dm thin: data block's ref count is not zero but not belong to any device.

dm thin: data block's ref count is not zero but not belong to any device.

[monty]

Hi! I met a problem of dm-thin: a thin-pool has no volumes but its
nr_free_blocks_data is not zero. I guess the scene of this problem like:
a. create a thin volume thin01, size is 10GB;
b. write 10GB to thin01;
c. create a snapshot thin01-snap, orig is thin01;
d. write 10GB to thin01-snap;
e. for a data block of thin01-snap, which will be written by user, will
experience following steps:
	1. alloc a data block m firstly;
	2. wirte data to data block m;
	3. bio callback and assign data block m to thin01-snap.
If power loss happens between step 2 and step 3, data block will not
belong to any device but its ref count is not zero.
Is it a problem? How to solve this problem?

Re: dm thin: data block's ref count is not zero but not belong to any device.

[Joe Thornber]

On Tue, Sep 25, 2018 at 11:13:17PM -0400, monty wrote:
> 
> Hi! I met a problem of dm-thin: a thin-pool has no volumes but its
> nr_free_blocks_data is not zero. I guess the scene of this problem like:
> a. create a thin volume thin01, size is 10GB;
> b. write 10GB to thin01;
> c. create a snapshot thin01-snap, orig is thin01;
> d. write 10GB to thin01-snap;
> e. for a data block of thin01-snap, which will be written by user, will
> experience following steps:
> 	1. alloc a data block m firstly;
> 	2. wirte data to data block m;
> 	3. bio callback and assign data block m to thin01-snap.
> If power loss happens between step 2 and step 3, data block will not
> belong to any device but its ref count is not zero.
> Is it a problem? How to solve this problem?

Yes, there are a couple of ways this can happen, a dump and restore of
the metadata will repair it.

- Joe



> 
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel