feat: enhance escrow security and optimize state storage - #373
Merged
godamongstmen897 merged 137 commits intoSep 1, 2026
Merged
feat: enhance escrow security and optimize state storage#373godamongstmen897 merged 137 commits into
godamongstmen897 merged 137 commits into
Conversation
|
@ayo-ola0710 Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits. You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀 |
tax_withholding_deductions already computes its split via the shared split_round_nearest helper (round-to-nearest, with the second part derived by subtraction rather than a second division), which already guarantees tax_amount + net_amount == gross_amount exactly and rounds to the nearest integer rather than always flooring. But no test exercised that guarantee directly: the two tests named after tax_withholding_deductions actually call the unrelated multisig_transfer_admin function, and the only test that does call tax_withholding_deductions only covers the zero-balance failure path. Adds 8 tests against the real entry point: value conservation across a spread of exact and inexact rates, round-to-nearest in both directions (proving it neither floors nor ceils), the 0%/100% boundaries, the smallest indivisible unit, rejection of a rate above 100%, and independence across separate milestones. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…tests (#290) Harden cancel_escrow with two missing validation rules and add a full test suite covering every guard, happy path, post-cancel state, event structure, and milestone isolation. Production changes (lib.rs): - Add EmergencyPaused guard: cancel_escrow now returns Error::Paused when the contract is emergency-paused, consistent with all other user-facing endpoints - Add duplicate-cancel guard: a second call while CancelLock is already active returns Error::EscrowLocked, preventing race conditions and redundant lock-sets Tests added (test.rs) — 20 new tests: Invalid address guards: - test_cancel_escrow_zero_account_address_rejected - test_cancel_escrow_zero_contract_address_rejected Not-initialized guard: - test_cancel_escrow_not_initialized_fails Not-funded guard: - test_cancel_escrow_not_funded_fails Unauthorized guards: - test_cancel_escrow_stranger_unauthorized - test_cancel_escrow_arbiter_unauthorized - test_cancel_escrow_admin_unauthorized Emergency-paused guard: - test_cancel_escrow_while_paused_fails Duplicate-cancel guard: - test_cancel_escrow_duplicate_call_fails - test_cancel_escrow_freelancer_duplicate_after_client_fails Happy paths: - test_cancel_escrow_client_succeeds - test_cancel_escrow_freelancer_succeeds Post-cancel state validation: - test_cancel_escrow_blocks_fund - test_cancel_escrow_blocks_mark_delivered - test_cancel_escrow_blocks_approve_milestone - test_cancel_escrow_blocks_raise_dispute Event validation: - test_cancel_escrow_emits_exactly_one_event - test_cancel_escrow_event_contains_correct_caller Milestone state isolation: - test_cancel_escrow_does_not_mutate_milestones - test_cancel_escrow_all_milestones_released_still_succeeds All 218 tests pass.
…tio-split-tests test(tax_withholding_deductions): verify ratio-split precision (#299)
…uctions-suite test(escrow): add unit test matrix for tax_withholding_deductions (#305)
…nsions-suite test(escrow): add unit test suite for milestone_time_extensions (#289)
…tests (#293) Expand CancelEscrowInitiatedEvent with full operational context so downstream indexers can reconstruct the complete cancellation state from the event payload alone, without querying contract storage. Production changes (lib.rs): - Expand CancelEscrowInitiatedEvent with six new fields: caller_is_client bool — true if initiator is the client, false if freelancer client Address — registered client address freelancer Address — registered freelancer address token Address — escrow token contract address milestone_count u32 — number of milestones at cancellation time total_amount i128 — aggregate milestone total (pre-release) - Update cancel_escrow event emission to populate all new fields, deriving caller_is_client from (caller == meta.client) No other code changed. Tests added (test.rs) — 14 new tests (396 -> 410): Event count: - test_cancel_escrow_event_emitted_exactly_once contract_id field: - test_cancel_escrow_event_contract_id_correct caller field: - test_cancel_escrow_event_caller_is_client_address - test_cancel_escrow_event_caller_is_freelancer_address caller_is_client role field: - test_cancel_escrow_event_caller_is_client_true_for_client - test_cancel_escrow_event_caller_is_client_false_for_freelancer client / freelancer / token fields: - test_cancel_escrow_event_client_field_correct - test_cancel_escrow_event_freelancer_field_correct - test_cancel_escrow_event_token_field_correct milestone_count field: - test_cancel_escrow_event_milestone_count_single - test_cancel_escrow_event_milestone_count_multiple total_amount field: - test_cancel_escrow_event_total_amount_correct_single_milestone - test_cancel_escrow_event_total_amount_correct_multi_milestone Full indexer round-trip: - test_cancel_escrow_event_full_indexer_parse All 410 tests pass.
