Rogue objects in routing databases

Mailman List Mirror (Read Only)
1

It appears that there is currently an influx of rogue route
objects created within the NTTCOM and RaDB IRR databases, in
connection to Quadranet (AS8100) and China Mobile
International (CMI).

Examples of affected networks are:

193.30.32.0/23
45.129.92.0/23
45.129.94.0/24

Networks, which have seemingly no affiliation with
Quadranet, nor China Mobile International (CMI), which
merely appears to be an upstream of Quadranet and hence
creates the route objects in an automated manner.

Another person has already reached out to Quadranet to find
out the root cause of the creation of these objects. Their
support gave an ETA of 24-72 hours.

The route objects are all identical:

route: 193.30.32.0/23
descr: CMI (Customer Route)
origin: AS8100
mnt-by: MAINT-AS58453
changed: qas_support@cmi.chinamobile.com 20200117
source: RADB

There appears to be a correlation with the affected
networks, a fair share of them is part of AS-SBAG, which in
turn is part of AS-VMHAUS, which in turn is part of AS-
QUADRANET and could yield the importing of these prefixes.
AS-VMHAUS appears to be a customer of Quadranet, listed
within AS-QUADRANET-CUSTOMER-ASSET.

These networks do however have no direct connection to
Quadranet, and are not affiliated with Quadranet, nor are
currently connected to Quadranet, which, entirely ignoring
that the `origin` points to Quadranet, makes the route
object illicit.

Basically this has given AS8100, whether that be
legitimately Quadranet, or somebody impersonating/spinning
up a rogue AS8100, theoretical control over a massive amount
of prefixes, as these can be advertised without restrictions
and very likely reach a fairly high percentage of global
visibility.

2

Hi!

This came up on our radar somewhere in the last 24 hours too. It indeed
does look very curious. Thank you for your analysis and report.

NTT is taking steps to figure out what is behind this. Our current
working theories are that perhaps the IRR maintainer account was
compromised, or some kind of automation script gone rogue, or perhaps
there is adverserial intent and this is stage setting.

I'm not sure we will be able to report our findings back to this group,
but we are actively investigating.

Kind regards,

Job

3

Hi Florian, NANOG,

While the symptom of (automatically) proxy registered route objects is problematic, perhaps we could also take this opportunity to discuss the underlying issue: we as an industry appear to place our trust in various IRR sources operated by entities that either can’t or don’t validate whether the actual owner of the involved resource approves the creation of the IRR database object.

We should start to push our customers to maintain their route origin information in databases operated by the RIR or NIR which assigned the resource, or even through RPKI ROAs that were optionally converted into IRR route objects for the ease of consumption. It’s also time for the RIRs to take their responsibility in this matter by facilitating services like IRR, RPKI, PTR, etc for legacy IP space under conditions which are palatable to corporate lawyers, if they haven’t already done so.

Finally, there doesn’t have to be a global “flip the switch” day where we decide to stop trusting 3rd party databases, but even if we start holding ourselves to a higher standard one customer at a time that’s still going to have the potential to make a big difference a couple of years down the road.

Best regards,
Martijn Schmidt

PS, a small disclaimer: none of the above are new ideas, nor did I come up with them myself - but it still makes sense to work towards implementing them…

4

Hi Martijn,

albeit a negligible amount of edge cases it can
indeed be stated that there is too much trust put
into alternative IRR sources operated by third
partys not affiliated to RIRs. Generally, usage of
such databases however is not mis-used in a larger
scope, and the complexity involved with creating
route objects (AltDB for example validates new
MAINTAINERS, RaDB charges) diminishes the vector
in a (barely influental) manner. An option to
combat this would perhaps be to run validation at
regular intervals, and brings invalid objects to
the attention of operators. I did similar for this
incident just a moment ago with a batch of self-
written scripts.

After further analysis, there are over 5390 IPv4
prefixes pointing with their origin towards
AS8100: https://pastebin.com/Zh1YZfEq

Out of 5390 prefixes, 2287 are currently not even
visible within the global routing table:
https://pastebin.com/cSepb7yS

Another 2714 prefixes are INVALID, that in
particular means that AS8100 is neither *within*
the announcing AS-PATH, nor originating the
prefix: https://pastebin.com/JhaxVeN0

Last but not least, there is 389 VALID prefixes
(in this case, perhaps only technically valid, as
I did probe for AS8100 within the AS-PATH
sequence, and not if AS8100 actually originates
the prefix): https://pastebin.com/UVt6nwGz

That's a conceivable 5001 IPv4 prefixes for a
potential bad actor right there. It can also
clearly be stated that, while initially mentioned,
the significance of ascendence caused by AS-SBAG
is negligible, as it appears, the entirity of
Quadranet and affiliates is affected.

Regards,
Florian Brandstetter

5

a message of 53 lines which said:

Examples of affected networks are:

193.30.32.0/23
45.129.92.0/23
45.129.94.0/24

Note that 193.30.32.0/23 has also a ROA (announces by 42198). So,
announces by AS8100 would be RPKI-invalid.

45.129.92.0/23 also has a ROA. Strangely, the prefix stopped being
announced on sunday 26.

45.129.94.0/24 has a ROA and is normally announced.

So, if AS8100 were to use its abnormal route objects , announces would
still be refused by ROA-validating routers.

6

Hi Stephane, NANOG –

Do the math for all pertained prefixes in the pastes, those 3 prefixes were just examples I had at hand,
and the event is still of quite some significance. Albeit ROA-validating routers being an argument that
extenuates probabilities and the ensuing effect, deployment of such still lacks, hence my mention of
reaching levels of (random guess) 90% global visibility still, taken the attacker understands ROA.

It is certainly unlikely that networks that are known for rather puerile filtering, or lack of adequate filtering
to filter the networks, so ultimately they will inevitably still transpire in the global tables. An impression
emerges that commitment in resolving this incident lacks, apart from the guys over at NTT which,
from what I gathered, suspended their IRR account temporarily to prevent further damage.