X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fptlrpc%2Fpinger.c;h=2d48a43ba3d1339eaed9884e0025e4a7d165df32;hb=ff0c89a73e141ce019ee2a94e5d01a8a37dd830a;hp=60e05550b998fca67b6ffbd9a03be30f487ba01b;hpb=b4c85cd5b6a4be39b240b39c5d0c1dbcbc12b93a;p=fs%2Flustre-release.git diff --git a/lustre/ptlrpc/pinger.c b/lustre/ptlrpc/pinger.c index 60e0555..2d48a43 100644 --- a/lustre/ptlrpc/pinger.c +++ b/lustre/ptlrpc/pinger.c @@ -1,29 +1,41 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: +/* + * GPL HEADER START * - * Portal-RPC reconnection and replay operations, for use in recovery. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). + * + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * Copyright (c) 2003 Cluster File Systems, Inc. - * Authors: Phil Schwan - * Mike Shaver + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * GPL HEADER END + */ +/* + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * Copyright (c) 2011, 2012, Whamcloud, Inc. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * lustre/ptlrpc/pinger.c * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * Portal-RPC reconnection and replay operations, for use in recovery. */ #ifndef __KERNEL__ @@ -36,16 +48,49 @@ #include #include "ptlrpc_internal.h" -struct semaphore pinger_sem; +cfs_mutex_t pinger_mutex; static CFS_LIST_HEAD(pinger_imports); +static cfs_list_t timeout_list = CFS_LIST_HEAD_INIT(timeout_list); +struct ptlrpc_request * +ptlrpc_prep_ping(struct obd_import *imp) +{ + struct ptlrpc_request *req; + + req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING, + LUSTRE_OBD_VERSION, OBD_PING); + if (req) { + ptlrpc_request_set_replen(req); + req->rq_no_resend = req->rq_no_delay = 1; + } + return req; +} + +int ptlrpc_obd_ping(struct obd_device *obd) +{ + int rc; + struct ptlrpc_request *req; + ENTRY; + + req = ptlrpc_prep_ping(obd->u.cli.cl_import); + if (req == NULL) + RETURN(-ENOMEM); + + req->rq_send_state = LUSTRE_IMP_FULL; + + rc = ptlrpc_queue_wait(req); + + ptlrpc_req_finished(req); + + RETURN(rc); +} +EXPORT_SYMBOL(ptlrpc_obd_ping); int ptlrpc_ping(struct obd_import *imp) { struct ptlrpc_request *req; ENTRY; - req = ptlrpc_request_alloc_pack(imp, &RQF_OBD_PING, LUSTRE_OBD_VERSION, - OBD_PING); + req = ptlrpc_prep_ping(imp); if (req == NULL) { CERROR("OOM trying to ping %s->%s\n", imp->imp_obd->obd_uuid.uuid, @@ -55,20 +100,22 @@ int ptlrpc_ping(struct obd_import *imp) DEBUG_REQ(D_INFO, req, "pinging %s->%s", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); - req->rq_no_resend = req->rq_no_delay = 1; - ptlrpc_request_set_replen(req); - req->rq_timeout = PING_INTERVAL; - ptlrpcd_add_req(req); + ptlrpcd_add_req(req, PDL_POLICY_ROUND, -1); RETURN(0); } -void ptlrpc_update_next_ping(struct obd_import *imp) +void ptlrpc_update_next_ping(struct obd_import *imp, int soon) { #ifdef ENABLE_PINGER - imp->imp_next_ping = cfs_time_shift( - (imp->imp_state == LUSTRE_IMP_DISCON ? - RECONNECT_INTERVAL : PING_INTERVAL)); + int time = soon ? PING_INTERVAL_SHORT : PING_INTERVAL; + if (imp->imp_state == LUSTRE_IMP_DISCON) { + int dtime = max_t(int, CONNECTION_SWITCH_MIN, + AT_OFF ? 