2 * Copyright (c) 2006, 2018 Oracle and/or its affiliates. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
33 #include <linux/kernel.h>
34 #include <linux/list.h>
35 #include <linux/slab.h>
36 #include <linux/export.h>
38 #include <net/inet6_hashtables.h>
39 #include <net/addrconf.h>
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)
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;
54 static struct hlist_head *rds_conn_bucket(const struct in6_addr *laddr,
55 const struct in6_addr *faddr)
57 static u32 rds6_hash_secret __read_mostly;
58 static u32 rds_hash_secret __read_mostly;
60 u32 lhash, fhash, hash;
62 net_get_random_once(&rds_hash_secret, sizeof(rds_hash_secret));
63 net_get_random_once(&rds6_hash_secret, sizeof(rds6_hash_secret));
65 lhash = (__force u32)laddr->s6_addr32[3];
66 #if IS_ENABLED(CONFIG_IPV6)
67 fhash = __ipv6_addr_jhash(faddr, rds6_hash_secret);
69 fhash = (__force u32)faddr->s6_addr32[3];
71 hash = __inet_ehashfn(lhash, 0, fhash, 0, rds_hash_secret);
73 return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
76 #define rds_conn_info_set(var, test, suffix) do { \
78 var |= RDS_INFO_CONNECTION_FLAG_##suffix; \
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,
89 struct rds_connection *conn, *ret = NULL;
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 net == rds_conn_net(conn) &&
96 conn->c_dev_if == dev_if) {
101 rdsdebug("returning conn %p for %pI6c -> %pI6c\n", ret,
107 * This is called by transports as they're bringing down a connection.
108 * It clears partial message state so that the transport can start sending
109 * and receiving over this connection again in the future. It is up to
110 * the transport to have serialized this call with its send and recv.
112 static void rds_conn_path_reset(struct rds_conn_path *cp)
114 struct rds_connection *conn = cp->cp_conn;
116 rdsdebug("connection %pI6c to %pI6c reset\n",
117 &conn->c_laddr, &conn->c_faddr);
119 rds_stats_inc(s_conn_reset);
120 rds_send_path_reset(cp);
123 /* Do not clear next_rx_seq here, else we cannot distinguish
124 * retransmitted packets from new packets, and will hand all
125 * of them to the application. That is not consistent with the
126 * reliability guarantees of RDS. */
129 static void __rds_conn_path_init(struct rds_connection *conn,
130 struct rds_conn_path *cp, bool is_outgoing)
132 spin_lock_init(&cp->cp_lock);
133 cp->cp_next_tx_seq = 1;
134 init_waitqueue_head(&cp->cp_waitq);
135 INIT_LIST_HEAD(&cp->cp_send_queue);
136 INIT_LIST_HEAD(&cp->cp_retrans);
139 atomic_set(&cp->cp_state, RDS_CONN_DOWN);
141 cp->cp_reconnect_jiffies = 0;
142 cp->cp_conn->c_proposed_version = RDS_PROTOCOL_VERSION;
143 INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker);
144 INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker);
145 INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker);
146 INIT_WORK(&cp->cp_down_w, rds_shutdown_worker);
147 mutex_init(&cp->cp_cm_lock);
152 * There is only every one 'conn' for a given pair of addresses in the
153 * system at a time. They contain messages to be retransmitted and so
154 * span the lifetime of the actual underlying transport connections.
156 * For now they are not garbage collected once they're created. They
157 * are torn down as the module is removed, if ever.
159 static struct rds_connection *__rds_conn_create(struct net *net,
160 const struct in6_addr *laddr,
161 const struct in6_addr *faddr,
162 struct rds_transport *trans,
167 struct rds_connection *conn, *parent = NULL;
168 struct hlist_head *head = rds_conn_bucket(laddr, faddr);
169 struct rds_transport *loop_trans;
172 int npaths = (trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
175 conn = rds_conn_lookup(net, head, laddr, faddr, trans, dev_if);
178 conn->c_trans != &rds_loop_transport &&
179 ipv6_addr_equal(laddr, faddr) &&
181 /* This is a looped back IB connection, and we're
182 * called by the code handling the incoming connect.
