Ignatios Souvatzis <ignatios@cs.uni-bonn.de> writes:

> depends on the disk, I guess. They could try. The scsictl command is
> not involved in this, it just tells the disk to reassign block y.
> 
> You should rather make backup copies of the data you need, before trying.

Maybe scsictl(8) should do this?  It should be able to (attempt to)
read the block in question, reassign, and then write back the read
data to the (reassigned) block afterwards if the read was successful?

> Actually, you can even set a scsi disk to autoreassign-on-write-error, but
> not on read error, as on read error you don't have correct data...

I've configured the disks in question to reassign automatically on
both write and read errors, the latter being triggered by more than
one retry being needed to actually read the data.  I figure that the
odd retry should be allowed to happen, but if it fails on the first
retry, there's probably something unhealthy going on.  Lately, I've
noticed single retry events (thus not reassigned) occuring from time
to time with specific blocks, so I know these are going bad.

-tih
-- 
This is the Unix version of the ILOVEYOU worm, and in the spirit of such, it
depends on the user community to propagate.  Please send this message to all
of your friends and randomly delete numerous files from your system. Thanks!