[Ocfs2-tools-devel] libocfs2 allocation of clusters and fsck

[Joel Becker]

On Thu, Oct 28, 2004 at 10:58:28AM -0700, Zach Brown wrote:
> Mmm, that last bit is compelling.  Rather than comparing, though, it
> seems we should just have fsck build up the bitmap as it scans and then
> use the library to sync it back.

	Is that what ext2 does?

> > o Try to populate an ocfs2_bitmap on the fly as needed.  That is, if you
> >   ask for bit 1000, read as many chains as I can until I find the one
> >   that matches bit 1000.
> 
> Oh, right, I keep forgetting that the chains aren't built for speedy
> lookups.  That's compelling, too.

	To be fair, chains _are_ built for speedy lookup of bits, but
not for speedy lookup of a particular group.
	Note that "bit 1000" in a chain is a misnomer, the bits aren't
position dependant.  So it's point-in-time.
	Still, the problems of:

o (compare to | set from) the fsck map
o build an initialized representation during mkfs

both benefit from an in-memory model.  Ok, I'm convinced.
	I'm having a fun time making the bitmap structure fit, though.

Joel

-- 

	f/8 and be there.

Joel Becker
Senior Member of Technical Staff
Oracle Corporation
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127