return 0;
master_here:
- mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
- task_pid_nr(dlm->dlm_reco_thread_task),
- dlm->name, dlm->reco.dead_node, dlm->node_num);
+ mlog(ML_NOTICE, "(%d) Node %u is the Recovery Master for the Dead Node "
+ "%u for Domain %s\n", task_pid_nr(dlm->dlm_reco_thread_task),
+ dlm->node_num, dlm->reco.dead_node, dlm->name);
status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
if (status < 0) {
(ml->type == LKM_EXMODE ||
memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
mlog(ML_ERROR, "mismatched lvbs!\n");
- __dlm_print_one_lock_resource(lock->lockres);
+ dlm_print_one_lock_resource(lock->lockres);
BUG();
}
memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
(struct dlm_migratable_lockres *)msg->buf;
int ret = 0;
u8 real_master;
+ u8 extra_refs = 0;
char *buf = NULL;
struct dlm_work_item *item = NULL;
struct dlm_lock_resource *res = NULL;
__dlm_insert_lockres(dlm, res);
spin_unlock(&dlm->spinlock);
+ /* Add an extra ref for this lock-less lockres lest the
+ * dlm_thread purges it before we get the chance to add
+ * locks to it */
+ dlm_lockres_get(res);
+
+ /* There are three refs that need to be put.
+ * 1. Taken above.
+ * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
+ * 3. dlm_lookup_lockres()
+ * The first one is handled at the end of this function. The
+ * other two are handled in the worker thread after locks have
+ * been attached. Yes, we don't wait for purge time to match
+ * kref_init. The lockres will still have atleast one ref
+ * added because it is in the hash __dlm_insert_lockres() */
+ extra_refs++;
+
/* now that the new lockres is inserted,
* make it usable by other processes */
spin_lock(&res->spinlock);
res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
spin_unlock(&res->spinlock);
wake_up(&res->wq);
-
- /* add an extra ref for just-allocated lockres
- * otherwise the lockres will be purged immediately */
- dlm_lockres_get(res);
}
/* at this point we have allocated everything we need,
dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
item->u.ml.lockres = res; /* already have a ref */
item->u.ml.real_master = real_master;
+ item->u.ml.extra_ref = extra_refs;
spin_lock(&dlm->work_lock);
list_add_tail(&item->list, &dlm->work_list);
spin_unlock(&dlm->work_lock);
queue_work(dlm->dlm_worker, &dlm->dispatched_work);
leave:
+ /* One extra ref taken needs to be put here */
+ if (extra_refs)
+ dlm_lockres_put(res);
+
dlm_put(dlm);
if (ret < 0) {
if (buf)
static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
{
- struct dlm_ctxt *dlm = data;
+ struct dlm_ctxt *dlm;
struct dlm_migratable_lockres *mres;
int ret = 0;
struct dlm_lock_resource *res;
u8 real_master;
+ u8 extra_ref;
dlm = item->dlm;
mres = (struct dlm_migratable_lockres *)data;
res = item->u.ml.lockres;
real_master = item->u.ml.real_master;
+ extra_ref = item->u.ml.extra_ref;
if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
/* this case is super-rare. only occurs if
}
leave:
+ /* See comment in dlm_mig_lockres_handler() */
+ if (res) {
+ if (extra_ref)
+ dlm_lockres_put(res);
+ dlm_lockres_put(res);
+ }
kfree(data);
mlog_exit(ret);
}
/* retry!? */
BUG();
}
- }
+ } else /* put.. incase we are not the master */
+ dlm_lockres_put(res);
spin_unlock(&res->spinlock);
}
spin_unlock(&dlm->spinlock);
"Recovering res %s:%.*s, is already on recovery list!\n",
dlm->name, res->lockname.len, res->lockname.name);
list_del_init(&res->recovering);
+ dlm_lockres_put(res);
}
/* We need to hold a reference while on the recovery list */
dlm_lockres_get(res);
assert_spin_locked(&dlm->spinlock);
assert_spin_locked(&res->spinlock);
+ /* We do two dlm_lock_put(). One for removing from list and the other is
+ * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
+
/* TODO: check pending_asts, pending_basts here */
list_for_each_entry_safe(lock, next, &res->granted, list) {
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
freed++;
}
}
}
}
+ /* Clean up join state on node death. */
+ if (dlm->joining_node == idx) {
+ mlog(0, "Clearing join state for node %u\n", idx);
+ __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+ }
+
/* check to see if the node is already considered dead */
if (!test_bit(idx, dlm->live_nodes_map)) {
mlog(0, "for domain %s, node %d is already dead. "
clear_bit(idx, dlm->live_nodes_map);
- /* Clean up join state on node death. */
- if (dlm->joining_node == idx) {
- mlog(0, "Clearing join state for node %u\n", idx);
- __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
- }
-
/* make sure local cleanup occurs before the heartbeat events */
if (!test_bit(idx, dlm->recovery_map))
dlm_do_local_recovery_cleanup(dlm, idx);
if (!dlm_grab(dlm))
return;
+ /*
+ * This will notify any dlm users that a node in our domain
+ * went away without notifying us first.
+ */
+ if (test_bit(idx, dlm->domain_map))
+ dlm_fire_domain_eviction_callbacks(dlm, idx);
+
spin_lock(&dlm->spinlock);
__dlm_hb_node_down(dlm, idx);
spin_unlock(&dlm->spinlock);