# Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero's P2P Network!

Crypto.world

> Prior eclipse attacks on Bitcoin, Ethereum, and Monero all assume the adversary can open an inbound connection to the victim, which excludes nodes behind NAT or firewalls — a class widely assumed to…

**arXiv**: 2609.10260

**Published**: 2026-09-09

**Authors**: Ruisheng Shi, Jiaqi Zeng, Lina Lan, Shihan Zhang, Bing Han, Xiapu Luo, Qishu Jin, Wenliang Du, Qin Wang

## AI summary

## Overview

**Research area:** Blockchain / cryptocurrency peer-to-peer (P2P) network security, specifically eclipse attacks on Monero nodes that sit behind NATs or firewalls ("unreachable" nodes).

**Technical level:** Advanced — the paper assumes familiarity with P2P peer discovery, NAT behavior, and Monero's internals (peerlists, timed sync, connection refresh). The prose is accessible, but the mechanism-level reasoning is dense.

**Scope:** The paper designs, models, and empirically evaluates the first eclipse attacks that fully take over all 12 outbound connections of *unreachable* Monero nodes without ever connecting to the victim, using indirect peerlist poisoning of reachable relay nodes and Monero's own outbound connection refresh logic.

## What This Paper Is About

Prior eclipse attacks on Bitcoin, Ethereum, and Monero all assume the adversary can open an inbound connection to the victim, which excludes nodes behind NAT or firewalls — a class widely assumed to be *more* secure because they accept no unsolicited connections. The authors argue the opposite: because unreachable nodes learn peers and pick connections exclusively through their outbound peers, compromising those outbound links yields an isolation state with no protocol-level recovery path. The goal is to show that unreachable nodes can be driven into complete, persistent eclipses across their entire lifecycle — both as long-running nodes and as brand-new nodes joining the network.

## Key Contributions

1. **Structural vulnerability analysis:** The paper identifies weaknesses in Monero's peerlist propagation and connection management — graylist deduplication only by ⟨IP, port⟩, whitelist insertion via PING/PONG validation, graylist-first replacement during outbound refresh, and seed-node bootstrapping — that disproportionately affect outbound-only nodes.
2. **First complete eclipse of unreachable Monero nodes:** The authors present an attack achieving full takeover of all 12 outbound connections without requiring inbound access to the victim, combining *indirect peerlist poisoning* (malicious records planted in reachable nodes and relayed to the target by timed sync) with *protocol-driven connection rotation* rather than adversary-triggered resets.
3. **Two lifecycle attacks:** *Nyx* targets existing unreachable nodes via network-wide poisoning; *Moros* targets newborn unreachable nodes by poisoning the six hardcoded seed nodes so a single bootstrap handshake delivers an attacker-dominated candidate set.
4. **Dual evaluation plus countermeasures:** *Nyx* is validated on a 1,200-node SEED Emulator network calibrated to mainnet latency, jitter, and loss; *Moros* is demonstrated on the Monero mainnet against a controlled target. The paper also discusses enabling protocol mechanisms and proposes mitigations, and reports the vulnerabilities to Monero via HackerOne (identifier #3702913).

