diff --git a/listener/tun/metadata.go b/listener/tun/metadata.go
index 0b0154d3..ded8cfe1 100644
--- a/listener/tun/metadata.go
+++ b/listener/tun/metadata.go
@@ -14,13 +14,11 @@ import (
 )
 
 const (
-	defaultMTU            = 1350
-	defaultReadBufferSize = 4096
+	defaultMTU = 1420
 )
 
 type metadata struct {
-	config         *tun_util.Config
-	readBufferSize int
+	config *tun_util.Config
 }
 
 func (l *tunListener) parseMetadata(md mdata.Metadata) (err error) {
@@ -34,11 +32,6 @@ func (l *tunListener) parseMetadata(md mdata.Metadata) (err error) {
 		gateway = "gw"
 	)
 
-	l.md.readBufferSize = mdutil.GetInt(md, "tun.rbuf", "rbuf", "readBufferSize")
-	if l.md.readBufferSize <= 0 {
-		l.md.readBufferSize = defaultReadBufferSize
-	}
-
 	config := &tun_util.Config{
 		Name:   mdutil.GetString(md, name),
 		Peer:   mdutil.GetString(md, peer),
diff --git a/listener/tun/tun.go b/listener/tun/tun.go
index ba710ee4..975f83d6 100644
--- a/listener/tun/tun.go
+++ b/listener/tun/tun.go
@@ -2,34 +2,58 @@ package tun
 
 import (
 	"io"
+	"math"
 
-	"github.com/go-gost/core/common/bufpool"
 	"golang.zx2c4.com/wireguard/tun"
 )
 
 const (
 	tunOffsetBytes = 16
+	MaxMessageSize = math.MaxUint16
 )
 
 type tunDevice struct {
-	dev            tun.Device
-	readBufferSize int
+	dev   tun.Device
+	elems [][MaxMessageSize]byte
+	rbufs [][]byte
+	wbuf  [tunOffsetBytes + MaxMessageSize]byte
 }
 
 func (d *tunDevice) Read(p []byte) (n int, err error) {
-	sizes := [1]int{}
-	_, err = d.dev.Read([][]byte{p}, sizes[:], 0)
-	n = sizes[0]
+	if len(d.rbufs) > 0 {
+		n = copy(p, d.rbufs[0])
+		d.rbufs = d.rbufs[1:]
+		return
+	}
+
+	sizes := make([]int, len(d.elems))
+	bufs := make([][]byte, len(d.elems))
+	for i := range d.elems {
+		bufs[i] = d.elems[i][:]
+	}
+
+	nn, err := d.dev.Read(bufs, sizes, 0)
+	if err != nil {
+		return
+	}
+
+	for i := 0; i < nn; i++ {
+		if sizes[i] <= 0 {
+			continue
+		}
+		d.rbufs = append(d.rbufs, bufs[i][:sizes[i]])
+	}
+
+	n = copy(p, d.rbufs[0])
+	d.rbufs = d.rbufs[1:]
+
 	return
 }
 
 func (d *tunDevice) Write(p []byte) (n int, err error) {
-	b := bufpool.Get(tunOffsetBytes + len(p))
-	defer bufpool.Put(b)
-
-	copy(b[tunOffsetBytes:], p)
-	_, err = d.dev.Write([][]byte{b}, tunOffsetBytes)
-	n = len(p)
+	buf := d.wbuf[:]
+	n = copy(buf[tunOffsetBytes:], p)
+	_, err = d.dev.Write([][]byte{buf[:tunOffsetBytes+n]}, tunOffsetBytes)
 	return
 }
 
@@ -44,8 +68,8 @@ func (l *tunListener) createTunDevice() (dev io.ReadWriteCloser, name string, er
 	}
 
 	dev = &tunDevice{
-		dev:            ifce,
-		readBufferSize: l.md.readBufferSize,
+		dev:   ifce,
+		elems: make([][MaxMessageSize]byte, ifce.BatchSize()),
 	}
 	name, err = ifce.Name()