(pursuing the conversation as the problem of transaction with perishable finality
in the mempool ain't solved)

> That's not really a comparable situation. If the HTLC-timeout transaction
> replaces a HTLC-preimage transaction in a mempool, it will do so under ordinary
> BIP125 rules, and is thus "paying for" the bandwidth with a higher fee.
> Equally, in a replace-by-fee-rate scenario, it would be "paying for" its
> bandwidth with a higher fee-rate. Either way, something will confirm.

> In the OP_Expire case, we're talking about a transaction that becomes entirely
> invalid after a point in time. If the transaction isn't mined with reasonably
> high probability (eg >10%, preferably higher) an attacker may be able to
> consume bandwidth indefinitely at little to no cost.

In my view, there is a generic concept of "perishable transactions", i.e transactions
of which the broadcast might be wasted by either a candidate replacement or expiring
due to future semantics like op_expire.

This is correct that the latter is a bit different as there is no guarantee that an on-chain
fee is paid (be it replace-by-fee or replace-by-fee-rate). However, from the viewpoint of the
transaction-relay node there is still a probability that when the HTLC-preimage broadcast time
is near `cltv_expiry`, the broadcast is "wasted" as it might be replaced immediately by the
HTLC-timeout, even before it's fully propagated on the whole transaction-relay topology.

I think you can reduce the observation in some LN configuration for an attacker to waste
transaction-relay bandwidth, though there is a notable minimum bar, the attacker might have
to bear channel on-chain funding cost.

> Similar to what I wrote above, in altruistic re-broadcasting, the attacker
> doing the replacement cycling attack has already paid for the bandwidth
> consumed in broadcasting the replaced transaction because they paid fees for
> the cycling attack. Nothing more needs to be done beyond existing RBF/RBFR
> rules to avoid DoS attacks.

And while I share the opinon that for a classic altruistic re-broadcasting, an
attacker doing the replacement is paying for the bandwidth, this might be restrained
by the caveat laid out above, namely exploiting the settlement timing of a payment
channel network to drift the altruistic re-broadcasting behavior to a maximum of
concurrent propagation situation where the bandwidth is wasted.

> I mean, that's still BIP157 working. It's just not supported by every node.
> It's easy to learn about lots of addrs from the addr distribution mechanisms,
> so I don't think that's a serious issue.
Yes, you can say it's still working though note how a limited number of BIP157 peers
can open the doors to other risks, e.g privacy issues.

> I'm very curious as to what those nodes are actually doing. Possibly some
> pre-made node distribution is in fact setting non-standard mempool size limits.
> Or these are fake spy nodes with unusual behavior.

I must say I have not investigated further myself in deep why some nodes sounds to
run with lower mempool size, a likely explanation like you're pointing to is pre-made
node running on raspi devices (with already other pre-configured services running), where
mempool transaction buffering can be an issue.

> If V3 transactions is such that a child tx can be replaced at a cost that
> doesn't "pay for" the bandwidth of the parent that is evicted, that is a
> straight-forward design flaw/bug in V3 transactions. Fixing that should be
> pretty straight forward, at which point the attacker is again paying "fairly"
> with fees on each cycle.

This is correct that V3 transactions might still have open design issues, w.r.t
parent package under mempool min transaction relay fees.

Best,
Antoine

Le mercredi 20 mars 2024 à 20:52:58 UTC, Peter Todd a écrit :
(replying manually with a cut-n-paste due to a mailing list delivery issue)

> > > > nodes should require higher minimum relay fees for transactions close to
> > > their expiration height to ensure we don’t waste bandwidth on transactions
> > > that have no potential to be mined
>
> I think this concern can be raised on _today_ LN second-stage transactions (HTLC-preimage / HTLC-timeout),
> when a HTLC-preimage is broadcast near "cltv_expiry". LN routing nodes will automatically go to broadcast an
> on-chain HTLC-timeout transaction. Probabilistically, we're wasting bandwidth on transactions that _might_ have
> lower odds to be mined.

That's not really a comparable situation. If the HTLC-timeout transaction
replaces a HTLC-preimage transaction in a mempool, it will do so under ordinary
BIP125 rules, and is thus "paying for" the bandwidth with a higher fee.
Equally, in a replace-by-fee-rate scenario, it would be "paying for" its
bandwidth with a higher fee-rate. Either way, something will confirm.

