Hello,

> This feels like someone should have published it before. But I can't find an
> obvious citation (eg in any of the documentation around keyless ephemeral
> anchors), so I'll publish one here. Maybe I'm the first to point this out
> explicitly? Probably not; I'd appreciate an earlier citation if one exists.

Same, I cannot remember about an earlier citation in doucmentation around keyless
ephemeral anchors, and while this is a concern I was aware of I was expecting
that some reviewer of ephemeral anchors would figure it out earlier (or at least
praying for that...).  

> tl;dr: _Anyone_ can do a replacement cycling attack on transactions where fees
> are paid via CPFP via keyless anchors and similar outputs that a third-party
> can double-spend. Secondly, for attackers who were already planning on making a
> transaction with a higher total fee and total fee-rate than the target, this
> attack is almost free.

Yes, about the second observation, see the section 5.2 about high-fee collaboration
transactions advantage in the write-up which was initially released at the
disclosure of replacement cycling attacks [0].

[0] https://github.com/ariard/mempool-research/blob/2023-10-replacement-paper/replacement-cycling.pdf

In my belief, the simple fact that any bitcoin economical actor originating
transaction traffic at regular interval (e.g LSP or exchange batching) can
leverage that position to engage in replacement cycling attacks is a real worry.

> # The Attack
>
> Suppose that Alice has created a 2 transaction package consisting of low-fee or
> zero-fee transaction A, whose fees are CPFP paid via a keyless ephemeral anchor
> spent by transaction B. For B to pay fees, obviously it must spend a second
> transaction output.
>
> Mallory can cycle A and B out of mempools by broadcasting transaction B2,
> spending his own output and the keyless ephemeral anchor of A, at a higher
> fee/fee-rate than B. Next, Mallory broadcasts B3, double-spending B2 by
> spending Mallory's input but not the ephemeral anchor of A. Assuming Mallory
> needed to mine B3 anyway, the only cost to this attack is the small chance that
> B2 will in fact be mined between the time that B2 is replaced by B3.
>
> At this point A is no longer economical to mine as B has been cycled out, and A
> may be dropped from mempools depending on the circumstances.

Yes, in my understanding this scenario works. I think can even maliciously batch
B3 to replace a N number of keyless spending B2 transactions at the same time.

> ## SIGHASH_ANYONECANPAY
>
> Obviously, a similar attack is possible against SIGHASH_ANYONECANPAY-using
> transactions, provided that _all_ signatures sign with SIGHASH_ANYONECANPAY.

Yes, a single SIGHASH_ALL locks-in the spent input, and as such avoid a third-party
to adds a higher-fee / feerate input, then to be ulteriorly conflited with. There
might be some data carriage style transactions using this pattern...

> # Countermeasures
>
> As with other replacement cycling attacks, rebroadcasting A and B fixes the
> issue. I think the existence of this additional type of replacement cycling
> attack suggests that adding an optional rebroadcasting module to Bitcoin Core
> that would keep track of dropped transactions and re-add them to mempools when
> they are again valid would make sense. This fixes all replacement cycling
> attacks and there's probably lots of nodes who have the memory and/or disk
> space to keep track of dropped transactions like this.
>
> Preventing the replacement of B2 with B3 is _not_ a viable countermeasure: if
> that replacement was prohibited, attackers could in turn exploit that rule as a
> new form of transaction pinning!

I still think altruistic rebroadcasting is a very limited mitigation as even beefy
nodes have computationally-limited memory and/or disk spaces. They would need to have
some ordering of the dropped transactions e.g caches by tiers of feerate group and
a replacement cycling attacker would just have to spam the top cache.

I believe it was already discussed there and in the previous email (I might be
missing something ?) [1].

[1] https://groups.google.com/g/bitcoindev/c/qEx4K8lGnLk/m/AMPqDi6MBwAJ

> # Privacy
>
> The fact that rebroadcasting is a countermeasure is a privacy concern. Each
> time a transaction is rebroadcast by the sender is a potential opportunity to
> track the origin of a transaction. Again, having third parties rebroadcasting
> transactions altruistically would mitigate this privacy concern.

Yes...On the other hand third parties rebroadcasting transactions are now
exposing themselves to be injected DoS traffic. Having full-nodes offering a
threshold of their memory / CPU cyles to process anonymizing traffic is great,
especially for the marginal lighnting user. When it's a lightning hub with
a lot of channel funds, I believe it's naive to expect those additional memory /
CPU cyles cannot be spammed for exploitation purposes.

