1. 41
    1. 27

      It’s worth linking to A&A’s (a British ISP) response to this: https://www.aa.net.uk/etc/news/bgp-and-rpki/

      1. 16

        Our (Cloudflare’s) director of networking responded to that on Twitter: https://twitter.com/Jerome_UZ/status/1251511454403969026

        there’s a lot of nonsense in this post. First, blocking our route statically to avoid receiving inquiries from customers is a terrible approach to the problem. Secondly, using the pandemic as an excuse to do nothing, when precisely the Internet needs to be more secure than ever. And finally, saying it’s too complicated when a much larger network than them like GTT is deploying RPKI on their customers sessions as we speak. I’m baffled.

        (And a long heated debate followed that.)

        A&A’s response on the one hand made sense - they might have fewer staff available - but on the other hand RPKI isn’t new and Cloudflare has been pushing carriers towards it for over a year, and route leaks still happen.

        Personally as an A&A customer I was disappointed by their response, and even more so by their GM and the official Twitter account “liking” some very inflammatory remarks (“cloudflare are knobs” was one, I believe). Very unprofessional.

      2. 15

        Hmm… I do appreciate the point that route signing means a court can order routes to be shut down, in a way that wouldn’t have been as easy to enforce without RPKI.

        I think it’s essentially true that this is CloudFlare pushing its own solution, which may not be the best. I admire the strategy of making a grassroots appeal, but I wonder how many people participating in it realize that it’s coming from a corporation which cannot be called a neutral party?

        I very much believe that some form of security enhancement to BGP is necessary, but I worry a lot about a trend I see towards the Internet becoming fragmented by country, and I’m not sure it’s in the best interests of humanity to build a technology that accelerates that trend. I would like to understand more about RPKI, what it implies for those concerns, and what alternatives might be possible. Something this important should be a matter of public debate; it shouldn’t just be decided by one company aggressively pushing its solution.

        1. 4

          This has been my problem with a few other instances of corporate messaging. Cloudflare and Google are giant players that control vast swathes of the internet, and they should be looked at with some suspicion when they pose as simply supporting consumers.

          1. 2

            Yes. That is correct, trust needs to be earned. During the years I worked on privacy at Google, I liked to remind my colleagues of this. It’s easy to forget it when you’re inside an organization like that, and surrounded by people who share not only your background knowledge but also your biases.

      3. 9

        While the timing might not have been the best, I would overall be on Cloudflare’s side on this. When would the right time to release this be? If Cloudflare had waited another 6-12 months, I would expect them to release a pretty much identical response then as well. And I seriously doubt that their actual actions and their associated risks would actually be different.

        And as ISPs keep showing over and over, statements like “we do plan to implement RPKI, with caution, but have no ETA yet” all too often mean that nothing will every happen without efforts like what Cloudflare is doing here.


        If we simply filtered invalid routes that we get from transit it is too late and the route is blocked. This is marginally better than routing to somewhere else (some attacker) but it still means a black hole in the Internet. So we need our transit providers sending only valid routes, and if they are doing that we suddenly need to do very little.

        Is some really suspicious reasoning to me. I would say that black hole routing the bogus networks is in every instance significantly rather than marginally better than just hoping that someone reports it to them so that they can then resolve it manually.

        Their transit providers should certainly be better at this, but that doesn’t remove any responsibility from the ISPs. Mistakes will always happen, which is why we need defense in depth.

      4. 6

        Their argument is a bit weak in my personal opinion. The reason in isolation makes sense: We want to uphold network reliability during a time when folks need internet access the most. I don’t think anyone can argue with that; we all want that!

        However they use it to excuse not doing anything, where they are actually in a situation where not implementing RPKI and implementing RPKI can both reduce network reliability.

        If you DO NOT implement RPKI, you allow route leaks to continue happening and reduce the reliability of other networks and maybe yours.

        If you DO implement RPKI, sure there is a risk that something goes wrong during the change/rollout of RPKI and network reliability suffers.

        So, with all things being equal, I would chose to implement RPKI, because at least with that option I would have greater control over whether or not the network will be reliable. Whereas in the situation of NOT implementing, you’re just subject to everyone else’s misconfigured routers.

