[Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Rusty Russell]

Title: Canonical Input and Output Ordering
Author: Rusty Russell <rusty@rustcorp.com.au>
Discussions-To: "Bitcoin Dev" <bitcoin-development@lists.sourceforge.net>
Status: Draft
Type: Standards Track
Created: 2015-06-06

Abstract

This BIP provides a canonical ordering of inputs and outputs when
creating transactions.

Motivation

Most bitcoin wallet implementations randomize the outputs of
transactions they create to avoid trivial linkage analysis (especially
change outputs), however implementations have made mistakes in this area
in the past.

Using a canonical ordering has the same effect, but is simpler, more
obvious if incorrect, and can eventually be enforced by IsStandard() and
even a soft-fork to enforce it.

Specification

Inputs should be ordered like so:
        index (lower value first)
        txid (little endian order, lower byte first)

Outputs should be ordered like so:
        amount (lower value first)
        script (starting from first byte, lower byte first, shorter wins)

Rationale

Any single wallet is already free to implement this, but if other
wallets do not it would reduce privacy by making those transactions
stand out.  Thus a BIP is appropriate, especially if this were to
become an IsStandard() rule once widely adopted.

Because integers are fast to compare, they're sorted first, before the
lexographical ordering.

The other input fields do not influence the sort order, as any valid
transactions cannot have two inputs with the same index and txid.

Reference Implementation

https://github.com/rustyrussell/bitcoin/tree/bip-in-out-ordering

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Mark Friedenbach]

[-- Attachment #1: Type: text/plain, Size: 2199 bytes --]

Rusty, this doesn't play well with SIGHASH_SINGLE which is used in
assurance contracts among other things. Sometimes the ordering is set by
the signing logic itself...
On Jun 5, 2015 9:43 PM, "Rusty Russell" <rusty@rustcorp.com.au> wrote:

> Title: Canonical Input and Output Ordering
> Author: Rusty Russell <rusty@rustcorp.com.au>
> Discussions-To: "Bitcoin Dev" <bitcoin-development@lists.sourceforge.net>
> Status: Draft
> Type: Standards Track
> Created: 2015-06-06
>
> Abstract
>
> This BIP provides a canonical ordering of inputs and outputs when
> creating transactions.
>
> Motivation
>
> Most bitcoin wallet implementations randomize the outputs of
> transactions they create to avoid trivial linkage analysis (especially
> change outputs), however implementations have made mistakes in this area
> in the past.
>
> Using a canonical ordering has the same effect, but is simpler, more
> obvious if incorrect, and can eventually be enforced by IsStandard() and
> even a soft-fork to enforce it.
>
> Specification
>
> Inputs should be ordered like so:
>         index (lower value first)
>         txid (little endian order, lower byte first)
>
> Outputs should be ordered like so:
>         amount (lower value first)
>         script (starting from first byte, lower byte first, shorter wins)
>
> Rationale
>
> Any single wallet is already free to implement this, but if other
> wallets do not it would reduce privacy by making those transactions
> stand out.  Thus a BIP is appropriate, especially if this were to
> become an IsStandard() rule once widely adopted.
>
> Because integers are fast to compare, they're sorted first, before the
> lexographical ordering.
>
> The other input fields do not influence the sort order, as any valid
> transactions cannot have two inputs with the same index and txid.
>
> Reference Implementation
>
> https://github.com/rustyrussell/bitcoin/tree/bip-in-out-ordering
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>

[-- Attachment #2: Type: text/html, Size: 3027 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Rusty Russell]

Mark Friedenbach <mark@friedenbach.org> writes:
> Rusty, this doesn't play well with SIGHASH_SINGLE which is used in
> assurance contracts among other things. Sometimes the ordering is set by
> the signing logic itself...

Ah, I forgot about that particular wart. Yech.  Implies that you can
order inputs or outputs, not both.

Something like "outputs must be in order, inputs which do not
CHECK(MULTI)SIG_(VERIFY) a SIGHASH_SINGLE sig must be in order with
respect to each other".  But that's much less trivial since it implies
script evaluation.

In other news, I noticed Kristov Atlas's concurrent proposal just after
I posted this (via reddit).  He used far more words, but didn't note
this issue either AFAICT.

Thanks!
Rusty.