> It kinda seems like this might introduce a DOS vector to the nodes running this
> module since you can keep cycling B3, B4 etc. and force the mempool to house
> all of these until one of them gets mined. Further, it would cause the mempool
> to have to decide which of these dead transactions gets priority upon the eviction
> of the conflicting one. Is this something you've given thought to? Admittedly I
> haven't dived deep into it.

I did, and effectively you're running into hierarchy of caches management styles of issues [2]

[2] https://github.com/bitcoin/bitcoin/issues/28699

Best,
Antoine
ots hash: eba1b16753146707a07dcd1aee1ca066fcb323e18222be7282088c49d9f44266
Le vendredi 2 août 2024 à 23:00:16 UTC+1, Keagan McClelland a écrit :
>  I think the existence of this additional type of replacement cycling
attack suggests that adding an optional rebroadcasting module to Bitcoin Core
that would keep track of dropped transactions and re-add them to mempools when
they are again valid would make sense. This fixes all replacement cycling
attacks and there's probably lots of nodes who have the memory and/or disk
space to keep track of dropped transactions like this.

It kinda seems like this might introduce a DOS vector to the nodes running this
module since you can keep cycling B3, B4 etc. and force the mempool to house
all of these until one of them gets mined. Further, it would cause the mempool
to have to decide which of these dead transactions gets priority upon the eviction
of the conflicting one. Is this something you've given thought to? Admittedly I
haven't dived deep into it.

Keags

On Fri, Aug 2, 2024 at 5:30 AM Peter Todd <pe...@petertodd.org> wrote:
This feels like someone should have published it before. But I can't find an
obvious citation (eg in any of the documentation around keyless ephemeral
anchors), so I'll publish one here. Maybe I'm the first to point this out
explicitly? Probably not; I'd appreciate an earlier citation if one exists.


tl;dr: _Anyone_ can do a replacement cycling attack on transactions where fees
are paid via CPFP via keyless anchors and similar outputs that a third-party
can double-spend. Secondly, for attackers who were already planning on making a
transaction with a higher total fee and total fee-rate than the target, this
attack is almost free.


# The Attack

Suppose that Alice has created a 2 transaction package consisting of low-fee or
zero-fee transaction A, whose fees are CPFP paid via a keyless ephemeral anchor
spent by transaction B. For B to pay fees, obviously it must spend a second
transaction output.

Mallory can cycle A and B out of mempools by broadcasting transaction B2,
spending his own output and the keyless ephemeral anchor of A, at a higher
fee/fee-rate than B. Next, Mallory broadcasts B3, double-spending B2 by
spending Mallory's input but not the ephemeral anchor of A. Assuming Mallory
needed to mine B3 anyway, the only cost to this attack is the small chance that
B2 will in fact be mined between the time that B2 is replaced by B3.

At this point A is no longer economical to mine as B has been cycled out, and A
may be dropped from mempools depending on the circumstances.


## SIGHASH_ANYONECANPAY

Obviously, a similar attack is possible against SIGHASH_ANYONECANPAY-using
transactions, provided that _all_ signatures sign with SIGHASH_ANYONECANPAY.


# Countermeasures

As with other replacement cycling attacks, rebroadcasting A and B fixes the
issue. I think the existence of this additional type of replacement cycling
attack suggests that adding an optional rebroadcasting module to Bitcoin Core
that would keep track of dropped transactions and re-add them to mempools when
they are again valid would make sense. This fixes all replacement cycling
attacks and there's probably lots of nodes who have the memory and/or disk
space to keep track of dropped transactions like this.

Preventing the replacement of B2 with B3 is _not_ a viable countermeasure: if
that replacement was prohibited, attackers could in turn exploit that rule as a
new form of transaction pinning!


# Privacy

The fact that rebroadcasting is a countermeasure is a privacy concern. Each
time a transaction is rebroadcast by the sender is a potential opportunity to
track the origin of a transaction. Again, having third parties rebroadcasting
transactions altruistically would mitigate this privacy concern.

--
https://petertodd.org 'peter'[:-1]@petertodd.org

--
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bitcoindev/ZqyQtNEOZVgTRw2N%40petertodd.org.

--
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bitcoindev/57c52a0c-c390-4b7b-8247-8612a489cb98n%40googlegroups.com.