net: reduce allocations for UDP send/recv on Windows

This brings the optimizations added in CLs 331489 and 331490 to Windows.

Updates #43451

Change-Id: I75cf520050325d9eb5c2785d6d8677cc864fcac8
Reviewed-on: https://go-review.googlesource.com/c/go/+/331511
Trust: Josh Bleecher Snyder <josharian@gmail.com>
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Damien Neil <dneil@google.com>

api/next.txt

@@ -102,3 +102,7 @@ pkg syscall (openbsd-amd64-cgo), func RecvfromInet4(int, []uint8, int, *Sockaddr
 pkg syscall (openbsd-amd64-cgo), func RecvfromInet6(int, []uint8, int, *SockaddrInet6) (int, error)
 pkg syscall (openbsd-amd64-cgo), func SendtoInet4(int, []uint8, int, SockaddrInet4) error
 pkg syscall (openbsd-amd64-cgo), func SendtoInet6(int, []uint8, int, SockaddrInet6) error
+pkg syscall (windows-386), func WSASendtoInet4(Handle, *WSABuf, uint32, *uint32, uint32, SockaddrInet4, *Overlapped, *uint8) error
+pkg syscall (windows-386), func WSASendtoInet6(Handle, *WSABuf, uint32, *uint32, uint32, SockaddrInet6, *Overlapped, *uint8) error
+pkg syscall (windows-amd64), func WSASendtoInet4(Handle, *WSABuf, uint32, *uint32, uint32, SockaddrInet4, *Overlapped, *uint8) error
+pkg syscall (windows-amd64), func WSASendtoInet6(Handle, *WSABuf, uint32, *uint32, uint32, SockaddrInet6, *Overlapped, *uint8) error

src/internal/poll/fd_windows.go

@@ -79,6 +79,8 @@ type operation struct {
 	buf    syscall.WSABuf
 	msg    windows.WSAMsg
 	sa     syscall.Sockaddr
+	sa4    syscall.SockaddrInet4
+	sa6    syscall.SockaddrInet6
 	rsa    *syscall.RawSockaddrAny
 	rsan   int32
 	handle syscall.Handle
@@ -595,7 +597,30 @@ func (fd *FD) ReadFrom(buf []byte) (int, syscall.Sockaddr, error) {
 
 // ReadFrom wraps the recvfrom network call for IPv4.
 func (fd *FD) ReadFromInet4(buf []byte, sa4 *syscall.SockaddrInet4) (int, error) {
-	n, sa, err := fd.ReadFrom(buf)
+	if len(buf) == 0 {
+		return 0, nil
+	}
+	if len(buf) > maxRW {
+		buf = buf[:maxRW]
+	}
+	if err := fd.readLock(); err != nil {
+		return 0, err
+	}
+	defer fd.readUnlock()
+	o := &fd.rop
+	o.InitBuf(buf)
+	n, err := execIO(o, func(o *operation) error {
+		if o.rsa == nil {
+			o.rsa = new(syscall.RawSockaddrAny)
+		}
+		o.rsan = int32(unsafe.Sizeof(*o.rsa))
+		return syscall.WSARecvFrom(o.fd.Sysfd, &o.buf, 1, &o.qty, &o.flags, o.rsa, &o.rsan, &o.o, nil)
+	})
+	err = fd.eofError(n, err)
+	if err != nil {
+		return n, err
+	}
+	sa, _ := o.rsa.Sockaddr()
 	if sa != nil {
 		*sa4 = *(sa.(*syscall.SockaddrInet4))
 	}
@@ -604,7 +629,30 @@ func (fd *FD) ReadFromInet4(buf []byte, sa4 *syscall.SockaddrInet4) (int, error)
 
 // ReadFrom wraps the recvfrom network call for IPv6.
 func (fd *FD) ReadFromInet6(buf []byte, sa6 *syscall.SockaddrInet6) (int, error) {
-	n, sa, err := fd.ReadFrom(buf)
+	if len(buf) == 0 {
+		return 0, nil
+	}
+	if len(buf) > maxRW {
+		buf = buf[:maxRW]
+	}
+	if err := fd.readLock(); err != nil {
+		return 0, err
+	}
+	defer fd.readUnlock()
+	o := &fd.rop
+	o.InitBuf(buf)
+	n, err := execIO(o, func(o *operation) error {
+		if o.rsa == nil {
+			o.rsa = new(syscall.RawSockaddrAny)
+		}
+		o.rsan = int32(unsafe.Sizeof(*o.rsa))
+		return syscall.WSARecvFrom(o.fd.Sysfd, &o.buf, 1, &o.qty, &o.flags, o.rsa, &o.rsan, &o.o, nil)
+	})
+	err = fd.eofError(n, err)
+	if err != nil {
+		return n, err
+	}
+	sa, _ := o.rsa.Sockaddr()
 	if sa != nil {
 		*sa6 = *(sa.(*syscall.SockaddrInet6))
 	}
@@ -810,8 +858,42 @@ func (fd *FD) WriteTo(buf []byte, sa syscall.Sockaddr) (int, error) {
 }
 
