[Ocfs2-tools-devel] [PATCH 0/3] Add extent list check for discontig block group.

[Tao Ma]

On 07/22/2010 09:18 AM, Joel Becker wrote:
> On Wed, Jul 21, 2010 at 03:24:45PM +0800, Tao Ma wrote:
>> 6. if one extent record looks broken while all the others look
>>     sane and complete, remove it(can we ever meet with it?).
>
> 	Do you mean that you have 5 records, but the first four add up
> to cpg, so you drop the 5th?
yes, that is what I mean, but it isn't limited to the 5th. So if the 1st
one has a e_leaf_clusters > total_clusters_in_the_volume, we will record 
it as to be fixed, and if the other 4 add up to cpg, we will remove the 
1st. if the other 4 e_leaf_cluster's sum is less than cpg, we will 
update 1st e_leaf_clusters accordingly.
>
>> There is still one issue I can't make sure. What about
>> e_leaf_clusters looks like sane. See the below example.
>> e_cpos	e_leaf_clusters	e_blkno
>> 0	12		10000
>> 10	10		20000
>>
>> Currently I drop the group because that is not easy to tell which
>> record needs fix. We may have a very complicated solution(say check
>> the global bitmap to see whether the intersection is allocated and
>> read the real block to see whether it contains valid data) etc, but
>> it is too much complicated and depends on other parts which should
>> be consistent and right. I think it doesn't worth. Joel, do you have
>> any idea that we can solve this easier and quickly or we can leave
>> it as-is?
>
> 	Throw it out.  It's corrupt, and we can't easily detect what
> happened.
Cool, that is what my patch does now.

Regards,
Tao