Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH] f2fs: let f2fs also gc atomic file to avoid loop gc
To:     Chao Yu <yuchao0@huawei.com>, <jaegeuk@kernel.org>,
        <chao@kernel.org>, <yunlong.song@icloud.com>
CC:     <miaoxie@huawei.com>, <bintian.wang@huawei.com>,
        <linux-fsdevel@vger.kernel.org>,
        <linux-f2fs-devel@lists.sourceforge.net>,
        <linux-kernel@vger.kernel.org>
References: <1510108497-58331-1-git-send-email-yunlong.song@huawei.com>
 <239f770e-53f0-69c6-fc1d-67aeb9a1065a@huawei.com>
 <a26c3d6a-5720-31d0-da29-73cf45e2ade0@huawei.com>
 <d5c26f31-2e0c-8b4d-c7f4-b11ff38c20fb@huawei.com>
From:   Yunlong Song <yunlong.song@huawei.com>
Message-ID: <383339b7-528d-3085-1399-1f1694178c15@huawei.com>
Date:   Fri, 17 Nov 2017 11:04:36 +0800
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:52.0) Gecko/20100101
 Thunderbird/52.2.1
MIME-Version: 1.0
In-Reply-To: <d5c26f31-2e0c-8b4d-c7f4-b11ff38c20fb@huawei.com>
Content-Type: text/plain; charset="windows-1252"; format=flowed
Content-Transfer-Encoding: 7bit
Content-Language: en-US
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

The atomic commit will trigger:
     -f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true)
         -file_write_and_wait_range(file, 0, LLONG_MAX)
         -fsync_node_pages
             -__write_node_page
                 -REQ_PREFLUSH | REQ_FUA

So data is flushed to non-volatile before  last node write with 
REQ_PREFLUSH | REQ_FUA,
we do not need to worry about the inconsistent problem. Right?

On 2017/11/17 10:49, Chao Yu wrote:
> On 2017/11/17 8:58, Yunlong Song wrote:
>> Is there any problem if just deleting the judgement condition in this patch?
> IIRC, dirty node comes from data segment GC can be writebacked & flushed during
> atomic commit, but related data will still be in inner bio cache, after later
> SPOR, data would be inconsistent.
>
> Thanks,
>
>> On 2017/11/8 17:28, Chao Yu wrote:
>>> On 2017/11/8 10:34, Yunlong Song wrote:
>>>> If some files are opened with atomic flag and have not commited yet, at
>>>> the same time, if all the target victim segments have at least one page
>>>> of these atomic files, then f2fs gc will fail to do gc and hangs in the
>>>> process of go to gc_more, since gc_date_segment will not move any data
>>>> and get_valid_blocks will never be 0, then do_garbage_collect will
>>>> always return 0.
>>> Oh, I added this judgment condition to avoid ruining atomic write by data
>>> GC, could we find another way to solve this issue? BTW, if there is direct
>>> IO, we will also skip data segment GC.
>>>
>>> Thanks,
>>>
>>>> Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
>>>> ---
>>>>    fs/f2fs/gc.c | 6 ------
>>>>    1 file changed, 6 deletions(-)
>>>>
>>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>>> index 5d5bba4..3fdcd04 100644
>>>> --- a/fs/f2fs/gc.c
>>>> +++ b/fs/f2fs/gc.c
>>>> @@ -621,9 +621,6 @@ static void move_data_block(struct inode *inode, block_t bidx,
>>>>    	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
>>>>    		goto out;
>>>>    
>>>> -	if (f2fs_is_atomic_file(inode))
>>>> -		goto out;
>>>> -
>>>>    	set_new_dnode(&dn, inode, NULL, NULL, 0);
>>>>    	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
>>>>    	if (err)
>>>> @@ -718,9 +715,6 @@ static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
>>>>    	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
>>>>    		goto out;
>>>>    
>>>> -	if (f2fs_is_atomic_file(inode))
>>>> -		goto out;
>>>> -
>>>>    	if (gc_type == BG_GC) {
>>>>    		if (PageWriteback(page))
>>>>    			goto out;
>>>>
>>> .
>>>
>
> .
>

-- 
Thanks,
Yunlong Song


From 1584307758154174969@xxx Fri Nov 17 10:13:38 +0000 2017
X-GM-THRID: 1583469797370502209
X-Gmail-Labels: Inbox,Category Forums,HistoricalUnread