…es (#295) Require both client and freelancer to approve before cancel_escrow sets CancelLock. A single-signature call records approval only and returns immediately — the escrow remains fully operational until the second party signs. Production changes (lib.rs): - Add DataKey::CancelApproval — persistent u32 bitmask (bit 0 = client, bit 1 = freelancer) tracking which parties have signed - Add CancelApprovalRecordedEvent — emitted on first/partial approval with the current mask (symbol: cxlappr) - Add CancelApprovalRevokedEvent — emitted when a party withdraws their approval (symbol: cancelrev) - Rewrite cancel_escrow two-phase logic: * Zero-address guard unchanged * Single call records caller_bit in CancelApproval mask, emits cxlappr, returns Ok — lock NOT set * Duplicate call from same party returns Error::InvalidStatus * Second call (mask == 3) clears CancelApproval, sets CancelLock, emits final cancel event * Call while lock already active returns Error::EscrowLocked - Add revoke_cancel_approval(env, caller) -> Result<(), Error> * Party can withdraw their approval before lock fires * Returns InvalidStatus if no approval recorded or lock already active * Returns Unauthorized for non-parties * Emits cancelrev event with updated mask cancel_escrow_test.rs — updated two existing tests: - test_cancel_escrow_sets_lock_and_emits_event: updated to call both parties and assert lock fires only after second signature - test_cancel_escrow_by_freelancer_succeeds: updated to use two-phase sequence (freelancer first, then client) New tests (test.rs) — 19 new tests (396 -> 415): Single-signature reverts: - test_cancel_escrow_single_client_signature_does_not_lock - test_cancel_escrow_single_freelancer_signature_does_not_lock - test_cancel_escrow_duplicate_same_party_reverts - test_cancel_escrow_duplicate_freelancer_reverts Two-party approval completes cancel: - test_cancel_escrow_client_then_freelancer_locks - test_cancel_escrow_freelancer_then_client_locks - test_cancel_escrow_third_call_after_lock_fails Approval bitmask and events: - test_cancel_escrow_first_signature_emits_approval_event - test_cancel_escrow_freelancer_first_emits_mask_2 - test_cancel_escrow_second_signature_emits_cancel_event - test_cancel_escrow_approval_key_cleared_after_lock revoke_cancel_approval: - test_revoke_cancel_approval_client_before_second_sig - test_revoke_then_reapprove_completes_cancel - test_revoke_cancel_approval_without_prior_approval_fails - test_revoke_cancel_approval_after_lock_fails - test_revoke_cancel_approval_non_party_unauthorized - test_revoke_cancel_approval_emits_event Unauthorized callers: - test_cancel_escrow_arbiter_cannot_sign - test_cancel_escrow_admin_cannot_sign All 415 tests pass.
Audited tax_withholding_deductions's existing validation against this
issue's requirement ("assert that bad setups are rejected immediately
with descriptive error types") and found the existing coverage already
extensive: NotInitialized, NotFunded, InvalidMilestone, InvalidRatio,
InvalidAmount (zero balance, overflow), and InvalidStatus for
Released/Refunded milestones are all implemented and tested (test.rs,
tax_withholding_tests.rs).
One real gap: the status check only excluded Released/Refunded, not
Disputed — unlike raise_dispute_inner and resolve_dispute elsewhere in
this file, which both treat Disputed as its own case via an exhaustive
match. A disputed milestone's funds are meant to be frozen pending
resolve_dispute; tax_withholding_deductions could still compute and
persist a TaxWithholdingRecord for one, moving money around that freeze.
No existing test combined raise_dispute with tax_withholding_deductions,
so this was uncovered on both the implementation and test side.
Fixes the gap by converting the two equality checks to an exhaustive
match over MilestoneStatus (so a future new variant fails to compile
here instead of silently falling through as allowed), matching the
established pattern elsewhere in this file, and adds a test exercising
it via the existing dual-signature test fixture.
Not touched: admin_tax_withholding_deductions, a separately-implemented
sibling function (per its own doc comment, "arrived from a separate PR
under the same name") that has no status check at all — a larger gap,
but outside this issue's named scope (Module/Component:
tax_withholding_deductions).
