* report_slashes plumbing in Substrate
Notably delays the SetRetired event until it provides a slash report or the set
after it becomes the set to report its slashes.
* Add dedicated AcceptedHandover event
* Add SlashReport TX to Tributary
* Create SlashReport TXs
* Handle SlashReport TXs
* Add logic to generate a SlashReport to the coordinator
* Route SlashReportSigner into the processor
* Finish routing the SlashReport signing/TX publication
* Add serai feature to processor's serai-client
* Use an extended timeout for DKGs specifically
* Add a log statement when message-queue connection fails
* Add a 60 second keep-alive to connections
* Use zalloc for processor/message-queue/coordinator
An additional layer which protects us against edge cases with Zeroizing
(objects which don't support it or don't miss it).
* Add further logs to message-queue
* Further increase re-attempt timeouts in CI
* Remove misplaced continue inmessage-queue client
Fixes observed CI failures.
* Revert "Further increase re-attempt timeouts in CI"
This reverts commit 3723530cf6.
Makes RemoveParticipantDueToDkg a voted-on event instead of a Provided.
This removes the requirement for offline parties to be able to fully validate
blame, yet unfortunately lets an dishonest supermajority have an honest node
label any arbitrary node as dishonest.
Corrects a variety of `.i(...)` calls which panicked when they shouldn't have.
Cleans up a couple no-longer-used storage values.
With a DKG removal comes a reduction in the amount of participants which was
ignored by re-attempts.
Now, we determine n/i based on the parties removed, and deterministically
obtain the context of who was removd.
* Schedule re-attempts and add a (not filled out) match statement to actually execute them
A comment explains the methodology. To copy it here:
"""
This is because we *always* re-attempt any protocol which had participation. That doesn't
mean we *should* re-attempt this protocol.
The alternatives were:
1) Note on-chain we completed a protocol, halting re-attempts upon 34%.
2) Vote on-chain to re-attempt a protocol.
This schema doesn't have any additional messages upon the success case (whereas
alternative #1 does) and doesn't have overhead (as alternative #2 does, sending votes and
then preprocesses. This only sends preprocesses).
"""
Any signing protocol which reaches sufficient participation will be
re-attempted until it no longer does.
* Have the Substrate scanner track DKG removals/completions for the Tributary code
* Don't keep trying to publish a participant removal if we've already set keys
* Pad out the re-attempt match a bit more
* Have CosignEvaluator reload from the DB
* Correctly schedule cosign re-attempts
* Actuall spawn new DKG removal attempts
* Use u32 for Batch ID in SubstrateSignableId, finish Batch re-attempt routing
The batch ID was an opaque [u8; 5] which also included the network, yet that's
redundant and unhelpful.
* Clarify a pair of TODOs in the coordinator
* Remove old TODO
* Final comment cleanup
* Correct usage of TARGET_BLOCK_TIME in reattempt scheduler
It's in ms and I assumed it was in s.
* Have coordinator tests drop BatchReattempts which aren't relevant yet may exist
* Bug fix and pointless oddity removal
We scheduled a re-attempt upon receiving 2/3rds of preprocesses and upon
receiving 2/3rds of shares, so any signing protocol could cause two re-attempts
(not one more).
The coordinator tests randomly generated the Batch ID since it was prior an
opaque byte array. While that didn't break the test, it was pointless and did
make the already-succeeded check before re-attempting impossible to hit.
* Add log statements, correct dead-lock in coordinator tests
* Increase pessimistic timeout on recv_message to compensate for tighter best-case timeouts
* Further bump timeout by a minute
AFAICT, GH failed by just a few seconds.
This also is worst-case in a single instance, making it fine to be decently long.
* Further further bump timeout due to lack of distinct error
* Move logic for evaluating if a cosign should occur to its own file
Cleans it up and makes it more robust.
* Have expected_next_batch return an error instead of retrying
While convenient to offer an error-free implementation, it potentially caused
very long lived lock acquisitions in handle_processor_message.
* Unify and clean DkgConfirmer and DkgRemoval
Does so via adding a new file for the common code, SigningProtocol.
Modifies from_cache to return the preprocess with the machine, as there's no
reason not to. Also removes an unused Result around the type.
Clarifies the security around deterministic nonces, removing them for
saved-to-disk cached preprocesses. The cached preprocesses are encrypted as the
DB is not a proper secret store.
Moves arguments always present in the protocol from function arguments into the
struct itself.
Removes the horribly ugly code in DkgRemoval, fixing multiple issues present
with it which would cause it to fail on use.
* Set SeraiBlockNumber in cosign.rs as it's used by the cosigning protocol
* Remove unnecessary Clone from lambdas in coordinator
* Remove the EventDb from Tributary scanner
We used per-Transaction DB TXNs so on error, we don't have to rescan the entire
block yet only the rest of it. We prevented scanning multiple transactions by
tracking which we already had.
This is over-engineered and not worth it.
* Implement borsh for HasEvents, removing the manual encoding
* Merge DkgConfirmer and DkgRemoval into signing_protocol.rs
Fixes a bug in DkgConfirmer which would cause it to improperly handle indexes
if any validator had multiple key shares.
