lock();
274	if (rds_destroy_pending(conn))
275		ret = -ENETDOWN;
276	else
277		ret = trans->conn_alloc(conn, GFP_ATOMIC);
278	if (ret) {
279		rcu_read_unlock();
280		kfree(conn->c_path);
281		kmem_cache_free(rds_conn_slab, conn);
282		conn = ERR_PTR(ret);
283		goto out;
284	}
285
286	rdsdebug("allocated conn %p for %pI6c -> %pI6c over %s %s\n",
287		 conn, laddr, faddr,
288		 strnlen(trans->t_name, sizeof(trans->t_name)) ?
289		 trans->t_name : "[unknown]", is_outgoing ? "(outgoing)" : "");
290
291	/*
292	 * Since we ran without holding the conn lock, someone could
293	 * have created the same conn (either normal or passive) in the
294	 * interim. We check while holding the lock. If we won, we complete
295	 * init and return our conn. If we lost, we rollback and return the
296	 * other one.
297	 */
298	spin_lock_irqsave(&rds_conn_lock, flags);
299	if (parent) {
300		/* Creating passive conn */
301		if (parent->c_passive) {
302			trans->conn_free(conn->c_path[0].cp_transport_data);
303			kfree(conn->c_path);
304			kmem_cache_free(rds_conn_slab, conn);
305			conn = parent->c_passive;
306		} else {
307			parent->c_passive = conn;
308			rds_cong_add_conn(conn);
309			rds_conn_count++;
310		}
311	} else {
312		/* Creating normal conn */
313		struct rds_connection *found;
314
315		found = rds_conn_lookup(net, head, laddr, faddr, trans,
316					tos, dev_if);
317		if (found) {
318			struct rds_conn_path *cp;
319			int i;
320
321			for (i = 0; i < npaths; i++) {
322				cp = &conn->c_path[i];
323				/* The ->conn_alloc invocation may have
324				 * allocated resource for all paths, so all
325				 * of them may have to be freed here.
326				 */
327				if (cp->cp_transport_data)
328					trans->conn_free(cp->cp_transport_data);
329			}
330			kfree(conn->c_path);
331			kmem_cache_free(rds_conn_slab, conn);
332			conn = found;
333		} else {
334			conn->c_my_gen_num = rds_gen_num;
335			conn->c_peer_gen_num = 0;
336			hlist_add_head_rcu(&conn->c_hash_node, head);
337			rds_cong_add_conn(conn);
338			rds_conn_count++;
339		}
340	}
341	spin_unlock_irqrestore(&rds_conn_lock, flags);
342	rcu_read_unlock();
343
344out:
345	return conn;
346}
347
348struct rds_connection *rds_conn_create(struct net *net,
349				       const struct in6_addr *laddr,
350				       const struct in6_addr *faddr,
351				       struct rds_transport *trans, u8 tos,
352				       gfp_t gfp, int dev_if)
353{
354	return __rds_conn_create(net, laddr, faddr, trans, gfp, tos, 0, dev_if);
355}
356EXPORT_SYMBOL_GPL(rds_conn_create);
357
358struct rds_connection *rds_conn_create_outgoing(struct net *net,
359						const struct in6_addr *laddr,
360						const struct in6_addr *faddr,
361						struct rds_transport *trans,
362						u8 tos, gfp_t gfp, int dev_if)
363{
364	return __rds_conn_create(net, laddr, faddr, trans, gfp, tos, 1, dev_if);
365}
366EXPORT_SYMBOL_GPL(rds_conn_create_outgoing);
367
368void rds_conn_shutdown(struct rds_conn_path *cp)
369{
370	struct rds_connection *conn = cp->cp_conn;
371
372	/* shut it down unless it's down already */
373	if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) {
374		/*
375		 * Quiesce the connection mgmt handlers before we start tearing
376		 * things down. We don't hold the mutex for the entire
377		 * duration of the shutdown operation, else we may be
378		 * deadlocking with the CM handler. Instead, the CM event