0 : + at_get(&imp->imp_at.iat_net_latency)); + time = min(time, dtime); + } + imp->imp_next_ping = cfs_time_shift(time); #endif /* ENABLE_PINGER */ } @@ -77,6 +124,12 @@ void ptlrpc_ping_import_soon(struct obd_import *imp) imp->imp_next_ping = cfs_time_current(); } +static inline int imp_is_deactive(struct obd_import *imp) +{ + return (imp->imp_deactive || + OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_IMP_DEACTIVE)); +} + static inline int ptlrpc_next_reconnect(struct obd_import *imp) { if (imp->imp_server_timeout) @@ -85,12 +138,29 @@ static inline int ptlrpc_next_reconnect(struct obd_import *imp) return cfs_time_shift(obd_timeout); } -static atomic_t suspend_timeouts = ATOMIC_INIT(0); +static cfs_atomic_t suspend_timeouts = CFS_ATOMIC_INIT(0); static cfs_time_t suspend_wakeup_time = 0; -#ifdef __KERNEL__ -static wait_queue_head_t suspend_timeouts_waitq; -#endif +cfs_duration_t pinger_check_timeout(cfs_time_t time) +{ + struct timeout_item *item; + cfs_time_t timeout = PING_INTERVAL; + + /* The timeout list is a increase order sorted list */ + cfs_mutex_lock(&pinger_mutex); + cfs_list_for_each_entry(item, &timeout_list, ti_chain) { + int ti_timeout = item->ti_timeout; + if (timeout > ti_timeout) + timeout = ti_timeout; + break; + } + cfs_mutex_unlock(&pinger_mutex); + + return cfs_time_sub(cfs_time_add(time, cfs_time_seconds(timeout)), + cfs_time_current()); +} + +static cfs_waitq_t suspend_timeouts_waitq; cfs_time_t ptlrpc_suspend_wakeup_time(void) { @@ -104,8 +174,9 @@ void ptlrpc_deactivate_timeouts(struct obd_import *imp) if (imp->imp_no_timeout) return; imp->imp_no_timeout = 1; - atomic_inc(&suspend_timeouts); - CDEBUG(D_HA|D_WARNING, "deactivate timeouts %u\n", atomic_read(&suspend_timeouts)); + cfs_atomic_inc(&suspend_timeouts); + CDEBUG(D_HA|D_WARNING, "deactivate timeouts %u\n", + cfs_atomic_read(&suspend_timeouts)); #endif } @@ -116,18 +187,19 @@ void ptlrpc_activate_timeouts(struct obd_import *imp) if (!imp->imp_no_timeout) return; imp->imp_no_timeout = 0; - LASSERT(atomic_read(&suspend_timeouts) > 0); - if (atomic_dec_and_test(&suspend_timeouts)) { + LASSERT(cfs_atomic_read(&suspend_timeouts) > 0); + if (cfs_atomic_dec_and_test(&suspend_timeouts)) { suspend_wakeup_time = cfs_time_current(); - wake_up(&suspend_timeouts_waitq); + cfs_waitq_signal(&suspend_timeouts_waitq); } - CDEBUG(D_HA|D_WARNING, "activate timeouts %u\n", atomic_read(&suspend_timeouts)); + CDEBUG(D_HA|D_WARNING, "activate timeouts %u\n", + cfs_atomic_read(&suspend_timeouts)); #endif } int ptlrpc_check_suspend(void) { - if (atomic_read(&suspend_timeouts)) + if (cfs_atomic_read(&suspend_timeouts)) return 1; return 0; } @@ -136,12 +208,12 @@ int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req) { struct l_wait_info lwi; - if (atomic_read(&suspend_timeouts)) { + if (cfs_atomic_read(&suspend_timeouts)) { DEBUG_REQ(D_NET, req, "-- suspend %d regular timeout", - atomic_read(&suspend_timeouts)); + cfs_atomic_read(&suspend_timeouts)); lwi = LWI_INTR(NULL, NULL); l_wait_event(suspend_timeouts_waitq, - atomic_read(&suspend_timeouts) == 0, &lwi); + cfs_atomic_read(&suspend_timeouts) == 0, &lwi); DEBUG_REQ(D_NET, req, "-- recharge regular timeout"); return 1; } @@ -155,12 +227,12 @@ static void ptlrpc_pinger_process_import(struct obd_import *imp, { int force, level; - spin_lock(&imp->imp_lock); + cfs_spin_lock(&imp->imp_lock); level = imp->imp_state; force = imp->imp_force_verify; if (force) imp->imp_force_verify = 0; - spin_unlock(&imp->imp_lock); + cfs_spin_unlock(&imp->imp_lock); CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA, "level %s/%u force %u deactive %u pingable %u\n", @@ -171,13 +243,14 @@ static void ptlrpc_pinger_process_import(struct obd_import *imp, this_ping) && force == 0) return; - if (level == LUSTRE_IMP_DISCON && !imp->imp_deactive) { - /* wait at least a timeout before trying recovery again */ + if (level == LUSTRE_IMP_DISCON && !imp_is_deactive(imp)) { + /* wait for a while before trying recovery again */ imp->imp_next_ping = ptlrpc_next_reconnect(imp); - ptlrpc_initiate_recovery(imp); + if (!imp->imp_no_pinger_recover) + ptlrpc_initiate_recovery(imp); } else if (level != LUSTRE_IMP_FULL || imp->imp_obd->obd_no_recov || - imp->imp_deactive) { + imp_is_deactive(imp)) { CDEBUG(D_HA, "not pinging %s (in recovery " " or recovery disabled: %s)\n", obd2cli_tgt(imp->imp_obd), @@ -189,28 +262,31 @@ static void ptlrpc_pinger_process_import(struct obd_import *imp, static int ptlrpc_pinger_main(void *arg) { - struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg; - struct ptlrpc_thread *thread = data->thread; - ENTRY; + struct ptlrpc_thread *thread = (struct ptlrpc_thread *)arg; + ENTRY; - cfs_daemonize(data->name); + cfs_daemonize(thread->t_name); /* Record that the thread is running */ - thread->t_flags = SVC_RUNNING; + thread_set_flags(thread, SVC_RUNNING); cfs_waitq_signal(&thread->t_ctl_waitq); /* And now, loop forever, pinging as needed. */ while (1) { cfs_time_t this_ping = cfs_time_current(); struct l_wait_info lwi; - cfs_duration_t time_to_next_ping; - struct list_head *iter; + cfs_duration_t time_to_next_wake; + struct timeout_item *item; + cfs_list_t *iter; - mutex_down(&pinger_sem); - list_for_each(iter, &pinger_imports) { + cfs_mutex_lock(&pinger_mutex); + cfs_list_for_each_entry(item, &timeout_list, ti_chain) { + item->ti_cb(item, item->ti_cb_data); + } + cfs_list_for_each(iter, &pinger_imports) { struct obd_import *imp = - list_entry(iter, struct obd_import, - imp_pinger_chain); + cfs_list_entry(iter, struct obd_import, + imp_pinger_chain); ptlrpc_pinger_process_import(imp, this_ping); /* obd_timeout might have changed */ @@ -218,47 +294,43 @@ static int ptlrpc_pinger_main(void *arg) cfs_time_after(imp->imp_next_ping, cfs_time_add(this_ping, cfs_time_seconds(PING_INTERVAL)))) - ptlrpc_update_next_ping(imp); + ptlrpc_update_next_ping(imp, 0); } - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); /* update memory usage info */ obd_update_maxusage(); /* Wait until the next ping time, or until we're stopped. */ - time_to_next_ping = cfs_time_sub(cfs_time_add(this_ping, - cfs_time_seconds(PING_INTERVAL)), - cfs_time_current()); - + time_to_next_wake = pinger_check_timeout(this_ping); /* The ping sent by ptlrpc_send_rpc may get sent out say .01 second after this. ptlrpc_pinger_sending_on_import will then set the next ping time to next_ping + .01 sec, which means we will SKIP the next ping at next_ping, and the ping will get sent 2 timeouts from now! Beware. */ - CDEBUG(D_INFO, "next ping in "CFS_DURATION_T" ("CFS_TIME_T")\n", - time_to_next_ping, - cfs_time_add(this_ping, - cfs_time_seconds(PING_INTERVAL))); - if (time_to_next_ping > 0) { + CDEBUG(D_INFO, "next wakeup in "CFS_DURATION_T" (" + CFS_TIME_T")\n", time_to_next_wake, + cfs_time_add(this_ping,cfs_time_seconds(PING_INTERVAL))); + if (time_to_next_wake > 0) { lwi = LWI_TIMEOUT(max_t(cfs_duration_t, - time_to_next_ping, + time_to_next_wake, cfs_time_seconds(1)), NULL, NULL); l_wait_event(thread->t_ctl_waitq, - thread->t_flags & (SVC_STOPPING|SVC_EVENT), + thread_is_stopping(thread) || + thread_is_event(thread), &lwi); - if (thread->t_flags & SVC_STOPPING) { - thread->t_flags &= ~SVC_STOPPING; + if (thread_test_and_clear_flags(thread, SVC_STOPPING)) { EXIT; break; - } else if (thread->t_flags & SVC_EVENT) { + } else { /* woken after adding import to reset timer */ - thread->t_flags &= ~SVC_EVENT; + thread_test_and_clear_flags(thread, SVC_EVENT); } } } - thread->t_flags = SVC_STOPPED; + thread_set_flags(thread, SVC_STOPPED); cfs_waitq_signal(&thread->t_ctl_waitq); CDEBUG(D_NET, "pinger thread exiting, process %d\n", cfs_curproc_pid()); @@ -270,7 +342,6 @@ static struct ptlrpc_thread *pinger_thread = NULL; int ptlrpc_start_pinger(void) { struct l_wait_info lwi = { 0 }; - struct ptlrpc_svc_data d; int rc; #ifndef ENABLE_PINGER return 0; @@ -286,12 +357,12 @@ int