News for hackers. Stuff that matters.
Re:IPv6 delegation, an unusually subtle failure mode by o_-_o on Wednesday November 18, 2026 @08:27AM (#50546057)
I think the most disturbing part of this article is; "I haven't replaced the dead CMOS battery on my router and the clock resets every time the router restarts." O_o
Re:IPv6 delegation, an unusually subtle failure mode by tosti on Wednesday November 18, 2026 @08:34AM (#50546134)
That would be bad for a server, not so much for a router.
Re:IPv6 delegation, an unusually subtle failure mode by tosapple on Wednesday November 18, 2026 @08:37AM (#50546164)
and if it has a deterministic rng, it will restart/initialize at seed[0] every time?
Re:IPv6 delegation, an unusually subtle failure mode by teraflop on Wednesday November 18, 2026 @09:41AM (#50546895)
If your system is relying on wall clock time for a random seed, it's already badly broken.
Re:IPv6 delegation, an unusually subtle failure mode by jervant on Wednesday November 18, 2026 @10:23AM (#50547415)
Re:IPv6 delegation, an unusually subtle failure mode by heavensteeth on Wednesday November 18, 2026 @10:12AM (#50547282)
Linux' RNG is far more complex than that.
Re:IPv6 delegation, an unusually subtle failure mode by RVuRnvbM2e on Wednesday November 18, 2026 @08:31AM (#50546104)
Isn't it a misconfiguration that the ISP is handing out a different prefix each time pppd restarts? Shouldn't the prefix be fairly static?
Re:IPv6 delegation, an unusually subtle failure mode by BadBadJellyBean on Wednesday November 18, 2026 @08:37AM (#50546169)
No. Not necessarily.
Re:IPv6 delegation, an unusually subtle failure mode by RVuRnvbM2e on Wednesday November 18, 2026 @08:54AM (#50546340)
Yes it is. The exact scenario from the article is a well known problem with dynamic prefixes: https://www.ietf.org/archive/id/draft-palet-v6ops-prefix-ass...
Re:IPv6 delegation, an unusually subtle failure mode by throw0101c on Wednesday November 18, 2026 @09:37AM (#50546826)
> https://www.ietf.org/archive/id/draft-palet-v6ops-prefix-ass...

PSA: You may wish to point to the most recent revision (or use a link that lists all revisions, and would point to the perhaps-eventual RFC), as opposed to an expired document from 2025:

* https://datatracker.ietf.org/doc/draft-ietf-v6ops-prefix-to-...

Re:IPv6 delegation, an unusually subtle failure mode by RVuRnvbM2e on Wednesday November 18, 2026 @09:39AM (#50546855)
Right, thanks. It was difficult finding a vaguely recalled RFC on my phone :D
Re:IPv6 delegation, an unusually subtle failure mode by Joker_vD on Wednesday November 18, 2026 @09:46AM (#50546952)
It boggles my mind that they would deliberately design Internet (!) Protocol that poorly handles the dynamic nature of the networks its supposed to interconnect. The prefixes are dynamic, that's the fact of life.

But then again, at least half of the IPv6 design looks as a deliberate and somewhat spiteful attempt to force the ISPs and ASs to do things in a radically different ways than they are doing already just fine.

Re:IPv6 delegation, an unusually subtle failure mode by throw0101c on Wednesday November 18, 2026 @09:51AM (#50547000)
> It boggles my mind that they would deliberately design Internet (!) Protocol that poorly handles the dynamic nature of the networks its supposed to interconnect. The prefixes are dynamic, that's the fact of life.

The 'dynamic' nature of these delegations is a policy decision of the ISP and not of the protocol: nothing is stopping them from giving the same prefix to the same customer.

It's no different than a DHCP(v4) server on a corporate network remembering which MAC got which IPv4 address in the past and giving the same one over the course of time, or of the IT folks telling the DHCP system to not remembering and give a new address each time.

Re:IPv6 delegation, an unusually subtle failure mode by Joker_vD on Wednesday November 18, 2026 @12:09PM (#50548886)
> The 'dynamic' nature of these delegations is a policy decision of the ISP and not of the protocol: nothing is stopping them from giving the same prefix to the same customer.

The 'static' delegation, however, is the policy decision of the protocol because it deals very poorly with a different policy. And it shouldn't; protocols should give mechanisms for implementing different reasonable policies — and there is nothing unreasonable about dynamic delegation except for the fact that the creators of IPv6 decided "yeah, no, those should be static, and if someone thinks otherwise, they can see figure 1".

Re:IPv6 delegation, an unusually subtle failure mode by unethical_ban on Wednesday November 18, 2026 @11:34AM (#50548387)
My read is that if things followed RFCs and best practices from the protocols correctly, a lot of things wouldn't be an issue.

I remember about 10 years ago trying to get IPv6 on an ISP and they were handing me a /128. And clients like home routers had/have bad UIs for managing IPv6.

Of course the infrastructure should handle dynamic allocations properly, and IPv6 delegations should not change every time a connection/daemon bounces just because it can.

Re:IPv6 delegation, an unusually subtle failure mode by ebiederm on Wednesday November 18, 2026 @11:01AM (#50547937)
Agreed.

That draft BCP doesn't put it as well as the RFC's unfortunately.

In short renumbering as little as possible is the current best practice because IPv6 renumbering is not as robust as it can be.

Flash renumbering (AKA renumbering without warning) has been documented as a problem.

The practical problem is the default lifetimes for prefixes are in days or weeks, and without something to notice and report the prefix is gone hosts continue to use it.

A number of RFCs are working their way through the system to get the handling of renumbering and especially flash renumbering as robust as possible.

Even when it is documented the best way to handle the sitation and the RFCs are published. It will take a little bit for all of the software to be updated.

All of that said renumbering somewhat regularly (monthly?) on IPv6 remains important so that bugs don't creep into the process unnecessarily.

Re:IPv6 delegation, an unusually subtle failure mode by rnhmjoj on Wednesday November 18, 2026 @08:48AM (#50546286)
It's likely not a misconfiguration, but it's a bad practice [1] to implement dynamic IPv6 prefixes like they were IPv4 addresses from a pool. I see no reason to do it like this other than "that's what we have always done with IPv4".

Non-sticky, dynamic prefixes cause all sort of mayhem because suddenly your RAs, firewall rules, DNS recods, NAT64 pools etc. are all outdated and you have to somehow update them.

[1]: https://www.ripe.net/publications/docs/ripe-690/

Re:IPv6 delegation, an unusually subtle failure mode by izacus on Wednesday November 18, 2026 @08:52AM (#50546319)
I hear that this is a common practice nowadays across worlds' ISPs though :/
Re:IPv6 delegation, an unusually subtle failure mode by lmz on Wednesday November 18, 2026 @09:34AM (#50546794)
Wouldn't want the users to be too attached to the prefix if they're not paying for a business plan.
Re:IPv6 delegation, an unusually subtle failure mode by throw0101c on Wednesday November 18, 2026 @09:47AM (#50546966)
> Wouldn't want the users to be too attached to the prefix if they're not paying for a business plan.

Residential assignments are often /56 (2^8 subnets) or even maybe /60 (2^4 subnets): this may or may not be enough for business needs, depending on the size of the organization.

Re:IPv6 delegation, an unusually subtle failure mode by lmz on Wednesday November 18, 2026 @09:58AM (#50547088)
Yes, but switching the prefixes regularly keeps them from easily putting the addresses in DNS, facilitating an upsell to the business plan.
Re:IPv6 delegation, an unusually subtle failure mode by kccqzy on Wednesday November 18, 2026 @09:01AM (#50546435)
Yeah it’s annoying enough that I had to set up my DNS server on a ULA network, so I can hardcode it in my RA without thinking about ISP prefixes. As for NAT64, I just stuck to the well-known 64:ff9b instead of using my own prefixes. It also simplifies DNS64.
Re:IPv6 delegation, an unusually subtle failure mode by wernam on Wednesday November 18, 2026 @11:14AM (#50548091)
NAT64 is for ipv4. How does relate to ipv6 routing?? Also, ULA is non routable and has lower precedence (3) than ipv4 (35), so ipv6 will never be used in your case
Re:IPv6 delegation, an unusually subtle failure mode by eqvinox on Wednesday November 18, 2026 @12:19PM (#50549081)
> ULA is non routable

ULA is absolutely routable. Are you confusing it with LLA?