 // WriteTo wraps the sendto network call for IPv4.
-func (fd *FD) WriteToInet4(buf []byte, sa syscall.SockaddrInet4) (int, error) {
+func (fd *FD) WriteToInet4(buf []byte, sa4 syscall.SockaddrInet4) (int, error) {
-	return fd.WriteTo(buf, &sa)
+	if err := fd.writeLock(); err != nil {
+		return 0, err
+	}
+	defer fd.writeUnlock()
+
+	if len(buf) == 0 {
+		// handle zero-byte payload
+		o := &fd.wop
+		o.InitBuf(buf)
+		o.sa4 = sa4
+		n, err := execIO(o, func(o *operation) error {
+			return syscall.WSASendtoInet4(o.fd.Sysfd, &o.buf, 1, &o.qty, 0, o.sa4, &o.o, nil)
+		})
+		return n, err
+	}
+
+	ntotal := 0
+	for len(buf) > 0 {
+		b := buf
+		if len(b) > maxRW {
+			b = b[:maxRW]
+		}
+		o := &fd.wop
+		o.InitBuf(b)
+		o.sa4 = sa4
+		n, err := execIO(o, func(o *operation) error {
+			return syscall.WSASendtoInet4(o.fd.Sysfd, &o.buf, 1, &o.qty, 0, o.sa4, &o.o, nil)
+		})
+		ntotal += int(n)
+		if err != nil {
+			return ntotal, err
+		}
+		buf = buf[n:]
+	}
+	return ntotal, nil
 }
 
 // WriteTo wraps the sendto network call for IPv6.

src/syscall/syscall_windows.go

@@ -924,6 +924,38 @@ func WSASendto(s Handle, bufs *WSABuf, bufcnt uint32, sent *uint32, flags uint32
 	return err
 }
 
+func WSASendtoInet4(s Handle, bufs *WSABuf, bufcnt uint32, sent *uint32, flags uint32, to SockaddrInet4, overlapped *Overlapped, croutine *byte) (err error) {
+	rsa, len, err := to.sockaddr()
+	if err != nil {
+		return err
+	}
+	r1, _, e1 := Syscall9(procWSASendTo.Addr(), 9, uintptr(s), uintptr(unsafe.Pointer(bufs)), uintptr(bufcnt), uintptr(unsafe.Pointer(sent)), uintptr(flags), uintptr(unsafe.Pointer(rsa)), uintptr(len), uintptr(unsafe.Pointer(overlapped)), uintptr(unsafe.Pointer(croutine)))
+	if r1 == socket_error {
+		if e1 != 0 {
+			err = errnoErr(e1)
+		} else {
+			err = EINVAL
+		}
+	}
+	return err
+}
+
+func WSASendtoInet6(s Handle, bufs *WSABuf, bufcnt uint32, sent *uint32, flags uint32, to SockaddrInet6, overlapped *Overlapped, croutine *byte) (err error) {
+	rsa, len, err := to.sockaddr()
+	if err != nil {
+		return err
+	}
+	r1, _, e1 := Syscall9(procWSASendTo.Addr(), 9, uintptr(s), uintptr(unsafe.Pointer(bufs)), uintptr(bufcnt), uintptr(unsafe.Pointer(sent)), uintptr(flags), uintptr(unsafe.Pointer(rsa)), uintptr(len), uintptr(unsafe.Pointer(overlapped)), uintptr(unsafe.Pointer(croutine)))
+	if r1 == socket_error {
+		if e1 != 0 {
+			err = errnoErr(e1)
+		} else {
+			err = EINVAL
+		}
+	}
+	return err
+}
+
 func LoadGetAddrInfo() error {
 	return procGetAddrInfoW.Find()
 }