Bug#863389: [Fwd: Re: Bug#863389: linux-image-4.9.0-0.bpo.3-amd64: After update to linux-image-4.9.0.0-0.bpo.3, no route to host via IPV4]

[Ben Hutchings <ben@decadent.org.uk>]

-------- Forwarded Message --------
From: Jason Wittlin-Cohen <jwittlincohen@gmail.com>
To: Ben Hutchings <ben@decadent.org.uk>
Subject: Re: Bug#863389: linux-image-4.9.0-0.bpo.3-amd64: After update to linux-image-4.9.0.0-0.bpo.3, no route to host via IPV4
Date: Fri, 26 May 2017 20:23:23 -0400
Message-id: <CADy0EvL4=pxFvbre3d=Q9k4BAigznCVVmvUXr9oe5S5xNZFfgw@mail.gmail.com>

Hi Ben,

Thanks for the reply. You're absolutely right about the ordering of the
iface statements.  I modified my interfaces file such that the bonding
parameters followed "iface bond0 inet static" and now it loads the bonding
module and adds a default gateway for IPV4.  The configuration below works
properly.  I'm pretty sure I know what happened.  I've been using bonding
with IPV4-only for quite a while; I used the Debian howto to set it up.  I
recently added a Hurricane Electric 6in4 tunnel to my pfsense router and
therefore added the "iface bond0 inet6 static" section.  Restarting the
bond using "ifdown bond0 && ifup bond0" worked, perhaps because the bonding
module was already loaded.  I wasn't able to find much documentation about
how to setup a bond with both IPV4 and IPV6.  The bonding guide is IPV4
only, and the IPV6 guide doesn't discuss bonding.

However, I still think there's something wrong with my configuration. If I
attempt to use "iface bond0 inet6 dhcp", I get an IP from the DHCPv6
server, but no default gateway is created, so attempting to access the
internet results in no route to host.  I know Router Advertisements and
DHCPv6 are working as other Windows and Linux clients are working (tested
with both SLAAC and DHCPv6 w/RA).  Do you know why I'm not getting a
default route when using DHCPv6?


auto lo
iface lo inet loopback
iface eth0 inet manual
iface eth1 inet manual

auto bond0
iface bond0 inet dhcp
   slaves eth0 eth1
   bond_mode 802.3ad
   bond_miimon 100
   bond_downdelay 200
   bond_updelay 200
   mtu 9000

iface bond0 inet6 static
   address 2001:470:8:1141::2
   netmask 64
   gateway 2001:470:8:1141::1

auto eth2
iface eth2 inet static
address 10.0.0.1
netmask 255.255.255.0
mtu 9000

Thanks,

Jason



On Fri, May 26, 2017 at 7:30 PM, Ben Hutchings <ben@decadent.org.uk> wrote:

> Control: tag -1 moreinfo
> 
> I don't see any changes in the kernel bonding driver, or any relevant
> changes to kernel networking.
> 
> On Fri, 2017-05-26 at 00:31 -0400, Jason Cohen wrote:
> > ** Network interface configuration:
> 
> [...]
> > auto bond0
> > iface bond0 inet static
> > address 192.168.1.200
> > netmask 255.255.255.0
> > network 192.168.1.0
> > gateway 192.168.1.1
> > iface bond0 inet6 static
> >       address 2001:470:8:1141::2
> >       netmask 64
> >       gateway 2001:470:8:1141::1
> > slaves eth0 eth1
> > bond_mode 802.3ad
> > bond_miimon 100
> > bond_downdelay 200
> > bond_updelay 200
> > mtu 9000
> 
> [...]
> 
> I'm pretty sure this configuration never worked, and that you just
> found that out by rebooting after the upgrade.
> 
> The problem is that ifupdown treats each 'iface' statement as
> introducing a separate interface configuration, so the above is parsed
> as:
> 
> iface bond0 inet static
>         address 192.168.1.200
>         netmask 255.255.255.0
>         network 192.168.1.0
>         gateway 192.168.1.1
> 
> iface bond0 inet6 static
>         address 2001:470:8:1141::2
>         netma
> sk 64
>         gateway 2001:470:8:1141::1
>         slaves eth0 eth1
> 
> bond_mode 802.3ad
>         bond_miimon 100
>         bond_downdelay 200
> 
> bond_updelay 200
>         mtu 9000
> 
> I think it will first try to apply the first interface configuration.
> There are no bonding parameters so the code to create a bonding
> interface (and load the bonding module) doesn't run.  As there is no
> such interface, the IPv4 configuration can't be applied either.
> 
> I don't know whether ifupdown tries to process the second interface
> configuration after this failure, but I would guess not.
> 
> If you move all the bonding and mtu parameters into the first interface
> configuration, does the bonding interface start working again?
> 
> Ben.
> 
> --
> Ben Hutchings
> The generation of random numbers is too important to be left to chance.
>                                                        - Robert Coveyou
>