Commit Graph

2222 Commits

Author SHA1 Message Date
durch 4317ad3031 Various fixes 2025-11-24 12:13:17 +01:00
Drazen Urch ecdeeb096e KKT + PSQ (#6203)
* Add nymkkt with KKT convenience wrappers for nym-lp integration

Integrates nymkkt module from georgio/noise-psq branch to enable
post-quantum key distribution for nym-lp.

Changes:
- Add common/nymkkt from georgio/noise-psq (KKT protocol implementation)
- Add convenience wrapper layer (kkt.rs) with simplified API:
  - request_kem_key() - Client requests gateway's KEM key
  - validate_kem_response() - Client validates signed response
  - handle_kem_request() - Gateway handles requests
- Add nymkkt to workspace members in root Cargo.toml
- Export kkt module in lib.rs

The KKT (Key Encapsulation Mechanism Transport) protocol enables efficient
distribution of post-quantum KEM public keys. Instead of storing large PQ
keys in the directory (1KB-500KB), we store 32-byte hashes and fetch actual
keys on-demand via this authenticated protocol.

Tests: All 5 unit tests passing (authenticated, anonymous, signature
verification, hash validation)

* feat(lp): add Ed25519 authentication to PSQ protocol

Replace basic PSQ v0 API with authenticated v1 API that includes
cryptographic authentication via Ed25519 signatures.

Changes:
- PSQ initiator now signs encapsulated keys with Ed25519 private key
- PSQ responder verifies Ed25519 signatures before deriving PSK
- Prevents MITM attacks through mutual authentication
- Fixed test helpers to use role-based Ed25519 keypair assignment
  (initiator uses [1u8;32], responder uses [2u8;32])

Security: This adds a critical authentication layer to the post-quantum
PSK derivation protocol, ensuring both parties can verify each other's
identity during the handshake.

Tests: All 77 tests passing (was 11 failures, now 0)

* feat(lp): integrate PSQ post-quantum PSK derivation

Complete integration of Post-Quantum Secure (PSQ) protocol for PSK
derivation in the Lewes Protocol, replacing simple Blake3 derivation
with cryptographically secure DHKEM-based PSK establishment.

This commit encompasses three completed tasks:

- Add KKTRequest/KKTResponse message types to LpMessage enum
- Update codec to handle KKT message serialization/deserialization
- Add kkt_orchestrator.rs with high-level KKT API wrappers
- Enable key exchange orchestration for PSQ protocol

- Add set_psk() method to NoiseProtocol for dynamic PSK injection
- Integrate PSQ derivation into LpSession handshake flow
- PSQ payload embedded in first Noise message (ClientHello)
- Derive PSK using libcrux-psq before Noise handshake completion
- Add helper functions for X25519 to KEM conversions

- Add comprehensive PSQ integration tests in session_integration/
- Test PSQ handshake end-to-end flow
- Validate PSK derivation correctness between initiator/responder
- Test PSQ + Noise combined protocol operation

Dependencies:
- libcrux-psq: Post-quantum PSK protocol implementation
- libcrux-kem: Key Encapsulation Mechanism primitives
- nym-kkt: KKT key exchange protocol wrappers
- rand 0.9: Required for KKT compatibility

Security: This adds Harvest-Now-Decrypt-Later (HNDL) resistance by
combining classical ECDH with post-quantum KEM for PSK derivation.
Even if X25519 is broken by quantum computers, the PSK remains secure.

Tests: All 77 tests passing

* feat(lp): add PSQ error handling documentation and tests (nym-bbi)

Formalize the "always abort" error handling strategy for PSQ failures.
PSQ errors indicate attacks, misconfigurations, or protocol violations
that should not be silently ignored or worked around.

Changes:
- Add comprehensive error handling documentation to psk.rs module
- Add diagnostic logging with error categorization:
  * CredError → warn about potential attack
  * TimestampElapsed → warn about potential replay
  * Other errors → log as errors
- Add 4 error scenario tests:
  * test_psq_deserialization_failure
  * test_handshake_abort_on_psq_failure
  * test_psq_invalid_signature
  * test_psq_state_unchanged_on_error
- Add log dependency to Cargo.toml

Error handling strategy: All PSQ failures abort the handshake cleanly
with no retry or fallback. This prevents silent security degradation
and ensures misconfigurations are detected early.

State guarantees: PSQ errors leave session in clean state - dummy PSK
remains, Noise HandshakeState unchanged, no partial data, no cleanup needed.

