Assemble-Resize-Stop loop doesn't work correctly

Hi Neil,

I've tested with the following script grow with "--assume-clean" in a loop. The first grow succeeds - following grows fail - mdadm 3.2.5, kernel 3.4.10.

#!/bin/bash

FIRST="/dev/sda"
SECON="/dev/sdd"
MDDEV="/dev/md0"
SIZE=1

mdadm --zero-superblock $FIRST
mdadm --zero-superblock $SECON
echo y | mdadm -C $MDDEV -e 1.2 \
--assume-clean -z "${SIZE}G" --force -l 1 -n 2 $FIRST $SECON sleep 3
mdadm -S $MDDEV
for ((i=0; i<4; i++)); do
mdadm -A $MDDEV $FIRST $SECON
let "SIZE++"
mdadm -G $MDDEV -z ${SIZE}G --assume-clean
cat /proc/mdstat
# mdadm -D $MDDEV > /dev/null
mdadm -S $MDDEV
done

Output looks like this:

mdadm: /dev/md0 has been started with 2 drives.
mdadm: component size of /dev/md0 has been set to 2097152K Personalities : [raid1]
md0 : active raid1 sda[0] sdd[1]

      2097152 blocks super 1.2 [2/2] [UU]

unused devices: <none>
mdadm: stopped /dev/md0
mdadm: /dev/md0 has been started with 2 drives.
mdadm: /dev/md0 is performing resync/recovery and cannot be reshaped Personalities : [raid1]
md0 : active raid1 sda[0] sdd[1]

      2097152 blocks super 1.2 [2/2] [UU]
      [==========>..........]  resync = 50.0% (1050624/2097152)

finish=8.4min speed=2048K/sec

Now the output with "Detail"-Mode after resize:

mdadm: /dev/md0 has been started with 2 drives.
mdadm: component size of /dev/md0 has been set to 2097152K Personalities : [raid1]
md0 : active raid1 sda[0] sdd[1]

      2097152 blocks super 1.2 [2/2] [UU]

unused devices: <none>
mdadm: stopped /dev/md0
mdadm: /dev/md0 has been started with 2 drives.
mdadm: component size of /dev/md0 has been set to 3145728K Personalities : [raid1]
md0 : active raid1 sda[0] sdd[1]

      3145728 blocks super 1.2 [2/2] [UU]

This one works. Is this wanted behaviour?

Cheers,
Sebastian

You've hit an unlikely corner-case there. Thanks.

This patch fixes it.

NeilBrown

From 2225a657ce9fb4a5390a4a82c03e6a0f937b4327 Mon Sep 17 00:00:00 2001
From: NeilBrown <[email protected]>
Date: Thu, 11 Oct 2012 11:41:14 +1100
Subject: [PATCH] md: make sure manual changes to recovery checkpoint are
saved.

If you make an array bigger but suppress resync of the new region with
mdadm --grow /dev/mdX --size=max --assume-clean

then stop the array before anything is written to it, the effect of the "--assume-clean" is lost and the array will resync the new space when restarted.
So ensure that we update the metadata in the case.

index e868f0c..dff013a 100644
@@ -3819,6 +3819,8 @@ resync_start_store(struct mddev *mddev, const char *buf, size_t len)
 		return -EINVAL;
 
 	mddev->recovery_cp = n;
+	if (mddev->pers)
+		set_bit(MD_CHANGE_CLEAN, &mddev->flags);
 	return len;
 }
 static struct md_sysfs_entry md_resync_start =