4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ptlrpc/pinger.c
38 * Portal-RPC reconnection and replay operations, for use in recovery.
42 #include <liblustre.h>
44 #define DEBUG_SUBSYSTEM S_RPC
47 #include <obd_support.h>
48 #include <obd_class.h>
49 #include "ptlrpc_internal.h"
51 static int suppress_pings;
52 CFS_MODULE_PARM(suppress_pings, "i", int, 0644, "Suppress pings");
54 struct mutex pinger_mutex;
55 static CFS_LIST_HEAD(pinger_imports);
56 static cfs_list_t timeout_list = CFS_LIST_HEAD_INIT(timeout_list);
58 int ptlrpc_pinger_suppress_pings()
60 return suppress_pings;
62 EXPORT_SYMBOL(ptlrpc_pinger_suppress_pings);
64 struct ptlrpc_request *
65 ptlrpc_prep_ping(struct obd_import *imp)
67 struct ptlrpc_request *req;
69 req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING,
70 LUSTRE_OBD_VERSION, OBD_PING);
72 ptlrpc_request_set_replen(req);
73 req->rq_no_resend = req->rq_no_delay = 1;
78 int ptlrpc_obd_ping(struct obd_device *obd)
81 struct ptlrpc_request *req;
84 req = ptlrpc_prep_ping(obd->u.cli.cl_import);
88 req->rq_send_state = LUSTRE_IMP_FULL;
90 rc = ptlrpc_queue_wait(req);
92 ptlrpc_req_finished(req);
96 EXPORT_SYMBOL(ptlrpc_obd_ping);
98 int ptlrpc_ping(struct obd_import *imp)
100 struct ptlrpc_request *req;
103 req = ptlrpc_prep_ping(imp);
105 CERROR("OOM trying to ping %s->%s\n",
106 imp->imp_obd->obd_uuid.uuid,
107 obd2cli_tgt(imp->imp_obd));
111 DEBUG_REQ(D_INFO, req, "pinging %s->%s",
112 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
113 ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1);
118 void ptlrpc_update_next_ping(struct obd_import *imp, int soon)
121 int time = soon ? PING_INTERVAL_SHORT : PING_INTERVAL;
122 if (imp->imp_state == LUSTRE_IMP_DISCON) {
123 int dtime = max_t(int, CONNECTION_SWITCH_MIN,
125 at_get(&imp->imp_at.iat_net_latency));
126 time = min(time, dtime);
128 imp->imp_next_ping = cfs_time_shift(time);
129 #endif /* ENABLE_PINGER */
132 void ptlrpc_ping_import_soon(struct obd_import *imp)
134 imp->imp_next_ping = cfs_time_current();
137 static inline int imp_is_deactive(struct obd_import *imp)
139 return (imp->imp_deactive ||
140 OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_IMP_DEACTIVE));
143 static inline int ptlrpc_next_reconnect(struct obd_import *imp)
145 if (imp->imp_server_timeout)
146 return cfs_time_shift(obd_timeout / 2);
148 return cfs_time_shift(obd_timeout);
151 cfs_duration_t pinger_check_timeout(cfs_time_t time)
153 struct timeout_item *item;
154 cfs_time_t timeout = PING_INTERVAL;
156 /* The timeout list is a increase order sorted list */
157 mutex_lock(&pinger_mutex);
158 cfs_list_for_each_entry(item, &timeout_list, ti_chain) {
159 int ti_timeout = item->ti_timeout;
160 if (timeout > ti_timeout)
161 timeout = ti_timeout;
164 mutex_unlock(&pinger_mutex);
166 return cfs_time_sub(cfs_time_add(time, cfs_time_seconds(timeout)),
174 void ptlrpc_pinger_ir_up(void)
176 CDEBUG(D_HA, "IR up\n");
179 EXPORT_SYMBOL(ptlrpc_pinger_ir_up);
181 void ptlrpc_pinger_ir_down(void)
183 CDEBUG(D_HA, "IR down\n");
186 EXPORT_SYMBOL(ptlrpc_pinger_ir_down);
188 static void ptlrpc_pinger_process_import(struct obd_import *imp,
189 unsigned long this_ping)
196 spin_lock(&imp->imp_lock);
198 level = imp->imp_state;
199 force = imp->imp_force_verify;
200 force_next = imp->imp_force_next_verify;
202 * This will be used below only if the import is "FULL".
