1 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
5 * Copyright (C) 2004 Oracle. All rights reserved.
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public
9 * License as published by the Free Software Foundation; either
10 * version 2 of the License, or (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public
18 * License along with this program; if not, write to the
19 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
20 * Boston, MA 021110-1307, USA.
24 * Callers for this were originally written against a very simple synchronus
25 * API. This implementation reflects those simple callers. Some day I'm sure
26 * we'll need to move to a more robust posting/callback mechanism.
28 * Transmit calls pass in kernel virtual addresses and block copying this into
29 * the socket's tx buffers via a usual blocking sendmsg. They'll block waiting
30 * for a failed socket to timeout. TX callers can also pass in a poniter to an
31 * 'int' which gets filled with an errno off the wire in response to the
34 * Handlers for unsolicited messages are registered. Each socket has a page
35 * that incoming data is copied into. First the header, then the data.
36 * Handlers are called from only one thread with a reference to this per-socket
37 * page. This page is destroyed after the handler call, so it can't be
38 * referenced beyond the call. Handlers may block but are discouraged from
41 * Any framing errors (bad magic, large payload lengths) close a connection.
43 * Our sock_container holds the state we associate with a socket. It's current
44 * framing state is held there as well as the refcounting we do around when it
45 * is safe to tear down the socket. The socket is only finally torn down from
46 * the container when the container loses all of its references -- so as long
47 * as you hold a ref on the container you can trust that the socket is valid
48 * for use with kernel socket APIs.
50 * Connections are initiated between a pair of nodes when the node with the
51 * higher node number gets a heartbeat callback which indicates that the lower
52 * numbered node has started heartbeating. The lower numbered node is passive
53 * and only accepts the connection if the higher numbered node is heartbeating.
56 #include <linux/kernel.h>
57 #include <linux/sched/mm.h>
58 #include <linux/jiffies.h>
59 #include <linux/slab.h>
60 #include <linux/idr.h>
61 #include <linux/kref.h>
62 #include <linux/net.h>
63 #include <linux/export.h>
66 #include <linux/uaccess.h>
68 #include "heartbeat.h"
70 #include "nodemanager.h"
71 #define MLOG_MASK_PREFIX ML_TCP
75 #include "tcp_internal.h"
77 #define SC_NODEF_FMT "node %s (num %u) at %pI4:%u"
78 #define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num, \
79 &sc->sc_node->nd_ipv4_address, \
80 ntohs(sc->sc_node->nd_ipv4_port)
83 * In the following two log macros, the whitespace after the ',' just
84 * before ##args is intentional. Otherwise, gcc 2.95 will eat the
85 * previous token if args expands to nothing.
87 #define msglog(hdr, fmt, args...) do { \
88 typeof(hdr) __hdr = (hdr); \
89 mlog(ML_MSG, "[mag %u len %u typ %u stat %d sys_stat %d " \
90 "key %08x num %u] " fmt, \
91 be16_to_cpu(__hdr->magic), be16_to_cpu(__hdr->data_len), \
92 be16_to_cpu(__hdr->msg_type), be32_to_cpu(__hdr->status), \
93 be32_to_cpu(__hdr->sys_status), be32_to_cpu(__hdr->key), \
94 be32_to_cpu(__hdr->msg_num) , ##args); \
97 #define sclog(sc, fmt, args...) do { \
98 typeof(sc) __sc = (sc); \
99 mlog(ML_SOCKET, "[sc %p refs %d sock %p node %u page %p " \
100 "pg_off %zu] " fmt, __sc, \
101 kref_read(&__sc->sc_kref), __sc->sc_sock, \
102 __sc->sc_node->nd_num, __sc->sc_page, __sc->sc_page_off , \
106 static DEFINE_RWLOCK(o2net_handler_lock
);
107 static struct rb_root o2net_handler_tree
= RB_ROOT
;
109 static struct o2net_node o2net_nodes
[O2NM_MAX_NODES
];
111 /* XXX someday we'll need better accounting */
112 static struct socket
*o2net_listen_sock
;
115 * listen work is only queued by the listening socket callbacks on the
116 * o2net_wq. teardown detaches the callbacks before destroying the workqueue.
117 * quorum work is queued as sock containers are shutdown.. stop_listening
118 * tears down all the node's sock containers, preventing future shutdowns
119 * and queued quroum work, before canceling delayed quorum work and
120 * destroying the work queue.
122 static struct workqueue_struct
*o2net_wq
;
123 static struct work_struct o2net_listen_work
;
125 static struct o2hb_callback_func o2net_hb_up
, o2net_hb_down
;
126 #define O2NET_HB_PRI 0x1
128 static struct o2net_handshake
*o2net_hand
;
129 static struct o2net_msg
*o2net_keep_req
, *o2net_keep_resp
;
131 static int o2net_sys_err_translations
[O2NET_ERR_MAX
] =
132 {[O2NET_ERR_NONE
] = 0,
133 [O2NET_ERR_NO_HNDLR
] = -ENOPROTOOPT
,
134 [O2NET_ERR_OVERFLOW
] = -EOVERFLOW
,
135 [O2NET_ERR_DIED
] = -EHOSTDOWN
,};
137 /* can't quite avoid *all* internal declarations :/ */
138 static void o2net_sc_connect_completed(struct work_struct
*work
);
139 static void o2net_rx_until_empty(struct work_struct
*work
);
140 static void o2net_shutdown_sc(struct work_struct
*work
);
141 static void o2net_listen_data_ready(struct sock
*sk
);
142 static void o2net_sc_send_keep_req(struct work_struct
*work
);
143 static void o2net_idle_timer(unsigned long data
);
144 static void o2net_sc_postpone_idle(struct o2net_sock_container
*sc
);
145 static void o2net_sc_reset_idle_timer(struct o2net_sock_container
*sc
);
147 #ifdef CONFIG_DEBUG_FS
148 static void o2net_init_nst(struct o2net_send_tracking
*nst
, u32 msgtype
,
149 u32 msgkey
, struct task_struct
*task
, u8 node
)
151 INIT_LIST_HEAD(&nst
->st_net_debug_item
);
153 nst
->st_msg_type
= msgtype
;
154 nst
->st_msg_key
= msgkey
;
158 static inline void o2net_set_nst_sock_time(struct o2net_send_tracking
*nst
)
160 nst
->st_sock_time
= ktime_get();
163 static inline void o2net_set_nst_send_time(struct o2net_send_tracking
*nst
)
165 nst
->st_send_time
= ktime_get();
168 static inline void o2net_set_nst_status_time(struct o2net_send_tracking
*nst
)
170 nst
->st_status_time
= ktime_get();
173 static inline void o2net_set_nst_sock_container(struct o2net_send_tracking
*nst
,
174 struct o2net_sock_container
*sc
)
179 static inline void o2net_set_nst_msg_id(struct o2net_send_tracking
*nst
,
185 static inline void o2net_set_sock_timer(struct o2net_sock_container
*sc
)
187 sc
->sc_tv_timer
= ktime_get();
190 static inline void o2net_set_data_ready_time(struct o2net_sock_container
*sc
)
192 sc
->sc_tv_data_ready
= ktime_get();
195 static inline void o2net_set_advance_start_time(struct o2net_sock_container
*sc
)
197 sc
->sc_tv_advance_start
= ktime_get();
200 static inline void o2net_set_advance_stop_time(struct o2net_sock_container
*sc
)
202 sc
->sc_tv_advance_stop
= ktime_get();
205 static inline void o2net_set_func_start_time(struct o2net_sock_container
*sc
)
207 sc
->sc_tv_func_start
= ktime_get();
210 static inline void o2net_set_func_stop_time(struct o2net_sock_container
*sc
)
212 sc
->sc_tv_func_stop
= ktime_get();
215 #else /* CONFIG_DEBUG_FS */
216 # define o2net_init_nst(a, b, c, d, e)
217 # define o2net_set_nst_sock_time(a)
218 # define o2net_set_nst_send_time(a)
219 # define o2net_set_nst_status_time(a)
220 # define o2net_set_nst_sock_container(a, b)
221 # define o2net_set_nst_msg_id(a, b)
222 # define o2net_set_sock_timer(a)
223 # define o2net_set_data_ready_time(a)
224 # define o2net_set_advance_start_time(a)
225 # define o2net_set_advance_stop_time(a)
226 # define o2net_set_func_start_time(a)
227 # define o2net_set_func_stop_time(a)
