MIME-Version: 1.0
References: <CAPv7TjaE1wF-25R=LaOES33A78ovDAp9-waiC7n5YLJnMmNs9A@mail.gmail.com>
	<uVQNn9hhpqlQuS-RzrUkpClVtegMRUoyIL6ITaYfNkjd_XYyu9Fh9vdAeLguzOyOrNx5FtuHk7yyZAdivqCVR2PKzF_PsoWJlsSY9oJTF7s=@protonmail.com>
	<CAPv7TjYE1rp2EEo247fKmmN=q9QBqCHPBy56xvBymOv418LFDw@mail.gmail.com>
	<y8brIayR1YQIud3TxdgiBJK3zrxclhcICAJgSw3tqqJM2jP9WA1SWiesXcNSrRxiIM7m8cRkZlLdu0sz2lNFEe5HbQI4TIRvkoA3yysMzKE=@protonmail.com>
In-Reply-To: <y8brIayR1YQIud3TxdgiBJK3zrxclhcICAJgSw3tqqJM2jP9WA1SWiesXcNSrRxiIM7m8cRkZlLdu0sz2lNFEe5HbQI4TIRvkoA3yysMzKE=@protonmail.com>
From: Ruben Somsen <rsomsen@gmail.com>
Date: Wed, 11 Sep 2019 06:58:57 +0200
Message-ID: <CAPv7TjaMm5q8Q-AXbVcjx5RoKt-5y7t4Fx+CMhJ=cabLysc9=w@mail.gmail.com>
To: ZmnSCPxj <ZmnSCPxj@protonmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Cc: Bitcoin Protocol Discussion <bitcoin-dev@lists.linuxfoundation.org>
Subject: Re: [bitcoin-dev] PoW fraud proofs without a soft fork
Precedence: list

Hi ZmnSCPxj,

>I suppose the critical difference is that invalid inflation can fool the S=
PV node, the fullnode will not be so fooled.

That is correct. If you sybil the SPV node, you can break any
consensus rule you like. I believe this is inherent to fraud proofs in
general, because you skip consensus checks unless you're able to
receive a fraud proof.

But note that my goal in the comparison was to assert that there is no
security difference between committing or not committing the utreexo
hash into a block. The attack your describe works in either situation,
so my conclusion remains that committing the hash adds no security.

Other weaknesses compared to full nodes are:
- the SPV nodes rely on the existence of a healthy network of utreexo
supporting full nodes
- at least one honest block needs to be mined
- consensus slows down, because you need to allow time for an honest
minority to produce a block

Cheers,
Ruben

On Mon, Sep 9, 2019 at 8:58 AM ZmnSCPxj <ZmnSCPxj@protonmail.com> wrote:
>
> Good morning Ruben,
>
> Yes, I suppose that is correct.
>
> I suppose the critical difference is that invalid inflation can fool the =
SPV node, the fullnode will not be so fooled.
>
> A somewhat larger-scale attack is to force a miner-supported miner-subsid=
y-increase / blocksize-increase hard fork.
> If enough such SPV nodes can be sybilled, they can be forced to use the h=
ard fork, which might incentivize them to support the hard fork rather than=
 back-compatible consensus chain.
>
> Regards,
> ZmnSCPxj
>
> > Hi ZmnSCPxj,
> >
> > Thank you for your comments. You raise an important point that I should=
 clarify.
> >
> > > 1.  In event of a sybil attack, a fullnode will stall and think the b=
lockchain has no more miners.
> >
> > You can still attack the full node by feeding it a minority PoW chain,
> > then it won't stall.
> >
> > > 2.  In event of a sybil attack, an SPV, even using this style, will f=
ollow the false blockchain.
> >
> > Correct, but this false blockchain does need to have valid PoW.
> >
> > So in both cases valid PoW is required to fool nodes. The one
> > difference is that for a full node, the blocks themselves also need to
> > be valid (except for the fact that they are in a minority chain), but
> > the end result is still that a victim can be successfully double spent
> > and lose money.
> >
> > I hope this clarifies why I consider the security for these two
> > situations to be roughly equivalent. In either situation, victims can
> > be fooled into accepting invalid payments.
> >
> > Cheers,
> > Ruben
> >
> > On Mon, Sep 9, 2019 at 6:14 AM ZmnSCPxj ZmnSCPxj@protonmail.com wrote:
> >
> > > Good morning Ruben,
> > >
> > > >     One might intuitively feel that the lack of a commitment is uns=
afe,
> > > >     but there seems to be no impact on security (only bandwidth). T=
he only
> > > >     way you can be fooled is if all peers lie to you (Sybil), causi=
ng you
> > > >     to follow a malicious minority chain. But even full nodes (or t=
he
> > > >     committed version of PoW fraud proofs) can be fooled in this wa=
y if
> > > >     they are denied access to the valid most PoW chain. If there ar=
e
> > > >     additional security concerns I overlooked, I=E2=80=99d love to =
hear them.
> > > >
> > >
> > > I think it would be better to more precisely say that:
> > >
> > > 1.  In event of a sybil attack, a fullnode will stall and think the b=
lockchain has no more miners.
> > > 2.  In event of a sybil attack, an SPV, even using this style, will f=
ollow the false blockchain.
> > >
> > > This has some differences when considering automated systems.
> > > Onchain automated payment processing systems, which use a fullnode, w=
ill refuse to acknowledge any incoming payments.
> > > This will lead to noisy complaints from clients of the automated paym=
ent processor, but this is a good thing since it warns the automated paymen=
t processor of the possibility of this attack occurring on them.
> > > The use of a timeout wherein if the fullnode is unable to see a new b=
lock for, say, 6 hours, could be done, to warn higher-layer management syst=
ems to pay attention.
> > > While it is sometimes the case that the real network will be unable t=
o find a new block for hours at a time, this warning can be used to confirm=
 if such an event is occurring, rather than a sybil attack targeting that f=
ullnode.
> > > On the other hand, such a payment processing system, which uses an SP=
V with PoW fraud proofs, will be able to at least see incoming payments, an=
d continue to release product in exchange for payment.
> > > Yet this is precisely a point of attack, where the automated payment =
processing system is sybilled and then false payments are given to the paym=
ent processor on the attack chain, which are double-spent on the global con=
sensus chain.
> > > And the automated system may very well not be able to notice this.
> > > Regards,
> > > ZmnSCPxj
>
>