Subject: Re: Patch to disallow mounts of unclean FFS unless forced
To: Charles M. Hannum <abuse@spamalicious.com>
From: Darrin B. Jewell <dbj@NetBSD.org>
List: tech-kern
Date: 10/14/2003 12:25:47
Perhaps I misunderstand the guarantees granted by a correctly
functioning softdep implementation, although I thought that apart from
unallocated blocks and unreferenced inodes the filesystem
appeared consistent after a softdep crash.

In particular, filesystem directory structure is consistent and
that filesystem metadata accesible from that directory structure
is consistent enough to not panic the kernel.

However, I admit that I haven't audited the code for this
functionality, and perhaps didn't read the softdep paper
with enough skepticisim.

Can you walk me through a potential problem case?

Darrin

"Charles M. Hannum" <abuse@spamalicious.com> writes:

> On Monday 13 October 2003 06:49 pm, Jason Thorpe wrote:
> > On Monday, October 13, 2003, at 11:09  AM, Jason Thorpe wrote:
> > > Comments?
> >
> > Darrin pointed out to me that it should be allowed to mount an unclean
> > FFS that was softdep-enabled, since the only inconsistencies in that
> > case should be lost file system blocks.
> 
> That theory dates back to something Kirk said years ago, which he has now 
> recanted because I pointed out several cases where he was wrong.