* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <obd_class.h>
#include "ptlrpc_internal.h"
-cfs_mutex_t pinger_mutex;
+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 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)
{
cfs_time_t timeout = PING_INTERVAL;
/* The timeout list is a increase order sorted list */
- cfs_mutex_lock(&pinger_mutex);
+ 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);
+ mutex_unlock(&pinger_mutex);
return cfs_time_sub(cfs_time_add(time, cfs_time_seconds(timeout)),
cfs_time_current());
#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;
- cfs_spin_lock(&imp->imp_lock);
- level = imp->imp_state;
- force = imp->imp_force_verify;
- if (force)
- imp->imp_force_verify = 0;
- cfs_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 for a while before trying recovery again */
} 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 timeout_item *item;
cfs_list_t *iter;
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
cfs_list_for_each_entry(item, &timeout_list, ti_chain) {
item->ti_cb(item, item->ti_cb_data);
}
cfs_time_seconds(PING_INTERVAL))))
ptlrpc_update_next_ping(imp, 0);
}
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
/* update memory usage info */
obd_update_maxusage();
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 };
- int rc;
+ struct l_wait_info lwi = { 0 };
+ int rc;
#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);
- 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);
+ cfs_waitq_init(&pinger_thread.t_ctl_waitq);
+ cfs_waitq_init(&suspend_timeouts_waitq);
- strcpy(pinger_thread->t_name, "ll_ping");
+ 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);
- OBD_FREE(pinger_thread, sizeof(*pinger_thread));
- pinger_thread = NULL;
- RETURN(rc);
- }
- l_wait_event(pinger_thread->t_ctl_waitq,
- thread_is_running(pinger_thread), &lwi);
-
- RETURN(0);
+ 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();
- cfs_mutex_lock(&pinger_mutex);
- thread_set_flags(pinger_thread, SVC_STOPPING);
- cfs_waitq_signal(&pinger_thread->t_ctl_waitq);
- cfs_mutex_unlock(&pinger_mutex);
+ ptlrpc_pinger_remove_timeouts();
- l_wait_event(pinger_thread->t_ctl_waitq,
- thread_is_stopped(pinger_thread), &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)
void ptlrpc_pinger_commit_expected(struct obd_import *imp)
{
- ptlrpc_update_next_ping(imp, 1);
+ 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)
if (!cfs_list_empty(&imp->imp_pinger_chain))
RETURN(-EALREADY);
- cfs_mutex_lock(&pinger_mutex);
+ 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 */
class_import_get(imp);
ptlrpc_pinger_wake_up();
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
if (cfs_list_empty(&imp->imp_pinger_chain))
RETURN(-ENOENT);
- cfs_mutex_lock(&pinger_mutex);
+ 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);
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
EXPORT_SYMBOL(ptlrpc_pinger_del_import);
{
struct timeout_item *ti;
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
ti = ptlrpc_pinger_register_timeout(time, event, cb, data);
if (!ti) {
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
return (-EINVAL);
}
cfs_list_add(obd_list, &ti->ti_obd_list);
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
return 0;
}
EXPORT_SYMBOL(ptlrpc_add_timeout_client);
if (cfs_list_empty(obd_list))
return 0;
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
cfs_list_del_init(obd_list);
/**
* If there are no obd attached to the timeout event
cfs_list_del(&ti->ti_chain);
OBD_FREE_PTR(ti);
}
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
return 0;
}
EXPORT_SYMBOL(ptlrpc_del_timeout_client);
{
struct timeout_item *item, *tmp;
- cfs_mutex_lock(&pinger_mutex);
+ 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);
+ mutex_unlock(&pinger_mutex);
return 0;
}
void ptlrpc_pinger_wake_up()
{
#ifdef ENABLE_PINGER
- thread_add_flags(pinger_thread, 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_state;
static cfs_waitq_t pet_waitq;
CFS_LIST_HEAD(pet_list);
-static cfs_spinlock_t pet_lock = CFS_SPIN_LOCK_UNLOCKED;
+static DEFINE_SPINLOCK(pet_lock);
int ping_evictor_wake(struct obd_export *exp)
{
- struct obd_device *obd;
-
- cfs_spin_lock(&pet_lock);
- if (pet_state != PET_READY) {
- /* eventually the new obd will call here again. */
- cfs_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);
- }
- cfs_spin_unlock(&pet_lock);
-
- cfs_waitq_signal(&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)
/* 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. */
- cfs_spin_lock(&pet_lock);
- obd = cfs_list_entry(pet_list.next, struct obd_device,
- obd_evict_list);
- cfs_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. */
- 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);
- cfs_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);
- cfs_spin_lock(&obd->obd_dev_lock);
- } else {
- /* List is sorted, so everyone below is ok */
- break;
- }
- }
- cfs_spin_unlock(&obd->obd_dev_lock);
-
- cfs_spin_lock(&pet_lock);
- cfs_list_del_init(&obd->obd_evict_list);
- cfs_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);
}
set = pd->pd_set;
/* add rpcs into set */
- cfs_mutex_lock(&pinger_mutex);
+ 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);
if (cfs_time_aftereq(pd->pd_this_ping,
imp->imp_next_ping - 5 * CFS_TICK)) {
/* Add a ping. */
- cfs_spin_lock(&imp->imp_lock);
- generation = imp->imp_generation;
- level = imp->imp_state;
- cfs_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;
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
/* Might be empty, that's OK. */
if (cfs_atomic_read(&set->set_remaining) == 0)
}
/* Expire all the requests that didn't come back. */
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
cfs_list_for_each(iter, &set->set_requests) {
req = cfs_list_entry(iter, struct ptlrpc_request,
rq_set_chain);
* 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);
- }
- cfs_mutex_unlock(&pinger_mutex);
+ 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
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
ptlrpc_update_next_ping(imp, 0);
if (pinger_args.pd_set == NULL &&
cfs_time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) {
imp->imp_next_ping, cfs_time_current());
pinger_args.pd_next_ping = imp->imp_next_ping;
}
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
#endif
}
void ptlrpc_pinger_commit_expected(struct obd_import *imp)
{
#ifdef ENABLE_PINGER
- cfs_mutex_lock(&pinger_mutex);
+ 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)) {
imp->imp_next_ping, cfs_time_current());
pinger_args.pd_next_ping = imp->imp_next_ping;
}
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
#endif
}
imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
ptlrpc_pinger_sending_on_import(imp);
- cfs_mutex_lock(&pinger_mutex);
+ mutex_lock(&pinger_mutex);
cfs_list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
class_import_get(imp);
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}
if (cfs_list_empty(&imp->imp_pinger_chain))
RETURN(-ENOENT);
- cfs_mutex_lock(&pinger_mutex);
+ 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);
- cfs_mutex_unlock(&pinger_mutex);
+ mutex_unlock(&pinger_mutex);
RETURN(0);
}