Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 01 Sep 2020 13:01:27 -0700 (PDT)
Message-Id: <20200901.130127.236989626732311083.davem@davemloft.net>
To:     rkovhaev@gmail.com
Cc:     kuba@kernel.org, netdev@vger.kernel.org, bpf@vger.kernel.org,
        linux-kernel@vger.kernel.org, gregkh@linuxfoundation.org
Subject: Re: [PATCH] veth: fix memory leak in veth_newlink()
From:   David Miller <davem@davemloft.net>
In-Reply-To: <20200830131336.275844-1-rkovhaev@gmail.com>
References: <20200830131336.275844-1-rkovhaev@gmail.com>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Rustam Kovhaev <rkovhaev@gmail.com>
Date: Sun, 30 Aug 2020 06:13:36 -0700

> when register_netdevice(dev) fails we should check whether struct
> veth_rq has been allocated via ndo_init callback and free it, because,
> depending on the code path, register_netdevice() might not call
> priv_destructor() callback
> 
> Reported-and-tested-by: syzbot+59ef240dd8f0ed7598a8@syzkaller.appspotmail.com
> Link: https://syzkaller.appspot.com/bug?extid=59ef240dd8f0ed7598a8
> Signed-off-by: Rustam Kovhaev <rkovhaev@gmail.com>

I think I agree with Toshiaki here.  There is no reason why the
rollback_registered() path of register_netdevice() should behave
differently from the normal control flow.

Any code path that invokes ->ndo_uninit() should probably also
invoke the priv destructor.

The question is why does the err_uninit: label of register_netdevice
behave differently from rollback_registered()?  If there is a reason,
it should be documented in a comment or similar.  If it is wrong,
it should be corrected.