On Tue, Jun 19, 2018 at 02:02:51PM -0400, David A. Harding wrote:
> Anyone can rewrite a SIGHASH_NOINPUT input's outpoint, but the actual
> transaction containing the settlement is expected to have (at least) two
> inputs, with the second one originating the fees. That second input's
> signature is (I assume) using SIGHASH_ALL to commit to all outpoints in
> the transaction, so it can't be arbitrarily rewritten by a third-party
> to apply to a different state outpoint
I realized that the fee-paying input could possibly be signed with
SIGHASH_ALL|SIGHASH_ANYONECANPAY to allow anyone to arbitrarily
rewrite the other input signed with SIGHASH_NOINPUT. However, this
reminded me of the well-known DoS against transactions signed with
SIGHASH_ANYONECANPAY[1], which seems to apply generally against
SIGHASH_NOINPUT as well and may allow theft from HTLCs.
## DoS against Eltoo settlements
Alice and Mallory have a channel with some state updates. Alice tries
to initiate a cooperate close, but Mallory stalls and instead broadcasts
the trigger transaction and the first state (state 0). Notably, the
first state is bundled into a very large vsize transaction with a low
feerate. State 1 is added to another very large low-feerate
transaction, as are states 2 through 9.
Alice could in theory RBF the state 0 transaction, but per BIP125 rule
#3, she needs to pay an absolute fee greater than all the transactions
being replaced (not just a higher feerate). That could cost a lot.
Alice could also create a transaction that binds the final state to the
state 9 transaction and attempt CPFP, but increasing the feerate for the
transaction ancestor group to a satisfactory degree would cost the same
amount as RBF.
So Alice is stuck waiting for states 0-9 to confirm before the final
state can be confirmed. During recent periods of full mempools on
default nodes, the waiting time for 10 nBTC/vbyte transactions has been
more than two weeks.
## HTLC theft
If Mallory is able to introduce significant settlement delays, HTLC
security is compromised. For example, imagine this route:
Mallory <-> Alice <-> Bob
Mallory orders a widget from Bob and pays via LN by sending 1 BTC to
Alice hashlocked and timelocked, which Alice forwards to Bob also
hashlocked and timelocked. Mallory releases the preimage to Bob, who
claims the funds from Alice and ships the widget, giving Alice the
preimage.
At this point, Mallory broadcasts the transactions described in the
preceding section.
If the low feerate of states 0-9 prevent them from confirming before the
timeout, Mallory can create a transaction containing a dishonest final
state that executes the refund branch. Like before, she can bury this
in an ancestor transaction chain that would be cost prohibitive for Alice
to RBF.
Considered independently, this is a very expensive attack for Mallory,
and so perhaps impractical. But Mallory can join forces with someone
already creating large low-feerate consolidation transactions. Better
yet, from Mallory's perspective, she can execute the attack against
hundreds of channels at once (creating long chains of ancestor
transactions that are large in aggregate rather than individually
large), using the aggregate size of all the victims' channels against
each of the individual victims.
Thanks,
-Dave
[1]
https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2014-August/006438.html
signature.asc
Description: PGP signature
___
Lightning-dev mailing list
Lightning-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/lightning-dev
