From email@example.com Tue Aug 23 19:43:19 1994
Date: Wed, 24 Aug 1994 08:50:42 +1000 (EST)
From: John Meehan
Subject: CCL:UNIX - rm
To: Shu-Chuan Jao
On Tue, 23 Aug 1994, Shu-Chuan Jao wrote:
> Hi! I am really in a big trouble. I accidentally
> typed "rm *" and, as you can image, all my files in
> that directory are gone. We did not do backup. So, I
> cannot restore those files. I know that there are some
> packages that can do "undelete" on DOS to recover files
> which are deleted. Is there package doing the same thing
> for unix?
> I appreciate any responds. 08/23/94
This was posted to comp.archives a little while ago. It may help....
Below is a "Dr. File System" article I wrote as commentary on
possible solutions to the problem of recovering lost files on
UNIX systems, especially those using the Berkeley Fast File
The article observes that it should be possible to write a program
which examines the cylinder group free fragment bit maps. Once
you have the bit maps, it is fairly easy to read the free blocks
of the file system and examine the old data in them.
A few months ago I removed some program sources that hadn't been
backed up yet and were long enough I didn't want to type them
in again. So, I wrote the program loosely described by the
article. Having been careful not to disturb the free list of
the disk while writing this program, I got back the previous
More recently, I dusted off the program, updated it to run
on DEC OSF/1 and fixed the stupidly slow parts to go faster.
A compressed archive of the program has been placed on:
Enjoy. If you have any questions about it, send me mail
* * *
Dear Dr. File System,
I just typed "rm -r" by accident in the wrong place and all my
files are gone. How do I get them back? Oh, by the way. I don't
have any backups...
You stupid twit! [ and off Dr. File System goes into his usual
rant about not keeping good backups... ]
* * *
While Dr. File System is ranting, perhaps we can help Clueless with
his problem (and make a bundle of money off software consulting
services in the process). First examine what rm(1) really does:
Process the argument list, making note of which options
For all remaining arguments, if the argument is ".."
continue. Otherwise "remove" it.
"Removing" consists of taking care of details like
not being able to remove directories, except as part
of a recursive remove, making sure the file is really
In the end, it comes down to doing an unlink(2) on the
The unlink(2) system call removes the file's name and inode number
from the directory and decrements the link count. It is worth
pointing out for those that didn't know it already that a file can
have many names, but only one inode. The inode includes a reference
count of how many names it has. These references are also called
When the link count goes to zero AND the file is no longer open,
then the file is removed. Thus it is possible to create a file,
unlink it (removing the name) and let the opening processes continue
to use it. When the file is closed or the process exits, the file
Actually getting rid of the file consists of putting the blocks
it was using back on the "free list" and clearing the inode. For
the fast file system this "free list" is actually a bit map for
each cylinder group. A simple system macro is used to calculate
the cylinder group number from the block number.
It's also worth noting that neither rm(1) nor the file system
do anything with the contents of the block when it removes the
file or puts the block on the "free list".
Using this collection of information it may be possible to
help Clueless recover some of his data. The first, absolute
most important thing to do is make sure the file system changes
as little as possible. This may require rather extreme measures
(like shutting down or crashing the system). The point of this
is to prevent the blocks that were part of Clueless' files from
getting allocated to other users. Making a PHYSICAL backup of
the partition holding the file system is a good idea. This
will let the recovery operate on a copy of the data, while
returning the disk to the service of the people that might be
using it. Now to the recovery.
First, recall that the inodes were cleared when the files
were removed. If they hadn't been, it would be fairly
simple matter to examine all the free inodes to determine
which had been owned by Clueless and look at the block
lists to determine where the data was. This avenue isn't
available to us, though.
What remains is the examination of the data on the disk in
the hopes of picking out the pieces that were part of Clueless'
files and putting them back together like many jigsaw puzzles.
The methods for this examination can vary from brute force
approach to a highly optimized one. Consider the choices.
1. Theme - Examine every LBN of the disk. Pick out the LBNs that
were probably interesting and ignore the rest. This has the
disadvantage of not throwing out blocks allocated to other
people and those that are part of the file system's data
2. Variation #1 - Observe that the Fast File System is organized
in pieces of data having two sizes; the fragment and the file
block size. Depending on the allocation scheme only the block
size may be interesting. The only advantage this has over the
Theme is that there are few blocks to examine.
3. Variation #2 - The skilled mechanic can easily identify what
file system block belong to the file system overhead and what
blocks are data. This is bound to help some.
4. Variation #3 - Observe that only free blocks are likely to
contain Clueless' data. Each cylinder group has a bit map
of the free space for that cylinder group. If the skilled
mechanic is able to filter out the file system overhead,
then limiting the search to the free list, shouldn't be that
5. Variation #4 - The value of this optimization depends on exactly
what Clueless removed and how the files were arranged. If the
files were in a directory and only the file in that directory
were removed (no subdirectories), then we can make a good guess
where to start the search.
Recall from previous discussions that when a file is created it
prefers to end up in the same cylinder group as the directory
in which it resides. If you can determine what cylinder group
the directory of interest is in, then you can start the search
in that cylinder group. While you may still have to search all
the cylinder groups to find all the data, there is a chance that
you can recover an interesting amount of the quickly.
A program to do such a search is not exceeding difficult with
enough study of the file system include files and a good
starting example. A good version of the program would perform
some data analysis of each block in the hopes of identifying
it. Sort of like file(1). A further enhancement would allow
formatting the data in various ways, like od(1). All of the
interesting recovery work is the examination of the data, not
getting the data to examine.
Alan Rollow firstname.lastname@example.org
John Meehan O CH2-COOH
Department of Biochemistry " /
University of Tasmania, HO-P-O-C-COOH
Australia | \
PHOSPHOCITRIC ACID ---- A powerful, natural
inhibitor of Pathological Biomineralization
---Administrivia: This message is automatically appended by the mail exploder:
CHEMISTRY@ccl.net -- everyone | CHEMISTRY-REQUEST@ccl.net -- coordinator
MAILSERV@ccl.net: HELP CHEMISTRY | Gopher: www.ccl.net 73
Anon. ftp www.ccl.net | CHEMISTRY-SEARCH@ccl.net -- archive search
http://www.ccl.net/chemistry.html | for info send: HELP SEARCH to MAILSERV