HomeIncident library
Incident library

INCIDENT REPLAY · 2017-08-25

Google leaks more-specifics to Verizon (Japan outage)

At 03:22 UTC Google leaked more than 160,000 prefixes to Verizon in Chicago, many of them more-specifics of other networks' space, including over 25,000 of NTT OCN's. KDDI and IIJ, as Verizon customers, accepted over 95,000 each, so Japanese domestic traffic headed for Chicago. Corrected within minutes, but services in Japan were disrupted for hours.

Route leakLeak readvertisementCatalog confidence: high

A reconstruction, not the operator's network. Hammerhead's authors rebuilt the change from the public record as a small network of their own. No configuration here is the operator's. What follows is what the tool reported on that reconstruction; it is not a claim that the tool would have prevented the outage.

What happened

Who: Google (AS15169) caused; Verizon (AS701) accepted and propagated; NTT OCN, KDDI, IIJ and Japanese users hurt

The summary at the top is paraphrased from the sources below. Hammerhead's catalog rates its confidence in this account high.

Sources

What the public record says

  • At 03:22 UTC Google leaked more than 160,000 prefixes to Verizon in Chicago, many of them more-specifics of other networks' space, including over 25,000 of NTT OCN's. KDDI and IIJ, as Verizon customers, accepted over 95,000 each, so Japanese domestic traffic headed for Chicago. Corrected within minutes, but services in Japan were disrupted for hours.
  • Who: Google (AS15169) caused; Verizon (AS701) accepted and propagated; NTT OCN, KDDI, IIJ and Japanese users hurt

What the reconstruction assumed

Decided by Hammerhead's authors to build a network that runs. None of it describes the operator's network.

Everything below is ours.

  • The family template (tools/replays/gen_family_leak.py): GOOGLE-1 (AS 15169) learns the victim's routes over a direct peering with the victim, and VERIZON-1 (AS 701) prefers what GOOGLE-1 sends it and filters nothing on that session. A transit carries the victim's routes to the same network the ordinary way, and a third party sits behind it.
  • The edit is assumed. Before the change GOOGLE-1's export toward VERIZON-1 is a prefix-list permitting its own block; after it, the list is no longer attached. No operator in this family has published what was changed.
  • Cisco IOS syntax, and stand-in address space except where a prefix is on the record.
  • The accepting network's other sessions are filtered and limited, so that the audit's finding is about this session and not the template.
  • Many of the leaked routes were more-specifics Google used internally for traffic engineering, which is not modelled; the leak of what it learned from its peer is.

What Hammerhead said

Each run below is pinned by Hammerhead's replay tests at commit 0ee9732: the verdict and the quoted text are checked against the tool's output, and a run that disagrees fails the test. Reason lines are copied verbatim; a pinned line can be part of a longer reason. Every preflight run uses --max-blast-radius 100, so a change is never counted as caught merely because it touched many devices of a small network.

every network in one snapshot

now: fail

On the reconstruction, hammerhead preflight on a before and an after snapshot returned fail, for the reasons below.

Reason text pinned (some reason contains each line)

  • GOOGLE-1 would announce to 172.16.3.1 (AS 701) routes it learns from other networks
  • path change:
  • now through AS 15169

First run, before the fixes this batch prompted

first: fail
  • forwarding-semantics gate: 1 device(s) changed configuration that reroutes or rewrites traffic without changing any route or ACL — GOOGLE-1: BGP peerings and their policy attachments pass --allow-flow-changes to accept it
  • path change: 1 destination(s) still delivered would be reached through a network their traffic does not cross today — 198.18.0.0/24 on OCN-1 (198.18.0.1) from VERIZON-1, X-1 now through AS 15169. No flow is lost, which is what a route leak looks like from inside the model. Pass --allow-path-changes if moving this traffic is intended

Pinned in tests/gauntlet/replays-leak.yaml.

GOOGLE-1's own router

now: fail

On the reconstruction, hammerhead preflight on a before and an after snapshot with --allow-flow-changes returned fail, for the reasons below.

Reason text pinned (some reason contains each line)

  • GOOGLE-1 would announce to 172.16.3.1 (AS 701) routes it learns from other networks

First run, before the fixes this batch prompted

first: pass

Pinned in tests/gauntlet/replays-leak.yaml.

VERIZON-1's own router, no change: the audit

now: fail (finding)

An audit of one snapshot, no change: hammerhead nqe-library --mode run, check nqe.bgp.ebgp_peer_unfiltered_and_unlimited. Pinned status: fail, a finding whose evidence contains the text below.

Evidence pinned (the evidence contains each line)

  • VERIZON-1
  • 15169

First run, before the fixes this batch prompted

first: pass (no finding)

Pinned in tests/gauntlet/replays-leak.yaml.

Limits

  • The configurations are a reconstruction. Where the record is silent, the assumptions above fill the gap; if one is wrong, the replay is wrong in that respect.
  • A verdict describes the reconstructed network's converged state. Timing, traffic load, and anything a device does that no configuration states, such as software reacting to a health check or a firmware bug, are outside it.
  • A pass is reported as a pass. A fail says a gate fired on the reconstruction, nothing more.
  • Why the catalog counts this outage as in scope: A change that lets peer-learned and internally generated more-specifics out over a peering session is a before-and-after advertised-routes diff; Verizon's session had no max-prefix able to stop 160,000 routes.

Imported from docs/replays/google-verizon-2017/FACTS.md and the replay manifests in the Hammerhead repository at commit 0ee9732.