On Mon, Feb 12, 2024 at 09:02:58AM +0200, Amir Goldstein wrote:
> If you are going to make kill_litter_super() an alias of kill_anon_super()
> I suggest going the extra mile and replacing the remaining 30 instances of
> kill_litter_super().
>
> If the rules become straight forward for the default ->kill_sb(),
> then maybe we should even make ->kill_sb() optional and do:
>
> diff --git a/fs/super.c b/fs/super.c
> index d35e85295489..6200cac0e4f8 100644
> --- a/fs/super.c
> +++ b/fs/super.c
> @@ -458,6 +458,18 @@ static void kill_super_notify(struct super_block *sb)
> super_wake(sb, SB_DEAD);
> }
>
> +static void kill_sb(struct super_block *s)
> +{
> + struct file_system_type *fs = s->s_type;
> +
> + if (fs->kill_sb)
> + fs->kill_sb(s);
> + else if (fs->fs_flags & FS_REQUIRES_DEV)
> + kill_block_super(s);
> + else
> + kill_anon_super(s);
> +}
Bloody bad idea, IMO. Note that straight use of kill_anon_super()
pretty much forces you into doing everything from ->put_super().
And that leads to rather clumsy failure exits in foo_fill_super(),
since you *won't* get ->put_super() called unless you've got to
setting ->s_root.
Considering how easily the failure exits rot, I'd rather discourage
that variant.
