Re: misc/60870: Downing a interface in the same subnet as the nfs-root freezes the system

["Robert Elz via gnats" <gnats-admin%NetBSD.org@localhost>]

The following reply was made to PR misc/60870; it has been noted by GNATS.

From: Robert Elz <kre%munnari.OZ.AU@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: 
Subject: Re: misc/60870: Downing a interface in the same subnet as the nfs-root freezes the system
Date: Tue, 13 Oct 2026 21:48:36 +0700

     Date:        Tue, 13 Oct 2026 12:10:01 +0000 (UTC)
     From:        "netbsd%lissners.de@localhost via gnats" <gnats-admin%NetBSD.org@localhost>
     Message-ID:  <20261013121001.22F5C1A923E%mollari.NetBSD.org@localhost>
 
   | Expected behaviour:
   | The ip-address is automatically removed from wm1 before
   | downing the interface.
 
 While that might make your problem go away, it isn't the right
 thing to do - interfaces are sometimes taken down & up again in
 order to do some sort of re-init on the hardware as the interface
 comes back up.  The configured IP address is intended to remain,
 it won't work when the network is down, but will again as soon as
 it comes back up.
 
 If you want to delete the address first, you can just do that.
 
   | My(!) expectation is not: A switchover to wm1 as the interface
   | handling the nfs-root when downing wm0. Would be nice, though :-)
 
 If routing were set up correctly, that would work.   In fact, with the
 dubious (really incorrect) config that you have, it is already working.
 That was the source of the problem.
 
 Your default route, and the route to the nfs server started out via your
 wm0 interface (and are shown that way).   When you (mis-)configured wm1
 you added an interface reaching the same destination, with a longer
 netmask (a more specific route) - the kernel would have switched to that
 one for all destinations reachable via that path - ie: everything on
 192.168.10/24 - that's why you have ARP entries:
 
 192.168.10.2       3c:a6:2f:3e:25:35  UHL         -        -      -  wm1
 192.168.10.6       08:00:27:b3:81:f0  UHL         -        -      -  wm1
 192.168.10.2       3c:a6:2f:3e:25:35  UHL         -        -      -  wm0
 192.168.10.6       08:00:27:b3:81:f0  UHL         -        -      -  wm0
 
 on both interfaces to the same destination, the wm0 ones were the
 originals, then the system switched to using wm1 when that one was
 configured.
 
 When wm1 goes down, without anything removing that routing (which
 removing the address would do), the system lacks a working path to
 the destination, and as that's your NFS server for / everything is
 just going to hang until wm1 comes back again - but without the
 ability to run anything any more that would be a very long time!
 
 But you have the solution already:
 
   | What also works is giving the interface a ip in another network:
   | >>>
   | ifconfig wm1 192.168.100.97
   | ifconfig wm1 down
   | <<<
 
 In that case, wm1 is not providing access to 192.168.10/24 addresses,
 so packets to those addresses will not be sent that way.   You could
 "fix" that with additional routing.
 
 This is the way that IP was designed to work - every network interface
 gets a distinct, and unique, network prefix, just for it, and for nothing
 else.   These days, in the IPv4 world, that's limited to being within your
 local NAT free area, as there just aren't enough unique prefixes for
 everyone, but there are plenty for you to use on your LAN that you don't
 need to cheat the addressing the way you are attempting.
 
 kre