[bitcoin-dev] BIP clearing house addresses

[Matthew Roberts via bitcoin-dev]

In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement layer
that can be used to revoke transactions?

You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a transaction's
progress was publicly visible so transparent fraud prevention and auditing
would become possible by anyone.

The reason why I bring this up is existing OP codes and TX types don't seem
suitable for a secure clearing mechanism; Nlocktimed TXs won't work for
this since you can't know ahead of time when and where a withdrawal needs
to be made, plus there's still the potential for key mismanagement; Similar
problems with OP_CHECKLOCKTIMEVERIFY apply too – unless you keep a private
key around on the server which would defeat the purpose. The main use case
here, would be specifically to improve centralized exchange security by
making it impossible for a hot wallet to be raided all at once.

Thoughts?

Some existing background:

http://hackingdistributed.com/2016/08/03/how-bitfinex-heist-could-have-been-avoided/
-- Proposed the basic idea for a time-based clearing house but using
blockchains directly, this is a much better idea than my own.

roberts.pm/timechain -- My original paper written in 2015 which proposed a
similar idea for secure wallet design but implemented using time-locked
ECDSA keys. Obviously a blockchain would work better for this.

Other -- if the idea has already been brought up by other people, I
apologize.

[Tier Nolan via bitcoin-dev]

On Wed, Aug 3, 2016 at 7:16 PM, Matthew Roberts via bitcoin-dev <

Post by Matthew Roberts via bitcoin-dev
The reason why I bring this up is existing OP codes and TX types don't
seem suitable for a secure clearing mechanism;

I think reversing transactions is not likely to be acceptable. You could
add an opcode that requires that an output be set to something.

[target script] SPENDTO

This would require that [target script] is the script for the corresponding
output. This is a purely local check.

For example, if SPENDTO executes as part of the script for input 3, then it
checks that output 3 uses the given script as its scriptPubKey. The value
of input 3 and output 3 would have to be the same too.

This allows check sequence verify to be used to lock the spending script
for a while. This doesn't allow reversal, but would give a 24 hour window
where the spenders can reverse the transaction.

[IF <1 day> CSV DROP <live public key> CHECKSIG ELSE <offline protected
key> CHECKSIG] SPENDTO <live public key2> CHECKSIG

Someone with the live public key can create a transaction that spends the
funds to the script in the square brackets.

Once that transaction hits the blockchain, then someone with the <offline
protected key> has 24 hours to spend the output before the person with the
live keys can send the funds onward.

[Matthew Roberts via bitcoin-dev]

This would honestly work. It forces the attacker to go through with the
clearing phase which simultaneously makes it possible to "cancel" the TX
through another logic branch before the timeout occurs. I'd say that would
meet the needs of a clearing mechanism / fraud prevention system for an
exchange perfectly while requiring minimal changes to the software.

Very, very smart idea. A++, would read again.

On Thu, Aug 4, 2016 at 9:55 AM, Tier Nolan via bitcoin-dev <

Post by Tier Nolan via bitcoin-dev
On Wed, Aug 3, 2016 at 7:16 PM, Matthew Roberts via bitcoin-dev <

Post by Matthew Roberts via bitcoin-dev
The reason why I bring this up is existing OP codes and TX types don't
seem suitable for a secure clearing mechanism;

I think reversing transactions is not likely to be acceptable. You could
add an opcode that requires that an output be set to something.
[target script] SPENDTO
This would require that [target script] is the script for the
corresponding output. This is a purely local check.
For example, if SPENDTO executes as part of the script for input 3, then
it checks that output 3 uses the given script as its scriptPubKey. The
value of input 3 and output 3 would have to be the same too.
This allows check sequence verify to be used to lock the spending script
for a while. This doesn't allow reversal, but would give a 24 hour window
where the spenders can reverse the transaction.
[IF <1 day> CSV DROP <live public key> CHECKSIG ELSE <offline protected
key> CHECKSIG] SPENDTO <live public key2> CHECKSIG
Someone with the live public key can create a transaction that spends the
funds to the script in the square brackets.
Once that transaction hits the blockchain, then someone with the <offline
protected key> has 24 hours to spend the output before the person with the
live keys can send the funds onward.
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[Luke Dashjr via bitcoin-dev]

On Wednesday, August 03, 2016 6:16:20 PM Matthew Roberts via bitcoin-dev

Post by Matthew Roberts via bitcoin-dev
In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement layer
that can be used to revoke transactions?

