Re: RAIDframe and bad blocks

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Fri, 20 Sep 2019, Emmanuel Dreyfus wrote:
> Consider a RAID 1 with wd0 and wd1
>
> wd1 has bad blocks and is marked failed. I replace it and reconstruct
> the RAID with a new disk. But wd0 also has bad blocks, and RAIDframe
> will give up reconstruction because of the read failure.
>
> Most likely the relevant blocks are oustide of used space, and could
> just be replaced by zeroes, but that needs to be done by cloning
> with dd conv=noerror,sync, with the RAID offline.
>
> Perhaps we could come up with a better RAIDframe behavior on
> reconstruction with read errors?

I wrote a tool I called diskfix which scan reads a given block device in 
1MB sections. If it gets an I/O error, it then scans the same section 
sector by sector and on encountering a sector I/O error, writes the sector 
back. I usually use it on the failed component (in your example, wd1) 
before initiating a reconstruction, but if you are brave/desperate it can 
be run on a live filesystem/RAIDframe array.

--
Stephen