need help with getting into a corrupted sub directory

need help with getting into a corrupted sub directory

[kyle]

Hi,

I have a ext3 filesystem created inside a problematic seagate ST3500320AS
drive. The drive will just shut itself down automatically whenever it hits
any read error. Only way to wake it up it to cold power off / on the
computer.

Now, my ext3 fs is /dev/sde2

I can mount it without any error:
mount /dev/sde2 /tmp

then, I can 'cd' into /tmp to get a sub directory listing.
let's says there's

Hello
World
users
public

now when I do 'ls public',  I get:
EXT3-fs error (device sde2): ext3_get_inode_loc: unable to read inode
block - inode: 26181633, block=26181634
ls: cannot access /tmp/public: Input/output error

'sde' will disappeared from the system, need power off / on can get it back.

After cold boot, I know there's a subdirectory called "EL" inside "public",
if I do a 'ls public/EL',
I get:
EXT3-fs error (device sde2): ext3_get_inode_loc: unable to read inode
block - inode: 26181633, block=26181634
ls: cannot access /tmp/public/EL: Input/output error

Is there any way I can get into subdirectory of "public" without the need of
read inode block 26181633/26181634 ?
Is it still possible to get a full subdirectory listing of "public" ?

Thanks a lot,
kyle

Re: need help with getting into a corrupted sub directory

[Andreas Dilger]

On 2010-01-29, at 23:07, kyle wrote:
> I have a ext3 filesystem created inside a problematic seagate  
> ST3500320AS
> drive. The drive will just shut itself down automatically whenever  
> it hits
> any read error. Only way to wake it up it to cold power off / on the
> computer.

Strange, I had to do the same for a friend, and I think it was the  
same drive.
You should put it into a USB enclosure - it speeds things up a lot.

> 'sde' will disappeared from the system, need power off / on can get  
> it back.

To get most of the data off the dying drive, use rsync -v, and create  
a list
of files to exclude, when you detect they cause the drive to die.

> After cold boot, I know there's a subdirectory called "EL" inside  
> "public",
> if I do a 'ls public/EL',
> I get:
> EXT3-fs error (device sde2): ext3_get_inode_loc: unable to read inode
> block - inode: 26181633, block=26181634
> ls: cannot access /tmp/public/EL: Input/output error
>
> Is there any way I can get into subdirectory of "public" without the  
> need of
> read inode block 26181633/26181634 ?
> Is it still possible to get a full subdirectory listing of "public" ?


debugfs is your friend.  It can open and list a directory by inode  
number, and dump the files to another filesystem.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

Re: need help with getting into a corrupted sub directory

[tytso]

On Sat, Jan 30, 2010 at 12:24:09AM -0700, Andreas Dilger wrote:
> On 2010-01-29, at 23:07, kyle wrote:
> >I have a ext3 filesystem created inside a problematic seagate
> >ST3500320AS
> >drive. The drive will just shut itself down automatically whenever
> >it hits
> >any read error. Only way to wake it up it to cold power off / on the
> >computer.
> 
> Strange, I had to do the same for a friend, and I think it was the
> same drive.
> You should put it into a USB enclosure - it speeds things up a lot.
> 
> >'sde' will disappeared from the system, need power off / on can
> >get it back.
> 
> To get most of the data off the dying drive, use rsync -v, and
> create a list
> of files to exclude, when you detect they cause the drive to die.

An image copy of the disk will tend to recover more than accessing the
disk via the file system.  I haven't run across the failure mode where
accessing a certain magic block causes the disk to die and require a
power cycle, but in that case what I'd probably do is enhance the
dd_rescue program to take a list of block numbers which it should skip
over....

This will require some programming, but it's probably the best way to
recover the most amount of data.  (Oh, and afterwards someone should
send the changes back to Kurt Garloff, since it would be a very useful
enhancement to dd_rescue.)

						- Ted

Re: need help with getting into a corrupted sub directory

[Andreas Dilger]

On 2010-01-30, at 10:44, tytso@mit.edu wrote:
> On Sat, Jan 30, 2010 at 12:24:09AM -0700, Andreas Dilger wrote:
>> On 2010-01-29, at 23:07, kyle wrote:
>>> I have a ext3 filesystem created inside a problematic seagate
>>> ST3500320AS drive. The drive will just shut itself down  
>>> automatically
>>> whenever it hits any read error.
>>
>> Strange, I had to do the same for a friend, and I think it was the
>> same drive.
>> You should put it into a USB enclosure - it speeds things up a lot.

I just checked the drive type that I had this problem on, it is  
ST3320620AS, as I kept it in case there needed to be more rescue work  
done.  Not exactly the same, but I don't know enough about Seagate  
model numbers to determine how related they are.

> An image copy of the disk will tend to recover more than accessing the
> disk via the file system.  I haven't run across the failure mode where
> accessing a certain magic block causes the disk to die and require a
> power cycle, but in that case what I'd probably do is enhance the
> dd_rescue program to take a list of block numbers which it should skip
> over....

Well, I thought the same thing initially, but like the poster I have a  
drive which dies (locks up internally? I don't know) as soon as  
certain files are accessed.  Since  I could get 95%+ of the files  
using the "rsync -av --exclude-from {bad_file_list}" method, and the  
files I couldn't recover were of marginal value, I did that, as it was  
expedient.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

Re: need help with getting into a corrupted sub directory

[kyle]

>
> To get most of the data off the dying drive, use rsync -v, and create
> a list
> of files to exclude, when you detect they cause the drive to die.

I need to get files from 'public/EL/', however, 'public' cause the drive die
........

>
>
> debugfs is your friend.  It can open and list a directory by inode
> number, and dump the files to another filesystem.

Is it possible to get the inode number of 'public/EL' without hitting the
'public' which cause the drive die?

Kyle

Re: need help with getting into a corrupted sub directory

[kyle]

>
> Well, I thought the same thing initially, but like the poster I have a
> drive which dies (locks up internally? I don't know) as soon as
> certain files are accessed.  Since  I could get 95%+ of the files
> using the "rsync -av --exclude-from {bad_file_list}" method, and the
> files I couldn't recover were of marginal value, I did that, as it was
> expedient.
>

90% of files I need is inside a subdirectly 'public' which cannot be read
........ :(

Kyle

Re: need help with getting into a corrupted sub directory

[Jan Kara]

> >
> > To get most of the data off the dying drive, use rsync -v, and create
> > a list
> > of files to exclude, when you detect they cause the drive to die.
> 
> I need to get files from 'public/EL/', however, 'public' cause the drive die
> ........
> 
> >
> >
> > debugfs is your friend.  It can open and list a directory by inode
> > number, and dump the files to another filesystem.
> 
> Is it possible to get the inode number of 'public/EL' without hitting the
> 'public' which cause the drive die?
  Not easily. But as Ted pointed out, you might try to dd blocks from /dev/sde
upto the problematic block and then next time starting after the problematic block.
If this works out, you can then just run fsck at the resulting image and it should
recover the filesystem reasonably...

									Honza
-- 
Jan Kara <jack@suse.cz>
SuSE CR Labs