This isn't something that makes sense at the address, since it represents the
recipient and not the sender. Transactions are not sent from addresses ever.

Post by Matthew Roberts via bitcoin-dev
You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a transaction's
progress was publicly visible so transparent fraud prevention and auditing
would become possible by anyone.

This is already possible. Just nLockTime your withdrawls for some future
block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip.

Luke

[Matthew Roberts via bitcoin-dev]

Post by Luke Dashjr via bitcoin-dev
This is already possible. Just nLockTime your withdrawls for some future

block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip.

The problem with nLockTimed transactions is a centralized exchange isn't
going to know ahead of time where those locked transactions need to go or
the amount that needs to be signed so you will end up having to keep the
private key around. If there was a way to create these transactions offline
with special SIG_HASH flags (and I don't think there is) there's nothing
about nLockTime that forces that the transactions be broadcast straight
away and plus: since the TXs aren't confirmed until the lock-time expires
they can be overwritten anyway.

I think given the requirements that a centralized exchange has: TierNolan's
idea is the best so far. Essentially, you have a new type of output script
that forces the redeemer to use a designated output script template in the
redeeming transaction, meaning that you can actually force people to send
coins into another transaction with "relative lock-timed" outputs. The new
transaction can then only be redeemed at the destination after the relative
lock-time expires OR it can be moved before-hand to a designated off-line
recovery address. This is much better than creating a new transaction
system, IMO.

Post by Luke Dashjr via bitcoin-dev
And the refund TXN would need to be able to go to a new address entirely.

Agreed.

Post by Luke Dashjr via bitcoin-dev
"This is already possible. Just nLockTime your withdrawls for some future
block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip."
This would have prevented the Bitfinex hack if BitGo did this, but it
wouldn't have helped if the Bitfinex offline key had been compromised
instead of BitGo doing the 2nd sig. In the BFX hack the TXNs were signed
by Bitfinex's hot key and BitGo's key, they required 2 of 2.
If I'm understanding correctly, what Matthew is proposing is a new type of
UTXO that is only valid to be spent as an nLockTime transaction and can be
reversed by some sort of RBF-type transaction within that time period, I
believe.
But I don't think this will work. What do you do if the keys are
compromised? What's to stop the attacker from locking the coins up
indefinitely by repeatedly broadcasting a refund transaction each time you
try to spend to an uncompromised address?
You'd need a third distinct key required for the refund TXN that's
separate from the keys used to sign the initial nLockTime TXN. And the
refund TXN would need to be able to go to a new address entirely.
On Aug 3, 2016 11:28 PM, "Luke Dashjr via bitcoin-dev" <

Post by Luke Dashjr via bitcoin-dev
On Wednesday, August 03, 2016 6:16:20 PM Matthew Roberts via bitcoin-dev

Post by Matthew Roberts via bitcoin-dev
In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement layer
that can be used to revoke transactions?

This isn't something that makes sense at the address, since it represents the
recipient and not the sender. Transactions are not sent from addresses ever.

Post by Matthew Roberts via bitcoin-dev
You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a

transaction's

Post by Matthew Roberts via bitcoin-dev
progress was publicly visible so transparent fraud prevention and

auditing

Post by Matthew Roberts via bitcoin-dev
would become possible by anyone.

This is already possible. Just nLockTime your withdrawls for some future
block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip.
Luke
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[Erik Aronesty via bitcoin-dev]

https://www.reddit.com/r/Bitcoin/comments/4w016b/use_of_payment_channels_to_mitigate_exchange_risk/