379		 * handler is supposed to check for state DISCONNECTING
380		 */
381		mutex_lock(&cp->cp_cm_lock);
382		if (!rds_conn_path_transition(cp, RDS_CONN_UP,
383					      RDS_CONN_DISCONNECTING) &&
384		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
385					      RDS_CONN_DISCONNECTING) &&
386		    !rds_conn_path_transition(cp, RDS_CONN_RESETTING,
387					      RDS_CONN_DISCONNECTING)) {
388			rds_conn_path_error(cp,
389					    "shutdown called in state %d\n",
390					    atomic_read(&cp->cp_state));
391			mutex_unlock(&cp->cp_cm_lock);
392			return;
393		}
394		mutex_unlock(&cp->cp_cm_lock);
395
396		wait_event(cp->cp_waitq,
397			   !test_bit(RDS_IN_XMIT, &cp->cp_flags));
398		wait_event(cp->cp_waitq,
399			   !test_bit(RDS_RECV_REFILL, &cp->cp_flags));
400
401		conn->c_trans->conn_path_shutdown(cp);
402		rds_conn_path_reset(cp);
403
404		if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING,
405					      RDS_CONN_DOWN) &&
406		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
407					      RDS_CONN_DOWN)) {
408			/* This can happen - eg when we're in the middle of tearing
409			 * down the connection, and someone unloads the rds module.
410			 * Quite reproducible with loopback connections.
411			 * Mostly harmless.
412			 *
413			 * Note that this also happens with rds-tcp because
414			 * we could have triggered rds_conn_path_drop in irq
415			 * mode from rds_tcp_state change on the receipt of
416			 * a FIN, thus we need to recheck for RDS_CONN_ERROR
417			 * here.
418			 */
419			rds_conn_path_error(cp, "%s: failed to transition "
420					    "to state DOWN, current state "
421					    "is %d\n", __func__,
422					    atomic_read(&cp->cp_state));
423			return;
424		}
425	}
426
427	/* Then reconnect if it's still live.
428	 * The passive side of an IB loopback connection is never added
429	 * to the conn hash, so we never trigger a reconnect on this
430	 * conn - the reconnect is always triggered by the active peer. */
431	cancel_delayed_work_sync(&cp->cp_conn_w);
432
433	clear_bit(RDS_RECONNECT_PENDING, &cp->cp_flags);
434	rcu_read_lock();
435	if (!hlist_unhashed(&conn->c_hash_node)) {
436		rcu_read_unlock();
437		rds_queue_reconnect(cp);
438	} else {
439		rcu_read_unlock();
440	}
441}
442
443/* destroy a single rds_conn_path. rds_conn_destroy() iterates over
444 * all paths using rds_conn_path_destroy()
445 */
446static void rds_conn_path_destroy(struct rds_conn_path *cp)
447{
448	struct rds_message *rm, *rtmp;
449
450	if (!cp->cp_transport_data)
451		return;
452
453	/* make sure lingering queued work won't try to ref the conn */
454	cancel_delayed_work_sync(&cp->cp_send_w);
455	cancel_delayed_work_sync(&cp->cp_recv_w);
456
457	rds_conn_path_drop(cp, true);
458	flush_work(&cp->cp_down_w);
459
460	/* tear down queued messages */
461	list_for_each_entry_safe(rm, rtmp,
462				 &cp->cp_send_queue,
463				 m_conn_item) {
464		list_del_init(&rm->m_conn_item);
465		BUG_ON(!list_empty(&rm->m_sock_item));
466		rds_message_put(rm);
467	}
468	if (cp->cp_xmit_rm)
469		rds_message_put(cp->cp_xmit_rm);
470
471	WARN_ON(delayed_work_pending(&cp->cp_send_w));
472	WARN_ON(delayed_work_pending(&cp->cp_recv_w));
473	WARN_ON(delayed_work_pending(&cp->cp_conn_w));
474	WARN_ON(work_pending(&cp->cp_down_w));
475
476	cp->cp_conn->c_trans->conn_free(cp->cp_transport_data);
477}
478
479/*
480 * Stop and free a connection.