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Wladimir J. van der Laan]

On Fri, Jun 05, 2015 at 09:46:17PM -0700, Mark Friedenbach wrote:
> Rusty, this doesn't play well with SIGHASH_SINGLE which is used in
> assurance contracts among other things. Sometimes the ordering is set by
> the signing logic itself...

But in that case (unconstrained) randomization can't be used either. This is posed as an alternative to randomization. So in that regard, the proposal still makes sense.
I think this move to verifyable, deterministic methods where possible is good.

Wladimir

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Mark Friedenbach]

[-- Attachment #1: Type: text/plain, Size: 675 bytes --]

Certainly, but I would drop discussion of IsStandard or consensus rules.
On Jun 6, 2015 1:24 AM, "Wladimir J. van der Laan" <laanwj@gmail.com> wrote:

> On Fri, Jun 05, 2015 at 09:46:17PM -0700, Mark Friedenbach wrote:
> > Rusty, this doesn't play well with SIGHASH_SINGLE which is used in
> > assurance contracts among other things. Sometimes the ordering is set by
> > the signing logic itself...
>
> But in that case (unconstrained) randomization can't be used either. This
> is posed as an alternative to randomization. So in that regard, the
> proposal still makes sense.
> I think this move to verifyable, deterministic methods where possible is
> good.
>
> Wladimir
>

[-- Attachment #2: Type: text/html, Size: 954 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Danny Thorpe]

[-- Attachment #1: Type: text/plain, Size: 2589 bytes --]

FWIW, The Open Assets colored coin protocol (CoinPrism) places special
significance on the zeroth input and the position of the OP_RETURN colored
coin marker output to distinguish colored coin issuance outputs from
transfer outputs. Reordering the inputs or the outputs breaks the colored
coin representation.

Recommending sorting of the inputs and outputs as a best practice is fine
(and better than random, IMO), but not as part of IsStandard() or consensus
rules.  There are cases where the order of the inputs and outputs is
significant.

-Danny

On Fri, Jun 5, 2015 at 9:42 PM, Rusty Russell <rusty@rustcorp.com.au> wrote:

> Title: Canonical Input and Output Ordering
> Author: Rusty Russell <rusty@rustcorp.com.au>
> Discussions-To: "Bitcoin Dev" <bitcoin-development@lists.sourceforge.net>
> Status: Draft
> Type: Standards Track
> Created: 2015-06-06
>
> Abstract
>
> This BIP provides a canonical ordering of inputs and outputs when
> creating transactions.
>
> Motivation
>
> Most bitcoin wallet implementations randomize the outputs of
> transactions they create to avoid trivial linkage analysis (especially
> change outputs), however implementations have made mistakes in this area
> in the past.
>
> Using a canonical ordering has the same effect, but is simpler, more
> obvious if incorrect, and can eventually be enforced by IsStandard() and
> even a soft-fork to enforce it.
>
> Specification
>
> Inputs should be ordered like so:
>         index (lower value first)
>         txid (little endian order, lower byte first)
>
> Outputs should be ordered like so:
>         amount (lower value first)
>         script (starting from first byte, lower byte first, shorter wins)
>
> Rationale
>
> Any single wallet is already free to implement this, but if other
> wallets do not it would reduce privacy by making those transactions
> stand out.  Thus a BIP is appropriate, especially if this were to
> become an IsStandard() rule once widely adopted.
>
> Because integers are fast to compare, they're sorted first, before the
> lexographical ordering.
>
> The other input fields do not influence the sort order, as any valid
> transactions cannot have two inputs with the same index and txid.
>
> Reference Implementation
>
> https://github.com/rustyrussell/bitcoin/tree/bip-in-out-ordering
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>

[-- Attachment #2: Type: text/html, Size: 3512 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Peter Todd]

[-- Attachment #1: Type: text/plain, Size: 969 bytes --]

On Mon, Jun 08, 2015 at 02:25:40PM -0700, Danny Thorpe wrote:
> FWIW, The Open Assets colored coin protocol (CoinPrism) places special
> significance on the zeroth input and the position of the OP_RETURN colored
> coin marker output to distinguish colored coin issuance outputs from
> transfer outputs. Reordering the inputs or the outputs breaks the colored
> coin representation.
> 
> Recommending sorting of the inputs and outputs as a best practice is fine
> (and better than random, IMO), but not as part of IsStandard() or consensus
> rules.  There are cases where the order of the inputs and outputs is
> significant.

Timestamping is another case where order matters: if you put the digest
in the last vout you can use SHA256 midstate's to reduce the size of the
timestamp proof.

