Received-SPF: pass (google.com: domain of linux-kernel+bounces-100513-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Date: Tue, 12 Mar 2024 09:27:29 -0700
From: "Darrick J. Wong" <djwong@kernel.org>
To: Christoph Hellwig <hch@infradead.org>
Cc: Zhang Yi <yi.zhang@huaweicloud.com>, linux-xfs@vger.kernel.org,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	brauner@kernel.org, david@fromorbit.com, tytso@mit.edu,
	jack@suse.cz, yi.zhang@huawei.com, chengzhihao1@huawei.com,
	yukuai3@huawei.com
Subject: Re: [PATCH 4/4] iomap: cleanup iomap_write_iter()
Message-ID: <20240312162729.GD1927156@frogsfrogsfrogs>
References: <20240311122255.2637311-1-yi.zhang@huaweicloud.com>
 <20240311122255.2637311-5-yi.zhang@huaweicloud.com>
 <20240311160739.GV1927156@frogsfrogsfrogs>
 <ZfBJYG5OHgLGewHv@infradead.org>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZfBJYG5OHgLGewHv@infradead.org>

On Tue, Mar 12, 2024 at 05:24:00AM -0700, Christoph Hellwig wrote:
> On Mon, Mar 11, 2024 at 09:07:39AM -0700, Darrick J. Wong wrote:
> > If at some point iomap_write_end actually starts returning partial write
> > completions (e.g. you wrote 250 bytes, but for some reason the pagecache
> > only acknowledges 100 bytes were written) then this code no longer
> > reverts the iter or truncates posteof pagecache correctly...
> 
> I don't think it makes sense to return a partial write from
> iomap_write_end.  But to make that clear it really should not return
> a byte count by a boolean.  I've been wanting to make that cleanup
> for a while, but it would reach all the way into buffer.c.

For now, can we change the return types of iomap_write_end_inline and
__iomap_write_end?  Then iomap can WARN_ON if the block_write_end return
value isn't 0 or copied:

	bool ret;

	if (srcmap->type == IOMAP_INLINE) {
		ret = iomap_write_end_inline(iter, folio, pos, copied);
	} else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) {
		size_t bh_written;

		bh_written = block_write_end(NULL, iter->inode->i_mapping,
				pos, len, copied, &folio->page, NULL);

		WARN_ON(bh_written != copied && bh_written != 0);
		ret = bh_written == copied;
	} else {
		ret = __iomap_write_end(iter->inode, pos, len, copied, folio);
	}

	...

	return ret;

Some day later we can circle back to bufferheads, or maybe they'll die
before we get to it. ;)

--D