481 *
482 * This can only be used in very limited circumstances.  It assumes that once
483 * the conn has been shutdown that no one else is referencing the connection.
484 * We can only ensure this in the rmmod path in the current code.
485 */
486void rds_conn_destroy(struct rds_connection *conn)
487{
488	unsigned long flags;
489	int i;
490	struct rds_conn_path *cp;
491	int npaths = (conn->c_trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
492
493	rdsdebug("freeing conn %p for %pI4 -> "
494		 "%pI4\n", conn, &conn->c_laddr,
495		 &conn->c_faddr);
496
497	/* Ensure conn will not be scheduled for reconnect */
498	spin_lock_irq(&rds_conn_lock);
499	hlist_del_init_rcu(&conn->c_hash_node);
500	spin_unlock_irq(&rds_conn_lock);
501	synchronize_rcu();
502
503	/* shut the connection down */
504	for (i = 0; i < npaths; i++) {
505		cp = &conn->c_path[i];
506		rds_conn_path_destroy(cp);
507		BUG_ON(!list_empty(&cp->cp_retrans));
508	}
509
510	/*
511	 * The congestion maps aren't freed up here.  They're
512	 * freed by rds_cong_exit() after all the connections
513	 * have been freed.
514	 */
515	rds_cong_remove_conn(conn);
516
517	kfree(conn->c_path);
518	kmem_cache_free(rds_conn_slab, conn);
519
520	spin_lock_irqsave(&rds_conn_lock, flags);
521	rds_conn_count--;
522	spin_unlock_irqrestore(&rds_conn_lock, flags);
523}
524EXPORT_SYMBOL_GPL(rds_conn_destroy);
525
526static void __rds_inc_msg_cp(struct rds_incoming *inc,
527			     struct rds_info_iterator *iter,
528			     void *saddr, void *daddr, int flip, bool isv6)
529{
530#if IS_ENABLED(CONFIG_IPV6)
531	if (isv6)
532		rds6_inc_info_copy(inc, iter, saddr, daddr, flip);
533	else
534#endif
535		rds_inc_info_copy(inc, iter, *(__be32 *)saddr,
536				  *(__be32 *)daddr, flip);
537}
538
539static void rds_conn_message_info_cmn(struct socket *sock, unsigned int len,
540				      struct rds_info_iterator *iter,
541				      struct rds_info_lengths *lens,
542				      int want_send, bool isv6)
543{
544	struct hlist_head *head;
545	struct list_head *list;
546	struct rds_connection *conn;
547	struct rds_message *rm;
548	unsigned int total = 0;
549	unsigned long flags;
550	size_t i;
551	int j;
552
553	if (isv6)
554		len /= sizeof(struct rds6_info_message);
555	else
556		len /= sizeof(struct rds_info_message);
557
558	rcu_read_lock();
559
560	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
561	     i++, head++) {
562		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
563			struct rds_conn_path *cp;
564			int npaths;
565
566			if (!isv6 && conn->c_isv6)
567				continue;
568
569			npaths = (conn->c_trans->t_mp_capable ?