On Thu, Aug 4, 2016 at 12:53 AM, Matthew Roberts via bitcoin-dev <

Post by Luke Dashjr via bitcoin-dev

Post by Luke Dashjr via bitcoin-dev
This is already possible. Just nLockTime your withdrawls for some future

block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip.
The problem with nLockTimed transactions is a centralized exchange isn't
going to know ahead of time where those locked transactions need to go or
the amount that needs to be signed so you will end up having to keep the
private key around. If there was a way to create these transactions offline
with special SIG_HASH flags (and I don't think there is) there's nothing
about nLockTime that forces that the transactions be broadcast straight
away and plus: since the TXs aren't confirmed until the lock-time expires
they can be overwritten anyway.
TierNolan's idea is the best so far. Essentially, you have a new type of
output script that forces the redeemer to use a designated output script
template in the redeeming transaction, meaning that you can actually force
people to send coins into another transaction with "relative lock-timed"
outputs. The new transaction can then only be redeemed at the destination
after the relative lock-time expires OR it can be moved before-hand to a
designated off-line recovery address. This is much better than creating a
new transaction system, IMO.

Post by Luke Dashjr via bitcoin-dev
And the refund TXN would need to be able to go to a new address entirely.

Agreed.

Post by Luke Dashjr via bitcoin-dev
"This is already possible. Just nLockTime your withdrawls for some future
block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip."
This would have prevented the Bitfinex hack if BitGo did this, but it
wouldn't have helped if the Bitfinex offline key had been compromised
instead of BitGo doing the 2nd sig. In the BFX hack the TXNs were signed
by Bitfinex's hot key and BitGo's key, they required 2 of 2.
If I'm understanding correctly, what Matthew is proposing is a new type
of UTXO that is only valid to be spent as an nLockTime transaction and can
be reversed by some sort of RBF-type transaction within that time period, I
believe.
But I don't think this will work. What do you do if the keys are
compromised? What's to stop the attacker from locking the coins up
indefinitely by repeatedly broadcasting a refund transaction each time you
try to spend to an uncompromised address?
You'd need a third distinct key required for the refund TXN that's
separate from the keys used to sign the initial nLockTime TXN. And the
refund TXN would need to be able to go to a new address entirely.
On Aug 3, 2016 11:28 PM, "Luke Dashjr via bitcoin-dev" <

Post by Luke Dashjr via bitcoin-dev
On Wednesday, August 03, 2016 6:16:20 PM Matthew Roberts via bitcoin-dev

Post by Matthew Roberts via bitcoin-dev
In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement

layer

Post by Matthew Roberts via bitcoin-dev
that can be used to revoke transactions?

This isn't something that makes sense at the address, since it represents the
recipient and not the sender. Transactions are not sent from addresses ever.

Post by Matthew Roberts via bitcoin-dev
You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a

transaction's

Post by Matthew Roberts via bitcoin-dev
progress was publicly visible so transparent fraud prevention and

auditing

Post by Matthew Roberts via bitcoin-dev
would become possible by anyone.

This is already possible. Just nLockTime your withdrawls for some future
block. Don't sign any transaction that isn't nLockTime'd at least N blocks
beyond the present tip.
Luke
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[Troy Benjegerdes via bitcoin-dev]

Post by Matthew Roberts via bitcoin-dev
In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement layer
that can be used to revoke transactions?

I think many of us who think about human - computer interactions see the
need for a well defined process to roll back unexpected behavior in a computer
system. My 2014 era proposal is https://bitbucket.org/tmagik/catoshi/issues/24

The fundamental assumption around cryptocoins is you have a secret (private
key) known only by you. Currently in bitcoin if that assumption changes, the
response is blame the user. 'Incompetence, etc, etc'

This is bad business. For any cryptocurrency to really get mass market, we
need to provide our users with key revocation, to be used when the assumption
about being the only holder of a secret is broken.

I think there's a hardfork-worthy choice here:

1) implement reversal/revocation as an add-on feature
2) implement reversal/revocation as a fundamental that every address gets.

Ethereum made a quick hardfork choice to reverse a *single* instance of
unexpected behavior, and looks a lot like a bank bailout. We have the chance
to learn from this mistake, and, apparently, make a lot of money trading
on both sides of the hardfork.

