Support full duplex traffic and half close socket in NAT localhost relay

[Feng Wang (chemwolf6922)]

Summary of the Pull Request

This is a continuation of wangxin12's work in #40171.

The original localhost relay, on both sides, would:

1. Block the other direction's traffic if one direction's write would block.
2. Close both sides if one side's read is closed. Which causes problems in half close socket scenarios.

This PR makes these changes to support full duplex traffic and half close socket in NAT localhost relay:

1. Refactor the windows bi-directional relay with MultiHandleWait with a new half close capable relay handle.
2. Refactor the Linux relay with a fully event driven non-blocking epoll loop. Which also avoids creating a new thread for each connection.

PR Checklist

• Closes: Blocking I/O behavior in localhost relay hangs with full duplex traffic due to half duplex forwarding logic. #10688 NAT localhost forwarding: wslrelay.exe --mode 1 deadlocks under concurrent bidirectional bulk transfer #41680
• Communication: I've discussed this with core contributors already. If work hasn't been agreed, this work might be rejected
• Tests: Added/updated if needed and all pass
• Localization: All end user facing strings can be localized
• Dev docs: Added/updated if needed
• Documentation updated: If checked, please file a pull request on our docs repo and link it here: #xxx

Detailed Description of the Pull Request / Additional comments

Validation Steps Performed

Add tests:
NetworkTests::NetworkTests::NatLocalhostRelayHalfClose
NetworkTests::NetworkTests::NatLocalhostRelayFullDuplex
UnitTests::UnitTests::BidirectionalRelayContinuesAfterHalfClose

[Copilot]

Pull request overview

This PR improves the NAT localhost relay implementation to correctly support full-duplex traffic and TCP half-close semantics, addressing hangs and premature connection teardown in bidirectional protocols.

Changes:

• Refactors the Windows bidirectional relay to use MultiHandleWait, introducing HalfCloseRelayHandle to propagate half-close (shutdown(SD_SEND)) without terminating the opposite direction.
• Rewrites the Linux localhost relay into a non-blocking, event-driven epoll loop to support full duplex and avoid per-connection threads.
• Adds Windows unit/integration tests covering half-close and full-duplex behavior through the localhost relay.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
test/windows/UnitTests.cpp	Adds a unit test validating relay continues after half-close.
test/windows/NetworkTests.cpp	Adds WSL2 NAT localhost relay tests for half-close and full-duplex behavior; extends socat helper.
src/windows/common/relay.cpp	Replaces legacy WaitForMultipleObjects relay loop with MultiHandleWait + relay handle abstractions.
src/windows/common/HandleIO.h	Extends ReadHandle to accept a buffer size; adds HalfCloseRelayHandle declaration.
src/windows/common/HandleIO.cpp	Implements buffer-sized ReadHandle construction and HalfCloseRelayHandle shutdown behavior.
src/linux/init/localhost.cpp	Replaces blocking/threaded relay with per-connection state machine driven by epoll on non-blocking sockets.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

Suppressed comments (1)

Previously missed (1) — in code that hasn't changed since the last review.

src/linux/init/localhost.cpp:521

• connect() can return EINTR when a signal interrupts the attempt; for this case Linux leaves the connection attempt in progress, but this branch currently tears down the relay connection. The previous blocking implementation retried interrupted connects, and the new accept/read/write paths also handle interruptions. Treat EINTR like EINPROGRESS and let epoll complete the connection instead.

        const int result = connect(m_tcpSocket.get(), socketAddress, socketAddressSize);
        if (result < 0 && errno != EINPROGRESS)

[Copilot]

Copilot review overview

🟡 Changes recommended

The full-duplex test must verify that queued forward traffic resumes and drains.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Verify queued data resumes and drains after unblocking the guest · New

[Copilot]

Copilot review overview

🟡 Changes recommended

The critical relay starvation and interrupted Linux connection handling issues must be resolved.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 1

Open (1)

• Prevent half-duplex starvation in MultiHandleWait scheduler · New

Resolved since last review (1)

• Verify queued data resumes and drains after unblocking the guest

Previously missed (1)

In code that hasn't changed since last review

Handle EINTR and EALREADY during nonblocking connect

src/​linux/​init/​localhost.cpp:522

The old blocking connect used TEMP_FAILURE_RETRY, but this nonblocking call now treats EINTR (and the EALREADY that can result when an interrupted connect is retried) as a terminal connection failure. A signal during setup will therefore drop an otherwise viable relay connection instead of waiting for it to complete; retry the call on EINTR and accept EALREADY as an in-progress state.