BGP hijack?

We just had every single prefix in AS378 start being announced by AS2027.

Every announcement by AS2027 is failing RPKI yet being propagated a bit.
Is this yet another misbehaving device or an actual attack?



Same stuff (with our ASN and our prefixes) detected here, coming from AS2027 (Milkywan), for a short time…

Hank, all exact match for prefix length? Or longer subnets covering the whole?
(Is this leakage of a optimizer or possibly censorship leakage?)

Hello everyone,
I'm working for MilkyWan / AS2027 and I wanted to give you some explanations regarding this incident. Last week-end, during an upgrade on our network configuration, it appears that some prefixes were announced with an incorrect AS Path. Based on our analysis, none of these routes seem to have been announced anywhere, but to some route-collectors (RIPE RIS, BGP.Tools, Qrator,, NLNOG RING), and therefore didn't effectively end up in the DFZ.
The issue was discovered quickly (in a few minutes) and corrected right away.
The incident is now closed on our side; please reach out to us should you see anything proving otherwise.

We deeply apologize for that and we can confirm it was not a BGP hijack attempt.

Wishing you a very pleasant day.

Vincent F. for Milkywan Team

Thanks for the transparency, Vincent. Are you able to share how the AS-Path became mangled to begin with? I’m assuming this was some kind of route optimizer, but maybe something else going on?

Hey everybody, I run, (And had a extremely busy alerting
engine for a few minutes)

From what can see it seems like they had a private asn in
the path like so

2027 4220270000 6696 6939 42615 212232

This can be valid for a number of reasons, ( they might have been
doing some homemade BGP confederation for example ), and I assume
then that they had enabled some kind of private asn filter that had
not quite done what they expected. I think what they are expecting
was the part to look like this:

2027 6696 6939 42615 212232

However instead the private AS stripping function instead did this,
and sent it to their customers/collector feeds:


This then obviously made everything look like a BGP origin hijack to
all of the route collectors and alerting systems.

It's worth noting that saw this from more than MilkyWan
directly, but also from what I can assume are their customers. But I
don't see any indication this faulty routing information propagated
anywhere else than that. ( To sort of backup the response that Vincent
has already provided us)

Hope this provides some interesting insight, and maybe some future heads up :slight_smile:


Because we were migrating our default table containing the DFZ into a VRF, we had a BGP session between 2 routers terminating on one side in the main table and on the other in the VRF. We had to remove the no-export from our redistribution route-map because of this private eBGP peering.

As we were concerned about reachability with transits and peerings on both sides, we tried to activate a route-leak between the main table and the newly installed VRF DFZ.

However, as a route-leak doesn’t retain BGP attributes, routes started to be learned in originate from the router containing the route-leak. So there was no hijacking on the DFZ. Moreover, my route collector listening on the DFZ did not identify any hijack during the entire migration.

On the Transit and Peering side, we are subject to max-pref on the provider side, and we have a route-map in prefix-list specific to our AS + customer community.

However, we consider Route Collectors as customers, who redistribute prefixes greater than or equal to /24 or /48, minus bogons, and based on no-export, we don’t send our internal routes. Except that in addition to the route leak, the no-export was removed, which let through all IGP originate routes to our customers, and route collectors.

The problem was quickly identified, we cut the route leak and stopped all transits and peering in main table to leave only the DFZ and work on the end of migration having only some LNS not configured in the good VRF, and finished the migration direction from the new VRF DFZ.

So don’t rely on the Route Collector, which is updated at the whim of the various operators, but compare what’s also in the DFZ before firing up the mailing lists. What’s more, route collectors are not necessarily configured in the same way as standard peers, depending on the operator. This remains a mistake on our part, which has only resulted in horrors on the monitoring route collectors and not on the DFZ routes. So Route Collectors don’t behave at all like transit or peering, given the lack of max pref, prefix-list or RPKI.

But hey, it’s quicker to send a flaming mail before typing your show ip route, I agree :wink:

My 2 cents,