Extrinsic Horizon

[gavofyork]

Goals

• Clean up the mess of types and impractical trait architecture used in transactions.
• Ensure that it is easy to introduce and integrate new transaction crypto.
• Facilitate use-cases which depend on altering the Origin prior to fee payment including proxied-pays-fee and multisig-pays-fee (eg. Allow transaction payment from proxy target account #348).
• Unify ValidateUnsigned and SignedExtension (fixes Switch ValidateUnsigned over to SignedExtension #365).

Explanation

Stage 1: TransactionExtension

New transaction type - General

Deprecate the terminology of "Unsigned" when used for transactions/extrinsics owing to there now being "proper" unsigned transactions which obey the extension framework and "old-style" unsigned which do not. Instead we have General for the former and Bare for the latter.

Types of extrinsic are now therefore:

• Bare (no hardcoded signature, no Extra data; used to be known as "Unsigned")
  • Bare transactions (DEPRECATED): Gossiped, validated with ValidateUnsigned (DEPRECATED) and the _unsigned bits of TransactionExtension (DEPRECATED).
  • Inherents: Not gossiped, validated with ProvideInherent.
• Extended (with a TxExtension (ne Extra) data): Gossiped, validated via TransactionExtension type.
  • Signed transactions (with a hardcoded signature).
  • General transactions (without a hardcoded signature).

Three extrinsic version discriminators ("versions") are now permissible, according to RFC84:

• 0b000000101: Bare (used to be called "Unsigned"): contains neither a Signature nor Extension (i.e. Extra data). After Bare transactions are no longer supported, this will strictly identify Inherents only and will be renamed as such.
• 0b100000101: Old-school "Signed" Transaction: contains Signature and Extension (Extra data).
• 0b010000101: New-school "General" Transaction: contains Extension (Extra data), but no Signature.

Origin mutation at the extension level

The introduction of the TransactionExtension interface over SignedExtension would now allow any origin, not just the signed/unsigned origin supported in frame_system. Furthermore, the origin can be mutated by any extension in the pipeline during the validation step, as now extensions receive a RuntimeOrigin as input instead of an AccountId, which is the de facto frame_system origin type.

Authorization of non-standard origins

With a TransactionExtension, for both New-school General and Old-school Signed Transactions, it becomes trivial for authors to publish extensions to the mechanism for authorizing an Origin, e.g. through new kinds of key-signing schemes, ZK proofs, pallet state, mutations over pre-authenticated origins or any combination of the above.

Free transactions

As of today, there are 2 specific requirements for users who want to run any (signed) transaction:

1. payment for the nonce, either through acquiring ED or having some on-chain entity provide for your account
2. payment for the transaction fees, which even if ultimately refunded, a user still needs to have enough funds to pay in the worst case scenario

Nonce storage is a concept inherently tied to the account model and authorization currently in use in substrate based chains. However, with the introduction of TransactionExtensions, any form of authorization that fits in the validate + prepare flow and mutates the origin is allowed. This means that the nonce will be checked and incremented only for traditional signed origins.

Charging transaction fees makes sense only when the origin is an account, as only an account can hold currency. Fees protect the chain against spam, but an arbitrary origin is incapable of paying fees. Therefore, as with the nonce, transaction fees should be charged only when the origin is a traditional signed origin.

Therefore, the responsibility of protection against replay attacks (nonce increments), Sybil resistance (storage of nonce on chain) and on-chain spam (transaction fees) fall on the extension authorizing the non-standard origin.

This implies a much greater compute burden on the validation phase of extensions, which is also variable depending on what authorization is needed for a particular call, so the current weightless model of extensions cannot work. TransactionExtensions now provide a fn weight function to factor in the cost of this computation. Also, the validation done by an extension followed by authorization through origin mutation should generally mean that the call requiring said special authorization does not do the same checks again during the actual dispatch of the call.

N.B. It becomes apparent that the use case for what are now unsigned transactions, which are naturally free and currently handled via ValidateUnsigned, fits nicely into non-standard origin types validated by the TransactionExtension interface.

