Return-path: Received: from arrakis.dune.hu ([78.24.191.176]:37301 "EHLO arrakis.dune.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1423419AbcBSQYk (ORCPT ); Fri, 19 Feb 2016 11:24:40 -0500 Subject: Re: [RFC] mac80211: add extap functionality To: Grzegorz Bajorski References: <1455710131-9967-1-git-send-email-grzegorz.bajorski@tieto.com> <56C4A5ED.4060802@openwrt.org> <56C5D066.8010405@openwrt.org> Cc: linux-wireless@vger.kernel.org, Johannes Berg From: Felix Fietkau Message-ID: <56C741C0.90002@openwrt.org> (sfid-20160219_172445_120767_E63584DB) Date: Fri, 19 Feb 2016 17:24:32 +0100 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: On 2016-02-19 13:05, Grzegorz Bajorski wrote: > 2016-02-18 15:08 GMT+01:00 Felix Fietkau : >> On 2016-02-18 14:36, Grzegorz Bajorski wrote: >>> 2016-02-17 17:55 GMT+01:00 Felix Fietkau : >>>> On 2016-02-17 12:55, Grzegorz Bajorski wrote: >>>>> Client interface briding was only possible when 4addr frames were used with >>>>> a 4addr/WDS aware AP. It was not possible to do it otherwise due to 3addr >>>>> frame limitation. >>>>> >>>>> The extap logic introduces a smart MAC address masking/translation >>>>> (including modyfing packets beyond SA/DA, e.g. DHCP broadcast flag is set). >>>>> >>>>> There are still some unsolved problems and bugs: >>>>> - due to bridge port routing and sk_buff payload sharing skb_copy() is >>>>> performed; this ideally should be reworked >>>>> - ipv6 support is still not finished >>>>> - extap is enabled by default currently; it should be configurable via >>>>> nl80211 the same way 4addr is >>>>> >>>>> There's also an idea to move this as a generic link driver (just like >>>>> macvlan, et al) which would allow unmodified cfg80211 drivers to enjoy the >>>>> extap functionality. Thoughts? >>>>> >>>>> Note: This changes cfg80211 file in this single patch only for reviewing >>>>> convienence. >>>>> >>>>> This is an early draft to solicit comments on the design. >>>>> >>>>> Signed-off-by: Grzegorz Bajorski >>>> You can get a lot of the same effect (sharing the same subnet between >>>> hosts behind multiple interfaces and having forwarding between them) >>>> without any changes to mac80211. >>>> >>>> OpenWrt uses a daemon called 'relayd' which I wrote some years ago. It >>>> does ARP translation, DHCP packet mangling and sets up policy routing to >>>> forward packets between multiple interfaces. >>>> >>>> You can find it here: >>>> http://git.openwrt.org/?p=project/relayd.git;a=summary >>>> git://git.openwrt.org/project/relayd.git >>>> >>>> Since you can cover the same use cases with user space code, I don't >>>> think it's a good idea to put bridge emulation hacks in the kernel's >>>> wireless stack. >>> >>> What about performance? Quick test show that is slow ~ 100-120 mbps >>> (UDP tests) and procesor is overloaded. Am I missing something? I >>> would expect it to be greater. (4 x 4 antena setup VHT80) >> What platform are you testing it on, and what kind of UDP test are you >> running? > > My setup it is as follows: > [laptop1] --eth-- [EA6500 AP] ~~rf~~ [AP148 STA w/ QCA99X0] --eth-- > [laptop2]. I run UDP traffic between Laptop1 and Laptop2 using iperf. > Laptop1: iperf -s -i1 -u > Laptop2: iperf -i1 -c 192.168.1.108 -b 200M -t100 -u -P5 (without -P5, > I got similar results) So what kind of kernel are you running? Maybe you should use perf to figure out why routing is so slow. - Felix