> and has lower precedence (3) than ipv4 (35)

This already got changed in https://datatracker.ietf.org/doc/html/draft-ietf-6man-rfc672..., the document is fully done with the RFC editor queue and only waiting for the [SNACBIT] reference to get published.

New preference table:

  Prefix        Precedence Label
  ::1/128               50     0
  $known_local/4x       45    14 (**)
  ::/0                  40     1
  fc00::/7              30    13 (*)
  ::ffff:0:0/96         20     4 (*)
  2002::/16              5     2 (*)
  2001::/32              5     5
  ::/96                  1     3
  fec0::/10              1    11
  3ffe::/16              1    12
Re:IPv6 delegation, an unusually subtle failure mode by wernam on Wednesday November 18, 2026 @01:41PM (#50550301)
ULA is routable, not internet routable, I stand corrected.

Yes many keep track of that RFC. However you're aware that every client device, tv, phone, esp32, iot device, has to apply this change? That will take decades

Re:IPv6 delegation, an unusually subtle failure mode by kccqzy on Wednesday November 18, 2026 @01:54PM (#50550478)
[delayed]
Re:IPv6 delegation, an unusually subtle failure mode by wernam on Wednesday November 18, 2026 @02:05PM (#50550603)
If you're using GUA then how are you handling the rotating prefix by the ISP, the problem mentioned in article? You don't have a fixed GUA. Most residential networks have this exact same problem as in the article.
Re:IPv6 delegation, an unusually subtle failure mode by kccqzy on Wednesday November 18, 2026 @02:10PM (#50550661)
Simple. The same way the original article solves it: using odhcpd from OpenWrt.
Re:IPv6 delegation, an unusually subtle failure mode by kccqzy on Wednesday November 18, 2026 @01:19PM (#50549963)
[delayed]
Re:IPv6 delegation, an unusually subtle failure mode by throw0101c on Wednesday November 18, 2026 @09:34AM (#50546784)
> It's likely not a misconfiguration, but it's a bad practice [1] to implement dynamic IPv6 prefixes like they were IPv4 addresses from a pool.

I don't necessarily disagree with the assessment, but to a certain extent I like getting new addresses/prefixes to help with privacy: with my previous ISP I had my own Asus router (now have to use ISP-provided), and it had built-in functionality to schedule reboots.

* https://www.asus.com/support/faq/1049258/

Re:IPv6 delegation, an unusually subtle failure mode by thedougd on Wednesday November 18, 2026 @10:20AM (#50547379)
This seems like a good use for DUID with the DHCPv6 request. The ISP would return the same prefix for the same DUID, but if you wanted a rotation, rotate your DUID.
Re:IPv6 delegation, an unusually subtle failure mode by pepperoni_pizza on Wednesday November 18, 2026 @09:38AM (#50546849)
They probably do it so you have to pay for a static one.
Re:IPv6 delegation, an unusually subtle failure mode by crote on Wednesday November 18, 2026 @02:50PM (#50551194)
I can easily see quasi-persistent prefixes being a giant support nightmare, because people will assume it is persistent, hardcode it everywhere, and then get upset when it suddenly changes after a year or two due to a big ISP network reconfig. Intentionally rotating it more often than strictly needed prevents this kind of ossification, because any user has to account for renumbering from the very start.

To me it seems the main issue is that the old prefix immediately stops being valid. There's not exactly a shortage of IPv6 addresses, so the ISP shouldn't be in any hurry to reuse the old prefix. If they were to still route it to the customer for, say, 24 hours, then it would give equipment time to switch over to the new one.

Re:IPv6 delegation, an unusually subtle failure mode by genghizkhan on Thursday November 19, 2026 @12:04AM (#50556262)
Ideally yes, the prefix should be static. However, in talking with a few engineers who work at ISPs, I've been told that the prefix is often handed out from a pool by whichever access server honours the customer's request, and that's somewhat random. My ISP would have to setup infrastructure to ensure the prefix isn't random and since no one's asking for it, it's not being built.

This is a guess on my part, of course, I don't know any network engineers at my own ISP.

