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;
req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL,
NULL);
if (req) {
- DEBUG_REQ(D_HA, req, "pinging %s->%s",
+ DEBUG_REQ(D_INFO, req, "pinging %s->%s",
imp->imp_obd->obd_uuid.uuid,
imp->imp_target_uuid.uuid);
req->rq_no_resend = req->rq_no_delay = 1;
RETURN(rc);
}
+static inline void ptlrpc_update_next_ping(struct obd_import *imp)
+{
+ imp->imp_next_ping = jiffies + HZ *
+ (imp->imp_state == LUSTRE_IMP_DISCON ? 10 : PING_INTERVAL);
+}
+
+void ptlrpc_ping_import_soon(struct obd_import *imp)
+{
+ imp->imp_next_ping = jiffies;
+}
+
#ifdef __KERNEL__
static int ptlrpc_pinger_main(void *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 */
while (1) {
unsigned long this_ping = jiffies;
long time_to_next_ping;
- struct l_wait_info lwi = LWI_TIMEOUT(obd_timeout * HZ,
+ struct l_wait_info lwi = LWI_TIMEOUT(PING_INTERVAL * HZ,
NULL, NULL);
struct list_head *iter;
spin_lock_irqsave(&imp->imp_lock, flags);
level = imp->imp_state;
force = imp->imp_force_verify;
- if (force)
- imp->imp_force_verify = 0;
+ 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
+ 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);
+
+ if (force ||
+ /* if the next ping is within, say, 5 jiffies from
+ now, go ahead and ping. See note below. */
+ time_after_eq(this_ping, imp->imp_next_ping - 5)) {
+ if (level == LUSTRE_IMP_DISCON &&
+ !imp->imp_deactive) {
+ /* wait at least a timeout before
trying recovery again. */
- imp->imp_next_ping = jiffies +
- (obd_timeout * HZ);
+ imp->imp_next_ping = jiffies +
+ obd_timeout * HZ;
ptlrpc_initiate_recovery(imp);
- }
- else if (level != LUSTRE_IMP_FULL ||
+ } 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",
+ CDEBUG(D_HA, "not pinging %s "
+ "(in recovery: %s or recovery "
+ "disabled: %u/%u)\n",
imp->imp_target_uuid.uuid,
- ptlrpc_import_state_name(level));
- }
- else if (imp->imp_pingable || force) {
+ ptlrpc_import_state_name(level),
+ imp->imp_deactive,
+ imp->imp_obd->obd_no_recov);
+ } 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);
+ if (!imp->imp_pingable)
+ continue;
+ CDEBUG(D_INFO,
+ "don't need to ping %s (%lu > %lu)\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_next_ping, this_ping);
}
+
+ /* obd_timeout might have changed */
+ if (time_after(imp->imp_next_ping,
+ this_ping + PING_INTERVAL * HZ))
+ ptlrpc_update_next_ping(imp);
}
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));
+ time_to_next_ping = this_ping + (PING_INTERVAL * HZ) - jiffies;
+ /* 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 %lu (%lu)\n", time_to_next_ping,
+ this_ping + PING_INTERVAL * HZ);
if (time_to_next_ping > 0) {
- lwi = LWI_TIMEOUT(time_to_next_ping, NULL, NULL);
+ lwi = LWI_TIMEOUT(max_t(long, time_to_next_ping, HZ),
+ NULL, NULL);
l_wait_event(thread->t_ctl_waitq,
thread->t_flags & (SVC_STOPPING|SVC_EVENT),
&lwi);
return 0;
}
+static struct ptlrpc_thread *pinger_thread = NULL;
+
int ptlrpc_start_pinger(void)
{
struct l_wait_info lwi = { 0 };
void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
{
- down(&pinger_sem);
- imp->imp_next_ping = jiffies + (obd_timeout * HZ);
- up(&pinger_sem);
+ ptlrpc_update_next_ping(imp);
}
int ptlrpc_pinger_add_import(struct obd_import *imp)
down(&pinger_sem);
CDEBUG(D_HA, "adding pingable import %s->%s\n",
imp->imp_obd->obd_uuid.uuid, imp->imp_target_uuid.uuid);
- imp->imp_next_ping = jiffies + (obd_timeout * HZ);
+ ptlrpc_update_next_ping(imp);
/* XXX sort, blah blah */
list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
class_import_get(imp);
static struct pinger_data {
int pd_recursion;
- unsigned long pd_this_ping;
- unsigned long pd_next_ping;
- int pd_force_check;
+ unsigned long pd_this_ping; /* jiffies */
+ unsigned long pd_next_ping; /* jiffies */
+ struct ptlrpc_request_set *pd_set;
} pinger_args;
static int pinger_check_rpcs(void *arg)
{
- unsigned long curtime = time(NULL);
+ unsigned long curtime = jiffies;
+ struct ptlrpc_request *req;
+ struct ptlrpc_request_set *set;
struct list_head *iter;
struct pinger_data *pd = &pinger_args;
+ int rc;
/* prevent recursion */
if (pd->pd_recursion++) {