228 #endif /* CONFIG_DEBUG_FS */
230 #ifdef CONFIG_OCFS2_FS_STATS
231 static ktime_t
o2net_get_func_run_time(struct o2net_sock_container
*sc
)
233 return ktime_sub(sc
->sc_tv_func_stop
, sc
->sc_tv_func_start
);
236 static void o2net_update_send_stats(struct o2net_send_tracking
*nst
,
237 struct o2net_sock_container
*sc
)
239 sc
->sc_tv_status_total
= ktime_add(sc
->sc_tv_status_total
,
240 ktime_sub(ktime_get(),
241 nst
->st_status_time
));
242 sc
->sc_tv_send_total
= ktime_add(sc
->sc_tv_send_total
,
243 ktime_sub(nst
->st_status_time
,
245 sc
->sc_tv_acquiry_total
= ktime_add(sc
->sc_tv_acquiry_total
,
246 ktime_sub(nst
->st_send_time
,
251 static void o2net_update_recv_stats(struct o2net_sock_container
*sc
)
253 sc
->sc_tv_process_total
= ktime_add(sc
->sc_tv_process_total
,
254 o2net_get_func_run_time(sc
));
260 # define o2net_update_send_stats(a, b)
262 # define o2net_update_recv_stats(sc)
264 #endif /* CONFIG_OCFS2_FS_STATS */
266 static inline unsigned int o2net_reconnect_delay(void)
268 return o2nm_single_cluster
->cl_reconnect_delay_ms
;
271 static inline unsigned int o2net_keepalive_delay(void)
273 return o2nm_single_cluster
->cl_keepalive_delay_ms
;
276 static inline unsigned int o2net_idle_timeout(void)
278 return o2nm_single_cluster
->cl_idle_timeout_ms
;
281 static inline int o2net_sys_err_to_errno(enum o2net_system_error err
)
284 BUG_ON(err
>= O2NET_ERR_MAX
);
285 trans
= o2net_sys_err_translations
[err
];
287 /* Just in case we mess up the translation table above */
288 BUG_ON(err
!= O2NET_ERR_NONE
&& trans
== 0);
292 static struct o2net_node
* o2net_nn_from_num(u8 node_num
)
294 BUG_ON(node_num
>= ARRAY_SIZE(o2net_nodes
));
295 return &o2net_nodes
[node_num
];
298 static u8
o2net_num_from_nn(struct o2net_node
*nn
)
301 return nn
- o2net_nodes
;
304 /* ------------------------------------------------------------ */
306 static int o2net_prep_nsw(struct o2net_node
*nn
, struct o2net_status_wait
*nsw
)
310 spin_lock(&nn
->nn_lock
);
311 ret
= idr_alloc(&nn
->nn_status_idr
, nsw
, 0, 0, GFP_ATOMIC
);
314 list_add_tail(&nsw
->ns_node_item
, &nn
->nn_status_list
);
316 spin_unlock(&nn
->nn_lock
);
320 init_waitqueue_head(&nsw
->ns_wq
);
321 nsw
->ns_sys_status
= O2NET_ERR_NONE
;
326 static void o2net_complete_nsw_locked(struct o2net_node
*nn
,
327 struct o2net_status_wait
*nsw
,
328 enum o2net_system_error sys_status
,
331 assert_spin_locked(&nn
->nn_lock
);
333 if (!list_empty(&nsw
->ns_node_item
)) {
334 list_del_init(&nsw
->ns_node_item
);
335 nsw
->ns_sys_status
= sys_status
;
336 nsw
->ns_status
= status
;
337 idr_remove(&nn
->nn_status_idr
, nsw
->ns_id
);
338 wake_up(&nsw
->ns_wq
);
342 static void o2net_complete_nsw(struct o2net_node
*nn
,
343 struct o2net_status_wait
*nsw
,
344 u64 id
, enum o2net_system_error sys_status
,
347 spin_lock(&nn
->nn_lock
);
352 nsw
= idr_find(&nn
->nn_status_idr
, id
);
357 o2net_complete_nsw_locked(nn
, nsw
, sys_status
, status
);
360 spin_unlock(&nn
->nn_lock
);
364 static void o2net_complete_nodes_nsw(struct o2net_node
*nn
)
366 struct o2net_status_wait
*nsw
, *tmp
;
367 unsigned int num_kills
= 0;
369 assert_spin_locked(&nn
->nn_lock
);
371 list_for_each_entry_safe(nsw
, tmp
, &nn
->nn_status_list
, ns_node_item
) {
372 o2net_complete_nsw_locked(nn
, nsw
, O2NET_ERR_DIED
, 0);
376 mlog(0, "completed %d messages for node %u\n", num_kills
,
377 o2net_num_from_nn(nn
));
380 static int o2net_nsw_completed(struct o2net_node
*nn
,
381 struct o2net_status_wait
*nsw
)
384 spin_lock(&nn
->nn_lock
);
385 completed
= list_empty(&nsw
->ns_node_item
);
386 spin_unlock(&nn
->nn_lock
);
390 /* ------------------------------------------------------------ */
392 static void sc_kref_release(struct kref
*kref
)
394 struct o2net_sock_container
*sc
= container_of(kref
,
395 struct o2net_sock_container
, sc_kref
);
396 BUG_ON(timer_pending(&sc
->sc_idle_timeout
));
398 sclog(sc
, "releasing\n");
401 sock_release(sc
->sc_sock
);
405 o2nm_undepend_item(&sc
->sc_node
->nd_item
);
406 o2nm_node_put(sc
->sc_node
);
409 o2net_debug_del_sc(sc
);
412 __free_page(sc
->sc_page
);
416 static void sc_put(struct o2net_sock_container
*sc
)
419 kref_put(&sc
->sc_kref
, sc_kref_release
);
421 static void sc_get(struct o2net_sock_container
*sc
)
424 kref_get(&sc
->sc_kref
);
426 static struct o2net_sock_container
*sc_alloc(struct o2nm_node
*node
)
428 struct o2net_sock_container
*sc
, *ret
= NULL
;
429 struct page
*page
= NULL
;
432 page
= alloc_page(GFP_NOFS
);
433 sc
= kzalloc(sizeof(*sc
), GFP_NOFS
);
434 if (sc
== NULL
|| page
== NULL
)
437 kref_init(&sc
->sc_kref
);
441 /* pin the node item of the remote node */
442 status
= o2nm_depend_item(&node
->nd_item
);
448 INIT_WORK(&sc
->sc_connect_work
, o2net_sc_connect_completed
);
449 INIT_WORK(&sc
->sc_rx_work
, o2net_rx_until_empty
);
450 INIT_WORK(&sc
->sc_shutdown_work
, o2net_shutdown_sc
);
451 INIT_DELAYED_WORK(&sc
->sc_keepalive_work
, o2net_sc_send_keep_req
);
453 init_timer(&sc
->sc_idle_timeout
);
454 sc
->sc_idle_timeout
.function
= o2net_idle_timer
;
455 sc
->sc_idle_timeout
.data
= (unsigned long)sc
;
457 sclog(sc
, "alloced\n");
461 o2net_debug_add_sc(sc
);
473 /* ------------------------------------------------------------ */
475 static void o2net_sc_queue_work(struct o2net_sock_container
*sc
,
476 struct work_struct
*work
)
479 if (!queue_work(o2net_wq
, work
))
482 static void o2net_sc_queue_delayed_work(struct o2net_sock_container
*sc
,
483 struct delayed_work
*work
,
487 if (!queue_delayed_work(o2net_wq
, work
, delay
))
490 static void o2net_sc_cancel_delayed_work(struct o2net_sock_container
*sc
,
491 struct delayed_work
*work
)
493 if (cancel_delayed_work(work
))
497 static atomic_t o2net_connected_peers
= ATOMIC_INIT(0);
499 int o2net_num_connected_peers(void)
501 return atomic_read(&o2net_connected_peers
);
504 static void o2net_set_nn_state(struct o2net_node
*nn
,
505 struct o2net_sock_container
*sc
,
506 unsigned valid
, int err
)
508 int was_valid
= nn
->nn_sc_valid
;
509 int was_err
= nn
->nn_persistent_error
;
510 struct o2net_sock_container
*old_sc
= nn
->nn_sc
;
512 assert_spin_locked(&nn
->nn_lock
);
515 atomic_dec(&o2net_connected_peers
);
516 else if (!old_sc
&& sc
)
517 atomic_inc(&o2net_connected_peers
);
519 /* the node num comparison and single connect/accept path should stop
520 * an non-null sc from being overwritten with another */
521 BUG_ON(sc
&& nn
->nn_sc
&& nn
->nn_sc
!= sc
);
522 mlog_bug_on_msg(err
&& valid
, "err %d valid %u\n", err
, valid
);
523 mlog_bug_on_msg(valid
&& !sc
, "valid %u sc %p\n", valid
, sc
);
525 if (was_valid
&& !valid
&& err
== 0)
528 mlog(ML_CONN
, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n",
529 o2net_num_from_nn(nn
), nn
->nn_sc
, sc
, nn
->nn_sc_valid
, valid
,
530 nn
->nn_persistent_error
, err
);
533 nn
->nn_sc_valid
= valid
? 1 : 0;
534 nn
->nn_persistent_error
= err
;
536 /* mirrors o2net_tx_can_proceed() */
537 if (nn
->nn_persistent_error
|| nn
->nn_sc_valid
)
538 wake_up(&nn
->nn_sc_wq
);
540 if (was_valid
&& !was_err
&& nn
->nn_persistent_error
) {
541 o2quo_conn_err(o2net_num_from_nn(nn
));
542 queue_delayed_work(o2net_wq
, &nn
->nn_still_up
,
543 msecs_to_jiffies(O2NET_QUORUM_DELAY_MS
));
546 if (was_valid
&& !valid
) {
548 printk(KERN_NOTICE
"o2net: No longer connected to "
549 SC_NODEF_FMT
"\n", SC_NODEF_ARGS(old_sc
));
550 o2net_complete_nodes_nsw(nn
);
553 if (!was_valid
&& valid
) {
554 o2quo_conn_up(o2net_num_from_nn(nn
));
555 cancel_delayed_work(&nn
->nn_connect_expired
);
556 printk(KERN_NOTICE
"o2net: %s " SC_NODEF_FMT
"\n",
557 o2nm_this_node() > sc
->sc_node
->nd_num
?
