{"type":"rich","version":"1.0","title":"MoneroResearchLab wrote","author_name":"MoneroResearchLab (npub1m0…gcfay)","author_url":"https://yabu.me/npub1m0ner0mtjgs74aghus7qxrxl7mqv06d4v8rz9pmnux2dpr4cmhmsjgcfay","provider_name":"njump","provider_url":"https://yabu.me","html":"rucknium could not run reproduction code but finds the attack plausible: poison peerlists so unreachable nodes only connect to adversary peers. Suspected flaws: inbound connections pace whitelist inserts; 1,000-peer whitelist (from CryptoNote https://github.com/monero-project/monero/blame/9e3a31032ee2cf3cb65c908e107a9952d03bbc4f/src/cryptonote_config.h#L138-L139). gingeropolous’s Shadow/monerosim runs partly confirm; unmodified monerods only filled 7/12 slots, hyper-injection still running. SEED Emulator used. boog900: weak eager whitelist; look at Bitcoin addrman https://bitcoincore.academy/addrman.html https://github.com/bitcoin/bitcoin/blob/master/doc/reduce-traffic.md ; Cuprate similar, planned change. Responsible disclosure via HackerOne; prior log https://libera.monerologs.net/monero-research-lab/20260913#c707269 . Prior Shi 2025 NDSS paper [44] already discussed whitelist mitigations unadopted. rbrunner’s /24 dedup PR https://github.com/monero-project/monero/pull/9939 raised attack cost. Ideas: 10x whitelist, inbound not writing whitelist, inbound/outbound insert pairing, incoming connection limits, only evict unreachable from whitelist. Eclipse impact: spy on tx origin or isolate nodes (majority unreachable, often GUI). Continue next week.\n\nrucknium: 8. Shi et al. (2026) \"Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero’s P2P Network!\" (https://arxiv.org/pdf/2609.10260)\nrucknium: Finally here. And we have held on to rbrunner !\nrbrunner: Indeed.\nrucknium: Probably best to have an initial discussion, with plans to continue in more depth next week.\nrucknium: Is boog900 here?\nboog900: Hi\nboog900: yes\nrucknium: I am skeptical of the paper because I cannot get their reproduction code to run. However, I find their theorized attack to be plausible, given monerod's design decisions.\nrucknium: Basics: An eclipse attack happens when an adversary (who wants to do something bad) completely isolates one or more honest nodes from the rest of the network. The honest node thinks it's connecting to other honest nodes, but actually all of its peer connections are to a node (or pseudo-node) controlled by the adversary.\nrucknium: So they say they can eclipse unreachable nodes (nodes that are behind firewalls so they do not accept inbound connections) by poisoning the peerlists of honest nodes. The honest nodes inadvertently give the targeted unreachable nodes the peerlists of the adversary.\nrucknium: I think it's caused by two design decisions that may be design flaws:\nrucknium: 1. Allowing inbound connections to control the pace of adding new IP addresses to honest nodes' peerlists, specifically their whitelists.\nrucknium: 2. Limiting the white list to 1,000 peers.\nrucknium: If you fix one or both of those, you would fix the attack probably. If the attack actually works.\ngingeropolous: so far my replication attempts are confirming the findings, but I'm running into issues with shadow. Currently running what I hope is the final run.\nrucknium: The paper uses \"SEED Emulator\", which is supposed to be simulator to Shadow.\nboog900: monerod does have a very weak address book, it is very eager to swap out peers in the white list\nrbrunner: Well, that magic number of 1000 sounds like a limit set in earlier time when RAM was smaller than today. Maybe no problem to lift that\ngingeropolous: Interestingly, the initial run of the entire 2.2k nodes used real monerods, and was only able to get 7/12 occupied. So if the attack was mounted using un-modified monerods, our protocol holds. But the peer list hyper injection that they did is different, and that sim is running.\nrbrunner: you mean special nodes that \"cook up\" special peer lists, and fast?\nboog900: Cuprate has this issue too, but that is planed to be changed. IMO we should look at Bitcoin's address book which is much more resilient\nrucknium: The 1000-peer limit in the white list was inherited from the original cryptonote codebase: https://github.com/monero-project/monero/blame/9e3a31032ee2cf3cb65c908e107a9952d03bbc4f/src/cryptonote_config.h#L138-L139\ngingeropolous: essentially a synthetic node - a python process that pretends to be a monerod, does the p2p dance, and then sends bad peers\nrbrunner: I see\nrucknium: boog900: The paper says that bitcoin's eviction design may be more vulnerable by nature. But they probably have patches to make it less vulnerable.\nrbrunner: Without looking into this closely, I find it kind of counter-intuitive that this should fully succeed, but may there is some re-inforcement, or runaway effect in play, until the daemons are totally poisoned\nrbrunner: I think simulations are very valuable here\nrucknium: My instinct is to just raise the white list by 10x. I don't know if that would have downsides. I haven't read all the bitcoin and ethereum eclipse papers. Maybe there is something valuable there. Or cautionary about raising the list size limit too much.\nrbrunner: Also for later testing any changes, it might be able to inadvertantly make things worse ...\nrucknium: Ideally, if there are X% adversary nodes, only X% of honest nodes' white lists should be composed of adversary node IP addresses.\nrucknium: But they can push that up by quickly connecting to honest nodes from 1,000 adversary \"nodes\". Then it fills their white list.\nabsinthium: unrelated, did the researchers that wrote the paper make any attempt to responsibly disclose their findings to us prior to releasing it?\nrucknium: absinthium: Yes. They submitted a HackerOne report. They state that in the paper, too.\nboog900: I'll have to read what they said there, but Bitcoin does not give up spaces in its address book as easily as monerod which should make it slightly more vulnerable at start up but less once running with a full peer list. \u003e rucknium: boog900: The paper says that bitcoin's eviction design may be more vulnerable by nature. But they probably have patches to make it less vulnerable.