BTRFS and old kernels.

BTRFS and old kernels.

[Miguel Sousa Filipe]

Hello,
I've been testing btrfs on a linux/ppc with a 2.6.20 kernel.
For that I needed to apply a minor patch for <v0.13 (related to the
definition of ,write_cache_pages()).
For v0.13 it required the patch for the endianess bug in crc32 due to
the change in the crc32c() implementation on linux-2.6.23, that I've
posted here before.

I've just pulled the unstable tree, and tried to build, but it doesn't
even compile, I've found yet one more changed function declaration :
linux-2.6.20: include/linux/bio.h:extern void bio_endio(struct bio *,
unsigned int, int);
linux-2.6.24: include/linux/bio.h:extern void bio_endio(struct bio *, int);

and also some missing functions: (bdi_ini()t, bdi_destroy()). (not
found in 2.6.23 and older)
linux-2.6.24: include/linux/backing-dev.h:int bdi_init(struct
backing_dev_info *bdi);
linux-2.6.24: include/linux/backing-dev.h:void bdi_destroy(struct
backing_dev_info *bdi);
Should I include the code from 2.6.24 for bdi_init and bdi_destroy in
btrfs, and wrap it around ifdefs ?

My question is, is there sill interestest in having btrfs compatible
with older kernels (like, 2.6.20 or 2.6.18)?
I'll post (or repost) patches that I need for btrfs-unstable to
build/work on this ppc system of mine.

Kind regards,
-- 
Miguel Sousa Filipe

Re: BTRFS and old kernels.

[Bron Gondwana]

On Sat, Apr 05, 2008 at 08:13:47PM +0100, Miguel Sousa Filipe wrote:
> My question is, is there sill interestest in having btrfs compatible
> with older kernels (like, 2.6.20 or 2.6.18)?
> I'll post (or repost) patches that I need for btrfs-unstable to
> build/work on this ppc system of mine.

It sort of depends who you ask.  Personally, I see value in being
compatible with as wide range of kernels as possible in a development
filesystem that you want people to be testing.  But then I'm not
doing any btrfs development, so I don't count for much.

Other people, well - Christoph Hellwig had this to say a little
while back in <20080207044835.GC4140@lst.de> (rewrapped):

  On Wed, Feb 06, 2008 at 05:14:51PM -0500, Jeff Mahoney wrote:
  > -----BEGIN PGP SIGNED MESSAGE-----
  > Hash: SHA1
  >
  >  Kernels prior to 2.6.19 don't export file_ra_state_init().
  >  This patch adds the file_ra_state_init() from 2.6.18.

  Please put this somewhere else, a separate file and hopefully 
  only in your backports tree.  This is kernels from stoneage
  after all, not just one or two revisions back.

Bron ( not being particularly helpful or even answering your 
       question really )

Re: BTRFS and old kernels.

[Bron Gondwana]

On Sun, Apr 06, 2008 at 10:57:12AM +1000, Bron Gondwana wrote:
> On Sat, Apr 05, 2008 at 08:13:47PM +0100, Miguel Sousa Filipe wrote:
> > My question is, is there sill interestest in having btrfs compatible
> > with older kernels (like, 2.6.20 or 2.6.18)?
> > I'll post (or repost) patches that I need for btrfs-unstable to
> > build/work on this ppc system of mine.
> 
> It sort of depends who you ask.  Personally, I see value in being
> compatible with as wide range of kernels as possible in a development
> filesystem that you want people to be testing.  But then I'm not
> doing any btrfs development, so I don't count for much.

I guess I should justify this a bit....

ASSUME: I have a bunch of production Debian Etch machines running
        the latest Debian Etch production kernel (2.6.18-6-amd64 
        sounds like a convincing sort of version number) and I want
        to benchmark/test btrfs in my environment.

CASE 1: I can upgrade the kernel on this system to 2.6.24+ and make
        sure all my hardware still works as expected, upgrade my 
        udev/hal/whatever to be compatible.  Possibly backport some
        userland tools that use new interfaces now... sounds like
        a whole pile of fun^H^H^Hwork.

CASE 2: I can build a standalone btrfs module against my current
        kernel and test, having changed nothing but the filesystem.


I know which one I would trust to give me a better idea of how
stable/performant[1] btrfs is on my platform.


(this all assuming that there aren't architectural changes in the
 intervening layers that make btrfs not work properly on the older
 kernels, but ext3/xfs/jfs/reiserfs/name-your-poison seem to have
 all remained stable right through those transitions[2].)

Bron.


[1] screw you http://boulter.com/blog/2004/08/19/performant-is-not-a-word/
    and friends...
        
[2] ok, there have been some exceptions to that, for example:
    http://oss.sgi.com/projects/xfs/faq.html#dir2

Re: BTRFS and old kernels.

[Chris Mason]

On Saturday 05 April 2008, Miguel Sousa Filipe wrote:

> My question is, is there sill interestest in having btrfs compatible
> with older kernels (like, 2.6.20 or 2.6.18)?
> I'll post (or repost) patches that I need for btrfs-unstable to
> build/work on this ppc system of mine.

Definitely.  Just to confirm what Zach said over the weekend, at this stage in 
the game it is much more important to me to keep btrfs easy for testers than 
it is to avoid small amounts of compat code.

So, any fixes would definitely be appreciated.

-chris