<4> post_create: setxatter failed

Russell Coker russell at coker.com.au
Wed May 25 12:00:37 UTC 2005


On Wednesday 25 May 2005 21:01, Dan Hollis <goemon at anime.net> wrote:
> > The following extract from reiserfsck(8) indicates that it won't:
> >        --rebuild-tree
> >               This option rebuilds the entire filesystem tree using leaf
> > nodes found  on the device.  Normally you only need this option if the
> > reiserfsck  --check  reports  "Running  with  --rebuild-tree  is
> > required".  You are strongly encouraged to make a backup copy of the
> > whole partition before attempting the --rebuild-tree option. Once 
> > reiserfsck  --rebuild-tree  is  started it must finish its work (and you
> > should not interrupt it), otherwise the filesystem will  be  left in the
> > unmountable state to avoid subsequent data corruptions.
>
> Why does that indicate reiserfs won't function as a root file system?

When (not if) the root file system needs a thorough fsck any interruption 
(such as power failure) will result in a non-bootable system.

But it's still a significant improvement, it used to be that --rebuild-tree 
was not permitted on a ro mounted file system so booting from an install CD 
was necessary to check the root fs.

> > > Our reiserfs results interested a business partner of ours who used
> > > to swear by ext3. They also found out reiserfs was better all round for
> > > them so they are also switching.
> >
> > Unless of course they want to do some common file system recovery options
> > such as putting an image of a file system on another file system.  The
> > last reports were that if you put a Reiser3 image as a regular file in a
> > Reiser3 file system then fsck would really mess things up.
>
> Er. This is true of _any_ journaled filesystem on top of an journaled
> filesystem (xfs, jfs, and iirc ext3) due to bad interactions with linux's
> caching. The general wisdom is to not do this, no matter what filesystem.

No it's not.  Put an ext3 image inside an ext3 file system and try to 
reproduce the problem.  You will discover that ext3 is not vulnerable to 
this.  Also try it on HPFS (non-journalled file system used by OS/2) and you 
will discover similar problems.

It's nothing to do with journalling.  It's a problem of recognising metadata, 
file systems which have fixed locations for metadata have a much easier time 
of this.

> Alternatively you could put a reiser3 image on ext2 if you really want to
> do that. It's not hard.

Sure it's not hard.  That's nice to know AFTER you've done it and had your 
data corrupted (as has happened to several people).

This is also a security issue.  For example you might be able to put a file in 
your home directory which contains a ReiserFS image of a file system 
containing a SUID file and then have a fsck make it a regular file owned by 
root with the SUID bit set which would allow taking over the machine.

In a quick test I managed to create a SUID root file in such a manner, but it 
seemed corrupted and could not be executed.

Hopefully Reiser4 will solve this issue.

> > > The only real caveat for reiserfs at the moment is the lack of selinux
> > > support.
> >
> > How is it lacking?  In a quick test it seems to work.
>
> reiserfs (at least up to 2.6.11.10) doesnt have working support for
> selinux attributes.

The latest rawhide kernel seems to have it.  Did you do any tests?


Finally, are there still known situations where a corrupt ReiserFS file system 
could corrupt kernel memory?  This was the bug that got me to convert the 
root file systems of machines I own to use Ext2/3.

-- 
http://www.coker.com.au/selinux/   My NSA Security Enhanced Linux packages
http://www.coker.com.au/bonnie++/  Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/    Postal SMTP/POP benchmark
http://www.coker.com.au/~russell/  My home page




More information about the fedora-test-list mailing list