558 "Connected to" : "Accepted connection from",
562 /* trigger the connecting worker func as long as we're not valid,
563 * it will back off if it shouldn't connect. This can be called
564 * from node config teardown and so needs to be careful about
565 * the work queue actually being up. */
566 if (!valid
&& o2net_wq
) {
568 /* delay if we're within a RECONNECT_DELAY of the
570 delay
= (nn
->nn_last_connect_attempt
+
571 msecs_to_jiffies(o2net_reconnect_delay()))
573 if (delay
> msecs_to_jiffies(o2net_reconnect_delay()))
575 mlog(ML_CONN
, "queueing conn attempt in %lu jiffies\n", delay
);
576 queue_delayed_work(o2net_wq
, &nn
->nn_connect_work
, delay
);
579 * Delay the expired work after idle timeout.
581 * We might have lots of failed connection attempts that run
582 * through here but we only cancel the connect_expired work when
583 * a connection attempt succeeds. So only the first enqueue of
584 * the connect_expired work will do anything. The rest will see
585 * that it's already queued and do nothing.
587 delay
+= msecs_to_jiffies(o2net_idle_timeout());
588 queue_delayed_work(o2net_wq
, &nn
->nn_connect_expired
, delay
);
591 /* keep track of the nn's sc ref for the caller */
592 if ((old_sc
== NULL
) && sc
)
594 if (old_sc
&& (old_sc
!= sc
)) {
595 o2net_sc_queue_work(old_sc
, &old_sc
->sc_shutdown_work
);
600 /* see o2net_register_callbacks() */
601 static void o2net_data_ready(struct sock
*sk
)
603 void (*ready
)(struct sock
*sk
);
604 struct o2net_sock_container
*sc
;
606 read_lock_bh(&sk
->sk_callback_lock
);
607 sc
= sk
->sk_user_data
;
609 sclog(sc
, "data_ready hit\n");
610 o2net_set_data_ready_time(sc
);
611 o2net_sc_queue_work(sc
, &sc
->sc_rx_work
);
612 ready
= sc
->sc_data_ready
;
614 ready
= sk
->sk_data_ready
;
616 read_unlock_bh(&sk
->sk_callback_lock
);
621 /* see o2net_register_callbacks() */
622 static void o2net_state_change(struct sock
*sk
)
624 void (*state_change
)(struct sock
*sk
);
625 struct o2net_sock_container
*sc
;
627 read_lock_bh(&sk
->sk_callback_lock
);
628 sc
= sk
->sk_user_data
;
630 state_change
= sk
->sk_state_change
;
634 sclog(sc
, "state_change to %d\n", sk
->sk_state
);
636 state_change
= sc
->sc_state_change
;
638 switch(sk
->sk_state
) {
639 /* ignore connecting sockets as they make progress */
643 case TCP_ESTABLISHED
:
644 o2net_sc_queue_work(sc
, &sc
->sc_connect_work
);
647 printk(KERN_INFO
"o2net: Connection to " SC_NODEF_FMT
648 " shutdown, state %d\n",
649 SC_NODEF_ARGS(sc
), sk
->sk_state
);
650 o2net_sc_queue_work(sc
, &sc
->sc_shutdown_work
);
654 read_unlock_bh(&sk
->sk_callback_lock
);
659 * we register callbacks so we can queue work on events before calling
660 * the original callbacks. our callbacks our careful to test user_data
661 * to discover when they've reaced with o2net_unregister_callbacks().
663 static void o2net_register_callbacks(struct sock
*sk
,
664 struct o2net_sock_container
*sc
)
666 write_lock_bh(&sk
->sk_callback_lock
);
668 /* accepted sockets inherit the old listen socket data ready */
669 if (sk
->sk_data_ready
== o2net_listen_data_ready
) {
670 sk
->sk_data_ready
= sk
->sk_user_data
;
671 sk
->sk_user_data
= NULL
;
674 BUG_ON(sk
->sk_user_data
!= NULL
);
675 sk
->sk_user_data
= sc
;
678 sc
->sc_data_ready
= sk
->sk_data_ready
;
679 sc
->sc_state_change
= sk
->sk_state_change
;
680 sk
->sk_data_ready
= o2net_data_ready
;
681 sk
->sk_state_change
= o2net_state_change
;
683 mutex_init(&sc
->sc_send_lock
);
685 write_unlock_bh(&sk
->sk_callback_lock
);
688 static int o2net_unregister_callbacks(struct sock
*sk
,
689 struct o2net_sock_container
*sc
)
693 write_lock_bh(&sk
->sk_callback_lock
);
694 if (sk
->sk_user_data
== sc
) {
696 sk
->sk_user_data
= NULL
;
697 sk
->sk_data_ready
= sc
->sc_data_ready
;
698 sk
->sk_state_change
= sc
->sc_state_change
;
700 write_unlock_bh(&sk
->sk_callback_lock
);
706 * this is a little helper that is called by callers who have seen a problem
707 * with an sc and want to detach it from the nn if someone already hasn't beat
708 * them to it. if an error is given then the shutdown will be persistent
709 * and pending transmits will be canceled.
711 static void o2net_ensure_shutdown(struct o2net_node
*nn
,
712 struct o2net_sock_container
*sc
,
715 spin_lock(&nn
->nn_lock
);
717 o2net_set_nn_state(nn
, NULL
, 0, err
);
718 spin_unlock(&nn
->nn_lock
);
722 * This work queue function performs the blocking parts of socket shutdown. A
723 * few paths lead here. set_nn_state will trigger this callback if it sees an
724 * sc detached from the nn. state_change will also trigger this callback
725 * directly when it sees errors. In that case we need to call set_nn_state
726 * ourselves as state_change couldn't get the nn_lock and call set_nn_state
729 static void o2net_shutdown_sc(struct work_struct
*work
)
731 struct o2net_sock_container
*sc
=
732 container_of(work
, struct o2net_sock_container
,
734 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
736 sclog(sc
, "shutting down\n");
738 /* drop the callbacks ref and call shutdown only once */
739 if (o2net_unregister_callbacks(sc
->sc_sock
->sk
, sc
)) {
740 /* we shouldn't flush as we're in the thread, the
741 * races with pending sc work structs are harmless */
742 del_timer_sync(&sc
->sc_idle_timeout
);
743 o2net_sc_cancel_delayed_work(sc
, &sc
->sc_keepalive_work
);
745 kernel_sock_shutdown(sc
->sc_sock
, SHUT_RDWR
);
748 /* not fatal so failed connects before the other guy has our
749 * heartbeat can be retried */
750 o2net_ensure_shutdown(nn
, sc
, 0);
754 /* ------------------------------------------------------------ */
756 static int o2net_handler_cmp(struct o2net_msg_handler
*nmh
, u32 msg_type
,
759 int ret
= memcmp(&nmh
->nh_key
, &key
, sizeof(key
));
762 ret
= memcmp(&nmh
->nh_msg_type
, &msg_type
, sizeof(msg_type
));
767 static struct o2net_msg_handler
*
768 o2net_handler_tree_lookup(u32 msg_type
, u32 key
, struct rb_node
***ret_p
,
769 struct rb_node
**ret_parent
)
771 struct rb_node
**p
= &o2net_handler_tree
.rb_node
;
772 struct rb_node
*parent
= NULL
;
773 struct o2net_msg_handler
*nmh
, *ret
= NULL
;
778 nmh
= rb_entry(parent
, struct o2net_msg_handler
, nh_node
);
779 cmp
= o2net_handler_cmp(nmh
, msg_type
, key
);
793 if (ret_parent
!= NULL
)
794 *ret_parent
= parent
;
799 static void o2net_handler_kref_release(struct kref
*kref
)
801 struct o2net_msg_handler
*nmh
;
802 nmh
= container_of(kref
, struct o2net_msg_handler
, nh_kref
);
807 static void o2net_handler_put(struct o2net_msg_handler
*nmh
)
809 kref_put(&nmh
->nh_kref
, o2net_handler_kref_release
);
812 /* max_len is protection for the handler func. incoming messages won't
813 * be given to the handler if their payload is longer than the max. */
814 int o2net_register_handler(u32 msg_type
, u32 key
, u32 max_len
,
815 o2net_msg_handler_func
*func
, void *data
,
816 o2net_post_msg_handler_func
*post_func
,
817 struct list_head
*unreg_list
)
819 struct o2net_msg_handler
*nmh
= NULL
;
820 struct rb_node
**p
, *parent
;
823 if (max_len
> O2NET_MAX_PAYLOAD_BYTES
) {
824 mlog(0, "max_len for message handler out of range: %u\n",
831 mlog(0, "no message type provided: %u, %p\n", msg_type
, func
);
837 mlog(0, "no message handler provided: %u, %p\n",
843 nmh
= kzalloc(sizeof(struct o2net_msg_handler
), GFP_NOFS
);
850 nmh
->nh_func_data
= data
;
851 nmh
->nh_post_func
= post_func
;
852 nmh
->nh_msg_type
= msg_type
;
853 nmh
->nh_max_len
= max_len
;
855 /* the tree and list get this ref.. they're both removed in
856 * unregister when this ref is dropped */
857 kref_init(&nmh
->nh_kref
);
858 INIT_LIST_HEAD(&nmh
->nh_unregister_item
);
860 write_lock(&o2net_handler_lock
);
861 if (o2net_handler_tree_lookup(msg_type
, key
, &p
, &parent
))
864 rb_link_node(&nmh
->nh_node
, parent
, p
);
865 rb_insert_color(&nmh
->nh_node
, &o2net_handler_tree
);
866 list_add_tail(&nmh
->nh_unregister_item
, unreg_list
);
868 mlog(ML_TCP
, "registered handler func %p type %u key %08x\n",
869 func
, msg_type
, key
);
870 /* we've had some trouble with handlers seemingly vanishing. */
871 mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type
, key
, &p
,
873 "couldn't find handler we *just* registered "
874 "for type %u key %08x\n", msg_type
, key
);
876 write_unlock(&o2net_handler_lock
);
886 EXPORT_SYMBOL_GPL(o2net_register_handler
);
888 void o2net_unregister_handler_list(struct list_head
*list
)
890 struct o2net_msg_handler
*nmh
, *n
;
892 write_lock(&o2net_handler_lock
);
893 list_for_each_entry_safe(nmh
, n
, list
, nh_unregister_item
) {
894 mlog(ML_TCP
, "unregistering handler func %p type %u key %08x\n",
895 nmh
->nh_func
, nmh
->nh_msg_type
, nmh
->nh_key
);
896 rb_erase(&nmh
->nh_node
, &o2net_handler_tree
);
897 list_del_init(&nmh
->nh_unregister_item
);
898 kref_put(&nmh
->nh_kref
, o2net_handler_kref_release
);
900 write_unlock(&o2net_handler_lock
);
902 EXPORT_SYMBOL_GPL(o2net_unregister_handler_list
);
904 static struct o2net_msg_handler
*o2net_handler_get(u32 msg_type
, u32 key
)
906 struct o2net_msg_handler
*nmh
;
908 read_lock(&o2net_handler_lock
);
909 nmh
= o2net_handler_tree_lookup(msg_type
, key
, NULL
, NULL
);
911 kref_get(&nmh
->nh_kref
);
912 read_unlock(&o2net_handler_lock
);
917 /* ------------------------------------------------------------ */
919 static int o2net_recv_tcp_msg(struct socket
*sock
, void *data
, size_t len
)
921 struct kvec vec
= { .iov_len
= len
, .iov_base
= data
, };
922 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
, };
923 return kernel_recvmsg(sock
, &msg
, &vec
, 1, len
, msg
.msg_flags
);
926 static int o2net_send_tcp_msg(struct socket
*sock
, struct kvec
*vec
,
927 size_t veclen
, size_t total
)
930 struct msghdr msg
= {.msg_flags
= 0,};
937 ret
= kernel_sendmsg(sock
, &msg
, vec
, veclen
, total
);
938 if (likely(ret
== total
))
940 mlog(ML_ERROR
, "sendmsg returned %d instead of %zu\n", ret
, total
);
942 ret
= -EPIPE
; /* should be smarter, I bet */
944 mlog(0, "returning error: %d\n", ret
);
948 static void o2net_sendpage(struct o2net_sock_container
*sc
,
949 void *kmalloced_virt
,
952 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
956 mutex_lock(&sc
->sc_send_lock
);
957 ret
= sc
->sc_sock
->ops
->sendpage(sc
->sc_sock
,
958 virt_to_page(kmalloced_virt
),
959 (long)kmalloced_virt
& ~PAGE_MASK
,
961 mutex_unlock(&sc
->sc_send_lock
);
964 if (ret
== (ssize_t
)-EAGAIN
) {
965 mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
966 " returned EAGAIN\n", size
, SC_NODEF_ARGS(sc
));
970 mlog(ML_ERROR
, "sendpage of size %zu to " SC_NODEF_FMT
971 " failed with %zd\n", size
, SC_NODEF_ARGS(sc
), ret
);
972 o2net_ensure_shutdown(nn
, sc
, 0);
977 static void o2net_init_msg(struct o2net_msg
*msg
, u16 data_len
, u16 msg_type
, u32 key
)
979 memset(msg
, 0, sizeof(struct o2net_msg
));
980 msg
->magic
= cpu_to_be16(O2NET_MSG_MAGIC
);
981 msg
->data_len
= cpu_to_be16(data_len
);
982 msg
->msg_type
= cpu_to_be16(msg_type
);
983 msg
->sys_status
= cpu_to_be32(O2NET_ERR_NONE
);
985 msg
->key
= cpu_to_be32(key
);
988 static int o2net_tx_can_proceed(struct o2net_node
*nn
,
989 struct o2net_sock_container
**sc_ret
,
994 spin_lock(&nn
->nn_lock
);
995 if (nn
->nn_persistent_error
) {
998 *error
= nn
->nn_persistent_error
;
999 } else if (nn
->nn_sc_valid
) {
1000 kref_get(&nn
->nn_sc
->sc_kref
);
1003 *sc_ret
= nn
->nn_sc
;
1006 spin_unlock(&nn
->nn_lock
);
1011 /* Get a map of all nodes to which this node is currently connected to */
1012 void o2net_fill_node_map(unsigned long *map
, unsigned bytes
)
1014 struct o2net_sock_container
*sc
;
1017 BUG_ON(bytes
< (BITS_TO_LONGS(O2NM_MAX_NODES
) * sizeof(unsigned long)));
1019 memset(map
, 0, bytes
);
1020 for (node
= 0; node
< O2NM_MAX_NODES
; ++node
) {
1021 if (!o2net_tx_can_proceed(o2net_nn_from_num(node
), &sc
, &ret
))
1029 EXPORT_SYMBOL_GPL(o2net_fill_node_map
);
1031 int o2net_send_message_vec(u32 msg_type
, u32 key
, struct kvec
*caller_vec
,
1032 size_t caller_veclen
, u8 target_node
, int *status
)
1035 struct o2net_msg
*msg
= NULL
;
1036 size_t veclen
, caller_bytes
= 0;
1037 struct kvec
*vec
= NULL
;
1038 struct o2net_sock_container
*sc
= NULL
;
1039 struct o2net_node
*nn
= o2net_nn_from_num(target_node
);
1040 struct o2net_status_wait nsw
= {
1041 .ns_node_item
= LIST_HEAD_INIT(nsw
.ns_node_item
),
1043 struct o2net_send_tracking nst
;
1045 o2net_init_nst(&nst
, msg_type
, key
, current
, target_node
);
1047 if (o2net_wq
== NULL
) {
1048 mlog(0, "attempt to tx without o2netd running\n");
1053 if (caller_veclen
== 0) {
1054 mlog(0, "bad kvec array length\n");
1059 caller_bytes
= iov_length((struct iovec
*)caller_vec
, caller_veclen
);
1060 if (caller_bytes
> O2NET_MAX_PAYLOAD_BYTES
) {
1061 mlog(0, "total payload len %zu too large\n", caller_bytes
);
1066 if (target_node
== o2nm_this_node()) {
1071 o2net_debug_add_nst(&nst
);
1073 o2net_set_nst_sock_time(&nst
);
1075 wait_event(nn
->nn_sc_wq
, o2net_tx_can_proceed(nn
, &sc
, &ret
));
1079 o2net_set_nst_sock_container(&nst
, sc
);
1081 veclen
= caller_veclen
+ 1;
1082 vec
= kmalloc(sizeof(struct kvec
) * veclen
, GFP_ATOMIC
);
1084 mlog(0, "failed to %zu element kvec!\n", veclen
);
1089 msg
= kmalloc(sizeof(struct o2net_msg
), GFP_ATOMIC
);
1091 mlog(0, "failed to allocate a o2net_msg!\n");
1096 o2net_init_msg(msg
, caller_bytes
, msg_type
, key
);
1098 vec
[0].iov_len
= sizeof(struct o2net_msg
);
1099 vec
[0].iov_base
= msg
;
1100 memcpy(&vec
[1], caller_vec
, caller_veclen
* sizeof(struct kvec
));
1102 ret
= o2net_prep_nsw(nn
, &nsw
);
1106 msg
->msg_num
= cpu_to_be32(nsw
.ns_id
);
1107 o2net_set_nst_msg_id(&nst
, nsw
.ns_id
);
1109 o2net_set_nst_send_time(&nst
);
1111 /* finally, convert the message header to network byte-order
1113 mutex_lock(&sc
->sc_send_lock
);
1114 ret
= o2net_send_tcp_msg(sc
->sc_sock
, vec
, veclen
,
1115 sizeof(struct o2net_msg
) + caller_bytes
);
1116 mutex_unlock(&sc
->sc_send_lock
);
1117 msglog(msg
, "sending returned %d\n", ret
);
1119 mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret
);
1123 /* wait on other node's handler */
1124 o2net_set_nst_status_time(&nst
);
1125 wait_event(nsw
.ns_wq
, o2net_nsw_completed(nn
, &nsw
));
1127 o2net_update_send_stats(&nst
, sc
);
1129 /* Note that we avoid overwriting the callers status return
1130 * variable if a system error was reported on the other
1131 * side. Callers beware. */
1132 ret
= o2net_sys_err_to_errno(nsw
.ns_sys_status
);
1134 *status
= nsw
.ns_status
;
1136 mlog(0, "woken, returning system status %d, user status %d\n",
1137 ret
, nsw
.ns_status
);
1139 o2net_debug_del_nst(&nst
); /* must be before dropping sc and node */
1144 o2net_complete_nsw(nn
, &nsw
, 0, 0, 0);
1147 EXPORT_SYMBOL_GPL(o2net_send_message_vec
);
1149 int o2net_send_message(u32 msg_type
, u32 key
, void *data
, u32 len
,
1150 u8 target_node
, int *status
)
1156 return o2net_send_message_vec(msg_type
, key
, &vec
, 1,
1157 target_node
, status
);
1159 EXPORT_SYMBOL_GPL(o2net_send_message
);
1161 static int o2net_send_status_magic(struct socket
*sock
, struct o2net_msg
*hdr
,
1162 enum o2net_system_error syserr
, int err
)
1166 .iov_len
= sizeof(struct o2net_msg
),
1169 BUG_ON(syserr
>= O2NET_ERR_MAX
);
1171 /* leave other fields intact from the incoming message, msg_num
1173 hdr
->sys_status
= cpu_to_be32(syserr
);
1174 hdr
->status
= cpu_to_be32(err
);
1175 hdr
->magic
= cpu_to_be16(O2NET_MSG_STATUS_MAGIC
); // twiddle the magic
1178 msglog(hdr
, "about to send status magic %d\n", err
);
1179 /* hdr has been in host byteorder this whole time */
1180 return o2net_send_tcp_msg(sock
, &vec
, 1, sizeof(struct o2net_msg
));
1183 /* this returns -errno if the header was unknown or too large, etc.
