3

I have a SW RAID1 and I just replaced /dev/sda with a new HDD as the old one failed.
Now, upon trying to recreate the RAID array I discovered that the "good" HDD (/dev/sdb) has bad blocks with prevent mdadm from resyncing the array.

While I could make backups, replace /dev/sdb as well and re-install the server completely I was wondering if there is any way I could "trick" mdadm into resyncing the RAID array and then replace /dev/sdb with a new HDD.
From what I can guess the badblocks are located in an unused area of /dev/sdb which is only used when trying to recreate the RAID array.

Alex Flo
  • 1,761
  • 3
  • 18
  • 23

2 Answers2

4

A better strategy would be to use ddrescue to copy bad drive to good. This tool tries hard to read entire drive, doing re-reads and "trimming" unreadable blocks. It also produces log, which is used for saving progress, but ultimately will contain the list of bad blocks. You can then parse this list and write to every bad block to see if the drive will survive remapping them all. If it will, you can mdadm --zero-superblock and then just add it as a new drive to your degraded raid1. BTW, for today's high capacity drives it is pretty normal to produce "soft" bad blocks from time to time. Just do your raid check weekly, and you'll be ok, unless the same block gets unreadable on two drives, which is very unlikely.

user208990
  • 41
  • 2
3

Can you verify whether the affected blocks and underlying bad sectors on the disk are reallocated to "spare sectors" area? The bad sector should be reallocated when write operation fails. Verify it with smartctl:

 smartctl -a /dev/sdb | grep -i reallocated

The last column should contain a number of total reallocated sectors. If there is zero try to read the bad sector:

hdparm –-read-sector XXXXXXXX /dev/sdb

It should return an I/O error otherwise I would recommend to skip next section.

The error means the sector was not reallocated yet. So you can try to reallocate it forcibly by writing it. Remember that any data stored in this sector will be lost after this step !!!:

hdparm –-write-sector XXXXXXXX --yes-i-know-what-i-am-doing /dev/sdb

By the way, the sector number XXXXXXXX should be possible to obtain from kernel messages (dmesg command or from /var/log/messages). As you had bad blocks during resynchronisation there should be some related messages similar to:

... end_request: I/O error, dev sdb, sector 1261071601

Then, try to verify it with smartctl again. Does the counter increased? If so try to read it with hdparm. Now, it should read it without any error as it is supposed to be reallocated. Done.

Finally, you can continue with mdadm and with adding the disk to your degraded mirror.

dsmsk80
  • 5,817
  • 18
  • 22
  • Excellent solution but I was wondering if there is any way I could test the whole disk for badblocks so I can reallocate them at once and then try to resync with mdadm. Right now mdadm stops each time it encounters bad blocks that I haven't yet reallocated. – Alex Flo Aug 16 '13 at 13:03
  • In my opinion, you can't do it without losing current data as you would need to rewrite the affected drive completely, e.g. with dd or badblocks tool. – dsmsk80 Aug 19 '13 at 06:55