Re: [bitcoin-dev] Hardware wallets and "advanced" Bitcoin features

From: Antoine Riard <antoine.riard@gmail•com>
To: Kevin Loaec <kevin@revault•dev>,
	 Bitcoin Protocol Discussion
	<bitcoin-dev@lists•linuxfoundation.org>
Subject: Re: [bitcoin-dev] Hardware wallets and "advanced" Bitcoin features
Date: Sat, 16 Jan 2021 20:31:07 -0500	[thread overview]
Message-ID: <CALZpt+EMHkYFxkO+SxFEjbiG2hU2Rm5kpMd8+71+Wjxc48J5ig@mail.gmail.com> (raw)
In-Reply-To: <da42f7e09cabcaa935bce4036340ef14f4bf454c.camel@revault.dev>

[-- Attachment #1: Type: text/plain, Size: 11935 bytes --]

Hello Kevin,

Thanks for starting this thread, that's a really relevant discussion
ecosystem-wise !

> * Proposed improvement: The HW should display the Bitcoin Script itself
when possible (including the unlock conditions).

What level of script literacy are you assuming on your users ? I can see
enterprise/hobbyist folks to know enough of Script to understand the
intended behavior but I don't think that's a reasonable assumption for your
average user. Of course, Miniscript Policy makes things easier, but IMHO, I
still hope to see some mature, higher-level language (e.g Ivy) to ease
script semantic understanding and thus widen the crowd of users.

Further, I would do a bit on UX research on the correctness model expected
by your users. I.e if they fail to verify accordingly, are they losing
funds, transaction doesn't confirm, transaction doesn't even propagate,
etc. You should also make assumptions on the mental resources you're
required from them. Time-sensitive L2 protocols have a wide scope to check,
e.g not verifying the nSequence/nLocktime fields can provoke funds failures.

> This applies to pre-signed transaction protocols especially well as the
template of these transactions could be known
and recognized by the HW. Typically for Revault, the HW could display:
"Unvault Transaction, all expected pubkeys
present in the script".

In the future, I would expect templates of high-security protocols like
vaults to be part of the trusted computing base of any decent HW. I think
good standards there would avoid HW vendors to come with some kind of
certified-templates scheme and thus having to bless custom scripts of every
vaults implementations.

> Proposed improvement: The HW could know pubkeys or xpubs it does not hold
the private keys
for, and display a label (or
understand it for logic reasons, such as "expected pubkeys" as the previous
example).

I don't think you even need user input on this, the absence of pubkeys
knowledge itself is a trigger to display a label or ask for further
information. Where absence of pubkeys knowledge can be interpreted as
devoid from key whitelisting or privkey ownership.

> Going further, the xpubs could be
aliased the first time they are entered/verified (as part of, say, an
initial setup ceremony) for instance with the
previously mentioned Miniscript policy: or(pk(Alice), and(pk(Bob),
after(42))).

I would be careful about accidental or malicious alias collisions. But yes
that can be something, you can even conserve a merkle tree root in the
Secure Element where the hashed element are
previously authenticated alias/pubkeys. And require from the non-trusted
challenger to come with a merkle branch to validate address inclusion.

> Then there is PSBT support and the maximum transaction size limit for
these: we need more transparency from HW manufacturers on their li
mitations.

I understand them, Script is full of subtleties, taproot is likely to have
more of them and if you take sighash malleability that's not something you
want your average user to play with. Maybe it
would be better to come up with a first wave of script features on which
you expect transparency ? For sure, OP_CSV is a good candidate.

> Once any input of a (pre-signed)transaction is
spent, this transaction isn't valid anymore. Most pre-signed transactions
protocols are used today as a form of defense
mechanism, spending any input would mean incapacitating the entire defense
mechanism.

I don't see the exact issue here. E.g in Lightning, even if you pre-sign a
justice transaction punishing every revokeable outputs on counterparty
transaction, and one input is spent, will current HWs prevent you to-resign
an updated justice transaction ?

> I understand some of these changes may be very difficult, especially
given the low memory and computational power of
secure elements.

Instead of relying on hand-sized devices, what about relying on HSMs for a
first-wave of adoptions, those ones have far enough resources to run a
reasonable L2 stack on the trusted-side ?

But overall agree, on the requirement to level-up HWs for L2. IMO, a first
step could be to list a  common set of features beyond
deployed/soon-to-be-deployed L2s, that would make things easier for HW
vendors to have a unique list of grievances. Before they engage in further,
dedicated tweaks to adapt for each protocol security model. OP_CSV/OP_CTLV
decoding/"burned" standard scripts support would be a good starter.

> Feel free to reply with your comments or adding suggestions, I am not a
hardware wallet expert and would take criticism
wit
hout being offended.