Post by Matthew Roberts via bitcoin-dev
You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a transaction's
progress was publicly visible so transparent fraud prevention and auditing
would become possible by anyone.
The reason why I bring this up is existing OP codes and TX types don't seem
suitable for a secure clearing mechanism; Nlocktimed TXs won't work for
this since you can't know ahead of time when and where a withdrawal needs
to be made, plus there's still the potential for key mismanagement; Similar
problems with OP_CHECKLOCKTIMEVERIFY apply too ??? unless you keep a private
key around on the server which would defeat the purpose. The main use case
here, would be specifically to improve centralized exchange security by
making it impossible for a hot wallet to be raided all at once.
Thoughts?
http://hackingdistributed.com/2016/08/03/how-bitfinex-heist-could-have-been-avoided/
-- Proposed the basic idea for a time-based clearing house but using
blockchains directly, this is a much better idea than my own.
roberts.pm/timechain -- My original paper written in 2015 which proposed a
similar idea for secure wallet design but implemented using time-locked
ECDSA keys. Obviously a blockchain would work better for this.
Other -- if the idea has already been brought up by other people, I
apologize.
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[s7r via bitcoin-dev]

Hi,

I can clearly see some advantages for such a feature, but it's kind of
in conflict with Bitcoin's fundamental design. This design might be
problematic when it comes to hacks/thefts, but it's what gives Bitcoin
strength and make it differentiate from other currencies:

* reversal of transactions is impossible
* keep private keys private and safe. Lose them, it's like losing cash,
you can just forget about it.
* while we try hard to make 0-conf as safe as possible (if there's no
RBF flag on the transaction), we make it almost impossible or very very
expensive to reverse a confirmed transaction.

Also, we don't have a clear way to properly decide a good settlement
period length. It doesn't fix the problem any more than nLockTime fixes
it -- you can't know ahead of time when a withdraw needs to be made.
Fair enough, but even if the withdraw is made with a settlement layer,
will the user be able to spend it further immediately? Who will accept
such an input and treat it as a payment if it can be reversed during the
settlement layer? So, if you can't know ahead of time when a withdraw
needs to be made (nLockTime) how can you know ahead of time+settlement
period when a transaction needs to be declared irrevocable?

The linked page describes that merchants will never accept payments from
'vaults', and it will take 24 hours for coins to be irreversible moved
outside the 'vault'. This covers the part "is the user able to spend a
transaction with settlement layer" but it has security properties equal
to nLockTime = 24 hours - you can't benefit and use the coins
immediately and in 24 hours price might go up or down in an undesirable
way for a certain user. It however raises a lot of other questions: what
if the attacker manages to steal both the private key and vault key (we
have strong reasons to assume this can happen: if you can't keep a
private key safe, why would you be able to keep the vault key any
safer?) and starts a race with the actual user to unlock and lock back
the vault?

I think this is a wrong approach. hacks and big losses are sad, but all
the time users / exchanges are to blame for wrong implementations or
terrible security practices.

Thanks!

Post by Matthew Roberts via bitcoin-dev
In light of the recent hack: what does everyone think of the idea of
creating a new address type that has a reversal key and settlement layer
that can be used to revoke transactions?
You could specify so that transactions "sent" from these addresses must
receive N confirmations before they can't be revoked, after which the
transaction is "settled" and the coins become redeemable from their
destination output. A settlement phase would also mean that a
transaction's progress was publicly visible so transparent fraud
prevention and auditing would become possible by anyone.
The reason why I bring this up is existing OP codes and TX types don't
seem suitable for a secure clearing mechanism; Nlocktimed TXs won't work
for this since you can't know ahead of time when and where a withdrawal
needs to be made, plus there's still the potential for key
mismanagement; Similar problems with OP_CHECKLOCKTIMEVERIFY apply too –
unless you keep a private key around on the server which would defeat
the purpose. The main use case here, would be specifically to improve
centralized exchange security by making it impossible for a hot wallet
to be raided all at once.
Thoughts?
http://hackingdistributed.com/2016/08/03/how-bitfinex-heist-could-have-been-avoided/
-- Proposed the basic idea for a time-based clearing house but using
blockchains directly, this is a much better idea than my own.
roberts.pm/timechain <http://roberts.pm/timechain> -- My original paper
written in 2015 which proposed a similar idea for secure wallet design
but implemented using time-locked ECDSA keys. Obviously a blockchain
would work better for this.
Other -- if the idea has already been brought up by other people, I
apologize.

