IPv6 support in XCP-ng for the management interface - feedback wanted
-
@jivanpal you wrote in the 8.3 beta thread:
There is no way to configure IPv6 on the management interface via xsconsole, such as if one wants to switch between static configuration, autoconf via RAs, or DHCPv6.
True but we'll soon release an new version of xsconsole adapted for IPV6 allowing to configure IPv6 for management interface
There is apparently no support for IPv6 DNS servers, only IPv4. For example, if I try to add an IPv6 address like fd00::1 or [fd00::1] as a DNS server via xsconsole, there is apparently no change to the configuration. Editing /etc/resolv.conf works to achieve this (e.g. adding the line nameserver fd00::1), but this is known not to persist across reboots.
Should be solved by the future xsconsole release as well
There is apparently no support for RDNSS (advertisement of DNS servers in RAs rather than via DHCPv6).
DHCPv6 is one of the major blindspot for now indeed, I'm working on it but I don't have much knowledge on this so any hints are welcome if you spot if something is missing somewhere.
The "autoconf" option (available during installation, after choosing IPv6-only or dual-stack, and then being asked which mode to use to configure IPv6 addresses) appears to only be used at installation time to determine values such as the gateway's link-local address, the available address prefixes, and perform SLAAC and DAD, but then the resulting values are hard-coded and don't change according to changes in the environment, such as an upstream change in network prefix. (I will need to do some more testing to really confirm this, but this seems to be the case in my experience.) Compare this to when IPv4 is configured to use DHCP(v4), in which the management interface may have a different IPv4 address at different times, namely if it's assigned a different address by the DHCP server when it attempts to get or renew a lease.
I'm not aware of this issue, i'll try to reproduce in our env.
Some repos are unreachable in IPv6-only environments, which I'm aware is already known, and I can get around this by using NAT64 (either with CLAT to perform 464XLAT; or with DNS64), but this fact is currently a blocker for me to move to being IPv6-only.
We contacted the mirrors many times, still trying to have'em all advertising IPv4 and 6 and also trying to find a solution that could "smartly" redirect towards a compatible mirror.
Speaking of NAT64, this is just a question, I haven't tested or looked into this myself: Does XCP-ng include a CLAT daemon and support for auto-configuring 464XLAT using either the "PREF64" RA option (RFC8781) or resolution of ipv4only.arpa via a DNS64 server (RFC7050)?
Haven't tested either for now, feel free to do and report if you get here before me.
Again, thank you for the report, this is greatly appreciated and any info about what's missing for IPv6 (and perhaps how to achieve it when possible) is welcomed.
Regards!
-
FYI, I have finally reviewed all mirrors that provide updates for XCP-ng and disabled the remaining 6 which didn't support IPv6 (and notified their owners. I'll enable them again if they enable IPv6).
So, if you experience any issues installing updates via IPv6, tell us so that we investigate faulty mirrors.
-
Stil having this issue:
Failed to set locale, defaulting to C Loaded plugins: fastestmirror Loading mirror speeds from cached hostfile Excluding mirror: updates.xcp-ng.org * xcp-ng-base: mirrors.xcp-ng.org Excluding mirror: updates.xcp-ng.org * xcp-ng-updates: mirrors.xcp-ng.org http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. http://mirrors.xcp-ng.org/8/8.3/base/x86_64/repodata/repomd.xml: [Errno 14] curl#7 - "Failed to connect to 2a01:240:ab08:2::2: Cannot assign requested address" Trying other mirror. One of the configured repositories failed (XCP-ng Base Repository), and yum doesn't have enough cached data to continue. At this point the only safe thing yum can do is fail. There are a few ways to work "fix" this: 1. Contact the upstream for the repository and get them to fix the problem. 2. Reconfigure the baseurl/etc. for the repository, to point to a working upstream. This is most often useful if you are using a newer distribution release than is supported by the repository (and the packages for the previous distribution release still work). 3. Run the command with the repository temporarily disabled yum --disablerepo=xcp-ng-base ... 4. Disable the repository permanently, so yum won't use it by default. Yum will then just ignore the repository until you permanently enable it again or use --enablerepo for temporary usage: yum-config-manager --disable xcp-ng-base or subscription-manager repos --disable=xcp-ng-base 5. Configure the failing repository to be skipped, if it is unavailable. Note that yum will try to contact the repo. when it runs most commands, so will have to try and fail each time (and thus. yum will be be much slower). If it is a very temporary problem though, this is often a nice compromise: yum-config-manager --save --setopt=xcp-ng-base.skip_if_unavailable=true
-
@TheFrisianClause This is not the right thread for your issue. I know it's tempting to think so because of the IPv6 address you see, but your host is not setup to use IPv6 for the management interface, right?