1184 * after this is called the buffer us reused for the next message */
1185 static int o2net_process_message(struct o2net_sock_container
*sc
,
1186 struct o2net_msg
*hdr
)
1188 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
1189 int ret
= 0, handler_status
;
1190 enum o2net_system_error syserr
;
1191 struct o2net_msg_handler
*nmh
= NULL
;
1192 void *ret_data
= NULL
;
1194 msglog(hdr
, "processing message\n");
1196 o2net_sc_postpone_idle(sc
);
1198 switch(be16_to_cpu(hdr
->magic
)) {
1199 case O2NET_MSG_STATUS_MAGIC
:
1200 /* special type for returning message status */
1201 o2net_complete_nsw(nn
, NULL
,
1202 be32_to_cpu(hdr
->msg_num
),
1203 be32_to_cpu(hdr
->sys_status
),
1204 be32_to_cpu(hdr
->status
));
1206 case O2NET_MSG_KEEP_REQ_MAGIC
:
1207 o2net_sendpage(sc
, o2net_keep_resp
,
1208 sizeof(*o2net_keep_resp
));
1210 case O2NET_MSG_KEEP_RESP_MAGIC
:
1212 case O2NET_MSG_MAGIC
:
1215 msglog(hdr
, "bad magic\n");
1221 /* find a handler for it */
1223 nmh
= o2net_handler_get(be16_to_cpu(hdr
->msg_type
),
1224 be32_to_cpu(hdr
->key
));
1226 mlog(ML_TCP
, "couldn't find handler for type %u key %08x\n",
1227 be16_to_cpu(hdr
->msg_type
), be32_to_cpu(hdr
->key
));
1228 syserr
= O2NET_ERR_NO_HNDLR
;
1232 syserr
= O2NET_ERR_NONE
;
1234 if (be16_to_cpu(hdr
->data_len
) > nmh
->nh_max_len
)
1235 syserr
= O2NET_ERR_OVERFLOW
;
1237 if (syserr
!= O2NET_ERR_NONE
)
1240 o2net_set_func_start_time(sc
);
1241 sc
->sc_msg_key
= be32_to_cpu(hdr
->key
);
1242 sc
->sc_msg_type
= be16_to_cpu(hdr
->msg_type
);
1243 handler_status
= (nmh
->nh_func
)(hdr
, sizeof(struct o2net_msg
) +
1244 be16_to_cpu(hdr
->data_len
),
1245 nmh
->nh_func_data
, &ret_data
);
1246 o2net_set_func_stop_time(sc
);
1248 o2net_update_recv_stats(sc
);
1251 /* this destroys the hdr, so don't use it after this */
1252 mutex_lock(&sc
->sc_send_lock
);
1253 ret
= o2net_send_status_magic(sc
->sc_sock
, hdr
, syserr
,
1255 mutex_unlock(&sc
->sc_send_lock
);
1257 mlog(0, "sending handler status %d, syserr %d returned %d\n",
1258 handler_status
, syserr
, ret
);
1261 BUG_ON(ret_data
!= NULL
&& nmh
->nh_post_func
== NULL
);
1262 if (nmh
->nh_post_func
)
1263 (nmh
->nh_post_func
)(handler_status
, nmh
->nh_func_data
,
1269 o2net_handler_put(nmh
);
1273 static int o2net_check_handshake(struct o2net_sock_container
*sc
)
1275 struct o2net_handshake
*hand
= page_address(sc
->sc_page
);
1276 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
1278 if (hand
->protocol_version
!= cpu_to_be64(O2NET_PROTOCOL_VERSION
)) {
1279 printk(KERN_NOTICE
"o2net: " SC_NODEF_FMT
" Advertised net "
1280 "protocol version %llu but %llu is required. "
1281 "Disconnecting.\n", SC_NODEF_ARGS(sc
),
1282 (unsigned long long)be64_to_cpu(hand
->protocol_version
),
1283 O2NET_PROTOCOL_VERSION
);
1285 /* don't bother reconnecting if its the wrong version. */
1286 o2net_ensure_shutdown(nn
, sc
, -ENOTCONN
);
1291 * Ensure timeouts are consistent with other nodes, otherwise
1292 * we can end up with one node thinking that the other must be down,
1293 * but isn't. This can ultimately cause corruption.
1295 if (be32_to_cpu(hand
->o2net_idle_timeout_ms
) !=
1296 o2net_idle_timeout()) {
1297 printk(KERN_NOTICE
"o2net: " SC_NODEF_FMT
" uses a network "
1298 "idle timeout of %u ms, but we use %u ms locally. "
1299 "Disconnecting.\n", SC_NODEF_ARGS(sc
),
1300 be32_to_cpu(hand
->o2net_idle_timeout_ms
),
1301 o2net_idle_timeout());
1302 o2net_ensure_shutdown(nn
, sc
, -ENOTCONN
);
1306 if (be32_to_cpu(hand
->o2net_keepalive_delay_ms
) !=
1307 o2net_keepalive_delay()) {
1308 printk(KERN_NOTICE
"o2net: " SC_NODEF_FMT
" uses a keepalive "
1309 "delay of %u ms, but we use %u ms locally. "
1310 "Disconnecting.\n", SC_NODEF_ARGS(sc
),
1311 be32_to_cpu(hand
->o2net_keepalive_delay_ms
),
1312 o2net_keepalive_delay());
1313 o2net_ensure_shutdown(nn
, sc
, -ENOTCONN
);
1317 if (be32_to_cpu(hand
->o2hb_heartbeat_timeout_ms
) !=
1318 O2HB_MAX_WRITE_TIMEOUT_MS
) {
1319 printk(KERN_NOTICE
"o2net: " SC_NODEF_FMT
" uses a heartbeat "
1320 "timeout of %u ms, but we use %u ms locally. "
1321 "Disconnecting.\n", SC_NODEF_ARGS(sc
),
1322 be32_to_cpu(hand
->o2hb_heartbeat_timeout_ms
),
1323 O2HB_MAX_WRITE_TIMEOUT_MS
);
1324 o2net_ensure_shutdown(nn
, sc
, -ENOTCONN
);
1328 sc
->sc_handshake_ok
= 1;
1330 spin_lock(&nn
->nn_lock
);
1331 /* set valid and queue the idle timers only if it hasn't been
1332 * shut down already */
1333 if (nn
->nn_sc
== sc
) {
1334 o2net_sc_reset_idle_timer(sc
);
1335 atomic_set(&nn
->nn_timeout
, 0);
1336 o2net_set_nn_state(nn
, sc
, 1, 0);
1338 spin_unlock(&nn
->nn_lock
);
1340 /* shift everything up as though it wasn't there */
1341 sc
->sc_page_off
-= sizeof(struct o2net_handshake
);
1342 if (sc
->sc_page_off
)
1343 memmove(hand
, hand
+ 1, sc
->sc_page_off
);
1348 /* this demuxes the queued rx bytes into header or payload bits and calls
1349 * handlers as each full message is read off the socket. it returns -error,
1350 * == 0 eof, or > 0 for progress made.*/
1351 static int o2net_advance_rx(struct o2net_sock_container
*sc
)
1353 struct o2net_msg
*hdr
;
1358 sclog(sc
, "receiving\n");
1359 o2net_set_advance_start_time(sc
);
1361 if (unlikely(sc
->sc_handshake_ok
== 0)) {
1362 if(sc
->sc_page_off
< sizeof(struct o2net_handshake
)) {
1363 data
= page_address(sc
->sc_page
) + sc
->sc_page_off
;
1364 datalen
= sizeof(struct o2net_handshake
) - sc
->sc_page_off
;
1365 ret
= o2net_recv_tcp_msg(sc
->sc_sock
, data
, datalen
);
1367 sc
->sc_page_off
+= ret
;
1370 if (sc
->sc_page_off
== sizeof(struct o2net_handshake
)) {
1371 o2net_check_handshake(sc
);
1372 if (unlikely(sc
->sc_handshake_ok
== 0))
1378 /* do we need more header? */
1379 if (sc
->sc_page_off
< sizeof(struct o2net_msg
)) {
1380 data
= page_address(sc
->sc_page
) + sc
->sc_page_off
;
1381 datalen
= sizeof(struct o2net_msg
) - sc
->sc_page_off
;
1382 ret
= o2net_recv_tcp_msg(sc
->sc_sock
, data
, datalen
);
1384 sc
->sc_page_off
+= ret
;
1385 /* only swab incoming here.. we can
1386 * only get here once as we cross from
1387 * being under to over */
1388 if (sc
->sc_page_off
== sizeof(struct o2net_msg
)) {
1389 hdr
= page_address(sc
->sc_page
);
1390 if (be16_to_cpu(hdr
->data_len
) >
1391 O2NET_MAX_PAYLOAD_BYTES
)
1399 if (sc
->sc_page_off
< sizeof(struct o2net_msg
)) {
1400 /* oof, still don't have a header */
1404 /* this was swabbed above when we first read it */
1405 hdr
= page_address(sc
->sc_page
);
1407 msglog(hdr
, "at page_off %zu\n", sc
->sc_page_off
);
1409 /* do we need more payload? */
1410 if (sc
->sc_page_off
- sizeof(struct o2net_msg
) < be16_to_cpu(hdr
->data_len
)) {
1411 /* need more payload */
1412 data
= page_address(sc
->sc_page
) + sc
->sc_page_off
;
1413 datalen