In the OP_Expire case, we're talking about a transaction that becomes entirely
invalid after a point in time. If the transaction isn't mined with reasonably
high probability (eg >10%, preferably higher) an attacker may be able to
consume bandwidth indefinitely at little to no cost.

> > If you already have a need to make such transactions, you can argue that the
> > marginal cost to also use up that bandwidth is low. But that's already the case
> > with RBF: we allow any transaction to be replaced with RBF for a (by default)
> > 1sat/vB additional cost to "pay for" the bandwidth of that replacement.
> > OP_EXPIRE does not change this situation: you're still paying for an additional
> > 1sat/vB cost over the replaced transaction, as eventually one of your
> > replacements will get mined.
>
> I think yes this is indeed more a replacement issue, nothing new introduced by OP_EXPIRE finality time-bounding semantics.
> However, I think it's more an issue if we introduce things like altruistic re-broadcasting.
>
> https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2023-December/022188.html
>
> Certainly, the re-broadcast could favor transactions with higher odds of being mined, which naively should match RBF rules.

Similar to what I wrote above, in altruistic re-broadcasting, the attacker
doing the replacement cycling attack has already paid for the bandwidth
consumed in broadcasting the replaced transaction because they paid fees for
the cycling attack. Nothing more needs to be done beyond existing RBF/RBFR
rules to avoid DoS attacks.

> And by the same way taking time to answer the open questions on this thread from the old mailing list:
> https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2023-December/022224.html
>
> > Are you claiming that BIP157 doesn't work well? In my experience it does.
>
> I've not checked recently, though from research memory a while back the numbers of BIP157 services offering peers
> was in the range of ~10 / 100.
>
> One can check by collecting nVersions messages from peers with `NODE_COMPACT_FILTERS`.

I mean, that's still BIP157 working. It's just not supported by every node.
It's easy to learn about lots of addrs from the addr distribution mechanisms,
so I don't think that's a serious issue.

> > Huh? Bitcoin nodes almost always use the same mempool limit, 300MB, so mempool min fees are very consistent across nodes. I just checked four different long running > nodes I have access to, running a variety of Bitcoin Core versions on different platforms and very different places in the world, and their minfees all agree to well within 1% > In fact, they agree on min fee much *more* closely than the size of their mempools (in terms of # of transactions). Which makes sense when you think about it, as the
> > slope of the supply/demand curve is fairly flat right now.
>
> See https://github.com/bitcoin/bitcoin/pull/28488 which is motivated from diverging mempool min fees from the ground iirc.

https://github.com/bitcoin/bitcoin/issues/28371#issuecomment-1939604817 is the
only actual data I could find in that link.

I'm very curious as to what those nodes are actually doing. Possibly some
pre-made node distribution is in fact setting non-standard mempool size limits.
Or these are fake spy nodes with unusual behavior.

> > From the point of view of a single node, an attacker can not reuse a UTXO in a replacement cycling attack. The BIP125 rules, in particular rule #4, ensure that each
> > replacement consumes liquidity because each replacement requires a higher fee, at least high enough to "pay for" the bandwidth of the replacement. An attacker trying to > use the same UTXO's to cycle out multiple victim transactions has to pay a higher fee for each victim cycled out. This is because at each step of the cycle, the attacker had > to broadcast a transaction with a higher fee than some other transaction.
>
> This does not stay true with nVersion=3, where a package parent can be signed with a feerate
> under min relay tx fee. See the second test attached in the initial full report email on replacement
> cycling attacks, one can replace the child of the package and the parent is automatically evicted,
> without the "pay for" bandwidth of the replacement fully covered.
>
> This is correct there is a minimal fee basis for each additional victim cycled out, while one can get
> a very advantageous scaling effect by RC'ing the child txn.

If V3 transactions is such that a child tx can be replaced at a cost that
doesn't "pay for" the bandwidth of the parent that is evicted, that is a
straight-forward design flaw/bug in V3 transactions. Fixing that should be
pretty straight forward, at which point the attacker is again paying "fairly"
with fees on each cycle.

--
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+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bitcoindev/d9f8f4dc-e772-4383-8024-1e67695f5685n%40googlegroups.com.