* Strictly type DataSpecification's Label
* Correct threshold_i_map_to_keys_and_musig_i_map
It didn't include the participant's own index and accordingly was offset.
* Create TributaryBlockHandler
This struct contains all variables prior passed to handle_block and stops them
from being passed around again and again.
This also ensures fatal_slash is only called while handling a block, as needed
as it expects to operate under perfect consensus.
* Inline accumulate, store confirmation nonces with shares
Inlining accumulate makes sense due to the amount of data accumulate needed to
be passed.
Storing confirmation nonces with shares ensures that both are available or
neither. Prior, one could be yet the other may not have been (requiring an
assert in runtime to ensure we didn't bungle it somehow).
* Create helper functions for handling DkgRemoval/SubstrateSign/Sign Tributary TXs
* Move Label into SignData
All of our transactions which use SignData end up with the same common usage
pattern for Label, justifying this.
Removes 3 transactions, explicitly de-duplicating their handlers.
* Remove CurrentlyCompletingKeyPair for the non-contextual DkgKeyPair
* Remove the manual read/write for TributarySpec for borsh
This struct doesn't have any optimizations booned by the manual impl. Using
borsh reduces our scope.
* Use temporary variables to further minimize LoC in tributary handler
* Remove usage of tuples for non-trivial Tributary transactions
* Remove serde from dkg
serde could be used to deserialize intenrally inconsistent objects which could
lead to panics or faults.
The BorshDeserialize derives have been replaced with a manual implementation
which won't produce inconsistent objects.
* Abstract Future generics using new trait definitions in coordinator
* Move published_signed_transaction to tributary/mod.rs to reduce the size of main.rs
* Split coordinator/src/tributary/mod.rs into spec.rs and transaction.rs
* Update ValidatorSets with a remove_participant call
* Add DkgRemoval, a sign machine for producing the relevant MuSig signatures
* Don't use position-dependent u8s yet Public when removing validators from the DKG
* Add DkgRemovalPreprocess, DkgRemovalShares
Implementation is via a new publish_tributary_tx lambda.
This is code is a copy-pasted mess which will need to be cleaned up.
* Only allow non-removed validators to vote for removals
Otherwise, it's risked that the remaining validators fall below 67% of the
original set.
* Correct publish_serai_tx, which was prior publish_set_keys in practice
* Remove NetworkId from processor-messages
Because intent binds to the sender/receiver, it's not needed for intent.
The processor knows what the network is.
The coordinator knows which to use because it's sending this message to the
processor for that network.
Also removes the unused zeroize.
* ProcessorMessage::Completed use Session instead of key
* Move SubstrateSignId to Session
* Finish replacing key with session
* Add a function to deterministically decide which Serai blocks should be co-signed
Has a 5 minute latency between co-signs, also used as the maximal latency
before a co-sign is started.
* Get all active tributaries we're in at a specific block
* Add and route CosignSubstrateBlock, a new provided TX
* Split queued cosigns per network
* Rename BatchSignId to SubstrateSignId
* Add SubstrateSignableId, a meta-type for either Batch or Block, and modularize around it
* Handle the CosignSubstrateBlock provided TX
* Revert substrate_signer.rs to develop (and patch to still work)
Due to SubstrateSigner moving when the prior multisig closes, yet cosigning
occurring with the most recent key, a single SubstrateSigner can be reused.
We could manage multiple SubstrateSigners, yet considering the much lower
specifications for cosigning, I'd rather treat it distinctly.
* Route cosigning through the processor
* Add note to rename SubstrateSigner post-PR
I don't want to do so now in order to preserve the diff's clarity.
* Implement cosign evaluation into the coordinator
* Get tests to compile
* Bug fixes, mark blocks without cosigners available as cosigned
* Correct the ID Batch preprocesses are saved under, add log statements
* Create a dedicated function to handle cosigns
* Correct the flow around Batch verification/queueing
Verifying `Batch`s could stall when a `Batch` was signed before its
predecessors/before the block it's contained in was cosigned (the latter being
inevitable as we can't sign a block containing a signed batch before signing
the batch).
Now, Batch verification happens on a distinct async task in order to not block
the handling of processor messages. This task is the sole caller of verify in
order to ensure last_verified_batch isn't unexpectedly mutated.
When the processor message handler needs to access it, or needs to queue a
Batch, it associates the DB TXN with a lock preventing the other task from
doing so.
This lock, as currently implemented, is a poor and inefficient design. It
should be modified to the pattern used for cosign management. Additionally, a
new primitive of a DB-backed channel may be immensely valuable.
Fixes a standing potential deadlock and a deadlock introduced with the
cosigning protocol.
* Working full-stack tests
After the last commit, this only required extending a timeout.
* Replace "co-sign" with "cosign" to make finding text easier
* Update the coordinator tests to support cosigning
* Inline prior_batch calculation to prevent panic on rotation
Noticed when doing a final review of the branch.
