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;
Date:   Fri, 2 Jun 2023 08:18:58 -0700
From:   "Darrick J. Wong" <djwong@kernel.org>
To:     zhanchengbin <zhanchengbin1@huawei.com>
Cc:     tytso@mit.edu, linux-ext4@vger.kernel.org, linfeilong@huawei.com,
        louhongxiang@huawei.com
Subject: Re: [PATCH] e2fsck: restore sb->s_state before journal recover
Message-ID: <20230602151858.GA16844@frogsfrogsfrogs>
References: <20230602082759.4062633-1-zhanchengbin1@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230602082759.4062633-1-zhanchengbin1@huawei.com>
Precedence: bulk

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.

--D

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