From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
MIME-Version: 1.0
In-Reply-To: <CALCETrUA2veiBYYbN1hCSkHxBFyDpe3b=PXg=j53_ZuQBt1FUw@mail.gmail.com>
References: <565394BE.4040506@codeaurora.org>
	<5696E366.2080605@codeaurora.org>
	<20160114045716.GB8006@kroah.com>
	<5697EF97.9020800@codeaurora.org>
	<CA+55aFwxaU0Y_UxyKFv6WaYsPWDDzZG-rhVtMuLSPcJ0WLLw2A@mail.gmail.com>
	<CALCETrWNQ9ytw1bCOOjFJRstauYc6DocQAmZCF61CErAJ5BF2g@mail.gmail.com>
	<CAB3-ZyT+xcU-0BnNvq-+GsDe6W8=MCWwxguS-+KQ39gajmM5pw@mail.gmail.com>
	<871t9i91e1.fsf@thinkpad.rath.org>
	<56994884.9060002@codeaurora.org>
	<CA+55aFw9N6Qt=5Q7z+rc_KsotiDgwizm-3Ym6Q70azkb2b3YOA@mail.gmail.com>
	<CALCETrUA2veiBYYbN1hCSkHxBFyDpe3b=PXg=j53_ZuQBt1FUw@mail.gmail.com>
Date: Fri, 15 Jan 2016 13:53:34 -0800
Message-ID: <CA+55aFyFLEkAthR=aUV2noQspgpZ8ascNCCftNBdEGaLvi199A@mail.gmail.com>
Subject: Re: [PATCH] fuse: Add support for fuse stacked I/O
From: Linus Torvalds <torvalds@linux-foundation.org>
To: Andy Lutomirski <luto@amacapital.net>
Cc: Nikhilesh Reddy <reddyn@codeaurora.org>,
	Nikolaus Rath <nikolaus@rath.org>,
	fuse-devel <fuse-devel@lists.sourceforge.net>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Greg KH <gregkh@linuxfoundation.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Linux API <linux-api@vger.kernel.org>, Jan Kara <jack@suse.cz>,
	"Theodore Ts'o" <tytso@mit.edu>, sven.utcke@gmx.de,
	Miklos Szeredi <miklos@szeredi.hu>,
	Richard Weinberger <richard@nod.at>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Antonio SJ Musumeci <trapexit@spawn.link>
Content-Type: text/plain; charset=UTF-8
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-fsdevel.vger.kernel.org>

On Fri, Jan 15, 2016 at 1:46 PM, Andy Lutomirski <luto@amacapital.net> wrote:

> If mmap sets vm_file to the underlying thing, wouldn't CRIU and
> anything else that uses map_files get confused?  Or did you have
> something else in mind?

Why would they care?

Also, I don't think you actually need to change vm_file - we have this
whole notion of "inode->i_data" vs "inode->i_mapping".

So I think you could set the "i_mapping" of the fuse inode to be the
i_mapping of the passed-through-to inode, and it should just work.

And no, I didn't look into this very deeply, and maybe there is some
annoying detail that would make that not work well. But that's part of
the whole point of the i_mapping indirection: so that you can share
the page cache when you have two separate anchor points. I think coda
uses it for the local caching, and block devices use it to not have
mapping aliases between different inodex that all are the same block
device.

                 Linus