204 suppress = ir_up && OCD_HAS_FLAG(&imp->imp_connect_data, PINGLESS);
206 imp->imp_force_verify = 0;
208 if (cfs_time_aftereq(imp->imp_next_ping - 5 * CFS_TICK, this_ping) &&
210 spin_unlock(&imp->imp_lock);
214 imp->imp_force_next_verify = 0;
216 spin_unlock(&imp->imp_lock);
218 CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA, "%s->%s: level %s/%u "
219 "force %u force_next %u deactive %u pingable %u suppress %u\n",
220 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd),
221 ptlrpc_import_state_name(level), level, force, force_next,
222 imp->imp_deactive, imp->imp_pingable, suppress);
224 if (level == LUSTRE_IMP_DISCON && !imp_is_deactive(imp)) {
225 /* wait for a while before trying recovery again */
226 imp->imp_next_ping = ptlrpc_next_reconnect(imp);
227 if (!imp->imp_no_pinger_recover)
228 ptlrpc_initiate_recovery(imp);
229 } else if (level != LUSTRE_IMP_FULL ||
230 imp->imp_obd->obd_no_recov ||
231 imp_is_deactive(imp)) {
232 CDEBUG(D_HA, "%s->%s: not pinging (in recovery "
233 "or recovery disabled: %s)\n",
234 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd),
235 ptlrpc_import_state_name(level));
236 } else if ((imp->imp_pingable && !suppress) || force_next || force) {
241 static int ptlrpc_pinger_main(void *arg)
243 struct ptlrpc_thread *thread = (struct ptlrpc_thread *)arg;
246 /* Record that the thread is running */
247 thread_set_flags(thread, SVC_RUNNING);
248 cfs_waitq_signal(&thread->t_ctl_waitq);
250 /* And now, loop forever, pinging as needed. */
252 cfs_time_t this_ping = cfs_time_current();
253 struct l_wait_info lwi;
254 cfs_duration_t time_to_next_wake;
255 struct timeout_item *item;
258 mutex_lock(&pinger_mutex);
259 cfs_list_for_each_entry(item, &timeout_list, ti_chain) {
260 item->ti_cb(item, item->ti_cb_data);
262 cfs_list_for_each(iter, &pinger_imports) {
263 struct obd_import *imp =
264 cfs_list_entry(iter, struct obd_import,
267 ptlrpc_pinger_process_import(imp, this_ping);
268 /* obd_timeout might have changed */
269 if (imp->imp_pingable && imp->imp_next_ping &&
270 cfs_time_after(imp->imp_next_ping,
271 cfs_time_add(this_ping,
272 cfs_time_seconds(PING_INTERVAL))))
273 ptlrpc_update_next_ping(imp, 0);
275 mutex_unlock(&pinger_mutex);
276 /* update memory usage info */
277 obd_update_maxusage();
279 /* Wait until the next ping time, or until we're stopped. */
280 time_to_next_wake = pinger_check_timeout(this_ping);
281 /* The ping sent by ptlrpc_send_rpc may get sent out
282 say .01 second after this.