= (sizeof(struct o2net_msg
) + be16_to_cpu(hdr
->data_len
)) -
1415 ret
= o2net_recv_tcp_msg(sc
->sc_sock
, data
, datalen
);
1417 sc
->sc_page_off
+= ret
;
1422 if (sc
->sc_page_off
- sizeof(struct o2net_msg
) == be16_to_cpu(hdr
->data_len
)) {
1423 /* we can only get here once, the first time we read
1424 * the payload.. so set ret to progress if the handler
1425 * works out. after calling this the message is toast */
1426 ret
= o2net_process_message(sc
, hdr
);
1429 sc
->sc_page_off
= 0;
1433 sclog(sc
, "ret = %d\n", ret
);
1434 o2net_set_advance_stop_time(sc
);
1438 /* this work func is triggerd by data ready. it reads until it can read no
1439 * more. it interprets 0, eof, as fatal. if data_ready hits while we're doing
1440 * our work the work struct will be marked and we'll be called again. */
1441 static void o2net_rx_until_empty(struct work_struct
*work
)
1443 struct o2net_sock_container
*sc
=
1444 container_of(work
, struct o2net_sock_container
, sc_rx_work
);
1448 ret
= o2net_advance_rx(sc
);
1451 if (ret
<= 0 && ret
!= -EAGAIN
) {
1452 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
1453 sclog(sc
, "saw error %d, closing\n", ret
);
1454 /* not permanent so read failed handshake can retry */
1455 o2net_ensure_shutdown(nn
, sc
, 0);
1461 static int o2net_set_nodelay(struct socket
*sock
)
1465 return kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
1466 (void *)&val
, sizeof(val
));
1469 static int o2net_set_usertimeout(struct socket
*sock
)
1471 int user_timeout
= O2NET_TCP_USER_TIMEOUT
;
1473 return kernel_setsockopt(sock
, SOL_TCP
, TCP_USER_TIMEOUT
,
1474 (void *)&user_timeout
, sizeof(user_timeout
));
1477 static void o2net_initialize_handshake(void)
1479 o2net_hand
->o2hb_heartbeat_timeout_ms
= cpu_to_be32(
1480 O2HB_MAX_WRITE_TIMEOUT_MS
);
1481 o2net_hand
->o2net_idle_timeout_ms
= cpu_to_be32(o2net_idle_timeout());
1482 o2net_hand
->o2net_keepalive_delay_ms
= cpu_to_be32(
1483 o2net_keepalive_delay());
1484 o2net_hand
->o2net_reconnect_delay_ms
= cpu_to_be32(
1485 o2net_reconnect_delay());
1488 /* ------------------------------------------------------------ */
1490 /* called when a connect completes and after a sock is accepted. the
1491 * rx path will see the response and mark the sc valid */
1492 static void o2net_sc_connect_completed(struct work_struct
*work
)
1494 struct o2net_sock_container
*sc
=
1495 container_of(work
, struct o2net_sock_container
,
1498 mlog(ML_MSG
, "sc sending handshake with ver %llu id %llx\n",
1499 (unsigned long long)O2NET_PROTOCOL_VERSION
,
1500 (unsigned long long)be64_to_cpu(o2net_hand
->connector_id
));
1502 o2net_initialize_handshake();
1503 o2net_sendpage(sc
, o2net_hand
, sizeof(*o2net_hand
));
1507 /* this is called as a work_struct func. */
1508 static void o2net_sc_send_keep_req(struct work_struct
*work
)
1510 struct o2net_sock_container
*sc
=
1511 container_of(work
, struct o2net_sock_container
,
1512 sc_keepalive_work
.work
);
1514 o2net_sendpage(sc
, o2net_keep_req
, sizeof(*o2net_keep_req
));
1518 /* socket shutdown does a del_timer_sync against this as it tears down.
1519 * we can't start this timer until we've got to the point in sc buildup
1520 * where shutdown is going to be involved */
1521 static void o2net_idle_timer(unsigned long data
)
1523 struct o2net_sock_container
*sc
= (struct o2net_sock_container
*)data
;
1524 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
1525 #ifdef CONFIG_DEBUG_FS
1526 unsigned long msecs
= ktime_to_ms(ktime_get()) -
1527 ktime_to_ms(sc
->sc_tv_timer
);
1529 unsigned long msecs
= o2net_idle_timeout();
1532 printk(KERN_NOTICE
"o2net: Connection to " SC_NODEF_FMT
" has been "
1533 "idle for %lu.%lu secs.\n",
1534 SC_NODEF_ARGS(sc
), msecs
/ 1000, msecs
% 1000);
1536 /* idle timerout happen, don't shutdown the connection, but
1537 * make fence decision. Maybe the connection can recover before
1538 * the decision is made.
1540 atomic_set(&nn
->nn_timeout
, 1);
1541 o2quo_conn_err(o2net_num_from_nn(nn
));
1542 queue_delayed_work(o2net_wq
, &nn
->nn_still_up
,
1543 msecs_to_jiffies(O2NET_QUORUM_DELAY_MS
));
1545 o2net_sc_reset_idle_timer(sc
);
1549 static void o2net_sc_reset_idle_timer(struct o2net_sock_container
*sc
)
1551 o2net_sc_cancel_delayed_work(sc
, &sc
->sc_keepalive_work
);
1552 o2net_sc_queue_delayed_work(sc
, &sc
->sc_keepalive_work
,
1553 msecs_to_jiffies(o2net_keepalive_delay()));
1554 o2net_set_sock_timer(sc
);
1555 mod_timer(&sc
->sc_idle_timeout
,
1556 jiffies
+ msecs_to_jiffies(o2net_idle_timeout()));
1559 static void o2net_sc_postpone_idle(struct o2net_sock_container
*sc
)
1561 struct o2net_node
*nn
= o2net_nn_from_num(sc
->sc_node
->nd_num
);
1563 /* clear fence decision since the connection recover from timeout*/
1564 if (atomic_read(&nn
->nn_timeout
)) {
1565 o2quo_conn_up(o2net_num_from_nn(nn
));
1566 cancel_delayed_work(&nn
->nn_still_up
);
1567 atomic_set(&nn
->nn_timeout
, 0);
1570 /* Only push out an existing timer */
1571 if (timer_pending(&sc
->sc_idle_timeout
))
1572 o2net_sc_reset_idle_timer(sc
);
1575 /* this work func is kicked whenever a path sets the nn state which doesn't
1576 * have valid set. This includes seeing hb come up, losing a connection,
1577 * having a connect attempt fail, etc. This centralizes the logic which decides
1578 * if a connect attempt should be made or if we should give up and all future
1579 * transmit attempts should fail */
1580 static void o2net_start_connect(struct work_struct
*work
)
1582 struct o2net_node
*nn
=
1583 container_of(work
, struct o2net_node
, nn_connect_work
.work
);
1584 struct o2net_sock_container
*sc
= NULL
;
1585 struct o2nm_node
*node
= NULL
, *mynode
= NULL
;
1586 struct socket
*sock
= NULL
;
1587 struct sockaddr_in myaddr
= {0, }, remoteaddr
= {0, };
1589 unsigned int timeout
;
1590 unsigned int noio_flag
;
1593 * sock_create allocates the sock with GFP_KERNEL. We must set
1594 * per-process flag PF_MEMALLOC_NOIO so that all allocations done
1595 * by this process are done as if GFP_NOIO was specified. So we
1596 * are not reentering filesystem while doing memory reclaim.
1598 noio_flag
= memalloc_noio_save();
1599 /* if we're greater we initiate tx, otherwise we accept */
1600 if (o2nm_this_node() <= o2net_num_from_nn(nn
))
1603 /* watch for racing with tearing a node down */
1604 node
= o2nm_get_node_by_num(o2net_num_from_nn(nn
));
1608 mynode
= o2nm_get_node_by_num(o2nm_this_node());
1612 spin_lock(&nn
->nn_lock
);
1614 * see if we already have one pending or have given up.
1615 * For nn_timeout, it is set when we close the connection
1616 * because of the idle time out. So it means that we have
1617 * at least connected to that node successfully once,
1618 * now try to connect to it again.
