[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZjIMQTAtxZ0NhCD2@dread.disaster.area>
Date: Wed, 1 May 2024 19:32:49 +1000
From: Dave Chinner <david@...morbit.com>
To: Zhang Yi <yi.zhang@...weicloud.com>
Cc: linux-ext4@...r.kernel.org, linux-fsdevel@...r.kernel.org,
linux-mm@...ck.org, linux-kernel@...r.kernel.org, tytso@....edu,
adilger.kernel@...ger.ca, jack@...e.cz, ritesh.list@...il.com,
hch@...radead.org, djwong@...nel.org, willy@...radead.org,
zokeefe@...gle.com, yi.zhang@...wei.com, chengzhihao1@...wei.com,
yukuai3@...wei.com, wangkefeng.wang@...wei.com
Subject: Re: [RFC PATCH v4 29/34] ext4: fall back to buffer_head path for
defrag
On Wed, Apr 10, 2024 at 11:03:08PM +0800, Zhang Yi wrote:
> From: Zhang Yi <yi.zhang@...wei.com>
>
> Online defrag doesn't support iomap path yet, we have to fall back to
> buffer_head path for the inode which has been using iomap. Changing
> active inode is dangerous, before we start, we must hold the inode lock
> and the mapping->invalidate_lock, and writeback all dirty folios and
> drop the inode's pagecache.
Even then, I don't think this is obviously safe. We went through
this with DAX and we couldn't make it work safely.
Just return EOPNOTSUPP to the online defrag ioctl if iomap is in use
- that avoids all the excitement involved in doing dangerous things
like swapping aops structures on actively referenced inodes...
-Dave.
--
Dave Chinner
david@...morbit.com
Powered by blists - more mailing lists