[Tier Nolan via bitcoin-dev]

On Sat, Aug 6, 2016 at 11:39 AM, s7r via bitcoin-dev <

Post by s7r via bitcoin-dev
* reversal of transactions is impossible

I think it would be more accurate to say that the requirement is that
reversal doesn't happen unexpectedly.

If it is clear in the script that reversal is possible, then obviously the
recipient can take that into consideration.

Post by s7r via bitcoin-dev
* keep private keys private and safe. Lose them, it's like losing cash,
you can just forget about it.

Key management is a thing. Managing risk by keeping some keys offline is
an important part of that.

Post by s7r via bitcoin-dev
* while we try hard to make 0-conf as safe as possible (if there's no
RBF flag on the transaction), we make it almost impossible or very very
expensive to reverse a confirmed transaction.

BitGo has an "instant" system where they promise to only sign one
transaction for a given output. If you trust BitGo, then this is safe from
double spending, since a double spender can't sign two transactions.

If BitGo had actually implemented a daily withdrawal limit, then their
system ends up similar to cold storage. Only 10% of the funds at Bitfinex
could have been withdrawn before manual intervention was required (with
offline keys).

Who will accept

Post by s7r via bitcoin-dev
such an input and treat it as a payment if it can be reversed during the
settlement layer?

Obviously, if a payment is reversible, then you treat it as a reversible
payment. The protection here relates to moving coins from the equivalent
of cold storage to hot storage.

It is OK if it takes longer, since security is more important than
convenience for coins in cold storage.

Post by s7r via bitcoin-dev
The linked page describes that merchants will never accept payments from
'vaults', and it will take 24 hours for coins to be irreversible moved
outside the 'vault'.

This relates to the reserves held by the exchange. A portion of the funds
are in hot storage with live keys. These funds can be stolen by anyone who
gets access to the servers. The remaining funds are held in cold storage
and they cannot be accessed unless you have the offline keys. These funds
are supposed to be hard to reach and require manual intervention.

I think this is a wrong approach. hacks and big losses are sad, but all

Post by s7r via bitcoin-dev
the time users / exchanges are to blame for wrong implementations or
terrible security practices.

Setting up offline keys to act as firebreaks is part of good security
practices.

[Matthew Roberts via bitcoin-dev]

I'm wondering if we're fully on the same page here. What I was thinking was
that this protection mechanism would be applied to the coins in the hot
wallet (I wasn't talking about moving coins from the cold wallet to the hot
wallet -- though such a mechanism is also needed.)

With the hot wallet you would have an output script that only allowed coins
to be sent to a new transaction whose output script was then only
redeemable after N confirmations (the output is relative time-locked) but
which can also be recovered to a fixed fail-safe address before the
time-lock is reached (exactly like TierNolan already listed only the
time-locked destination shouldn't be completely fixed.) So the private key
for this hot wallet can still sign valid transactions to withdraw coins to
any known destination and these transactions still reach the blockchain.

The key difference from a regular transaction is that the destination only
has access to the coins -after- the relative time-lock is reached (N blocks
after first confirm) so everyone knows where withdrawals are suppose to be
going and how many coins are being withdrawn at any given time. Deposits to
the hot wallet would therefore need to be encumbered by the same protection
so that from then on this time-lock to redeem coins can be applied to every
new transaction trying to move coins (withdrawn by a user of the exchange
or sent to the cold wallet.)

Notice we don't care about the destination in the TX script for the hot
wallet because to process user's withdrawals we can't know ahead of time
where they need to be sent (so it isn't possible to use a fixed address
here – though you might want to remove the clearing phase and set a fixed
address for coins sent from the hot wallet to the cold wallet.) The benefit
here comes from being able to see what withdrawals are being cleared,
matching those up to our expectations, and being able to "cancel"
withdrawals if they look suspicious, and you get the benefits for transfers
made from the hot wallet to the cold wallet and visa-versa.


This approach is good for a number of crucial services:

1. Wallets could be built that grouped coins into different "accounts" with
different time-frames required for clearing / unlocking coins. Your savings
or investment account would say -- take up to a week to clear -- whereas
your everyday account used for smaller purchases (with less money) would
only take a few hours. This could all be linked up to services that
notified you of your money being moved + made any phone calls needed to
verify any larger transfers.

