Axelar 백서

Axelar

원문

bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 3쪽 중 2쪽

y are necessary. Pre-signatures, non-interactive online signing. In an effort to reduce the burden of communication upon the parties to sign a message, several recent protocols have identified a significant portion of the work for a signature that can be done “offline”, before the message to sign is known [18, 13]. The output of this offline phase is called a pre-signature. The production of pre-signatures is viewed as a separate protocol T .Presign distinct from T .Keygen and T .Sign. The outputs of the pre-signature protocol must be kept private by the parties until they use them at the signing phase. Later, when the message to sign becomes known, only a small amount of additional “online” work remains to be done in T .Sign in order to complete the signature. The online T .Sign phase does not require any communication among the parties. Each party simply does a local computation on the message and pre-signature and then announces her share si of the signature. (Once public, these signature shares s1, . . . , st+1 are easily combined by anyone to reveal the actual signature s.) This property is called non-interactive online signing . Robustness. Threshold schemes guarantee only that a subset of malicious parties cannot sign messages or learn the secret key. This guarantee does not, however, preclude the possibility that bad actors can block everyone else from producing keys or signatures. In some schemes, malicious behaviour by even a single party can cause T .Keygen or T .Sign to abort with no useful output. The only recourse is to restart the protocol, possibly with different parties. Instead, for decentralized networks, we want T .Keygen and T .Sign to succeed if at least t + 1 of the parties are honest, even if some malicious parties send malformed messages or drop messages in the protocols. This property is called robustness. F ault attribution. The ability to identify bad actors in T .Keygen or T .Sign is called fault attribution . Without fault attribution it is difficult to reliably exclude or punish bad actors, in which case the costs imposed by bad actors must be borne by everyone. This property is also important for decentralized networks where malicious behavior should be identifiable and economically disincentivized via slashing rules. Security in concurrent settings. The signature scheme needs to be secure in a concurrent setting, where multiple instances of the keygen and signing algorithms can be involved in parallel. (Drijvers et al. [16] for instance, showed an attack against Schnorr multisignature schemes in these settings). There are versions of both ECDSA and Schnorr schemes that satisfy these properties [13, 22]. ECDSA and EdDSA are by far the most widely deployed signature schemes in the blockchain space. As such, threshold versions of both schemes have been the focus of a recent resurgence in research and development. Readers interested in state-of-the-art can refer to [22, 13, 18] and a recent survey paper [12]. 6 5 Axelar Network 5.1 Designing an Open Cross-Chain Network The bridges that Axelar network maintains are backed up by threshold accounts such that (almost) all valida- tors must collectively authorize any cross-chain request. Designing a network where anyone can participate to secure these bridges requires meeting the following technical requirements: • Open membership. Any user should be able to become a validator (following the rules of the network). • Updates to membership. When a validator leaves the system honestly, their key needs to be revoked appropriately. • Incentives and slashing. Malicious validators should be identifiable and their actions must be identified and addressed by the protocol. • Consensus. Threshold schemes on their own are defined as stand-alone protocols. To propagate mes- sages between nodes we need both broadcast and point-to-point private channels. Moreover, validators need to agree on the latest state of each invocation of threshold schemes since they often have multiple round of interactions. • Key-management. Just as ordinary validators in any PoS system must carefully guard their keys, so too must Axelar validators guard their threshold shares. Keys need to be rotated, split between online and offline parts, etc. Axelar starts with Delegated Proof-of-Stake model, where the community elects a set of validators to run the consensus. Note that standard threshold schemes treat every player identically and have no notion of “weight” in the consensus. Hence, the network must adapt them to take validators’ weight into account. A simple approach is to assign multiple threshold shares to larger validators. Outlined below are three basic functions that validators collectively perform. • Threshold Key Generation. Existing threshold key generation algorithms for standard blockchain signature schemes (ECDSA, Ed25519) are interactive protocols between multiple participants (see Section 4). A special transaction on the Axelar network instructs the validators to commence execution of this stateful protocol. Each validator runs a threshold daemon process that is responsible for the secure keeping of the secret state. For each phase of the protocol: 1. A validator keeps the state of the protocol in its local memory. 