        Disclosure: Current Cloudflare employee/engineer, but opinions are my own, not employers; also not a network engineer, hopefully my comment does not have any glaring ignorance.

        1. 4

          Agreed. A&A does have a point regarding Cloudflare’s argumentum in terrorem, especially the name and shame “strategy” via their website as well as twitter. Personally, I think is is a dick move. This is the kind of stuff you get as a result:

          This website shows that @VodafoneUK are still using a very old routing method called Border Gateway Protocol (BGP). Possible many other ISP’s in the UK are doing the same.

          1. 1

            I’m sure the team would be happy to take feedback on better wording.

            The website is open sourced: https://github.com/cloudflare/isbgpsafeyet.com

            1. 1

              The website is open sourced: […]

              There’s no open source license in sight so no, it is not open sourced. You, like many other people confuse and/or conflate anything being made available on GitHub as being open source. This is not the case - without an associated license (and please don’t use a viral one - we’ve got enough of that already!), the code posted there doesn’t automatically become public domain. As it stands, we can see the code, and that’s that!

              1. 7

                There’s no open source license in sight so no, it is not open sourced.

                This is probably a genuine mistake. We never make projects open until they’ve been vetted and appropriately licensed. I’ll raise that internally.

                You, like many other people confuse and/or conflate anything being made available on GitHub as being open source.

                You are aggressively assuming malice or stupidity. Please don’t do that. I am quite sure this is just a mistake nevertheless I will ask internally.

                1. 1

                  There’s no open source license in sight so no, it is not open sourced.

                  This is probably a genuine mistake. We never make projects open until they’ve been vetted and appropriately licensed.

                  I don’t care either way - not everything has to be open source everywhere, i.e. a website. I was merely stating a fact - nothing else.

                  You are aggressively […]

                  Not sure why you would assume that.

                  […] assuming malice or stupidity.

                  Neither - ignorance at most. Again, this is purely statement of a fact - no more, no less. Most people know very little about open source and/or nothing about licenses. Otherwise, GitHub would not have bother creating https://choosealicense.com/ - which itself doesn’t help the situation much.

              2. 1

                It’s true that there’s no license so it’s not technically open-source. That being said I think @jamesog’s overall point is still valid: they do seem to be accepting pull requests, so they may well be happy to take feedback on the wording.

                Edit: actually, it looks like they list the license as MIT in their package.json. Although given that there’s also a CloudFlare copyright embedded in the index.html, I’m not quite sure what to make of it.

                1. -1

                  If part of your (dis)service is to publically name and shame ISPs, then I very much doubt it.

      5. 2

        While I think that this is ultimately a shit response, I’d like to see a more well wrought criticism about the centralized signing authority that they mentioned briefly in this article. I’m trying to find more, but I’m not entirely sure of the best places to look given my relative naïvete of BGP.

        1. 4

          So as a short recap, IANA is the top level organization that oversees the assignment of e.g. IP addresses. IANA then delegates large IP blocks to the five Regional Internet Registries, AFRINIC, APNIC, ARIN, LACNIC, and RIPE NCC. These RIRs then further assigns IP blocks to LIRs, which in most cases are the “end users” of those IP blocks.

          Each of those RIRs maintain an RPKI root certificate. These root certificates are then used to issue certificates to LIRs that specify which IPs and ASNs that LIR is allowed to manage routes for. Those LIR certificates are then used to sign statements that specify which ASNs are allowed to announce routes for the IPs that the LIR manages.

          So their stated worry is then that the government in the country in which the RIR is based might order the RIR to revoke a LIR’s RPKI certificate.

          This might be a valid concern, but if it is actually plausible, wouldn’t that same government already be using the same strategy to get the RIR to just revoke the IP block assignment for the LIR, and then compel the relevant ISPs to black hole route it?

          And if anything this feels even more likely to happen, and be more legally viable, since it could target a specific IP assignment, whereas revoking the RPKI certificate would make the RoAs of all of the LIRs IP blocks invalid.

          1. 1

            Thanks for the explanation! That helps a ton to clear things up for me, and I see how it’s not so much a valid concern.

      6. 1

        I get a ‘success’ message using AAISP - did something change?

        1. 1

          They are explicitly dropping the Cloudflare route that is being checked.