1620 timeout
= atomic_read(&nn
->nn_timeout
);
1621 stop
= (nn
->nn_sc
||
1622 (nn
->nn_persistent_error
&&
1623 (nn
->nn_persistent_error
!= -ENOTCONN
|| timeout
== 0)));
1624 spin_unlock(&nn
->nn_lock
);
1628 nn
->nn_last_connect_attempt
= jiffies
;
1630 sc
= sc_alloc(node
);
1632 mlog(0, "couldn't allocate sc\n");
1637 ret
= sock_create(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
);
1639 mlog(0, "can't create socket: %d\n", ret
);
1642 sc
->sc_sock
= sock
; /* freed by sc_kref_release */
1644 sock
->sk
->sk_allocation
= GFP_ATOMIC
;
1646 myaddr
.sin_family
= AF_INET
;
1647 myaddr
.sin_addr
.s_addr
= mynode
->nd_ipv4_address
;
1648 myaddr
.sin_port
= htons(0); /* any port */
1650 ret
= sock
->ops
->bind(sock
, (struct sockaddr
*)&myaddr
,
1653 mlog(ML_ERROR
, "bind failed with %d at address %pI4\n",
1654 ret
, &mynode
->nd_ipv4_address
);
1658 ret
= o2net_set_nodelay(sc
->sc_sock
);
1660 mlog(ML_ERROR
, "setting TCP_NODELAY failed with %d\n", ret
);
1664 ret
= o2net_set_usertimeout(sock
);
1666 mlog(ML_ERROR
, "set TCP_USER_TIMEOUT failed with %d\n", ret
);
1670 o2net_register_callbacks(sc
->sc_sock
->sk
, sc
);
1672 spin_lock(&nn
->nn_lock
);
1673 /* handshake completion will set nn->nn_sc_valid */
1674 o2net_set_nn_state(nn
, sc
, 0, 0);
1675 spin_unlock(&nn
->nn_lock
);
1677 remoteaddr
.sin_family
= AF_INET
;
1678 remoteaddr
.sin_addr
.s_addr
= node
->nd_ipv4_address
;
1679 remoteaddr
.sin_port
= node
->nd_ipv4_port
;
1681 ret
= sc
->sc_sock
->ops
->connect(sc
->sc_sock
,
1682 (struct sockaddr
*)&remoteaddr
,
1685 if (ret
== -EINPROGRESS
)
1690 printk(KERN_NOTICE
"o2net: Connect attempt to " SC_NODEF_FMT
1691 " failed with errno %d\n", SC_NODEF_ARGS(sc
), ret
);
1692 /* 0 err so that another will be queued and attempted
1693 * from set_nn_state */
1694 o2net_ensure_shutdown(nn
, sc
, 0);
1699 o2nm_node_put(node
);
1701 o2nm_node_put(mynode
);
1703 memalloc_noio_restore(noio_flag
);
1707 static void o2net_connect_expired(struct work_struct
*work
)
1709 struct o2net_node
*nn
=
1710 container_of(work
, struct o2net_node
, nn_connect_expired
.work
);
1712 spin_lock(&nn
->nn_lock
);
1713 if (!nn
->nn_sc_valid
) {
1714 printk(KERN_NOTICE
"o2net: No connection established with "
1715 "node %u after %u.%u seconds, check network and"
1716 " cluster configuration.\n",
1717 o2net_num_from_nn(nn
),
1718 o2net_idle_timeout() / 1000,
1719 o2net_idle_timeout() % 1000);
1721 o2net_set_nn_state(nn
, NULL
, 0, 0);
1723 spin_unlock(&nn
->nn_lock
);
1726 static void o2net_still_up(struct work_struct
*work
)
1728 struct o2net_node
*nn
=
1729 container_of(work
, struct o2net_node
, nn_still_up
.work
);
1731 o2quo_hb_still_up(o2net_num_from_nn(nn
));
1734 /* ------------------------------------------------------------ */
1736 void o2net_disconnect_node(struct o2nm_node
*node
)
1738 struct o2net_node
*nn
= o2net_nn_from_num(node
->nd_num
);
1740 /* don't reconnect until it's heartbeating again */
1741 spin_lock(&nn
->nn_lock
);
1742 atomic_set(&nn
->nn_timeout
, 0);
1743 o2net_set_nn_state(nn
, NULL
, 0, -ENOTCONN
);
1744 spin_unlock(&nn
->nn_lock
);
1747 cancel_delayed_work(&nn
->nn_connect_expired
);
1748 cancel_delayed_work(&nn
->nn_connect_work
);
1749 cancel_delayed_work(&nn
->nn_still_up
);
1750 flush_workqueue(o2net_wq
);
1754 static void o2net_hb_node_down_cb(struct o2nm_node
*node
, int node_num
,
1757 o2quo_hb_down(node_num
);
1762 if (node_num
!= o2nm_this_node())
1763 o2net_disconnect_node(node
);
1765 BUG_ON(atomic_read(&o2net_connected_peers
) < 0);
1768 static void o2net_hb_node_up_cb(struct o2nm_node
*node
, int node_num
,
1771 struct o2net_node
*nn
= o2net_nn_from_num(node_num
);
1773 o2quo_hb_up(node_num
);
1777 /* ensure an immediate connect attempt */
1778 nn
->nn_last_connect_attempt
= jiffies
-
1779 (msecs_to_jiffies(o2net_reconnect_delay()) + 1);
1781 if (node_num
!= o2nm_this_node()) {
1782 /* believe it or not, accept and node hearbeating testing
1783 * can succeed for this node before we got here.. so
1784 * only use set_nn_state to clear the persistent error
1785 * if that hasn't already happened */
1786 spin_lock(&nn
->nn_lock
);
1787 atomic_set(&nn
->nn_timeout
, 0);
1788 if (nn
->nn_persistent_error
)
1789 o2net_set_nn_state(nn
, NULL
, 0, 0);
1790 spin_unlock(&nn
->nn_lock
);
1794 void o2net_unregister_hb_callbacks(void)
1796 o2hb_unregister_callback(NULL
, &o2net_hb_up
);
1797 o2hb_unregister_callback(NULL
, &o2net_hb_down
);
1800 int o2net_register_hb_callbacks(void)
1804 o2hb_setup_callback(&o2net_hb_down
, O2HB_NODE_DOWN_CB
,
1805 o2net_hb_node_down_cb
, NULL
, O2NET_HB_PRI
);
1806 o2hb_setup_callback(&o2net_hb_up
, O2HB_NODE_UP_CB
,
1807 o2net_hb_node_up_cb
, NULL
, O2NET_HB_PRI
);
1809 ret
= o2hb_register_callback(NULL
, &o2net_hb_up
);
1811 ret
= o2hb_register_callback(NULL
, &o2net_hb_down
);
1814 o2net_unregister_hb_callbacks();
1819 /* ------------------------------------------------------------ */
1821 static int o2net_accept_one(struct socket
*sock
, int *more
)
1824 struct sockaddr_in sin
;
1825 struct socket
*new_sock
= NULL
;
1826 struct o2nm_node
*node
= NULL
;
1827 struct o2nm_node
*local_node
= NULL
;
1828 struct o2net_sock_container
*sc
= NULL
;
1829 struct o2net_node
*nn
;
1830 unsigned int noio_flag
;
1833 * sock_create_lite allocates the sock with GFP_KERNEL. We must set
1834 * per-process flag PF_MEMALLOC_NOIO so that all allocations done
1835 * by this process are done as if GFP_NOIO was specified. So we
1836 * are not reentering filesystem while doing memory reclaim.
1838 noio_flag
= memalloc_noio_save();
1840 BUG_ON(sock
== NULL
);
1842 ret
= sock_create_lite(sock
->sk
->sk_family
, sock
->sk
->sk_type
,
1843 sock
->sk
->sk_protocol
, &new_sock
);
1847 new_sock
->type
= sock
->type
;
1848 new_sock
->ops
= sock
->ops
;
1849 ret
= sock
->ops
->accept(sock
, new_sock
, O_NONBLOCK
, false);
1854 new_sock
->sk
->sk_allocation
= GFP_ATOMIC
;
1856 ret
= o2net_set_nodelay(new_sock
);
1858 mlog(ML_ERROR
, "setting TCP_NODELAY failed with %d\n", ret
);
1862 ret
= o2net_set_usertimeout(new_sock
);
1864 mlog(ML_ERROR
, "set TCP_USER_TIMEOUT failed with %d\n", ret
);
1869 ret
= new_sock
->ops
->getname(new_sock
, (struct sockaddr
*) &sin
,
1874 node
= o2nm_get_node_by_ip(sin
.sin_addr
.s_addr
);
1876 printk(KERN_NOTICE
"o2net: Attempt to connect from unknown "
1877 "node at %pI4:%d\n", &sin
.sin_addr
.s_addr
,
1878 ntohs(sin
.sin_port
));
1883 if (o2nm_this_node() >= node
->nd_num
) {
1884 local_node
= o2nm_get_node_by_num(o2nm_this_node());
1886 printk(KERN_NOTICE
"o2net: Unexpected connect attempt "
1887 "seen at node '%s' (%u, %pI4:%d) from "
1888 "node '%s' (%u, %pI4:%d)\n",
1889 local_node
->nd_name
, local_node
->nd_num
,
1890 &(local_node
->nd_ipv4_address
),
1891 ntohs(local_node
->nd_ipv4_port
),
1893 node
->nd_num
, &sin
.sin_addr
.s_addr
,
1894 ntohs(sin
.sin_port
));
1899 /* this happens all the time when the other node sees our heartbeat
1900 * and tries to connect before we see their heartbeat */
1901 if (!o2hb_check_node_heartbeating_from_callback(node
->nd_num
)) {
1902 mlog(ML_CONN
, "attempt to connect from node '%s' at "
1903 "%pI4:%d but it isn't heartbeating\n",
1904 node
->nd_name
, &sin
.sin_addr
.s_addr
,
1905 ntohs(sin
.sin_port
));
1910 nn
= o2net_nn_from_num(node
->nd_num
);
1912 spin_lock(&nn
->nn_lock
);
1917 spin_unlock(&nn
->nn_lock
);
1919 printk(KERN_NOTICE
"o2net: Attempt to connect from node '%s' "
1920 "at %pI4:%d but it already has an open connection\n",
1921 node
->nd_name
, &sin
.sin_addr
.s_addr
,
1922 ntohs(sin
.sin_port
));
1926 sc
= sc_alloc(node
);
1932 sc
->sc_sock
= new_sock
;
1935 spin_lock(&nn
->nn_lock
);
1936 atomic_set(&nn
->nn_timeout
, 0);
1937 o2net_set_nn_state(nn
, sc
, 0, 0);
1938 spin_unlock(&nn
->nn_lock
);
1940 o2net_register_callbacks(sc
->sc_sock
->sk
, sc
);
1941 o2net_sc_queue_work(sc
, &sc
->sc_rx_work
);
1943 o2net_initialize_handshake();
1944 o2net_sendpage(sc
, o2net_hand
, sizeof(*o2net_hand
));
1948 sock_release(new_sock
);
1950 o2nm_node_put(node
);
1952 o2nm_node_put(local_node
);
1956 memalloc_noio_restore(noio_flag
);
1961 * This function is invoked in response to one or more
1962 * pending accepts at softIRQ level. We must drain the
1963 * entire que before returning.
