From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Ahern Subject: Re: switchdev fib offload issues Date: Mon, 18 Apr 2016 10:59:37 -0600 Message-ID: <57151279.4020806@cumulusnetworks.com> References: <20160418154757.GA2059@nanopsycho.amit.cz> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Cc: davem@davemloft.net, idosch@mellanox.com, eladr@mellanox.com, yotamg@mellanox.com, ogerlitz@mellanox.com, roopa@cumulusnetworks.com, nikolay@cumulusnetworks.com, jhs@mojatatu.com, john.fastabend@gmail.com, rami.rosen@intel.com, gospo@cumulusnetworks.com, stephen@networkplumber.org, sfeldma@gmail.com, f.fainelli@gmail.com, andrew@lunn.ch, vivien.didelot@savoirfairelinux.com, tgraf@suug.ch, aduyck@mirantis.com To: Jiri Pirko , netdev@vger.kernel.org Return-path: Received: from mail-io0-f175.google.com ([209.85.223.175]:35781 "EHLO mail-io0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752535AbcDRQ7l (ORCPT ); Mon, 18 Apr 2016 12:59:41 -0400 Received: by mail-io0-f175.google.com with SMTP id g185so200778308ioa.2 for ; Mon, 18 Apr 2016 09:59:41 -0700 (PDT) In-Reply-To: <20160418154757.GA2059@nanopsycho.amit.cz> Sender: netdev-owner@vger.kernel.org List-ID: On 4/18/16 9:47 AM, Jiri Pirko wrote: > Proposed solutions (ideas): > 1) per-netns. Add a procfs file: > /proc/sys/net/ipv4/route/fib_offload_error_policy > with values: "evict" - default, current behaviour > "fail" - propagate offload error to user > The policy value would be stored in struct net. > > 2) per-VRF/table > When user creates a VRF master, he specifies a table ID > this VRF is going to use. I propose to extend this so > he can pass a policy ("evict"/"fail"). > The policy value would be stored in struct fib_table or > struct fib6_table. The problem is that vfr only saves > table ID, allocates dst but does not actually create > table. That might be created later. But I think this > could be resolved. Yes, we have a local patch where I do create the table for IPv6. Can do that for IPv4 as well. Some other clean ups are needed in this area - like the ability to delete a table > > 3) per-VFR/master_netdev > In this case, the policy would be also set during > the creation of VFR master. From user perspective, > this looks same as 2) > The policy value would be stored in struct net_vrf (vrf private). The VRF device is really only used for guiding lookups, not inserting routes. A per table/VRF policy (option 2) seems more appropriate.