static DECLARE_MUTEX(pinger_sem);
static struct list_head pinger_imports = LIST_HEAD_INIT(pinger_imports);
-#ifdef __KERNEL__
static struct ptlrpc_thread *pinger_thread = NULL;
int ptlrpc_ping(struct obd_import *imp)
RETURN(rc);
}
+#ifdef __KERNEL__
static int ptlrpc_pinger_main(void *arg)
{
struct ptlrpc_svc_data *data = (struct ptlrpc_svc_data *)arg;
#endif
}
-#else
+#else /* !__KERNEL__ */
+
/* XXX
* the current implementation of pinger in liblustre is not optimized
*/
int pd_recursion;
unsigned long pd_this_ping;
unsigned long pd_next_ping;
- struct ptlrpc_request_set *pd_set;
+ int pd_force_check;
} pinger_args;
static int pinger_check_rpcs(void *arg)
{
unsigned long curtime = time(NULL);
- 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_set && pd->pd_next_ping > curtime) {
+ if (pd->pd_next_ping > curtime && !pd->pd_force_check) {
pd->pd_recursion--;
return 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;
- }
+ if (pd->pd_force_check)
+ pd->pd_force_check = 0;
pd->pd_this_ping = curtime;
- pd->pd_set = ptlrpc_prep_set();
- set = pd->pd_set;
/* add rpcs into set */
down(&pinger_sem);
struct obd_import *imp =
list_entry(iter, struct obd_import,
imp_pinger_chain);
- int generation, level;
+ int level, force;
unsigned long flags;
- if (imp->imp_next_ping <= pd->pd_this_ping) {
- /* 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);
- if (level != LUSTRE_IMP_FULL) {
- CDEBUG(D_HA,
- "not pinging %s (in recovery)\n",
- imp->imp_target_uuid.uuid);
- continue;
+ 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);
}
- 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);
+ 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);
+ }
}
}
- 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 && curtime < pd->pd_this_ping + obd_timeout) {
- 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);
- ptlrpc_set_destroy(set);
- pd->pd_set = NULL;
-
- pd->pd_next_ping = pd->pd_this_ping + obd_timeout;
- pd->pd_this_ping = 0; /* XXX for debug */
+ pd->pd_next_ping = pd->pd_this_ping + (obd_timeout * HZ);
CDEBUG(D_HA, "finished a round ping\n");
pd->pd_recursion--;
{
down(&pinger_sem);
imp->imp_next_ping = time(NULL) + obd_timeout;
- if (pinger_args.pd_set == NULL &&
- pinger_args.pd_next_ping > imp->imp_next_ping) {
+ if (pinger_args.pd_next_ping > imp->imp_next_ping) {
CDEBUG(D_HA, "set next ping to %ld(cur %ld)\n",
imp->imp_next_ping, time(NULL));
pinger_args.pd_next_ping = imp->imp_next_ping;
void ptlrpc_pinger_wake_up()
{
-#ifdef ENABLE_PINGER
- /* XXX force pinger to run, if needed */
-#endif
+ pinger_args.pd_force_check = 1;
}
#endif /* !__KERNEL__ */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff