IPv6 support in XCP-ng for the management interface - feedback wanted

BenjiReis

@AtaxyaNetwork so infact the issue was with our IPv6 lab config (lol) so XOA is reachable in fact with an IPv6 address ahah.
So you can play with it.

Now i'm back on my DHCPv6/Autoconf/SLAAC investigations

BenjiReis

Hi all!

8.2.1 IPv6 ISO available!

Here's a new ISO for IPv6 based on XCP-ng 8.2.1!
The ISO can be used to upgrade an existing server installed with the previous IPv6 test ISO or install a brand new XCP-ng 8.2.1 with IPv6 support on management interface.

A non-IPv6 8.2.0 would remain non-IPv6 after an upgrade as it's not possible to edit the management interface's primary adress type.

An 8.2.0 IPv6 hosts can also be upgraded via yum: yum upgrade --enablerepo=xcp-ng-updates,xcp-ng-ipv6.

What's new

All 8.2.1 fixes
Better DNS management in the case of both IPv4 and IPv6 configured on a PIF
Partial support of IPv6 DHCP and autoconf

What to test

Your daily uses of XCP-ng but with IPv6
DHCP and autoconf (I have reached the limits of my knowledge so help from the community with more IPv6 expertise would be very VERY VERY helpful! :D)
The goal of this ISO release is mainly to get help and leads about what's missing in DHCP and Autoconf.

Any issue encountered (and what works fine also) can be reported in this thread.

Usual warning

This a test ISO with an experimental feature still in development.
IPv6 on management interface is not officially supported by XCP-ng yet and so, we do not recommend to use it for a production environment.

Thanks a lot for the help and I hope the ISO will work well for everyone.

AtaxyaNetwork

@BenjiReis Hello !
Thank you for the ISO !

I just tested the install in a VM (for the moment, soon I will have a physical machine available)

First review:
So far, the autoconf seem to be working !
But, during the installation, I provide a IPv6 DNS (the Cloudflare one, 2606:4700:4700::1111), but DNS is not working, as I have 1.1.1.1 in my /etc/resolv.conf
I don't know if is the autoconf who is pushing the 1.1.1.1 (i need to check my router first), but I think is better if when we give a DNS, it bypass the autoconf

More test is coming the next few day

(sorry if my English is a bit bad)

Thank you and the team for all you work !

BenjiReis

@AtaxyaNetwork thanks for the report.

I reproduced the issue, for some reason at first boot XCP-ng launch an IPv4 dhclient request (even though IPv4 is not configured on the management interface...) which overrides the DNS set after the request is replied to.

mbunkus

@BenjiReis I've just started giving the IPv6-enabled 8.2.1 a try. Right within the first hour I've stumbled across the following two issues on an IPv6-only server:

Repository mirrors

The preconfigured repositories use mirrors.xcp-ng.org. That one returns the address of an actual mirror. And if that mirror happens not to have an IPv6 address, doing anything (e.g. yum makecache) fails.

Re-running it might return on with AAAA records; then it does work — or maybe it'll be another AAAA-less mirror.

NFS mounting via host name

NFS mounting doesn't work if I use a host name that has both A and AAAA records (the problem isn't the A record, though). I've tried to do this via XOA. After entering everything the list of exports available on the server is actually populated, but selecting one will result in the following error in /var/log/SMlog:

Jun  9 19:38:55 ul SM: [7165] ['mount.nfs', 'sweet-chili.int.bunkus.org:/srv/nfs4/home/', '/var/run/sr-mount/probe', '-o', 'soft,proto=tcp,vers=3,acdirmin=0,acdirmax=0']
Jun  9 19:38:55 ul SM: [7165] FAILED in util.pread: (rc 32) stdout: '', stderr: 'mount.nfs: Network is unreachable
Jun  9 19:38:55 ul SM: [7165] '
Jun  9 19:38:55 ul SM: [7165] Raising exception [73, NFS mount error [opterr=mount failed with return code 32]]
Jun  9 19:38:55 ul SM: [7165] lock: released /var/lock/sm/sr
Jun  9 19:38:55 ul SM: [7165] ***** generic exception: sr_probe: EXCEPTION <class 'SR.SROSError'>, NFS mount error [opterr=mount failed with return code 32]
Jun  9 19:38:55 ul SM: [7165]   File "/opt/xensource/sm/SRCommand.py", line 110, in run
Jun  9 19:38:55 ul SM: [7165]     return self._run_locked(sr)
Jun  9 19:38:55 ul SM: [7165]   File "/opt/xensource/sm/SRCommand.py", line 159, in _run_locked
Jun  9 19:38:55 ul SM: [7165]     rv = self._run(sr, target)
Jun  9 19:38:55 ul SM: [7165]   File "/opt/xensource/sm/SRCommand.py", line 332, in _run
Jun  9 19:38:55 ul SM: [7165]     txt = sr.probe()
Jun  9 19:38:55 ul SM: [7165]   File "/opt/xensource/sm/NFSSR", line 170, in probe
Jun  9 19:38:55 ul SM: [7165]     self.mount(temppath, self.remotepath)
Jun  9 19:38:55 ul SM: [7165]   File "/opt/xensource/sm/NFSSR", line 133, in mount
Jun  9 19:38:55 ul SM: [7165]     raise xs_errors.XenError('NFSMount', opterr=exc.errstr)

