- (neequ57) Merged !330 - geolocation feature (off by default) to allow excluding/denylisting route nodes based on geography
- (evelyn) Merged !267 - adding the ability to create dht records with a specified owner key (rust only currently)
- (rivka segan) Merged !335 - fix logic error that used wss when not tls
- New startup/shutdown initialization and component system
- Logs are tagged with the program_name+namespace they are collected in
- No more per-facility log macros, one unified `veilid_log!()` macro for all events
- Switch between subnodes in veilid-cli switches which logs you're viewing
- All global logs and subnode 0 go to console, all other subnodes are accessible via veilid-cli
- Major refactor to add VeilidComponentRegistry as the 'owner' of all components
- Access to components now uses scoped guards for lifetime management rather than loose Arc clones
- VeilidComponent trait makes adding common per-component functions easier
- Unified initialize, post-initialize, pre-terminate, and terminate phase harness
- AsyncCryptoSystemGuard added to make heavy operations happier in async environments
- UDP hole punch needed TTL setting to keep routers from incorrectly making conntracks
- Public address detection was getting stuck in a lock contention, regression from refactor
- PeerInfo caching to eliminate some repeated cloning
- NodeContactMethod cache improvements
- Symmetric NAT and NetworkClass::OutboundOnly were broken. When routing domain address types are known, but there is no dialinfo, that should be OutboundOnly and not Invalid. It's valid to have no dialinfo. Added network class 'confirmation'.
veilid-tools:
- replaced deprecated serde_yaml crate with maintained serde_yaml_ng crate
- Start of VirtualRouter network virtualization
- standalone virtual router binary in veilid-tools (`cargo run --bin virtual_router`)
- IAC-style configuration system for repeatable virtualized network generation
- Add a publish() as well as a commit() for routing domain editor
- Should only publish our peer info after we're sure we done editing it (end of public address detection task)
- Publish should happen after relay selection as well
- Publish should happen if the relay's peerinfo has changed
- Publish should not do anything if the peerinfo hasn't changed
- PeerInfo -> Arc<PeerInfo> everywhere to minimize deep clones and ensure read-only PeerInfo
- Routing domain editing is now more atomic
- When a node selects a relay it now immediately protects its connections.
- Made dial info port (for port restricted nat) more resilient to changes, in the case there are multiple mappings
- Relays that drop protected connections should be deprioritized for relay selection (table saturation detection)
- clear_network_callback in do_public_dial_info_check is a kludge, removed
- Raised the bar for dialinfo changes when its just the port
- Pinging node on the same network works again
- resolve_node() never returns a dead node even when we want to try to communicate with it again
- Removed 'bad public address' detection as it wasn't working anyway
- Added separate parallelism lanes for relay keepalive pings from peer liveness check pings, as they are higher priority
- Change send_data to always check cache for contact method first instead of going with filtered active flows first, avoids choosing UDP when a preferable TCP connection could be made
- Nodes that are not relay capable should drop relayed packets
- DHT issues:
- Make setvalue more likely to succeed by accepting a getvalue consensus if a full setvalue consensus is not reached.
- Offline subkey writes are cleared too fast and should be thought as 'subkeys not yet synchronized'
- If set_value is partial / in-flight, it should still be in offline_subkey_writes
- Make inflight_subkey_writes list and probably some bit for 'written_while_inflight' so we dont clear the offline_subkey_writes until they're really written
- Watching values incorrectly categorized 'owner' keys as anonymous watchers
- Setting a dht value with the same sequence number as what is on the network, but with a conflicting value, did not result in the current value being returned to the api caller as it should have been
- DHT cleanup
- Proper application of DHT capabilities
- Fanout debugging log target
- Performance measurement / timing of veilid_api log target
- Fix DHT Rust integration test
- ValueChanged Optional
- Allow value changed data to be optional in rpc schema
- Make valuechanged update no longer happen when value hasn't changed or is older
- on `RoutingContext`: `with_privacy()` renamed to `with_default_safety()`
- on `RoutingContext`: `with_custom_privacy()` renamed to `with_safety()`
- on `RoutingContext`: `safety()` method added that returns the current `SafetySelection`
- Routing contexts are now safety-route-enabled by default. To disable, use `with_safety()` with `SafetySelection::Unsafe`.
- WASM now works better with updated connection manager code
- Async-std flavor of veilid-core now builds correctly again
- Safety route allocation is bidirectional
- Connection table LRU cache now has protection for relays and in-use RPC question/answers
- Dead route notifications are now sent only for manually allocated routes
- Allocated routes that fail tests now have their nodes marked as 'failure to send' so they go 'unreliable' and get re-tested. Also the same route will not immediately be reallocated as a result.
- DHT tests ported from Python to Rust
- Rustls updated to latest release
- Protected connections (such as relays) that drop result in marking the node as 'failure to send' so a different relay gets chosen
* DESCRIPTION: Decompression was occurring in an unbounded way upon envelope receipt.
* IMPACT: Node crashes resulting in downtime. There was no risk of RCE or compromise due to Rust's memory protections and no use of unsafe code near the site of the error.
* INDICATIONS: This resulted in an out-of-memory abort on nodes. Issue first identified on the bootstrap servers.
* REMEDIATION: Length check added to decompression on envelopes.
- Earthfile support for generating a debug executable