The service could also be entrusted with the “cancellation” key which can
only be used to move money to your offline fail-safe address. This would be
quite an interesting way to mitigate fraud without the user having to be
trusted to do anything (except I suppose – not storing their recovery keys
online 
 but this could be partially solved with BIP 32-style “master”
public keys + hardware wallets + multi-sig, N factor auth, etc ...)

2. Gambling websites that process a lot of Bitcoins also have a hot wallet
which could be better protected by this.

3. Various other e-commerce websites also accept Bitcoins directly. (Deep
web markets come to mind -- hey, people breaking the law need good security
too.)

4. Provable dead man's switches on the protocol level is another idea -- no
need to keep special time-locked transactions around and rely on them to be
broadcast = more reliable escrow services.

5. And obviously exchange hot (and cold) wallets - enemy number 1.

I hope that makes sense. I think I initially managed to confuse a lot of
people by talking about revoking transactions / “settlement layers”, etc.
But IMO: all of this needs to take place on the blockchain with a new set
of OP_CODES and other than the fixed address issue with OP_SPENDTO, I think
the general idea would still work.


tl; dr, A pseudo-reversal mechanism for transactions would mean that stolen
private keys were no longer such an issue. This is desperately needed for
exchanges, wallets, and other services that are forced to manage private
keys, and whose users (I argue) already expect for this to be possible (or
at least will when they're hacked.)




On Sat, Aug 6, 2016 at 9:13 PM, Tier Nolan via bitcoin-dev <

Post by Tier Nolan via bitcoin-dev

Post by s7r via bitcoin-dev
* reversal of transactions is impossible

I think it would be more accurate to say that the requirement is that
reversal doesn't happen unexpectedly.
If it is clear in the script that reversal is possible, then obviously the
recipient can take that into consideration.

Post by s7r via bitcoin-dev
* keep private keys private and safe. Lose them, it's like losing cash,
you can just forget about it.

Key management is a thing. Managing risk by keeping some keys offline is
an important part of that.

Post by s7r via bitcoin-dev
* while we try hard to make 0-conf as safe as possible (if there's no
RBF flag on the transaction), we make it almost impossible or very very
expensive to reverse a confirmed transaction.

BitGo has an "instant" system where they promise to only sign one
transaction for a given output. If you trust BitGo, then this is safe from
double spending, since a double spender can't sign two transactions.
If BitGo had actually implemented a daily withdrawal limit, then their
system ends up similar to cold storage. Only 10% of the funds at Bitfinex
could have been withdrawn before manual intervention was required (with
offline keys).
Who will accept

Post by s7r via bitcoin-dev
such an input and treat it as a payment if it can be reversed during the
settlement layer?

Obviously, if a payment is reversible, then you treat it as a reversible
payment. The protection here relates to moving coins from the equivalent
of cold storage to hot storage.
It is OK if it takes longer, since security is more important than
convenience for coins in cold storage.

Post by s7r via bitcoin-dev
The linked page describes that merchants will never accept payments from
'vaults', and it will take 24 hours for coins to be irreversible moved
outside the 'vault'.

This relates to the reserves held by the exchange. A portion of the funds
are in hot storage with live keys. These funds can be stolen by anyone who
gets access to the servers. The remaining funds are held in cold storage
and they cannot be accessed unless you have the offline keys. These funds
are supposed to be hard to reach and require manual intervention.
I think this is a wrong approach. hacks and big losses are sad, but all

Post by s7r via bitcoin-dev
the time users / exchanges are to blame for wrong implementations or
terrible security practices.

Setting up offline keys to act as firebreaks is part of good security
practices.
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[Erik Aronesty via bitcoin-dev]

I still feel like you're better off getting rid of "hot wallets" and use
lightning-esqe networks to route orders. I don't think either speed or
flexibility is an issue there.

IMO, the point of Bitcoin is to avoid the centralization that seems to be
happening on the network now. By making "hot wallets" more "secure", we
encourage things to keep heading downhill with massive centralized
crappy-security exchanges.

Because, ultimately, there's no security that will prevent an inside job.
And all of these thefts have, in my opinion, been at least partly inside
jobs.

