When two EasyTier instances run on the same machine and share the same
network, the direct connector would expand a remote peer's 0.0.0.0
listener into local interface IPs and then attempt to connect to
itself, causing an infinite loop of failed connection attempts.
The existing `peer_id != my_peer_id` guard does not cover this case
because the two instances have different peer IDs despite sharing the
same physical network interfaces.
Fix by adding a self-connection check in `spawn_direct_connect_task`:
before spawning a connect task, compare the candidate (scheme, IP,
port) against the local running listeners. If a local listener matches
on all three dimensions — accounting for 0.0.0.0/:: wildcards by
checking membership in the local interface IP sets — the candidate is
silently dropped with a DEBUG log message.
The fix covers all four code paths:
- IPv4 unspecified (0.0.0.0) expansion loop
- IPv4 specific-address branch
- IPv6 unspecified (::) expansion loop
- IPv6 specific-address branch
The TESTING flag logic is untouched so existing unit tests are
unaffected.
* refactor(connector): replace is_self_connect closure with GlobalCtx::should_deny_proxy (#1954)
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
* feat: separate faketcp into a feature
* fix: no need to initialize out_len
* feat: separate zstd into a feature
* clippy: remove unnecessary cast, because for unix size_t always equals usize
This allows HTTP redirect servers to provide network-specific node
lists based on the client's network identity. Updated unit tests
to verify the header is correctly sent.
Use noise protocol on handshake. Check peer's public key if needed. Also support rekey and replay attack prevention.
E2EE and temporary password will be implemented based on this.
support faketcp to avoid tcp-over-tcp problem.
linux/macos/windows are supported.
better to be used in internet env, the maximum
performance is majorly limited by windivert/raw socket.
This change introduces a major refactoring of the RPC service layer to improve modularity, unify the API, and simplify the overall architecture.
Key changes:
- Replaced per-network-instance RPC services with a single global RPC server, reducing resource usage and simplifying management.
- All clients (CLI, Web UI, etc.) now interact with EasyTier core through a unified RPC entrypoint, enabling consistent authentication and control.
- RPC implementation logic has been moved to `easytier/src/rpc_service/` and organized by functionality (e.g., `instance_manage.rs`, `peer_manage.rs`, `config.rs`) for better maintainability.
- Standardized Protobuf API definitions under `easytier/src/proto/` with an `api_` prefix (e.g., `cli.proto` → `api_instance.proto`) to provide a consistent interface.
- CLI commands now require explicit `--instance-id` or `--instance-name` when multiple network instances are running; the parameter is optional when only one instance exists.
BREAKING CHANGE:
RPC portal configuration (`rpc_portal` and `rpc_portal_whitelist`) has been removed from per-instance configs and the Web UI. The RPC listen address must now be specified globally via the `--rpc-portal` command-line flag or the `ET_RPC_PORTAL` environment variable, as there is only one RPC service for the entire application.
1. avoid dns query hangs the thread
2. avoid deadloop when stun query failed because of no ipv4 addr.
3. make quic input error non-fatal.
4. remove ring tunnel from connection map to avoid mem leak.
5. limit listener retry count.
Large memory usage comes from:
Mimalloc hold large thread cache, causing abort 13M+ usage.
QUIC endpoint occupy 3M when GRO is enabled.
Smoltcp 64 tcp listener use 2MB.
Some devices have ipv6 but don't allow input connection, this patch add hole punching for these devices.
- **add v6 hole punch msg to udp tunnel**
- **send hole punch packet when do ipv6 direct connect**
1. dns resolver should be global unique so dns cache can work. avoid dns query influence hole punching.
2. when system dns failed, fallback to hickory dns.
* optimize memory issues
1. introduce jemalloc support, which can dump current memory usage
2. reduce the GlobalEvent broadcaster memory usage.
3. reduce tcp & udp tunnel memory usage
TODO: if peer conn tunnel hangs, the unbounded channel of peer rpc
may consume lots of memory, which should be improved.
* select a port from 15888+ when port is 0
* support ipv6 stun
* show interface and public ip in cli node info
* direct conn should keep trying unless already direct connected
* peer should use conn with smallest latency
* deprecate ipv6_listener, use -l instead
fanyang