Anyway, there's no reason to rush re: changes to IsStandard()

-- 
'peter'[:-1]@petertodd.org
0000000000000000127ab1d576dc851f374424f1269c4700ccaba2c42d97e778

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 650 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Gregory Maxwell]

On Mon, Jun 8, 2015 at 9:25 PM, Danny Thorpe <danny.thorpe@gmail.com> wrote:
> Recommending sorting of the inputs and outputs as a best practice is fine
> (and better than random, IMO), but not as part of IsStandard() or consensus
> rules.  There are cases where the order of the inputs and outputs is
> significant.

Is it your opinion that its fine if the result is that it makes the
usage trivially distinguishable e.g. where it might be subjected to
higher tx fees, or might break some software which incorrectly expects
all transactions to be ordered since most are?

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Gregory Maxwell]

On Sat, Jun 6, 2015 at 4:42 AM, Rusty Russell <rusty@rustcorp.com.au> wrote:
> Title: Canonical Input and Output Ordering
> Author: Rusty Russell <rusty@rustcorp.com.au>
> Discussions-To: "Bitcoin Dev" <bitcoin-development@lists.sourceforge.net>
> Status: Draft
> Type: Standards Track
> Created: 2015-06-06
>
> Abstract
>
> This BIP provides a canonical ordering of inputs and outputs when
> creating transactions.
>
> Motivation
>
> Most bitcoin wallet implementations randomize the outputs of
> transactions they create to avoid trivial linkage analysis (especially
> change outputs), however implementations have made mistakes in this area
> in the past.
>
> Using a canonical ordering has the same effect, but is simpler, more
> obvious if incorrect, and can eventually be enforced by IsStandard() and
> even a soft-fork to enforce it.
>
> Specification
>
> Inputs should be ordered like so:
>         index (lower value first)
>         txid (little endian order, lower byte first)
>
> Outputs should be ordered like so:
>         amount (lower value first)
>         script (starting from first byte, lower byte first, shorter wins)
>
> Rationale
>
> Any single wallet is already free to implement this, but if other
> wallets do not it would reduce privacy by making those transactions
> stand out.  Thus a BIP is appropriate, especially if this were to
> become an IsStandard() rule once widely adopted.
>
> Because integers are fast to compare, they're sorted first, before the
> lexographical ordering.
>
> The other input fields do not influence the sort order, as any valid
> transactions cannot have two inputs with the same index and txid.
>
> Reference Implementation
>
> https://github.com/rustyrussell/bitcoin/tree/bip-in-out-ordering


Sorry I wasn't a part of the IRC conversation where this was first
discussed, though I'm very happy to see a concrete implementation
along with the proposal.

I'm not a great fan of this proposal for two reasons: The first is
that the strict ordering requirements is incompatible with future
soft-forks that may expose additional ordering constraints. Today we
have _SINGLE, which as noted this interacts with poorly, but there
have been other constraints proposed that this would also interact
with poorly.

The second is that even absent consensus rules there may be invisible
constraints in systems-- e.g. hardware wallets that sign top down, or
future transaction covenants that have constraints about ordering,  or
proof systems that use (yuck) midstate compression for efficiency.
Right now, with random ordering these applications are fairly
indistinguishable from other random uses (since their imposed order
could come about by chance) but if everyone else was ordered, even if
wasn't enforced.. these would be highly distinguishable. Which would
be unfortunate.  Worse, if most transactions are ordered the few that
aren't could be mishandled (which is, I assume, part of why you
propose requiring the ordering-- but I think the soft fork constraints
there hurt it more).

[Sorry for the delay in stating my views here; I wanted to talk them
over with a few other people to see if I was just being stupid and
misunderstanding the proposal]

I think perhaps the motivations here are understated. We have not seen
any massive deployments of accidentally broken ordering that I'm aware
of-- and an implementation that got this wrong in a harmful way would
likely make far more fatal mistakes (e.g. non random private keys).
As an alternative to this proposal the ordering can be privately
derandomized in the same way DSA is, to avoid the need for an actual
number source.  If getting the randomness right were really the only
motivation, I'd suggest we propose a simple derandomized randomization
algorithm--- e.g. take the order from (H(input ids||client secret)).

I think there is actually an unstated motivation also driving this
(and the other) proposal related to collaborative transaction systems
like coinjoins or micropayment channels; where multiple clients need
to agree on the same ordering. Is this the case? If so we should
probably talk through some of the requirements there and see if there
isn't a better way to address it.

