secondary IPV6 address add gateway and interface problem in VPS ON / openNebula.

Solved1.45K viewsVirtual servers#address #ipv6 #vpson

After several manual attempts directly in the shell on the vps I decided to ask for support via email. A nice and friendly technician gave me the following link

i followed the guide and i finally managed to add additional ipv6 to the ::/56 subnet assigned in the opennebula panel. As you can see from the following images I have attached and added a total of 5 ipv6.

After updating the changes and restarting the server, the situation is as follows. Given that it is absurd from my point of view that openNebula raises 5 interfaces eth1 for ::1, eth2 for ::2, eth3 ::3 .. in the config and assigning a single ipv6 address for each interface being precisely from the same subnet and not there is an attempt to delegate to a user outside the network. Having said that, I rebooted the server and was expecting that /etc/network/interfaces had already taken the configuration but to my surprise it didn’t.

So I tried to add static addresses at startup without specifying the gateway for all addresses but only for the starting ::1 (see example). Result ::1 responds to pings and all others ::2 ::3 ::4 ::5 do not respond. I then tried to add the gateway for each ip with the result that ::5 replied and ::4 ::3 ::2 ::1 no longer replied.

1) vim /etc/network/interfaces

auto eth1
iface eth1 inet6 static
  address 2a02:x:x:x::1
  netmask 56
  autoconf 0
  accept_ra 0
  gateway 2a02:x:x:x::1 # opennebula specified in the panel
 auto eth2
iface eth2 inet6 static
  address 2a02:x:x:x::2
  netmask 56
# gateway 2a02:x:x:x::1 (test add ipv6 without gateway)
continue up to 5

At this point I went to add by hand following information from the main opennebula panel and then to check the routes on the vps and I realized that the situation is the following.

~$ ip -6 r s
2a02:x:x:x::/56 dev eth1 proto kernel metric 256 pref medium
2a02:x:x:x::/56 dev eth2 proto kernel metric 256 pref medium
fe80::/64 dev eth0 proto kernel metric 256 pref medium
fe80::/64 dev eth1 proto kernel metric 256 pref medium
fe80::/64 dev eth2 proto kernel metric 256 pref medium
default via 2a02:x:x:x::1 dev eth1 metric 1024 onlink pref medium
~$ route -6
Kernel IPv6 routing table
Destination                    Next Hop                   Flag Met Ref Use If
2a02:x:x:x::/56           [::]                       U    256 2     0 eth1
2a02:x:x:x::/56           [::]                       U    256 1     0 eth2
fe80::/64                      [::]                       U    256 2     0 eth0
fe80::/64                      [::]                       U    256 2     0 eth1
fe80::/64                      [::]                       U    256 1     0 eth2
[::]/0                         2a02:2b88:2:500::1         UGH  1024 2     0 eth1
ip6-localhost/128              [::]                       Un   0   4     0 lo
address/128               [::]                       Un   0   4     0 eth1
2a02:x:x:x::2/128         [::]                       Un   0   2     0 eth2
fe80::59ff:fedd:d89e/128       [::]                       Un   0   4     0 eth0
fe80::e8ff:fe41:9a4/128        [::]                       Un   0   4     0 eth1
fe80::e8ff:fe41:9a5/128        [::]                       Un   0   2     0 eth2
ff00::/8                       [::]                       U    256 3     0 eth0
ff00::/8                       [::]                       U    256 2     0 eth1
ff00::/8                       [::]                       U    256 2     0 eth2
[::]/0                         [::]                       !n   -1  1     0 lo


I realized then and I tried add manually add the gateway for eth2, eth3 … obviously with result:

~$ sudo ip  -6 r a default via 2a02:x:x:x::1 dev eth2 
RTNETLINK answers: File exists

at this point I wonder how it is right to do. Theoretically I should remove the gateway which is only assigned to eth1 and globally assign it to everything. Correct?


~$ sudo ip -6 r d default via 2a02:x:x:x::1 dev eth1

and add new global gateway without dev

~$ sudo ip -6 r a default via 2a02:x:x:x::1

I have a feeling it won’t work. Did you have the same problem? do you have any ideas about it?

Thanks for your help.

Role: Customer
Question is closed for new answers.
MB313456 Selected answer as best 26. 11. 2021

For the record, I sorted it out, dusting off my old routing notes.

I keep thinking and arguing that the address management of an ipv6 subnet as it is now handled by your opennebula system is not correct and wedos admin would have the duty to correct this behavior by reporting directly to opennebula project via github.

I am referring to the latest RFC articles on ipv6 routing of a subnet. At present with leases he creates an eth * interface for each single ipv6 address to which he obviously associates a different mac address which is unnecessary when you are in the same network without external delegation. Usually the most obvious would be to have an example class 2001:a:b:c::/56 on eth*. To add other addresses, a simple ip -6 a a 2001:a:b:c::2 dev eth* is sufficient! it is even superfluous to indicate the type of address ::/64 ::/80 ::/128.

Having said that I describe without going into detail the solution for those who have the same problem. You have to act with iproute2 and create the default routes on single tables. In order:
1) edit /etc/iproute2/rt_table
2) raise the network interfaces related to the additional ipv6
3) add the ipv6 delegate for each interface
4) set the default route for each interface referred to the previously created table.
5) set rule add from :: yourIPV6 table $YourNameTable

Solution two “TO TEST”: opennebula has the option to attach nic as an alias theoretically this should give the possibility to raise the ipv6 starting from eth1 on eth1:1 eth1:2 and so on thus exploiting the same general gateway of eth1 for all aliases. For obvious reasons I have not tried. As for me, the discussion can be closed here. Best Regards.

Role: Customer
MB313456 Selected answer as best 26. 11. 2021
You are viewing 1 out of 2 answers, click here to view all answers.