From mboxrd@z Thu Jan  1 00:00:00 1970
From: qindehua <13691222965@163.com>
Subject: Problem with disk replacement
Date: Fri, 19 Jul 2013 00:46:22 +0800 (CST)
Message-ID: <7945ee0b.7d5e.13ff2acdc3f.Coremail.13691222965@163.com>
References: <45ae92bb.4ffb.13ff2092e1d.Coremail.13691222965@163.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=GBK
Content-Transfer-Encoding: 7bit
Return-path: <linux-raid-owner@vger.kernel.org>
In-Reply-To: <45ae92bb.4ffb.13ff2092e1d.Coremail.13691222965@163.com>
Sender: linux-raid-owner@vger.kernel.org
To: linux-raid@vger.kernel.org, neilb@suse.de
List-Id: linux-raid.ids

Hi Neil,
I use kernel 3.9.6 and found data may lost while replacing a RAID disk using --replace.

Here is the steps of the test case:
1. Create 3-drives RAID5 and wait for resync done (I use 1GB disks in VirtualBox machine)
2. Calculate the md5sum on the whole raid: dd if=/dev/md1 bs=1M iflag=direct | md5sum
3. add two spare disks to the raid: mdadm /dev/md1 -a /dev/sdi -a /dev/sdj
4. mdadm --replace a disk then --fail another disk:
   mdadm /dev/md1 --replace /dev/sdb; sleep 3; mdadm /dev/md1 -f /dev/sdc
5. wait for recovery done
6. recalculate the md5sum on the whole raid: dd if=/dev/md1 bs=1M iflag=direct | md5sum

The md5sum of step 6 is NOT identical with that of step 2, this means data is corrupted.
I found in step 4, when another disk was made failed, the replacing process was stopped,
and started recovering of the failed disk, and afterwards the replacing not continue.

Regards,
Qin Dehua