Shi et al. (2026) "Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero’s P2P Network!": https://arxiv.org/pdf/2609.10260
nevent1qqs8h38gkfa38apr90ysuqkayn2c9uk5y9awc5kvn7auycgxhhu04lcxusw9h
Login to reply
Replies (1)
rucknium presented a minimal PR intended to give the network a defense against the Nyx attack without prolonged experimentation. The PR does not enlarge the gray_list; participants noted that gray_list entries are not initially connectivity-checked and that multiple ports per IP are allowed, making adversarial filling easy. ofrnxmr’s work-in-progress changes bucket both lists and prevent direct injection into the white list; the patches are stacked and will be tested separately and together. White-list housekeeping is currently only “evict oldest over limit”; rucknium suggested adding proper eviction-on-no-response behavior and possibly housekeeping on every insertion to mitigate temporary VPN-based peer injection. jeffro256 proposed tracking the originator of each peer-list entry and evicting by originator proximity (ASN sorting as one possible ordering). Caching of peer lists was suggested to reduce CPU cost of repeated reshuffling and to limit enumeration attacks. IPv6 was flagged as difficult to secure without something like proof-of-storage. plowsof reported that roughly 60% of reachable nodes appear suspicious and that an adversary already controls more than 2500 addresses according to observatory data. jeffro256 viewed the current PR changes as almost strictly better in most scenarios (speed impact still to be measured). rbrunner anticipated a prolonged period of competing approaches. Ban-list updates will require a reliable fingerprint with negligible false-positive risk.
rucknium: I made this PR:
. There are some details and possible additions I want to discuss.
rucknium: But first I want to give anyone the opportunity to respond to it as-is.
rucknium: 1. In the PR I don't increase the gray_list size. Maybe it would make sense to do that. Increasing the gray_list isn't a way to defend against the Nyx attack, but maybe it's a good idea to increase it to defend against other attacks.
ofrnxmr: My changes bucket both lists
rucknium: It's hard to prevent an adversary from filling the gray_list with its own peers. They are not (initially) checked for connectivity by the node. And the node allows multiple ports on the same IP address.,
ofrnxmr: max sizes 4096 white and 8192 gray
ofrnxmr: 256 buckets of 16 and 32
rucknium: I thought maybe the node could hold only one port per IP address in the gray_list, but that would allow an adversary to bump an honest node out of the list by feeding false ports for an honest IP address.
ofrnxmr: I think blindly increasing sizes has diminishing returns under peerlists that are not static and all online
rucknium: It seems obvious that a node with a wide view of the network would be harder to eclipse than a node with a narrow view of the network.
ofrnxmr: Lists fill will dead nodes amd are never evicted
rucknium: Sometimes things that seem obvious are not actually true.
vtnerd: but if this were ipv6 mode, couldn’t the attacker just spam with valid ips and still eventually replace the valid peers?
ofrnxmr: Its not a biew of the network is 40k/45k are dead
ofrnxmr: If* the
rucknium: That's one of the additions that I think could help. I assumed that the white_list had a housekeeping behavior and evict-on-no-response behavior like the gray list does, but it doesn't. Probably some housekeeping for the white-list should be added.
ofrnxmr: No, its blindly evicting oldest over limit
ofrnxmr: I added
rucknium: I don't see how IPv6 will be safe until something like proof-of-storage, with all its downsides, is deployed.
rucknium: I mean widespread use of IPv6, maybe as the default.
jeffro256: An simple-ish partial solution I've been thinking about is to, for each peer list entry, keep track of the node which told you about this entry (originator). When the limit is hit, instead of erasing based on insert time, erase the entry with the closest originator
rucknium: I think it could even make sense to do white_list housekeeping whenever a new peer is added to the white_list. Why? To prevent this augmentation of the Nyx attack:
ofrnxmr: Part of bucketing additons added
jeffro256: Is there a PR somewhere?
ofrnxmr: No, still wip
rucknium: An adversary can temporarily control an IP address with open ports by using a commercial VPN that permits port opening. Then, the adversary could put a lot of peers into the white_list that would respond to a temporary ping-pong. So the adversary does not even need permanent control of all the adversary IP addresses.
ofrnxmr: A bunch of changes stacked on too of each other, want to test combined and separate before concluding x is better than y
rbrunner: Looks like we may enter a quite prolonged phase of experimentation and competing approaches until things finally settle ...