183 * We need a second connection object into which we
184 * can stick the other QP. */
186 conn = parent->c_passive;
192 conn = kmem_cache_zalloc(rds_conn_slab, gfp);
194 conn = ERR_PTR(-ENOMEM);
197 conn->c_path = kcalloc(npaths, sizeof(struct rds_conn_path), gfp);
199 kmem_cache_free(rds_conn_slab, conn);
200 conn = ERR_PTR(-ENOMEM);
204 INIT_HLIST_NODE(&conn->c_hash_node);
205 conn->c_laddr = *laddr;
206 conn->c_isv6 = !ipv6_addr_v4mapped(laddr);
207 conn->c_faddr = *faddr;
208 conn->c_dev_if = dev_if;
210 #if IS_ENABLED(CONFIG_IPV6)
211 /* If the local address is link local, set c_bound_if to be the
212 * index used for this connection. Otherwise, set it to 0 as
213 * the socket is not bound to an interface. c_bound_if is used
214 * to look up a socket when a packet is received
216 if (ipv6_addr_type(laddr) & IPV6_ADDR_LINKLOCAL)
217 conn->c_bound_if = dev_if;
220 conn->c_bound_if = 0;
222 rds_conn_net_set(conn, net);
224 ret = rds_cong_get_maps(conn);
227 kmem_cache_free(rds_conn_slab, conn);
233 * This is where a connection becomes loopback. If *any* RDS sockets
234 * can bind to the destination address then we'd rather the messages
235 * flow through loopback rather than either transport.
237 loop_trans = rds_trans_get_preferred(net, faddr, conn->c_dev_if);
239 rds_trans_put(loop_trans);
240 conn->c_loopback = 1;
241 if (is_outgoing && trans->t_prefer_loopback) {
242 /* "outgoing" connection - and the transport
243 * says it wants the connection handled by the
244 * loopback transport. This is what TCP does.
246 trans = &rds_loop_transport;
250 conn->c_trans = trans;
252 init_waitqueue_head(&conn->c_hs_waitq);
253 for (i = 0; i < npaths; i++) {
254 __rds_conn_path_init(conn, &conn->c_path[i],
256 conn->c_path[i].cp_index = i;
259 if (rds_destroy_pending(conn))
262 ret = trans->conn_alloc(conn, GFP_ATOMIC);
266 kmem_cache_free(rds_conn_slab, conn);
271 rdsdebug("allocated conn %p for %pI6c -> %pI6c over %s %s\n",
273 strnlen(trans->t_name, sizeof(trans->t_name)) ?
274 trans->t_name : "[unknown]", is_outgoing ? "(outgoing)" : "");
277 * Since we ran without holding the conn lock, someone could
278 * have created the same conn (either normal or passive) in the
279 * interim. We check while holding the lock. If we won, we complete
280 * init and return our conn. If we lost, we rollback and return the
283 spin_lock_irqsave(&rds_conn_lock, flags);
285 /* Creating passive conn */
286 if (parent->c_passive) {
287 trans->conn_free(conn->c_path[0].cp_transport_data);
289 kmem_cache_free(rds_conn_slab, conn);
290 conn = parent->c_passive;
292 parent->c_passive = conn;
293 rds_cong_add_conn(conn);
297 /* Creating normal conn */
298 struct rds_connection *found;
300 found = rds_conn_lookup(net, head, laddr, faddr, trans,
303 struct rds_conn_path *cp;
306 for (i = 0; i < npaths; i++) {
307 cp = &conn->c_path[i];
308 /* The ->conn_alloc invocation may have
309 * allocated resource for all paths, so all
310 * of them may have to be freed here.
