[Top] [All Lists]

Re: RFC: log record CRC validation

To: Barry Naujok <bnaujok@xxxxxxx>
Subject: Re: RFC: log record CRC validation
From: David Chinner <dgc@xxxxxxx>
Date: Wed, 1 Aug 2007 12:43:40 +1000
Cc: David Chinner <dgc@xxxxxxx>, Michael Nishimoto <miken@xxxxxxxxx>, markgw@xxxxxxx, xfs-dev <xfs-dev@xxxxxxx>, xfs-oss <xfs@xxxxxxxxxxx>
In-reply-to: <op.twcz6vvu3jf8g2@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>
References: <20070725092445.GT12413810@xxxxxxx> <46A7226D.8080906@xxxxxxx> <46A8DF7E.4090006@xxxxxxxxx> <20070726233129.GM12413810@xxxxxxx> <46A94963.7000103@xxxxxxxxx> <20070727065930.GT12413810@xxxxxxx> <46AFD88E.9070403@xxxxxxxxx> <20070801022418.GR31489@xxxxxxx> <op.twcz6vvu3jf8g2@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Sender: xfs-bounce@xxxxxxxxxxx
User-agent: Mutt/
On Wed, Aug 01, 2007 at 12:36:21PM +1000, Barry Naujok wrote:
> On Wed, 01 Aug 2007 12:24:18 +1000, David Chinner <dgc@xxxxxxx> wrote:
> >On Tue, Jul 31, 2007 at 05:49:18PM -0700, Michael Nishimoto wrote:
> >
> >>What are your plans for adding CRCs to other metadata objects?
> >
> >Other objects will require some on-disk format change, and that will
> >require a feature bit to be set.
> >
> >Basically, we can add CRCs to individual inodes without a version bump
> >as we have some empty space in the v2 inode core. That will make v2  
> >inodes
> >the default, but we already have a superblock bit for that and all  
> >versions
> >of linux support v2 inodes so there is no issue there. This will require
> >userspace tool changes, though, because tools like repair will revert v2
> >inodes back to v1 format if the are not using project id's or the link  
> >count
> >fits into 16 bits.....
> No, repair does not revert v2 inodes back to v1. Currently, inodes
> are created as v1 in the kernel and moved to v2 as required.

Ah, yes, you're right - it's too easy to make flow mistakes in 1200
line functions....


Dave Chinner
Principal Engineer
SGI Australian Software Group

<Prev in Thread] Current Thread [Next in Thread>