Date: Sat, 24 Feb 2018 07:15:27 -0800 (PST) From: "Rodney W. Grimes" <freebsd-rwg@pdx.rh.CN85.dnsmgr.net> To: Eugene Grosbein <eugen@grosbein.net> Cc: Andreas Longwitz <longwitz@incore.de>, freebsd-net@freebsd.org Subject: Re: loopback routes pinned by commit r326012 for FreeBSD 10 cannot be deleted Message-ID: <201802241515.w1OFFRKw001282@pdx.rh.CN85.dnsmgr.net> In-Reply-To: <5A9162C9.5050206@grosbein.net>
next in thread | previous in thread | raw e-mail | index | archive | help
> 24.02.2018 19:37, Andreas Longwitz wrote: > > >> Please test the follwoind patch for the kernel: > >> > >> --- sys/net/if.c.orig 2018-02-24 08:00:34.578670000 +0700 > >> +++ sys/net/if.c 2018-02-24 11:07:57.014318000 +0700 > >> @@ -1699,7 +1699,7 @@ ifa_del_loopback_route(struct ifaddr *if > >> null_sdl.sdl_type = ifa->ifa_ifp->if_type; > >> null_sdl.sdl_index = ifa->ifa_ifp->if_index; > >> bzero(&info, sizeof(info)); > >> - info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC; > >> + info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC | RTF_PINNED; > >> info.rti_info[RTAX_DST] = ia; > >> info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl; > >> error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib); > >> > > > > This patch patch solves the problem for "ifconfig fxp0 delete": now the > > kernel removes the loopback route, tested on FreeBSD 10 r328260. > > Glad to know that. > > > But deleting the loopback route manually does not work anymore, is this > > intended behaviour ? > > > > I would like to control the routing table as much as possible. > > This is side effect of loopback routes being intentionally protected with RTF_PINNED flag > in recent FreeBSD versions so that link routes can override ones installed by routing daemons. > > Perhaps, ifconfig(8) utility should be extended to use RTF_PINNED to be able to modify such routes. Perhaps it is a rather serious and undesired side effect to have such routes without simple user ability to control them. I am going to write a patch that wraps whis whole loopback_route management in a kernel compile time option, which well leaves things as they are now, but allow a user to remove this non-sensical route policy code from there kernel. It shall also have a sysctl that can turn these routes off so that a GENERIC kernel can be used without this non-sense kernel routing policy. The default of this sysctl shall leave things as they are now. -- Rod Grimes rgrimes@freebsd.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201802241515.w1OFFRKw001282>