And centralization is the actually demon that needs slaying here.

A client-side library with P2P order routing, tether.to + bitcoin .... and
you've got a decentralized exchange... with orders matched to users
directly, and channel-trades executed instantly. And "market makers"
running nodes to facilitate routing, etc.

No center... nothing to shut down or sue... and no one holds your funds.
That's a real Bitcoin exchange.



On Sun, Aug 7, 2016 at 1:35 AM, Matthew Roberts via bitcoin-dev <

Post by Matthew Roberts via bitcoin-dev
I'm wondering if we're fully on the same page here. What I was thinking
was that this protection mechanism would be applied to the coins in the hot
wallet (I wasn't talking about moving coins from the cold wallet to the hot
wallet -- though such a mechanism is also needed.)
With the hot wallet you would have an output script that only allowed
coins to be sent to a new transaction whose output script was then only
redeemable after N confirmations (the output is relative time-locked) but
which can also be recovered to a fixed fail-safe address before the
time-lock is reached (exactly like TierNolan already listed only the
time-locked destination shouldn't be completely fixed.) So the private key
for this hot wallet can still sign valid transactions to withdraw coins to
any known destination and these transactions still reach the blockchain.
The key difference from a regular transaction is that the destination only
has access to the coins -after- the relative time-lock is reached (N blocks
after first confirm) so everyone knows where withdrawals are suppose to be
going and how many coins are being withdrawn at any given time. Deposits to
the hot wallet would therefore need to be encumbered by the same protection
so that from then on this time-lock to redeem coins can be applied to every
new transaction trying to move coins (withdrawn by a user of the exchange
or sent to the cold wallet.)
Notice we don't care about the destination in the TX script for the hot
wallet because to process user's withdrawals we can't know ahead of time
where they need to be sent (so it isn't possible to use a fixed address
here – though you might want to remove the clearing phase and set a fixed
address for coins sent from the hot wallet to the cold wallet.) The benefit
here comes from being able to see what withdrawals are being cleared,
matching those up to our expectations, and being able to "cancel"
withdrawals if they look suspicious, and you get the benefits for transfers
made from the hot wallet to the cold wallet and visa-versa.
1. Wallets could be built that grouped coins into different "accounts"
with different time-frames required for clearing / unlocking coins. Your
savings or investment account would say -- take up to a week to clear --
whereas your everyday account used for smaller purchases (with less money)
would only take a few hours. This could all be linked up to services that
notified you of your money being moved + made any phone calls needed to
verify any larger transfers.
The service could also be entrusted with the “cancellation” key which can
only be used to move money to your offline fail-safe address. This would be
quite an interesting way to mitigate fraud without the user having to be
trusted to do anything (except I suppose – not storing their recovery keys
online 
 but this could be partially solved with BIP 32-style “master”
public keys + hardware wallets + multi-sig, N factor auth, etc ...)
2. Gambling websites that process a lot of Bitcoins also have a hot wallet
which could be better protected by this.
3. Various other e-commerce websites also accept Bitcoins directly. (Deep
web markets come to mind -- hey, people breaking the law need good security
too.)
4. Provable dead man's switches on the protocol level is another idea --
no need to keep special time-locked transactions around and rely on them to
be broadcast = more reliable escrow services.
5. And obviously exchange hot (and cold) wallets - enemy number 1.
I hope that makes sense. I think I initially managed to confuse a lot of
people by talking about revoking transactions / “settlement layers”, etc.
But IMO: all of this needs to take place on the blockchain with a new set
of OP_CODES and other than the fixed address issue with OP_SPENDTO, I think
the general idea would still work.
tl; dr, A pseudo-reversal mechanism for transactions would mean that
stolen private keys were no longer such an issue. This is desperately
needed for exchanges, wallets, and other services that are forced to manage
private keys, and whose users (I argue) already expect for this to be
possible (or at least will when they're hacked.)
On Sat, Aug 6, 2016 at 9:13 PM, Tier Nolan via bitcoin-dev <

Post by Tier Nolan via bitcoin-dev
On Sat, Aug 6, 2016 at 11:39 AM, s7r via bitcoin-dev <

Post by s7r via bitcoin-dev
* reversal of transactions is impossible

I think it would be more accurate to say that the requirement is that
reversal doesn't happen unexpectedly.
If it is clear in the script that reversal is possible, then obviously
the recipient can take that into consideration.

