[dm-devel] potential dm-thin recovery issue

[Jim Minter]

Hello,

Following on from my previous e-mail ("dm-thin discard issue"), I'm having trouble recovering from the metadata corruption state I got into.  I'm using the latest userland tools cloned from https://github.com/jthornber/thin-provisioning-tools .  Here's what I've tried; am I going wrong somewhere, or is there potentially an issue with the tools?

# thin_dump /dev/mapper/vg-meta >/tmp/meta
# cat /tmp/meta
<superblock uuid="" time="1" transaction="0" data_block_size="128" nr_data_blocks="16384">
  <device dev_id="0" mapped_blocks="0" transaction="0" creation_time="0" snap_time="1">
  </device>
  <device dev_id="1" mapped_blocks="8192" transaction="0" creation_time="1" snap_time="1">
    <range_mapping origin_begin="0" data_begin="0" length="8192" time="0"/>
  </device>
</superblock>
# dmsetup table
vg-pool: 0 2097152 linear 252:16 264192
fedora-swap: 0 8257536 linear 252:2 2048
fedora-root: 0 32653312 linear 252:2 8259584
vg-meta: 0 262144 linear 252:16 2048
# thin_restore -i /tmp/meta -o /dev/mapper/vg-meta
recursive 'new_block()' not supported
#

# thin_dump -r /dev/mapper/vg-meta >/tmp/meta2
# diff /tmp/meta /tmp/meta2
#

(so the -r option doesn't help me here)

# dd if=/dev/zero of=/dev/mapper/vg-meta bs=1M &>/dev/null
# thin_restore -i /tmp/meta -o /dev/mapper/vg-meta
recursive 'new_block()' not supported
#

(so wiping the metadata volume before restoring doesn't help either)

Cheers,

Jim