Re: [PATCH -next v2 4/6] ext4: avoid to re-read mmp check data get from page cache

[Jan Kara <jack@suse.cz>]

On Sat 11-09-21 17:00:57, Ye Bin wrote:
> As call read_mmp_block pass bh_check which value is NULL, then call
> sb_getblk to get buffer_head. But mmp_block's buffer_head is already exist
>  which also is uptodate. Lead to compare the same data.
> 
> Signed-off-by: Ye Bin <yebin10@huawei.com>

This looks good, except that read_mmp_block() also releases bh it is passed
in case of error. So it can free buffer head which is still referenced from
EXT4_SB(sb)->s_mmp_bh and cause use-after-free issues.

I guess I would just get rid of sb_getblk() in read_mmp_block() and always
expect valid bh passed. The only place that passes NULL bh after this
patch is one case in ext4_multi_mount_protect() and that can call
sb_getblk() on its own. That way we can also simplify read_mmp_block()
prototype to:

static int read_mmp_block(struct super_block *sb, struct buffer_head *bh);

								Honza

> ---
>  fs/ext4/mmp.c | 17 ++++++-----------
>  1 file changed, 6 insertions(+), 11 deletions(-)
> 
> diff --git a/fs/ext4/mmp.c b/fs/ext4/mmp.c
> index 4433fe7e9e86..007bde3c12b8 100644
> --- a/fs/ext4/mmp.c
> +++ b/fs/ext4/mmp.c
> @@ -213,10 +213,7 @@ static int kmmpd(void *data)
>  		 * we need to check if the MMP block is as we write it.
>  		 */
>  		if (jiffies - last_check_time > mmp_check_interval * HZ) {
> -			struct buffer_head *bh_check = NULL;
> -			struct mmp_struct *mmp_check;
> -
> -			retval = read_mmp_block(sb, &bh_check, mmp_block);
> +			retval = read_mmp_block(sb, &bh, mmp_block);
>  			if (retval) {
>  				ext4_error_err(sb, -retval,
>  					       "error reading MMP data: %d",
> @@ -224,20 +221,18 @@ static int kmmpd(void *data)
>  				goto wait_to_exit;
>  			}
>  
> -			mmp_check = (struct mmp_struct *)(bh_check->b_data);
> -			if (seq != mmp_check->mmp_seq ||
> -			    memcmp(nodename, mmp_check->mmp_nodename,
> -				   sizeof(mmp->mmp_nodename))) {
> -				dump_mmp_msg(sb, mmp_check,
> +			mmp = (struct mmp_struct *)(bh->b_data);
> +			if (seq != le32_to_cpu(mmp->mmp_seq) ||
> +			    memcmp(nodename, mmp->mmp_nodename,
> +				    sizeof(nodename))) {
> +				dump_mmp_msg(sb, mmp,
>  					     "Error while updating MMP info. "
>  					     "The filesystem seems to have been"
>  					     " multiply mounted.");
>  				ext4_error_err(sb, EBUSY, "abort");
> -				put_bh(bh_check);
>  				retval = -EBUSY;
>  				goto wait_to_exit;
>  			}
> -			put_bh(bh_check);
>  			last_check_time = jiffies;
>  		}
>  
> -- 
> 2.31.1
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR