Re: Your backup is unsafe!
Nathan Hand (nathanh@chirp.com.au)
Tue, 3 Aug 1999 14:06:38 +1000
On Mon, Aug 02, 1999 at 08:16:51AM -0800, Kent Overstreet wrote:
> Nathan Hand wrote:
> >
> > On Mon, Aug 02, 1999 at 11:31:38AM +0100, Robert de Bath wrote:
> > > On Mon, 2 Aug 1999, Nathan Hand wrote:
> > >
> > > > The problem isn't straightforward. Simply stated, VFAT has two names for
> > > > a file, both are valid, both must be preserved, only one is visible, but
> > > > both are usable at all times. This does not map onto any UNIX filesystem
> > > > so some magic is needed.
> > >
> > > Under _Windows_ both are valid. Do you really think this brain dammage
> > > should be kept in Linux? There is no reason for the short name to be
> > > visable _except_ when there are interactions with DOS, this comes down
> > > to dosemu, samba and backups.
> >
> > In retrospect I have to agree. It's clearly better to only show the long
> > filenames and have special "VFAT aware" programs to handle backups. This
> > can be done via ioctl, a /proc interface, etc. It's a robust solution.
> >
> > > > How about a magic file in each VFAT directory which contains mappings of
> > > > long to short filenames. You only see the long names using standard UNIX
> > > > file I/O. Then backups work, because you backup the magic file too, so a
> > > > restore will put the correct short/long mappings back.
> > >
> > > Yes, I thought of this but the coding would be horrific. I think this would
> > > work out to be just the same as using a directory by directory sfn_backup,
> > > and IMO it's better to do it in userspace and keep those horrors from the
> > > kernel.
> >
> > Yes, I agree. The existing Linux VFAT filesystem is clean, UNIX-like, it
> > hides the major braindamages, and it fails in very few cases. Far better
> > to solve specific problems (like VFAT backups) with specialised tools.
>
> Ok, I've been following this thread for awhile, and I think I missed the
> first email, but I don't get something: Why do we need to store the 8.3
> version in a backup? It's a pretty simply algorithm to get the 8.3
> filename from the LFN one, why not put the 8.3 version back when you
> restore to the vfat partition, and completely hide the 8.3 filename?
$ touch /dosc/longfilename # longfilename -> longfi~1
$ touch /dosc/longfilename234 # longfliename234 -> longfi~2
$ rm /dosc/longfilename
$ yourbackup
$ rm /dosc/*
$ yourrestore # longfilename234 -> longfi~1
And you pick longfilename234 to generate the same encoding (it's not
actually 234). Silly, contrived, proves the point.
> Most of the other ideas - symlinks, hardlinks, are *extremely* messy and
> kludgy, and special vfat aware tools. It shouldn't be hard for Linux to
> generate the 8.3 name from the LFN version - when you restore (or are
> writing files to a vfat filesystem in general) you make sure Linux puts
> in both names. In a previous post, I showed how Windows uses directory
> entries with the hidden, system, and volume label attributes to store
> the LFN name, the entries are then ignored if you boot into, say DOS
> 6.22. Now I could be talking out of my ass, but this whole thing just
> doesn't make sense to me, especially some of the solutions that have
> been proposed. If I'm wrong, feel free to flame me with better
> information :)
The whole point is that the proposed solutions are trying to make an
idiotic design into a sensible one.
It's not possible. VFAT is stupid. Let it fail. Trying to fix it can
only make Linux-VFAT idiotic as well.
Linux-VFAT makes the existing idiotic design look like UNIX. This is
as good as it needs to get. Do any VFAT backups in Windows.
My opinion only, etc. I just talk a lot, I don't write VFAT :-)
--
Nathan Hand - Chirp Web Design - http://www.chirp.com.au/ - $e^{i\pi}+1 = 0$
Phone: +61 2 6230 1871 Fax: +61 2 6230 1515 E-mail: nathanh@chirp.com.au
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/