From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dan Williams Subject: Re: [patch net-next V2] bond: have random dev address by default instead of zeroes Date: Tue, 29 Jan 2013 11:54:22 -0600 Message-ID: <1359482062.1635.17.camel@dcbw.foobar.com> References: <542237717.4200599.1359136587903.JavaMail.root@redhat.com> <22106.1359138692@death.nxdomain> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: Pavel Simerda , netdev@vger.kernel.org, davem@davemloft.net, andy@greyhouse.net, stephen@networkplumber.org, Jiri Pirko To: Jay Vosburgh Return-path: Received: from mx1.redhat.com ([209.132.183.28]:16866 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754398Ab3A2R4L (ORCPT ); Tue, 29 Jan 2013 12:56:11 -0500 In-Reply-To: <22106.1359138692@death.nxdomain> Sender: netdev-owner@vger.kernel.org List-ID: On Fri, 2013-01-25 at 10:31 -0800, Jay Vosburgh wrote: > Pavel Simerda wrote: > > >----- Original Message ----- > >> From: "Jay Vosburgh" > >> but I don't think it should be changed. > > > >Just a short question. Is there any reason for bonding interfaces to > >behave differently from bridging interfaces in this respect? > > To clarify, what I don't think should change is that a manually > set MAC on the bonding master should override the automatic copy of the > first slave's MAC to the bonding master. The fail_over_mac active and > follow settings are an exception to this, but those are special cases > for unusual network hardware. > > As for the random MAC vs. zero MAC, I've always thought that the > all zero MAC was a clear indicator that the device (the bonding master > in this case) was not in a usable state (in the sense that it could not > send or receive actual traffic). It's not a really big deal, though, so > if the trend these days is for everything to have a MAC all the time, > that's fine, as long as doing so doesn't break anything. Isn't that exactly what carrier means? If the carrier bit is off, nothing should expect that traffic can pass. The bond will only set its carrier ON if at least one slave exists and at least one slave has a carrier that's ON. We have overrides for buggy driver carrier checking already. Zero-MAC is somewhat redundant here as a mechanism for detecting that the bond is usable or not. Dan > I think the patch under discussion should be fine with the > addition of the last notifier call previously discussed. Some > documentation updates would be nice, too. > > -J > > --- > -Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com >