283 ptlrpc_pinger_sending_on_import will then set the
284 next ping time to next_ping + .01 sec, which means
285 we will SKIP the next ping at next_ping, and the
286 ping will get sent 2 timeouts from now! Beware. */
287 CDEBUG(D_INFO, "next wakeup in "CFS_DURATION_T" ("
288 CFS_TIME_T")\n", time_to_next_wake,
289 cfs_time_add(this_ping,cfs_time_seconds(PING_INTERVAL)));
290 if (time_to_next_wake > 0) {
291 lwi = LWI_TIMEOUT(max_t(cfs_duration_t,
293 cfs_time_seconds(1)),
295 l_wait_event(thread->t_ctl_waitq,
296 thread_is_stopping(thread) ||
297 thread_is_event(thread),
299 if (thread_test_and_clear_flags(thread, SVC_STOPPING)) {
303 /* woken after adding import to reset timer */
304 thread_test_and_clear_flags(thread, SVC_EVENT);
309 thread_set_flags(thread, SVC_STOPPED);
310 cfs_waitq_signal(&thread->t_ctl_waitq);
312 CDEBUG(D_NET, "pinger thread exiting, process %d\n", cfs_curproc_pid());
316 static struct ptlrpc_thread pinger_thread;
318 int ptlrpc_start_pinger(void)
320 struct l_wait_info lwi = { 0 };
322 #ifndef ENABLE_PINGER
327 if (!thread_is_init(&pinger_thread) &&
328 !thread_is_stopped(&pinger_thread))
331 cfs_waitq_init(&pinger_thread.t_ctl_waitq);
333 strcpy(pinger_thread.t_name, "ll_ping");
335 /* CLONE_VM and CLONE_FILES just avoid a needless copy, because we
336 * just drop the VM and FILES in kthread_run() right away. */
337 rc = PTR_ERR(kthread_run(ptlrpc_pinger_main,
338 &pinger_thread, pinger_thread.t_name));
339 if (IS_ERR_VALUE(rc)) {
340 CERROR("cannot start thread: %d\n", rc);
343 l_wait_event(pinger_thread.t_ctl_waitq,
344 thread_is_running(&pinger_thread), &lwi);
347 CWARN("Pings will be suppressed at the request of the "
348 "administrator. The configuration shall meet the "
349 "additional requirements described in the manual. "
350 "(Search for the \"suppress_pings\" kernel module "
356 int ptlrpc_pinger_remove_timeouts(void);
358 int ptlrpc_stop_pinger(void)
360 struct l_wait_info lwi = { 0 };
361 #ifndef ENABLE_PINGER
366 if (thread_is_init(&pinger_thread) ||
367 thread_is_stopped(&pinger_thread))
370 ptlrpc_pinger_remove_timeouts();
372 thread_set_flags(&pinger_thread, SVC_STOPPING);
373 cfs_waitq_signal(&pinger_thread.t_ctl_waitq);
375 l_wait_event(pinger_thread.t_ctl_waitq,
376 thread_is_stopped(&pinger_thread), &lwi);
380 void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
382 ptlrpc_update_next_ping(imp, 0);
384 EXPORT_SYMBOL(ptlrpc_pinger_sending_on_import);
386 void ptlrpc_pinger_commit_expected(struct obd_import *imp)
388 ptlrpc_update_next_ping(imp, 1);
389 LASSERT_SPIN_LOCKED(&imp->imp_lock);
391 * Avoid reading stale imp_connect_data. When not sure if pings are
392 * expected or not on next connection, we assume they are not and force
393 * one anyway to guarantee the chance of updating
394 * imp_peer_committed_transno.
396 if (imp->imp_state != LUSTRE_IMP_FULL ||
397 OCD_HAS_FLAG(&imp->imp_connect_data, PINGLESS))
398 imp->imp_force_next_verify = 1;
401 int ptlrpc_pinger_add_import(struct obd_import *imp)
404 if (!cfs_list_empty(&imp->imp_pinger_chain))
407 mutex_lock(&pinger_mutex);
408 CDEBUG(D_HA, "adding pingable import %s->%s\n",
409 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
410 /* if we add to pinger we want recovery on this import */
411 imp->imp_obd->obd_no_recov = 0;
412 ptlrpc_update_next_ping(imp, 0);
413 /* XXX sort, blah blah */
414 cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
415 class_import_get(imp);
417 ptlrpc_pinger_wake_up();
418 mutex_unlock(&pinger_mutex);
422 EXPORT_SYMBOL(ptlrpc_pinger_add_import);
424 int ptlrpc_pinger_del_import(struct obd_import *imp)
427 if (cfs_list_empty(&imp->imp_pinger_chain))
430 mutex_lock(&pinger_mutex);
431 cfs_list_del_init(&imp->imp_pinger_chain);
432 CDEBUG(D_HA, "removing pingable import %s->%s\n",
433 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
434 /* if we remove from pinger we don't want recovery on this import */
435 imp->imp_obd->obd_no_recov = 1;
436 class_import_put(imp);
437 mutex_unlock(&pinger_mutex);
440 EXPORT_SYMBOL(ptlrpc_pinger_del_import);
443 * Register a timeout callback to the pinger list, and the callback will
444 * be called when timeout happens.
446 struct timeout_item* ptlrpc_new_timeout(int time, enum timeout_event event,
447 timeout_cb_t cb, void *data)
449 struct timeout_item *ti;
455 CFS_INIT_LIST_HEAD(&ti->ti_obd_list);
456 CFS_INIT_LIST_HEAD(&ti->ti_chain);
457 ti->ti_timeout = time;
458 ti->ti_event = event;
460 ti->ti_cb_data = data;
466 * Register timeout event on the the pinger thread.
467 * Note: the timeout list is an sorted list with increased timeout value.
469 static struct timeout_item*
470 ptlrpc_pinger_register_timeout(int time, enum timeout_event event,
471 timeout_cb_t cb, void *data)
473 struct timeout_item *item, *tmp;
475 LASSERT_MUTEX_LOCKED(&pinger_mutex);
477 cfs_list_for_each_entry(item, &timeout_list, ti_chain)
478 if (item->ti_event == event)
481 item = ptlrpc_new_timeout(time, event, cb, data);
483 cfs_list_for_each_entry_reverse(tmp, &timeout_list, ti_chain) {
484 if (tmp->ti_timeout < time) {
485 cfs_list_add(&item->ti_chain, &tmp->ti_chain);
489 cfs_list_add(&item->ti_chain, &timeout_list);
495 /* Add a client_obd to the timeout event list, when timeout(@time)
496 * happens, the callback(@cb) will be called.
498 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
499 timeout_cb_t cb, void *data,
500 cfs_list_t *obd_list)
502 struct timeout_item *ti;
504 mutex_lock(&pinger_mutex);
505 ti = ptlrpc_pinger_register_timeout(time, event, cb, data);
507 mutex_unlock(&pinger_mutex);
510 cfs_list_add(obd_list, &ti->ti_obd_list);
511 mutex_unlock(&pinger_mutex);
514 EXPORT_SYMBOL(ptlrpc_add_timeout_client);
516 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
517 enum timeout_event event)
519 struct timeout_item *ti = NULL, *item;
521 if (cfs_list_empty(obd_list))
523 mutex_lock(&pinger_mutex);
524 cfs_list_del_init(obd_list);
526 * If there are no obd attached to the timeout event
527 * list, remove this timeout event from the pinger
529 cfs_list_for_each_entry(item, &timeout_list, ti_chain) {
530 if (item->ti_event == event) {
535 LASSERTF(ti != NULL, "ti is NULL ! \n");
536 if (cfs_list_empty(&ti->ti_obd_list)) {
537 cfs_list_del(&ti->ti_chain);
540 mutex_unlock(&pinger_mutex);
543 EXPORT_SYMBOL(ptlrpc_del_timeout_client);
545 int ptlrpc_pinger_remove_timeouts(void)
547 struct timeout_item *item, *tmp;
549 mutex_lock(&pinger_mutex);
550 cfs_list_for_each_entry_safe(item, tmp, &timeout_list, ti_chain) {
551 LASSERT(cfs_list_empty(&item->ti_obd_list));
552 cfs_list_del(&item->ti_chain);
555 mutex_unlock(&pinger_mutex);
559 void ptlrpc_pinger_wake_up()
562 thread_add_flags(&pinger_thread, SVC_EVENT);
563 cfs_waitq_signal(&pinger_thread.t_ctl_waitq);
567 /* Ping evictor thread */
569 #define PET_TERMINATE 2
571 static int pet_refcount = 0;
572 static int pet_state;
573 static cfs_waitq_t pet_waitq;
574 CFS_LIST_HEAD(pet_list);
575 static DEFINE_SPINLOCK(pet_lock);
577 int ping_evictor_wake(struct obd_export *exp)
579 struct obd_device *obd;
581 spin_lock(&pet_lock);
582 if (pet_state != PET_READY) {
583 /* eventually the new obd will call here again. */
584 spin_unlock(&pet_lock);
588 obd = class_exp2obd(exp);
589 if (cfs_list_empty(&obd->obd_evict_list)) {
590 class_incref(obd, "evictor", obd);
591 cfs_list_add(&obd->obd_evict_list, &pet_list);
593 spin_unlock(&pet_lock);
595 cfs_waitq_signal(&pet_waitq);
599 static int ping_evictor_main(void *arg)
601 struct obd_device *obd;
602 struct obd_export *exp;
603 struct l_wait_info lwi = { 0 };
609 CDEBUG(D_HA, "Starting Ping Evictor\n");
610 pet_state = PET_READY;
612 l_wait_event(pet_waitq, (!cfs_list_empty(&pet_list)) ||
613 (pet_state == PET_TERMINATE), &lwi);
615 /* loop until all obd's will be removed */
616 if ((pet_state == PET_TERMINATE) && cfs_list_empty(&pet_list))
619 /* we only get here if pet_exp != NULL, and the end of this
620 * loop is the only place which sets it NULL again, so lock
621 * is not strictly necessary. */
622 spin_lock(&pet_lock);
623 obd = cfs_list_entry(pet_list.next, struct obd_device,
625 spin_unlock(&pet_lock);
627 expire_time = cfs_time_current_sec() - PING_EVICT_TIMEOUT;
629 CDEBUG(D_HA, "evicting all exports of obd %s older than %ld\n",
630 obd->obd_name, expire_time);
632 /* Exports can't be deleted out of the list while we hold
633 * the obd lock (class_unlink_export), which means we can't
634 * lose the last ref on the export. If they've already been
635 * removed from the list, we won't find them here. */
636 spin_lock(&obd->obd_dev_lock);
637 while (!cfs_list_empty(&obd->obd_exports_timed)) {
638 exp = cfs_list_entry(obd->obd_exports_timed.next,
640 exp_obd_chain_timed);
641 if (expire_time > exp->exp_last_request_time) {
642 class_export_get(exp);
643 spin_unlock(&obd->obd_dev_lock);
644 LCONSOLE_WARN("%s: haven't heard from client %s"
645 " (at %s) in %ld seconds. I think"
646 " it's dead, and I am evicting"
647 " it. exp %p, cur %ld expire %ld"
650 obd_uuid2str(&exp->exp_client_uuid),
651 obd_export_nid2str(exp),
652 (long)(cfs_time_current_sec() -
653 exp->exp_last_request_time),
654 exp, (long)cfs_time_current_sec(),
656 (long)exp->exp_last_request_time);
657 CDEBUG(D_HA, "Last request was at %ld\n",
658 exp->exp_last_request_time);
659 class_fail_export(exp);
660 class_export_put(exp);
661 spin_lock(&obd->obd_dev_lock);
663 /* List is sorted, so everyone below is ok */
667 spin_unlock(&obd->obd_dev_lock);
669 spin_lock(&pet_lock);
670 cfs_list_del_init(&obd->obd_evict_list);
671 spin_unlock(&pet_lock);
673 class_decref(obd, "evictor", obd);
675 CDEBUG(D_HA, "Exiting Ping Evictor\n");
680 void ping_evictor_start(void)
684 if (++pet_refcount > 1)
687 cfs_waitq_init(&pet_waitq);
689 task = kthread_run(ping_evictor_main, NULL, "ll_evictor");
692 CERROR("Cannot start ping evictor thread: %ld\n",
696 EXPORT_SYMBOL(ping_evictor_start);
698 void ping_evictor_stop(void)
700 if (--pet_refcount > 0)
703 pet_state = PET_TERMINATE;
704 cfs_waitq_signal(&pet_waitq);
706 EXPORT_SYMBOL(ping_evictor_stop);
707 #else /* !__KERNEL__ */
710 * the current implementation of pinger in liblustre is not optimized
714 static struct pinger_data {
716 cfs_time_t pd_this_ping; /* jiffies */
717 cfs_time_t pd_next_ping; /* jiffies */
718 struct ptlrpc_request_set *pd_set;
721 static int pinger_check_rpcs(void *arg)
723 cfs_time_t curtime = cfs_time_current();
724 struct ptlrpc_request *req;
725 struct ptlrpc_request_set *set;
727 struct obd_import *imp;
728 struct pinger_data *pd = &pinger_args;
731 /* prevent recursion */
732 if (pd->pd_recursion++) {
733 CDEBUG(D_HA, "pinger: recursion! quit\n");
739 /* have we reached ping point? */
740 if (!pd->pd_set && cfs_time_before(curtime, pd->pd_next_ping)) {
745 /* if we have rpc_set already, continue processing it */
747 LASSERT(pd->pd_this_ping);
752 pd->pd_this_ping = curtime;
753 pd->pd_set = ptlrpc_prep_set();
754 if (pd->pd_set == NULL)
758 /* add rpcs into set */
759 mutex_lock(&pinger_mutex);
760 cfs_list_for_each(iter, &pinger_imports) {
761 struct obd_import *imp = cfs_list_entry(iter, struct obd_import,
763 int generation, level;
765 if (cfs_time_aftereq(pd->pd_this_ping,
766 imp->imp_next_ping - 5 * CFS_TICK)) {
768 spin_lock(&imp->imp_lock);
769 generation = imp->imp_generation;
770 level = imp->imp_state;
771 spin_unlock(&imp->imp_lock);
773 if (level != LUSTRE_IMP_FULL) {
775 "not pinging %s (in recovery)\n",
776 obd2cli_tgt(imp->imp_obd));
780 req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING,
784 CERROR("OOM trying to ping %s->%s\n",
785 imp->imp_obd->obd_uuid.uuid,
786 obd2cli_tgt(imp->imp_obd));
790 req->rq_no_resend = 1;
791 ptlrpc_request_set_replen(req);
792 req->rq_send_state = LUSTRE_IMP_FULL;
793 ptlrpc_rqphase_move(req, RQ_PHASE_RPC);
794 req->rq_import_generation = generation;
795 ptlrpc_set_add_req(set, req);
797 CDEBUG(D_INFO, "don't need to ping %s ("CFS_TIME_T
798 " > "CFS_TIME_T")\n", obd2cli_tgt(imp->imp_obd),
799 imp->imp_next_ping, pd->pd_this_ping);
802 pd->pd_this_ping = curtime;
803 mutex_unlock(&pinger_mutex);
805 /* Might be empty, that's OK. */
806 if (cfs_atomic_read(&set->set_remaining) == 0)
807 CDEBUG(D_RPCTRACE, "nothing to ping\n");
809 cfs_list_for_each(iter, &set->set_requests) {
810 struct ptlrpc_request *req =
811 cfs_list_entry(iter, struct ptlrpc_request,
813 DEBUG_REQ(D_RPCTRACE, req, "pinging %s->%s",
814 req->rq_import->imp_obd->obd_uuid.uuid,
815 obd2cli_tgt(req->rq_import->imp_obd));
816 (void)ptl_send_rpc(req, 0);
820 rc = ptlrpc_check_set(NULL, set);
822 /* not finished, and we are not expired, simply return */
823 if (!rc && cfs_time_before(curtime, cfs_time_add(pd->pd_this_ping,
824 cfs_time_seconds(PING_INTERVAL)))) {
825 CDEBUG(D_RPCTRACE, "not finished, but also not expired\n");
830 /* Expire all the requests that didn't come back. */
831 mutex_lock(&pinger_mutex);
832 cfs_list_for_each(iter, &set->set_requests) {
833 req = cfs_list_entry(iter, struct ptlrpc_request,
836 if (req->rq_phase == RQ_PHASE_COMPLETE)
839 CDEBUG(D_RPCTRACE, "Pinger initiate expire request(%p)\n",
842 /* This will also unregister reply. */
843 ptlrpc_expire_one_request(req, 0);
845 /* We're done with this req, let's finally move it to complete
846 * phase and take care of inflights. */
847 ptlrpc_rqphase_move(req, RQ_PHASE_COMPLETE);
848 imp = req->rq_import;
849 spin_lock(&imp->imp_lock);
850 if (!cfs_list_empty(&req->rq_list)) {
851 cfs_list_del_init(&req->rq_list);
852 cfs_atomic_dec(&imp->imp_inflight);
854 spin_unlock(&imp->imp_lock);
855 cfs_atomic_dec(&set->set_remaining);
857 mutex_unlock(&pinger_mutex);
859 ptlrpc_set_destroy(set);
863 pd->pd_next_ping = cfs_time_add(pd->pd_this_ping,