## Main Findings

- **Unreachability does not imply resilience:** The authors challenge the common assumption that NATed nodes have a reduced attack surface, showing the opposite — the absence of inbound connections prevents honest nodes from reconnecting to break an eclipse, creating a "self-healing dilemma" that makes isolation stable.
- **Port diversity defeats graylist deduplication:** Reusing a single announced port across all reachable nodes would collapse injected records into at most 1,000 distinct ⟨IP, port⟩ entries, dropping the injection rate to zero after round 1; assigning each reachable node a dedicated port lets all 12 × 250 = 3,000 records per round pass deduplication.
- **Theoretical occupation bound of ≈98.8%:** Because each reachable node's 12 stable outbound-connection records have `last_seen` refreshed every 60 s, at most 1,000 − 12 = 988 of 1,000 whitelist slots can be occupied, giving OR_max = (1,000 − 12)/1,000 ≈ 98.8%.
- **Malicious graylist fraction of at least ≈97.1%:** The benign-record upper bound is B_max = 12² = 144 (the 12 outbound peers each draining a 1,000-entry whitelist via the `sent_addresses` mechanism), giving IR_min = (5,000 − 144)/5,000 ≈ 97.1%.
- **Initial takeover probability ≥ 87.4%:** With M = 1,000 malicious candidates across 1,000 distinct /24 subnets and B ≤ 144 benign candidates, P₀ ≥ 1,000/(1,000 + 144) ≈ 87.4%; convergence requires only P_k > q_k = k/12, and the boundary case k = 11 satisfies P₁₁ → 1 > 11/12.
- **Poisoning scales quickly in emulation:** In the 1,200-node network, one complete filling round over 1,199 reachable nodes in 10 batches of 120 (30 s per-batch timeout) takes about 5 min; at T+5 min, 89.6% (1,074/1,199) of reachable nodes reached ≥95% malicious whitelist occupation, and by T+10 min 99.5% (1,193/1,199) did.
- **Measured occupation matches theory, with measured decay:** Median whitelist occupation across reachable nodes reached 98.5% at T+10 min, matching the 98.8% bound; over monitoring lasting 18.2 h, median occupation declined to 97.2% at T+1020 min and the mean from 98.16% to 96.90%, which the authors attribute to routine timed sync redistribution among reachable nodes — motivating continuous re-filling.
- **Full eclipse times:** *Nyx* reaches full takeover within 27 min and *Moros* eclipses newborn unreachable nodes within 8 s, per the paper's stated contributions.
- **A large malicious peer pool:** The port-diversity design leaves the attacker holding roughly 1,000 × 3,000 = 3,000,000 distinct ⟨IP, port⟩ pairs, about 600× the graylist's 5,000-slot capacity.
- **Mainnet feasibility of eclipse-at-birth:** *Moros* was demonstrated on the Monero mainnet against a server running with `--hide-my-port` to simulate unreachability, using 1,000 public IPs against the six official seed nodes; graylist poisoning was omitted because a newborn's graylist is limited to the 250 records from a single seed handshake.
- **Evaluation limits in the supplied content:** The paper text provided is truncated mid-way through the persistence analysis of N-I; later results (e.g., full N-II/N-III emulation curves, detailed mainnet timelines, and the countermeasure evaluation) are not reported here and are therefore not summarized.

## Methodology in Plain English

The authors first reverse-engineer the parts of Monero's networking that matter: each node keeps a whitelist of up to 1,000 verified peers and a graylist of up to 5,000 unverified candidates; peers exchange up to 250 records per handshake or timed sync; and every ~101 s a node drops one outbound peer and picks a replacement, preferring the graylist. Newborn nodes with empty lists ask a hardcoded seed node for peers and build their first 12 connections from that single response.

The attack chain uses these rules rather than fighting them. First, the attacker floods reachable nodes' whitelists with 1,000 attacker IPs (one per /24) that complete PING/PONG validation, while separately flooding their graylists with 5,000 randomly generated, non-connectable "trash" records so that graylist housekeeping wastes its promotion slots. Crucially, each reachable node is told a different port for the same attacker IPs, so when a victim later receives those records they look unique and survive deduplication.

Second, when the unreachable victim runs its routine timed sync with its 12 outbound peers, it pulls in up to 3,000 attacker records per round, filling its graylist with malicious entries; because it cannot accept inbound connections, nothing can be written directly. Third, the victim's own scheduled connection-refresh cycle replaces benign peers with malicious ones — each malicious peer keeps returning fresh malicious records, and reports the genesis block so it stays classified as `state_normal` and keeps the refresh trigger alive. The authors model this as a probabilistic process over k (number of attacker-controlled outbound slots) and show it converges.

For newborn nodes, the same filling procedure is aimed at only the six seed nodes, so the first handshake a new node performs returns an almost entirely malicious peerlist, and its initial 12 connections are drawn from it.

