[bitcoin-dev] New serialization/encoding format for key material

[Jonas Schnelli <dev@jonasschnelli.ch>]

[-- Attachment #1.1: Type: text/plain, Size: 6958 bytes --]

Hi

Extended public and private keys are defined in BIP32 [1].

Encoded extended private keys should not be confused with a wallet „seed“
(proposals like BIP39) while they can also partially serve the purpose to
„seed“ a wallet (there may be an overlap in the use-case).

Recovering a wallet by its extended private master key (xpriv; may or may not
be at depth 0) is a complex task with risks of failing to recover all available
funds.

It may be reasonable to consider that recovering a wallet purely based on the
existence of an extended private master key is a forensic funds recovery
process and should probably be the last resort in case of a backup-recovery
situation. A simple example here is, that it was/is possible to have used an
xpriv (referring to extended private master key) in production that is/was used
to derive BIP45 based P2SH multisig addresses (1of1, used by Bitpays BWS for
while), later used for bare BIP45ish multisig 1of1 as well as for P2PKH after
BIP44 & vanilla BIP32 P2WPKH (m/0’/k’).
I’m not aware of any wallet that would recover 100% of those funds, leading to
the risk that forwarding the unspents and destroying the extended master key
may result in coins forever lost.

The case above may be an edge case, but I’m generally under the assumption that
recovering funds based on the sole existence of an xpriv (or seed) without further
metadata is a fragile concept.

Second, the missing birthday-metadata tend to lead to non-optimal blockchain
scans (eventually increased p2p traffic). Recovering funds can take hours.

Additionally, the BIP44 gap limit seems to be a weak construct. The current gap
limit in BIP44 is set to 20 [2] which basically means, handing out more then 20
incoming payment requests (addresses) results in taking the risks that funds
may be destroyed (or at least not detected) during a recovery.
The Gap limit value may also depend on the use case, but the current proposals
do not allow to set an arbitrary value. High load merchants very likely need a
different gap limit value then individuals create a transaction once a year.

During creation time of an xpriv/xpub, it is impossible to know if the created
xpriv will be used for an unforeseen derivation scheme. Future proposals may
want to limit an extended key to a single derivation scheme.


This is an early draft in order to allow discussion that may lead to a possible
proposal.
This proposals could also make BIP 178 obsolete since it can be replace the
WIF[3] standard.


Thanks for feedback
/jonas


------------------------------------


Titel
######
Bech32 encoded key material including metadata

Abstract
########
An error tolerant encoding format for key material up to 520bits with a minimal
amount of metadata.

Motivation
##########
(See above; intro text)


Specification
#############

## Serialization format

1 bit version bit
15 bits (bit 1 to 16) key-birthday (0-32767)
(12 bit gap limit)
3 or 5 bits script type
256 or 512 or 520 bits key material
= Total 275, 545, 553 bits

The initial version bit allows extending the serialization-format in future.
The encoding format must hint the total length and thus allow to calculate the
length of the key material.

The total length for 256 or 512 bit key material is optimised for Bech32 (power
of 5).

### Key material
If the key material length is 520 bits, it must contain an extended public key
If the key material length is 512 bits, it must contain an extended private key
Key material length other then 256, 512, 520 bits and invalid.

If 520 bits are present, first 256 bits are the BIP32 chain code, to second 264
bits (33 bytes) define the public key (according to BIP32)

If 512 bits are present, first 256 bits are the BIP32 chain code, to second 256
bits define the private key

If 256 bits are present, those bits represent a pure private key (or seed)

### Key birthday
A 15 bit timestamp expressed in days since genesis (valid up to ~2098). The
birthday must be set to the first possible derivation of the according extended
key, if unknown, the used seed birthday must be used. If both unknown, 0
(16x0bit) must be used.

### Gap limit delta
12 bits, results in a possible range from 0 to 4095.

If the total decoded serialization length is 275 bits (decode) or if the key
material is 256 bits (encode), the gap limit must not be present.

The base gap limit value is 20 (to disallow insane gap limits). The final gap
limit is the base value + the gap limit delta stored in those 12 bits.
Key derivation gap limit must not be exceeded when deriving child keys and must
be respected during transaction rescans.
Child key derivation must not be possible if gap limit is hit.

### Script type restriction
3 or 5 bits (range 0-7 / 0-31)
0 no restriction
1 P2PKH compressed
2 P2PKH | P2SH
3 P2WPKH P2WSH nested in P2SH
4 P2WPKH | P2WSH

If the total decoded serialization length is 275 bits (decode) or if the key
material is 256 bits (encode), 3 bits are used for the script type. 5 bits are
used for key material with the size of 512, 520 bits.

If the script type restriction is set, the according extended key must only be
used to derive addresses with the selected script type.
This does not stands in contradiction to derivation path proposals ([4]). It
does allow to derive and encode an extended key at a keypath where users assume
restricted script types in derivation due to other supported proposals.


Encoding
########

Bech32 must be used as encoding format (see the Bech32 rational [5]). Encoding
545 or 553 bits (results in 109 resp. 111 x 5 bits) will exceed the Bech32 property of a
guaranteed detection of 4 errors (only 3 are).
It is possible that there are more efficient BCH codes, especially for encoding
extended private keys. Since a Bech32 implementation needs to be present in
modern Bitcoin software, re-using Bech32 will allow to migrate to this proposal
with a minimal implementation effort.
Forensic, cpu-intense key-recovery (including brute-force techniques) may allow
to recover keys beyond the guaranteed error detection limits.

Bech32 HRPs
Mainnet Private Extended: xp
Mainnet Public Extended: xpu
Testnet Private Extended: tp
Testnet Public Extended: tpu
Mainnet Key: pk-
Testnet Key: tk-

Compatibility
###########
Only new software will be able to use these serialization and encoding format.

References
##########


[1] https://github.com/bitcoin/bips/blob/master/bip-0032.mediawiki <https://github.com/bitcoin/bips/blob/master/bip-0032.mediawiki>
[2] https://github.com/bitcoin/bips/blob/master/bip-0044.mediawiki
[3] https://github.com/bitcoin/bips/blob/master/bip-0178.mediawiki
[4] https://github.com/bitcoin/bips/blob/master/bip-0049.mediawiki
[5] https://github.com/bitcoin/bips/blob/master/bip-0173.mediawiki#rationale


[-- Attachment #1.2: Type: text/html, Size: 11484 bytes --]

[-- Attachment #2: Message signed with OpenPGP --]
[-- Type: application/pgp-signature, Size: 833 bytes --]