Filesystem Mutation Tool

Filesystem Mutation Tool

[Darrick J. Wong]

Hi all,

As I might've mentioned to a few of you at OLS, I've hacked up a quick
and dirty program to study the effects of what happens to a filesystem
when certain blocks mutate underneath it (think malice, your RAID5
controller goes berserk, etc).  Said program is now posted in a crude
form here:

http://sweaglesw.net/~djwong/programs/fs_mutate/

I've run this program against ext3 and reiserfs; so far, ext3 seems to
be the stability winner, as it tends to stay up the longest (about 30-35
minutes) even with destroy mode turned on.  reiserfs lasts a few minutes
under such a beating.  Of course, "stays up" is a long way from "works
properly" -- overwriting things like indirect blocks has the rather
amusing effect of generating lots of messages about falling off the end
of a drive.  As with the folks who used carefully crafted ISO9660
filesystems to crash arbitrary machines demonstrated last year, it's not
so hard to get Linux to automount filesystems.  To my knowledge,
nobody's tried a similar thing against the other filesystems, though I
could just be ignorant.

What do you think?  Useful tool?  Or am I the one being the tool? ;)

--D

Re: Filesystem Mutation Tool

[Greg KH]

On Fri, Aug 04, 2006 at 07:07:02PM -0700, Darrick J. Wong wrote:
> Hi all,
> 
> As I might've mentioned to a few of you at OLS, I've hacked up a quick
> and dirty program to study the effects of what happens to a filesystem
> when certain blocks mutate underneath it (think malice, your RAID5
> controller goes berserk, etc).  Said program is now posted in a crude
> form here:
> 
> http://sweaglesw.net/~djwong/programs/fs_mutate/
> 
> I've run this program against ext3 and reiserfs; so far, ext3 seems to
> be the stability winner, as it tends to stay up the longest (about 30-35
> minutes) even with destroy mode turned on.  reiserfs lasts a few minutes
> under such a beating.  Of course, "stays up" is a long way from "works
> properly" -- overwriting things like indirect blocks has the rather
> amusing effect of generating lots of messages about falling off the end
> of a drive.  As with the folks who used carefully crafted ISO9660
> filesystems to crash arbitrary machines demonstrated last year, it's not
> so hard to get Linux to automount filesystems.  To my knowledge,
> nobody's tried a similar thing against the other filesystems, though I
> could just be ignorant.

People have done it in the past, and found lots of bugs that have
hopefully been fixed (although the iso issues have not been fixed...)

> What do you think?  Useful tool?  Or am I the one being the tool? ;)

I think it's useful, especially if it causes things to be fixed up in
the kernel :)

Try running it against a vfat filesystem and see if you can create some
good oopses.  That would be a good place to start, as USB flash sticks
are more common these days than cdroms...

thanks,

greg k-h