…uth-preconditions Feat/issue 330 341 harden auth preconditions
…lding-validation fix(tax_withholding_deductions): reject Disputed milestones (#298)
The conflict was just a module declaration; it now sits under #[cfg(test)] alongside the others, matching what #429 established. Two compile fixes in the new suite: - setup_funded_escrow was not in scope. Imported from crate::test, the same way admin_override_cancel_tests.rs does it. - DataKey::YieldRateBps does not exist. admin_set_yield_rate persists the rate as the yield_rate field of the YieldConfig entry under DataKey::YieldConfig, so read_yield_rate reads that instead. None still means "never written", which is what the no-mutation cases want. Three of the new tests then failed, and they were right to. They pause with emergency_pause and expect admin_set_yield_rate to return Paused, but it called only assert_not_paused, which reads DataKey::Paused -- the flag admin_pause_escrow sets. The emergency pause is a separate, stronger freeze recorded under DataKey::Ep, and nothing was checking it here, so a yield-rate change went straight through an emergency pause while the weaker admin pause blocked it. admin_set_yield_rate now rejects under either flag. That is the hardening this PR set out to add; the tests had simply reached for the pause that was not wired up. No existing test asserted the old behaviour. 558 tests passing / WASM release build OK
…te-guards feat: harden caller auth and precondition guards in admin_set_yield_rate
Dropping the EmergencyPauseLock dance from emergency_pause_admin_override is correct and is the point of #399: the lock exists to close a reentrancy window around external calls, and this path makes none -- it reads the flag, compares, writes it back, and emits an event. EpLk is still taken by the paths that do call out. Three names had to be corrected against the enum as it actually exists: - DataKey::EmergencyPaused -> DataKey::Ep in lib.rs - DataKey::EmergencyPauseLock -> DataKey::EpLk in the new test suite One test then failed on the event tally. It calls the override twice and expects the count to go 1 then 2, but env.events().all() reports the most recent contract invocation rather than a running total, so the second call reports 1. Adjusted to assert the second call emits exactly one event, with the existing payload check confirming it is the new one. This is the same env behaviour that #418 and #438 ran into. 571 tests passing / WASM release build OK
…-override-storage-footprint #399 - fix(emergency_pause_admin_override): remove unnecessary EmergencyPauseLock overhead
The branch's own copy of admin_pause_escrow does not parse. A line from
the previous version was left dangling after the new writes:
.set(&DataKey::EmergencyPauseLock, &true);
env.storage().instance().set(&DataKey::Paused, &true);
.set(&DataKey::EpLk, &true); <- orphaned continuation
That produced 90 errors, all downstream of "expected expression, found
`.`". DataKey::EmergencyPauseLock is also not a variant -- the enum
calls it EpLk -- in lib.rs and twice in the new test suite.
Rebuilt the function around the two guards this PR is actually for:
assert_emergency_pause_not_locked before any write, and an early return
when the escrow is already paused so a redundant call mutates nothing.
Because that early return now handles the repeat case, the inner
`if !already_paused` that main used to gate the event is redundant, and
the publish is unconditional inside the lock.
Two of the new tests then failed on event tallies. They read
pause_event_count after is_paused / is_lock_held, and those helpers go
through env.as_contract -- env.events().all() reports the most recent
invocation, not a running total. Reordered so the tally is read first.
The idempotency case now asserts the second call emits nothing at all,
which is what "no-op" means here. Same env behaviour as #418, #428
and #438.
576 tests passing / WASM release build OK
…use-escrow-guards feat: harden caller authorization and precondition guards in admin_pause_escrow (closes #349)
…admin_override Resolve the merge against main, which had already landed a narrower version of this event under the `pf_ovr` topic. The branch declared a second `PlatformFeeAllocationOverrideEvent` struct (same name, plus `contract_id` and `locked`) and published it a second time under a `pfovrride` topic after the allocation lock was released -- two identically named types will not compile, and the two publishes are the same event emitted twice. Unified to one struct carrying the branch's richer payload and one publish, kept inside the lock guard where main emits it. Retained main's already-published `pf_ovr` topic rather than renaming a topic that is live on the default branch. Updated main's existing assertion for the two new fields and repointed the branch's four new tests at `pf_ovr`. 580 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat(#396): emit PlatformFeeAllocationOverrideEvent from pf_alloc_admin_override
…in_override Resolve the merge against main. Three fixes were needed on top of the branch: * The branch predates #434, so taking its rewritten pf_alloc_admin_override body verbatim would have silently dropped the PlatformFeeAllocationOverrideEvent that call now emits. Kept the single-write optimisation and the event. * The branch deletes `mod admin_override_cancel_tests;` -- stale-branch damage that would drop a whole test module from the build. Kept ours. * Its new test referenced `DataKey::EmergencyPauseLock`, which does not exist; the variant is the short key `EpLk`. Also folded the branch's `setup_pf_alloc_escrow` helper together with #434's `setup_locked_pf_alloc` instead of landing two near-identical copies. 591 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
perf(#397): reduce ledger storage footprint of pf_alloc_admin_override
Main had already migrated MilestoneTimeExtension writes to temporary
storage, so the branch's remaining substance is the read fallback --
`load_time_extension` now checks temporary first and falls back to the
persistent key. Without it, any escrow whose extension was written before
the migration reads back 0 and auto-releases on the unextended deadline.
