2 files changed, 182 insertions, 42 deletions

diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
index 9f21dd78536..ad2e41d6879 100644
--- a/fs/ocfs2/cluster/heartbeat.c
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -132,6 +132,33 @@ char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
 unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
 unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
 
+/*
+ * o2hb_dependent_users tracks the number of registered callbacks that depend
+ * on heartbeat. o2net and o2dlm are two entities that register this callback.
+ * However only o2dlm depends on the heartbeat. It does not want the heartbeat
+ * to stop while a dlm domain is still active.
+ */
+unsigned int o2hb_dependent_users;
+
+/*
+ * In global heartbeat mode, all regions are pinned if there are one or more
+ * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
+ * regions are unpinned if the region count exceeds the cut off or the number
+ * of dependent users falls to zero.
+ */
+#define O2HB_PIN_CUT_OFF		3
+
+/*
+ * In local heartbeat mode, we assume the dlm domain name to be the same as
+ * region uuid. This is true for domains created for the file system but not
+ * necessarily true for userdlm domains. This is a known limitation.
+ *
+ * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
+ * works for both file system and userdlm domains.
+ */
+static int o2hb_region_pin(const char *region_uuid);
+static void o2hb_region_unpin(const char *region_uuid);
+
 /* Only sets a new threshold if there are no active regions.
  *
  * No locking or otherwise interesting code is required for reading
@@ -186,7 +213,9 @@ struct o2hb_region {
 	struct config_item	hr_item;
 
 	struct list_head	hr_all_item;
-	unsigned		hr_unclean_stop:1;
+	unsigned		hr_unclean_stop:1,
+				hr_item_pinned:1,
+				hr_item_dropped:1;
 
 	/* protected by the hr_callback_sem */
 	struct task_struct 	*hr_task;
@@ -702,6 +731,14 @@ static void o2hb_set_quorum_device(struct o2hb_region *reg,
 	       config_item_name(&reg->hr_item));
 
 	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
+
+	/*
+	 * If global heartbeat active, unpin all regions if the
+	 * region count > CUT_OFF
+	 */
+	if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
+		o2hb_region_unpin(NULL);
 }
 
 static int o2hb_check_slot(struct o2hb_region *reg,
@@ -1316,6 +1353,8 @@ int o2hb_init(void)
 	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
 	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
 
+	o2hb_dependent_users = 0;
+
 	return o2hb_debug_init();
 }
 
@@ -2003,16 +2042,20 @@ static void o2hb_heartbeat_group_drop_item(struct config_group *group,
 {
 	struct task_struct *hb_task;
 	struct o2hb_region *reg = to_o2hb_region(item);
+	int quorum_region = 0;
 
 	/* stop the thread when the user removes the region dir */
 	spin_lock(&o2hb_live_lock);
 	if (o2hb_global_heartbeat_active()) {
 		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
 		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
+		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
+			quorum_region = 1;
 		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
 	}
 	hb_task = reg->hr_task;
 	reg->hr_task = NULL;
+	reg->hr_item_dropped = 1;
 	spin_unlock(&o2hb_live_lock);
 
 	if (hb_task)
@@ -2030,7 +2073,27 @@ static void o2hb_heartbeat_group_drop_item(struct config_group *group,
 	if (o2hb_global_heartbeat_active())
 		printk(KERN_NOTICE "o2hb: Heartbeat stopped on region %s\n",
 		       config_item_name(&reg->hr_item));
+
 	config_item_put(item);
+
+	if (!o2hb_global_heartbeat_active() || !quorum_region)
+		return;
+
+	/*
+	 * If global heartbeat active and there are dependent users,
+	 * pin all regions if quorum region count <= CUT_OFF
+	 */
+	spin_lock(&o2hb_live_lock);
+
+	if (!o2hb_dependent_users)
+		goto unlock;
+
+	if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
+		o2hb_region_pin(NULL);
+
+unlock:
+	spin_unlock(&o2hb_live_lock);
 }
 
 struct o2hb_heartbeat_group_attribute {
@@ -2216,63 +2279,138 @@ void o2hb_setup_callback(struct o2hb_callback_func *hc,
 }
 EXPORT_SYMBOL_GPL(o2hb_setup_callback);
 
-static struct o2hb_region *o2hb_find_region(const char *region_uuid)
+/*
+ * In local heartbeat mode, region_uuid passed matches the dlm domain name.
+ * In global heartbeat mode, region_uuid passed is NULL.
+ *
+ * In local, we only pin the matching region. In global we pin all the active
+ * regions.
+ */
+static int o2hb_region_pin(const char *region_uuid)
 {
-	struct o2hb_region *p, *reg = NULL;
+	int ret = 0, found = 0;
+	struct o2hb_region *reg;
+	char *uuid;
 
 	assert_spin_locked(&o2hb_live_lock);
 
-	list_for_each_entry(p, &o2hb_all_regions, hr_all_item) {
-		if (!strcmp(region_uuid, config_item_name(&p->hr_item))) {
-			reg = p;
-			break;
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
+		uuid = config_item_name(&reg->hr_item);
+
+		/* local heartbeat */
+		if (region_uuid) {
+			if (strcmp(region_uuid, uuid))
+				continue;
+			found = 1;
 		}
+
+		if (reg->hr_item_pinned || reg->hr_item_dropped)
+			goto skip_pin;
+
+		/* Ignore ENOENT only for local hb (userdlm domain) */
+		ret = o2nm_depend_item(&reg->hr_item);
+		if (!ret) {
+			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
+			reg->hr_item_pinned = 1;
+		} else {
+			if (ret == -ENOENT && found)
+				ret = 0;
+			else {
+				mlog(ML_ERROR, "Pin region %s fails with %d\n",
+				     uuid, ret);
+				break;
+			}
+		}
+skip_pin:
+		if (found)
+			break;
 	}
 