CDEBUG(D_HA, "pinger: recursion! quit\n");
+ LASSERT(pd->pd_set);
pd->pd_recursion--;
return 0;
}
/* have we reached ping point? */
- if (pd->pd_next_ping > curtime && !pd->pd_force_check) {
+ if (!pd->pd_set && time_before(curtime, pd->pd_next_ping)) {
pd->pd_recursion--;
return 0;
}
- if (pd->pd_force_check)
- pd->pd_force_check = 0;
+ /* if we have rpc_set already, continue processing it */
+ if (pd->pd_set) {
+ LASSERT(pd->pd_this_ping);
+ set = pd->pd_set;
+ goto do_check_set;
+ }
pd->pd_this_ping = curtime;
+ pd->pd_set = ptlrpc_prep_set();
+ set = pd->pd_set;
/* add rpcs into set */
down(&pinger_sem);
list_for_each(iter, &pinger_imports) {
struct obd_import *imp =
- list_entry(iter, struct obd_import,
- imp_pinger_chain);
- int level, force;
+ list_entry(iter, struct obd_import, imp_pinger_chain);
+ int generation, level;
unsigned long flags;
+ if (time_after_eq(pd->pd_this_ping, imp->imp_next_ping - 5)) {
+ /* Add a ping. */
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ generation = imp->imp_generation;
+ level = imp->imp_state;
+ spin_unlock_irqrestore(&imp->imp_lock, 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 <= pd->pd_this_ping || force) {
- if (level == LUSTRE_IMP_DISCON) {
- /* wait at least a timeout before
- trying recovery again. */
- imp->imp_next_ping = time(NULL) +
- (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);
+ if (level != LUSTRE_IMP_FULL) {
+ CDEBUG(D_HA,
+ "not pinging %s (in recovery)\n",
+ imp->imp_target_uuid.uuid);
+ continue;
}
- } 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, pd->pd_this_ping);
+ req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL,
+ NULL);
+ if (!req) {
+ CERROR("out of memory\n");
+ break;
}
+ req->rq_no_resend = 1;
+ req->rq_replen = lustre_msg_size(0, NULL);
+ req->rq_send_state = LUSTRE_IMP_FULL;
+ req->rq_phase = RQ_PHASE_RPC;
+ req->rq_import_generation = generation;
+ ptlrpc_set_add_req(set, req);
+ } else {
+ CDEBUG(D_HA, "don't need to ping %s (%lu > "
+ "%lu)\n", imp->imp_target_uuid.uuid,
+ imp->imp_next_ping, pd->pd_this_ping);
}
}
+ pd->pd_this_ping = curtime;
+ up(&pinger_sem);
+
+ /* Might be empty, that's OK. */
+ if (set->set_remaining == 0)
+ CDEBUG(D_HA, "nothing to ping\n");
+
+ list_for_each(iter, &set->set_requests) {
+ struct ptlrpc_request *req =
+ list_entry(iter, struct ptlrpc_request,
+ rq_set_chain);
+ DEBUG_REQ(D_HA, req, "pinging %s->%s",
+ req->rq_import->imp_obd->obd_uuid.uuid,
+ req->rq_import->imp_target_uuid.uuid);
+ (void)ptl_send_rpc(req);
+ }
+do_check_set:
+ rc = ptlrpc_check_set(set);
+
+ /* not finished, and we are not expired, simply return */
+ if (!rc && time_before(curtime, pd->pd_this_ping + PING_INTERVAL * HZ)) {
+ CDEBUG(D_HA, "not finished, but also not expired\n");
+ pd->pd_recursion--;
+ return 0;
+ }
+
+ /* Expire all the requests that didn't come back. */
+ down(&pinger_sem);
+ list_for_each(iter, &set->set_requests) {
+ req = list_entry(iter, struct ptlrpc_request,
+ rq_set_chain);
+
+ if (req->rq_replied)
+ continue;
+
+ req->rq_phase = RQ_PHASE_COMPLETE;
+ 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_HA, "pinger initiate expire_one_request\n");
+ ptlrpc_expire_one_request(req);
+ }
up(&pinger_sem);
- pd->pd_next_ping = pd->pd_this_ping + (obd_timeout * HZ);
+ ptlrpc_set_destroy(set);
+ pd->pd_set = NULL;
+
+ pd->pd_next_ping = pd->pd_this_ping + PING_INTERVAL * HZ;
+ pd->pd_this_ping = 0; /* XXX for debug */
CDEBUG(D_HA, "finished a round ping\n");
pd->pd_recursion--;
void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
{
down(&pinger_sem);
- imp->imp_next_ping = time(NULL) + obd_timeout;
- if (pinger_args.pd_next_ping > imp->imp_next_ping) {
+ ptlrpc_update_next_ping(imp);
+ if (pinger_args.pd_set == NULL &&
+ time_before(imp->imp_next_ping, pinger_args.pd_next_ping)) {
CDEBUG(D_HA, "set next ping to %ld(cur %ld)\n",
- imp->imp_next_ping, time(NULL));
+ imp->imp_next_ping, jiffies);
pinger_args.pd_next_ping = imp->imp_next_ping;
}
up(&pinger_sem);
void ptlrpc_pinger_wake_up()
{
- pinger_args.pd_force_check = 1;
+#ifdef ENABLE_PINGER
+ /* XXX force pinger to run, if needed */
+#endif
}
#endif /* !__KERNEL__ */