ofrnxmr: Also changed that they cant inject into white
rucknium: But what if an adversary controls 1,000 IP addresses, like in the Nyx attack paper?
ofrnxmr: According to plowsof's obaervatory, they control over 2500
jeffro256: Sorting originators by ASN may help in this case. The originator order function could be anything, really
rucknium: rbrunner: I want to avoid that because this is a vulnerability that the network has no defense against yet.
rbrunner: Sounds reasonable
plowsof: 60% of reachable nodes are implicated in being sus
rucknium: Which is one of the reasons why I went for minimal changes.
boog900: semi-related to the discussion:
jeffro256: AFAICT, Ruck's change are almost strictly better in most cases that I can think of (except perhaps maybe speed). I need to check the time signature of the output selection algo again. Because an effectively ~1950x bigger list might have >1950x impact on speed, depending on time signature
plowsof: jeffro the problem is if everyone would use ASN - and the adversary defeats it, its possible that we would be worse off. 9/10 ai research papers agree that its great fix if me, you and a handful of others use it but not the whole network
boog900: To update we will need to coordinate signatures again ...
syntheticbird: it wasn't that bad
syntheticbird: I was just late of one month
plowsof: said ai research
rucknium: Did you see my reason for picking the specific white_list size limit? I am not committed to that value, but at least it has a non-arbitrary reason for it, instead of the reason being humans having 10 fingers.
boog900: no but they will just switch IPs again I don't really want to keep this cat and mouse going forever
ofrnxmr: yes
rucknium: I would be OK with lowering alpha. Or some other reasonable value.
jeffro256: The code can afford to be really inefficient when operating on a list of size 20
rucknium: The third complementary change (the first two were gray_list size and whte_list housekeeping) would be caching the peerlist that peers send to each other. It would:
rucknium: 1. Prevent the CPU load of re-shuffling the white_list every time a (potential adversary) peer requests it
rucknium: 2. Prevent white_list "enumeration" by an adversary, where the adversary can get the whole contents of the white_list by just requesting it many times.
rucknium: By the way, some of these additional suggestions come from the LLM review of my PR (
) and the Shi et al. research team.
rucknium: I mean, the LLM review told me that the white_list doesn't have housekeeping, which I should have realized before.
rucknium: Where do we go from here?
rucknium: You or I could run the Nyx simulation with your patch once you finish it.
rucknium: AFAIK, the code was already, in theory, processing the whole gray_list in that section of code, which can get up to 5000 records. It just limited the white_list to 20.
rucknium: I think there has to be a good fingerprint with negligible risk of false positives if the ban list is to be updated.
rucknium: Oops. Meant to ping boog900 in that message.
rucknium: Discussion can continue in the PR and on Matrix/IRC
rucknium: More on this topic for now?
jeffro256: Caching and algo improvement, I guess. I need some more time to think about it. > Where do we go from here?
ofrnxmr: Sounds good > You or I could run the Nyx simulation with your patch once you finish it.
ofrnxmr: (but its ~8 patches fwiw)
rucknium: I need to fix these monerosim config files by the way. Accidentally got the ones for the 26 hour run instead of the 50 hour run:
View quoted note →
GitHub
p2p: increase white_list size and change connection selection algorithm by Rucknium · Pull Request #11527 · monero-project/monero
TL;DR: An adversary with 1,000 or more IP addresses under its control can pack its IP addresses into honest nodes' peer lists and successfully ...
GitHub
suggested additions · Issue #18 · Boog900/monero-ban-list
31.58.215.0/24 87.229.79.0/24 143.20.137.0/24 66.92.61.2 69.17.52.1 (which is also in issue #16 69.17.52.10 104.234.118.2 also abuse coming from Pe...
GitHub
crawler-research/Monero-Peer-AS-Deduplication/pdf/monero-peer-asn-deduplication.pdf at main · plowsof/crawler-research
Contribute to plowsof/crawler-research development by creating an account on GitHub.
GitHub
Review: monero-project/monero#11527 @ 903d6b555e86 · Issue #1088 · xmrack/monero-review
Security review of p2p: increase white_list size and change connection selection algorithm Result: 3 findings: 1 MEDIUM, 2 LOW · 2 introduced here...
GitHub
misc-research/Monero-Eclipse-Attack/nyx/code/monerosim-config at main · Rucknium/misc-research
Small research projects. Contribute to Rucknium/misc-research development by creating an account on GitHub.
#monero-research-lab