864 cfs_time_seconds(PING_INTERVAL));
865 pd->pd_this_ping = 0; /* XXX for debug */
867 CDEBUG(D_INFO, "finished a round ping\n");
872 static void *pinger_callback = NULL;
873 #endif /* ENABLE_PINGER */
875 int ptlrpc_start_pinger(void)
878 memset(&pinger_args, 0, sizeof(pinger_args));
879 pinger_callback = liblustre_register_wait_callback("pinger_check_rpcs",
886 int ptlrpc_stop_pinger(void)
890 liblustre_deregister_wait_callback(pinger_callback);
895 void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
898 mutex_lock(&pinger_mutex);
899 ptlrpc_update_next_ping(imp, 0);
900 if (pinger_args.pd_set == NULL &&
901 cfs_time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) {
902 CDEBUG(D_HA, "set next ping to "CFS_TIME_T"(cur "CFS_TIME_T")\n",
903 imp->imp_next_ping, cfs_time_current());
904 pinger_args.pd_next_ping = imp->imp_next_ping;
906 mutex_unlock(&pinger_mutex);
910 void ptlrpc_pinger_commit_expected(struct obd_import *imp)
913 mutex_lock(&pinger_mutex);
914 ptlrpc_update_next_ping(imp, 1);
915 if (pinger_args.pd_set == NULL &&
916 cfs_time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) {
917 CDEBUG(D_HA,"set next ping to "CFS_TIME_T"(cur "CFS_TIME_T")\n",
918 imp->imp_next_ping, cfs_time_current());
919 pinger_args.pd_next_ping = imp->imp_next_ping;
921 mutex_unlock(&pinger_mutex);
925 int ptlrpc_add_timeout_client(int time, enum timeout_event event,
926 timeout_cb_t cb, void *data,
927 cfs_list_t *obd_list)
932 int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
933 enum timeout_event event)
938 int ptlrpc_pinger_add_import(struct obd_import *imp)
941 if (!cfs_list_empty(&imp->imp_pinger_chain))
944 CDEBUG(D_HA, "adding pingable import %s->%s\n",
945 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
946 ptlrpc_pinger_sending_on_import(imp);
948 mutex_lock(&pinger_mutex);
949 cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
950 class_import_get(imp);
951 mutex_unlock(&pinger_mutex);
956 int ptlrpc_pinger_del_import(struct obd_import *imp)
959 if (cfs_list_empty(&imp->imp_pinger_chain))
962 mutex_lock(&pinger_mutex);
963 cfs_list_del_init(&imp->imp_pinger_chain);
964 CDEBUG(D_HA, "removing pingable import %s->%s\n",
965 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
966 class_import_put(imp);
967 mutex_unlock(&pinger_mutex);
971 void ptlrpc_pinger_wake_up()
974 /* XXX force pinger to run, if needed */
975 struct obd_import *imp;
977 cfs_list_for_each_entry(imp, &pinger_imports, imp_pinger_chain) {
978 CDEBUG(D_RPCTRACE, "checking import %s->%s\n",
979 imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
980 #ifdef ENABLE_LIBLUSTRE_RECOVERY
981 if (imp->imp_state == LUSTRE_IMP_DISCON &&
982 !imp_is_deactive(imp))
984 /*XXX only recover for the initial connection */
985 if (!lustre_handle_is_used(&imp->imp_remote_handle) &&
986 imp->imp_state == LUSTRE_IMP_DISCON &&
987 !imp_is_deactive(imp))
989 ptlrpc_initiate_recovery(imp);
990 else if (imp->imp_state != LUSTRE_IMP_FULL)
991 CDEBUG(D_HA, "Refused to recover import %s->%s "
992 "state %d, deactive %d\n",
993 imp->imp_obd->obd_uuid.uuid,
994 obd2cli_tgt(imp->imp_obd), imp->imp_state,
995 imp_is_deactive(imp));
1000 #endif /* !__KERNEL__ */