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Rusty Russell]

Gregory Maxwell <gmaxwell@gmail.com> writes:
> I'm not a great fan of this proposal for two reasons: The first is
> that the strict ordering requirements is incompatible with future
> soft-forks that may expose additional ordering constraints. Today we
> have _SINGLE, which as noted this interacts with poorly, but there
> have been other constraints proposed that this would also interact
> with poorly.

Yes, I hit this when I implemented an IsStandard change; upon input
evaluation the scriptsigs which used _SINGLE get disregarded from
ordering.  

> The second is that even absent consensus rules there may be invisible
> constraints in systems-- e.g. hardware wallets that sign top down,

I think that one's pretty easy to fix (and they should fix it anyway, to
avoid leaking information due to ordering): they can receive an
unordered tx and sign it as if it were ordered canonically.

> or
> future transaction covenants that have constraints about ordering,  or
> proof systems that use (yuck) midstate compression for efficiency.

The softfork argument I find the most compelling, though it's tempting
to argue that every ordering use (including SIGHASH_SINGLE) is likely
a mistake.

> I think perhaps the motivations here are understated. We have not seen
> any massive deployments of accidentally broken ordering that I'm aware
> of-- and an implementation that got this wrong in a harmful way would
> likely make far more fatal mistakes (e.g. non random private keys).

I was prompted to propose something by this:

https://blog.blocktrail.com/2015/05/getting-your-change-in-order/

If that's the only one though, it's less compelling.

> As an alternative to this proposal the ordering can be privately
> derandomized in the same way DSA is, to avoid the need for an actual
> number source.  If getting the randomness right were really the only
> motivation, I'd suggest we propose a simple derandomized randomization
> algorithm--- e.g. take the order from (H(input ids||client secret)).
>
> I think there is actually an unstated motivation also driving this
> (and the other) proposal related to collaborative transaction systems
> like coinjoins or micropayment channels; where multiple clients need
> to agree on the same ordering. Is this the case? If so we should
> probably talk through some of the requirements there and see if there
> isn't a better way to address it.

Indeed.  I was implementing deterministic permutations for lightning
(signature exchange requires both sides agree on ordering).

Cheers,
Rusty.

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Gregory Maxwell]

On Mon, Jun 15, 2015 at 2:29 AM, Rusty Russell <rusty@rustcorp.com.au> wrote:
> The softfork argument I find the most compelling, though it's tempting
> to argue that every ordering use (including SIGHASH_SINGLE) is likely
> a mistake.

Oh.

Hm.

It is the case that the generalized sighash flag design I was thinking
about was actually completely neutral about ordering, and yet still
replaced SINGLE.

I need to think a bit on that.

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Mark Friedenbach]

[-- Attachment #1: Type: text/plain, Size: 1578 bytes --]

There's another important use case which you mentioned Greg, that also
requires special exemption: compact commitments via mid-state compression.

The use case is an OP_RETURN output sorted last, whose last N bytes are a
commitment of some kind. A proof of the commitment can then use mid state
compression to elide the beginning of the transaction.

How do you make a special exemption for this category of outputs? I can't
think of a very clean way of doing so that doesn't require an ugly
advertising of sort-order exemptions.

The fact that we have two different existing use cases which conflict with
soft-fork enforcement, I'm quiet concerned that there are either other
things we aren't thinking of or haven't invented yet which would be
affected.

On Sun, Jun 14, 2015 at 7:33 PM, Gregory Maxwell <gmaxwell@gmail.com> wrote:

> On Mon, Jun 15, 2015 at 2:29 AM, Rusty Russell <rusty@rustcorp.com.au>
> wrote:
> > The softfork argument I find the most compelling, though it's tempting
> > to argue that every ordering use (including SIGHASH_SINGLE) is likely
> > a mistake.
>
> Oh.
>
> Hm.
>
> It is the case that the generalized sighash flag design I was thinking
> about was actually completely neutral about ordering, and yet still
> replaced SINGLE.
>
> I need to think a bit on that.
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>

[-- Attachment #2: Type: text/html, Size: 2305 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Rusty Russell]

Mark Friedenbach <mark@friedenbach.org> writes:
> There's another important use case which you mentioned Greg, that also
> requires special exemption: compact commitments via mid-state compression.
>
> The use case is an OP_RETURN output sorted last, whose last N bytes are a
> commitment of some kind. A proof of the commitment can then use mid state
> compression to elide the beginning of the transaction.
>
> How do you make a special exemption for this category of outputs? I can't
> think of a very clean way of doing so that doesn't require an ugly
> advertising of sort-order exemptions.

