From mboxrd@z Thu Jan 1 00:00:00 1970 From: Brian Haley Subject: Re: [PATCH] bonding: send IPv6 neighbor advertisement on failover Date: Wed, 08 Oct 2008 15:01:58 -0400 Message-ID: <48ED03A6.2060705@hp.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Alex Sidorenko , David Miller , fubar@linux.vnet.ibm.com, Simon Horman , "netdev@vger.kernel.org" , netdev-owner@vger.kernel.org, Vladislav Yasevich To: David Stevens Return-path: Received: from g5t0007.atlanta.hp.com ([15.192.0.44]:7513 "EHLO g5t0007.atlanta.hp.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753170AbYJHTCb (ORCPT ); Wed, 8 Oct 2008 15:02:31 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: David Stevens wrote: > Well, actually, it looks like I'm suggesting you to re-use something that > doesn't > exist. :-) > > MLD (and IGMP) has such a thing where unsolicited advertisements are sent > multiple times, with delays in between, to account for lossy networks > possibly > dropping the first one. There are configurable counts associated with > probes > and retransmit intervals for solicits, but I don't see the equivalent yet > for > unsolicited NA's. I don't see an equivalent either, since the only unsolicited NA the kernel sends is for DAD, which uses dad_transmits. I left the MLD changes out of this patch so I could work on it separately, when I get to it I'll make sure to look at the issues you raised in your other email so it follows the RFC, or at least the Linux behavior. > So, instead, what I suggest is that you add (or find!) THAT knob, instead > of a > bonding-specific one. Because adding an address that wasn't there before > has identical issues with unsolicited NA's as bonding has with activating > a > new address. The default should probably be 1, but if you ever need to > send multiple unsolicited NA's for bonding, you probably also need it for > adding a normal address on the same network. dad_transmits is similar, > but not really the same thing. The problem is that dad_transmits can be set to zero, although not recommended, so if we used that value then bonding failover would be just as broken. I think having this new tunable stay in the bonding code is useful since that's the code that's actually doing the transmit. -Brian