Permalink
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
138 lines (86 sloc) 11.4 KB

Ticket Matcher Service Design

Glossary

  • Participant: A wallet contributing some amount of DCR to the ticket
  • Session: A group of participants in the process of creating a shared ticket
  • Matcher: The backend service that collects outputs and generates the ticket

General Flow

Session Execution

  • Participate() informs the matcher service that a wallet can contribute at most X DCR
    • The matcher waits for as many participants as needed, until the ticket price is met
    • Replies with the ticket amount + fees that the participant will enter session
  • AddOutputs() sends the sstxsubmission (amount from previous step), sstxchange (usually 0), the split transaction outputs and input outpoints for the participant
    • The matcher waits for all participants to send their outputs and checks the validity of them
    • After all outputs are received, the matcher generates the ticket and split transaction templates plus a signed revocation for the ticket
    • Replies with the transactions for the participants
  • FundTicket() signs the ticket inputs and sends the corresponding scriptSigs to the matcher
    • The matcher waits for all participants to send their corresponding scriptSigs
    • Replies with the fully funded ticket
  • FundSplit() signs the split transaction inputs and sends the corresponding scriptSigs to the matcher
    • The matcher waits for all participants to send the corresponding scriptSigs for all inputs used in the split tx
    • Replies with the fully funded split tx

Notes

  • A single split transaction is created to reduce (though not eliminate) the chances of unilateral fee drain
  • Funding the split transaction should only happen after the participant has confirmed the ticket, revocation and split txs are all valid and conform to the expected for the ticket purchase
  • Stakepool fee is specified as one of the outputs of the single split tx (funded by the participants)

Voting Rights

The matcher decides the voting address for the ticket and more generally (if it is also acting as a stake pool) decides the actual vote choices. In the case of a split ticket, many strategies for deciding the participants influence over this process are possible (and may be matcher-dependent):

  • Majority rule (choose whatever > 50% of what the participants want; abstain otherwise - vulnerable to influence amplification)
  • Supermajority rule (choose whatever > 75% of what the participants want; abstain otherwise - vulnerable to influence amplification)
  • Participant with highest contribution chooses (vulnerable to influence amplification)
  • Randomly choose a participant (vulnerable to influence amplification)
  • Randomly choose a participant, weighted by the contribution %
  • Pool always chooses the option (delegate authority)

The actual voting right (i.e. the private key used for the ticket submission output that would allow someone to create the vote transaction) can either be maintained by the pool, with the pool deciding the vote choices based on its internal state and/or the private key or a multisig script can be used to allow one of the participants to vote.

Care must be taken when using the second option though, because if the matcher discloses the voting participant before the split and ticket transactions are fully funded and transmitted it becomes vulnerable to attack by participants closing the session early if they do not receive the voting rights.

Influence Amplification

We call influence amplification a possible attack where a big decred holder can gain more influence on agenda decisions (more voting power) by exploiting the matcher vote decision strategy.

A malicious matcher can trivially gain more influence over the network the same way the existing stake pools can: by overriding voters' preferences and always voting a certain way. Due to the centralized nature of stakepools, we assume users trust (but periodically verify) that the stakepools are honoring their vote choices and the same consideration applies to this service: we assume the users trust the matcher they connect to (which will probably be a stakepool by itself).

The attack we describe here, however, is slightly more subtle: it tries to craft a participation strategy that exploits the voting rights decision of honest matchers/stakepools uses that to try and gain more influence over the network.

Some of alternatives for deciding the vote choice outlined on the previous section are vulnerable to this attack. For a concrete example, assume the following:

  • Attacker has 1e6 DCR available for purchasing influence in any way and wants to have biggest impact
  • Ticket price remains stable at 100 DCR (attack is carried over a long period of time)
  • Ticket, transaction and stakepool fees are negligible and ignored (best case for an attacker)
  • Attacker can craft any participation level (measured as % of a split ticket) desired

Given the previous assumptions, the attacker could buy 1e4 tickets outright and have that amount of influence on the agenda. OR it could try to exploit vulnerable pools.

Pools using the strategy of majority rule (participant with >51% of participation level choose the the vote) are vulnerable to this attack by having the attacker buy exactly 51% of each split ticket and deciding the vote of the full ticket. In the example, the attacker can buy voting rights for up to (1e6 / (51% * 1e2)) 19607 tickets or almost double its influence compared to buying full tickets.

In general, any pool using the rule of granting voting rights to the participant with x% of a ticket will allow the attacker to increase its voting influence to 1/x%, so this strategy is not suitable to prevent influence amplification.

Pools that use the strategy of selecting a random participant to vote are also vulnerable to this, because an attacker can simply simulate a large number of participants of a split (increasing its chances of being selected with minimal increase in usage of funds).

Pools that use the strategy of selecting a random participant to vote, weighted by their participation % are not vulnerable to this attack. Working out the previous example, assuming an attacker buying 51% of tickets, they can purchase up to 19607 tickets. However, in each individual ticket the attacker only has 51% chance of being selected to decide the vote choice, therefore we expect it will actually decide the vote in only (19607*51%) ~10000 tickets, bringing it back in line with the voting influence it could have if it used its original funds to purchase individual tickets.

Therefore, pools should preferably use this method of vote choice.

Revocation Rights

Revocation of missed and expired tickets is one of the most important concerns of the split ticket purchase service. Users must be sure that they will be able to redeem their funds if the pool is disabled (possibly forever).

The strategy for allowing revocation by any participant is the following:

  • The matcher generates and signs a revocation transaction and includes it in the reply to the addOutputs() call to all participants
  • Participants should ensure this revocation tx validates and then store it until the corresponding ticket is either voted, misses or expires
  • Either the matcher or one or multiple participants may transmit the tx as appropriate

Revocation Tx Storage

The requirement of having to store the revocation transaction until the ticket is voted, missed or expired is less than ideal as it introduces another piece of information an user needs to store along with their seed to have access to all of their funds.

There are two ways to mitigate this requirement:

  • Have an independent (third-party to the participants and the matcher/pool - possibly as a community service) store the revocation transaction and monitor the blockchain to issue the revocation as needed
  • Use a smart contract to allow anyone to revoke the transaction (indefinite revocation rights - see below)

Indefinite Revocation Rights

The above strategy assumes at least one participant will not lose the revocation transaction and will be interested in actually publishing it. It breaks, preventing return of the funds if all participants plus the matcher lose the transaction (or the matcher loses the keys to the voting wallet or any similar situation).

To allow for ticket revocation even in the face of failures from all participants and matcher, we can design a script for the ticket submission output (output 0 of the SSTx) redeemable either by the matcher (signing the corresponding input of an SSGen or SSRTx) or by anyone. That is achieved by using a P2SH address in the ticket and a specially crafted and known in advance redeem script:

OP_IF
  OP_DUP OP_HASH160 [pool-phk-hash] OP_EQUALVERIFY OP_CHECKSIG
OP_ELSE
  [revocationRelLockTime] OP_CHECKSEQUENCEVERIFY
OP_ENDIF

The semantics for this smart contract is that a revocation transaction can be crafted either by signing the input with the appropriate private key OR by setting an appropriate value for the txin sequence field.

Ordinarily, an UTXO using this script would be spendable by anyone and allow the coins to be sent to any other address, however the consensus rules for stake transactions ensure that the funds of an SSTx can be spent only on a very specific way, to wit:

  • Funds from the outputs of an SSTx can only be spent by an SSGen or SSRTx
  • Funds from an SSTx can only be sent to the addresses of the stake commitment outputs (odd-numbered outputs)
  • Rewards of an SSGen must always be distributed proportionally to the commitment amounts
  • Fees payable on an SSRtx are limited to a maximum amount (fee allowance) encoded in the commitment output

Stake transactions not following these rules cannot be mined on a block and are considered invalid. Therefore, the above script can be used as a redeem script of tickets, such that the ticket can be revoked if missed/expired by anyone (as long as the particular script used by a pool is known).

The advantage of using the above script over simply storing the revocation is that the script can be known ahead of time (or is otherwise deducible from previous submitted transactions), it is smaller to store, may be reused among multiple tickets and removes the need for the pool to ever disclose any of its private keys.

The only drawback of having to execute the ELSE branch of the above script is the possibility of paying a slightly increased amount of transaction fees (up to the individual limits of the ticket commitments) which in a default ticket commitment as generated by the current version of dcrwallet is ~0.16 DCR (0x18 limit byte).

The revocationRelLockTime value must be chosen such that there is opportunity to send the previously signed revocation transaction (with the reduced tx fee) before anyone (possibly a miner interested in the higher fees) sends the unlocked revocation tx. Also, it must ensure that a vote cannot be cast before end of the ticket expiration window. Therefore, a value of TicketExpiry + 7 days-worth-of-blocks (or some other similar time) could be selected.

Example of revocation using the above script:

https://testnet.dcrdata.org/api/tx/decoded/52cc60a2008b64d9e558be34a25820995a38b0c05fd735c578f238d7517752f0?indent=true

(notice on the above the sequence value is not the one that should be used in production - it was chosen only to test the script in a more timely fashion).

Outstanding Issues

  • How to prevent spam?
    • Require signing the outpoints to be used during addOutputs()? Or require the outpoints on Participate()?