Re: ZFS errors with two HEX numbers.

On Sat, Sep 25, 2021 at 4:01 PM Miroslav Lachman <000.fbsd@quip.cz> wrote:
>
> On 25/09/2021 21:28, Zaphod Beeblebrox wrote:
> > I've got a ZFS pool --- 3 x 8 x 4T disks in a RAIDZ2 configuration.  Every
> > few months I scrub them.  This time, curiously, it came up with an error.
> > All the disks in a vdev simultaneously give an error when this is
> > scanned.... thusly:
> >
> >            raidz2-3      ONLINE       0     0     0
> >              gpt/v1-f0   ONLINE       0     0     1
> >              gpt/v1-f1   ONLINE       0     0     1
> >              gpt/v1-f2   ONLINE       0     0     1
> >              gpt/v1-f3   ONLINE       0     0     1
> >              gpt/v1-f4   ONLINE       0     0     1
> >              gpt/v1-f5   ONLINE       0     0     1
> >              gpt/v1-f6a  ONLINE       0     0     1
> >              gpt/v1-f7b  ONLINE       0     0     1
> >
> > ... and if I add -v to the status, I see the following right now:
> >
> > errors: Permanent errors have been detected in the following files:
> >          <0x57b2>:<0x73ab46>
> >
> > Now... since that happened, I have scrubbed twice more.  The first number
> > changes each time.  Last scrub it was: <0x4223>:<0x73ab46>.  Note the last
> > number is not changing.
> >
> > I have done some googling and found that this might be something that got
> > corrupted on delete or somesuch.  My OCD would like to know what 0x4223
> > maps to (or now 0x57b2) s.t. I could find it and remove it.  I _think_ this
> > is a snapshot reference?
>
> I have seen this in the past. It was corruption on some deleted file
> preserved on snapshot. Once I deleted snapshots this error disappeared.
>
> [...]
>
> Miroslav Lachman

I believe the first number is the objset number, where the objset can
be either a dataset or a snapshot.  The corrupt block likely exists in
multiple snapshots, but "zpool status" will show only one, whichever
it happened to find first.  You can see the objset numbers for
datasets (but not snapshots) in the output of "sysctl kstat.zfs".  The
second number I believe is the object number, and that should never
change.