ptlrpc_start_pinger(void) cfs_waitq_init(&pinger_thread->t_ctl_waitq); cfs_waitq_init(&suspend_timeouts_waitq); - d.name = "ll_ping"; - d.thread = pinger_thread; + strcpy(pinger_thread->t_name, "ll_ping"); - /* CLONE_VM and CLONE_FILES just avoid a needless copy, because we - * just drop the VM and FILES in ptlrpc_daemonize() right away. */ - rc = cfs_kernel_thread(ptlrpc_pinger_main, &d, CLONE_VM | CLONE_FILES); + /* CLONE_VM and CLONE_FILES just avoid a needless copy, because we + * just drop the VM and FILES in cfs_daemonize_ctxt() right away. */ + rc = cfs_create_thread(ptlrpc_pinger_main, + pinger_thread, CFS_DAEMON_FLAGS); if (rc < 0) { CERROR("cannot start thread: %d\n", rc); OBD_FREE(pinger_thread, sizeof(*pinger_thread)); @@ -299,11 +370,13 @@ int ptlrpc_start_pinger(void) RETURN(rc); } l_wait_event(pinger_thread->t_ctl_waitq, - pinger_thread->t_flags & SVC_RUNNING, &lwi); + thread_is_running(pinger_thread), &lwi); RETURN(0); } +int ptlrpc_pinger_remove_timeouts(void); + int ptlrpc_stop_pinger(void) { struct l_wait_info lwi = { 0 }; @@ -315,13 +388,15 @@ int ptlrpc_stop_pinger(void) if (pinger_thread == NULL) RETURN(-EALREADY); - mutex_down(&pinger_sem); - pinger_thread->t_flags = SVC_STOPPING; + + ptlrpc_pinger_remove_timeouts(); + cfs_mutex_lock(&pinger_mutex); + thread_set_flags(pinger_thread, SVC_STOPPING); cfs_waitq_signal(&pinger_thread->t_ctl_waitq); - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); l_wait_event(pinger_thread->t_ctl_waitq, - (pinger_thread->t_flags & SVC_STOPPED), &lwi); + thread_is_stopped(pinger_thread), &lwi); OBD_FREE_PTR(pinger_thread); pinger_thread = NULL; @@ -330,27 +405,32 @@ int ptlrpc_stop_pinger(void) void ptlrpc_pinger_sending_on_import(struct obd_import *imp) { - ptlrpc_update_next_ping(imp); + ptlrpc_update_next_ping(imp, 0); +} + +void ptlrpc_pinger_commit_expected(struct obd_import *imp) +{ + ptlrpc_update_next_ping(imp, 1); } int ptlrpc_pinger_add_import(struct obd_import *imp) { ENTRY; - if (!list_empty(&imp->imp_pinger_chain)) + if (!cfs_list_empty(&imp->imp_pinger_chain)) RETURN(-EALREADY); - mutex_down(&pinger_sem); + cfs_mutex_lock(&pinger_mutex); CDEBUG(D_HA, "adding pingable import %s->%s\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); /* if we add to pinger we want recovery on this import */ imp->imp_obd->obd_no_recov = 0; - ptlrpc_update_next_ping(imp); + ptlrpc_update_next_ping(imp, 0); /* XXX sort, blah blah */ - list_add_tail(&imp->imp_pinger_chain, &pinger_imports); + cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports); class_import_get(imp); ptlrpc_pinger_wake_up(); - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); RETURN(0); } @@ -358,24 +438,139 @@ int ptlrpc_pinger_add_import(struct obd_import *imp) int ptlrpc_pinger_del_import(struct obd_import *imp) { ENTRY; - if (list_empty(&imp->imp_pinger_chain)) + if (cfs_list_empty(&imp->imp_pinger_chain)) RETURN(-ENOENT); - mutex_down(&pinger_sem); - list_del_init(&imp->imp_pinger_chain); + cfs_mutex_lock(&pinger_mutex); + cfs_list_del_init(&imp->imp_pinger_chain); CDEBUG(D_HA, "removing pingable import %s->%s\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); /* if we remove from pinger we don't want recovery on this import */ imp->imp_obd->obd_no_recov = 1; class_import_put(imp); - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); RETURN(0); } +/** + * Register a timeout callback to the pinger list, and the callback will + * be called when timeout happens. + */ +struct timeout_item* ptlrpc_new_timeout(int time, enum timeout_event event, + timeout_cb_t cb, void *data) +{ + struct timeout_item *ti; + + OBD_ALLOC_PTR(ti); + if (!ti) + return(NULL); + + CFS_INIT_LIST_HEAD(&ti->ti_obd_list); + CFS_INIT_LIST_HEAD(&ti->ti_chain); + ti->ti_timeout = time; + ti->ti_event = event; + ti->ti_cb = cb; + ti->ti_cb_data = data; + + return ti; +} + +/** + * Register timeout event on the the pinger thread. + * Note: the timeout list is an sorted list with increased timeout value. + */ +static struct timeout_item* +ptlrpc_pinger_register_timeout(int time, enum timeout_event event, + timeout_cb_t cb, void *data) +{ + struct timeout_item *item, *tmp; + + LASSERT_MUTEX_LOCKED(&pinger_mutex); + + cfs_list_for_each_entry(item, &timeout_list, ti_chain) + if (item->ti_event == event) + goto out; + + item = ptlrpc_new_timeout(time, event, cb, data); + if (item) { + cfs_list_for_each_entry_reverse(tmp, &timeout_list, ti_chain) { + if (tmp->ti_timeout < time) { + cfs_list_add(&item->ti_chain, &tmp->ti_chain); + goto out; + } + } + cfs_list_add(&item->ti_chain, &timeout_list); + } +out: + return item; +} + +/* Add a client_obd to the timeout event list, when timeout(@time) + * happens, the callback(@cb) will be called. + */ +int ptlrpc_add_timeout_client(int time, enum timeout_event event, + timeout_cb_t cb, void *data, + cfs_list_t *obd_list) +{ + struct timeout_item *ti; + + cfs_mutex_lock(&pinger_mutex); + ti = ptlrpc_pinger_register_timeout(time, event, cb, data); + if (!ti) { + cfs_mutex_unlock(&pinger_mutex); + return (-EINVAL); + } + cfs_list_add(obd_list, &ti->ti_obd_list); + cfs_mutex_unlock(&pinger_mutex); + return 0; +} + +int ptlrpc_del_timeout_client(cfs_list_t *obd_list, + enum timeout_event event) +{ + struct timeout_item *ti = NULL, *item; + + if (cfs_list_empty(obd_list)) + return 0; + cfs_mutex_lock(&pinger_mutex); + cfs_list_del_init(obd_list); + /** + * If there are no obd attached to the timeout event + * list, remove this timeout event from the pinger + */ + cfs_list_for_each_entry(item, &timeout_list, ti_chain) { + if (item->ti_event == event) { + ti = item; + break; + } + } + LASSERTF(ti != NULL, "ti is NULL ! \n"); + if (cfs_list_empty(&ti->ti_obd_list)) { + cfs_list_del(&ti->ti_chain); + OBD_FREE_PTR(ti); + } + cfs_mutex_unlock(&pinger_mutex); + return 0; +} + +int ptlrpc_pinger_remove_timeouts(void) +{ + struct timeout_item *item, *tmp; + + cfs_mutex_lock(&pinger_mutex); + cfs_list_for_each_entry_safe(item, tmp, &timeout_list, ti_chain) { + LASSERT(cfs_list_empty(&item->ti_obd_list)); + cfs_list_del(&item->ti_chain); + OBD_FREE_PTR(item); + } + cfs_mutex_unlock(&pinger_mutex); + return 0; +} + void ptlrpc_pinger_wake_up() { #ifdef ENABLE_PINGER - pinger_thread->t_flags |= SVC_EVENT; + thread_add_flags(pinger_thread, SVC_EVENT); cfs_waitq_signal(&pinger_thread->t_ctl_waitq); #endif } @@ -386,27 +581,29 @@ void ptlrpc_pinger_wake_up() static int pet_refcount = 0; static int pet_state; -static wait_queue_head_t pet_waitq; -static struct obd_export *pet_exp = NULL; -static spinlock_t pet_lock = SPIN_LOCK_UNLOCKED; +static cfs_waitq_t pet_waitq; +CFS_LIST_HEAD(pet_list); +static cfs_spinlock_t pet_lock = CFS_SPIN_LOCK_UNLOCKED; int ping_evictor_wake(struct obd_export *exp) { - spin_lock(&pet_lock); - if (pet_exp || (pet_state != PET_READY)) { + struct obd_device *obd; + + cfs_spin_lock(&pet_lock); + if (pet_state != PET_READY) { /* eventually the new obd will call here again. */ - spin_unlock(&pet_lock); + cfs_spin_unlock(&pet_lock); return 1; } - /* We have to make sure the obd isn't destroyed between now and when - * the ping evictor runs. We'll take a reference here, and drop it - * when we finish in the evictor. We don't really care about this - * export in particular; we just need one to keep the obd alive. */ - pet_exp = class_export_get(exp); - spin_unlock(&pet_lock); + obd = class_exp2obd(exp); + if (cfs_list_empty(&obd->obd_evict_list)) { + class_incref(obd, "evictor", obd); + cfs_list_add(&obd->obd_evict_list, &pet_list); + } + cfs_spin_unlock(&pet_lock); - wake_up(&pet_waitq); + cfs_waitq_signal(&pet_waitq); return 0; } @@ -418,25 +615,27 @@ static int ping_evictor_main(void *arg) time_t expire_time; ENTRY; - ptlrpc_daemonize("ll_evictor"); + cfs_daemonize_ctxt("ll_evictor"); CDEBUG(D_HA, "Starting Ping Evictor\n"); - pet_exp = NULL; pet_state = PET_READY; while (1) { - l_wait_event(pet_waitq, pet_exp || + l_wait_event(pet_waitq, (!cfs_list_empty(&pet_list)) || (pet_state == PET_TERMINATE), &lwi); - if (pet_state == PET_TERMINATE) + + /* loop until all obd's will be removed */ + if ((pet_state == PET_TERMINATE) && cfs_list_empty(&pet_list)) break; /* we only get here if pet_exp != NULL, and the end of this * loop is the only place which sets it NULL again, so lock * is not strictly necessary. */ - spin_lock(&pet_lock); - obd = pet_exp->exp_obd; - spin_unlock(&pet_lock); + cfs_spin_lock(&pet_lock); + obd = cfs_list_entry(pet_list.next, struct obd_device, + obd_evict_list); + cfs_spin_unlock(&pet_lock); - expire_time = cfs_time_current_sec() - (3 * obd_timeout / 2); + expire_time = cfs_time_current_sec() - PING_EVICT_TIMEOUT; CDEBUG(D_HA, "evicting all exports of obd %s older than %ld\n", obd->obd_name, expire_time); @@ -445,13 +644,14 @@ static int ping_evictor_main(void *arg) * the obd lock (class_unlink_export), which means we can't * lose the last ref on the export. If they've already been * removed from the list, we won't find them here. */ - spin_lock(&obd->obd_dev_lock); - while (!list_empty(&obd->obd_exports_timed)) { - exp = list_entry(obd->obd_exports_timed.next, - struct obd_export,exp_obd_chain_timed); + cfs_spin_lock(&obd->obd_dev_lock); + while (!cfs_list_empty(&obd->obd_exports_timed)) { + exp = cfs_list_entry(obd->obd_exports_timed.next, + struct obd_export, + exp_obd_chain_timed); if (expire_time > exp->exp_last_request_time) { class_export_get(exp); - spin_unlock(&obd->obd_dev_lock); + cfs_spin_unlock(&obd->obd_dev_lock); LCONSOLE_WARN("%s: haven't heard from client %s" " (at %s) in %ld seconds. I think" " it's dead, and I am evicting" @@ -469,19 +669,19 @@ static int ping_evictor_main(void *arg) exp->exp_last_request_time); class_fail_export(exp); class_export_put(exp); - spin_lock(&obd->obd_dev_lock); + cfs_spin_lock(&obd->obd_dev_lock); } else { /* List is sorted, so everyone below is ok */ break; } } - spin_unlock(&obd->obd_dev_lock); + cfs_spin_unlock(&obd->obd_dev_lock); - class_export_put(pet_exp); + cfs_spin_lock(&pet_lock); + cfs_list_del_init(&obd->obd_evict_list); + cfs_spin_unlock(&pet_lock); - spin_lock(&pet_lock); - pet_exp = NULL; - spin_unlock(&pet_lock); + class_decref(obd, "evictor", obd); } CDEBUG(D_HA, "Exiting Ping Evictor\n"); @@ -495,9 +695,9 @@ void ping_evictor_start(void) if (++pet_refcount > 1) return; - init_waitqueue_head(&pet_waitq); + cfs_waitq_init(&pet_waitq); - rc = cfs_kernel_thread(ping_evictor_main, NULL, CLONE_VM | CLONE_FILES); + rc = cfs_create_thread(ping_evictor_main, NULL, CFS_DAEMON_FLAGS); if (rc < 0) { pet_refcount--; CERROR("Cannot start ping evictor thread: %d\n", rc); @@ -511,7 +711,7 @@ void ping_evictor_stop(void) return; pet_state = PET_TERMINATE; - wake_up(&pet_waitq); + cfs_waitq_signal(&pet_waitq); } EXPORT_SYMBOL(ping_evictor_stop); #else /* !