xfs
[Top] [All Lists]

Re: [RFC, PATCH] xfs: make superblock version checks reflect reality

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [RFC, PATCH] xfs: make superblock version checks reflect reality
From: Christoph Hellwig <hch@xxxxxxxxxxxxx>
Date: Fri, 7 Mar 2014 02:16:37 -0800
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20140307083430.GQ6851@dastard>
References: <1394088890-10713-1-git-send-email-david@xxxxxxxxxxxxx> <20140306180534.GA305@xxxxxxxxxxxxx> <20140306225541.GL6851@dastard> <20140307083430.GQ6851@dastard>
User-agent: Mutt/1.5.21 (2010-09-15)
On Fri, Mar 07, 2014 at 07:34:30PM +1100, Dave Chinner wrote:
> Ok, we don't reject filesystems that don't have the NLINK bit set.
> Older filesystems that have only v1 inodes won't have that bit
> set, and we didn't set NLINK by default in mkfs until late 2007.
> Hence we need to keep some form of NLINK support around.
> 
> The alternative is to simply set the bit in the superblock if it is
> not set, and then just assume everywhere that it is set and we are
> using v2 inodes. That will get rid of the hasnlink/addnlink code
> needed to modify the superblock when the link count goes above
> MAX_NLINK_1, and will result in filesystems always converting v1
> inodes to v2 inodes on writeback of dirty inodes. I don't see a
> problem with taking this approach, bt maybe I'm missing something?

I'd love to get rid of v1 support sprinkled all over.  Obviously this
should be a separate patch, but making both the code and the in-use
filesystems less diverse is a good idea.

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