-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <obd_class.h>
#include "ptlrpc_internal.h"
-struct semaphore pinger_sem;
+static int suppress_pings;
+CFS_MODULE_PARM(suppress_pings, "i", int, 0644, "Suppress pings");
+
+struct mutex pinger_mutex;
static CFS_LIST_HEAD(pinger_imports);
-static struct list_head timeout_list = CFS_LIST_HEAD_INIT(timeout_list);
+static cfs_list_t timeout_list = CFS_LIST_HEAD_INIT(timeout_list);
+
+int ptlrpc_pinger_suppress_pings()
+{
+ return suppress_pings;
+}
+EXPORT_SYMBOL(ptlrpc_pinger_suppress_pings);
+
struct ptlrpc_request *
ptlrpc_prep_ping(struct obd_import *imp)
{
DEBUG_REQ(D_INFO, req, "pinging %s->%s",
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
- ptlrpcd_add_req(req, PSCOPE_OTHER);
+ 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
- int time = 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 :
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;
cfs_duration_t pinger_check_timeout(cfs_time_t time)
cfs_time_t timeout = PING_INTERVAL;
/* The timeout list is a increase order sorted list */
- mutex_down(&pinger_sem);
- list_for_each_entry(item, &timeout_list, ti_chain) {
+ 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;
}
- mutex_up(&pinger_sem);
+ mutex_unlock(&pinger_mutex);
return cfs_time_sub(cfs_time_add(time, cfs_time_seconds(timeout)),
cfs_time_current());
}
-#ifdef __KERNEL__
-static wait_queue_head_t suspend_timeouts_waitq;
-#endif
+static cfs_waitq_t suspend_timeouts_waitq;
cfs_time_t ptlrpc_suspend_wakeup_time(void)
{
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
}
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;
}
{
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;
}
#ifdef __KERNEL__
+static bool ir_up;
+
+void ptlrpc_pinger_ir_up(void)
+{
+ CDEBUG(D_HA, "IR up\n");
+ ir_up = true;
+}
+EXPORT_SYMBOL(ptlrpc_pinger_ir_up);
+
+void ptlrpc_pinger_ir_down(void)
+{
+ CDEBUG(D_HA, "IR down\n");
+ ir_up = false;
+}
+EXPORT_SYMBOL(ptlrpc_pinger_ir_down);
+
static void ptlrpc_pinger_process_import(struct obd_import *imp,
unsigned long this_ping)
{
- int force, level;
+ int level;
+ int force;
+ int force_next;
+ int suppress;
- 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);
+ spin_lock(&imp->imp_lock);
- CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA,
- "level %s/%u force %u deactive %u pingable %u\n",
- ptlrpc_import_state_name(level), level,
- force, imp->imp_deactive, imp->imp_pingable);
+ level = imp->imp_state;
+ force = imp->imp_force_verify;
+ force_next = imp->imp_force_next_verify;
+ /*
+ * This will be used below only if the import is "FULL".
+ */
+ suppress = ir_up && OCD_HAS_FLAG(&imp->imp_connect_data, PINGLESS);
- if (cfs_time_aftereq(imp->imp_next_ping - 5 * CFS_TICK,
- this_ping) && force == 0)
- return;
+ imp->imp_force_verify = 0;
+
+ if (cfs_time_aftereq(imp->imp_next_ping - 5 * CFS_TICK, this_ping) &&
+ !force) {
+ spin_unlock(&imp->imp_lock);
+ return;
+ }
+
+ imp->imp_force_next_verify = 0;
+
+ spin_unlock(&imp->imp_lock);
+
+ CDEBUG(level == LUSTRE_IMP_FULL ? D_INFO : D_HA, "%s->%s: level %s/%u "
+ "force %u force_next %u deactive %u pingable %u suppress %u\n",
+ imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd),
+ ptlrpc_import_state_name(level), level, force, force_next,
+ imp->imp_deactive, imp->imp_pingable, suppress);
if (level == LUSTRE_IMP_DISCON && !imp_is_deactive(imp)) {
- /* wait at least a timeout before trying recovery again */
+ /* 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_is_deactive(imp)) {
- CDEBUG(D_HA, "not pinging %s (in recovery "
- " or recovery disabled: %s)\n",
- obd2cli_tgt(imp->imp_obd),
- ptlrpc_import_state_name(level));
- } else if (imp->imp_pingable || force) {
- ptlrpc_ping(imp);
- }
+ CDEBUG(D_HA, "%s->%s: not pinging (in recovery "
+ "or recovery disabled: %s)\n",
+ imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd),
+ ptlrpc_import_state_name(level));
+ } else if ((imp->imp_pingable && !suppress) || force_next || force) {
+ ptlrpc_ping(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. */
struct l_wait_info lwi;
cfs_duration_t time_to_next_wake;
struct timeout_item *item;
- struct list_head *iter;
+ cfs_list_t *iter;
- mutex_down(&pinger_sem);
- list_for_each_entry(item, &timeout_list, ti_chain) {
+ mutex_lock(&pinger_mutex);
+ cfs_list_for_each_entry(item, &timeout_list, ti_chain) {
item->ti_cb(item, item->ti_cb_data);
}
- list_for_each(iter, &pinger_imports) {
+ 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 */
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);
+ mutex_unlock(&pinger_mutex);
/* update memory usage info */
obd_update_maxusage();
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 wakeup in "CFS_DURATION_T" ("CFS_TIME_T")\n",
- time_to_next_wake,
- cfs_time_add(this_ping, cfs_time_seconds(PING_INTERVAL)));
+ 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_wake, cfs_time_seconds(1)),
- NULL, NULL);
+ lwi = LWI_TIMEOUT(max_t(cfs_duration_t,
+ 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());
return 0;
}
-static struct ptlrpc_thread *pinger_thread = NULL;
+static struct ptlrpc_thread pinger_thread;
int ptlrpc_start_pinger(void)
{
- struct l_wait_info lwi = { 0 };
- struct ptlrpc_svc_data d;
- int rc;
+ struct l_wait_info lwi = { 0 };
+ int rc;
#ifndef ENABLE_PINGER
- return 0;
+ return 0;
#endif
- ENTRY;
-
- if (pinger_thread != NULL)
- RETURN(-EALREADY);
-
- OBD_ALLOC_PTR(pinger_thread);
- if (pinger_thread == NULL)
- RETURN(-ENOMEM);
- cfs_waitq_init(&pinger_thread->t_ctl_waitq);
- cfs_waitq_init(&suspend_timeouts_waitq);
-
- d.name = "ll_ping";
- d.thread = pinger_thread;
-
- /* 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_kernel_thread(ptlrpc_pinger_main, &d, CLONE_VM | CLONE_FILES);
- if (rc < 0) {
- CERROR("cannot start thread: %d\n", rc);
- OBD_FREE(pinger_thread, sizeof(*pinger_thread));
- pinger_thread = NULL;
- RETURN(rc);
- }
- l_wait_event(pinger_thread->t_ctl_waitq,
- pinger_thread->t_flags & SVC_RUNNING, &lwi);
-
- RETURN(0);
+ ENTRY;
+
+ if (!thread_is_init(&pinger_thread) &&
+ !thread_is_stopped(&pinger_thread))
+ RETURN(-EALREADY);
+
+ cfs_waitq_init(&pinger_thread.t_ctl_waitq);
+ cfs_waitq_init(&suspend_timeouts_waitq);
+
+ 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 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);
+ RETURN(rc);
+ }
+ l_wait_event(pinger_thread.t_ctl_waitq,
+ thread_is_running(&pinger_thread), &lwi);
+
+ if (suppress_pings)
+ CWARN("Pings will be suppressed at the request of the "
+ "administrator. The configuration shall meet the "
+ "additional requirements described in the manual. "
+ "(Search for the \"suppress_pings\" kernel module "
+ "parameter.)\n");
+
+ RETURN(0);
}
int ptlrpc_pinger_remove_timeouts(void);
int ptlrpc_stop_pinger(void)
{
- struct l_wait_info lwi = { 0 };
- int rc = 0;
+ struct l_wait_info lwi = { 0 };
#ifndef ENABLE_PINGER
- return 0;
+ return 0;
#endif
- ENTRY;
+ ENTRY;
- if (pinger_thread == NULL)
- RETURN(-EALREADY);
+ if (thread_is_init(&pinger_thread) ||
+ thread_is_stopped(&pinger_thread))
+ RETURN(-EALREADY);
- ptlrpc_pinger_remove_timeouts();
- mutex_down(&pinger_sem);
- pinger_thread->t_flags = SVC_STOPPING;
- cfs_waitq_signal(&pinger_thread->t_ctl_waitq);
- mutex_up(&pinger_sem);
+ ptlrpc_pinger_remove_timeouts();
- l_wait_event(pinger_thread->t_ctl_waitq,
- (pinger_thread->t_flags & SVC_STOPPED), &lwi);
+ thread_set_flags(&pinger_thread, SVC_STOPPING);
+ cfs_waitq_signal(&pinger_thread.t_ctl_waitq);
- OBD_FREE_PTR(pinger_thread);
- pinger_thread = NULL;
- RETURN(rc);
+ l_wait_event(pinger_thread.t_ctl_waitq,
+ thread_is_stopped(&pinger_thread), &lwi);
+ RETURN(0);
}
void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
{
- ptlrpc_update_next_ping(imp);
+ ptlrpc_update_next_ping(imp, 0);
+}
+EXPORT_SYMBOL(ptlrpc_pinger_sending_on_import);
+
+void ptlrpc_pinger_commit_expected(struct obd_import *imp)
+{
+ ptlrpc_update_next_ping(imp, 1);
+ LASSERT_SPIN_LOCKED(&imp->imp_lock);
+ /*
+ * Avoid reading stale imp_connect_data. When not sure if pings are
+ * expected or not on next connection, we assume they are not and force
+ * one anyway to guarantee the chance of updating
+ * imp_peer_committed_transno.
