[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZjUTtuNqi9o9xgwg@infradead.org>
Date: Fri, 3 May 2024 09:41:26 -0700
From: Christoph Hellwig <hch@...radead.org>
To: "Ritesh Harjani (IBM)" <ritesh.list@...il.com>
Cc: linux-ext4@...r.kernel.org, linux-xfs@...r.kernel.org,
linux-fsdevel@...r.kernel.org, Dave Chinner <david@...morbit.com>,
Matthew Wilcox <willy@...radead.org>,
Christoph Hellwig <hch@...radead.org>,
"Darrick J . Wong" <djwong@...nel.org>,
Christian Brauner <brauner@...nel.org>,
Ojaswin Mujoo <ojaswin@...ux.ibm.com>, Jan Kara <jack@...e.cz>
Subject: Re: [RFC] Documentation: Add initial iomap document
> +i.e. at a high level how iomap does write iter is [1]::
> + user IO
> + loop for file IO range
> + loop for each mapped extent
> + if (buffered) {
> + loop for each page/folio {
> + instantiate page cache
> + copy data to/from page cache
> + update page cache state
> + }
> + } else { /* direct IO */
> + loop for each bio {
> + pack user pages into bio
> + submit bio
> + }
> + }
> + }
> + }
This would probably be easier to follow (and match the actual code)
with two separate flows for buffered vs direct I/O.
The loop for each bio in direct I/O is also not wrong, but maybe a
bit confusing as we really map over a pinned memory range. The code
just has a little optimization to directly place that into a bio.
> +Guideline for filesystem conversion to iomap
> +=============================================
Maybe split this into a separate file?
Powered by blists - more mailing lists