312 if (cp->cp_transport_data)
313 trans->conn_free(cp->cp_transport_data);
316 kmem_cache_free(rds_conn_slab, conn);
319 conn->c_my_gen_num = rds_gen_num;
320 conn->c_peer_gen_num = 0;
321 hlist_add_head_rcu(&conn->c_hash_node, head);
322 rds_cong_add_conn(conn);
326 spin_unlock_irqrestore(&rds_conn_lock, flags);
333 struct rds_connection *rds_conn_create(struct net *net,
334 const struct in6_addr *laddr,
335 const struct in6_addr *faddr,
336 struct rds_transport *trans, gfp_t gfp,
339 return __rds_conn_create(net, laddr, faddr, trans, gfp, 0, dev_if);
341 EXPORT_SYMBOL_GPL(rds_conn_create);
343 struct rds_connection *rds_conn_create_outgoing(struct net *net,
344 const struct in6_addr *laddr,
345 const struct in6_addr *faddr,
346 struct rds_transport *trans,
347 gfp_t gfp, int dev_if)
349 return __rds_conn_create(net, laddr, faddr, trans, gfp, 1, dev_if);
351 EXPORT_SYMBOL_GPL(rds_conn_create_outgoing);
353 void rds_conn_shutdown(struct rds_conn_path *cp)
355 struct rds_connection *conn = cp->cp_conn;
357 /* shut it down unless it's down already */
358 if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) {
360 * Quiesce the connection mgmt handlers before we start tearing
361 * things down. We don't hold the mutex for the entire
362 * duration of the shutdown operation, else we may be
363 * deadlocking with the CM handler. Instead, the CM event
364 * handler is supposed to check for state DISCONNECTING
366 mutex_lock(&cp->cp_cm_lock);
367 if (!rds_conn_path_transition(cp, RDS_CONN_UP,
368 RDS_CONN_DISCONNECTING) &&
369 !rds_conn_path_transition(cp, RDS_CONN_ERROR,
370 RDS_CONN_DISCONNECTING)) {
371 rds_conn_path_error(cp,
372 "shutdown called in state %d\n",
373 atomic_read(&cp->cp_state));
374 mutex_unlock(&cp->cp_cm_lock);
377 mutex_unlock(&cp->cp_cm_lock);
379 wait_event(cp->cp_waitq,
380 !test_bit(RDS_IN_XMIT, &cp->cp_flags));
381 wait_event(cp->cp_waitq,
382 !test_bit(RDS_RECV_REFILL, &cp->cp_flags));
384 conn->c_trans->conn_path_shutdown(cp);
385 rds_conn_path_reset(cp);
387 if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING,
389 !rds_conn_path_transition(cp, RDS_CONN_ERROR,
391 /* This can happen - eg when we're in the middle of tearing
392 * down the connection, and someone unloads the rds module.
393 * Quite reproducible with loopback connections.
396 * Note that this also happens with rds-tcp because
397 * we could have triggered rds_conn_path_drop in irq
398 * mode from rds_tcp_state change on the receipt of
399 * a FIN, thus we need to recheck for RDS_CONN_ERROR
402 rds_conn_path_error(cp, "%s: failed to transition "
403 "to state DOWN, current state "
405 atomic_read(&cp->cp_state));
410 /* Then reconnect if it's still live.
411 * The passive side of an IB loopback connection is never added
412 * to the conn hash, so we never trigger a reconnect on this
413 * conn - the reconnect is always triggered by the active peer. */
414 cancel_delayed_work_sync(&cp->cp_conn_w);
416 if (!hlist_unhashed(&conn->c_hash_node)) {
418 rds_queue_reconnect(cp);
424 /* destroy a single rds_conn_path. rds_conn_destroy() iterates over
425 * all paths using rds_conn_path_destroy()
427 static void rds_conn_path_destroy(struct rds_conn_path *cp)
429 struct rds_message *rm, *rtmp;
431 if (!cp->cp_transport_data)
434 /* make sure lingering queued work won't try to ref the conn */
435 cancel_delayed_work_sync(&cp->cp_send_w);
436 cancel_delayed_work_sync(&cp->cp_recv_w);
438 rds_conn_path_drop(cp, true);
439 flush_work(&cp->cp_down_w);
441 /* tear down queued messages */
442 list_for_each_entry_safe(rm, rtmp,
445 list_del_init(&rm->m_conn_item);
446 BUG_ON(!list_empty(&rm->m_sock_item));
450 rds_message_put(cp->cp_xmit_rm);
452 WARN_ON(delayed_work_pending(&cp->cp_send_w));
453 WARN_ON(delayed_work_pending(&cp->cp_recv_w));
454 WARN_ON(delayed_work_pending(&cp->cp_conn_w));
455 WARN_ON(work_pending(&cp->cp_down_w));
457 cp->cp_conn->c_trans->conn_free(cp->cp_transport_data);
461 * Stop and free a connection.