Fixes on top of the branch:
* Both new i128-extreme tests wrote storage from outside a contract
context, which panics in soroban-sdk 22 ("not accessible outside of a
contract"). Wrapped the seeding writes in `env.as_contract`.
* The branch reorders `mod admin_override_cancel_tests;` above `mod test;`
and drops the `#[cfg(test)]` attributes main added; kept ours.
593 tests pass; wasm32 release build is clean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
fix-checked-arithmetic-footprint
Three footprint reductions, all consistent with the pattern main already applies elsewhere: * admin_tax_withholding_deductions no longer sets and immediately removes TaxWithholdingExecutionLock. Main had already hoisted every guard above the lock, leaving a set/remove pair that protected nothing. * admin_override_tax_release no longer writes the MilestoneReleased temporary flag. Nothing reads it (is_milestone_released_flag is dead_code) and the persistent status already carries Released -- the same rationale main documents for two other override paths under #383. * multisig_admin_override_refund removes MultisigLocked instead of writing `false`, so no stale entry is left on the ledger. Two fixes on top of the branch: it re-added guards main had already hoisted (kept ours), and its new test read storage through env.as_contract before checking the event tally -- env.events().all() only reflects the most recent invocation, so the buffer was already cleared. Reordered the assertions. 596 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Minor Admin Changes
admin_override_refund and multisig_admin_override_refund now reject their illegal source state (paused / not multisig-locked) before touching any job or milestone storage, so a rejected call writes nothing. Resolved against main: unioned the test module lists, kept main's crate-scoped `setup_funded_escrow`, and dropped the branch's duplicate imports. Replaced the branch's inlined Paused read with the existing `assert_not_paused` helper rather than adding a third hand-rolled copy. 600 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…verride-refund-guards Harden caller authorization and precondition guards in admin_override…
…eEvent EmergencyPauseAdminOverrideEvent gains a `previous` field, so an indexer can see which transition an override actually performed rather than only its result. The branch predates #428, which removed the re-entrancy guard from emergency_pause_admin_override, so it reinstated the lock write and published the event a second time. Kept main's single publish and added `previous: current` to it. Its new tests referenced `DataKey::EmergencyPaused` and `DataKey::EmergencyPauseLock`; neither variant exists (the short keys are `Ep` and `EpLk`). The lock assertion also expected `Some(false)` -- after #428 the guard key is never written at all, so it now asserts `None`. 603 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat(#398): structured event for emergency_pause_admin_override outcome
The same author's #432 branch already carried this work -- the ArbitrationSplitAppliedEvent struct, its publish at the end of apply_dispute_arbitration_split, and arbitration_split_event_tests.rs -- so it landed with that merge. The only remaining difference here was a set of stale test snapshots; taking main's regenerated copies leaves the tree identical to main. Merging so the commits are recorded against this PR rather than closing it unmerged. 603 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…350) Four entrypoints now reject unauthorised callers and illegal source states before touching the ledger: * transfer_admin rejects a zero address (InvalidAddress) and refuses to rotate the admin while a multisig proposal is pending (AdminTransferPending). * emergency_unpause rejects NotPaused before taking the transition lock. * admin_resume_escrow rejects NotPaused instead of silently succeeding, so a mistaken call no longer reads as a completed recovery. * multisig_admin_override_release requires MultisigLocked to be active before any JobMeta or milestone read. lib.rs merged cleanly (main's DataKey renames Ep/EpLk applied to the branch's guards). test.rs needed hand-resolution: a three-way apply anchored the four multisig_lock insertions onto unrelated admin_override_release / admin_override_refund tests, which would have rewritten those tests into copies of the multisig ones. Reverted and placed each insertion on its intended test. 613 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
