net/http: fix when writeLoop exited still has request in writech

When there is still a request in writech after
writeLoop exits, the request body will not be
closed and will not be retried.
The request is not sent to the server, so it can
be safely retried without regard to idempotence.

Fixes golang#49621

Author: Aoang

src/net/http/transport.go

@@ -719,6 +719,13 @@ func (pc *persistConn) shouldRetryRequest(req *Request, err error) bool {
 		// can "rewind" the body with GetBody.
 		return req.outgoingLength() == 0 || req.GetBody != nil
 	}
+	if len(pc.writech) != 0 {
+		// Request was sent successfully into writech, but isn't read from it
+		// because writeLoop exited.
+		// In this case, you can retry without considering idempotence.
+		// see https://github.com/golang/go/issues/49621
+		return true
+	}
 	if !req.isReplayable() {
 		// Don't retry non-idempotent requests.
 		return false

src/net/http/transport_test.go

@@ -40,6 +40,7 @@ import (
 	"strings"
 	"sync"
 	"sync/atomic"
+	"syscall"
 	"testing"
 	"testing/iotest"
 	"time"
@@ -6654,3 +6655,133 @@ func testHandlerAbortRacesBodyRead(t *testing.T, mode testMode) {
 	}
 	wg.Wait()
 }
+
+type issue49621Listener struct {
+	net.Listener
+}
+type issue49621Conn struct {
+	net.Conn // base
+
+	mtx sync.Mutex
+	err error
+	n   int
+}
+
+func (l *issue49621Listener) Accept() (net.Conn, error) {
+	c, err := l.Listener.Accept()
+	if err == nil {
+		c = &issue49621Conn{Conn: c}
+	}
+	return c, err
+}
+
+func (c *issue49621Conn) inject(n int) bool {
+	c.mtx.Lock()
+	defer c.mtx.Unlock()
+	c.n += n
+	return c.n > 1000
+}
+
+func (c *issue49621Conn) isFailed() error {
+	c.mtx.Lock()
+	defer c.mtx.Unlock()
+	return c.err
+}
+
+func (c *issue49621Conn) markFailed(err error) error {
+	c.mtx.Lock()
+	defer c.mtx.Unlock()
+	c.err = err
+	return c.err
+}
+
+func (c *issue49621Conn) Read(b []byte) (n int, err error) {
+	if err = c.isFailed(); err != nil {
+		return 0, err
+	}
+
+	n, err = c.Conn.Read(b)
+	if err != nil {
+		c.markFailed(err)
+		return n, err
+	}
+
+	if c.inject(n) {
+		return 0, c.markFailed(syscall.ECONNRESET)
+	}
+
+	return n, nil
+}
+
+func (c *issue49621Conn) Write(b []byte) (n int, err error) {
+	if err = c.isFailed(); err != nil {
+		return 0, err
+	}
+
+	if c.inject(len(b)) {
+		return 0, c.markFailed(syscall.ECONNRESET)
+	}
+
+	n, err = c.Conn.Write(b)
+	if err != nil {
+		return n, c.markFailed(err)
+	}
+	return n, nil
+}
+
+// Issue 49621: request not retry or close body after writeLoop exited.
+func TestIssue49621(t *testing.T) {
+	ln := &issue49621Listener{newLocalListener(t)}
+	defer ln.Close()
+
+	addr := ln.Addr().String()
+	handler := HandlerFunc(func(w ResponseWriter, r *Request) {
+		if r.Method != MethodPost {
+			w.WriteHeader(StatusMethodNotAllowed)
+			return
+		}
+
+		_, _ = io.Copy(io.Discard, r.Body)
+		_ = r.Body.Close()
+		w.WriteHeader(StatusOK)
+	})
+	s := &Server{
+		Addr:    addr,
+		Handler: handler,
+	}
+	go func() { s.Serve(ln) }()
+
+	for i := 0; i < 10; i++ {
+		testIssue49621Request(t, addr)
+	}
+	s.Close()
+}
+
+func testIssue49621Request(t *testing.T, addr string) {
+	var wg sync.WaitGroup
+
+	for i := 0; i < 50; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+
+			body := bytes.NewBuffer([]byte(`
+				hello world hello world hello world hello world
+				hello world hello world hello world hello world
+				hello world hello world hello world hello world
+				hello world hello world hello world hello world
+				hello world hello world hello world hello world
+			`))
+
+			resp, err := Post(addr, "plain", body)
+			if err == nil {
+				_, _ = io.Copy(io.Discard, resp.Body)
+				_ = resp.Body.Close()
+				return
+			}
+
+			t.Errorf("resp = %v, err = %v\n", resp, err)
+		}()
+	}
+	wg.Wait()
+}