2. It calls the secret daemon to generate the messages as per the protocol description for other validators. 3. It propagates the messages either via the broadcast or via the private channels to other validators. 4. Each validator executes state transition functions to update its state, proceed to the next phase of the protocol, and repeat the above steps. At the end of the protocol, a threshold public key is generated on the Axelar chain, and it can be displayed back to the user (e.g., for deposits) or to the application that generated the initial request. • Threshold Signing. Signing requests on the Axelar network are processed similarly to the key-generation requests. These are invoked, for instance, when a user wants to withdraw an asset from one of the chains. These are interactive protocols, and state transition between the rounds is triggered as a function of the messages propagated via the Axelar blockchain view and every validator’s local memory. • Handling Validator Membership Changes. The validator set needs to be rotated periodically to allow for new stakeholders to join the set. Upon a validator set update, we need to update the threshold key to be shared across the new set. Thus if we allowed anyone to join at any time, we would have to update the threshold key very frequently. To prevent this, we rotate validators every T blocks. Within intervals of T rounds, the set V R and the threshold key are fixed. At every round that is an integral multiple of the parameter T , we update the validator set as follows: 7 1. At any round R, the Axelar state keeps track of the current validator set V R. V R+1 = V R unless R + 1 is a multiple of T . 2. During rounds (( i− 1)T, iT ], users post bonding/unbonding messages. 3. At the end of round iT , these messages are applied to V iT−1 to get V iT . • Threshold Key Generation and Signing in the Presence of Rotating Validators. Axelar blockchain may issue a request for a new key or a threshold signature at roundR. The signing process takes longer than one round, and we don’t want to slow down consensus, so we request that the signature is produced before round R + 10 starts. In particular, validators start round R + 10 only after seeing a certificate for round R + 9 and a signature for each keygen/signature request issued at round R. The outcome of all round R requests must be included in block R + 11. In other words, a round R block proposal that does not contain the outcomes from a round R− 11 is considered invalid, and validators don’t vote on it. To ensure that all threshold messages are signed before a validator set update, Axelar does not issue any threshold requests during a round equal to −1,−2, . . . ,−9 modulo T . 5.2 Network Security The security of blockchain systems relies on various cryptographic and game theoretic protocols, as well as the decentralization of the network. For instance, in proof-of-stake blockchains, without the proper incentives validators may collude and rewrite the history, stealing other users’ funds in the process. In proof-of-work networks, without sufficient decentralization, it is quite easy to create long forks and double spend, as the multiple attacks on Bitcoin Gold and Ethereum Classic have proven. Most of the research on blockchain security has focused on sovereign chains. But once chains interoperate, new attack vectors have to be considered. For instance, assume that Ethereum talks to a small blockchain X through a direct bridge controlled by two smart contracts, one on Ethereum and one on X. Besides the engineering challenges we summarized in Section 1.1, one must decide what happens when the trust assump- tions of X are violated. In this case, if ETH has moved to X, the validators of X may collude to forge a history of X where they hold all the ETH, post the forged consensus proofs on Ethereum and steal the ETH. The situation is even worse when X is connected with multiple other chains through direct bridges, where if X forks the effects propagate through every bridge. Setting up recovery governance guidelines for each pairwise bridge is an overwhelming task for any individual project. Axelar network addresses the security concerns using the following mechanisms: • Maximum Safety. Axelar sets the safety threshold to 90%, meaning that almost all validators will need to collude to withdraw any funds that are locked by its network or forge state proofs 1. In practice, it has been observed that PoS validators have very high up-time (close to 100%), assuming they are properly incentivized. Hence, Axelar network will produce blocks even despite this high threshold. However, in the rare case that something goes wrong and the network stalls, the network needs robust fall-back mechanisms to reboot the system described next. • Maximum Decentralization. Since the network uses threshold signature schemes, the number of val- idators can be as large as possible. The network is not bounded by the number of validators we can support, transaction limits or fees that would arise from using, for instance, multi-signatures on different chains where the complexity (and fees) increase linearly with the number of validators. 