570				 RDS_MPATH_WORKERS : 1);
571
572			for (j = 0; j < npaths; j++) {
573				cp = &conn->c_path[j];
574				if (want_send)
575					list = &cp->cp_send_queue;
576				else
577					list = &cp->cp_retrans;
578
579				spin_lock_irqsave(&cp->cp_lock, flags);
580
581				/* XXX too lazy to maintain counts.. */
582				list_for_each_entry(rm, list, m_conn_item) {
583					total++;
584					if (total <= len)
585						__rds_inc_msg_cp(&rm->m_inc,
586								 iter,
587								 &conn->c_laddr,
588								 &conn->c_faddr,
589								 0, isv6);
590				}
591
592				spin_unlock_irqrestore(&cp->cp_lock, flags);
593			}
594		}
595	}
596	rcu_read_unlock();
597
598	lens->nr = total;
599	if (isv6)
600		lens->each = sizeof(struct rds6_info_message);
601	else
602		lens->each = sizeof(struct rds_info_message);
603}
604
605static void rds_conn_message_info(struct socket *sock, unsigned int len,
606				  struct rds_info_iterator *iter,
607				  struct rds_info_lengths *lens,
608				  int want_send)
609{
610	rds_conn_message_info_cmn(sock, len, iter, lens, want_send, false);
611}
612
613#if IS_ENABLED(CONFIG_IPV6)
614static void rds6_conn_message_info(struct socket *sock, unsigned int len,
615				   struct rds_info_iterator *iter,
616				   struct rds_info_lengths *lens,
617				   int want_send)
618{
619	rds_conn_message_info_cmn(sock, len, iter, lens, want_send, true);
620}
621#endif
622
623static void rds_conn_message_info_send(struct socket *sock, 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, 1);
628}
629
630#if IS_ENABLED(CONFIG_IPV6)
631static void rds6_conn_message_info_send(struct socket *sock, unsigned int len,
632					struct rds_info_iterator *iter,
633					struct rds_info_lengths *lens)
634{
635	rds6_conn_message_info(sock, len, iter, lens, 1);
636}
637#endif
638
639static void rds_conn_message_info_retrans(struct socket *sock,
640					  unsigned int len,
641					  struct rds_info_iterator *iter,
642					  struct rds_info_lengths *lens)
643{
644	rds_conn_message_info(sock, len, iter, lens, 0);
645}
646
647#if IS_ENABLED(CONFIG_IPV6)
648static void rds6_conn_message_info_retrans(struct socket *sock,
649					   unsigned int len,
650					   struct rds_info_iterator *iter,
651					   struct rds_info_lengths *lens)
652{
653	rds6_conn_message_info(sock, len, iter, lens, 0);
654}
655#endif
656
657void rds_for_each_conn_info(struct socket *sock, unsigned int len,
658			  struct rds_info_iterator *iter,
659			  struct rds_info_lengths *lens,
660			  int (*visitor)(struct rds_connection *, void *),
661			  u64 *buffer,
662			  size_t item_len)
663{
664	struct hlist_head *head;
665	struct rds_connection *conn;
666	size_t i;
667
668	rcu_read_lock();
669
670	lens->nr = 0;
671	lens->each = item_len;
672
673	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
674	     i++, head++) {
675		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
676
677			/* XXX no c_lock usage.. */
678			if (!visitor(conn, buffer))
679				continue;
680
681			/* We copy as much as we can fit in the buffer,
682			 * but we count all items so that the caller
683			 * can resize the buffer. */
684			if (len >= item_len) {
685				rds_info_copy(iter, buffer, item_len);
686				len -= item_len;
687			}
688			lens->nr++;
689		}
690	}
691	rcu_read_unlock();
692}
693EXPORT_SYMBOL_GPL(rds_for_each_conn_info);
694
695static void rds_walk_conn_path_info(struct socket *sock, unsigned int len,
696				    struct rds_info_iterator *iter,
697				    struct rds_info_lengths *lens,
698				    int (*visitor)(struct rds_conn_path *, void *),
699				    u64 *buffer,
700				    size_t item_len)
701{
702	struct hlist_head *head;
703	struct rds_connection *conn;
704	size_t i;
705
706	rcu_read_lock();
707
708	lens->nr = 0;
709	lens->each = item_len;
710
711	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
712	     i++, head++) {
713		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
714			struct rds_conn_path *cp;
715
716			/* XXX We only copy the information from the first
717			 * path for now.  The problem is that if there are
718			 * more than one underlying paths, we cannot report
719			 * information of all of them using the existing
720			 * API.  For example, there is only one next_tx_seq,
721			 * which path's next_tx_seq should we report?  It is
722			 * a bug in the design of MPRDS.