• FRAME: Create TransactionExtension as a replacement for SignedExtension #2280
• FRAME: Reintroduce TransactionExtension as a replacement for SignedExtension #3685

Stage 2: Begin removal of concept of Unsigned transactions:

Unsigned transactions should be validated by the TransactionExtension pipeline just like any other transaction. This will allow "bare" transactions to just mean inherents.

Current state of ValidateUnsigned

Each extrinsic call that is now validated using ValidateUnsigned instead of the SignedExtension pipeline has a particular set of arbitrary checks that are performed, described by the implementation of ValidateUnsigned::validate_unsigned.

Moving the logic of ValidateUnsigned into an instance of TransactionExtension

In principle, the logic of each ValidateUnsigned::validate_unsigned and ValidateUnsigned::pre_dispatch can be done in TransactionExtension::validate and TransactionExtension::prepare respectively. The call validation logic can be injected into the functions in TransactionExtension similarly to how SkipCheckIfFeeless is using the fn feeless_if closure decorator over pallet calls.

This validation logic must:

• mutate the origin to a bespoke, pallet specific origin which must be validated during call dispatch;
• return a Weight consumed (and benchmarked) by the validation logic, to be later used when calculating the weight of the overarching extension.

Creating bespoke origins for each call within a pallet that wants to use this new "unsigned" interface can be done during the pallet macro expansion. It could either be a single origin per pallet with a predefined name, such as Origin::Preauthorized, or it could be one origin variant for each call that uses the interface, named after the call name, e.g. fn foo_bar(origin: OriginFor<T>) -> DispatchResult -> Origin::FooBar.

In short, an extension that validates calls that were previously unsigned now validates free calls that have a non-standard origin. The extension should run the validation logic defined in the pallet call closure and mutate the origin to a pre-approved pallet origin, to be later checked during dispatch. Most of this can be hidden away behind procedural macros so users don't have to manually define and interact these extensions and origins.

All transactions have an origin

With the change to fee payment and ValidateUnsigned, all valid transactions must reach the call dispatch phase of extrinsic application with some origin, as no origin would mean that nobody authorized this transaction.

In order to do this, there could either be an extension in the pipeline invalidating transactions without an origin. The extension could be anywhere in the pipeline as long as there are no possible origin mutations after its own validate cycle, so somewhere towards the end of it. Alternatively, this can be enforced in the implementation of Applyable::apply, but it would need special handling when validating transaction for the transaction pool and, given that extensions have versioning now through RFC99, it's probably cleaner overall to have it as an extension, as any changes to this logic would be reflected in a version bump.

Other action items

• Deprecate ValidateUnsigned & #[validate_unsigned] (logic should be moved to TransactionExtension).
• Remove (if unused) or migrate uses of validate_unsigned.
• Deprecate Extrinsic and break it down into logical components:
  • Use ExtrinsicLike as the interface with Block, capable of differentiating inherents from transactions;
  • Break down CreateSignedTransaction into a CreateTransaction trait family, as described in issue 3571;
  • Remove the SendTransactionTypes interface;
  • Remove TestXt and use concrete UncheckedExtrinsic types in tests.

Stage 2.5: Integrated semantic description

Transactions should include a metadata extension VerifyMetadata

• Extension places optional hash in the signed payload which corresponds to the hash of the metadata for the chain.
• Online signers simply hash the metadata and place it in the transaction.
• Offline signers should generally be provided with witness data to be able to construct a sparse Merkle tree of the metadata and display the transactions meaning to the user at the time of signing.

Stage 3 PR: New APIs, deprecate old-school.

All extrinsics are now validated either by ProvideInherent (which is an aggregated type of the runtime) or TransactionExtension (which is an explicitly configured type of the runtime).

Bare transactions are removed, so:

• "Extended transactions" are synonymous with "transactions" and we drop the "Extended" nomenclature.
• Similarly, "Bare extrinsics" will be synonymous with "Inherents" and we will drop the "Bare" nomenclature.

This leaves us with three types/versions of Extrinsics:

• 0b000000101: Inherent: contains neither a Signature nor Extension (i.e. Extra data). Validated by ProvideInherent.
• 0b100000101: Old-school "Signed" Transaction: contains Signature and Extension.
• 0b010000101: New-school "General" Transaction: contains Extension, but no Signature.

