From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jiri Benc Subject: Re: [PATCH net] openvswitch: Fix egress tunnel info. Date: Wed, 7 Oct 2015 10:09:55 +0200 Message-ID: <20151007100955.3f13120e@griffin> References: <1444067897-1951-1-git-send-email-pshelar@nicira.com> <20151005204020.56f50496@griffin> <20151006175634.200d69a0@griffin> <20151006204558.5f2f56fc@griffin> <20151006210358.65750aeb@griffin> <20151006213245.74810d81@griffin> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: netdev To: Pravin Shelar Return-path: Received: from mx1.redhat.com ([209.132.183.28]:55741 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752000AbbJGIJ7 (ORCPT ); Wed, 7 Oct 2015 04:09:59 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Tue, 6 Oct 2015 14:16:02 -0700, Pravin Shelar wrote: > When OVS needs to add a vxlan tunnel device, OVS can create the vxlan > device and check the parameters for the flag. If the flag does not > exist then fallback to vport-create. You omitted the very important step, delete the device before falling back. This means that with older kernels, there will be an interface created and destroyed shortly after. With all netlink notifications sent, all tools listening for netlink events seeing the interface appear and go, with this being logged in various places, etc. Sounds very hacky and confusing to anyone watching their logs or output of such tools. That is the confusion I was talking about. > This only needs to be done for > very first tunnel device. Thats how most of features are supported in > OVS. The big difference to the other features is this cannot be detected until half way through the setup. What I'm proposing instead is to introduce a way to clearly and unambiguously detect whether lwtunnels are supported or not. We'll need this anyway: kernel 4.3 won't really support IPv6 tunneling with ovs, yet there's currently no way to determine whether it's supported or not (and, unlike with lwtunnel detection, there's not even a hacky way). Querying the datapath for the supported features is needed nevertheless; it's only logical to use it for the lwtunnel vs. old vport decision, too. I don't understand why you're opposed to this: it's much cleaner and there's no problem with lwtunnels not being used with the 4.3 kernel, everything should work just fine. Jiri -- Jiri Benc