Skip to contentSkip to conversation

ocfs2: fix a tiny race case when fire callbacks

PATCH New Joseph Qi 2 participants 4 messages Aug 27, 2013 – Aug 28, 2013 OCFS2

lore
Share & tools
lore

Joseph Qi

author
#0 lore raw

PATCH 1/1 ocfs2: fix a tiny race case when fire callbacks

In o2hb_shutdown_slot() and o2hb_check_slot(), since event is defined as local, it is only valid during the call stack. So the following tiny race case may happen in a multi-volumes mounted environment:

o2hb-vol1 o2hb-vol2
1) o2hb_shutdown_slot
allocate local event1
2) queue_node_event
add event1 to global o2hb_node_events
3) o2hb_shutdown_slot allocate local event2
4) queue_node_event
add event2 to global o2hb_node_events
5) o2hb_run_event_list delete event1 from o2hb_node_events
6) o2hb_run_event_list
event1 empty, return
7) o2hb_shutdown_slot
event1 lifecycle ends
8) o2hb_fire_callbacks event1 is already *invalid*

This patch lets it wait o2hb_callback_sem when another thread is firing callbacks. And for performance consideration, we only call o2hb_run_event_list when there is an event queued.

---
 fs/ocfs2/cluster/heartbeat.c |   18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)
fs/ocfs2/cluster/heartbeat.c +9 -9
index 42252bf..af5cd3b 100644
@@ -641,16 +641,9 @@ static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
 /* Will run the list in order until we process the passed event */
 static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
 {
-	int empty;
 	struct o2hb_callback *hbcall;
 	struct o2hb_node_event *event;
 
-	spin_lock(&o2hb_live_lock);
-	empty = list_empty(&queued_event->hn_item);
-	spin_unlock(&o2hb_live_lock);
-	if (empty)
-		return;
-
 	/* Holding callback sem assures we don't alter the callback
 	 * lists when doing this, and serializes ourselves with other
 	 * processes wanting callbacks. */
@@ -709,6 +702,7 @@ static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 	struct o2hb_node_event event =
 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
 	struct o2nm_node *node;
+	int queued = 0;
 
 	node = o2nm_get_node_by_num(slot->ds_node_num);
 	if (!node)
@@ -726,11 +720,13 @@ static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
 
 			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
 					      slot->ds_node_num);
+			queued = 1;
 		}
 	}
 	spin_unlock(&o2hb_live_lock);
 
-	o2hb_run_event_list(&event);
+	if (queued)
+		o2hb_run_event_list(&event);
 
 	o2nm_node_put(node);
 }
@@ -790,6 +786,7 @@ static int o2hb_check_slot(struct o2hb_region *reg,
 	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
 	unsigned int slot_dead_ms;
 	int tmp;
+	int queued = 0;
 
 	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
 
@@ -883,6 +880,7 @@ fire_callbacks:
 					      slot->ds_node_num);
 
 			changed = 1;
+			queued = 1;
 		}
 
 		list_add_tail(&slot->ds_live_item,
@@ -934,6 +932,7 @@ fire_callbacks:
 					      node, slot->ds_node_num);
 
 			changed = 1;
+			queued = 1;
 		}
 
 		/* We don't clear this because the node is still
@@ -949,7 +948,8 @@ fire_callbacks:
 out:
 	spin_unlock(&o2hb_live_lock);
 
-	o2hb_run_event_list(&event);
+	if (queued)
+		o2hb_run_event_list(&event);
 
 	if (node)
 		o2nm_node_put(node);

-- 1.7.9.7

author
#2 lore raw

PATCH 1/1 ocfs2: fix a tiny race case when fire callbacks

Quoted 9 lines · Andrew Morton
Quoted 2 lines · On Tue, 27 Aug 2013 14:17:17 +0800 Joseph Qi <[email protected]> wrote:

This signoff trail makes me wonder: which of you was the primary author of this patch?

Sorry for the misleading signoff trail. Joyce fixed this bug and I rearranged the patch and sent it.
Please refer Joyce as the primary author, thanks.

#3 lore raw

PATCH 1/1 ocfs2: fix a tiny race case when fire callbacks

Quoted 6 lines · On Wed, 28 Aug 2013 09:59:21 +0800 Joseph Qi <[email protected]> wrote:
Quoted 5 lines · Andrew Morton
Quoted 2 lines · On Tue, 27 Aug 2013 14:17:17 +0800 Joseph Qi <[email protected]> wrote:
Quoted 7 lines
Quoted 4 lines

This signoff trail makes me wonder: which of you was the primary author of this patch?

Sorry for the misleading signoff trail. Joyce fixed this bug and I rearranged the patch and sent it.
Please refer Joyce as the primary author, thanks.

No problems. Damn, am I good, or what ;)

The way to handle this situation is to put an explicit From:Joyce line at the very start of the changelog. If that is present, the person who receives the patch should prioritize that From: line over the one which is present in the email envelope.