From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1760402Ab3EBQXK (ORCPT ); Thu, 2 May 2013 12:23:10 -0400 Received: from mail-ia0-f178.google.com ([209.85.210.178]:45658 "EHLO mail-ia0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758743Ab3EBQXI convert rfc822-to-8bit (ORCPT ); Thu, 2 May 2013 12:23:08 -0400 MIME-Version: 1.0 In-Reply-To: <87zjwda9bl.fsf@nemi.mork.no> References: <20130502.041625.412316202038784117.davem@davemloft.net> <20130502083619.GA22010@macbook.localnet> <87d2t9bvj1.fsf@nemi.mork.no> <20130502.051727.949429506738234701.davem@davemloft.net> <1367490491.5106.137.camel@deadeye.wl.decadent.org.uk> <87zjwda9bl.fsf@nemi.mork.no> From: =?ISO-8859-2?Q?Micha=B3_Miros=B3aw?= Date: Thu, 2 May 2013 18:22:47 +0200 Message-ID: Subject: Re: [GIT] Networking To: =?ISO-8859-1?Q?Bj=F8rn_Mork?= , =?UTF-8?Q?Pavel_=C5=A0imerda?= Cc: Ben Hutchings , David Miller , kaber@trash.net, torvalds@linux-foundation.org, hayeswang@realtek.com, akpm@linux-foundation.org, netdev@vger.kernel.org, linux-kernel@vger.kernel.org Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 2013/5/2 Bjørn Mork : > Ben Hutchings writes: >> On Thu, 2013-05-02 at 05:17 -0400, David Miller wrote: >>> From: Bjørn Mork >>> Date: Thu, 02 May 2013 11:06:42 +0200 >>> >>> > Adding the new netdev features will make it go from 1 to 2: >>> >>> We already had more than 31 feature bits before Patrick's >>> changes, and I'm pretty sure this was the case when we added >>> that ethtool API. >> >> It wasn't, but this should be OK. Userland is supposed to query the >> number of features using ETHTOOL_GSSET_INFO and then work out the number >> of words/blocks using FEATURE_BITS_TO_BLOCKS(). > > > Looking at > http://cgit.freedesktop.org/NetworkManager/NetworkManager/tree/src/platform/nm-linux-platform.c#n1025 > there seems to be a couple of bugs in this area. This is certainly > abusing the exported API, but it does mean that NM breaks if you ever > move NETIF_F_VLAN_CHALLENGED (like the 802.1ad patch did): > > ---- > #define NETIF_F_VLAN_CHALLENGED (1 << 10) > > static gboolean > link_supports_vlans (NMPlatform *platform, int ifindex) > { > auto_nl_object struct rtnl_link *rtnllink = link_get (platform, ifindex); > const char *name = nm_platform_link_get_name (ifindex); > struct { > struct ethtool_gfeatures features; > struct ethtool_get_features_block features_block; > } edata = { .features = { .cmd = ETHTOOL_GFEATURES, .size = 1 } }; > > /* Only ARPHRD_ETHER links can possibly support VLANs. */ > if (!rtnllink || rtnl_link_get_arptype (rtnllink) != ARPHRD_ETHER) > return FALSE; > > if (!name || !ethtool_get (name, &edata)) > return FALSE; > > return !(edata.features.features[0].active & NETIF_F_VLAN_CHALLENGED); > } > ---- > > > Not that I see how this particular bug matters unless you need VLAN > support in NM. But there could be similar issues around. I guess > avoiding unnecessary renumbering of the NETIF_F bits can save us some > trouble. Although you can certainly argue that those bits never were > intended to be part of the API, and that using them like this is a user > application bug. This is certainly a bug in NM, and a fresh one: commit b636ea86b1c0a28b77eda311c84d3b2417cad22e from 2013-04-10 14:40:58 (GMT). Userspace is expected to use ETHTOOL_GSTRINGS for ETH_SS_FEATURES and find a corresponding bit position by feature name ("vlan-challenged" in this case). Cc: commit's author. Best Regards, Michał Mirosław