Yes, we can suit any one use case, but not all of them.

For example, outputs shall be sorted by:
        1.  First byte (or 0 if script is zero length) minus 107.
        2.  The remainder of the script in lexographical order.

This would put OP_RETURN outputs last.

Though Peter Todd's more general best-effort language might make more
sense.  It's not like you can hide an OP_RETURN transaction to make it
look like something else, so that transaction not going to be
distinguished by non-canonical ordering.

Cheers,
Rusty.

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Jorge Timón]

[-- Attachment #1: Type: text/plain, Size: 993 bytes --]

On Jun 15, 2015 11:43 PM, "Rusty Russell" <rusty@rustcorp.com.au> wrote:

> Though Peter Todd's more general best-effort language might make more
> sense.  It's not like you can hide an OP_RETURN transaction to make it
> look like something else, so that transaction not going to be
> distinguished by non-canonical ordering.

What about commitments that don't use op_return (ie pay2contract
commitments)?

In any case, if the motivation is ordering in multi-party transactions
there should be ways to do it without any consequences for other
transaction types' privacy. For example you could have a deterministic
method that depends on a random seed all parties in the transaction
previously share. That way the ordering is deterministic for all parties
involved in the transaction (which can use whatever channel they're using
to send the parts to also send this random seed) while at the same time the
order looks random (or at least not cannonical in a recognisable way) to
everyone else.

[-- Attachment #2: Type: text/html, Size: 1185 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Rusty Russell]

Jorge Timón <jtimon@jtimon.cc> writes:
> On Jun 15, 2015 11:43 PM, "Rusty Russell" <rusty@rustcorp.com.au> wrote:
>
>> Though Peter Todd's more general best-effort language might make more
>> sense.  It's not like you can hide an OP_RETURN transaction to make it
>> look like something else, so that transaction not going to be
>> distinguished by non-canonical ordering.
>
> What about commitments that don't use op_return (ie pay2contract
> commitments)?

I have no idea what they are? :)

> In any case, if the motivation is ordering in multi-party transactions
> there should be ways to do it without any consequences for other
> transaction types' privacy. For example you could have a deterministic
> method that depends on a random seed all parties in the transaction
> previously share. That way the ordering is deterministic for all parties
> involved in the transaction (which can use whatever channel they're using
> to send the parts to also send this random seed) while at the same time the
> order looks random (or at least not cannonical in a recognisable way) to
> everyone else.

Yes, my plan B would be an informational bip with simple code,
suggesting a way to permute a transaction based on some secret.  No
point everyone reinventing the wheel, badly.

Cheers,
Rusty.

[Bitcoin-development] Fwd: Re: [RFC] Canonical input and output ordering in transactions

[Jorge Timón]

[-- Attachment #1: Type: text/plain, Size: 1340 bytes --]

---------- Forwarded message ----------
From: "Jorge Timón" <jtimon@jtimon.cc>
Date: Jun 17, 2015 6:59 PM
Subject: Re: [Bitcoin-development] [RFC] Canonical input and output
ordering in transactions
To: "Rusty Russell" <rusty@rustcorp.com.au>
Cc:

On Tue, Jun 16, 2015 at 10:06 AM, Rusty Russell <rusty@rustcorp.com.au>
wrote:
> Jorge Timón <jtimon@jtimon.cc> writes:
>> On Jun 15, 2015 11:43 PM, "Rusty Russell" <rusty@rustcorp.com.au> wrote:
>>
>>> Though Peter Todd's more general best-effort language might make more
>>> sense.  It's not like you can hide an OP_RETURN transaction to make it
>>> look like something else, so that transaction not going to be
>>> distinguished by non-canonical ordering.
>>
>> What about commitments that don't use op_return (ie pay2contract
>> commitments)?
>
> I have no idea what they are? :)

Here's a short explanation and the code:

https://github.com/Blockstream/contracthashtool

Here's a longer explanation with a concrete use case (the contract is
the invoice):

https://www.youtube.com/watch?v=qwyALGlG33Q

> Yes, my plan B would be an informational bip with simple code,
> suggesting a way to permute a transaction based on some secret.  No
> point everyone reinventing the wheel, badly.

