From mboxrd@z Thu Jan  1 00:00:00 1970
From: Al Viro <viro@ZenIV.linux.org.uk>
Subject: Re: [PATCH] vfs namespace: Don't assume mount namespace has valid
 root
Date: Wed, 14 Mar 2012 20:10:48 +0000
Message-ID: <20120314201048.GI23916@ZenIV.linux.org.uk>
References: <1331758710-16400-1-git-send-email-levinsasha928@gmail.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
	davej@redhat.com
To: Sasha Levin <levinsasha928@gmail.com>
Return-path: <linux-kernel-owner@vger.kernel.org>
Content-Disposition: inline
In-Reply-To: <1331758710-16400-1-git-send-email-levinsasha928@gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
List-Id: linux-fsdevel.vger.kernel.org

On Wed, Mar 14, 2012 at 04:58:30PM -0400, Sasha Levin wrote:
> This patch fixes the assumption that a mnt namespace will always have a valid
> root object.

It's not an assumption, it's an invariant that should hold unless you have
run into a bug somewhere.

Instances of struct mnt_namespace should *all* come from alloc_mnt_ns().
There are only two callers - dup_mnt_namespace() and create_mnt_ns().
The latter will assign non-NULL vfsmount to ->root or die NULL pointer
dereference in
                mnt->mnt_ns = new_ns;
The former will either assign non-NULL to ->root or kfree() mnt_namespace
before anyone can see it.

And nothing should modify ->root after that assignment for as long as
the instance of struct mnt_namespace is allocated.

Mind explaining how have you managed to get mnt_namespace with NULL ->root
passed to dup_mnt_ns()?