Re: [PATCH] fs: reintroduce freezing nesting

[Mateusz Guzik]

On Tue, Apr 19, 2016 at 10:48:41PM +0200, Florian Margaine wrote:
> The behavior was removed inÂ18e9e5104fcd9a973ffe3eed3816c87f2a1b6cd2
> noting that this was a better idea than using a counter. However, this
> behavior is actually wanted if multiple applications want to freeze
> concurrently while remaining non-racy.
> 
> This patch reintroduces this feature by using a counter.
> 

This patch is wrong.

It uses non-atomic ops to modify the counter and no locks are
held to protect it.

I would argue the code should track that freezing has started and
additional freezers must only return when the state is
SB_FREEZE_COMPLETE.

> ---
> Âfs/super.cÂÂÂÂÂÂÂÂÂ| 15 +++++++++++----
> Âinclude/linux/fs.h |ÂÂ1 +
> Â2 files changed, 12 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/super.c b/fs/super.c
> index 74914b1..9fa8ca1 100644
> --- a/fs/super.c
> +++ b/fs/super.c
> @@ -231,6 +231,7 @@ static struct super_block *alloc_super(struct
> file_system_type *type, int flags)
> Â	Â*/
> Â	down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
> Â	s->s_count = 1;
> +	s->s_freezers = 0;
> Â	atomic_set(&s->s_active, 1);
> Â	mutex_init(&s->s_vfs_rename_mutex);
> Â	lockdep_set_class(&s->s_vfs_rename_mutex, &type-
> >s_vfs_rename_key);
> @@ -1275,12 +1276,12 @@ int freeze_super(struct super_block *sb)
> Â{
> Â	int ret;
> Â
> +	sb->s_freezers++;
> +	if (sb->s_writers.frozen != SB_UNFROZEN)
> +		return 0;
> +
> Â	atomic_inc(&sb->s_active);
> Â	down_write(&sb->s_umount);
> -	if (sb->s_writers.frozen != SB_UNFROZEN) {
> -		deactivate_locked_super(sb);
> -		return -EBUSY;
> -	}
> Â
> Â	if (!(sb->s_flags & MS_BORN)) {
> Â		up_write(&sb->s_umount);
> @@ -1338,14 +1339,20 @@ EXPORT_SYMBOL(freeze_super);
> Â * @sb: the super to thaw
> Â *
> Â * Unlocks the filesystem and marks it writeable again after
> freeze_super().
> + * Since nesting freezes is allowed, only the last freeze actually
> unlocks.
> Â */
> Âint thaw_super(struct super_block *sb)
> Â{
> Â	int error;
> Â
> +	sb->s_freezers--;
> +	if (sb->s_freezers > 0)
> +		return 0;
> +
> Â	down_write(&sb->s_umount);
> Â	if (sb->s_writers.frozen == SB_UNFROZEN) {
> Â		up_write(&sb->s_umount);
> +		sb->s_freezers++;
> Â		return -EINVAL;
> Â	}
> Â
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index e514f76..c045e2a 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -1333,6 +1333,7 @@ struct super_block {
> Â	struct quota_info	s_dquot;	/* Diskquota specific
> options */
> Â
> Â	struct sb_writers	s_writers;
> +	int			s_freezers;
> Â
> Â	char s_id[32];				/* Informational
> name */
> Â	u8 s_uuid[16];				/* UUID */
> --Â
> 2.8.0
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Mateusz Guzik