[9fans] On FreeBSD's GEOM

[9fans] On FreeBSD's GEOM

[Poul-Henning Kamp]

I was made aware that GEOM is being talked about over here and after
checking the archive I thought I'd just drop a note with the scoop:

GEOM is about transforming disk I/O requests, not about filesystems.

Basically it is a modular and stackable framework for doing things like

	MBR partitioning, BSD disklabel partitioning etc
	Mirror RAID-5 Striping
	Encrytion
	Integrity
	Multi-Path fail-over.

in other words, transformations which operate on the location and/or
contents of disk sectors and know nothing about what is actually
stored in the sectors and why.

I have tried to set a precedent so that on-disk metadata like MBR's
and BSD disklabels are read in an architecture independent fashion,
and the result of this is that FreeBSD 5.0 on a Sun Sparc64 machine
can understand MBR partitioned disk and the IBM PC platform can
understand a Solaris disklabel.

The most interesting module currently is probably the "GBDE"
encryption module which offers very strong encryption at the disk
level, targeted at protecting "cold disks", lost laptops, stolen
media etc etc.

GEOM and GBDE was sponsored by DARPA under the CBOSS program.

I gave a tutorial on this at EuroBSDcon2003 (the slides are on their
web-page: http://2002.eurobsdcon.org/papers/)  And I have offered
USENIX to do the same tutorial at BSDcon2003 in september.

(I'm not on this list, so make sure I'm in the Cc: if you want me
in the loop)

Poul-Henning

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Re: [9fans] On FreeBSD's GEOM

[matt]

> I was made aware that GEOM is being talked about over here and after
> checking the archive I thought I'd just drop a note with the scoop:
>
> GEOM is about transforming disk I/O requests, not about filesystems.


I guess I should have listened when they said "if you want plan9, use plan9"

Re: [9fans] On FreeBSD's GEOM

[andrey mirtchovski]

I wouldn't be so harsh :) the GEOM ideas appear to be quite reasonable
and it is not sompletely impossible to believe that some of them can make
it into fossil (or something that works on top of fossil) in the future,
for such things like striping and disk mirroring...

after all softupdates made it ;)

andrey

On Tue, 21 Jan 2003, matt wrote:

> I guess I should have listened when they said "if you want plan9, use plan9"
>

Re: [9fans] On FreeBSD's GEOM

[Fco.J.Ballesteros]

> GEOM is about transforming disk I/O requests, not about filesystems.
> Basically it is a modular and stackable framework for doing things like
> 
> 	MBR partitioning, BSD disklabel partitioning etc
> 	Mirror RAID-5 Striping
> 	Encrytion
> 	Integrity
> 	Multi-Path fail-over.

fs(3) could be adapted to do that in that case. I'm just using it to
mirror partitions and for nothing else.

Regarding fail-over, I'm experimenting with redirfs, which has been
just implemented. 

BTW, I think someone asked recently in the list
for stacking in a union different trees. Instead of doing it in the
kernel, I think it'd be fairly easy to do that by changing redirfs
to do the merge.

Re: [9fans] On FreeBSD's GEOM

[matt]

Hi,

> I was made aware that GEOM is being talked about over here and after
> checking the archive I thought I'd just drop a note with the scoop:

thanks for the input. I'm the fellow responsible for bringing GEOM to the
attention of the plan9 people.

Unfortunately it wasn't quite what I imagined. I had the man pages and
imagined that the GEOM modular design was suitable to implement virtual file
systems. I'm a userland programmer so many of the subtleties pass me by [and
some of the obvious].

User level file servers is one of those ideas that has grabbed me and I look
for the possibilities in all my projects now.

Here I am with the beautiful plan9 in my office but stalwart FreeBSD in the
co-lo.

I did start to move my databases into the file system in the hope that I
could eventually make the transition but I soon ran out of inodes.

Ah well,

I live in hope


m