[PATCH]QNX6 filesystem (RO) driver

[Kai Bankett <chaosman@ontika.net>]

Hi,

Onhttp://a6.ontika.net/patches/patch-3.2.2-qnx6.gz  you find a patch for
adding QNX6 FS readonly support (against 3.2.2) to the linux kernel.
(well, at least I would like to open the discussion about it ... ;))

It's the first linux kernel driver I wrote, so please be kind to me in
case some of my design decisions do violate some "common practice" well
known to any regular reader of this list.
I am very happy to work on the source in the "right" direction, if you
suggest me the direction.
So, suggestions, feedback, input, testing is highly welcome!

I reverse engineered - I would say - whole QNX6 Filesystem from ground
up. (hexedit, from superblock identification to adressing scheme to node
kinds etc.)
Successfully completed writing a FS resizing application before starting
work on this kernel driver.
(so you can expect more than just readonly to come ...)

Some sidenotes worth mentioning:
- Audi MMI FS support
I've included support for the Audi MMI HDD filesystem (who's driving a
recent Audi with 3G or 3G+ HDD multimedia system?)
So I've added a "mmi_fs" mount option to tell the driver to use the
right superblock layout and -offset for mounting that "modified" or "old
development state" QNX6 derivate filesystem.
Not sure if a mount option is the usual approach, but it workes well -
at least for me.
I took the decision to make that specific driver support configurable
via kernel config.

- Endianness support
QNX6 offers to create filesystems either little- or big endian. That's
an advantage if the created filesystem later is used on an architecture
that got a different endianness.
I introduced some TO_CPUxx() macros and a superblock detection mechanism
for detecting the FS endianness.
Could not test it on a different architecture than x86 and therefore I'm
unsure if it works on a different endian linux system, but from (my)
theory it should do.
At least on x86 it's no problem to mount a different endian QNX6 FS and
read from it - that's what I've tested.
If the macros are the right approach ... well, happy to receive feedback ;)

I also made the endianness support configurable - just to be on the safe
side.

- Checksums
Superblock checksums are validated.
Longfilenames in a qnx6fs got a checksum in the directory inode.
That checksum algorithm is also build in, but just logs to the kernel
log if it detects a wrong checksum.
(so it's not a hard check)
mmi_fs does not have that longfilename protection, so it's switched off
there.

So much for now ...

Kai