Tests: 81 tests passing (77 original + 4 new error tests)

Closes: nym-bbi

* feat(lp): add PSK injection tracking to prevent dummy PSK usage (nym-ep2)

Add safety mechanism to ensure real post-quantum PSK was injected before
allowing transport mode operations (encrypt/decrypt). This prevents
accidentally using the insecure dummy PSK [0u8; 32] if PSQ injection fails.

Changes:
- Add `psk_injected: AtomicBool` field to LpSession
- Initialize to `false` in LpSession::new()
- Set to `true` after successful PSK injection:
  * Initiator: In prepare_handshake_message() after set_psk()
  * Responder: In process_handshake_message() after set_psk()
- Add NoiseError::PskNotInjected error variant
- Add PSK injection checks in encrypt_data() and decrypt_data()
  * Check happens before handshake completion check
  * Returns PskNotInjected if flag is false
- Add comprehensive PSK injection lifecycle documentation to LpSession
- Add test_transport_fails_without_psk_injection test
- Update test_encrypt_decrypt_before_handshake to expect PskNotInjected

PSK Injection Lifecycle:
1. Session created with dummy PSK [0u8; 32] in Noise HandshakeState
2. During handshake, PSQ runs and derives real post-quantum PSK
3. Real PSK injected via set_psk() - psk_injected flag set to true
4. Handshake completes, transport mode available
5. Transport operations check psk_injected flag for safety

This is defensive programming - normal PSQ flow always injects the real PSK.
The safety check prevents transport mode if PSQ somehow fails silently or is
bypassed due to implementation bugs.

Tests: 82 tests passing (81 original + 1 new)

Closes: nym-ep2

* docs(lp): fix PSK state documentation inaccuracy

Correct error handling documentation to clarify that PSK slot 3
remains unmodified only on error, not in all cases.

Previous: "PSK slot 3 = dummy [0u8; 32] (never modified)"
Corrected: "PSK slot 3 = dummy [0u8; 32] (not modified on error)"

This is more accurate since:
- On error: PSK remains as dummy value (never injected)
- On success: PSK is replaced with real post-quantum PSK

Documentation-only change, no functional impact.

* feat(lp): add KKTExchange state to state machine for pre-handshake KEM key transfer (nym-4za)

Add KKTExchange state to LpStateMachine to properly orchestrate KKT (KEM Key Transfer)
protocol before Noise handshake begins. This enables dynamic KEM public key exchange,
allowing post-quantum KEM algorithms to be used without pre-published keys.

Changes:
- Add KKTExchange state and KKTComplete action to state machine
- Implement automatic KKT exchange on StartHandshake:
  * Initiator: sends KKT request → waits for response → validates signature
  * Responder: waits for request → validates → sends signed KEM key
- Update process_kkt_response() to accept Option<&[u8]> for hash validation:
  * Some(hash): full KKT validation with directory hash (future)
  * None: signature-only mode (current deployment)
- Add local_x25519_public() helper for responder KEM key derivation
- Update state flow: ReadyToHandshake → KKTExchange → Handshaking → Transport
- Add PSK handle storage (psk_handle) for future re-registration
- Export generate_fresh_salt() for session creation
- Update psq_responder_process_message() to return encrypted PSK handle (ctxt_B)
- Add comprehensive tests:
  * test_kkt_exchange_initiator_flow
  * test_kkt_exchange_responder_flow
  * test_kkt_exchange_full_roundtrip
  * test_kkt_exchange_close
  * test_kkt_exchange_rejects_invalid_inputs
  * Updated test_state_machine_simplified_flow for KKT phase

All tests passing. Ready for nym-8y5 (PSQ handshake KKT integration).

* docs(lp): add state machine and post-quantum security protocol documentation

Add comprehensive documentation of the Lewes Protocol state machine and
post-quantum security architecture to LP_PROTOCOL.md.

New sections:
- State Machine and Security Protocol overview
- Detailed state transition diagram (ReadyToHandshake → KKTExchange → Handshaking → Transport)
- Complete message sequence diagram showing KKT + PSQ + Noise flow
- KKT (KEM Key Transfer) protocol specification
- PSQ (Post-Quantum Secure PSK) protocol details
- Security guarantees and implementation status
- Algorithm choices (current X25519, future ML-KEM-768)
- Message type specifications for KKT
- Version 1.1 changelog entry documenting KKT/PSQ integration

Documentation includes:
- ASCII art state machine diagram
- Message sequence diagram with all protocol phases
- PSK derivation formulas
- Security properties checklist
- Migration path to post-quantum KEMs
- Integration details (PSQ embedded in Noise, no extra round-trips)

Related to nym-4za (KKTExchange state implementation).

