Fix relay loops spinning at 100% CPU on synchronous EOF by benhillis · Pull Request #40725 · microsoft/WSL

benhillis · 2026-06-05T22:07:28Z

Summary

Fixes wsl.exe relay threads spinning at 100% CPU after the peer (VM-side hvsocket) closes gracefully.

Problem

When ReadFile completes synchronously with 0 bytes (the canonical EOF signal on stream sockets and pipes), three relay code paths failed to recognize this as EOF:

BidirectionalRelay — used by wslrelay.exe for localhost forwarding and socket relay
ScopedMultiRelay::Run — used for set-version stderr relay

Instead of breaking out of the loop, they re-issued ReadFile on the EOF'd handle, which returned TRUE + 0 bytes again immediately, producing a pure CPU-bound spin.

Fix

Add bytesRead == 0 checks on the synchronous-success path in:

BidirectionalRelay (both left and right read sites)
ScopedMultiRelay::Run (both the sync-completion and overlapped-completion paths)

Validation

Added RelayEofDetection unit test that creates pipe pairs, closes the write end (producing synchronous EOF), and verifies each relay function terminates within 5 seconds rather than spinning.

Fixes #40651

Copilot

Pull request overview

This PR addresses a high-CPU spin in WSL relay loops by treating synchronous ReadFile success with bytesRead == 0 as EOF, ensuring relay threads terminate instead of repeatedly re-issuing reads after a graceful peer close.

Changes:

Add explicit bytesRead == 0 EOF handling on synchronous-success ReadFile paths in BidirectionalRelay.
Add explicit zero-byte EOF handling in ScopedMultiRelay::Run for both synchronous and overlapped-completion paths.
Add a new RelayEofDetection unit test to validate relay termination on EOF scenarios.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
`src/windows/common/relay.cpp`	Adds zero-byte read detection to prevent CPU-bound spinning on synchronous EOF in relay loops.
`test/windows/UnitTests.cpp`	Introduces a unit test intended to ensure relay functions terminate promptly on EOF.

OneBlue

LGTM. I think at some point we should move those to use RelayHandle and factor all the IO scheduling in one place, but for now let's merge this

benhillis · 2026-06-05T23:41:35Z

LGTM. I think at some point we should move those to use RelayHandle and factor all the IO scheduling in one place, but for now let's merge this

yep I considered doing that as part of this change, but we can do it as a follow-up.

Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

+            // Wait up to 5 seconds for the relay to finish. If it doesn't, the EOF check is broken.
+            auto threadHandle = wil::unique_handle(OpenThread(SYNCHRONIZE, FALSE, GetThreadId(relayThread.native_handle())));
+            VERIFY_ARE_NOT_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_TIMEOUT);
+            relayThread.join();


+            readHandle.reset(CreateNamedPipeW(
+                pipeName.c_str(), PIPE_ACCESS_INBOUND | FILE_FLAG_OVERLAPPED, PIPE_TYPE_BYTE | PIPE_WAIT, 1, 4096, 4096, 0, &sa));
+            VERIFY_IS_NOT_NULL(readHandle.get());
+


+            writeHandle.reset(
+                CreateFileW(pipeName.c_str(), GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, nullptr));
+            VERIFY_IS_NOT_NULL(writeHandle.get());


+            VERIFY_ARE_NOT_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_TIMEOUT);
+            relayThread.join();


Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 7 comments.

+        // Helper: create an overlapped pipe pair (named pipe server + client).
+        auto createOverlappedPipe = [](wil::unique_handle& readHandle, wil::unique_handle& writeHandle) {
+            static std::atomic<int> pipeCounter{0};
+            auto pipeName = std::format(L"\\\\.\\pipe\\WslTest_RelayEof_{}", pipeCounter++);
+
+            SECURITY_ATTRIBUTES sa{sizeof(sa), nullptr, TRUE};
+            readHandle.reset(CreateNamedPipeW(
+                pipeName.c_str(), PIPE_ACCESS_INBOUND | FILE_FLAG_OVERLAPPED, PIPE_TYPE_BYTE | PIPE_WAIT, 1, 4096, 4096, 0, &sa));
+            VERIFY_IS_NOT_NULL(readHandle.get());
+
+            writeHandle.reset(CreateFileW(pipeName.c_str(), GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_FLAG_OVERLAPPED, nullptr));
+            VERIFY_IS_NOT_NULL(writeHandle.get());
+        };


+            wil::unique_handle readPipe, writePipe;
+            createOverlappedPipe(readPipe, writePipe);
+


+            // Create an output pipe to capture relayed data.
+            wil::unique_handle outputRead, outputWrite;
+            createOverlappedPipe(outputRead, outputWrite);
+


+            // Wait up to 5 seconds for the relay to finish. If it doesn't, the EOF check is broken.
+            VERIFY_ARE_NOT_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_TIMEOUT);
+            relayThread.join();


+            wil::unique_handle leftRead, leftWrite, rightRead, rightWrite;
+            createOverlappedPipe(leftRead, leftWrite);
+            createOverlappedPipe(rightRead, rightWrite);
+


+
+            VERIFY_ARE_NOT_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_TIMEOUT);
+            relayThread.join();


+            wil::unique_handle read1, write1, read2, write2;
+            createOverlappedPipe(read1, write1);
+            createOverlappedPipe(read2, write2);
+


When ReadFile completes synchronously with 0 bytes read (graceful close on sockets/pipes), the relay loops failed to detect this as EOF and re-issued the read in a tight loop, burning CPU indefinitely. Fix by checking for zero bytes on the synchronous-success path in: - BidirectionalRelay (both left and right sides) - ScopedMultiRelay::Run (both sync and overlapped completion paths) Fixes #40651 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 8 comments.

