---

yaml
---
r: 212795
b: refs/heads/master
c: b1c5ebf
h: refs/heads/master
i:
  212793: 2009620
  212791: a721579
v: v3

[refs]

@@ -1,2 +1,2 @@
 ---
-refs/heads/master: 43182d2a799865872041b6e4d8387131e9462f56
+refs/heads/master: b1c5ebfbe398b3360614a4788c02061cd153e60a

trunk/fs/ocfs2/cluster/heartbeat.c

@@ -68,10 +68,12 @@ static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
  * 	- o2hb_live_region_bitmap tracks live regions (seen steady iterations).
  * 	- o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
  * 		heartbeat on it.
+ * 	- o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
  */
 static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
 static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
 static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
+static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
 
 #define O2HB_DB_TYPE_LIVENODES		0
 struct o2hb_debug_buf {
@@ -217,15 +219,48 @@ struct o2hb_bio_wait_ctxt {
 	int               wc_error;
 };
 
+static int o2hb_pop_count(void *map, int count)
+{
+	int i = -1, pop = 0;
+
+	while ((i = find_next_bit(map, count, i + 1)) < count)
+		pop++;
+	return pop;
+}
+
 static void o2hb_write_timeout(struct work_struct *work)
 {
+	int failed, quorum;
+	unsigned long flags;
 	struct o2hb_region *reg =
 		container_of(work, struct o2hb_region,
 			     hr_write_timeout_work.work);
 
 	mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
 	     "milliseconds\n", reg->hr_dev_name,
 	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
+
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock_irqsave(&o2hb_live_lock, flags);
+		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
+			set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
+		failed = o2hb_pop_count(&o2hb_failed_region_bitmap,
+					O2NM_MAX_REGIONS);
+		quorum = o2hb_pop_count(&o2hb_quorum_region_bitmap,
+					O2NM_MAX_REGIONS);
+		spin_unlock_irqrestore(&o2hb_live_lock, flags);
+
+		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
+		     quorum, failed);
+
+		/*
+		 * Fence if the number of failed regions >= half the number
+		 * of  quorum regions
+		 */
+		if ((failed << 1) < quorum)
+			return;
+	}
+
 	o2quo_disk_timeout();
 }
 
@@ -234,6 +269,11 @@ static void o2hb_arm_write_timeout(struct o2hb_region *reg)
 	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
 	     O2HB_MAX_WRITE_TIMEOUT_MS);
 
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock(&o2hb_live_lock);
+		clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
+		spin_unlock(&o2hb_live_lock);
+	}
 	cancel_delayed_work(&reg->hr_write_timeout_work);
 	reg->hr_last_timeout_start = jiffies;
 	schedule_delayed_work(&reg->hr_write_timeout_work,
@@ -1173,6 +1213,7 @@ int o2hb_init(void)
 	memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
 	memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
 	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
+	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
 
 	return o2hb_debug_init();
 }