Re:IPv6 delegation, an unusually subtle failure mode by arcfour on Wednesday November 18, 2026 @08:44AM (#50546223)
Why do you even need networkd for a router? Setting one up on Alpine (so not all this cruft) is easy as pie and the whole thing can cold boot in under 10 seconds... Would make the whole setup much more understandable + smaller attack surface...
Re:IPv6 delegation, an unusually subtle failure mode by o_-_o on Wednesday November 18, 2026 @08:53AM (#50546327)
Yeah this is why I have that Arch Linux uses systemd by default. Will have to switch to non-systemd os soon.
Re:IPv6 delegation, an unusually subtle failure mode by amaccuish on Wednesday November 18, 2026 @03:16PM (#50551550)
networkd doesn't run as PID1?
Re:IPv6 delegation, an unusually subtle failure mode by hylaride on Wednesday November 18, 2026 @10:19AM (#50547358)
Networking is the biggest cause of issues I've had with systemd, mostly because of the systemd maintainer's insistence on strict adherence to modern RFC's to the detriment of legacy support. This has resulted in my having to deal with gross workarounds for various issues, including this 032 issue with multiple domains: https://docs.aws.amazon.com/linux/al2023/ug/deprecated-al2.h...
Re:IPv6 delegation, an unusually subtle failure mode by lmz on Wednesday November 18, 2026 @10:44AM (#50547703)
That multiple domain issue doesn't seem to stem from "modern RFCs", just a decision to not be bug compatible with the previous implementation.

> Windows and other Linux operating systems treat the value as a single domain, which results in unexpected behavior.

The parent poster's suggestion to use Alpine is amusing in the context of your problem because musl libc is notorious for sticking to the spec when compared to glibc.

Re:IPv6 delegation, an unusually subtle failure mode by eqvinox on Wednesday November 18, 2026 @12:12PM (#50548956)
> maintainer's insistence on strict adherence to modern RFC

…except when they don't and then issues like this are the result. I guess they don't consider themselves viable for a CPE device and don't "feel" like following CPE requirements.

(In itself, good to know that systemd does consider some things out of scope, maybe?)

Re:IPv6 delegation, an unusually subtle failure mode by doublepg23 on Wednesday November 18, 2026 @11:32AM (#50548365)
> smaller attack surface

I literally run alpine in prod and home but I think the recent exploits in musl should give us pause if this is a fact or if it's a heuristic.

Re:IPv6 delegation, an unusually subtle failure mode by arcfour on Wednesday November 18, 2026 @05:21PM (#50553106)
The number of packages and amount of code in an Alpine distribution is quantitatively a smaller attack surface; that doesn't mean there are never going to be any vulnerabilities. Also - it's not like they're coming out with new ones every day - what was it, 1 or 2?
Re:IPv6 delegation, an unusually subtle failure mode by miladyincontrol on Thursday November 19, 2026 @12:43AM (#50556464)
Is systemd required? No. But it certainly makes sane system configuration easier and more reliable. Try it!
Re:IPv6 delegation, an unusually subtle failure mode by jon128 on Wednesday November 18, 2026 @09:23AM (#50546677)
Part of how some of my devices cope with ISPs doing things the wrong way is to ping IP addresses belonging to nearby AS and some remote AS from the source addresses of each local IP prefix.

When a local IP prefix is detected as stale relative to other local IP prefixes, my local router(s) sound out appropriate router announcements to squash the stale prefixes.

Devices I travel with do the same.

It seems these days, most eyeball networks aren't too technical, rely on whatever closed source software vendors poop out, and for the most part they are just plugging things in. That, and billing.

Re:IPv6 delegation, an unusually subtle failure mode by wowczarek on Wednesday November 18, 2026 @09:53AM (#50547029)
I thought PD in Linux and *BSDs was a solved issue. I think I've been spoilt by dedicated NOSes.
Re:IPv6 delegation, an unusually subtle failure mode by thedougd on Wednesday November 18, 2026 @10:18AM (#50547343)
This is a narrow issue related to PD when a PPP connection is used. I can attest that PD works well for me with systemd-networkd.
Re:IPv6 delegation, an unusually subtle failure mode by miyuru on Wednesday November 18, 2026 @10:19AM (#50547359)
One my LTE router's had a similar issue, which i documented for anyone(or AI agents) stumbling on it here.

https://blog.miyuru.lk/zlt-ipv6-fixes/