* feat(lp): use KKT-authenticated KEM key in PSQ handshake (nym-8y5)

Replace direct X25519→KEM conversion with KKT-derived authenticated key
in PSQ initiator flow. This ensures PSQ uses the responder's authenticated
KEM public key obtained via KKT protocol instead of blindly converting
their X25519 key, properly completing the post-quantum security chain.

Changes:
- session.rs: Extract KEM key from KKTState::Completed in prepare_handshake_message()
- session.rs: Add set_kkt_completed_for_test() helper for test initialization
- session.rs: Update create_handshake_test_session() to initialize KKT state
- session.rs: Fix test_handshake_abort_on_psq_failure and test_psq_invalid_signature
- session_manager.rs: Add init_kkt_for_test() for integration test setup
- session_integration/mod.rs: Update tests for KKT-first flow (6 rounds total)
- session_integration/mod.rs: Fix state machine test expectations for KKTExchange state

All 87 tests passing. Unblocks nym-w8f (KKT tests) and nym-m15 (production integration).

* feat(lp): simplify API to Ed25519-only, derive X25519 internally

Refactored LP state machine to use Ed25519 keys exclusively in the public
API, with X25519 keys derived internally via RFC 7748. This simplifies the
API from 6 parameters to 4 while maintaining protocol security.

**Core API Changes:**
- LpStateMachine::new(): Removed explicit X25519 keypair parameters
- Old: new(is_initiator, local_keypair, local_ed25519_keypair,
         remote_public_key, remote_ed25519_key, salt)
- New: new(is_initiator, local_ed25519_keypair, remote_ed25519_key, salt)
- X25519 keys now derived internally from Ed25519 using RFC 7748
- lp_id calculation moved inside state machine (uses derived X25519 keys)

**Protocol Changes:**
- ClientHello message extended from 65 to 97 bytes
- Now includes client_ed25519_public_key field (32 bytes)
- Required for PSQ authentication in KKT + PSQ handshake flow
- Breaking change: gateway must extract Ed25519 from ClientHello

**Gateway Updates:**
- receive_client_hello() now extracts Ed25519 public key
- LpGatewayHandshake::new_responder() accepts Ed25519 keys only
- Removed manual X25519 conversion (handled by state machine)

**Registration Client Updates:**
- LpRegistrationClient now uses Ed25519 keypairs
- Generate fresh ephemeral Ed25519 keys for LP registration
- ClientHello includes Ed25519 public key for gateway authentication
- Fixed 7 pre-existing build errors:
  * mixnet_client_startup_timeout field removal
  * IprClientConnect API change (async → sync)
  * Error variant renames (use helper function)
  * LP client key type mismatches (X25519 → Ed25519)

**Test Suite:**
- Updated 16+ test functions to use new 4-parameter constructor
- Fixed 5 integration test failures caused by lp_id mismatch
- Tests now derive X25519 from Ed25519 (matching production behavior)
- Added missing PublicKey imports in test modules
- All 87 tests passing (100% success rate)

**Implementation Details:**
- Added Ed25519RecoveryError variant to LpError enum
- Type conversion: nym_crypto X25519 → nym_lp keypair types
- Maintained backward compatibility for PSQ/KKT protocol flow
- Session manager updated to use new API signature

This change completes the Ed25519-only API migration, hiding X25519 as an
implementation detail while preserving all security properties of the
KKT-authenticated PSQ handshake protocol.

* chore: run cargo fmt

* chore: run cargo clippy --fix to resolve simple linter issues

* Basic handshake working

* Final tweaks

* Wrap PR comments, 2024

---------

Co-authored-by: Jędrzej Stuczyński <jedrzej.stuczynski@gmail.com>
2025-11-21 18:37:38 +01:00
durch 6d0e4f65f2 Simplify, clean up 2025-11-21 18:25:47 +01:00
durch 1f6daa7fd3 Bits and bobs to make everything work 2025-11-21 18:25:47 +01:00
durch fbcc9e4782 lp-reg gw flow working-ish 2025-11-21 18:25:47 +01:00
durch 10405c7dc1 more metrics 2025-11-21 18:18:30 +01:00
durch de06f4a5c0 fmt and metrics 2025-11-21 18:17:32 +01:00
durch ec90a218df Cleanup 2025-11-21 18:16:34 +01:00
durch 5f2122688f KDF and tests 2025-11-21 18:16:34 +01:00
durch dd6b7b6a34 Remove notes 2025-11-21 13:49:22 +01:00
durch cae63877a4 Client bits 2025-11-21 13:49:22 +01:00
durch 542e56044a Gateway side things 2025-11-21 13:40:50 +01:00
Jędrzej Stuczyński 440aadf124 chore: updated default endpoint for retrieving attestation.json (#6207) 2025-11-14 15:04:51 +00:00
Jędrzej Stuczyński d126d8e5a0 feat: upgrade mode: VPN adjustments (#6189)
* placeholder handling of wg registration with upgrade mode token

