Re: [PATCH] btrfs: prevent subvol with swapfile from being deleted

[David Sterba <dsterba@suse.cz>]

On Tue, Mar 22, 2022 at 06:47:59PM +0800, Qu Wenruo wrote:
> > @@ -10419,7 +10425,17 @@ static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
> >   	 * before walking the extents because we don't want a concurrent
> >   	 * snapshot to run after we've already checked the extents.
> >   	 */
> > +	spin_lock(&root->root_item_lock);
> > +	if (btrfs_root_dead(root)) {
> 
> This looks a little weird to me.
> 
> If the root is already dead, it means we should not be able to access
> any file inside the subvolume.
> 
> How could we go into btrfs_swap_activate() while the root is already dead?
> 
> Or is there some special race I missed?

The deletion has a few steps, eg. the dentry is removed, root is marked
as dead and different thing locking protection.

I was wondering about file descriptor access to the subvolume and
calling swapon on that, but swapon/swapoff is a syscall and work with a
path argument so that can't happen. I haven't checked in what order are
the dentry removal and dead flag done, it could be theoretically
possible that there's a short window where the dentry is accessible and
subvolume already marked.