+ */
+ if (imp->imp_state != LUSTRE_IMP_FULL ||
+ OCD_HAS_FLAG(&imp->imp_connect_data, PINGLESS))
+ imp->imp_force_next_verify = 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);
+ 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);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
+EXPORT_SYMBOL(ptlrpc_pinger_add_import);
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);
+ 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);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
+EXPORT_SYMBOL(ptlrpc_pinger_del_import);
/**
* Register a timeout callback to the pinger list, and the callback will
{
struct timeout_item *item, *tmp;
- LASSERT_SEM_LOCKED(&pinger_sem);
+ LASSERT_MUTEX_LOCKED(&pinger_mutex);
- list_for_each_entry(item, &timeout_list, ti_chain)
+ 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) {
- list_for_each_entry_reverse(tmp, &timeout_list, ti_chain) {
+ cfs_list_for_each_entry_reverse(tmp, &timeout_list, ti_chain) {
if (tmp->ti_timeout < time) {
- list_add(&item->ti_chain, &tmp->ti_chain);
+ cfs_list_add(&item->ti_chain, &tmp->ti_chain);
goto out;
}
}
- list_add(&item->ti_chain, &timeout_list);
+ cfs_list_add(&item->ti_chain, &timeout_list);
}
out:
return item;
*/
int ptlrpc_add_timeout_client(int time, enum timeout_event event,
timeout_cb_t cb, void *data,
- struct list_head *obd_list)
+ cfs_list_t *obd_list)
{
struct timeout_item *ti;
- mutex_down(&pinger_sem);
+ mutex_lock(&pinger_mutex);
ti = ptlrpc_pinger_register_timeout(time, event, cb, data);
if (!ti) {
- mutex_up(&pinger_sem);
+ mutex_unlock(&pinger_mutex);
return (-EINVAL);
}
- list_add(obd_list, &ti->ti_obd_list);
- mutex_up(&pinger_sem);
+ cfs_list_add(obd_list, &ti->ti_obd_list);
+ mutex_unlock(&pinger_mutex);
return 0;
}
+EXPORT_SYMBOL(ptlrpc_add_timeout_client);
-int ptlrpc_del_timeout_client(struct list_head *obd_list,
+int ptlrpc_del_timeout_client(cfs_list_t *obd_list,
enum timeout_event event)
{
struct timeout_item *ti = NULL, *item;
- if (list_empty(obd_list))
+ if (cfs_list_empty(obd_list))
return 0;
- mutex_down(&pinger_sem);
- list_del_init(obd_list);
+ 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
*/
- list_for_each_entry(item, &timeout_list, ti_chain) {
+ 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 (list_empty(&ti->ti_obd_list)) {
- list_del(&ti->ti_chain);
+ if (cfs_list_empty(&ti->ti_obd_list)) {
+ cfs_list_del(&ti->ti_chain);
OBD_FREE_PTR(ti);
}
- mutex_up(&pinger_sem);
+ mutex_unlock(&pinger_mutex);
return 0;
}
+EXPORT_SYMBOL(ptlrpc_del_timeout_client);
int ptlrpc_pinger_remove_timeouts(void)
{
struct timeout_item *item, *tmp;
- mutex_down(&pinger_sem);
- list_for_each_entry_safe(item, tmp, &timeout_list, ti_chain) {
- LASSERT(list_empty(&item->ti_obd_list));
- list_del(&item->ti_chain);
+ 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);
}
- mutex_up(&pinger_sem);
+ mutex_unlock(&pinger_mutex);
return 0;
}
void ptlrpc_pinger_wake_up()
{
#ifdef ENABLE_PINGER
- pinger_thread->t_flags |= SVC_EVENT;
- cfs_waitq_signal(&pinger_thread->t_ctl_waitq);
+ thread_add_flags(&pinger_thread, SVC_EVENT);
+ cfs_waitq_signal(&pinger_thread.t_ctl_waitq);
#endif
}
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 DEFINE_SPINLOCK(pet_lock);
int ping_evictor_wake(struct obd_export *exp)
{
- spin_lock(&pet_lock);
- if (pet_exp || (pet_state != PET_READY)) {
- /* eventually the new obd will call here again. */
- 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);
-
- wake_up(&pet_waitq);
- return 0;
+ struct obd_device *obd;
+
+ spin_lock(&pet_lock);
+ if (pet_state != PET_READY) {
+ /* eventually the new obd will call here again. */
+ spin_unlock(&pet_lock);
+ return 1;
+ }
+
+ 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);
+ }
+ spin_unlock(&pet_lock);
+
+ cfs_waitq_signal(&pet_waitq);
+ return 0;
}
static int ping_evictor_main(void *arg)
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);