cancel_escrow no longer lets one party freeze the escrow on its own. The first signature records a bit in a CancelApproval mask and emits CancelApprovalRecordedEvent; the CancelLock fires only when the counter-party signs. revoke_cancel_approval lets a party withdraw its signature before the lock. Resolved against main: * The branch deletes the public cancel_escrow_split_refund entrypoint, which is unrelated to this change. Kept it. * It reverts admin_override_tax_release to the two-equality status check that #411 replaced with an exhaustive match rejecting Disputed. Kept ours. * The textual merge spliced main's enriched CancelEscrowInitiatedEvent fields into the branch's new CancelApprovalRecordedEvent (which does not declare them) while leaving the real CancelEscrowInitiatedEvent publish with only two. Swapped them back. * Dropped a duplicate CancelLock read the branch added below the one main already performs. Removing the standalone `admin.require_auth()` from admin_override_tax_release is kept: require_admin already calls require_auth, so it was a duplicated auth requirement. Nineteen existing tests asserted the old one-signature lock -- they were written after the branch diverged, so it could not have adapted them. Updated each to supply the counter-party signature, and reinterpreted the two duplicate-call tests: a repeat from the same party is now InvalidStatus, and EscrowLocked applies once both have signed. revoke_cancel_approval arrived with no tests (the branch carries snapshot files for tests that were never committed). Added six covering the event payload, mask clearing, revoke-then-reapprove, revoking after the lock, revoking without a prior approval, and a non-party caller. 619 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ty-auth feat(cancel_escrow): enforce two-party authentication before lock fir…
…eld/time/streaming Adds TimeExtExecutionLock and PaymentStreamingExecutionLock re-entrancy guards across the settlement entrypoints, a time_extensions_consent dual-signature path, and an interest_yield_split_refund allocation entrypoint. Resolved against main: * Error discriminant 32 is already taken by main's EmptyBalance, so the branch's TimeExtInProgress / PaymentStreamingInProgress moved to 33 and 34. Leaving them at 32/33 would have renumbered a live error code. * The branch reintroduces DataKey::EmergencyPauseLock, which main shortened to EpLk; dropped its copy and kept its two genuinely new lock keys. * Both sides added a helper in the same place -- kept main's assert_nonzero_balance alongside the branch's two lock assertions. * test.rs is purely additive on the branch, so took main's file and reapplied the module declaration and the #255 test block rather than resolving the interleaved hunks. 651 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
feat: locks, split-refund, and dual consent for yield/time/streaming
… fees, shorter time-extension key Closes #306, #308, #315, #322. * emergency_pause now takes the client and freelancer instead of the admin and requires both signatures (#322). The admin's unilateral path remains emergency_pause_admin_override. * validate_fee_allocation enforces MAX_TREASURY_FEE_BPS (2000) and MAX_CLIENT_FEE_BPS (5000) with a new FeeTooHigh error (#306). * split_refund_net_distribution applies the platform fee to the freelancer's payout only, leaving the client refund fee-exempt (#308). * DataKey::MilestoneTimeExtension becomes TimeExt and its value narrows from u64 to u32, matching the EpLk/Ep shortening already in the enum (#315). Resolved against main: * The branch inserts FeeTooHigh at 29 and shifts AlreadyPaused, NotPaused and InvalidAllocationWeights up by one. Those codes are live and main has since added 32-34, so FeeTooHigh took 35 and every existing discriminant kept its value. * Its time-extension read/write reverted main's move to temporary storage and #426's persistent fallback. Kept both, under the new TimeExt key. * split_refund_net_distribution called multisig_split_refund with four arguments; that function takes five and additionally requires the admin key and an active multisig lock. Repointed at cancel_escrow_split_refund, the pure allocator with the same shape. * Moved the cap check below the sum check, so a malformed ratio still reports InvalidRatio rather than FeeTooHigh -- structural validity first, policy bound second. * Updated thirteen emergency_pause call sites the branch could not see. Three of them (from #378) used emergency_pause as the probe for "the admin key is unchanged"; that no longer tests admin authority, so they now probe emergency_unpause, which still distinguishes the rightful admin (NotPaused) from anyone else (Unauthorized). * Two fee tests configured an even-thirds split, now above the treasury cap; moved to 40/40/20, which still floors every share to zero. 654 tests pass; wasm32 release build is clean. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What was done
This update significantly hardens the smart contract’s security model and optimizes its storage footprint on the ledger. We introduced robust dual-signature validation for the emergency pause functionality and reduced state overhead by streamlining the milestone streaming storage representations. Additionally, we enhanced the platform fee logic by correctly exempting client refunds from platform fees during split refunds, and introduced strict new business bounds—enforced by a new FeeTooHigh error—to prevent malicious or invalid fee configurations. All new features are fully verified, bringing the passing test suite up to 328 tests.
Close #306
Close #308
Close #315
Close #322