463 * This can only be used in very limited circumstances. It assumes that once
464 * the conn has been shutdown that no one else is referencing the connection.
465 * We can only ensure this in the rmmod path in the current code.
467 void rds_conn_destroy(struct rds_connection *conn)
471 struct rds_conn_path *cp;
472 int npaths = (conn->c_trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
474 rdsdebug("freeing conn %p for %pI4 -> "
475 "%pI4\n", conn, &conn->c_laddr,
478 /* Ensure conn will not be scheduled for reconnect */
479 spin_lock_irq(&rds_conn_lock);
480 hlist_del_init_rcu(&conn->c_hash_node);
481 spin_unlock_irq(&rds_conn_lock);
484 /* shut the connection down */
485 for (i = 0; i < npaths; i++) {
486 cp = &conn->c_path[i];
487 rds_conn_path_destroy(cp);
488 BUG_ON(!list_empty(&cp->cp_retrans));
492 * The congestion maps aren't freed up here. They're
493 * freed by rds_cong_exit() after all the connections
496 rds_cong_remove_conn(conn);
499 kmem_cache_free(rds_conn_slab, conn);
501 spin_lock_irqsave(&rds_conn_lock, flags);
503 spin_unlock_irqrestore(&rds_conn_lock, flags);
505 EXPORT_SYMBOL_GPL(rds_conn_destroy);
507 static void __rds_inc_msg_cp(struct rds_incoming *inc,
508 struct rds_info_iterator *iter,
509 void *saddr, void *daddr, int flip, bool isv6)
511 #if IS_ENABLED(CONFIG_IPV6)
513 rds6_inc_info_copy(inc, iter, saddr, daddr, flip);
516 rds_inc_info_copy(inc, iter, *(__be32 *)saddr,
517 *(__be32 *)daddr, flip);
520 static void rds_conn_message_info_cmn(struct socket *sock, unsigned int len,
521 struct rds_info_iterator *iter,
522 struct rds_info_lengths *lens,
523 int want_send, bool isv6)
525 struct hlist_head *head;
526 struct list_head *list;
527 struct rds_connection *conn;
528 struct rds_message *rm;
529 unsigned int total = 0;
535 len /= sizeof(struct rds6_info_message);
537 len /= sizeof(struct rds_info_message);
541 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
543 hlist_for_each_entry_rcu(conn, head, c_hash_node) {
544 struct rds_conn_path *cp;
547 if (!isv6 && conn->c_isv6)
550 npaths = (conn->c_trans->t_mp_capable ?
551 RDS_MPATH_WORKERS : 1);
553 for (j = 0; j < npaths; j++) {
554 cp = &conn->c_path[j];
556 list = &cp->cp_send_queue;
558 list = &cp->cp_retrans;
560 spin_lock_irqsave(&cp->cp_lock, flags);
562 /* XXX too lazy to maintain counts.. */
563 list_for_each_entry(rm, list, m_conn_item) {
566 __rds_inc_msg_cp(&rm->m_inc,
573 spin_unlock_irqrestore(&cp->cp_lock, flags);
581 lens->each = sizeof(struct rds6_info_message);
583 lens->each = sizeof(struct rds_info_message);
586 static void rds_conn_message_info(struct socket *sock, unsigned int len,
587 struct rds_info_iterator *iter,
588 struct rds_info_lengths *lens,
591 rds_conn_message_info_cmn(sock, len, iter, lens, want_send, false);
594 #if IS_ENABLED(CONFIG_IPV6)
595 static void rds6_conn_message_info(struct socket *sock, unsigned int len,
596 struct rds_info_iterator *iter,
597 struct rds_info_lengths *lens,
600 rds_conn_message_info_cmn(sock, len, iter, lens, want_send, true);
604 static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
605 struct rds_info_iterator *iter,
606 struct rds_info_lengths *lens)
608 rds_conn_message_info(sock, len, iter, lens, 1);
611 #if IS_ENABLED(CONFIG_IPV6)
612 static void rds6_conn_message_info_send(struct socket *sock, unsigned int len,
613 struct rds_info_iterator *iter,
614 struct rds_info_lengths *lens)
616 rds6_conn_message_info(sock, len, iter, lens, 1);
620 static void rds_conn_message_info_retrans(struct socket *sock,
622 struct rds_info_iterator *iter,
623 struct rds_info_lengths *lens)
625 rds_conn_message_info(sock, len, iter, lens, 0);
