corrupted 3.5 format fs

corrupted 3.5 format fs

[Philippe Gramoullé]

Hi,

Here are the results of some tests i made, playing with my home PC
(nothing to do with production servers this time ;o)

fs is /var , a 3.5 format 300Mo partition, reiserfsck from reiserfsprogs-3.6.3 

I dd'ed /var to a file, losetup'ed it and made successive --check ,
--fixable, --check and finally --rebuild-tree.

After the first --check , it reported 2657 corruptions which could be fixed by
--fix-fixable.
So , i did and re-ran --check and came with still 1 corruption left that was
said to be able to be corrected with --fix-fixable.
Re-ran --fix-fixable and --check , and i still had this '1 corruption found
which can be fixed with --fix-fixable'

bad_stat_data: block 38372, [3955 5371 0x0 SD (0)], SD has bad nlink number

I ran --rebuild-tree and the final corruption finally disappeared.

I've packed the metadata, if you want to have a look at it.

You can find the shell output here :

http://philippemm.multimania.com/reiserfs/3.5_var_corrupted_fs.txt

and the gzipped metadata here;

http://philippemm.multimania.com/reiserfs/metadata.gz

and the log files here :

http://philippemm.multimania.com/reiserfs/

Thanks

Philippe

Re: corrupted 3.5 format fs

[Oleg Drokin]

Hello!

On Thu, Aug 22, 2002 at 06:35:12AM +0200, Philippe Gramoull? wrote:

> So , i did and re-ran --check and came with still 1 corruption left that was
> said to be able to be corrected with --fix-fixable.
> Re-ran --fix-fixable and --check , and i still had this '1 corruption found
> which can be fixed with --fix-fixable'
> bad_stat_data: block 38372, [3955 5371 0x0 SD (0)], SD has bad nlink number

SOmetimes it is hard to tell whenever corruption that looks like it can be fixed
at --fix-fixable will be fixed at this stage or not.
Of course it is a bug taht fix-fixable does not recommend to use
--rebuild-tere.
Currently this is a rule of thumb that if you get corruptions after
--fix-fixable stage, you should run --rebuild-tree.

> http://philippemm.multimania.com/reiserfs/3.5_var_corrupted_fs.txt
> and the gzipped metadata here;
> http://philippemm.multimania.com/reiserfs/metadata.gz

Got it, thank you.

Bye,
    Oleg

Re: corrupted 3.5 format fs

[Vitaly Fertman]

On Thursday 22 August 2002 09:43, Oleg Drokin wrote:
> Hello!
>
> On Thu, Aug 22, 2002 at 06:35:12AM +0200, Philippe Gramoull? wrote:
> > So , i did and re-ran --check and came with still 1 corruption left that
> > was said to be able to be corrected with --fix-fixable.
> > Re-ran --fix-fixable and --check , and i still had this '1 corruption
> > found which can be fixed with --fix-fixable'
> > bad_stat_data: block 38372, [3955 5371 0x0 SD (0)], SD has bad nlink
> > number

Yes, the idea to fix nlinks on fix-fixable has some problems and we decided
to leave it for rebuild-tree as it was before.

-- 

Thanks,
Vitaly Fertman

Re: corrupted 3.5 format fs

[Hans Reiser]

Vitaly Fertman wrote:

>On Thursday 22 August 2002 09:43, Oleg Drokin wrote:
>  
>
>>Hello!
>>
>>On Thu, Aug 22, 2002 at 06:35:12AM +0200, Philippe Gramoull? wrote:
>>    
>>
>>>So , i did and re-ran --check and came with still 1 corruption left that
>>>was said to be able to be corrected with --fix-fixable.
>>>Re-ran --fix-fixable and --check , and i still had this '1 corruption
>>>found which can be fixed with --fix-fixable'
>>>bad_stat_data: block 38372, [3955 5371 0x0 SD (0)], SD has bad nlink
>>>number
>>>      
>>>
>
>Yes, the idea to fix nlinks on fix-fixable has some problems and we decided
>to leave it for rebuild-tree as it was before.
>
>  
>
It seems to me you have a UI bug here, and until you fix those problems 
you need to tell users that it requires a --rebuild-tree to be fixed.

Hans