Great. Well, then all I'm saying is that I like this as plan A.

[-- Attachment #2: Type: text/html, Size: 2002 bytes --]

Re: [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Kristov Atlas]

[-- Attachment #1: Type: text/plain, Size: 2913 bytes --]

On Sun, Jun 14, 2015 at 7:02 PM, Gregory Maxwell <gmaxwell@gmail.com> wrote:

> I'm not a great fan of this proposal for two reasons: The first is
> that the strict ordering requirements is incompatible with future
> soft-forks that may expose additional ordering constraints. Today we
> have _SINGLE, which as noted this interacts with poorly, but there
> have been other constraints proposed that this would also interact
> with poorly.
>

I'm not clear on why this is a problem, so long as the canonical ordering
BIP is *optional*. Unless there is a specific plan to soft fork a change
that would break the BIP and it is fairly imminent, I see this only as a
reason not to integrate it into isStandard().


> The second is that even absent consensus rules there may be invisible
> constraints in systems-- e.g. hardware wallets that sign top down, or
> future transaction covenants that have constraints about ordering,  or
> proof systems that use (yuck) midstate compression for efficiency. Right
> now, with random ordering these applications are fairly
> indistinguishable from other random uses (since their imposed order
> could come about by chance) but if everyone else was ordered, even if
> wasn't enforced.. these would be highly distinguishable. Which would
> be unfortunate.


Maybe they shouldn't be doing that. :-P


> I think perhaps the motivations here are understated. We have not seen
> any massive deployments of accidentally broken ordering that I'm aware
> of-- and an implementation that got this wrong in a harmful way would
> likely make far more fatal mistakes (e.g. non random private keys).
>

In my reading of various wallet client sources, it is common that wallet
clients will use cryptographically weak sources of randomness to sort
outputs -- that is, the ones that actually bother to randomly sort. I can
hunt down some examples if this would substantially contribute to the
discussion.

As an alternative to this proposal the ordering can be privately
> derandomized in the same way DSA is, to avoid the need for an actual
> number source.  If getting the randomness right were really the only
> motivation, I'd suggest we propose a simple derandomized randomization
> algorithm--- e.g. take the order from (H(input ids||client secret)).
>

This sounds similar to an idea that Sergio pitched to me privately, which
was for wallets to have a private sorting key that they can use to sort
inputs and outputs. However, I suspect that adding yet another key which
will necessarily require special key rotation rules and maybe special
backup procedures will mean that this standard will not be widely adopted
any time soon. Ideally, I'd like to see someone write a different BIP with
the sorting key idea and let them compete in the wallet client market
rather than trying to anticipate what is best for all clients and
distilling two good ideas into one artificially.

-Kristov

[-- Attachment #2: Type: text/html, Size: 3860 bytes --]

Re: [bitcoin-dev] [Bitcoin-development] [RFC] Canonical input and output ordering in transactions

[Kristov Atlas]

[-- Attachment #1: Type: text/plain, Size: 6597 bytes --]

Following up on this topic...

gmaxwell has reserved BIP 69 for my proposal.

It has been implemented by Electrum in v2.3.2:
https://github.com/spesmilo/electrum/blob/master/RELEASE-NOTES

Rusty has kindly tweaked his original canonical ordering proposal
implementation for Bitcoin Core's wallet client to fit my proposal:
https://github.com/rustyrussell/bitcoin/tree/bip-69 (needs testing)

Outstanding objections appear to me to boil down to two points:

1) Some transactions cannot comply with this BIP because there are input
and/or put index dependencies.

My response: No big deal, it's informational. They simply won't be
compliant with the BIP, and that's fine with me.

2) If we set a standard now for transactions that is not apparently random
ordering from the perspective of passive blockchain observers, transactions
that can't comply with this BIP will stand out. Also, if we change our
collective minds in the future about how ordering should be handled, those
future transactions would stand out. Therefore, the "safe" course of action
is to come up with another scheme that appears to be random ordering from
the perspective of a passive blockchain observer.