The problem here is mount.nfs -o proto=tcp. As can be seen in man 5 nfs the udp and tcp protocols only use IPv4 where as udp6 and tcp6 only use IPv6. I'm not aware of a way of saying "use TCP as the protocol, resolve the name, prefer IPv6 over IPv4", unfortunately.

This isn't limited to XOA, obviously; the corresponding call xe sr-probe type=nfs device-config:server=sweet-chili.int.bunkus.org device-config:serverpath=/srv/nfs4/space fails the same way.

One possible way of addressing this could be to resolve the host name right before constructing the mount commands & using the correct proto depending on whether the management interface is IPv6 enabled.

Note that using an IPv6 address instead of a host name does not work either: even though sr-create works as proto=tcp6 is used in the mount calls according to /var/log/SMlog, the later sr-create does not work with similar error messages.

I can file issues for both on Github, if that helps. The second one in xcp-ng/xcp, I guess, but where would I file the first one? vatesfr/xen-orchestra?

BenjiReis

@mbunkus thanks for the report.

About entering an IPv6 address for NFS in XOA: did you put the [] around the IPv6?
If so and it still failed you can indeed create an issue on vatesfr/xen-orchestra repo (make sure to reference this thread if you do).

For the rest, no need to create issues, i'm aware of them and I'll note them in our internal board for next devs.

Regards

mathieu-gilloots

Thanks for your work !
I'm little new in xcp-ng, I was on xen few year ago.

I'm trying xcp-ng with an ipv6 only server.
ISO file is Ok for install.
Just a little thing, mirrors of packages don't all have an ipv6 record, and on a ipv6 installation I have some error.

I just change the file /etc/yum.repos.d/xcp-ng.repo :
Before :

...
baseurl=http://mirrors.xcp-ng.org/8/8.2/base/x86_64/ http://updates.xcp-ng.org/8/8.2/base/x86_64/
...

After :

...
baseurl=https://updates.xcp-ng.org/8/8.2/base/x86_64/
...

Regards

olivierlambert

Hey good catch! Let me check and fix that ASAP!

olivierlambert

Are you sure about this?

olivier@test:~$ ping mirrors.xcp-ng.org
PING mirrors.xcp-ng.org(alpha.xcp-ng.org (2a01:240:ab08:2::2)) 56 data bytes
64 bytes from alpha.xcp-ng.org (2a01:240:ab08:2::2): icmp_seq=1 ttl=63 time=0.386 ms
64 bytes from alpha.xcp-ng.org (2a01:240:ab08:2::2): icmp_seq=2 ttl=63 time=0.264 ms
^C
--- mirrors.xcp-ng.org ping statistics ---
2 packets transmitted, 2 received, 0% packet loss, time 1002ms
rtt min/avg/max/mdev = 0.264/0.325/0.386/0.061 ms

olivier@test:~$ ping updates.xcp-ng.org
PING updates.xcp-ng.org(alpha.xcp-ng.org (2a01:240:ab08:2::2)) 56 data bytes
64 bytes from alpha.xcp-ng.org (2a01:240:ab08:2::2): icmp_seq=1 ttl=63 time=0.672 ms
64 bytes from alpha.xcp-ng.org (2a01:240:ab08:2::2): icmp_seq=2 ttl=63 time=0.295 ms
^C
--- updates.xcp-ng.org ping statistics ---
2 packets transmitted, 2 received, 0% packet loss, time 1010ms
rtt min/avg/max/mdev = 0.295/0.483/0.672/0.188 ms

edit: aaaaah I see! It's just that SOME mirrors aren't IPv6 ready (ours are). This is indeed less trivial to solve. We'll discuss that with @stormi

mathieu-gilloots

@olivierlambert Hello
Yes, the problem is on http://mirrors.xcp-ng.org/8/8.2/base/x86_64/repodata/repomd.xml, there is 302 Found to non-IPv6 urls

You can try with :

curl -IvL http://mirrors.xcp-ng.org/8/8.2/base/x86_64/repodata/repomd.xml

You will see the differents mirrors associated (Link), and some of them redirect to ipv4.