2 • Robust Fall-back Mechanisms. The first question that must be addressed in a network with high safety thresholds as above is what happens when the network itself stalls. Suppose Axelar network itself stalls. Can we have a fall-back mechanism that would allow users to recover their funds? To address any potential stall of the Axelar network itself, each threshold bridge account on a blockchain X that the Axelar validators collectively control has an “emergency unlock key”. This key can be shared 1The final parameter that will be chosen for the network deployment may be adjusted. 2For some blockchains, multi-signatures offer a reasonable alternative where gas fess are small and supported message formats are appropriate. But they do not scale for two of the most largest platforms like Bitcoin and Ethereum. 8 across thousands of parties and may even be a custom key for blockchain X that is shared across the community of that chain. Hence, if Axelar network stalls, this key will act as a fall-back and enable recovery of the assets (see below for more details). • Maximum Decentralization of Fall-Back Mechanisms. This fall-back mechanism includes a secondary recovery set of users, in which just anyone can participate without any cost. These users do not need to be online, run nodes, or coordinate with each other. They are only “called on duty” if Axelar network stalls and cannot recover. The network’s security is enhanced by a very high threshold on the primary validator set and a maximally decentralized secondary recovery set. • Shared Governance. A common protocol governs the Axelar network. Collectively, the users can vote on which chain should be supported through its network. The network will also allocate a pool of funds that can be used to reimburse users in case of unexpected emergencies, controlled via the governance protocols as well. Various security mechanisms are discussed below. F all-Back Mechanisms. When Axelar stalls due to the high threshold, an “emergency unlock key” takes control of the network. There are multiple ways to instantiate this unlock key, and certain chains/applications may opt to utilize a different variation for the “recovery set” or opt-out completely: 3 • Option a. Share the key across foundations of blockchain projects and reputable people in the com- munity. • Option b. Share the across parties elected through the delegated PoS mechanism. • Option c. For accounts managing assets and information for chain/application X, share a custom key across the stakeholders/validators of X. Assuming X has governance mechanisms in place, the same governance mechanisms can be applied to determine a course of action if Axelar stalls. Now, given the recovery users’ identities and their public keys, a simple protocol generates shares of the recovery key that no-one knows. Moreover, the users of recovery set do not need to be online until called to recover via the governance mechanisms. Following the standard distributed key-generation protocols, each Axelar validator shares a random value. The recovery secret key is generated by summing up these values. Instead of doing the summations in the clear, all shares are encrypted under the public keys of the recovery users and then added up homomorphically (this assumes additively homomorphic encryption and an additional layer of zero-knowledge, both of which are easily obtainable). The result of this protocol is a recovery public key RP K and potentially thousands of encryptions (under the public keys of the recovery users) of the shares of the corresponding secret key Enci(si) that are distributed to their owners (e.g., posted on chain). Axelar bridge contracts include an option to recover funds using RP K under certain conditions. Finally, it is also possible to update this recovery key and even change the set of users holding its shares without requiring any work from the participating shareholders. If chain X that is connected to Axelar breaks, there are a couple options: • Impose limits on the USD value of assets that can be moved in/out of X on any single day. Thus a malicious chain X can only steal a small fraction of all assets that are bridged to it before Axelar validators detect this, and the governance mechanisms from the following bullets kick in. • The Axelar governance module can be used to vote on what happens in those situations. For instance, if there is a benign bug and the community restarts X, Axelar governance can determine to restart the connection from where it left off. • If ETH had moved to X, a custom Ethereum recovery key can determine what happens to the ETH assets. 