To test this without harming real users, the authors built a 1,200-node Monero v0.18.4.3 network on the SEED Emulator spanning 100 ASes with BGP routing, injected mainnet-calibrated delay, jitter, and packet loss, firewalled the target to block inbound connections, and implemented malicious peers with py-levin, sampling the target's peerlists and connections via RPC every 30 s. *Moros* was additionally run on the mainnet against their own controlled target, with the target's /24 diversity restriction disabled so their narrow set of public IPs could be selected.

## Why This Matters

**Impact on research:** The paper broadens eclipse attacks from "nodes with public IPs" to the much larger population of outbound-only nodes, and reframes unreachability as an *amplifier* of eclipse durability rather than a defense. It also introduces a new attack primitive — transitive peerlist poisoning through honest relays — that likely generalizes to other P2P systems that rely on outbound-only peers (Tor bridges, Dandelion-style relays, other cryptocurrency clients).

**Real-world applications and implications:**

- **Wallet and node operators:** Users running Monero nodes at home behind consumer routers should not assume NAT protects them from network-level isolation; the same setup is exactly what *Nyx* and *Moros* target.
- **Chain privacy and deanonymization risk:** An eclipsed node's view of the network is attacker-controlled, enabling follow-on deanonymization, network partitioning, and selfish-mining style advantages, which matters especially for a privacy-focused currency.
- **Protocol hardening for client developers:** The findings point to concrete fixes — seed-node response diversity, per-subnet limits on peerlist insertion, and less predictable outbound refresh.
- **Measurement and monitoring communities:** The work shows that "invisible" NATed nodes are a silent majority with distinct failure modes that external crawlers cannot observe, motivating new monitoring approaches.

**Industry relevance:** Any deployed P2P client with a bootstrapping seed list and graylist-first peer selection shares the structural pattern exploited here. The responsible disclosure to Monero (HackerOne #3702913) means node operators and maintainers can act on the findings directly.

## Future Directions

- **Robust mitigations and their measured cost:** The paper proposes countermeasures but the supplied content does not report their evaluation; quantifying their effectiveness against both *Nyx* and *Moros*, and their overhead on normal peer discovery, is the obvious next step.
- **Generalizing beyond Monero:** Does the three-phase model (network-wide poisoning → indirect contamination → protocol-driven takeover) transfer to other outbound-only P2P designs, and what structural features make a network immune?
- **Detecting sustained poisoning:** The attack requires continuous re-filling because occupation decayed from a median 98.5% to 97.2% over the 18.2 h monitoring window — whether reachable nodes or seed operators can detect such persistent mass handshake traffic remains open.
- **Long-run stability under adversarial and defensive churn:** Only partial stability data appears in the supplied text; long-horizon measurement of eclipse duration under differing re-filling rates, network sizes, and partial defender adoption is unresolved.

## Target Audience

Blockchain protocol researchers and Monero contributors, P2P networking and anonymity researchers studying eclipse and partitioning attacks, node-operator and wallet-infrastructure engineers who need to understand realistic NAT threat models, and graduate students in security who want a worked example of translating protocol quirks into a formally modeled, empirically validated attack chain.

## Authors’ abstract

Eclipse attacks isolate a blockchain node by monopolizing its network connections. Existing attacks on Monero (NDSS'25), Bitcoin (USENIX'15/21, S&amp;P'20) and Ethereum (WWW'26) implicitly assume that the adversary can establish inbound connections, thereby excluding a large and practically dominant class of nodes: \textit{unreachable nodes} operating behind NATs. Such nodes are widely believed to enjoy stronger networks. We challenge this assumption and show that…

[The full abstract on arXiv](https://arxiv.org/abs/2609.10260)

[Read the original paper](https://arxiv.org/abs/2609.10260)

---

Canonical: https://crypto.world/paper/are-unreachable-nodes-truly-safe-fully-eclipsing-monero-s-p2p-network-2609.10260