* Have processor report errors during the DKG to the coordinator
* Add RemoveParticipant, InvalidDkgShare to coordinator
* Route DKG blame around coordinator
* Allow public construction of AdditionalBlameMachine
Necessary for upcoming work on handling DKG blame in the processor and
coordinator.
Additionally fixes a publicly reachable panic when commitments parsed with one
ThresholdParams are used in a machine using another set of ThresholdParams.
Renames InvalidProofOfKnowledge to InvalidCommitments.
* Remove unused error from dleq
* Implement support for VerifyBlame in the processor
* Have coordinator send the processor share message relevant to Blame
* Remove desync between processors reporting InvalidShare and ones reporting GeneratedKeyPair
* Route blame on sign between processor and coordinator
Doesn't yet act on it in coordinator.
* Move txn usage as needed for stable Rust to build
* Correct InvalidDkgShare serialization
* Update the coordinator to give key shares based on weight, not based on existence
Participants are now identified by their starting index. While this compiles,
the following is unimplemented:
1) A conversion for DKG `i` values. It assumes the threshold `i` values used
will be identical for the MuSig signature used to confirm the DKG.
2) Expansion from compressed values to full values before forwarding to the
processor.
* Add a fn to the DkgConfirmer to convert `i` values as needed
Also removes TODOs regarding Serai ensuring validator key uniqueness +
validity. The current infra achieves both.
* Have the Tributary DB track participation by shares, not by count
* Prevent a node from obtaining 34% of the maximum amount of key shares
This is actually mainly intended to set a bound on message sizes in the
coordinator. Message sizes are amplified by the amount of key shares held, so
setting an upper bound on said amount lets it determine constants. While that
upper bound could be 150, that'd be unreasonable and increase the potential for
DoS attacks.
* Correct the mechanism to detect if sufficient accumulation has occured
It used to check if the latest accumulation hit the required threshold. Now,
accumulations may jump past the required threshold. The required mechanism is
to check the threshold wasn't prior met and is now met.
* Finish updating the coordinator to handle a multiple key share per validator environment
* Adjust stategy re: preventing noce reuse in DKG Confirmer
* Add TODOs regarding dropped transactions, add possible TODO fix
* Update tests/coordinator
This doesn't add new multi-key-share tests, it solely updates the existing
single key-share tests to compile and run, with the necessary fixes to the
coordinator.
* Update processor key_gen to handle generating multiple key shares at once
* Update SubstrateSigner
* Update signer, clippy
* Update processor tests
* Update processor docker tests
* add slash tx
* ignore unsigned tx replays
* verify that provided evidence is valid
* fix clippy + fmt
* move application tx handling to another module
* partially handle the tendermint txs
* fix pr comments
* support unsigned app txs
* add slash target to the votes
* enforce provided, unsigned, signed tx ordering within a block
* bug fixes
* add unit test for tendermint txs
* bug fixes
* update tests for tendermint txs
* add tx ordering test
* tidy up tx ordering test
* cargo +nightly fmt
* Misc fixes from rebasing
* Finish resolving clippy
* Remove sha3 from tendermint-machine
* Resolve a DoS in SlashEvidence's read
Also moves Evidence from Vec<Message> to (Message, Option<Message>). That
should meet all requirements while being a bit safer.
* Make lazy_static a dev-depend for tributary
* Various small tweaks
One use of sort was inefficient, sorting unsigned || signed when unsigned was
already properly sorted. Given how the unsigned TXs were given a nonce of 0, an
unstable sort may swap places with an unsigned TX and a signed TX with a nonce
of 0 (leading to a faulty block).
The extra protection added here sorts signed, then concats.
* Fix Tributary tests I broke, start review on tendermint/tx.rs
* Finish reviewing everything outside tests and empty_signature
* Remove empty_signature
empty_signature led to corrupted local state histories. Unfortunately, the API
is only sane with a signature.
We now use the actual signature, which risks creating a signature over a
malicious message if we have ever have an invariant producing malicious
messages. Prior, we only signed the message after the local machine confirmed
it was okay per the local view of consensus.
This is tolerated/preferred over a corrupt state history since production of
such messages is already an invariant. TODOs are added to make handling of this
theoretical invariant further robust.
* Remove async_sequential for tokio::test
There was no competition for resources forcing them to be run sequentially.
* Modify block order test to be statistically significant without multiple runs
* Clean tests
---------
Co-authored-by: Luke Parker <lukeparker5132@gmail.com>
Removes last_block as an argument from Tendermint. It now loads from the DB as
needed. While slightly less performant, it's easiest and should be fine.
Impls a LocalP2p for testing.
Moves rebroadcasting into Tendermint, since it's what knows if a message is
fully valid + original.
Removes TributarySpec::validators() HashMap, as its non-determinism caused
different instances to have different round robin schedules. It was already
prior moved to a Vec for this issue, so I'm unsure why this remnant existed.
Also renames the GH no-std workflow from the prior commit.
The existing code was almost entirely applicable. It just needed to be scoped
with an ID. While the handle function is now a bit convoluted, I don't see a
better option.
