#include <linux/obd_class.h>
#include "ptlrpc_internal.h"
+#define PINGER_RATE 3 /* how many pings we'll do in obd_timeout period */
+
static DECLARE_MUTEX(pinger_sem);
static struct list_head pinger_imports = LIST_HEAD_INIT(pinger_imports);
-static struct ptlrpc_thread *pinger_thread = NULL;
-
int ptlrpc_ping(struct obd_import *imp)
{
struct ptlrpc_request *req;
int rc = 0;
ENTRY;
- req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL,
- NULL);
+ req = ptlrpc_prep_req(imp, LUSTRE_OBD_VERSION, OBD_PING, 0, NULL, NULL);
if (req) {
DEBUG_REQ(D_HA, req, "pinging %s->%s",
imp->imp_obd->obd_uuid.uuid,
imp->imp_target_uuid.uuid);
req->rq_no_resend = req->rq_no_delay = 1;
- req->rq_replen = lustre_msg_size(0,
- NULL);
+ req->rq_replen = lustre_msg_size(0, NULL);
+ req->rq_timeout = obd_timeout / PINGER_RATE;
ptlrpcd_add_req(req);
} else {
CERROR("OOM trying to ping %s->%s\n",
- imp->imp_obd->obd_uuid.uuid,
- imp->imp_target_uuid.uuid);
+ imp->imp_obd->obd_uuid.uuid,
+ imp->imp_target_uuid.uuid);
rc = -ENOMEM;
}
}
#ifdef __KERNEL__
+static inline int ptlrpc_next_ping(struct obd_import *imp)
+{
+ return jiffies + (obd_timeout / PINGER_RATE * HZ);
+}
+
+static inline int ptlrpc_next_reconnect(struct obd_import *imp)
+{
+ if (imp->imp_server_timeout)
+ return jiffies + (obd_timeout / 2 * HZ);
+ else
+ return jiffies + (obd_timeout * HZ);
+}
+
+static atomic_t suspend_timeouts = ATOMIC_INIT(0);
+static wait_queue_head_t suspend_timeouts_waitq;
+
+void ptlrpc_deactivate_timeouts(void)
+{
+ CDEBUG(D_HA, "deactivate timeouts\n");
+ atomic_inc(&suspend_timeouts);
+}
+
+void ptlrpc_activate_timeouts(void)
+{
+ CDEBUG(D_HA, "activate timeouts\n");
+ LASSERT(atomic_read(&suspend_timeouts) > 0);
+ if (atomic_dec_and_test(&suspend_timeouts))
+ wake_up(&suspend_timeouts_waitq);
+}
+
+int ptlrpc_check_suspend(void)
+{
+ if (atomic_read(&suspend_timeouts))
+ return 1;
+ return 0;
+}
+
+int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req)
+{
+ struct l_wait_info lwi;
+
+ if (atomic_read(&suspend_timeouts)) {
+ DEBUG_REQ(D_NET, req, "-- suspend %d regular timeout",
+ atomic_read(&suspend_timeouts));
+ lwi = LWI_INTR(NULL, NULL);
+ l_wait_event(suspend_timeouts_waitq,
+ atomic_read(&suspend_timeouts) == 0, &lwi);
+ DEBUG_REQ(D_NET, req, "-- recharge regular timeout");
+ return 1;
+ }
+ return 0;
+}
+
+static void ptlrpc_pinger_process_import(struct obd_import *imp,
+ unsigned long this_ping)
+{
+ unsigned long flags;
+ int force, level;
+
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ level = imp->imp_state;
+ force = imp->imp_force_verify;
+ if (force)
+ imp->imp_force_verify = 0;
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+
+ if ((imp->imp_next_ping - this_ping > 0) &&
+ (imp->imp_next_ping - this_ping > obd_timeout * HZ)) {
+ CWARN("wrong ping time %lu (current %lu)\n",
+ imp->imp_next_ping, this_ping);
+ imp->imp_next_ping = ptlrpc_next_reconnect(imp);
+ }
+
+ if (imp->imp_next_ping > this_ping && force == 0)
+ return;
+
+ if (level == LUSTRE_IMP_DISCON && !imp->imp_deactive) {
+ /* wait at least a timeout before trying recovery again */
+ imp->imp_next_ping = ptlrpc_next_reconnect(imp);
+ ptlrpc_initiate_recovery(imp);
+ } else if (level != LUSTRE_IMP_FULL || imp->imp_obd->obd_no_recov) {
+ CDEBUG(D_HA, "not pinging %s (in recovery "
+ " or recovery disabled: %s)\n",
+ imp->imp_target_uuid.uuid,
+ ptlrpc_import_state_name(level));