__KERNEL__ */ @@ -533,7 +733,8 @@ static int pinger_check_rpcs(void *arg) cfs_time_t curtime = cfs_time_current(); struct ptlrpc_request *req; struct ptlrpc_request_set *set; - struct list_head *iter; + cfs_list_t *iter; + struct obd_import *imp; struct pinger_data *pd = &pinger_args; int rc; @@ -546,7 +747,7 @@ static int pinger_check_rpcs(void *arg) } /* have we reached ping point? */ - if (!pd->pd_set && time_before(curtime, pd->pd_next_ping)) { + if (!pd->pd_set && cfs_time_before(curtime, pd->pd_next_ping)) { pd->pd_recursion--; return 0; } @@ -565,19 +766,19 @@ static int pinger_check_rpcs(void *arg) set = pd->pd_set; /* add rpcs into set */ - mutex_down(&pinger_sem); - list_for_each(iter, &pinger_imports) { - struct obd_import *imp = - list_entry(iter, struct obd_import, imp_pinger_chain); + cfs_mutex_lock(&pinger_mutex); + cfs_list_for_each(iter, &pinger_imports) { + struct obd_import *imp = cfs_list_entry(iter, struct obd_import, + imp_pinger_chain); int generation, level; if (cfs_time_aftereq(pd->pd_this_ping, imp->imp_next_ping - 5 * CFS_TICK)) { /* Add a ping. */ - spin_lock(&imp->imp_lock); + cfs_spin_lock(&imp->imp_lock); generation = imp->imp_generation; level = imp->imp_state; - spin_unlock(&imp->imp_lock); + cfs_spin_unlock(&imp->imp_lock); if (level != LUSTRE_IMP_FULL) { CDEBUG(D_HA, @@ -599,7 +800,7 @@ static int pinger_check_rpcs(void *arg) req->rq_no_resend = 1; ptlrpc_request_set_replen(req); req->rq_send_state = LUSTRE_IMP_FULL; - req->rq_phase = RQ_PHASE_RPC; + ptlrpc_rqphase_move(req, RQ_PHASE_RPC); req->rq_import_generation = generation; ptlrpc_set_add_req(set, req); } else { @@ -609,16 +810,16 @@ static int pinger_check_rpcs(void *arg) } } pd->pd_this_ping = curtime; - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); /* Might be empty, that's OK. */ - if (set->set_remaining == 0) + if (cfs_atomic_read(&set->set_remaining) == 0) CDEBUG(D_RPCTRACE, "nothing to ping\n"); - list_for_each(iter, &set->set_requests) { + cfs_list_for_each(iter, &set->set_requests) { struct ptlrpc_request *req = - list_entry(iter, struct ptlrpc_request, - rq_set_chain); + cfs_list_entry(iter, struct ptlrpc_request, + rq_set_chain); DEBUG_REQ(D_RPCTRACE, req, "pinging %s->%s", req->rq_import->imp_obd->obd_uuid.uuid, obd2cli_tgt(req->rq_import->imp_obd)); @@ -626,7 +827,7 @@ static int pinger_check_rpcs(void *arg) } do_check_set: - rc = ptlrpc_check_set(set); + rc = ptlrpc_check_set(NULL, set); /* not finished, and we are not expired, simply return */ if (!rc && cfs_time_before(curtime, cfs_time_add(pd->pd_this_ping, @@ -637,27 +838,33 @@ do_check_set: } /* Expire all the requests that didn't come back. */ - mutex_down(&pinger_sem); - list_for_each(iter, &set->set_requests) { - req = list_entry(iter, struct ptlrpc_request, - rq_set_chain); + cfs_mutex_lock(&pinger_mutex); + cfs_list_for_each(iter, &set->set_requests) { + req = cfs_list_entry(iter, struct ptlrpc_request, + rq_set_chain); if (req->rq_phase == RQ_PHASE_COMPLETE) continue; - req->rq_phase = RQ_PHASE_COMPLETE; - atomic_dec(&req->rq_import->imp_inflight); - set->set_remaining--; - /* If it was disconnected, don't sweat it. */ - if (list_empty(&req->rq_import->imp_pinger_chain)) { - ptlrpc_unregister_reply(req); - continue; - } + CDEBUG(D_RPCTRACE, "Pinger initiate expire request(%p)\n", + req); - CDEBUG(D_RPCTRACE, "pinger initiate expire_one_request\n"); - ptlrpc_expire_one_request(req); + /* This will also unregister reply. */ + ptlrpc_expire_one_request(req, 0); + + /* We're done with this req, let's finally move it to complete + * phase and take care of inflights. */ + ptlrpc_rqphase_move(req, RQ_PHASE_COMPLETE); + imp = req->rq_import; + cfs_spin_lock(&imp->imp_lock); + if (!cfs_list_empty(&req->rq_list)) { + cfs_list_del_init(&req->rq_list); + cfs_atomic_dec(&imp->imp_inflight); + } + cfs_spin_unlock(&imp->imp_lock); + cfs_atomic_dec(&set->set_remaining); } - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); ptlrpc_set_destroy(set); pd->pd_set = NULL; @@ -698,32 +905,60 @@ int ptlrpc_stop_pinger(void) void ptlrpc_pinger_sending_on_import(struct obd_import *imp) { #ifdef ENABLE_PINGER - mutex_down(&pinger_sem); - ptlrpc_update_next_ping(imp); + cfs_mutex_lock(&pinger_mutex); + ptlrpc_update_next_ping(imp, 0); if (pinger_args.pd_set == NULL && - time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) { + cfs_time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) { CDEBUG(D_HA, "set next ping to "CFS_TIME_T"(cur "CFS_TIME_T")\n", imp->imp_next_ping, cfs_time_current()); pinger_args.pd_next_ping = imp->imp_next_ping; } - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); +#endif +} + +void ptlrpc_pinger_commit_expected(struct obd_import *imp) +{ +#ifdef ENABLE_PINGER + cfs_mutex_lock(&pinger_mutex); + ptlrpc_update_next_ping(imp, 1); + if (pinger_args.pd_set == NULL && + cfs_time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) { + CDEBUG(D_HA,"set next ping to "CFS_TIME_T"(cur "CFS_TIME_T")\n", + imp->imp_next_ping, cfs_time_current()); + pinger_args.pd_next_ping = imp->imp_next_ping; + } + cfs_mutex_unlock(&pinger_mutex); #endif } +int ptlrpc_add_timeout_client(int time, enum timeout_event event, + timeout_cb_t cb, void *data, + cfs_list_t *obd_list) +{ + return 0; +} + +int ptlrpc_del_timeout_client(cfs_list_t *obd_list, + enum timeout_event event) +{ + return 0; +} + int ptlrpc_pinger_add_import(struct obd_import *imp) { ENTRY; - if (!list_empty(&imp->imp_pinger_chain)) + if (!cfs_list_empty(&imp->imp_pinger_chain)) RETURN(-EALREADY); CDEBUG(D_HA, "adding pingable import %s->%s\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); ptlrpc_pinger_sending_on_import(imp); - mutex_down(&pinger_sem); - list_add_tail(&imp->imp_pinger_chain, &pinger_imports); + cfs_mutex_lock(&pinger_mutex); + cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports); class_import_get(imp); - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); RETURN(0); } @@ -731,15 +966,15 @@ int ptlrpc_pinger_add_import(struct obd_import *imp) int ptlrpc_pinger_del_import(struct obd_import *imp) { ENTRY; - if (list_empty(&imp->imp_pinger_chain)) + if (cfs_list_empty(&imp->imp_pinger_chain)) RETURN(-ENOENT); - mutex_down(&pinger_sem); - list_del_init(&imp->imp_pinger_chain); + cfs_mutex_lock(&pinger_mutex); + cfs_list_del_init(&imp->imp_pinger_chain); CDEBUG(D_HA, "removing pingable import %s->%s\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); class_import_put(imp); - mutex_up(&pinger_sem); + cfs_mutex_unlock(&pinger_mutex); RETURN(0); } @@ -749,15 +984,17 @@ void ptlrpc_pinger_wake_up() /* XXX force pinger to run, if needed */ struct obd_import *imp; ENTRY; - list_for_each_entry(imp, &pinger_imports, imp_pinger_chain) { + cfs_list_for_each_entry(imp, &pinger_imports, imp_pinger_chain) { CDEBUG(D_RPCTRACE, "checking import %s->%s\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd)); #ifdef ENABLE_LIBLUSTRE_RECOVERY - if (imp->imp_state == LUSTRE_IMP_DISCON && !imp->imp_deactive) + if (imp->imp_state == LUSTRE_IMP_DISCON && + !imp_is_deactive(imp)) #else /*XXX only recover for the initial connection */ if (!lustre_handle_is_used(&imp->imp_remote_handle) && - imp->imp_state == LUSTRE_IMP_DISCON && !imp->imp_deactive) + imp->imp_state == LUSTRE_IMP_DISCON && + !imp_is_deactive(imp)) #endif ptlrpc_initiate_recovery(imp); else if (imp->imp_state != LUSTRE_IMP_FULL) @@ -765,7 +1002,7 @@ void ptlrpc_pinger_wake_up() "state %d, deactive %d\n", imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd), imp->imp_state, - imp->imp_deactive); + imp_is_deactive(imp)); } EXIT; #endif