Received-SPF: pass (google.com: domain of linux-kernel+bounces-159622-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:40f1:3f00::1 as permitted sender) client-ip=2604:1380:40f1:3f00::1;
Subject: Re: [PATCH v5 4/9] xfs: convert delayed extents to unwritten when
 zeroing post eof blocks
To: Christoph Hellwig <hch@infradead.org>
Cc: "Darrick J. Wong" <djwong@kernel.org>, linux-xfs@vger.kernel.org,
 linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
 brauner@kernel.org, david@fromorbit.com, chandanbabu@kernel.org,
 tytso@mit.edu, jack@suse.cz, yi.zhang@huawei.com, chengzhihao1@huawei.com,
 yukuai3@huawei.com
References: <20240425131335.878454-1-yi.zhang@huaweicloud.com>
 <20240425131335.878454-5-yi.zhang@huaweicloud.com>
 <20240425182904.GA360919@frogsfrogsfrogs>
 <3be86418-e629-c7e6-fd73-f59f97a73a89@huaweicloud.com>
 <ZitKncYr0cCmU0NG@infradead.org>
From: Zhang Yi <yi.zhang@huaweicloud.com>
Message-ID: <5b6228ce-c553-3387-dfc4-2db78e3bd810@huaweicloud.com>
Date: Fri, 26 Apr 2024 15:18:17 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101
 Thunderbird/78.12.0
Precedence: bulk
MIME-Version: 1.0
In-Reply-To: <ZitKncYr0cCmU0NG@infradead.org>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit

On 2024/4/26 14:33, Christoph Hellwig wrote:
> On Fri, Apr 26, 2024 at 02:24:19PM +0800, Zhang Yi wrote:
>> Yeah, it looks more reasonable. But from the original scene, the
>> xfs_bmap_extsize_align() aligned the new extent that added to the cow fork
>> could overlaps the unreflinked range, IIUC, I guess that spare range is
>> useless exactly, is there any situation that would use it?
> 
> I've just started staring at this (again) half an hour ago, and I fail
> to understand the (pre-existing) logic in xfs_reflink_zero_posteof.
> 
> We obviously need to ensure data between i_size and the end of the
> block that i_size sits in is zeroed (but IIRC we already do that
> in write and truncate anyway).  But what is the point of zeroing
> any speculative preallocation beyond the last block that actually
> contains data?  Just truncating the preallocation and freeing
> the delalloc and unwritten blocks seems like it would be way
> more efficient.
> 

I've had the same idea before, I asked Dave and he explained that Linux
could leak data beyond EOF page for some cases, e.g. mmap() can write to
the EOF page beyond EOF without failing, and the data in that EOF page
could be non-zeroed by mmap(), so the zeroing is still needed now.

OTOH, if we free the delalloc and unwritten blocks beyond EOF blocks, he
said it could lead to some performance problems and make thinks
complicated to deal with the trimming of EOF block. Please see [1]
for details and maybe Dave could explain more.

[1] https://lore.kernel.org/linux-xfs/ZeERAob9Imwh01bG@dread.disaster.area/

Thanks,
Yi.