Is SMART really that dumb?

[Chris Murphy]

On Sat, Mar 14, 2015 at 3:48 PM, Tom Horsley <horsley1953 at gmail.com> wrote:
> So, SMART reports it has N pending sectors. This must mean
> it knows exactly which sectors those are, but nothing in
> the SMART interface is willing to tell you what sectors it
> is talking about?

If there's a definite latent sector error, this shows up with a
'smarctl -t long' which will be aborted at the first error found. The
LBA for this shows up under LBA_of_first_error.

The other way it will show up is in dmesg, libata will report the read
error with the affected LBA. There's some chance of the drive
attempting long recoveries, and the kernel SCSI/ATA command timer
times out and does a link reset which stops the recovery and finding
out what sector is affected.

>
> You could maybe correlate them with the filesystem
> structures and find out what files they might be
> affecting (or if they are just in free space),
> but that's not useful information for SMART to report?
>
> Please tell me I'm the one who is an idiot and I've
> just overlooked the obvious here :-).

No, it requires esoteric knowledge. It's completely non-obvious and
non-discoverable. Fedora actually has smartd running by default and
it'll report some things into the journal; if it's minimally
configured it can do smart -t long on a schedule and report more
things.


-- 
Chris Murphy