Re: [PATCH 0/3] fanotify support for btrfs sub-volumes

From: Christian Brauner <brauner@kernel.org>
To: Christoph Hellwig <hch@infradead.org>
Cc: Josef Bacik <josef@toxicpanda.com>,
	Amir Goldstein <amir73il@gmail.com>, Jan Kara <jack@suse.cz>,
	Chris Mason <clm@fb.com>, David Sterba <dsterba@suse.com>,
	linux-btrfs@vger.kernel.org, linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH 0/3] fanotify support for btrfs sub-volumes
Date: Tue, 31 Oct 2023 13:50:46 +0100	[thread overview]
Message-ID: <20231031-anorak-sammeln-8b1c4264f0db@brauner> (raw)
In-Reply-To: <ZUDxli5HTwDP6fqu@infradead.org>

On Tue, Oct 31, 2023 at 05:22:46AM -0700, Christoph Hellwig wrote:
> On Tue, Oct 31, 2023 at 01:14:42PM +0100, Christian Brauner wrote:
> > What happens in the kernel right now I've mentiond in the mount api
> > conversion patch for btrfs I sent out in June at [1] because I tweaked
> > that behavior. Say I mount both subvolumes:
> > 
> > mount /dev/sda -o subvol=subvol1 /vol1 # sb1@vfsmount1
> > mount /dev/sda -o subvol=subvol2 /vol2 # sb1@vfsmount2
> > 
> > It creates a superblock for /dev/sda. It then creates two vfsmounts: one
> > for subvol1 and one for subvol2. So you end up with two subvolumes on
> > the same superblock.
> > 
> > So if you mount a subvolume today then you already get separate
> > vfsmounts. To put it another way. If you start 10,000 containers each
> > using a separate btrfs subvolume then you get 10,000 vfsmounts.
> 
> But only if you mount them explicitly, which you don't have to.

Yep, I'm aware.

> 
> > Or is it that you want a separate superblock per subvolume?
> 
> Does "you" refer to me here?  No, I don't.
> 
> > Because only
> > if you allocate a new superblock you'll get clean device number
> > handling, no? Or am I misunderstanding this?
> 
> If you allocate a super block you get it for free.  If you don't
> you have to manually allocate it report it in ->getattr.

So this is effectively a request for:

btrfs subvolume create /mnt/subvol1

to create vfsmounts? IOW,

mkfs.btrfs /dev/sda
mount /dev/sda /mnt
btrfs subvolume create /mnt/subvol1
btrfs subvolume create /mnt/subvol2

would create two new vfsmounts that are exposed in /proc/<pid>/mountinfo
afterwards?

That might be odd. Because these vfsmounts aren't really mounted, no?
And so you'd be showing potentially hundreds of mounts in
/proc/<pid>/mountinfo that you can't unmount?

And even if you treat them as mounted what would unmounting mean? I'm
not saying that it's a show stopper but we would need a clear
understanding what the semantics are were after?

My knee-jerk reaction is that if we wanted each btrfs subvolume to be a
vfsmount then we don't want to have them show up in
/proc/<pid>/mountinfo _unless_ they're actually mounted.