Re: copy on write for splice() from file to pipe?

From: Jeremy Allison <jra@samba.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andy Lutomirski <luto@kernel.org>, Jens Axboe <axboe@kernel.dk>,
	Linux API Mailing List <linux-api@vger.kernel.org>,
	Dave Chinner <david@fromorbit.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Matthew Wilcox <willy@infradead.org>,
	Stefan Metzmacher <metze@samba.org>,
	Al Viro <viro@zeniv.linux.org.uk>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Samba Technical <samba-technical@lists.samba.org>,
	io-uring <io-uring@vger.kernel.org>
Subject: Re: copy on write for splice() from file to pipe?
Date: Fri, 10 Feb 2023 11:27:51 -0800	[thread overview]
Message-ID: <Y+aat8sggTtgff+A@jeremy-acer> (raw)
In-Reply-To: <CAHk-=wjuXvF1cA=gJod=-6k4ypbEmOczFFDKriUpOVKy9dTJWQ@mail.gmail.com>

On Fri, Feb 10, 2023 at 11:18:05AM -0800, Linus Torvalds via samba-technical wrote:
>
>We should point the fingers at either the _user_ of splice - as Jeremy
>Allison has done a couple of times - or we should point it at the sink
>that cannot deal with unstable sources.
> ....
> - it sounds like the particular user in question (samba) already very
>much has a reasonable model for "I have exclusive access to this" that
>just wasn't used

Having said that, I just had a phone discussion with Ralph Boehme
on the Samba Team, who has been following along with this in
read-only mode, and he did point out one case I had missed.

1). Client opens file with a lease. Hurrah, we think we can use splice() !
2). Client writes into file.
3). Client calls SMB_FLUSH to ensure data is on disk.
4). Client reads the data just wrtten to ensure it's good.
5). Client overwrites the previously written data.

Now when client issues (4), the read request, if we
zero-copy using splice() - I don't think theres a way
we get notified when the data has finally left the
system and the mapped splice memory in the buffer cache
is safe to overwrite by the write (5).

So the read in (4) could potentially return the data
written in (5), if the buffer cache mapped memory has
not yet been sent out over the network.

That is certainly unexpected behavior for the client,
even if the client leased the file.

If that's the case, then splice() is unusable for
Samba even in the leased file case.

>   Maybe this thread raised some awareness of it for some people, but
>more realistically - maybe we can really document this whole issue
>somewhere much more clearly

Complete comprehensive documentation on this would
be extremely helpful (to say the least :-).