Post by s7r via bitcoin-dev
* keep private keys private and safe. Lose them, it's like losing cash,
you can just forget about it.

Key management is a thing. Managing risk by keeping some keys offline is
an important part of that.

Post by s7r via bitcoin-dev
* while we try hard to make 0-conf as safe as possible (if there's no
RBF flag on the transaction), we make it almost impossible or very very
expensive to reverse a confirmed transaction.

BitGo has an "instant" system where they promise to only sign one
transaction for a given output. If you trust BitGo, then this is safe from
double spending, since a double spender can't sign two transactions.
If BitGo had actually implemented a daily withdrawal limit, then their
system ends up similar to cold storage. Only 10% of the funds at Bitfinex
could have been withdrawn before manual intervention was required (with
offline keys).
Who will accept

Post by s7r via bitcoin-dev
such an input and treat it as a payment if it can be reversed during the
settlement layer?

Obviously, if a payment is reversible, then you treat it as a reversible
payment. The protection here relates to moving coins from the equivalent
of cold storage to hot storage.
It is OK if it takes longer, since security is more important than
convenience for coins in cold storage.

Post by s7r via bitcoin-dev
The linked page describes that merchants will never accept payments from
'vaults', and it will take 24 hours for coins to be irreversible moved
outside the 'vault'.

This relates to the reserves held by the exchange. A portion of the
funds are in hot storage with live keys. These funds can be stolen by
anyone who gets access to the servers. The remaining funds are held in
cold storage and they cannot be accessed unless you have the offline keys.
These funds are supposed to be hard to reach and require manual
intervention.
I think this is a wrong approach. hacks and big losses are sad, but all

Post by s7r via bitcoin-dev
the time users / exchanges are to blame for wrong implementations or
terrible security practices.

Setting up offline keys to act as firebreaks is part of good security
practices.
_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

_______________________________________________
bitcoin-dev mailing list
https://lists.linuxfoundation.org/mailman/listinfo/bitcoin-dev

[Tier Nolan via bitcoin-dev]

Not everyone who uses centralized exchanges are there to obtain the
currency though. A large portion are speculators who need to be able to
enter and exit complex positions in milliseconds and don't care about
decentralization, security, and often even the asset that they're buying.

Centralized exchanges also allow for things like limit orders. You don't
even have to be logged in and they can execute trades. This couldn't be
done with channels.

Try telling everyone who currently uses Btc-e to go do their margin
trading over lightning channels, for example.

Using channels and a centralized exchange gets many of the benefits of a
distributed exchange.

The channel allows instant funding while allowing the customer to have full
control over the funds. The customer could fund the channel and then move
money to the exchange when needed.

Even margin account holders might like the fact that it is clear which
funds are under their direct control and which funds are held by the
exchange.

If they are using bitcoin funds as collateral for a margin trade, then
inherently the exchange has to have control over those funds. A 2 of 3
system where the customer, exchange and a 3rd party arbitration agency
holds keys might be acceptable to the exchange.

[Erik Aronesty via bitcoin-dev]

I'm not convinced you need to hold people's funds to provide those
features. Maybe the millisecond thing. But 99 out of 100 traders would
accept a 100 millisecond latency in exchange for 0 counterparty risk.

[Tier Nolan via bitcoin-dev]

With channels and the exchange acting as hub, you can do instant trades
between altcoins.

This doesn't work with fiat accounts. A "100% reserve" company could issue
fiat tokens. The exchange could then trade those tokens.

This eliminates the counter-party risk for the exchange. If the exchange
dies, you still have your (alt)coins and also fiat tokens.

There is still risk that the token company could go bankrupt though. This
could be mitigated by that company requiring only "cashing out" tokens to
accounts which have been verified.

The company could set up a blockchain where it signed the blocks rather
than mining and could get money from transaction fees and also minting fees
(say it charges 1% for minting new tokens)

I wonder what how the law would work for that. It isn't actually doing
trading, it is just issuing tokens and redeeming them.