< Link: <https://xcpng-mirror.as208069.net/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=1; geo=fr
Link: <https://xcpng-mirror.as208069.net/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=1; geo=fr
< Link: <https://mirror.as50046.net/xcp-ng/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=2; geo=fr
Link: <https://mirror.as50046.net/xcp-ng/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=2; geo=fr
< Link: <https://mirror-xcpng.torontot.fr/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=3; geo=fr
Link: <https://mirror-xcpng.torontot.fr/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=3; geo=fr
< Link: <https://updates.xcp-ng.org/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=4; geo=fr
Link: <https://updates.xcp-ng.org/8/8.2/base/x86_64/repodata/repomd.xml>; rel=duplicate; pri=4; geo=fr
< Location: https://rg2-xcpng-mirror.reptigo.fr/8/8.2/base/x86_64/repodata/repomd.xml
Location: https://rg2-xcpng-mirror.reptigo.fr/8/8.2/base/x86_64/repodata/repomd.xml
...
couldn't connect to host at rg2-xcpng-mirror.reptigo.fr:443:
 Failed to connect to 45.152.69.252

jivanpal

Howdy, all, just wondering what the status of this feature is as I'm looking to go IPv6-only on my LAN. If it's complete, is there a way for me to add it to an existing installation of 8.2.1 (stable, i.e. an installation that was not made using one of test ISOs mentioned in this thread)?

Cheers

EDIT: Just my luck that I see this feature mentioned in the 8.3 Beta 1 blog post minutes after I post this! If there's any recommended path to enter the beta so that I can upgrade my existing 8.2.1 installation to it and get this feature, I'd love to know how

stormi

Hi. Check XCP-ng 8.3 beta 1: https://xcp-ng.org/blog/2023/06/22/xcp-ng-8-3-beta-1/

stormi

Sorry, I replied based on the e-mail notification I got, and missed your EDIT

jivanpal

@stormi Am I required to install version 8.3-beta1 from scratch, rather than upgrading, in order to get the new IPv6 functionality? I just ran the upgrade from 8.2.1, but am not seeing any change, nor was I prompted to choose which IP versions to enable during the upgrade process.

If I'm required to upgrade from scratch, is there a recommended way to do this without losing my VM data, given that my pool consists of a single host running all VMs using local storage?

stormi

@jivanpal The only way to enable IPv6 for the management interface is at boot time. It's something that can't be modified afterwards in XAPI.

Reinstalling without overwriting your VM disks is possible. You'd just 1. make sure you have backups, just in case and 2. carefully avoid selecting any disks for local SR creation during installation. Otherwise they would be wiped. I have doubts about what happens when they are on a partition which is on the same disk as the system, though. I'd have to test to be sure.

However, this is not a complete solution, because you'd still have your VM disks, but all VM metadata will be gone so you'd have to re-create all VMs one by one and associate the disks to them. There is a way to export and reimport metadata, but here's the catch: the metadata also contain information about the management interface, and restoring them might also overwrite your brand new IPv6 configuration.

So, to me, the best would be to export the VMs (VM exports do include VM metadata for the exported VM), reinstall the server, and then reimport the VM.

This considering that you only have one server. With more servers, you can use live migration.

jivanpal

@stormi Thanks for the thorough explanation. I will test whether SR partition on system disk is overwritten by installing and reinstalling XCP-ng on a VM on my laptop.

Loss of VM metadata doesn't concern me as I have relatively few VMs and am happy to just recreate these and attach the retained VDIs to them. The only question that remains is whether those VDIs (and some raw/non-sparse VHDs I have that were created by cloning old disks for data recovery tasks) will show up under the Local Storage repo with a simple click of the refresh button in XO, or whether metadata for those also needs to recreated manually.

Rest assured that I have backups I'd obviously just prefer to avoid needing to restore from them as it's time-consuming.

stormi

@jivanpal There's another gotcha, IIRC: after reinstalling, make sure you attach the SR, not create a new one, because SR creation is destructive.

jivanpal

@stormi Thanks, my testing in a VM should reveal how to make sure I do this properly.

lethedata

The only way to enable IPv6 for the management interface is at boot time. It's something that can't be modified afterwards in XAPI.

@stormi Wouldn't it possible to change afterwards by following a similar procedure to replacing an interface card that has the management interface on it? Disable host management, modify PIF with v6 as primary address type (forget/reintroduce/reconfigure), then reset the host management.

It's definitely a process but seems possible

BenjiReis

@lethedata Hi!

Yes indeed it's impossible to change the primary address type of the management interface, it's forbidden by design by XAPI for now. Maybe in the future it'll be allowed but before there still some work to do and then it'd need to be discuss with the XAPI dev team.

I'll try to find a workaround and post here if found but i'm not sure there is for now.

Regards!