1966 static void o2net_accept_many(struct work_struct
*work
)
1968 struct socket
*sock
= o2net_listen_sock
;
1973 * It is critical to note that due to interrupt moderation
1974 * at the network driver level, we can't assume to get a
1975 * softIRQ for every single conn since tcp SYN packets
1976 * can arrive back-to-back, and therefore many pending
1977 * accepts may result in just 1 softIRQ. If we terminate
1978 * the o2net_accept_one() loop upon seeing an err, what happens
1979 * to the rest of the conns in the queue? If no new SYN
1980 * arrives for hours, no softIRQ will be delivered,
1981 * and the connections will just sit in the queue.
1985 err
= o2net_accept_one(sock
, &more
);
1992 static void o2net_listen_data_ready(struct sock
*sk
)
1994 void (*ready
)(struct sock
*sk
);
1996 read_lock_bh(&sk
->sk_callback_lock
);
1997 ready
= sk
->sk_user_data
;
1998 if (ready
== NULL
) { /* check for teardown race */
1999 ready
= sk
->sk_data_ready
;
2003 /* This callback may called twice when a new connection
2004 * is being established as a child socket inherits everything
2005 * from a parent LISTEN socket, including the data_ready cb of
2006 * the parent. This leads to a hazard. In o2net_accept_one()
2007 * we are still initializing the child socket but have not
2008 * changed the inherited data_ready callback yet when
2009 * data starts arriving.
2010 * We avoid this hazard by checking the state.
2011 * For the listening socket, the state will be TCP_LISTEN; for the new
2012 * socket, will be TCP_ESTABLISHED. Also, in this case,
2013 * sk->sk_user_data is not a valid function pointer.
2016 if (sk
->sk_state
== TCP_LISTEN
) {
2017 queue_work(o2net_wq
, &o2net_listen_work
);
2023 read_unlock_bh(&sk
->sk_callback_lock
);
2028 static int o2net_open_listening_sock(__be32 addr
, __be16 port
)
2030 struct socket
*sock
= NULL
;
2032 struct sockaddr_in sin
= {
2033 .sin_family
= PF_INET
,
2034 .sin_addr
= { .s_addr
= addr
},
2038 ret
= sock_create(PF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
);
2040 printk(KERN_ERR
"o2net: Error %d while creating socket\n", ret
);
2044 sock
->sk
->sk_allocation
= GFP_ATOMIC
;
2046 write_lock_bh(&sock
->sk
->sk_callback_lock
);
2047 sock
->sk
->sk_user_data
= sock
->sk
->sk_data_ready
;
2048 sock
->sk
->sk_data_ready
= o2net_listen_data_ready
;
2049 write_unlock_bh(&sock
->sk
->sk_callback_lock
);
2051 o2net_listen_sock
= sock
;
2052 INIT_WORK(&o2net_listen_work
, o2net_accept_many
);
2054 sock
->sk
->sk_reuse
= SK_CAN_REUSE
;
2055 ret
= sock
->ops
->bind(sock
, (struct sockaddr
*)&sin
, sizeof(sin
));
2057 printk(KERN_ERR
"o2net: Error %d while binding socket at "
2058 "%pI4:%u\n", ret
, &addr
, ntohs(port
));
2062 ret
= sock
->ops
->listen(sock
, 64);
2064 printk(KERN_ERR
"o2net: Error %d while listening on %pI4:%u\n",
2065 ret
, &addr
, ntohs(port
));
2069 o2net_listen_sock
= NULL
;
2077 * called from node manager when we should bring up our network listening
2078 * socket. node manager handles all the serialization to only call this
2079 * once and to match it with o2net_stop_listening(). note,
2080 * o2nm_this_node() doesn't work yet as we're being called while it
2083 int o2net_start_listening(struct o2nm_node
*node
)
2087 BUG_ON(o2net_wq
!= NULL
);
2088 BUG_ON(o2net_listen_sock
!= NULL
);
2090 mlog(ML_KTHREAD
, "starting o2net thread...\n");
2091 o2net_wq
= alloc_ordered_workqueue("o2net", WQ_MEM_RECLAIM
);
2092 if (o2net_wq
== NULL
) {
2093 mlog(ML_ERROR
, "unable to launch o2net thread\n");
2094 return -ENOMEM
; /* ? */
2097 ret
= o2net_open_listening_sock(node
->nd_ipv4_address
,
2098 node
->nd_ipv4_port
);
2100 destroy_workqueue(o2net_wq
);
2103 o2quo_conn_up(node
->nd_num
);
2108 /* again, o2nm_this_node() doesn't work here as we're involved in
2109 * tearing it down */
2110 void o2net_stop_listening(struct o2nm_node
*node
)
2112 struct socket
*sock
= o2net_listen_sock
;
2115 BUG_ON(o2net_wq
== NULL
);
2116 BUG_ON(o2net_listen_sock
== NULL
);
2118 /* stop the listening socket from generating work */
2119 write_lock_bh(&sock
->sk
->sk_callback_lock
);
2120 sock
->sk
->sk_data_ready
= sock
->sk
->sk_user_data
;
2121 sock
->sk
->sk_user_data
= NULL
;
2122 write_unlock_bh(&sock
->sk
->sk_callback_lock
);
2124 for (i
= 0; i
< ARRAY_SIZE(o2net_nodes
); i
++) {
2125 struct o2nm_node
*node
= o2nm_get_node_by_num(i
);
2127 o2net_disconnect_node(node
);
2128 o2nm_node_put(node
);
2132 /* finish all work and tear down the work queue */
2133 mlog(ML_KTHREAD
, "waiting for o2net thread to exit....\n");
2134 destroy_workqueue(o2net_wq
);
2137 sock_release(o2net_listen_sock
);
2138 o2net_listen_sock
= NULL
;
2140 o2quo_conn_err(node
->nd_num
);
2143 /* ------------------------------------------------------------ */
2145 int o2net_init(void)
2151 if (o2net_debugfs_init())
2154 o2net_hand
= kzalloc(sizeof(struct o2net_handshake
), GFP_KERNEL
);
2155 o2net_keep_req
= kzalloc(sizeof(struct o2net_msg
), GFP_KERNEL
);
2156 o2net_keep_resp
= kzalloc(sizeof(struct o2net_msg
), GFP_KERNEL
);
2157 if (!o2net_hand
|| !o2net_keep_req
|| !o2net_keep_resp
)
2160 o2net_hand
->protocol_version
= cpu_to_be64(O2NET_PROTOCOL_VERSION
);
2161 o2net_hand
->connector_id
= cpu_to_be64(1);
2163 o2net_keep_req
->magic
= cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC
);
2164 o2net_keep_resp
->magic
= cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC
);
2166 for (i
= 0; i
< ARRAY_SIZE(o2net_nodes
); i
++) {
2167 struct o2net_node
*nn
= o2net_nn_from_num(i
);
2169 atomic_set(&nn
->nn_timeout
, 0);
2170 spin_lock_init(&nn
->nn_lock
);
2171 INIT_DELAYED_WORK(&nn
->nn_connect_work
, o2net_start_connect
);
2172 INIT_DELAYED_WORK(&nn
->nn_connect_expired
,
2173 o2net_connect_expired
);
2174 INIT_DELAYED_WORK(&nn
->nn_still_up
, o2net_still_up
);
2175 /* until we see hb from a node we'll return einval */
2176 nn
->nn_persistent_error
= -ENOTCONN
;
2177 init_waitqueue_head(&nn
->nn_sc_wq
);
2178 idr_init(&nn
->nn_status_idr
);
2179 INIT_LIST_HEAD(&nn
->nn_status_list
);
2186 kfree(o2net_keep_req
);
2187 kfree(o2net_keep_resp
);
2188 o2net_debugfs_exit();
2193 void o2net_exit(void)
2197 kfree(o2net_keep_req
);
2198 kfree(o2net_keep_resp
);
2199 o2net_debugfs_exit();