-
@stormi I have made a different thread, but the reply you posted. Made me feel like I could put my reply here.
-
@TheFrisianClause What's the other thread?
-
@stormi This is the link to it: https://xcp-ng.org/forum/topic/8459/yum-update-ipv6-issue
-
@BenjiReis I've finally taken the time to review this again now that I've updated to 8.3-rc1 via
yum update
, so here's some follow-up on the points I brought up previously:There is no way to configure IPv6 on the management interface via xsconsole, such as if one wants to switch between static configuration, autoconf via RAs, or DHCPv6.
True but we'll soon release an new version of xsconsole adapted for IPV6 allowing to configure IPv6 for management interface
There is apparently no support for IPv6 DNS servers, only IPv4. For example, if I try to add an IPv6 address like fd00::1 or [fd00::1] as a DNS server via xsconsole, there is apparently no change to the configuration. Editing /etc/resolv.conf works to achieve this (e.g. adding the line nameserver fd00::1), but this is known not to persist across reboots.
Should be solved by the future xsconsole release as well
Still not seeing any enhancements/changes in behaviour as of xsconsole 11.0.6-1.1.xcpng8.3.
There is apparently no support for RDNSS (advertisement of DNS servers in RAs rather than via DHCPv6).
DHCPv6 is one of the major blindspot for now indeed, I'm working on it but I don't have much knowledge on this so any hints are welcome if you spot if something is missing somewhere.
Just to clarify, this isn't related to DHCPv6, but RAs (Router Advertisement packets). I personally don't have a DHCPv6 server on my network at all. RDNSS is described in RFC8106.
Others may want to advertise DNS servers using DHCPv6, though, so that should still be tested as well.
The "autoconf" option (available during installation, after choosing IPv6-only or dual-stack, and then being asked which mode to use to configure IPv6 addresses) appears to only be used at installation time to determine values such as the gateway's link-local address, the available address prefixes, and perform SLAAC and DAD, but then the resulting values are hard-coded and don't change according to changes in the environment, such as an upstream change in network prefix. (I will need to do some more testing to really confirm this, but this seems to be the case in my experience.) Compare this to when IPv4 is configured to use DHCP(v4), in which the management interface may have a different IPv4 address at different times, namely if it's assigned a different address by the DHCP server when it attempts to get or renew a lease.
I'm not aware of this issue, i'll try to reproduce in our env.
I haven't been able to reproduce this either, and my prefix has changed a couple of times since I said this was an issue. Perhaps I just imagined it, hit a weird edge case, or didn't wait for the valid lifetime of the old prefix to expire; my router doesn't reliably advertise the fact that an old prefix is no longer valid.
Some repos are unreachable in IPv6-only environments, which I'm aware is already known, and I can get around this by using NAT64 (either with CLAT to perform 464XLAT; or with DNS64), but this fact is currently a blocker for me to move to being IPv6-only.
We contacted the mirrors many times, still trying to have'em all advertising IPv4 and 6 and also trying to find a solution that could "smartly" redirect towards a compatible mirror.
@stormi said in IPv6 support in XCP-ng for the management interface - feedback wanted:
FYI, I have finally reviewed all mirrors that provide updates for XCP-ng and disabled the remaining 6 which didn't support IPv6 (and notified their owners. I'll enable them again if they enable IPv6).
So, if you experience any issues installing updates via IPv6, tell us so that we investigate faulty mirrors.
