Bug 1834454

+++ This bug was initially created as a clone of Bug #1829448 +++

According to Harald's email to dnsmasq ML http://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2020q2/014038.html there is regression in latest dnsmasq version dnsmasq-2.79-11.el8

Details from Harald:

We discovered an issue introduced in this commit:
137286e9baecf6a3ba97722ef1b49c851b531810

Prior to this commit one could have two dhcp-host entries, one for IPv4
and another for IPv6, for example:

 dhcp-host=52:54:00:bc:c3:fd,172.20.0.11,host2
 dhcp-host=52:54:00:bc:c3:fd,[fd12:3456:789a:1::aadd],host2

This no longer works. In the above example dhcpv6 client succedes, but
the dhcpv4 client get 'no address available'. Swapping the order of the
two entries in the config file allow the dhcpv4 client to succeed, but
then the dhcpv5 client fails.

Alternative configurations that do work in 2.81:

 dhcp-host=52:54:00:bc:c3:fd,172.20.0.11,host2
 dhcp-host=tag:dhcpv6,52:54:00:bc:c3:fd,[fd12:3456:789a:1::aadd],host2

or: 

 dhcp-host=52:54:00:bc:c3:fd,172.20.0.11,[fd12:3456:789a:1::aadd],host2


I'm not sure using two dhcp-host entries was ever intended to work, as
the manual page states that both ipv4 and ipv6 address can be defined
in a single entry.

A first tought for possible fix would be to internally set the
'tag:dhcpv6' for any dhcp-host entry with only ipv6 addresse(s).

It impacts e.g. OpenStack Neutron, see https://bugzilla.redhat.com/show_bug.cgi?id=1829201 for details.

--- Additional comment from Petr Menšík on 2020-05-11 12:25:00 CEST ---

Patch posted to upstream list by Harald[1].

1. http://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2020q2/014051.html