files missing

Jeff Dinisco jeff at
Wed Nov 24 18:39:08 UTC 2004

So I copied the image then ran e2fsck -fy on it.  Good news is that it
restored over 6,000 files.  Bad news is it restored them to lost+found
w/ no file names.  

It's my understanding that I should be looking for directory inodes w/
tables mapping file names to inode #'s.  My guess is that these inodes
have been destroyed at some point which is the reason that I discovered
missing files in the 1st place.  They don't appear to be in lost+found.
Is there a method to recover them?  Thanks.

-----Original Message-----
From: Theodore Ts'o [mailto:tytso at] On Behalf Of Theodore Ts'o
Sent: Tuesday, November 23, 2004 12:48 PM
To: Jeff Dinisco
Cc: ext3-users at
Subject: Re: files missing

On Mon, Nov 22, 2004 at 06:05:56PM -0800, Jeff Dinisco wrote:
> I have a 1 TB ext3 filesystem mounted via iscsi on a redhat 9 system
> kernel version - 2.4.20-30.9.  I'm not sure when it happened, but
> there appears to be about 7,000 files (600GB) missing.  The output
> df implies that the files are still there.  It shows 861 GB utilized.
> But du shows only 300 GB of data.  I'm sure that there are no
> holding onto deleted files because I have unmounted/mounted the
> filesystem several times, synced, etc.  Here's an excerpt from e2fsck
> -nf /dev/sdf ...

The output from df is calculated by using the free blocks field in the
superblock, and is only as accurate as the superblock statistics will

It does look like garbage has been written into your filesystem, and
the situation is probably pretty grim; there will almost certainly be
some data loss.  It may not be as bad as you think, however, since
there may be disconnected inodes (and possibly entire directory
hierarchies) which e2fsck can recover and link into the lost+found

> Things look pretty grim right now.  As soon as I run e2fsck -p, these
> inodes will be deleted and I will lose data, correct?  

Actually, e2fsck -y will be needed to fix the data.  E2fsck -p will
make "safe" fixes, but anything that might require human judgement
will cause e2fsck -p to abort (since it is intended to be used in
unattended boot scripts).  You will need to answer each of e2fsck's
questions manually, or use e2fsck -y to cause e2fsck to automatically
assume an answer of "yes" for each question.  I'd strongly suggest
making a bit-for-bit image backup of the filesystem before proceeding,
since it might be possible for an expert to try other means of
recoverying data after getting back what data you can with e2fsck -y.

> I suspect the
> cause of this issue was multiple hosts mounting this filesystem r/w.

Ah, yup, that would do it.  How long was the filesystem mounted by
multiple hosts?

> Does anyone know of any method that could be used to recover this
> Any help would be greatly appreciated.

You can try e2fsck -y and hope for the best.  Aside from that, you can
grep the disk looking for a specific text pattern if there is a few
extremely valuable files that have to be recovered at all costs.
Failing that, it will be restore from backup times.....  (you did keep
regular backups, right?  :-)

						- Ted

More information about the Ext3-users mailing list