+ } else if (imp->imp_pingable || force) {
+ imp->imp_next_ping = ptlrpc_next_ping(imp);
+ ptlrpc_ping(imp);
+ }
+}
+
static int ptlrpc_pinger_main(void *arg)
{
struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
RECALC_SIGPENDING;
SIGNAL_MASK_UNLOCK(current, flags);
- THREAD_NAME(current->comm, "%s", data->name);
+ LASSERTF(strlen(data->name) < sizeof(current->comm),
+ "name %d > len %d\n",
+ (int)strlen(data->name), (int)sizeof(current->comm));
+ THREAD_NAME(current->comm, sizeof(current->comm) - 1, "%s", data->name);
unlock_kernel();
/* Record that the thread is running */
NULL, NULL);
struct list_head *iter;
+ time_to_next_ping = this_ping + (obd_timeout * HZ) - jiffies;
down(&pinger_sem);
list_for_each(iter, &pinger_imports) {
struct obd_import *imp =
list_entry(iter, struct obd_import,
imp_pinger_chain);
- int force, level;
- unsigned long flags;
-
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- level = imp->imp_state;
- force = imp->imp_force_verify;
- if (force)
- imp->imp_force_verify = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (imp->imp_next_ping <= this_ping || force) {
- if (level == LUSTRE_IMP_DISCON) {
- /* wait at least a timeout before
- trying recovery again. */
- imp->imp_next_ping = jiffies +
- (obd_timeout * HZ);
- ptlrpc_initiate_recovery(imp);
- }
- else if (level != LUSTRE_IMP_FULL ||
- imp->imp_obd->obd_no_recov) {
- CDEBUG(D_HA,
- "not pinging %s (in recovery "
- " or recovery disabled: %s)\n",
- imp->imp_target_uuid.uuid,
- ptlrpc_import_state_name(level));
- }
- else if (imp->imp_pingable || force) {
- ptlrpc_ping(imp);
- }
-
- } else {
- if (imp->imp_pingable)
- CDEBUG(D_HA, "don't need to ping %s "
- "(%lu > %lu)\n",
- imp->imp_target_uuid.uuid,
- imp->imp_next_ping, this_ping);
- }
+
+ ptlrpc_pinger_process_import(imp, this_ping);
+
+ CDEBUG(D_OTHER, "%s: pingable %d, next_ping %lu(%lu)\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_pingable, imp->imp_next_ping, jiffies);
+
+ if (imp->imp_pingable && imp->imp_next_ping &&
+ imp->imp_next_ping - jiffies < time_to_next_ping &&
+ imp->imp_next_ping > jiffies)
+ time_to_next_ping = imp->imp_next_ping - jiffies;
}
up(&pinger_sem);
/* Wait until the next ping time, or until we're stopped. */
- time_to_next_ping = this_ping + (obd_timeout * HZ) - jiffies;
CDEBUG(D_HA, "next ping in %lu (%lu)\n", time_to_next_ping,
this_ping + (obd_timeout * HZ));
if (time_to_next_ping > 0) {
return 0;
}
+static struct ptlrpc_thread *pinger_thread = NULL;
+
int ptlrpc_start_pinger(void)
{
struct l_wait_info lwi = { 0 };
#endif
ENTRY;
+ LASSERT(obd_timeout > PINGER_RATE);
+
if (pinger_thread != NULL)
RETURN(-EALREADY);
if (pinger_thread == NULL)
RETURN(-ENOMEM);
init_waitqueue_head(&pinger_thread->t_ctl_waitq);
+ init_waitqueue_head(&suspend_timeouts_waitq);
d.name = "ll_ping";
d.thread = pinger_thread;
void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
{
down(&pinger_sem);
- imp->imp_next_ping = jiffies + (obd_timeout * HZ);
+ imp->imp_next_ping = ptlrpc_next_ping(imp);
up(&pinger_sem);
}
if (level == LUSTRE_IMP_DISCON) {
/* wait at least a timeout before
trying recovery again. */
+ unsigned long timeout = obd_timeout;
+ if (imp->imp_server_timeout)
+ timeout = obd_timeout / 2;
imp->imp_next_ping = time(NULL) +
- (obd_timeout * HZ);
+ (timeout * HZ);
ptlrpc_initiate_recovery(imp);
}
else if (level != LUSTRE_IMP_FULL ||
git://git.whamcloud.com / fs / lustre-release.git / blobdiff