723			 */
724			cp = conn->c_path;
725
726			/* XXX no cp_lock usage.. */
727			if (!visitor(cp, buffer))
728				continue;
729
730			/* We copy as much as we can fit in the buffer,
731			 * but we count all items so that the caller
732			 * can resize the buffer.
733			 */
734			if (len >= item_len) {
735				rds_info_copy(iter, buffer, item_len);
736				len -= item_len;
737			}
738			lens->nr++;
739		}
740	}
741	rcu_read_unlock();
742}
743
744static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
745{
746	struct rds_info_connection *cinfo = buffer;
747	struct rds_connection *conn = cp->cp_conn;
748
749	if (conn->c_isv6)
750		return 0;
751
752	cinfo->next_tx_seq = cp->cp_next_tx_seq;
753	cinfo->next_rx_seq = cp->cp_next_rx_seq;
754	cinfo->laddr = conn->c_laddr.s6_addr32[3];
755	cinfo->faddr = conn->c_faddr.s6_addr32[3];
756	cinfo->tos = conn->c_tos;
757	strscpy_pad(cinfo->transport, conn->c_trans->t_name);
758	cinfo->flags = 0;
759
760	rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
761			  SENDING);
762	/* XXX Future: return the state rather than these funky bits */
763	rds_conn_info_set(cinfo->flags,
764			  atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
765			  CONNECTING);
766	rds_conn_info_set(cinfo->flags,
767			  atomic_read(&cp->cp_state) == RDS_CONN_UP,
768			  CONNECTED);
769	return 1;
770}
771
772#if IS_ENABLED(CONFIG_IPV6)
773static int rds6_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
774{
775	struct rds6_info_connection *cinfo6 = buffer;
776	struct rds_connection *conn = cp->cp_conn;
777
778	cinfo6->next_tx_seq = cp->cp_next_tx_seq;
779	cinfo6->next_rx_seq = cp->cp_next_rx_seq;
780	cinfo6->laddr = conn->c_laddr;
781	cinfo6->faddr = conn->c_faddr;
782	strscpy_pad(cinfo6->transport, conn->c_trans->t_name);
783	cinfo6->flags = 0;
784
785	rds_conn_info_set(cinfo6->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
786			  SENDING);
787	/* XXX Future: return the state rather than these funky bits */
788	rds_conn_info_set(cinfo6->flags,
789			  atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
790			  CONNECTING);
791	rds_conn_info_set(cinfo6->flags,
792			  atomic_read(&cp->cp_state) == RDS_CONN_UP,
793			  CONNECTED);
794	/* Just return 1 as there is no error case. This is a helper function
795	 * for rds_walk_conn_path_info() and it wants a return value.
