https://github.com/hackenproof-public/somnia
Description:
EIP-2718 and EIP-1559 introduced typed transactions with strict signature rules, specifically requiring the yParity (v) field to be either 0 or 1. This design choice was meant to permanently eliminate signature malleability from typed transactions.
However, during testing I discovered that the client accepts non-canonical values such as v=27 for typed transactions. When broadcast, these transactions are treated as valid and mined successfully. This results in the same logical transaction being valid under multiple hashes - one with canonical v=0/1 and another with legacy-style v=27.
This breaks the typed transaction invariant that they are malleability-free. Any application, bridge, or smart contract system relying on canonical transaction hashes can be misled, enabling replay or double-spend style inconsistencies.
Impact
It Breaks Canonicality Guarantees: Typed transactions are expected to be malleability-free. This bug reintroduces multiple valid encodings. Potential Replay/Duplication Risk: Bridges, relayers, and indexers relying on unique hashes may process the same transaction twice under different IDs. Consensus/State Inconsistencies: Applications assuming typed txs cannot be mutated may break. Practical Exploit Scenario: An attacker can front-run by replaying the same transaction with a different v value to create confusion or bypass replay protection.
Why I Marked This as High Severity
The issue directly undermines one of the key protections introduced with typed transactions: eliminating signature malleability. By accepting v=27, the network allows the same logical transaction to have multiple valid encodings. This means a transaction can be replayed or represented differently, which breaks the assumption that every signed transaction maps to one unique hash.
From an attacker’s perspective, this is serious because:
It’s remotely exploitable - anyone who can send a transaction to the network can abuse this, no special access needed.
It requires no privileges and no user interaction - the attacker doesn’t need tokens, prior access, or a victim to click something. They just craft and broadcast a non-standard signature.
It threatens transaction integrity - downstream systems like bridges, relayers, or indexers that depend on transaction uniqueness could be tricked into double processing or mis-accounting. In some cases, that could escalate into fund loss or denial of service.
I didn’t mark it as Critical because it doesn’t by itself let an attacker directly drain balances from the base chain. But it’s still more than a minor inconsistency - it opens the door for replay and double-spend style attacks in higher-level protocols. That makes it a clear fit for High severity under the usual bug bounty scoring.
My Setup
RPC node: http://127.0.0.1:6500 Test account (private key available to me)
I ran the RPC node in terminal 1
NUMVALIDATORS=1
NUMGENERATEDTESTACCOUNTS=101
NUMGENERATEDTRANSACTIONSPERSECOND=0
NETWORK_PRESET=mainnet-small
bash ./ci/run-local-deployment.sh
I Craft a Typed Transaction [** Full runnable poc I uploaded as a zip file in the attachment section **]
I Constructed a valid EIP-1559 (typed) transaction with all required fields.
Instead of using v=0 or v=1 for yParity, I set v=27. and then invoked the transaction in terminal 2 :
Below is the summary of the output:
PoC: typed yParity accepts non-{0,1} (used 27) and mines the tx.
Address: 0x12a76e09BaE1934265a1aeA2812E0c772F372f8d ChainId: 5031 Nonce before: 1 Hashes: canonical: 0x3fb350747eed8a921e5df360e884bb3e135107d5d01bb0314f39baba5eb297cb mutated : 0xf74e37e7dcf402e3e4f60f8e848ecb31005a76636c72720111a677fd5c84771f
Receipt (mutated): status: 1 block : 564 from : 0x12a76e09BaE1934265a1aeA2812E0c772F372f8d nonce : 1 type : 2
This proves my claim. all the evidences and hashes are also attached in the attachments
Attachments are hidden
This information is not publicly available.
Comments are hidden
This information is not publicly available.