I personally haven't had any issues reaching repos since then, but I haven't explicitly tested this or looked through the mirrorlist. I also don't think this is much of an issue in practice, since 464XLAT can be used; this is no longer a blocker from me, as I've reviewed the way I'm deploying IPv6-only. It's very nice to see you motivate / put pressure on mirror maintainers to make their sites accessible over IPv6 though, especially indirectly by simply removing such sites from the mirrorlist.
Speaking of NAT64, this is just a question, I haven't tested or looked into this myself: Does XCP-ng include a CLAT daemon and support for auto-configuring 464XLAT using either the "PREF64" RA option (RFC8781) or resolution of ipv4only.arpa via a DNS64 server (RFC7050)?
Haven't tested either for now, feel free to do and report if you get here before me.
I've got this working pretty easily by manually installing clatd from GitHub and its dependencies from EPEL and the other RHEL repos. It works, but isn't native. That being said, I don't know of any other Linux distros that natively support this yet. To my knowledge, there is ongoing work to implement this directly in Systemd. Clatd supports RFC7050, but doesn't support PREF64/RFC8781 as it's not particualrly feasible for it to do so, but hopefully Systemd is able to if/when it implements a CLAT.
This also isn't reliable across reboots / DHCP lease renewals because I have no simple way to disable IPv4 on the management interface. I haven't tried this with an installation where I've selected "IPv6-only" in the installer.
One practical issue I've experienced when using 464XLAT in this way is that XO Lite tries to contact the pool server in the frontend / client / web browser using JS fetch calls for URLs falling under
https://localhost/
, which would instead usually be underhttps://<pool server IPv4 address>/
. These are the addresses that XO Lite will prompt the user to ensure that the browser trusts TLS certificates for if they are self-signed and no known CA has issued/signed them. As such, these don't work, since "localhost" from the XO Lite user's perspective isn't the same machine as the "localhost" that XO Lite is running on. If XO Lite supported making these calls using any of the pool servers' routable IPv6 addresses (e.g. ULAs or GUAs, but not LLAs), this would work just fine.I may find some time to test these things on an "IPv6-only" installation, but I expect that will be after 8.3 has reached general release.
-
RE: https://xcp-ng.org/blog/2021/02/09/ipv6-in-xcp-ng/
Couple questions regarding future IPv6 support:
- Is it currently on the docket to support multiple IPv6 addresses on the management interface? In my environment I use a ULA (Unique-Local Address) and GUA (Globally-Unique Address) for private and public access respectively on all of my servers. I would personally be OK if this supported via static assignments initially, but at present it appears only one address may be assigned. Multiple IPv6 addresses should be supported per-interface regardless because...
- There is no LLA (Link-Local Address) address assigned to the management interface which violates some IPv6 RFC and will be required for SLAAC in the future. Are there plans or a bug filed to correct this issue?
Per https://datatracker.ietf.org/doc/html/rfc4291:
"All interfaces are required to have at least one Link-Local unicast address (see Section 2.8 for additional required addresses). A single interface may also have multiple IPv6 addresses of any type (unicast, anycast, and multicast) or scope. Unicast addresses with a scope greater than link-scope are not needed for interfaces that are not used as the origin or destination of any IPv6 packets to or from non-neighbors."
-
Question for @Team-XAPI-Network
-
@bnerickson
About 1, That's something that I'd like to support in the future, but it will take some time to get there. About 2, currently LLAs are not allowed to be the used as the management IPs, and I wouldn't like them to be used as such: it can lead to hosts not being to connect to one another if they're not in the same physical network. They still could be configured for the interfaces and shown to the user, but not as the only IP for the interface, or the main one for that matter. -
@psafont Sounds good on point #1. On point #2 I agree LLA shouldn't be the primary management IPv6 address on the interface, but you could run into trouble by not having a LLA address assigned at all. All of the IPv6 standards assume a LLA is assigned to an interface running IPv6 for things like NDP or RA to work ergo mysterious things could break as a result of an LLA address not being assigned.
Just spitballing here, but if you're concerned that only an LLA would be assigned to an interface then perhaps there could be logic to disable IPv6 for said management interface if no non-LLA address is assigned, or IPv4 could be preferred if only LLA addresses as assigned?