+ spin_lock(&pet_lock);
+ obd = cfs_list_entry(pet_list.next, struct obd_device,
+ obd_evict_list);
+ spin_unlock(&pet_lock);
expire_time = cfs_time_current_sec() - PING_EVICT_TIMEOUT;
* 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);
- if (expire_time > exp->exp_last_request_time) {
- class_export_get(exp);
- spin_unlock(&obd->obd_dev_lock);
- LCONSOLE_WARN("%s: haven't heard from client %s"
+ 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);
+ LCONSOLE_WARN("%s: haven't heard from client %s"
" (at %s) in %ld seconds. I think"
" it's dead, and I am evicting"
" it. exp %p, cur %ld expire %ld"
exp->exp_last_request_time);
class_fail_export(exp);
class_export_put(exp);
- spin_lock(&obd->obd_dev_lock);
- } else {
- /* List is sorted, so everyone below is ok */
- break;
- }
- }
- spin_unlock(&obd->obd_dev_lock);
-
- class_export_put(pet_exp);
-
- spin_lock(&pet_lock);
- pet_exp = NULL;
- spin_unlock(&pet_lock);
+ spin_lock(&obd->obd_dev_lock);
+ } else {
+ /* List is sorted, so everyone below is ok */
+ break;
+ }
+ }
+ spin_unlock(&obd->obd_dev_lock);
+
+ spin_lock(&pet_lock);
+ cfs_list_del_init(&obd->obd_evict_list);
+ spin_unlock(&pet_lock);
+
+ class_decref(obd, "evictor", obd);
}
CDEBUG(D_HA, "Exiting Ping Evictor\n");
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);
return;
pet_state = PET_TERMINATE;
- wake_up(&pet_waitq);
+ cfs_waitq_signal(&pet_waitq);
}
EXPORT_SYMBOL(ping_evictor_stop);
#else /* !__KERNEL__ */
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;
}
/* 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;
}
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);
+ 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);
- generation = imp->imp_generation;
- level = imp->imp_state;
- spin_unlock(&imp->imp_lock);
+ spin_lock(&imp->imp_lock);
+ generation = imp->imp_generation;
+ level = imp->imp_state;
+ spin_unlock(&imp->imp_lock);
if (level != LUSTRE_IMP_FULL) {
CDEBUG(D_HA,
}
}
pd->pd_this_ping = curtime;
- mutex_up(&pinger_sem);
+ 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));
}
/* 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);
+ 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;
* phase and take care of inflights. */
ptlrpc_rqphase_move(req, RQ_PHASE_COMPLETE);
imp = req->rq_import;
- spin_lock(&imp->imp_lock);
- if (!list_empty(&req->rq_list)) {
- list_del_init(&req->rq_list);
- atomic_dec(&imp->imp_inflight);
- }
- spin_unlock(&imp->imp_lock);
- set->set_remaining--;
- }
- mutex_up(&pinger_sem);
+ 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);
+ }
+ spin_unlock(&imp->imp_lock);
+ cfs_atomic_dec(&set->set_remaining);
+ }
+ mutex_unlock(&pinger_mutex);
ptlrpc_set_destroy(set);
pd->pd_set = NULL;
void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
{
#ifdef ENABLE_PINGER
- mutex_down(&pinger_sem);
- ptlrpc_update_next_ping(imp);
+ 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);
+ mutex_unlock(&pinger_mutex);
+#endif
+}
+
+void ptlrpc_pinger_commit_expected(struct obd_import *imp)
+{
+#ifdef ENABLE_PINGER
+ 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;
+ }
+ mutex_unlock(&pinger_mutex);
#endif
}
int ptlrpc_add_timeout_client(int time, enum timeout_event event,
timeout_cb_t cb, void *data,
- struct list_head *obd_list)
+ cfs_list_t *obd_list)
{
return 0;
}
-int ptlrpc_del_timeout_client(struct list_head *obd_list,
+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);
+ mutex_lock(&pinger_mutex);
+ cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
class_import_get(imp);
- mutex_up(&pinger_sem);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
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);
+ 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);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
/* 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
git://git.whamcloud.com / fs / lustre-release.git / blobdiff