Eclair Lightning Bug Let Any Connected Peer Exhaust a Node’s Memory

Original colored-pencil illustration of an Asian network engineer studying a Bitcoin Lightning node memory-alert display in a rack-filled laboratory, with the saved neon-green BitcoinVersus.tech watermark on solid black at the bottom left.

A publicly disclosed software flaw allowed a connected peer to force an Eclair Bitcoin Lightning node to accumulate messages until its memory ran out. The denial-of-service issue affected Eclair 0.14.0; ACINQ fixed it in version 0.14.1 on July 29, 2026. Researcher rorp published the technical disclosure on November 8, and Bitcoin Optech highlighted it on November 20. The finding does not establish an ongoing attack on patched nodes.

Eclair is ACINQ’s Scala implementation of the Lightning Network, Bitcoin’s payment-channel network. Unlike a wallet-only software problem, the flaw was reachable from a network peer without any Lightning channel shared with the victim. The researcher did not report direct theft of private keys or coins; the immediate impact was disruption of node availability.

Background video: Bitcoin Amsterdam 2025 discussion of Lightning channel splicing, which requires coordinated commitment messages. The recording predates the disclosure and is not a demonstration of an exploit. Watch on YouTube.

The error was in a message count

Lightning splicing lets two parties adjust a channel’s funding through an on-chain transaction without first closing their channel. When multiple possible funding transactions are pending, a node may need several related commitment_signed messages. The start_batch announcement tells the receiver how many messages form one logical update.

The BOLT 2 protocol specification requires a batch size greater than one and no larger than 20. Eclair 0.14.0 enforced the ceiling but omitted the lower-bound check. A zero count could therefore be accepted, after which the implementation kept buffering messages while waiting for an impossible match between a positive received count and zero. A connected attacker could exhaust the process’s memory.

The original disclosure identifies additional weaknesses: a peer did not need a recognized channel identifier, buffered payloads were not yet cryptographically checked, and ordinary connection limits did not bound the memory used by each connected peer. None of those observations proves the vulnerability was exploited against a production operator.

Why a crash matters even without stolen keys

A Lightning node must stay available to monitor on-chain channel closures and enforce payment timelocks. Repeated crashes can prevent timely responses to disputed channel states or expiring hash-time-locked contracts. Consequently, a denial-of-service problem can create an indirect financial risk despite not handing an attacker spending credentials.

Background video: Rusty Russell explains dual funding and splicing. The discussion describes protocol mechanics relevant to batch handling, not the Eclair security report. Watch on YouTube.

Similar reliability considerations motivated another recent LND payment-state correction. The site’s coverage of Core Lightning’s next release candidate also demonstrates how distinct implementations can introduce improvements and expose separate software risks.

Patch, timeline and operator action

ACINQ’s patch rejects batch sizes of one or fewer as well as values above 20. Its July 29 release notes describe a compatible upgrade that does not require closing channels: administrators stop Eclair, install the release, and restart it. Operators still running 0.14.0 should move to 0.14.1 or a later supported version and verify the running binary after the change.

Users of a Lightning application relying on an Eclair backend should check which party operates the node before attempting changes. Independent operators control their own upgrades; custodial services typically manage the infrastructure. The topic connects to our earlier Bitcoin Core privacy-fix reporting and coverage of Lightning payment software that removes the need for users to run a node.

The lesson is straightforward: consensus rules and cryptography cannot compensate for a node accepting an invalid protocol field. The vulnerability was already patched months before its public analysis, and the open disclosure now allows independent engineers to examine the failure and confirm protection in their own deployments.

Leave a Reply