628 #if IS_ENABLED(CONFIG_IPV6)
629 static void rds6_conn_message_info_retrans(struct socket *sock,
631 struct rds_info_iterator *iter,
632 struct rds_info_lengths *lens)
634 rds6_conn_message_info(sock, len, iter, lens, 0);
638 void rds_for_each_conn_info(struct socket *sock, unsigned int len,
639 struct rds_info_iterator *iter,
640 struct rds_info_lengths *lens,
641 int (*visitor)(struct rds_connection *, void *),
645 struct hlist_head *head;
646 struct rds_connection *conn;
652 lens->each = item_len;
654 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
656 hlist_for_each_entry_rcu(conn, head, c_hash_node) {
658 /* XXX no c_lock usage.. */
659 if (!visitor(conn, buffer))
662 /* We copy as much as we can fit in the buffer,
663 * but we count all items so that the caller
664 * can resize the buffer. */
665 if (len >= item_len) {
666 rds_info_copy(iter, buffer, item_len);
674 EXPORT_SYMBOL_GPL(rds_for_each_conn_info);
676 static void rds_walk_conn_path_info(struct socket *sock, unsigned int len,
677 struct rds_info_iterator *iter,
678 struct rds_info_lengths *lens,
679 int (*visitor)(struct rds_conn_path *, void *),
683 struct hlist_head *head;
684 struct rds_connection *conn;
690 lens->each = item_len;
692 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
694 hlist_for_each_entry_rcu(conn, head, c_hash_node) {
695 struct rds_conn_path *cp;
697 /* XXX We only copy the information from the first
698 * path for now. The problem is that if there are
699 * more than one underlying paths, we cannot report
700 * information of all of them using the existing
701 * API. For example, there is only one next_tx_seq,
702 * which path's next_tx_seq should we report? It is
703 * a bug in the design of MPRDS.
707 /* XXX no cp_lock usage.. */
708 if (!visitor(cp, buffer))
711 /* We copy as much as we can fit in the buffer,
712 * but we count all items so that the caller
713 * can resize the buffer.
715 if (len >= item_len) {
716 rds_info_copy(iter, buffer, item_len);
725 static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
727 struct rds_info_connection *cinfo = buffer;
728 struct rds_connection *conn = cp->cp_conn;
733 cinfo->next_tx_seq = cp->cp_next_tx_seq;
734 cinfo->next_rx_seq = cp->cp_next_rx_seq;
735 cinfo->laddr = conn->c_laddr.s6_addr32[3];
736 cinfo->faddr = conn->c_faddr.s6_addr32[3];
737 strncpy(cinfo->transport, conn->c_trans->t_name,
738 sizeof(cinfo->transport));
741 rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
743 /* XXX Future: return the state rather than these funky bits */
744 rds_conn_info_set(cinfo->flags,
745 atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
747 rds_conn_info_set(cinfo->flags,
748 atomic_read(&cp->cp_state) == RDS_CONN_UP,
753 #if IS_ENABLED(CONFIG_IPV6)
754 static int rds6_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
756 struct rds6_info_connection *cinfo6 = buffer;
757 struct rds_connection *conn = cp->cp_conn;
759 cinfo6->next_tx_seq = cp->cp_next_tx_seq;
760 cinfo6->next_rx_seq = cp->cp_next_rx_seq;
761 cinfo6->laddr = conn->c_laddr;
762 cinfo6->faddr = conn->c_faddr;
763 strncpy(cinfo6->transport, conn->c_trans->t_name,
764 sizeof(cinfo6->transport));
767 rds_conn_info_set(cinfo6->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
769 /* XXX Future: return the state rather than these funky bits */
770 rds_conn_info_set(cinfo6->flags,
771 atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
773 rds_conn_info_set(cinfo6->flags,
774 atomic_read(&cp->cp_state) == RDS_CONN_UP,
776 /* Just return 1 as there is no error case. This is a helper function
777 * for rds_walk_conn_path_info() and it wants a return value.
