Message-Id: <200106282302.f5SN2BX30955@jen.americas.sgi.com>
To: "Holger Smolinski" <HSmolinski@de.ibm.com>
cc: Steve Lord <lord@sgi.com>, linux-xfs@oss.sgi.com
Subject: Re: Contraint to Blksize 512? 
In-Reply-To: Message from "Holger Smolinski" <HSmolinski@de.ibm.com> 
   of "Thu, 28 Jun 2001 16:49:13 +0200." <OFC80C900A.5B01308A-ONC1256A79.0051271F@de.ibm.com> 
Date: Thu, 28 Jun 2001 18:02:11 -0500
From: Steve Lord <lord@sgi.com>
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk

> 
> >> Hello,
> >> I tried to use XFS with devices of sector sizes larger than 512 Byte and
> >> failed.
> >> Apparently there are dependencies to a fixed blocksize of 512 Byte all
> over
> >> the code of XFS.
> >> Are these contraints unavoidable by the design of XFS or could you
> imagine
> >> running an XFS e.g. on a device with 4k blocksize and what needs to be
> done
> >> to the code in order to do so?
> >Yes, unfortunately XFS is built around assumptions about 512 byte block
> >sizes. There was an internal project to clean this up which was shelved
> >a while back, I will ask around and see if the code still exists. I
> >think this would end up with an incompatible on disk format, given that
> >some structures would change size, but I think that would be an acceptable
> >solution for devices with a larger block size.
> 
> For sure we would have incompatible layouts between XFS(512) and XFS(4k).
> I'd imagine sth like replacing any 512 in the XFS code by blksize(device)
> or blksize(xfs-type) rsp.
> Do you see any design obstacles in doing so?

I have just taken a look at an old project which was doing exactly this, XFS
was extended to be runtime switchable between blocksizes. The code is against
an old version of Irix, so it is not so easy to extract into a linux tree,
but it looks like most of the changes are there. It is a little more complex
than just looking for uses of 512 though. I will dig into this some more.

Steve

> 
> Gruesse / Regards
>         Dr. Holger Smolinski
> 
> 
> 
> 
>