* include upgrade mode credentials as part of credential storage

* introduce helper for decoding JWT payload

* expose methods for removing emergency credentials from the storage

* don't allow duplicate emergency credentials with the same content

* added authenticator ClientMessage for upgrade mode check

* retrieve credentials with longest expiration first

* post rebasing fixes

* fixed gateway config

* feat: allow specifying minimum node performance for client init

* nym-node UM improvements

* fixed upgrade mode bandwidth on initial authentication

* fix: logs and thresholds

* expose attestation information from nym-node http api

* additional logs

* post rebasing fixes

* make @simonwicky happy by removing empty lines in emergency_credential table definition

* chore: remove '_' prefix for internal counters within in-mem ecash storage

* improved import of 'UpgradeModeState' within the nym-node

* use explicit time dependency within credential-storage

* re-order imports within the gateway-client

* moved 'AvailableBandwidth' definition to the monorepo
2025-11-14 13:34:36 +00:00
Jędrzej Stuczyński 6b2bb3029b feat: merge intermediate upgrade mode changes (#6174)
* squashing feat: merge intermediate upgrade mode changes #6174 to more easily resolve merge conflicts during rebasing

added additional v2 query for metadata endpoint for requesting upgrade mode recheck

added additional message to v6 authenticator to request explicit upgrade mode recheck

clippy

test fixes due to updated keys

updated assertion for upgrading v1 top up request to v2

compare attester public key against the expected value within the credential proxy

use pre-generated attestation public keys within nym-nodes

remove version deprecation

bugfix: default bandwidth response for authenticator

expose upgrade mode information in authenticator responses

adding tests for new v2 server

passing upgrade mode information in metadata endpoint

v2 wireguard private metadata

bugfix: make sure to immediately poll for attestation after spawning task

fix gateway probe and remove code duplication for finalizing registration

squashing before rebasing

post rebasing fixes

AuthenticatorVersion helpers

additional nits

allow unwraps in mocks

fixed linux build

clippy

integrating upgrade mode into authenticator

fixed build after adding wrappers to response types

conditionally updating peer handle bandwidth

cleanup

negotiate initial protocol during registration

change auth to use highest protocol

handler for JWT message

dont meter client bandwidth in upgrade mode

handling recheck requests

sending information about upgrade_mode on client messages

gateway watching for upgrade mode attestation

wip: gateways to disable bandwidth metering on upgrade mode

* fixed ServerResponse deserialisation

* fixed incorrect swagger path for upgrade mode check endpoint

* moved upgrade mode endpoint out of bandwidth routes

* chore: remove unused error variant

* removed re-export of UpgradeModeAttestation from credentials-interface

* chore: define single source of truth for minimum bandwidth threshold value

* moved type definitions out of traits.rs

* updated v6 versioning to point to niolo release instead

* fixed incorrect error mapping
2025-11-14 13:13:15 +00:00
Jędrzej Stuczyński 350d244032 bugfix: fix credential proxy upgrade mode attestation url arg (#6202)
this includes bringing over changes introduced in #6174
2025-11-14 08:19:21 +00:00
Jack Wampler 17ca000782 HTTP API resilience enable & domain rotation conditions (#6200)
* http url fallback conditions

* include changes and tests for fronted

* Allow for explicit DNS error Handling in HTTP client  (#6201)

when sending http reqs add manual DNS so we can handle errors directly

* Address PR nits

---------

Co-authored-by: durch <durch@users.noreply.github.com>
2025-11-14 08:59:36 +01:00
jmwample 10951d4cd3 clippy, fmt, minor fix 2025-11-11 10:40:25 -07:00
jmwample 5acce42c64 add some staic hosts and switch server strategy 2025-11-11 09:14:26 -07:00
jmwample 2f752a6c42 fix things related to interface changes 2025-11-06 18:37:50 -07:00
Jack Wampler 806f807f02 Implement Static DNS fallback (#6178) 2025-11-06 16:46:39 -07:00
Jack Wampler 1400db6156 DNS Reliability Fixes (#6175) 2025-11-06 12:37:27 -07:00
Jędrzej Stuczyński d9c2f6ebda Feature/credential proxy jwt (#5957)
* squashed feature/credential-proxy-jwt [#5957]

