Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 23 Nov 2018 14:07:22 +0000
From:   Al Viro <viro@zeniv.linux.org.uk>
To:     Pan Bian <bianpan2016@163.com>
Cc:     Amir Goldstein <amir73il@gmail.com>,
        Miklos Szeredi <mszeredi@redhat.com>,
        "J. Bruce Fields" <bfields@fieldses.org>,
        Christoph Hellwig <hch@lst.de>,
        linux-kernel <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH V2] exportfs: do not read dentry after free
Message-ID: <20181123140722.GA2217@ZenIV.linux.org.uk>
References: <1542959793-118334-1-git-send-email-bianpan2016@163.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1542959793-118334-1-git-send-email-bianpan2016@163.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Nov 23, 2018 at 03:56:33PM +0800, Pan Bian wrote:
> The function dentry_connected calls dput(dentry) to drop the previously
> acquired reference to dentry. In this case, dentry can be released.
> After that, IS_ROOT(dentry) checks the condition
> (dentry == dentry->d_parent), which may result in a use-after-free bug.
> This patch directly compares dentry with its parent obtained before
> dropping the reference.

It's a bit more subtle than the description implies (the race has
dentry connected during dget_parent() and the child we'd reached it
through moved elsewhere during the dput()), but you are right - the
race is there and that patch fixes it.

I wonder if we could avoid those dget_parent()/dput() completely -
looks like we might be able to with rcu_read_lock() and some
care.  OTOH, that's not going to be a hot path, anyway...

Applied.