SU+J, journaled softupdates
Jeff Roberson
jroberson at jroberson.net
Tue Dec 22 01:17:44 UTC 2009
On Tue, 22 Dec 2009, Robert Watson wrote:
> I'm CC'ing Jeff Roberson, who perhaps can comment on his on-going project to
> merge journaling techniques with soft updates in UFS (which is just meta-data
> journaling, but hopefully will address many of the fsck/bgfsck-related
> concerns people have).
Hi folks,
I have blogged a little bit about my plans at
http://jeffr_tech.livejournal.com/. Briefly, I have created a small
intent journal that works in concert with softupdates to eliminate the
requirement for fsck on boot. Instead fsck has been augmented with a
recovery process that reads the journal and corrects the filesystem. The
recovery process is very quick and scales with the size of the journal,
not the filesystem.
The journal is enabled with tunefs on an existing ffs filesystem. The
filesystem must be clean to enable. The filesystem must be clean before
mounting with a legacy implementation but it is otherwise metadata
compatible. A legacy fsck will destroy the journal but otherwise there
are no other problems.
Presently the code does not support non-journaled softupdates but I will
rectify that. Peter Holm (pho@) has been helping me test and once I've
fixed the final bugs revealed by his stress2 suite I will make a release
and provide instructions. I'm hoping many people will be interested so we
can quickly get to some confidence with the patch.
One thing to keep in mind is that at runtime this can only slow down
softupdates. It is additional overhead which I have attempted to minimize
but there is no way around it. To eliminate fsck you must pay a price.
This also can not cope with disks that lie about write ordering coupled
with power failures. In this case the normal fsck can recover the file
system.
Cheers,
Jeff
>
> Robert N M Watson
> Computer Laboratory
> University of Cambridge
>
More information about the freebsd-fs
mailing list