From owner-p4-projects@FreeBSD.ORG Wed Aug 12 22:35:13 2009 Return-Path: Delivered-To: p4-projects@freebsd.org Received: by hub.freebsd.org (Postfix, from userid 32767) id 944671065672; Wed, 12 Aug 2009 22:35:12 +0000 (UTC) Delivered-To: perforce@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 52949106566B; Wed, 12 Aug 2009 22:35:12 +0000 (UTC) (envelope-from zec@freebsd.org) Received: from labs3.cc.fer.hr (labs3.cc.fer.hr [161.53.72.21]) by mx1.freebsd.org (Postfix) with ESMTP id C236E8FC52; Wed, 12 Aug 2009 22:35:11 +0000 (UTC) Received: from sluga.fer.hr (sluga.cc.fer.hr [161.53.72.14]) by labs3.cc.fer.hr (8.13.8+Sun/8.12.10) with ESMTP id n7CMZ5al013451; Thu, 13 Aug 2009 00:35:10 +0200 (CEST) Received: from localhost ([161.53.19.8]) by sluga.fer.hr with Microsoft SMTPSVC(6.0.3790.3959); Thu, 13 Aug 2009 00:35:04 +0200 From: Marko Zec To: Julian Elischer Date: Thu, 13 Aug 2009 00:34:57 +0200 User-Agent: KMail/1.9.10 References: <200908122108.n7CL8uhJ058398@repoman.freebsd.org> <4A833B16.2040301@elischer.org> In-Reply-To: <4A833B16.2040301@elischer.org> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200908130034.57133.zec@freebsd.org> X-OriginalArrivalTime: 12 Aug 2009 22:35:05.0068 (UTC) FILETIME=[235E52C0:01CA1B9D] Cc: Perforce Change Reviews , Robert Watson Subject: Re: PERFORCE change 167260 for review X-BeenThere: p4-projects@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: p4 projects tree changes List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 12 Aug 2009 22:35:13 -0000 On Wednesday 12 August 2009 23:58:46 Julian Elischer wrote: > Marko Zec wrote: ... > > @@ -710,22 +715,36 @@ > > .pr_input = div_input, > > .pr_ctlinput = div_ctlinput, > > .pr_ctloutput = ip_ctloutput, > > - .pr_init = NULL, > > + .pr_init = div_init, > > .pr_usrreqs = &div_usrreqs > > If you are going to make pr_init() called for every vnet then > pr_destroy should be as well. But in fact that is not really safe. > (either of them) > > The trouble is that we can not guarantee that other protocols can > handle being called multiple times in their init and destroy methods. > Especially 3rd party protocols. > > We need to ensure only protocols that have been converted to run > with multiple vnets are ever called with multiple vnets. > > for this reason the only safe way to do this is via the VNET_SYSINIT > and VNET_SYSUNINIT calls. That would mean you would have to convert most if not all of the existing things that hang off of protosw-s in netinet, netinet6 etc. to use VNET_SYSINT / VNET_SYSUNIT instead of protosw->pr_init(). So the short answer is no. I cannot recall that we ever discussed or planned to be able to mix virtualized with non-virtualized protocols in the same kernel. That would be a horrible mess, and I cannot even imagine having say a multi-instance INET with a single-instance INET6 kernel, shared among all the vnets. To start with, how would you decide that you're not allowed to process an IPv6 packet received on the wire in a non-default vnet in such an environment? Do we have the infrastructure in place necessary for preventing doing say a ifconfig lo0 ::1 in a non-default vnet in such an hypotetical setup? The answer is no. VNET_SYSINIT is nice, but proper special-casing changes required to support single-instance protocols to work only with vnet0 and not with the other protocols are simply not there, and I hope will never be, because I fear they would be highly intrusive, difficult to verify and maintain, and probably also have an impact on performance. A proper solution for the issue you are raising could be something that would prevent modules assuming our stack is compiled as single-instance to be kldloaded if the kernel was actually built with multi-instance stack support. I think Robert (cc-ed) had some ideas on how to accomplish this by having such modules depend on a magic global variable (say __no_vnet_support) to be available. All the current "base" protocols are already using pr_init() in multi-instance mode in options VIMAGE case. So I see no reason for ip_divert not being allowed to leverage on the same mechanism. Re. pr_destroy(), you're right, patch already submitted to p4... Marko