796	 */
797	return 1;
798}
799#endif
800
801static void rds_conn_info(struct socket *sock, unsigned int len,
802			  struct rds_info_iterator *iter,
803			  struct rds_info_lengths *lens)
804{
805	u64 buffer[(sizeof(struct rds_info_connection) + 7) / 8];
806
807	rds_walk_conn_path_info(sock, len, iter, lens,
808				rds_conn_info_visitor,
809				buffer,
810				sizeof(struct rds_info_connection));
811}
812
813#if IS_ENABLED(CONFIG_IPV6)
814static void rds6_conn_info(struct socket *sock, unsigned int len,
815			   struct rds_info_iterator *iter,
816			   struct rds_info_lengths *lens)
817{
818	u64 buffer[(sizeof(struct rds6_info_connection) + 7) / 8];
819
820	rds_walk_conn_path_info(sock, len, iter, lens,
821				rds6_conn_info_visitor,
822				buffer,
823				sizeof(struct rds6_info_connection));
824}
825#endif
826
827int rds_conn_init(void)
828{
829	int ret;
830
831	ret = rds_loop_net_init(); /* register pernet callback */
832	if (ret)
833		return ret;
834
835	rds_conn_slab = KMEM_CACHE(rds_connection, 0);
836	if (!rds_conn_slab) {
837		rds_loop_net_exit();
838		return -ENOMEM;
839	}
840
841	rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
842	rds_info_register_func(RDS_INFO_SEND_MESSAGES,
843			       rds_conn_message_info_send);
844	rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
845			       rds_conn_message_info_retrans);
846#if IS_ENABLED(CONFIG_IPV6)
847	rds_info_register_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
848	rds_info_register_func(RDS6_INFO_SEND_MESSAGES,
849			       rds6_conn_message_info_send);
850	rds_info_register_func(RDS6_INFO_RETRANS_MESSAGES,
851			       rds6_conn_message_info_retrans);
852#endif
853	return 0;
854}
855
856void rds_conn_exit(void)
857{
858	rds_loop_net_exit(); /* unregister pernet callback */
859	rds_loop_exit();
860
861	WARN_ON(!hlist_empty(rds_conn_hash));
862
863	kmem_cache_destroy(rds_conn_slab);
864
865	rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
866	rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
867				 rds_conn_message_info_send);
868	rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
869				 rds_conn_message_info_retrans);
870#if IS_ENABLED(CONFIG_IPV6)
871	rds_info_deregister_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
872	rds_info_deregister_func(RDS6_INFO_SEND_MESSAGES,
873				 rds6_conn_message_info_send);
874	rds_info_deregister_func(RDS6_INFO_RETRANS_MESSAGES,
875				 rds6_conn_message_info_retrans);
876#endif
877}
878
879/*
880 * Force a disconnect
881 */
882void rds_conn_path_drop(struct rds_conn_path *cp, bool destroy)
883{
884	atomic_set(&cp->cp_state, RDS_CONN_ERROR);
885
886	rcu_read_lock();
887	if (!destroy && rds_destroy_pending(cp->cp_conn)) {
888		rcu_read_unlock();
889		return;
890	}
891	queue_work(rds_wq, &cp->cp_down_w);
892	rcu_read_unlock();
893}
894EXPORT_SYMBOL_GPL(rds_conn_path_drop);
895
896void rds_conn_drop(struct rds_connection *conn)
897{
898	WARN_ON(conn->c_trans->t_mp_capable);
899	rds_conn_path_drop(&conn->c_path[0], false);
900}
901EXPORT_SYMBOL_GPL(rds_conn_drop);
902
903/*
904 * If the connection is down, trigger a connect. We may have scheduled a
905 * delayed reconnect however - in this case we should not interfere.
906 */
907void rds_conn_path_connect_if_down(struct rds_conn_path *cp)
908{
909	rcu_read_lock();
910	if (rds_destroy_pending(cp->cp_conn)) {
911		rcu_read_unlock();
912		return;
913	}
914	if (rds_conn_path_state(cp) == RDS_CONN_DOWN &&
915	    !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
916		queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
917	rcu_read_unlock();
918}
919EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down);
920
921/* Check connectivity of all paths
922 */
923void rds_check_all_paths(struct rds_connection *conn)
924{
925	int i = 0;
926
927	do {
928		rds_conn_path_connect_if_down(&conn->c_path[i]);
929	} while (++i < conn->c_npaths);
930}
931
932void rds_conn_connect_if_down(struct rds_connection *conn)
933{
934	WARN_ON(conn->c_trans->t_mp_capable);
935	rds_conn_path_connect_if_down(&conn->c_path[0]);
936}
937EXPORT_SYMBOL_GPL(rds_conn_connect_if_down);
938
939void
940__rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...)
941{
942	va_list ap;
943
944	va_start(ap, fmt);
945	vprintk(fmt, ap);
946	va_end(ap);
947
948	rds_conn_path_drop(cp, false);
949}