I don't know yet any *L2* hardware wallet expert :)

Cheers,
Antoine

Le jeu. 14 janv. 2021 à 13:46, Kevin Loaec via bitcoin-dev <
bitcoin-dev@lists•linuxfoundation.org> a écrit :

> Hello everyone,
>
> I would like to start a discussion on improving Hardware Wallets.
>
> My approach to this right now is from a vault protocol we are developing
> (Revault, [1]), and its Hardware Wallet
> requirements. I started working on a Github Issue in our repo [2], other
> people recommended us to do a more general
> discussion on the mailing list instead as it could benefit many other
> protocols and users.
> This email discusses improvements that would benefit everyone, and some
> that are more suitable for "layer 2" or pre-
> signed transactions protocols.
> The goal is to spark discussions and hopefully iterate to a more secure
> and more usable hardware ecosystem for all
> bitcoiners.
> While I mainly foresee issues/improvements that may affect Revault, I
> would be really happy to see people joining this
> thread with any other ideas and remarks that would benefit some parts of
> Bitcoin that I overlooked.
>
>
>
> Prior work on similar problematics:
> ===================================
> - ZmnSCPx
> j: [Lightning-dev] Speculations on hardware wallet support for Lightning
> [3]
> - mflaxman: Known Issues: Verifying a Receive Address [4]
> - ksedgwic and devrandom01: Lightning-signer [5]
> - benma: How nearly all personal hardware wallet multisig setups are
> insecure [6]
>
>
> The postulate we start from is that Hardware Wallets (HW) are useful to
> mitigate the compromission of the day-to-day
> device of a user. They mainly prevent private-key extraction today, and
> aren't very suitable against an attack on the
> transaction being signed, as explained further.
> To make this discussion security-focused, let's assume the general purpose
> device (laptop, for example) is compromised
> with a malware implanted by an attacker, capable of modifying PSBTs,
> displayed addresses, etc.
>
> Our study so far:
>
>
> Output Script Parsing:
> ======================
> Problem: A typical HW today would display the "destination" of a
> transaction in the form of a bitcoin address. A user
> would generally compare this wit
> h the address displayed on his laptop screen... which might have been
> compromised
> already. The correct usage would be for a user to verify this address on a
> third device (mobile phone, for example).
> This is weak security and bad user experience.
> Proposed improvement: The HW should display the Bitcoin Script itself when
> possible (including the unlock conditions).
> The best way to do so would be to lift this Script to a more user-friendly
> format such as a MiniScript Policy display,
> but anything would be better than an "address".
> This applies to pre-signed transaction protocols especially well as the
> template of these transactions could be known
> and recognized by the HW. Typically for Revault, the HW could display:
> "Unvault Transaction, all expected pubkeys
> present in the script".
>
>
> Pubkey Interpretation:
> ======================
> Problem: currently HW cannot "identify" addresses or keys.
> Proposed improvement: The HW could know pubkeys or xpubs it does not hold
> the private keys
> for, and display a label (or
> understand it for logic reasons, such as "expected pubkeys" as the
> previous example). Going further, the xpubs could be
> aliased the first time they are entered/verified (as part of, say, an
> initial setup ceremony) for instance with the
> previously mentioned Miniscript policy: or(pk(Alice), and(pk(Bob),
> after(42))).
> This should be done in the Secure Element if possible to avoid physical
> compromission, but would be a strong improvement
> versus a day-to-day laptop in any case.
>
>
> Better Bitcoin Compatibility:
> =============================
> Problem: most HW cannot interpret some Script OPs such as OP_CSV, or any
> conditional outputs. This is a major issue for
> anyone using Bitcoin "advanced" features. Related to this are the Sighash
> flags: most HW do not support most Sighash
> flags. Kind of annoying for a signing device. Then there is PSBT support
> and the maximum transaction size limit for
> these: we need more transparency from HW manufacturers on their li
> mitations.
> Solution: Make Bitcoin HW actually compatible with Bitcoin :D
>
>
> Inputs (mainly for pre-signed Tx):
> ==================================
> Problem: Poisoned inputs are a major risk for HW as they don't know the
> UTXO set. While this can be exploited for fee
> attacks, it is a bigger threat to pre-signed transactions protocols. Once
> any input of a (pre-signed)transaction is
> spent, this transaction isn't valid anymore. Most pre-signed transactions
> protocols are used today as a form of defense
> mechanism, spending any input would mean incapacitating the entire defense
> mechanism.
> Proposed improvement: for protocols that requires it, keeping track of
> inputs already signed once would be extremely
> helpful. Going further, most of these protocols require to follow a
> specific signing order (typically the "clawback"
> first, then the regular spend path) so adding a way to check that a
> "clawback" has been signed first, with the same
> input, would be very helpful. All of this on the dev
> ice itself.
>
>
>
>
> I understand some of these changes may be very difficult, especially given
> the low memory and computational power of
> secure elements. However I truly believe most of these points are a MUST
> have for any decent security. If you don't
> assume the computer on which the transaction is crafted is compromised,
> then you don't need a hardware wallet. If you
> assume it may be compromised, then the HW needs to be able to defend
> against those.
>
> Revault does not plan on building hardware wallets, we hope existing and
> upcoming manufacturers will implement a strong
> security that we could use for the Revault protocol users. Vault users
> will likely hold very large sums and would be
> happy to pay a high premium for more secure HW. This will hopefully
> encourage existing players to keep on improving
> their devices and that will ultimately benefit us all.
>
> Feel free to reply with your comments or adding suggestions, I am not a
> hardware wallet expert and would take criticism
> wit
> hout being offended.
>
> Kind Regards,
> Kevin Loaec
>
> [1]:
> https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2020-May/017835.html
> [2]: https://github.com/re-vault/practical-revault/issues/59
> [3]:
> https://lists.linuxfoundation.org/pipermail/lightning-dev/2020-January/002425.html
> [4]: https://btcguide.github.io/known-issues/verify-receive-address
> [5]: https://gitlab.com/lightning-signer/docs
> [6]:
> https://shiftcrypto.ch/blog/how-nearly-all-personal-hardware-wallet-multisig-setups-are-insecure/
>
> _______________________________________________
> bitcoin-dev mailing list
> bitcoin-dev@lists•linuxfoundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev
>

[-- Attachment #2: Type: text/html, Size: 13534 bytes --]