From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx3-rdu2.redhat.com ([66.187.233.73]:41232 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1750876AbeBZQf0 (ORCPT ); Mon, 26 Feb 2018 11:35:26 -0500 Date: Mon, 26 Feb 2018 17:35:23 +0100 From: Jiri Benc To: Andrew Cann Cc: netdev@vger.kernel.org Subject: Re: Automatic TAP destruction/Monitoring namespace destruction Message-ID: <20180226173523.63d48632@redhat.com> In-Reply-To: <20180223093937.GA8120@canndrew.org> References: <20180223093937.GA8120@canndrew.org> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: netdev-owner@vger.kernel.org List-ID: On Fri, 23 Feb 2018 04:39:37 -0500, Andrew Cann wrote: > In a program I'm writing I have a network namespace with a virtual (TAP) > network interface assigned to it. I would like it so that the interface is > automatically destroyed when the namespace is destroyed (ie. when the last > process in the namespace exits). I can't see any way to implement this.. This should just work. > As I understand it, when a namespace is destroyed all its interfaces are moved > to the root namespace. If this is the case, is there anyway to detect when an > interface is moved so that I can close it manually? It is the case only for interfaces backed by a physical device. Virtual interfaces are deleted when the netns is destroyed. That includes tun/tap interfaces. > Alternatively, is there a way to detect when a namespace is destroyed? I don't think we emit any netlink event on netns exit. > I figured it might possible to use inotify to do this, but it won't let me > watch directories under /proc. Also the files under /proc/*/ns/ seem to be some > kind of wierd symlink-to-a-raw-inode-thing (?) - is there a way to detect when > an inode is destroyed that I can use with these? You'd need this patchset: https://lkml.org/lkml/2016/10/15/40 but I don't think it went anywhere. Plus it probably wouldn't be enough anyway. > I also thought it might be possible to use a netlink socket to detect when an > interface changes namespace. But the netlink docs don't seem to suggest that > this is possible. Yes, that's possible. You'll need a recent kernel with commit e8368d9ebb94 included. > Basically I'm looking for any event the Linux kernel can give me that I can use > to implement what I want. Does anyone have any ideas? What you want should already be happening automatically. Have you tried? ip netns add ns0 ip -n ns0 tuntap add name tap0 mode tap ip -n ns0 link show dev tap0 ip netns del ns0 ip a # no tap interface Jiri