[Qemu-devel] a tiny tool for checking and repairing damaged qcow2 image

[张友加]

Hi,

I want to introduce a tiny tool(https://github.com/zhangyoujia/qcow2-dump) for 
checking and repairing damaged qcow2 image.

I met many damaged qcow2 images when I work.

It does't always work by using the qemu-img command to check and repair these 
damaged qcow2 images.

The functions of the qemu-img check command are too limited for developers.

It's difficult to known where is damaged, and difficult to repair the damage 
too.

For example: more than one active L2 entries point to the same active 
cluster(mostly caused by refcount error), such as:

Active Cluster:
----------------------------------------------------------------
REUSED cluster offset: 0x44000000 | l1_index:    2, l2_index: 1017 | refcount: 2
REUSED cluster offset: 0x44000000 | l1_index:   12, l2_index: 7780

REUSED cluster offset: 0x45000000 | l1_index:    2, l2_index: 1273 | refcount: 2
REUSED cluster offset: 0x45000000 | l1_index:   12, l2_index: 6742

REUSED cluster offset: 0x4a000000 | l1_index:    2, l2_index: 2553 | refcount: 2
REUSED cluster offset: 0x4a000000 | l1_index:   12, l2_index: 7068

The results of qemu-img check command are wrong which only show refcount errors 
and copied errors, but do not show the active cluster reused errors. so if you 
use qemu-img check -r sub-command to repair this qcow2 image, it only increases 
refcount of the reused cluster, but does not rebuild the reused active cluster. 
that does't work. (eg: reused_active_cluster.png)

That's why I want to make a tool has more functions.

qcow2-dump has some improvements compare with qemu-img check command:

1. -m check: output the status of qcow2 image and statistics of metadata. (eg: 
rebuild_refcount_table.png)

2. -m error: output of -m check + the damaged positions of qcow2 image. (eg: 
reused_active_cluster.png & error.tgz)

3. -m dump: output of -m error + dump all metadata of qcow2 image. (eg: 
dump.tgz)


qcow2-dump also has some functions for repairing damaged qcow2 image:

1. rebuild damaged refcount table. (eg: rebuild_refcount_table.png)
   qcow2-dump -R all rebuild_refcount_table.qcow2

2. rebuild reused active cluster. (eg: reused_active_cluster.png)
   qcow2-dump -R all -r reused_active_cluster.qcow2

3. delete damaged qcow2 snapshot header. (eg: invalid_snapshot_header.png)
   qcow2-dump -D snapshot[0] vm-disk-2.qcow2

4. -m edit / -m copy: for repairing some special damage.  (eg: comments of 
qcow2_edit_modify is a instance)

...

But this tool is not good enough yet, hope to receive your advices to improve 
it.

Best wishes
youplus

qcow2-dump.tar.gz
Description: GNU Zip compressed data