Action items:

Remove deprecated items:

• Remove ValidateUnsigned and #[validate_unsigned]; Bare extrinsics are always Inherents and therefore always fail validation/pre-dispatch in UncheckedExtrinsic.

Tidy up language:

• Rename Signed to Account and ensure_signed to ensure_account.

Action items done in PR3685

Move test code to regular types:

• Remove old test transaction types and just use the regular CheckedExtrinsic and UncheckedExtrinsic.
• Remove ExtrinsicWrapper and use the regular CheckedExtrinsic and UncheckedExtrinsic.

Remove all needless reliance on traits with a defunct model of transactions:

• Stop using and deprecate old items in Extrinsic trait:
  • Extrinsic::is_signed, Extrinsic::new, Extrinsic::SignaturePayload.
• Deprecate trait SignaturePayload, trait CreateSignedTransaction.
• Introduce new CreateTransaction traits, simplifying transaction creation for tests & OCW.
  • fn createTransaction(from: Signer, call: Call) -> UncheckedExtrinsic;
    • called inside a state-queryable environment (so pallet storage can be gotten).
  • Implementing this is the requirement of the runtime, but can be mostly automated with utility structs.

Stage 4: Remove rest of deprecated API

• Remove SignaturePayload, CreateSignedTransaction and deprecated items in Extrinsic.

Other notes:

• This isn't really what we mean - we really want a tx.is_transaction, since bare transactions may be gossipped as in the case of Frontier txs or claims. This will be sorted once we dispense with the concept of bare transactions and make inherents the only possible type of extrinsics which are bare. At this point we can change this to tx.is_transaction(). - substrate/client/transaction-pool/src/lib.rs

[bkchr]

Transactions should include a metadata extension VerifyMetadata

Wouldn't this also not just be a TransactionExtension?

[ggwpez]

Remove or migrate uses of validate_unsigned:
Frontier: migrate.
Claims: remove.

What is the criteria for migration vs removal in the other pallets that use it?

[gavofyork]

If the feature is out of use (as I expected claims was), then they can be deprecated and removed. But since claim was actually used only 5 days ago last, I guess we ought to keep it around 😕.

[bkchr]

We should also change SignedPayload to just always hash the data that we sign/verify. I don't really see the advantage of doing it this way.

[bkchr]

	fn validate(
		&self,
		origin: OriginOf<Call>,
		call: &Call,
		info: &DispatchInfoOf<Call>,
		len: usize,
		context: &mut Context,
		self_implicit: Self::Implicit,
		inherited_implication: &impl Encode,
	) -> ValidateResult<Self::Val, Call>;

	pub type ValidateResult<Val, Call> =
		Result<(ValidTransaction, Val, OriginOf<Call>), TransactionValidityError>;

Instead of this, I think it would be better to change the interface to:

	fn validate(
		&self,
		origin: Option<OriginOf<Call>>,
                signer: Option<Account>,
		call: &Call,
		info: &DispatchInfoOf<Call>,
		len: usize,
		context: &mut Context,
		self_implicit: Self::Implicit,
		inherited_implication: &impl Encode,
	) -> ValidateResult<Self::Val, Call>;

	pub type ValidateResult<Val, Call> =
		Result<(ValidTransaction, Val, Option<OriginOf<Call>>), TransactionValidityError>;

So, basically we would make the origin an Option and go back to passing the signer. This would bring us the advantage of chaining logic, but first let's look at the current structure. If I want to use the SkipFeelessIf with the TransactionPayment extension, I need to nest them. Aka SkipFeelessIf<TransactionPayment>. Now let's assume we actually want to have General transactions that are created by an offchain worker for example. The problem would be that we need to skip at least TransactionPayment. So, we would need to nest again CustomExt<SkipFeelessIf<TransactionPayment>>>. (We can probably continue this for multiple different extensions).

IMO it would be better to do this like:

Exts = (
     CustomExt,
     SkipFeelessIf,
     TransactionPayment,
);

At the beginning origin would be None. The extensions can set the origin to Some, which would signal the following extensions like TransactionPayment to not try to deduct any fees. If at the end of executing all extensions, origin would still be None, we would reject the transaction. This way we would also get back the old behavior of rejecting unsigned/general transactions "by default".