From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1754742AbcETF0m (ORCPT <rfc822;w@1wt.eu>);
	Fri, 20 May 2016 01:26:42 -0400
Received: from zeniv.linux.org.uk ([195.92.253.2]:54020 "EHLO
	ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1750851AbcETF0l (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Fri, 20 May 2016 01:26:41 -0400
Date: Fri, 20 May 2016 06:26:39 +0100
From: Al Viro <viro@ZenIV.linux.org.uk>
To: Badhri Jagan Sridharan <badhri@google.com>
Cc: LKML <linux-kernel@vger.kernel.org>, linux-fsdevel@vger.kernel.org
Subject: Re: path-lookup inconsistency ?
Message-ID: <20160520052639.GT14480@ZenIV.linux.org.uk>
References: <CAPTae5+0v_Bec_=53Abuzxfq4raNZCt8eJHYMNeF=Pdbu===yA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAPTae5+0v_Bec_=53Abuzxfq4raNZCt8eJHYMNeF=Pdbu===yA@mail.gmail.com>
User-Agent: Mutt/1.6.0 (2016-04-01)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Thu, May 19, 2016 at 07:20:26PM -0700, Badhri Jagan Sridharan wrote:
> Hi,
> 
> I mounted overlayfs at /
> 
> My cat /proc/mounts looks like the following.
> # cat /proc/mounts
> /dev/root / squashfs ro,seclabel,relatime 0 0
> ..
> overlayfs / overlay
> rw,relatime,lowerdir=/,upperdir=/cache/upper,workdir=/cache/working 0
> 0
> 
> The original blockdevice at fs root is squashfs formatted so doesnt
> support write operations. I mounted overlayfs on fs root to cache the
> writes made.
> 
> While in /, the filesystem does not allow me to create files/directories,
> if I dont prefix it with ".." directive.

Chroot to the thing that overmounts root.  All that .. is doing here is
triggering the traversal to whatever covers the parent (== whatever
covers the root itself, since the parent of root is the root itself).
And we do *not* cross a mountpoint if the starting point of lookup
happens to be covered.  Might've been better if we did a different semantics,
but this one is userland-exposed and, worse yet, a bunch of early boot
userland code relies upon it.

What you want is to be chrooted to whatever overmounts the global root.
The easiest way is probably chroot("/..");chdir("/"); when you are
setting the things up.  Before you exec the final /sbin/init...