Re: [PATCH v5 04/11] blksnap: header file of the module interface

From: Christoph Hellwig <hch@infradead.org>
To: Dave Chinner <david@fromorbit.com>
Cc: Sergei Shtepa <sergei.shtepa@veeam.com>,
	axboe@kernel.dk, hch@infradead.org, corbet@lwn.net,
	snitzer@kernel.org, viro@zeniv.linux.org.uk, brauner@kernel.org,
	dchinner@redhat.com, willy@infradead.org, dlemoal@kernel.org,
	linux@weissschuh.net, jack@suse.cz, ming.lei@redhat.com,
	linux-block@vger.kernel.org, linux-doc@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
	Donald Buczek <buczek@molgen.mpg.de>
Subject: Re: [PATCH v5 04/11] blksnap: header file of the module interface
Date: Tue, 13 Jun 2023 23:26:25 -0700	[thread overview]
Message-ID: <ZIldkb1pwhNsSlfl@infradead.org> (raw)
In-Reply-To: <ZIjsywOtHM5nIhSr@dread.disaster.area>

On Wed, Jun 14, 2023 at 08:25:15AM +1000, Dave Chinner wrote:
> > + * Return: 0 if succeeded, negative errno otherwise.
> > + */
> > +#define IOCTL_BLKSNAP_SNAPSHOT_APPEND_STORAGE					\
> > +	_IOW(BLKSNAP, blksnap_ioctl_snapshot_append_storage,			\
> > +	     struct blksnap_snapshot_append_storage)
> 
> That's an API I'm extremely uncomfortable with. We've learnt the
> lesson *many times* that userspace physical mappings of underlying
> file storage are unreliable.
> 
> i.e.  This is reliant on userspace telling the kernel the physical
> mapping of the filesystem file to block device LBA space and then
> providing a guarantee (somehow) that the mapping will always remain
> unchanged. i.e. It's reliant on passing FIEMAP data from the
> filesystem to userspace and then back into the kernel without it
> becoming stale and somehow providing a guarantee that nothing (not
> even the filesystem doing internal garbage collection) will change
> it.

Hmm, I never thought of this API as used on files that somewhere
had a logical to physical mapping applied to them.

Sergey, is that the indtended use case?  If so we really should
be going through the file system using direct I/O.