Stream: git-wasmtime

Topic: wasmtime / issue #12402 SO_REUSEADDR should be set on TCP...

Wasmtime GitHub notifications bot (Jan 23 2026 at 14:06):

wingo opened issue #12402:

When writing a test that sockets have SO_REUSEADDR set for wasip3, @saulecabrera ran into a bug that shows up only on Linux (not mac, not windows):

async fn test_reuseaddr(family: IpAddressFamily) {
    let client = TcpSocket::create(family).unwrap();
    let local_addr = {
        let server = TcpSocket::create(family).unwrap();
        let addr = IpSocketAddress::localhost(family, 0);
        server.bind(addr).unwrap();
        let local_addr = server.get_local_address().unwrap();
        let mut accept = server.listen().unwrap();
        join!(
            // Change the state to connected.
            async {
                client.connect(local_addr).await.unwrap();
            },
            async {
                let sock = accept.next().await.unwrap();
                let (mut send_tx, send_rx) = wit_stream::new();
                join!(
                    async {
                        sock.send(send_rx).await.unwrap();
                    },
                    async {
                        let remaining = send_tx.write_all(vec![0; 1]).await;
                        assert!(remaining.is_empty());
                        drop(send_tx);
                    }
                );
            }
        );
        local_addr
    };

    // Immediately try to connect to the same after the connection is
    // dropped.  According to the spec, `SO_REUSEADDR` should be set
    // by default, so the next connection should not be affected by
    // the `TIME_WAIT` state.
    let next = TcpSocket::create(family).unwrap();
    next.bind(local_addr).unwrap();
    next.listen().unwrap();
}

When next is bound to the same local_addr, the bind fails. It is apparently because the first socket server wasn't opened with REUSEADDR, because port is 0 : https://github.com/bytecodealliance/wasmtime/blob/main/crates/wasi/src/sockets/util.rs#L345. If I change the condition on that line to unconditional "true", Saúl's wasi test passes, and no wasmtime test fails.

So, I propose that wasmtime should just always set REUSEADDR (unless we are on a windows system).

Full test here: https://github.com/WebAssembly/wasi-testsuite/pull/205

Wasmtime GitHub notifications bot (Jan 23 2026 at 14:06):

wingo added the bug label to Issue #12402.

Wasmtime GitHub notifications bot (Jan 23 2026 at 23:30):

alexcrichton commented on issue #12402:

cc @badeend, do you know why this explicitly excludes port 0 from setting reuseaddr? Changing that gets this test passing, and naively I'd expect true to be unconditional there, but you're far more knowledgable about this all than I

Wasmtime GitHub notifications bot (Jan 23 2026 at 23:30):

alexcrichton added the wasi label to Issue #12402.

Wasmtime GitHub notifications bot (Jan 23 2026 at 23:32):

alexcrichton commented on issue #12402:

Also @rvolosatovs you might be able to shed more light on my above question too

Wasmtime GitHub notifications bot (Feb 02 2026 at 18:30):

badeend commented on issue #12402:

Yes, this is indeed a consequence of the current setup.

For listener sockets I think it's fine to set it unconditionally.
For client sockets I'm not sure.
Let me think about it

Wasmtime GitHub notifications bot (Feb 15 2026 at 09:17):

badeend commented on issue #12402:

I think it should be fine :+1:

PTAL at the two PRs I just opened

Wasmtime GitHub notifications bot (Feb 15 2026 at 16:48):

alexcrichton closed issue #12402:

When writing a test that sockets have SO_REUSEADDR set for wasip3, @saulecabrera ran into a bug that shows up only on Linux (not mac, not windows):

async fn test_reuseaddr(family: IpAddressFamily) {
    let client = TcpSocket::create(family).unwrap();
    let local_addr = {
        let server = TcpSocket::create(family).unwrap();
        let addr = IpSocketAddress::localhost(family, 0);
        server.bind(addr).unwrap();
        let local_addr = server.get_local_address().unwrap();
        let mut accept = server.listen().unwrap();
        join!(
            // Change the state to connected.
            async {
                client.connect(local_addr).await.unwrap();
            },
            async {
                let sock = accept.next().await.unwrap();
                let (mut send_tx, send_rx) = wit_stream::new();
                join!(
                    async {
                        sock.send(send_rx).await.unwrap();
                    },
                    async {
                        let remaining = send_tx.write_all(vec![0; 1]).await;
                        assert!(remaining.is_empty());
                        drop(send_tx);
                    }
                );
            }
        );
        local_addr
    };

    // Immediately try to connect to the same after the connection is
    // dropped.  According to the spec, `SO_REUSEADDR` should be set
    // by default, so the next connection should not be affected by
    // the `TIME_WAIT` state.
    let next = TcpSocket::create(family).unwrap();
    next.bind(local_addr).unwrap();
    next.listen().unwrap();
}

When next is bound to the same local_addr, the bind fails. It is apparently because the first socket server wasn't opened with REUSEADDR, because port is 0 : https://github.com/bytecodealliance/wasmtime/blob/main/crates/wasi/src/sockets/util.rs#L345. If I change the condition on that line to unconditional "true", Saúl's wasi test passes, and no wasmtime test fails.

So, I propose that wasmtime should just always set REUSEADDR (unless we are on a windows system).

Full test here: https://github.com/WebAssembly/wasi-testsuite/pull/205

Last updated: Feb 24 2026 at 05:28 UTC