My response: Apparently-random but owner-verifiable ordering is doable.
Discussion of this has revolved around what I have called a "sorting key"
-- sort lexicographically, and then reorder according to the bits in a
sorting key that is impossible to predict by an attacker. This means
passive observers cannot determine anything meaningful about the
transaction (e.g. which output is change, information leaked based on utxo
selection algorithm for inputs, etc.) but the owner of the funds and the
sorting key can verify that his transaction matches the canonical
specification. Ideally, I think the key should rotate for each transaction
to avoid the possibility that a static key can link multiple transactions
together. The key should be rotated in such a fashion that the next
iteration is not predictable to anyone except the key holder (e.g. put the
key through a secure pseudo-random function for each new iteration). This
could be done by generating a few bytes of entropy upon wallet creation and
keeping track of the current iteration of rotation. HD wallets could derive
the initial state of the sorting key by deriving it from the HD seed. There
are a variety of schemes that could work here.

My main objection to this family of approaches at present is complexity. I
suspect that many wallet clients will not want to implement the BIP if they
have to maintain a sorting key.

A second objection is that no one will be able to detect anomalies in BIP
compliance except for the sorting key holder. Most users probably will not
bother to verify this. For code reviewers, this means that the sorting key
is yet another aspect of the code base that must be scrutinized to ensure
it is not being used as a covert channel or has been underhandedly weakened
in some fashion.

Also, I will mention an ancillary benefit of a non-random canonical
ordering: it makes unit testing of transactions for Bitcoin wallets simpler.

Given all of the above, I will reiterate my preference to keep the proposal
as it is now. The pull request is here:
https://github.com/bitcoin/bips/pull/157

If there is market demand for it, a separate sorting key-based proposal
could be written which can compete with this BIP and over time successfully
deprecate it. I would currently envision that as an HD BIP with a new
purpose code.

-Kristov


On Mon, Jun 15, 2015 at 12:01 AM, Kristov Atlas <
kristovatlas.lists@gmail.com> wrote:

> On Sun, Jun 14, 2015 at 7:02 PM, Gregory Maxwell <gmaxwell@gmail.com>
> wrote:
>
>> I'm not a great fan of this proposal for two reasons: The first is
>> that the strict ordering requirements is incompatible with future
>> soft-forks that may expose additional ordering constraints. Today we
>> have _SINGLE, which as noted this interacts with poorly, but there
>> have been other constraints proposed that this would also interact
>> with poorly.
>>
>
> I'm not clear on why this is a problem, so long as the canonical ordering
> BIP is *optional*. Unless there is a specific plan to soft fork a change
> that would break the BIP and it is fairly imminent, I see this only as a
> reason not to integrate it into isStandard().
>
>
>> The second is that even absent consensus rules there may be invisible
>> constraints in systems-- e.g. hardware wallets that sign top down, or
>> future transaction covenants that have constraints about ordering,  or
>> proof systems that use (yuck) midstate compression for efficiency. Right
>> now, with random ordering these applications are fairly
>> indistinguishable from other random uses (since their imposed order
>> could come about by chance) but if everyone else was ordered, even if
>> wasn't enforced.. these would be highly distinguishable. Which would
>> be unfortunate.
>
>
> Maybe they shouldn't be doing that. :-P
>
>
>> I think perhaps the motivations here are understated. We have not seen
>> any massive deployments of accidentally broken ordering that I'm aware
>> of-- and an implementation that got this wrong in a harmful way would
>> likely make far more fatal mistakes (e.g. non random private keys).
>>
>
> In my reading of various wallet client sources, it is common that wallet
> clients will use cryptographically weak sources of randomness to sort
> outputs -- that is, the ones that actually bother to randomly sort. I can
> hunt down some examples if this would substantially contribute to the
> discussion.
>
> As an alternative to this proposal the ordering can be privately
>> derandomized in the same way DSA is, to avoid the need for an actual
>> number source.  If getting the randomness right were really the only
>> motivation, I'd suggest we propose a simple derandomized randomization
>> algorithm--- e.g. take the order from (H(input ids||client secret)).
>>
>
> This sounds similar to an idea that Sergio pitched to me privately, which
> was for wallets to have a private sorting key that they can use to sort
> inputs and outputs. However, I suspect that adding yet another key which
> will necessarily require special key rotation rules and maybe special
> backup procedures will mean that this standard will not be widely adopted
> any time soon. Ideally, I'd like to see someone write a different BIP with
> the sorting key idea and let them compete in the wallet client market
> rather than trying to anticipate what is best for all clients and
> distilling two good ideas into one artificially.
>
> -Kristov
>
>

[-- Attachment #2: Type: text/html, Size: 8608 bytes --]