Date:   Tue, 14 Feb 2023 12:48:35 +0100
From:   Jan Kara <jack@suse.cz>
To:     zhanchengbin <zhanchengbin1@huawei.com>
Cc:     tytso@mit.edu, jack@suse.com, linux-ext4@vger.kernel.org,
        yi.zhang@huawei.com, linfeilong@huawei.com,
        liuzhiqiang26@huawei.com
Subject: Re: [PATCH v4 1/2] ext4: fix inode tree inconsistency caused by
 ENOMEM in ext4_split_extent_at
Message-ID: <20230214114835.hpjr4zgofrcp7hyy@quack3>
References: <20230213040522.3339406-1-zhanchengbin1@huawei.com>
 <20230213040522.3339406-2-zhanchengbin1@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230213040522.3339406-2-zhanchengbin1@huawei.com>
Precedence: bulk

On Mon 13-02-23 12:05:21, zhanchengbin wrote:
> If ENOMEM fails when the extent is splitting, we need to restore the length
> of the split extent.
> In the call stack of the ext4_split_extent_at function, only in
> ext4_ext_create_new_leaf will it alloc memory and change the shape of the
> extent tree,even if an ENOMEM is returned at this time, the extent tree is
> still self-consistent, Just restore the split extent lens in the function
> ext4_split_extent_at.
> 
> ext4_split_extent_at
>  ext4_ext_insert_extent
>   ext4_ext_create_new_leaf
>    1)ext4_ext_split
>      ext4_find_extent
>    2)ext4_ext_grow_indepth
>      ext4_find_extent
> 
> Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
> ---
>  fs/ext4/extents.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
> index 9de1c9d1a13d..0f95e857089e 100644
> --- a/fs/ext4/extents.c
> +++ b/fs/ext4/extents.c
> @@ -935,6 +935,7 @@ ext4_find_extent(struct inode *inode, ext4_lblk_t block,
>  
>  		bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
>  		if (IS_ERR(bh)) {
> +			EXT4_ERROR_INODE(inode, "IO error reading extent block");

Why have you added this? Usually we don't log any additional errors for IO
errors because the storage layer already reports it... Furthermore this
would potentialy panic the system / remount the fs RO which we also usually
don't do in case of IO errors, only in case of FS corruption.

								Honza

>  			ret = PTR_ERR(bh);
>  			goto err;
>  		}
> @@ -3251,7 +3252,7 @@ static int ext4_split_extent_at(handle_t *handle,
>  		ext4_ext_mark_unwritten(ex2);
>  
>  	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
> -	if (err != -ENOSPC && err != -EDQUOT)
> +	if (err != -ENOSPC && err != -EDQUOT && err != -ENOMEM)
>  		goto out;
>  
>  	if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
> -- 
> 2.31.1
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR