[AusNOG] Experiences with RPKI

Thu May 23 16:34:45 AEST 2024

Hey Joe,

I don’t believe this to be the case - there are many networks much larger than yours that use max-length as coverage extensively.
5.1 in RFC9319 is precisely the reason why these networks choose to do this - the propagation latency for ad-hoc records is too slow for effective DDoS mitigation.

If your network is relatively small and you have infrequent route changes, I recommend creating minimal ROAs for exactly what you intend to announce. If that means creating a parent /22 ROA and child /24 ROAs (because you intend to advertise more specifics as well) then that’s the best approach. You’re limiting your exposure for hijacks on other prefix lengths but giving enough flexibility for your advertisements. Making ROA creation part of the “new route advertisement” process should just become the norm.

Larger networks with more frequent changes should, ideally, use automation to create ROAs on the fly - e.g. driven from Netbox - once the prefix is created, hit up your friendly RIR API to create the ROA. 

And finally, I don’t believe there is validation logic to invalidate perfectly valid routes - if a ROA exists for a /24 and that’s what you’re advertising, it will be considered valid. Creating ROAs for your existing advertisements is the absolute lowest risk thing you can do.

Cheers,

Tim

> On 23 May 2024, at 16:08, Joseph Goldman <joseph at goldman.id.au> wrote:
> 
> Hi Phil,
> 
>  Thanks - I 100% understand the point of best practice of not using maxLength and the potential for hijacking, what im most worried about is the statement i've been told about 'unused' ROAs affecting used ROAs, which just seems counter-intuitive to me, in terms of (as per the example given) having ROAs ready for failover or TE purposes.
> 
>  This is the RFC im referring to:
> 
> https://datatracker.ietf.org/doc/html/rfc9319
> 
> Specifically 5.1 which indicates having dormant ROAs available for use, but I am being told having any dormant ROAs is grounds for all routes to be marked invalid based on stricter validators, which is the answer im trying to ascertain right now.
> 
> In a failover sense, say for example I am advertising /22 out of my Brisbane POP and some /24's out of my Sydney POP, my Sydney POP goes down and im no longer originating those /24's, by what i've been told my /22 would now become invalid on next check as my /24's are not being advertised, even though I have ROAs for the /22 and /24's.
> 
>  I believe i've been given somewhat wrong information but they were basing it off experiences with other APNIC members, I just dont want to end up in a position where our routes are dropped after implementing our ROAs, and maintain flexibility to not wait multiple hours before we can advertise a new prefix if required.
> 
> Thanks,
> Joe
> 
> ------ Original Message ------
> From: "Phil Mawson" <phil.mawson at gmail.com <mailto:phil.mawson at gmail.com>>
> To: "Joseph Goldman" <joseph at goldman.id.au <mailto:joseph at goldman.id.au>>
> Cc: "ausnog at lists.ausnog.net <mailto:ausnog at lists.ausnog.net>" <ausnog at lists.ausnog.net <mailto:ausnog at lists.ausnog.net>>
> Sent: 23/05/2024 3:52:54 PM
> Subject: Re: [AusNOG] Experiences with RPKI
> 
>> Hi Joe,
>> 
>> First up, well done on working on your RPKI roll out.  Signing your own routes is the most important step you can take to protect your own network.
>> 
>> In regards to using max length, I do advise against that as what it means someone can still hijack one of your un-advertised routes and it would be treated as real.
>> 
>> IE: If you advertise and sign a /19, but have a ROA created for each route up to a /24.  If you are not advertising all of those, a third party could advertise one of our /24s and spoof your ASN and it would be treated as valid on the internet.
>> 
>> APNIC portal make it very easy to create ROAs for your own routes as it has the route table view so it can see what is already publicly advertise.  I recommend looking at that and then signing routes based on that.  
>> 
>> DDoS  mitigation providers and ROAs is an interesting topic and not sure the community can agree on what is the best technical solution for that.
>> 
>> Regards,
>> Phil
>> 
>> 
>>> On 23 May 2024, at 3:46 PM, Joseph Goldman <joseph at goldman.id.au <mailto:joseph at goldman.id.au>> wrote:
>>> 
>>> G'day list,
>>> 
>>>  In the process of rolling out RPKI - and while I thought I had a good grasp on everything, there is one niggling piece of information that I've come against and can't verify. Was hoping people can share their experiences.
>>> 
>>>  We are only doing our ROA's to begin with and not implementing validation until later, the initial thought was to create an ROA for all our 'supernets' and use maxLength to 24 to help cover any prefix we may want to advertise. We are a much simpler setup, single AS only and we do advertise many of our ranges down to /24 but not all of them. I do know of the best practices of not using maxLength based on a draft rfc doc, but I am personally not super concerned for our relatively small use-case to the issues brought up in that doc.
>>> 
>>>  Where I have come into trouble is a source (APNIC helpdesk) indicating that if we have any ROAs that exist for prefixes we are not directly advertising - it may lend some validators to mark all our routes as invalid?
>>> 
>>> i.e. say we had /22 ROA, 2x /23 ROAs and 4x /24 ROAs - are currently advertising the /22 and 2x /24's, so 2x /23's and 2x /24 ROAs are 'unused' in that we are not advertising those specific resources - would that cause issues with strict validators out in the wild?
>>> 
>>>  My understanding reading through the RFC's is this should not be the case. If any ROA that matches the prefix for the origin AS exists it should be valid, regardless of other ROAs signed by the same resource holder etc.
>>> 
>>>  Matching ROAs to exact advertisements is great, but it seems to lend itself to much less flexibility in traffic engineering and failover scenarios - a good scenario is having dormant /24 ROAs for say a DDoS mitigation service to use when needed, so you dont have to wait for RPKI propagation before scrubbing kicks in.
>>> 
>>>  Based on your experience, is having all-encompassing (using maxLength), or unused ROAs an acceptable way to use RPKI or will we run into issues?
>>> 
>>> All help appreciated :)
>>> 
>>> Thanks,
>>> Joe
>>> _______________________________________________
>>> AusNOG mailing list
>>> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
>>> https://lists.ausnog.net/mailman/listinfo/ausnog
>> 
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> https://lists.ausnog.net/mailman/listinfo/ausnog

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ausnog.net/pipermail/ausnog/attachments/20240523/b7018022/attachment.htm>