3The final deployment on the Axelar network will be finalized closer to the network launch. 9 6 Cross-Chain Gateway Protocol (CGP) In this section, we explain the cross-chain gateway protocol and routing mechanisms on two core examples common between many applications’ needs: State synchronization (Section 6.2). Post information about the state of a source blockchain S into the state of a destination blockchain D. (For example, post a Bitcoin block header to the Ethereum blockchain.) Asset transfer (Section 6.3). Transfer a digital asset from S to D and back again. (For example, transfer bitcoins from the Bitcoin blockchain to the Ethereum blockchain, and then back to the Bitcoin blockchain.) For simplicity we assume that chain D has at least minimal support for smart contracts but S can be any blockchain whatsoever. 6.1 Accounts on other chains To bridge different chains, threshold accounts are created on each chain that control the flow of value and information across them. For chain Chain, denote the account by Chain Axelar. Bitcoin account. For Bitcoin and other non-smart contract chains Axelar validators create a threshold ECDSA key as per section 5.1. This key controls the ECDSA account on Bitcoin, and is the destination address where users send deposits. Personalized threshold keys may be created per user request. The key may be updated periodically, and the latest key and personalized keys can be found by querying an Axelar node. Threshold bridge account on chains with smart contracts. Denote the chain by SC. the validators create a threshold ECDSA or ED25519 key as per section 5.1, depending on which key type the chain supports. We denote this key by P KAxelar, when there is no ambiguity as to which chain we are referring to. This key controls a smart contract account on SC, denoted by SCAxelar, and any application on SC can query SCAxelar to learn the PK address of that key. This way, any SC application can recognize messages signed by SKAxelar. The protocol also needs to account for rotating values of P KAxelar. This happens as follows: 1. Initialize SCAxelar on SC. It stores P KAxelar as part of its state, which is initialized as its genesis value on Axelar. SCAxelar also includes rules for updating the PK. 2. To update P KAxelar, a transaction of the format (update, P Knew) must be submitted with a signature from the current SKAxelar. Then the contract sets P KAxelar = P Knew. 3. Every time the validators update the threshold key for SC from P Ki to P Ki+1, Axelar requests that validators use SK i to sign ( update, P Ki+1). Subsequently this signature is posted to SCAxelar which updates P KAxelar. 6.2 State synchronization Let qS denote an arbitrary question about the state of chain S. Examples of such questions include: • “At what block round, if any, did a transaction tx appear?” • “What is the value of a certain data field?” • “What is the Merkle root hash of the entire state of S at block round 314159?” Let aS denote the correct answer to qS and suppose an end-user or application demands that aS be posted to chain D. Axelar network meets this demand as follows: 10 1. The user posts a request qS on one of the bridge accounts (which are subsequently picked up by the the validators) or directly to the Axelar blockchain. 2. As part of Axelar consensus, each validator must run node software for chains S, D. Axelar validators query the API of their chain S node software for the answer aS and report the answer to the Axelar chain. 3. Once > F weighted validators report the same answer at round R, Axelar asks validators to sign aS. 4. Using threshold cryptography the validators sign aS. The signature is included in block R + 11. 5. Anyone can take the signed value aS from block R + 11 and post it to D. 6. The request has been serviced. Any application on D may now take the signed value aS, query DAxelar for the latest P KAxelar, and verify that the signature of aS corresponds to P KAxelar. The validators also post aS to the bridge account on chain D, which applications can retrieve. 6.3 Cross-Chain Asset Transfer The network enables cross-chain transfers of digital assets by extending the state synchronization workflow of Section 6.2. A sufficient supply of pegged-S tokens is printed and controlled byDAxelar upon its initialization. Suppose a user demands to exchange x amount of tokens on source chain S for x amount of pegged- S tokens on destination chain D, to be deposited at a D-address wD of the user’s choice. We present the fully general workflow, which supports arbitrary source chains S—even chains such as Bitcoin that do not support smart contracts: 1. The user (or an application acting on the user’s behalf) posts a transfer request (x, wD) to the threshold bridge account which is subsequently routed to the Axelar network. 2. Axelar validators use threshold cryptography to collectively create a fresh deposit address dS for S. T
프로젝트로 돌아가기