\nrbrunner: Formulated this way, this sounds surprisingly logical :)\nrucknium: absinthium: See https://libera.monerologs.net/monero-research-lab/20260913#c707269\nplowsof: https://bitcoincore.academy/addrman.html\nrucknium: Say that inbound connections do not get added the to the white list. Would that have an unacceptable affect on new nodes getting added to the \"network-wide\" white list?\nrucknium: Or say that for every peer added to the white list by in inbound connection, you would have to wait for one to be added by an outbound connection (i.e. through graylist promotion). Would that slow down the pace of the adversary packing its IPs into white lists?\nrbrunner: Don't know, my gut feeling tells me that with drastic changes like these you can easily make things worse. But not really much more than a gut feeling.\nrbrunner: Anyway, I guess things also have to work in edge cases, like only seed nodes ready to accept incomming connections and everybody else closed against that\nrucknium: There is a balance between letting the honest nodes establish sufficient connectivity and not letting the adversary nodes establish excessive connectivity.\nrucknium: The paper says\nrucknium: \u003e The whitelist-filling primitive is reused from [ 44 ], and its mitigations, such as blocking whitelist insertion from incoming connections or increasing whitelist capacity, have already been discussed but remain unadopted by Monero.\nboog900: that is for the incoming connection limit from what I can see \u003e rucknium: boog900: The paper says that bitcoin's eviction design may be more vulnerable by nature. But they probably have patches to make it less vulnerable.\nrucknium: [44] is Ruisheng Shi, Zhiyuan Peng, Lina Lan, Yulian Ge, Peng Liu, Qin Wang, and Juan Wang. 2025. Eclipse Attacks on Monero’s peer-to-peer Network. In Network and Distributed System Security (NDSS) Symposium.\ngingeropolous: i gg, but will post results when i have them\nrucknium: Thanks, gingeropolous\nboog900: monerod doesn't have a limit on incoming connections, so an adversary can't fill all incoming slots. But this isn't Bitcoin's address book\nboog900: Although IMO monerod should also have an incoming connection limit just for DoS prevention\nrucknium: IIRC the discussion about Shi et al. (2025) went like this: Vulnerabilities were patched, but the bigger discussion was not instigated because there was little to worry about after the patches.\nrucknium: boog900: I see. They mean \"eviction\" as in connection eviction, not address book eviction. I was confused because they mention Ethereum's incoming-connection limits like it is something different from what bitcoin does.\nrucknium: \u003e s. In Monero, for example, the lack of\nan eviction mechanism [44 ] prevents an attacker from actively\nexpelling honest connections, and the absence of a strict cap on in-\ncoming connections makes it difficult to block new honest incoming\npeers. In contrast, Bitcoin’s eviction policy [ 2, 25 ] and Ethereum’s\nincoming-connection [ 43] limits can be more readily “leveraged” by\nattackers to complete an eclipse\nrbrunner: Seems to me that there is quite some stuff to study and fish for good ideas :)\nrucknium: boog900: I am hesitant to just have Monero copy what Bitcoin does. Maybe by accident a more resilient design could come out of the cryptonote protocol :)\nrucknium: Not to mention the risk of replacing big chunks of critical code\nboog900: Yeah I only got it from the link [2]: https://github.com/bitcoin/bitcoin/blob/master/doc/reduce-traffic.md\nrbrunner: By the way, if they want maximum harm, or maximum damage, what will the adversary do with those fully eclipsed nodes?\nrbrunner: I mean, yeah, stopping all transactions from propagating, but is there more?\nrucknium: rbrunner: Either complete spying on origin of tx or completely make the nodes unusable.\nboog900: rucknium: I agree but Bitcoin does have a much better address book IMO I have looked into its logic from a high level. We don't need to copy everything but some core parts would help us a lot.\nrucknium: Unreachable nodes are the majority of the network, probably. Many of them are users at home with the Monero GUI wallet.\nboog900: like only removing unreachable peers from the white list\nrucknium: This is a good paper on many network-level attacks, including eclipse: Franzoni, F., \u0026 Daza, V. (2022). \"SoK: Network-Level Attacks on the Bitcoin P2P Network.\" (https://moneroresearch.info/232)\nrucknium: We can continue discussion on this paper next week. Any last thoughts?\nrbrunner: Hmmm. So worst case somebody could quietly establish something like a network on top of the real network, and over time spy on everything.\nrbrunner: Ha, we are all living in the Monero network matrix already :)\nrucknium: Like mentioned prior to the meeting, the paper says rbrunner 's recent patch increased the cost of the attack: p2p: Improved peer selection with /24 subnet deduplication to disadvantage 'spy nodes' https://github.com/monero-project/monero/pull/9939\nrbrunner: But maybe hard to achieve further improvements on the selection front. Better list management is probably wide open for improvements compared to that\nrucknium: We can end the meeting here. Thanks everyone.\nrucknium: I guess you could figure out if you are eclipsed by doing a full network crawl once daily. Then if most of the network disappears from one day to the next, you know you've been eclipsed.\nrucknium: But what if the adversary does it gradually?\nvtnerd: You can ask anyone in the chat of hash+block#. Assuming the goal is fake blocks, it's hard to completely fake all avenues of checking that. Additionally there's the hashpower which would be lower\nhttps://libera.monerologs.net/monero-research-lab/20260916#c708293\nnostr:nevent1qvzqqqqqqypzpkl8jxlkhy3pat630epuqvxdlakqcl5m2cwxy2rh8cv56z8t3h0hqqs92wdlle6gprtjtem5patgngus4jlpeqy74062alkrs72qqnlkw8c36qugw "}
