Account keys are keys that are meant to control funds. They can be either:
ed25519 implementation using Schnorr signatures.
sr25519 variant using Schnorr signatures.
ECDSA signatures on secp256k1
There are no differences in security between
sr25519 for simple signatures.
ed25519 to be much better supported by commercial HSMs for the foreseeable future.
At the same time,
sr25519 makes implementing more complex protocols safer. In particular,
sr25519 comes with safer version of many protocols like HDKD common in the Bitcoin and Ethereum ecosystem.
When we talk about "controller" and "stash" keys, we usually talk about them in the context of running a validator or nominating DNAR, but they are useful concepts for all users to know. Both keys are types of account keys. They are distinguished by their intended use, not by an underlying cryptographic difference. All the info mentioned in the parent section applies to these keys. When creating new controller or stash keys, all cryptography supported by account keys are an available option.
The controller key is a semi-online key that will be in the direct control of a user, and used to submit manual extrinsics. For validators or nominators, this means that the controller key will be used to start or stop validating or nominating. Controller keys should hold some DNAR to pay for fees, but they should not be used to hold huge amounts or life savings. Since they will be exposed to the internet with relative frequency, they should be treated carefully and occasionally replaced with new ones.
The stash key is a key that will, in most cases, be a cold wallet, existing on a piece of paper in a safe or protected by layers of hardware security. It should rarely, if ever, be exposed to the internet or used to submit extrinsics. The stash key is intended to hold a large amount of funds. It should be thought of as a saving's account at a bank, which ideally is only ever touched in urgent conditions. Or, perhaps a more apt metaphor is to think of it as buried treasure, hidden on some random island and only known by the pirate who originally hid it.
Since the stash key is kept offline, it must be set to have its funds bonded to a particular controller. For non-spending actions, the controller has the funds of the stash behind it. For example, in nominating, staking, or voting, the controller can indicate its preference with the weight of the stash. It will never be able to actually move or claim the funds in the stash key. However, if someone does obtain your controller key, they could use it for slashable behavior, so you should still protect it and change it regularly.
Session keys are hot keys that must be kept online by a validator to perform network operations. Session keys are typically generated in the client, although they don't have to be. They are not meant to control funds and should only be used for their intended purpose. They can be changed regularly; your controller only need create a certificate by signing a session public key and broadcast this certificate via an extrinsic.
Setheum uses three session keys:
I'm Online: sr25519
BABE requires keys suitable for use in a Verifiable Random Function as well as for digital signatures. Sr25519 keys have both capabilities and so are used for BABE.
In the future, we plan to use a BLS key for GRANDPA because it allows for more efficient signature aggregation.
The original key derivation cryptography that was implemented for Polkadot and Substrate chains including Setheum was
ed25519, which is a Schnorr signature algorithm implemented over the Edward's Curve 25519 (so named due to the parameters of the curve equation).
Most cryptocurrencies, including Bitcoin and Ethereum, currently use ECDSA signatures on the secp256k1 curve. This curve is considered much more secure than NIST curves, which have possible backdoors from the NSA. The Curve25519 is considered possibly even more secure than this one and allows for easier implementation of Schnorr signatures. A recent patent expiration on it has made it the preferred choice for use in Polkadot and Setheum.
The choice of using Schnorr signatures over using ECDSA is not so cut and dry. As stated in Jeff Burdges' (a Web3 researcher) original forum post on the topic:
There is one sacrifice we make by choosing Schnorr signatures over ECDSA signatures for account keys: Both require 64 bytes, but only ECDSA signatures communicate their public key. There are obsolete Schnorr variants that support recovering the public key from a signature, but they break important functionality like hierarchical deterministic key derivation. In consequence, Schnorr signatures often take an extra 32 bytes for the public key.
But ultimately the benefits of using Schnorr signatures outweigh the tradeoffs, and future optimizations may resolve the inefficiencies pointed out in the quote above.
Some context: The Schnorr signatures over the Twisted Edward's Curve25519 are considered secure, however Ed25519 has not been completely devoid of its bugs. Most notably, Monero and all other CryptoNote currencies were vulnerable to a double spend exploit that could have potentially led to undetected, infinite inflation.
These exploits were due to one peculiarity in Ed25519, which is known as its cofactor of 8. The cofactor of a curve is an esoteric detail that could have dire consequences for the security of implementation of more complex protocols.
Conveniently, Mike Hamburg's Decaf paper provides a possible path forward to solving this potential bug. Decaf is basically a way to take Twisted Edward's Curves cofactor and mathematical change it with little cost to performance and gains to security.
The Decaf paper approach by the Ristretto Group was extended and implemented in Rust to include cofactor-8 curves like the Curve25519 and makes Schnorr signatures over the Edward's curve more secure.
The Web3 Foundation has implemented a Schnorr signature library using the more secure Ristretto compression over the Curve25519 in the Schnorrkel repository. Schnorrkel implements related protocols on top of this curve compression such as HDKD, MuSig, and a verifiable random function (VRF). It also includes various minor improvements such as the hashing scheme STROBE that can theoretically process huge amounts of data with only one call across the Wasm boundary.
The implementation of Schnorr signatures that is used in Polkadot and Setheum and implements the Schnorrkel protocols over the Ristretto compression of the Curve25519 is known as sr25519.
Not yet, but they will be. BLS signatures allow more efficient signature aggregation. Because GRANDPA validators are usually signing the same thing (e.g. a block), it makes sense to aggregate them, which can allow for other protocol optimizations.
As stated in the BLS library's README,
Boneh-Lynn-Shacham (BLS) signatures have slow signing, very slow verification, require slow and much less secure pairing friendly curves, and tend towards dangerous malleability. Yet, BLS permits a diverse array of signature aggregation options far beyond any other known signature scheme, which makes BLS a preferred scheme for voting in consensus algorithms and for threshold signatures.
Even though Schnorr signatures allow for signature aggregation, BLS signatures are much more efficient in some fashions. For this reason it will be one of the session keys that will be used by validators on the Setheum network and critical to the GRANDPA finality gadget.
Key discovery attack on BIP32-Ed25519 - Forum post detailing a potential attack on BIP32-Ed25519. A motivation for transition to the sr25519 variant.
Account signatures and keys in Polkadot - Original forum post by Web3 researcher Jeff Burdges.
From the introduction of Curve25519 into
The reason is the following: During summer of 2013, revelations from ex-consultant at [the] NSA Edward Snowden gave proof that [the] NSA willingly inserts backdoorsinto software, hardware components and published standards. While it is stillbelieved that the mathematics behind ECC (Elliptic-curve cryptography) are still sound and solid,some people (including Bruce Schneier [SCHNEIER]), showed their lack of confidencein NIST-published curves such as nistp256, nistp384, nistp521, for which constantparameters (including the generator point) are defined without explanation. Itis also believed that [the] NSA had a word to say in their definition. These curvesare not the most secure or fastest possible for their key sizes [DJB], andresearchers think it is possible that NSA have ways of cracking NIST curves.It is also interesting to note that SSH belongs to the list of protocols the NSAclaims to be able to eavesdrop. Having a secure replacement would make passiveattacks much harder if such a backdoor exists.However an alternative exists in the form of Curve25519. This algorithm has beenproposed in 2006 by DJB [Curve25519]. Its main strengths are its speed, itsconstant-time run time (and resistance against side-channel attacks), and itslack of nebulous hard-coded constants.