783 static void rds_conn_info(struct socket *sock, unsigned int len,
784 struct rds_info_iterator *iter,
785 struct rds_info_lengths *lens)
787 u64 buffer[(sizeof(struct rds_info_connection) + 7) / 8];
789 rds_walk_conn_path_info(sock, len, iter, lens,
790 rds_conn_info_visitor,
792 sizeof(struct rds_info_connection));
795 #if IS_ENABLED(CONFIG_IPV6)
796 static void rds6_conn_info(struct socket *sock, unsigned int len,
797 struct rds_info_iterator *iter,
798 struct rds_info_lengths *lens)
800 u64 buffer[(sizeof(struct rds6_info_connection) + 7) / 8];
802 rds_walk_conn_path_info(sock, len, iter, lens,
803 rds6_conn_info_visitor,
805 sizeof(struct rds6_info_connection));
809 int rds_conn_init(void)
813 ret = rds_loop_net_init(); /* register pernet callback */
817 rds_conn_slab = kmem_cache_create("rds_connection",
818 sizeof(struct rds_connection),
820 if (!rds_conn_slab) {
825 rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
826 rds_info_register_func(RDS_INFO_SEND_MESSAGES,
827 rds_conn_message_info_send);
828 rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
829 rds_conn_message_info_retrans);
830 #if IS_ENABLED(CONFIG_IPV6)
831 rds_info_register_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
832 rds_info_register_func(RDS6_INFO_SEND_MESSAGES,
833 rds6_conn_message_info_send);
834 rds_info_register_func(RDS6_INFO_RETRANS_MESSAGES,
835 rds6_conn_message_info_retrans);
840 void rds_conn_exit(void)
842 rds_loop_net_exit(); /* unregister pernet callback */
845 WARN_ON(!hlist_empty(rds_conn_hash));
847 kmem_cache_destroy(rds_conn_slab);
849 rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
850 rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
851 rds_conn_message_info_send);
852 rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
853 rds_conn_message_info_retrans);
854 #if IS_ENABLED(CONFIG_IPV6)
855 rds_info_deregister_func(RDS6_INFO_CONNECTIONS, rds6_conn_info);
856 rds_info_deregister_func(RDS6_INFO_SEND_MESSAGES,
857 rds6_conn_message_info_send);
858 rds_info_deregister_func(RDS6_INFO_RETRANS_MESSAGES,
859 rds6_conn_message_info_retrans);
866 void rds_conn_path_drop(struct rds_conn_path *cp, bool destroy)
868 atomic_set(&cp->cp_state, RDS_CONN_ERROR);
871 if (!destroy && rds_destroy_pending(cp->cp_conn)) {
875 queue_work(rds_wq, &cp->cp_down_w);
878 EXPORT_SYMBOL_GPL(rds_conn_path_drop);
880 void rds_conn_drop(struct rds_connection *conn)
882 WARN_ON(conn->c_trans->t_mp_capable);
883 rds_conn_path_drop(&conn->c_path[0], false);
885 EXPORT_SYMBOL_GPL(rds_conn_drop);
888 * If the connection is down, trigger a connect. We may have scheduled a
889 * delayed reconnect however - in this case we should not interfere.
891 void rds_conn_path_connect_if_down(struct rds_conn_path *cp)
894 if (rds_destroy_pending(cp->cp_conn)) {
898 if (rds_conn_path_state(cp) == RDS_CONN_DOWN &&
899 !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
900 queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
903 EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down);
905 void rds_conn_connect_if_down(struct rds_connection *conn)
907 WARN_ON(conn->c_trans->t_mp_capable);
908 rds_conn_path_connect_if_down(&conn->c_path[0]);
910 EXPORT_SYMBOL_GPL(rds_conn_connect_if_down);
913 __rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...)
921 rds_conn_path_drop(cp, false);