Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <3b0c9923-e12f-abc7-b0d9-ac96b52ec88b@huawei.com>
Date:   Sat, 3 Jun 2023 11:49:38 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:91.0) Gecko/20100101
 Thunderbird/91.3.2
Subject: Re: [PATCH] e2fsck: restore sb->s_state before journal recover
To:     "Darrick J. Wong" <djwong@kernel.org>
CC:     <tytso@mit.edu>, <linux-ext4@vger.kernel.org>,
        <linfeilong@huawei.com>, <louhongxiang@huawei.com>
References: <20230602082759.4062633-1-zhanchengbin1@huawei.com>
 <20230602151858.GA16844@frogsfrogsfrogs>
From:   zhanchengbin <zhanchengbin1@huawei.com>
In-Reply-To: <20230602151858.GA16844@frogsfrogsfrogs>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 2023/6/2 23:18, Darrick J. Wong wrote:
> On Fri, Jun 02, 2023 at 04:27:59PM +0800, zhanchengbin wrote:
>> ext4_handle_error
>>      EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
>>      if remount-ro
>>          ext4_commit_super(sb);
>> As you can see, when the filesystem error in the kernel, the last sb commit
>> not record the journal, So sb->s_state will be overwritten by journal recover.
>> In some cases , modifying metadata and superblock data are placed in two
>> transactions, if the previous transaction is already in the journal, and
>> ext4_handle_error occurs when updating sb, the filesystem is still error even
>> if the journal is recovered(I know that this situation should not occur in
>> theory, but I encountered this error when testing quota. Therefore, I think
>> we cannot fully rely on the kernel).
>> So when the filesystem is error before the journal recover, keep the error
>> state and perform deep check later.
>>
>> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
>> ---
>>   e2fsck/journal.c | 4 ++++
>>   1 file changed, 4 insertions(+)
>>
>> diff --git a/e2fsck/journal.c b/e2fsck/journal.c
>> index c7868d89..6f49321d 100644
>> --- a/e2fsck/journal.c
>> +++ b/e2fsck/journal.c
>> @@ -1683,6 +1683,7 @@ errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
>>   	errcode_t	retval, recover_retval;
>>   	io_stats	stats = 0;
>>   	unsigned long long kbytes_written = 0;
>> +	__u16 state = ctx->fs->super->s_state;
>>   
>>   	printf(_("%s: recovering journal\n"), ctx->device_name);
>>   	if (ctx->options & E2F_OPT_READONLY) {
>> @@ -1722,6 +1723,9 @@ errcode_t e2fsck_run_ext3_journal(e2fsck_t ctx)
>>   	ctx->fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
>>   	ctx->fs->super->s_kbytes_written += kbytes_written;
>>   
>> +	if (EXT2_ERROR_FS | state)
> 
> Isn't this  ^^^^^^^^^^^^^^^^^^^^^ expression always nonzero? >
>> +		ctx->fs->super->s_state = state | EXT2_ERROR_FS;
> 
> /me doesn't understand this bit logic at all.

You can check this stack:
ext4_handle_error
     ext4_commit_super
         ext4_update_super
             if (sbi->s_add_error_count > 0) {
                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
  - bin.

> 
> --D
> 
>> +
>>   	/* Set the superblock flags */
>>   	e2fsck_clear_recover(ctx, recover_retval != 0);
>>   
>> -- 
>> 2.31.1
>>
> .
>