xfs
[Top] [All Lists]

Re: [PATCH 00/22] xfs: metadata CRCs, fourth version

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 00/22] xfs: metadata CRCs, fourth version
From: Ben Myers <bpm@xxxxxxx>
Date: Sat, 27 Apr 2013 15:42:58 -0500
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1364965892-19623-1-git-send-email-david@xxxxxxxxxxxxx>
References: <1364965892-19623-1-git-send-email-david@xxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
Hi Dave,

On Wed, Apr 03, 2013 at 04:11:10PM +1100, Dave Chinner wrote:
> New CRC patchset. Previous versions:
> 
> http://oss.sgi.com/archives/xfs/2013-01/msg00328.html
> http://oss.sgi.com/archives/xfs/2013-02/msg00451.html
> http://oss.sgi.com/archives/xfs/2013-03/msg00291.html
> 
> This version is based on 3.9-rc4 + TOT xfsdev. 3.9-rc5 has loopback
> bugs that make it useless for testing, so I've just kept my tree on
> -rc4.
> 
> New in this patch set:
> 
>         - numerous bug fixes
>       - cleanups to addresse Bens review comments
>       - logs entire inode allocation buffers
>       - reworks the buffer type tracking for log recovery
>       - fixes the endian issues reported by sparse
>       - splits out the symlink code rearrangement
>       - adds support for v5 superblock feature masks
>         - add mount warnings about CRC support being experimental
> 
> Still to do:
> 
>         - Documentation (half written, not in series)
>         - DT_* type fields in the directory entries. This can be
>         done with a feature bit if not done in time.
>         - storage of attributes larger than 256 bytes in shortform
>           attribute forks. Same comment about a feature bit.
> 
> The addition of the feature mask support to the superblock added a
> new field to the superblock - a log compatibility feature mask. This
> is to allow new transactions and recovery features to be added and
> prevent kernels that don't understand those features from performing
> log recovery. The idea is that clean logs can be mounted by kernels
> that don't support the new feature if everything else is compatible,
> but if they require log recovery the mount will be aborted. This
> means pure log changes don't require a compat/incompat/rocompat
> feature bit to be set.
> 
> The result of adding this is that all the old CRC enabled
> filesystems will not work with this kernel - the CRC field location
> changed, and that makes the mount fail hard. So a new userspace will
> be needed to test the CRC side of this patchset. (Coming soon!)
> 
> Comments, testing welcome...

I've pulled in this series.  There are still a few outstanding questions
from review.  I'm confident that we can get them sorted in the coming week.

Reviewed-by: Ben Myers <bpm@xxxxxxx>

Cheers,
        Ben

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