Re: [PATCHv3 net] ipv6: do not match device when remove source route

[Ido Schimmel <idosch@idosch.org>]

On Sun, Jul 23, 2023 at 12:12:00PM -0600, David Ahern wrote:
> On 7/23/23 2:13 AM, Ido Schimmel wrote:
> > 
> > I don't know, but when I checked the code and tested it I noticed that
> > the kernel doesn't care on which interface the address is configured.
> > Therefore, in order for deletion to be consistent with addition and with
> > IPv4, the preferred source address shouldn't be removed from routes in
> > the VRF table as long as the address is configured on one of the
> > interfaces in the VRF.
> > 
> 
> Deleting routes associated with device 2 when an address is deleted from
> device 1 is going to introduce as many problems as it solves. The VRF
> use case is one example.

This already happens in IPv4:

# ip link add name dummy1 up type dummy
# ip link add name dummy2 up type dummy
# ip address add 192.0.2.1/24 dev dummy1
# ip route add 198.51.100.0/24 dev dummy2 src 192.0.2.1
# ip -4 r s
192.0.2.0/24 dev dummy1 proto kernel scope link src 192.0.2.1 
198.51.100.0/24 dev dummy2 scope link src 192.0.2.1 
# ip address del 192.0.2.1/24 dev dummy1
# ip -4 r s

IPv6 only removes the preferred source address from routes, but doesn't
delete them. The patch doesn't change that.

Another difference from IPv4 is that IPv6 only removes the preferred
source address from routes whose first nexthop device matches the device
from which the address was removed:

# ip link add name dummy1 up type dummy
# ip link add name dummy2 up type dummy
# ip address add 2001:db8:1::1/64 dev dummy1
# ip route add 2001:db8:2::/64 dev dummy2 src 2001:db8:1::1
# ip -6 r s
2001:db8:1::/64 dev dummy1 proto kernel metric 256 pref medium
2001:db8:2::/64 dev dummy2 src 2001:db8:1::1 metric 1024 pref medium
fe80::/64 dev dummy1 proto kernel metric 256 pref medium
fe80::/64 dev dummy2 proto kernel metric 256 pref medium
# ip address del 2001:db8:1::1/64 dev dummy1
# ip -6 r s
2001:db8:2::/64 dev dummy2 src 2001:db8:1::1 metric 1024 pref medium
fe80::/64 dev dummy1 proto kernel metric 256 pref medium
fe80::/64 dev dummy2 proto kernel metric 256 pref medium

And this is despite the fact that the kernel only allowed the route to
be programmed because the preferred source address was present on
another interface in the same L3 domain / VRF:

# ip link add name dummy1 up type dummy
# ip link add name dummy2 up type dummy
# ip route add 2001:db8:2::/64 dev dummy2 src 2001:db8:1::1
Error: Invalid source address.

The intent of the patch (at least with the changes I proposed) is to
remove the preferred source address from routes in a VRF when the
address is no longer configured on any interface in the VRF.

Note that the above is true for addresses with a global scope. The
removal of a link-local address from a device should not affect other
devices. This restriction also applies when a route is added:

# ip link add name dummy1 up type dummy
# ip link add name dummy2 up type dummy
# ip -6 address add fe80::1/64 dev dummy1
# ip -6 route add 2001:db8:2::/64 dev dummy2 src fe80::1
Error: Invalid source address.
# ip -6 address add fe80::1/64 dev dummy2
# ip -6 route add 2001:db8:2::/64 dev dummy2 src fe80::1