-	return reg;
+	return ret;
 }
 
-static int o2hb_region_get(const char *region_uuid)
+/*
+ * In local heartbeat mode, region_uuid passed matches the dlm domain name.
+ * In global heartbeat mode, region_uuid passed is NULL.
+ *
+ * In local, we only unpin the matching region. In global we unpin all the
+ * active regions.
+ */
+static void o2hb_region_unpin(const char *region_uuid)
 {
-	int ret = 0;
 	struct o2hb_region *reg;
+	char *uuid;
+	int found = 0;
 
-	spin_lock(&o2hb_live_lock);
+	assert_spin_locked(&o2hb_live_lock);
 
-	reg = o2hb_find_region(region_uuid);
-	if (!reg)
-		ret = -ENOENT;
-	spin_unlock(&o2hb_live_lock);
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
+		uuid = config_item_name(&reg->hr_item);
+		if (region_uuid) {
+			if (strcmp(region_uuid, uuid))
+				continue;
+			found = 1;
+		}
 
-	if (ret)
-		goto out;
+		if (reg->hr_item_pinned) {
+			mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
+			o2nm_undepend_item(&reg->hr_item);
+			reg->hr_item_pinned = 0;
+		}
+		if (found)
+			break;
+	}
+}
 
-	ret = o2nm_depend_this_node();
-	if (ret)
-		goto out;
+static int o2hb_region_inc_user(const char *region_uuid)
+{
+	int ret = 0;
 
-	ret = o2nm_depend_item(&reg->hr_item);
-	if (ret)
-		o2nm_undepend_this_node();
+	spin_lock(&o2hb_live_lock);
 
-out:
+	/* local heartbeat */
+	if (!o2hb_global_heartbeat_active()) {
+	    ret = o2hb_region_pin(region_uuid);
+	    goto unlock;
+	}
+
+	/*
+	 * if global heartbeat active and this is the first dependent user,
+	 * pin all regions if quorum region count <= CUT_OFF
+	 */
+	o2hb_dependent_users++;
+	if (o2hb_dependent_users > 1)
+		goto unlock;
+
+	if (o2hb_pop_count(&o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
+		ret = o2hb_region_pin(NULL);
+
+unlock:
+	spin_unlock(&o2hb_live_lock);
 	return ret;
 }
 
-static void o2hb_region_put(const char *region_uuid)
+void o2hb_region_dec_user(const char *region_uuid)
 {
-	struct o2hb_region *reg;
-
 	spin_lock(&o2hb_live_lock);
 
-	reg = o2hb_find_region(region_uuid);
+	/* local heartbeat */
+	if (!o2hb_global_heartbeat_active()) {
+	    o2hb_region_unpin(region_uuid);
+	    goto unlock;
+	}
 
-	spin_unlock(&o2hb_live_lock);
+	/*
+	 * if global heartbeat active and there are no dependent users,
+	 * unpin all quorum regions
+	 */
+	o2hb_dependent_users--;
+	if (!o2hb_dependent_users)
+		o2hb_region_unpin(NULL);
 
-	if (reg) {
-		o2nm_undepend_item(&reg->hr_item);
-		o2nm_undepend_this_node();
-	}
+unlock:
+	spin_unlock(&o2hb_live_lock);
 }
 
 int o2hb_register_callback(const char *region_uuid,
@@ -2293,9 +2431,11 @@ int o2hb_register_callback(const char *region_uuid,
 	}
 
 	if (region_uuid) {
-		ret = o2hb_region_get(region_uuid);
-		if (ret)
+		ret = o2hb_region_inc_user(region_uuid);
+		if (ret) {
+			mlog_errno(ret);
 			goto out;
+		}
 	}
 
 	down_write(&o2hb_callback_sem);
@@ -2313,7 +2453,7 @@ int o2hb_register_callback(const char *region_uuid,
 	up_write(&o2hb_callback_sem);
 	ret = 0;
 out:
-	mlog(ML_HEARTBEAT, "returning %d on behalf of %p for funcs %p\n",
+	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
 	     ret, __builtin_return_address(0), hc);
 	return ret;
 }
@@ -2324,7 +2464,7 @@ void o2hb_unregister_callback(const char *region_uuid,
 {
 	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
 
-	mlog(ML_HEARTBEAT, "on behalf of %p for funcs %p\n",
+	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
 	     __builtin_return_address(0), hc);
 
 	/* XXX Can this happen _with_ a region reference? */
@@ -2332,7 +2472,7 @@ void o2hb_unregister_callback(const char *region_uuid,
 		return;
 
 	if (region_uuid)
-		o2hb_region_put(region_uuid);
+		o2hb_region_dec_user(region_uuid);
 
 	down_write(&o2hb_callback_sem);
 
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index fcc40c33489..7e38a072d72 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -1659,8 +1659,8 @@ bail:
 
 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
 {
-	o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
-	o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
+	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
+	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
 	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
 }
 
@@ -1672,13 +1672,13 @@ static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
 
 	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
 			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
-	status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
+	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
 	if (status)
 		goto bail;
 
 	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
 			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
-	status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
+	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
 	if (status)
 		goto bail;