> Date: Tue, 17 Oct 1995 16:50:54 -0500
> From: Michael K. Johnson <johnsonm@nigel.vnet.net>
> To: Dan Merillat <harik@accessorl.net>
> Cc: linux-kernel@vger.rutgers.edu
> Subject: Re: Windows 95 (VFAT) file systems?
>
>
> Dan Merillat writes:
> > Anyway, why do the drivers like ntfs/hpfs etc allways go readonly?
> > If they have the structures etc. done enough to read the dir entrys
> > and the files, shouldn't they be able to update them?
>
> Short answer: no
> Long answer: Not necessarily.
>
> HPFS I know in particular uses B-trees that are very simple to read,
> and much harder to update. I don't know of the other challenges,
> although I assume they exist.
> In general, being able to parse a data structure doesn't imply
> knowledge of the invariants used while creating it. As an example
> from a different domain, as I'm working on my scanner driver, I have
> not yet been able (even with documentation) to get it to scan correctly.
> However, if I had a SCSI sniffer between my box and the scanner, I would
> be able to understand everything going on between the two.
> Understanding what is going on parallels readonly, being able
> to scan parallels read-write.
>
> Does that help?
Sort of. I can understand the problems when dealing with a scanner
(or any unknown interface) But if you have the structure data, and
you can read it, why are the magic incantations to write it so hard?
Of course, it wouldn't be a first try thing, but you make a small
hpfs partition and write to it, if you can read it back, you got it
right. unless the ability to read the drive dosn't requre the full
structure of the drive. (In which case it is more like data-extraction
with a nice interface rather then a read-only interface, if you get
the distinction.)
If there is extra information there that is being silently ignored,
then I could understand why the system is harder to write to.
> > Oh well. Just a personal pet peeve.
>
> Short answer: So fix it...
At times, I have worked on it. I once made a FAT-table editor for
dog (don't laugh, It was an early project) by reading what
nortun utilitys etc told me about the fat table. (knowing it was 12 or
16 bits was helpful) This was of course before I had heard of great
things like technical manuals, documentation and published source code.
Did I mention it was in pascal? But it did correctly write a file
to the dos file system, that it had un-deleted. Then it trashed my
drive and all my source and it took weeks to rebuild the disk.
(that's when I discoverd backups)
> Long answer: Do you really think that the people writing the
> filesystems would stop at read-only if it were actually only
> a little bit more work to add read-write, or if they had the
> info right there, or if they trusted their knowledge of the
> data structures enough? The filesystem authors I know really
> aren't trying to annoy people...
No, that's not what I was implying. I was pointing out that
all the backups in the world won't make a bit of difference if you
can't write them back. Of course, he could allways dd the whole
dammed partition to the tape.
Well, at this point, now that I've put my foot in my mouth enough,
I'm going to dive into the fs code and find out what it is that
makes it so hard to write back. (quick count, how many read-only
drivers are there out there? hpfs off the top of my head, what about
mac?)
Dan Merillat <Dan.Merillat@eola.accessorl.net>
Animaniac! Linux Activist and net.junkie
Also at: <drm89121@pegasus.cc.ucf.edu>