post rebasing fixes

clippy

changed obtain-async endpoint to conditionally return jwt instead of pending zk-nym

watching for the attestation file and issuing jwt

* changed attestation starting time serialisation into rfc3339

* including authorised JWT issuers in attestation

* reduce attestation retrieval error log
2025-11-03 16:42:39 +00:00
Jędrzej Stuczyński f5a4dbc555 removed useless use of vec! 2025-10-31 11:42:42 +00:00
Jędrzej Stuczyński 4480534e4d derived Default impl for EpochState 2025-10-31 11:39:58 +00:00
Simon Wicky fd5a95fa4d allow overwriting existing sdk shutdown manager 2025-10-24 16:17:29 +02:00
Simon Wicky c61df79182 typo 2025-10-24 14:11:56 +02:00
Simon Wicky 08559a7660 calling for shutdown from the MixTrafficController 2025-10-24 14:07:15 +02:00
Jędrzej Stuczyński 6dce55a99b using same hierarchy of trackers for client shutdown control 2025-10-24 14:03:18 +02:00
Tommy Verrall bc0b89b31c Internal comments 2025-10-24 12:44:10 +02:00
Tommy Verrall 67c32faa11 Fix comments 2025-10-23 19:22:26 +02:00
Tommy Verrall aa0d15ee67 Better message to come in the PR description 2025-10-23 19:06:27 +02:00
Tommy Verrall e6f4bae895 Last failing test - fix 2025-10-21 19:34:20 +02:00
Tommy Verrall d71742af32 Use explicit Vec<ApiUrl> handling in BaseClientBuilder
- Replace NymNetworkDetails with explicit API URL handling
- Fix deprecated from_network() usage and improve fallback logic
- Add URL validation and remove unused backwards compatibility
2025-10-21 19:15:24 +02:00
Tommy Verrall 3b7c07e249 Actually commit the recommended changes 2025-10-21 18:12:38 +02:00
Tommy Verrall 3b429dde69 Fix broken tests in CI 2025-10-21 16:29:26 +02:00
Tommy Verrall 3a29c296da Replace deprecated from_network() with new_with_fronted_urls() 2025-10-21 16:05:41 +02:00
Tommy Verrall 8544c54f8f Merge develop into feature/domain-fronting-v2
- Use new_with_fronted_urls() for explicit domain fronting
- Deprecate from_network() in favor of explicit method
2025-10-21 15:58:20 +02:00
Jędrzej Stuczyński 9f9639950b feat: expose more explicit new_with_fronted_urls builder for http API client (#6136) 2025-10-21 14:47:58 +01:00
Tommy Verrall 67b300d0b8 Fix new_from_env() to populate nym_api_urls for domain fronting 2025-10-21 12:22:51 +02:00
Tommy Verrall f6800aff0a fix all clippy messages 2025-10-21 11:32:47 +02:00
Tommy Verrall 0c7c927ca2 Add more tests for retry logic 2025-10-21 11:32:47 +02:00
Tommy Verrall a69c8b1660 Fix confusing tracing logs 2025-10-21 11:32:47 +02:00
Tommy Verrall f3ea310a46 Fix retries - this is working 2025-10-21 11:32:46 +02:00
Tommy Verrall 92f9ff035d Add configuration-based domain fronting support
Changes:
- Add network_details field to BaseClientBuilder (optional, backwards compatible)
- Add with_network_details() method for opt-in domain fronting
- Update construct_nym_api_client() to use from_network() when network_details provided
- Enable network-defaults feature in nym-client-core Cargo.toml
- SDK passes network_details to BaseClientBuilder
2025-10-21 11:32:46 +02:00
Tommy Verrall 35ea7e4926 - Add DEFAULT_NYM_API_RETRIES constant (replaces magic number 3)
- Run cargo fmt on all affected packages
- All clippy warnings resolved
2025-10-20 16:51:07 +02:00
Tommy Verrall d1cb9afaf0 not sure what happened but it's fixed 2025-10-20 15:20:24 +02:00
Tommy Verrall 79d4b4b2e3 Merge branch 'develop' into multiple-fall-back-urls 2025-10-20 15:16:36 +02:00
Tommy Verrall ae6539e07c Merge resolution 2025-10-20 15:14:48 +02:00
Tommy Verrall 6f4dfd1dab fix conversion type && make the retry count configurable 2025-10-20 11:15:31 +02:00