+            readHandle.reset(CreateNamedPipeW(
+                pipeName.c_str(), PIPE_ACCESS_INBOUND | FILE_FLAG_OVERLAPPED, PIPE_TYPE_BYTE | PIPE_WAIT, 1, 4096, 4096, 0, &sa));
+            VERIFY_IS_NOT_NULL(readHandle.get());


+            writeHandle.reset(CreateFileW(pipeName.c_str(), GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_FLAG_OVERLAPPED, nullptr));
+            VERIFY_IS_NOT_NULL(writeHandle.get());


+            serverHandle.reset(CreateNamedPipeW(
+                pipeName.c_str(), PIPE_ACCESS_DUPLEX | FILE_FLAG_OVERLAPPED, PIPE_TYPE_BYTE | PIPE_WAIT, 1, 4096, 4096, 0, &sa));
+            VERIFY_IS_NOT_NULL(serverHandle.get());


+            clientHandle.reset(CreateFileW(pipeName.c_str(), GENERIC_READ | GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_FLAG_OVERLAPPED, nullptr));
+            VERIFY_IS_NOT_NULL(clientHandle.get());


+                std::thread([&]() { wsl::windows::common::relay::InterruptableRelay(readPipe.get(), outputWrite.get()); });
+
+            // Wait up to 5 seconds for the relay to finish. If it doesn't, the EOF check is broken.
+            VERIFY_ARE_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_OBJECT_0);
+            relayThread.join();


+            outputWrite.reset();
+            char buf[64]{};
+            DWORD bytesRead{};
+            ReadFile(outputRead.get(), buf, sizeof(buf), &bytesRead, nullptr);
+            VERIFY_ARE_EQUAL(bytesRead, static_cast<DWORD>(testData.size()));
+            VERIFY_ARE_EQUAL(std::string_view(buf, bytesRead), testData);


+            auto relayThread =
+                std::thread([&]() { wsl::windows::common::relay::BidirectionalRelay(leftServer.get(), rightServer.get()); });
+
+            VERIFY_ARE_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_OBJECT_0);
+            relayThread.join();


+                // Sync should return promptly once both inputs hit EOF.
+                relay.Sync();
+            }


Copilot AI review requested due to automatic review settings June 5, 2026 22:07

benhillis requested a review from a team as a code owner June 5, 2026 22:07

Copilot started reviewing on behalf of benhillis June 5, 2026 22:07 View session

Copilot AI reviewed Jun 5, 2026

View reviewed changes

Comment thread test/windows/UnitTests.cpp

Comment thread test/windows/UnitTests.cpp

Comment thread src/windows/common/relay.cpp

benhillis force-pushed the benhill/fix-relay-eof-spin branch from 7a41534 to c06f4bc Compare June 5, 2026 22:58

OneBlue previously approved these changes Jun 5, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings June 6, 2026 02:51

benhillis dismissed OneBlue’s stale review via af6a873 June 6, 2026 02:51

benhillis force-pushed the benhill/fix-relay-eof-spin branch from c06f4bc to af6a873 Compare June 6, 2026 02:51

Copilot started reviewing on behalf of benhillis June 6, 2026 02:51 View session

Copilot AI reviewed Jun 6, 2026

View reviewed changes

benhillis force-pushed the benhill/fix-relay-eof-spin branch from af6a873 to a4a502a Compare June 7, 2026 20:01

Copilot AI review requested due to automatic review settings June 7, 2026 21:15

benhillis force-pushed the benhill/fix-relay-eof-spin branch from a4a502a to 0cfd705 Compare June 7, 2026 21:15

Copilot started reviewing on behalf of benhillis June 7, 2026 21:15 View session

Copilot AI reviewed Jun 7, 2026

View reviewed changes

benhillis force-pushed the benhill/fix-relay-eof-spin branch from 0cfd705 to d89dffc Compare June 8, 2026 14:28

Copilot AI review requested due to automatic review settings June 8, 2026 14:46

benhillis force-pushed the benhill/fix-relay-eof-spin branch from d89dffc to 820c01f Compare June 8, 2026 14:46

Copilot started reviewing on behalf of benhillis June 8, 2026 14:46 View session

Copilot AI reviewed Jun 8, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix relay loops spinning at 100% CPU on synchronous EOF#40725

Fix relay loops spinning at 100% CPU on synchronous EOF#40725
benhillis wants to merge 1 commit into
masterfrom
benhill/fix-relay-eof-spin

benhillis commented Jun 5, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

OneBlue left a comment

Uh oh!

benhillis commented Jun 5, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		VERIFY_ARE_NOT_EQUAL(WaitForSingleObject(relayThread.native_handle(), 5000), WAIT_TIMEOUT);
		relayThread.join();

		wil::unique_handle readPipe, writePipe;
		createOverlappedPipe(readPipe, writePipe);

		writeHandle.reset(CreateFileW(pipeName.c_str(), GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_FLAG_OVERLAPPED, nullptr));
		VERIFY_IS_NOT_NULL(writeHandle.get());

		clientHandle.reset(CreateFileW(pipeName.c_str(), GENERIC_READ \| GENERIC_WRITE, 0, &sa, OPEN_EXISTING, FILE_FLAG_OVERLAPPED, nullptr));
		VERIFY_IS_NOT_NULL(clientHandle.get());

Conversation

benhillis commented Jun 5, 2026

Summary

Problem

Fix

Validation

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

OneBlue left a comment

Choose a reason for hiding this comment

Uh oh!

benhillis commented Jun 5, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants