[Ocfs2-commits] mfasheh commits r2501 - trunk/fs/ocfs2/cluster

[svn-commits at oss.oracle.com]

Author: mfasheh
Signed-off-by: manish
Date: 2005-08-08 15:31:18 -0500 (Mon, 08 Aug 2005)
New Revision: 2501

Modified:
   trunk/fs/ocfs2/cluster/heartbeat.c
Log:
* Set a higher priority for the heartbeat thread. this hopefully alleviates 
  a problem with heartbeat where we were seeing the thread not being
  scheduled for over 12 seconds (current heartbeat timeout)

* Report the actual timing of a heartbeat timeout more accurately

* Report more information about bad slot write - this may help in future
  debugging.

* Use a block copy when computing checksums as we have to temporarily zero  
  out the hb_cksum field. This fixes a bug seen where our own write would hit
  disk with a zero checksum.

Signed-off-by: manish



Modified: trunk/fs/ocfs2/cluster/heartbeat.c
===================================================================
--- trunk/fs/ocfs2/cluster/heartbeat.c	2005-08-08 20:12:46 UTC (rev 2500)
+++ trunk/fs/ocfs2/cluster/heartbeat.c	2005-08-08 20:31:18 UTC (rev 2501)
@@ -20,6 +20,7 @@
  */
 
 #include <linux/kernel.h>
+#include <linux/sched.h>
 #include <linux/jiffies.h>
 #include <linux/module.h>
 #include <linux/fs.h>
@@ -124,6 +125,12 @@
 	u64			hr_generation;
 
 	struct work_struct	hr_write_timeout_work;
+	unsigned long		hr_last_timeout_start;
+
+	/* Used during o2hb_check_slot to hold a copy of the block
+	 * being checked because we temporarily have to zero out the
+	 * crc field. */
+	struct o2hb_disk_heartbeat_block *hr_tmp_block;
 };
 
 struct o2hb_bio_wait_ctxt {
@@ -136,8 +143,8 @@
 	struct o2hb_region *reg = arg;
 
 	mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
-	     "milliseconds\n", reg->hr_dev_name, O2HB_MAX_WRITE_TIMEOUT_MS);
-
+	     "milliseconds\n", reg->hr_dev_name,
+	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start)); 
 	o2quo_disk_timeout();
 }
 
@@ -419,6 +426,15 @@
 	return ret;
 }
 
+static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
+{
+	mlog(ML_ERROR, "Dump slot information: seq = 0x%"MLFx64", node = %u, "
+	     "cksum = 0x%x, generation 0x%"MLFx64"\n",
+	     le64_to_cpu(hb_block->hb_seq), hb_block->hb_node,
+	     le32_to_cpu(hb_block->hb_cksum),
+	     le64_to_cpu(hb_block->hb_generation));
+}
+
 static int o2hb_verify_crc(struct o2hb_region *reg,
 			   struct o2hb_disk_heartbeat_block *hb_block)
 {
@@ -598,9 +614,11 @@
 	struct o2hb_node_event event =
 		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
 	struct o2nm_node *node;
-	struct o2hb_disk_heartbeat_block *hb_block = slot->ds_raw_block;
+	struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
 	u64 cputime;
 
+	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
+
 	/* Is this correct? Do we assume that the node doesn't exist
 	 * if we're not configured for him? */
 	node = o2nm_get_node_by_num(slot->ds_node_num);
@@ -623,6 +641,7 @@
 		 * other values as they may be junk. */
 		mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
 		     slot->ds_node_num, reg->hr_dev_name);
+		o2hb_dump_slot(hb_block);
 
 		slot->ds_equal_samples++;
 		goto fire_callbacks;
@@ -826,6 +845,8 @@
 
 	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
 
+	set_user_nice(current, -20);
+
 	while (!kthread_should_stop() && !reg->hr_unclean_stop) {
 		o2hb_do_disk_heartbeat(reg);
 
@@ -917,6 +938,9 @@
 	struct page *page;
 	struct o2hb_region *reg = to_o2hb_region(item);
 
+	if (reg->hr_tmp_block)
+		kfree(reg->hr_tmp_block);
+
 	if (reg->hr_slot_data) {
 		for (i = 0; i < reg->hr_num_pages; i++) {
 			page = reg->hr_slot_data[i];
@@ -1081,6 +1105,12 @@
 	char *raw;
 	struct o2hb_disk_slot *slot;
 
+	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
+	if (reg->hr_tmp_block == NULL) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
 	reg->hr_slots = kcalloc(reg->hr_blocks,
 				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
 	if (reg->hr_slots == NULL) {