fdbased: Vectorized write for packet; relax writev syscall filter.

Now it calls pkt.Data.ToView() when writing the packet. This may require
copying when the packet is large, which puts the worse case in an even worse
situation.

This sent out in a separate preparation change as it requires syscall filter
changes. This change will be followed by the change for the adoption of the new
PacketHeader API.

PiperOrigin-RevId: 321447003

3 files changed, 26 insertions, 76 deletions

diff --git a/pkg/tcpip/link/fdbased/BUILD b/pkg/tcpip/link/fdbased/BUILD
index aa6db9aea..507b44abc 100644
--- a/pkg/tcpip/link/fdbased/BUILD
+++ b/pkg/tcpip/link/fdbased/BUILD
@@ -15,6 +15,7 @@ go_library(
     visibility = ["//visibility:public"],
     deps = [
         "//pkg/binary",
+        "//pkg/iovec",
         "//pkg/sync",
         "//pkg/tcpip",
         "//pkg/tcpip/buffer",
diff --git a/pkg/tcpip/link/fdbased/endpoint.go b/pkg/tcpip/link/fdbased/endpoint.go
index 32abe2a13..6aa1badc7 100644
--- a/pkg/tcpip/link/fdbased/endpoint.go
+++ b/pkg/tcpip/link/fdbased/endpoint.go
@@ -45,6 +45,7 @@ import (
 
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/binary"
+	"gvisor.dev/gvisor/pkg/iovec"
 	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
@@ -406,6 +407,8 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne
 		eth.Encode(ethHdr)
 	}
 
+	var builder iovec.Builder
+
 	fd := e.fds[pkt.Hash%uint32(len(e.fds))]
 	if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
 		vnetHdr := virtioNetHdr{}
@@ -430,29 +433,25 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, protocol tcpip.Ne
 		}
 
 		vnetHdrBuf := binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr)
-		return rawfile.NonBlockingWrite3(fd, vnetHdrBuf, pkt.Header.View(), pkt.Data.ToView())
+		builder.Add(vnetHdrBuf)
 	}
 
-	if pkt.Data.Size() == 0 {
-		return rawfile.NonBlockingWrite(fd, pkt.Header.View())
-	}
-	if pkt.Header.UsedLength() == 0 {
-		return rawfile.NonBlockingWrite(fd, pkt.Data.ToView())
+	builder.Add(pkt.Header.View())
+	for _, v := range pkt.Data.Views() {
+		builder.Add(v)
 	}
 
-	return rawfile.NonBlockingWrite3(fd, pkt.Header.View(), pkt.Data.ToView(), nil)
+	return rawfile.NonBlockingWriteIovec(fd, builder.Build())
 }
 
 func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tcpip.Error) {
 	// Send a batch of packets through batchFD.
 	mmsgHdrs := make([]rawfile.MMsgHdr, 0, len(batch))
 	for _, pkt := range batch {
-		var ethHdrBuf []byte
-		iovLen := 0
+		var eth header.Ethernet
 		if e.hdrSize > 0 {
 			// Add ethernet header if needed.
-			ethHdrBuf = make([]byte, header.EthernetMinimumSize)
-			eth := header.Ethernet(ethHdrBuf)
+			eth = make(header.Ethernet, header.EthernetMinimumSize)
 			ethHdr := &header.EthernetFields{
 				DstAddr: pkt.EgressRoute.RemoteLinkAddress,
 				Type:    pkt.NetworkProtocolNumber,
@@ -465,12 +464,11 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc
 				ethHdr.SrcAddr = e.addr
 			}
 			eth.Encode(ethHdr)
-			iovLen++
 		}
 
-		vnetHdr := virtioNetHdr{}
 		var vnetHdrBuf []byte
 		if e.Capabilities()&stack.CapabilityHardwareGSO != 0 {
+			vnetHdr := virtioNetHdr{}
 			if pkt.GSOOptions != nil {
 				vnetHdr.hdrLen = uint16(pkt.Header.UsedLength())
 				if pkt.GSOOptions.NeedsCsum {
@@ -491,45 +489,20 @@ func (e *endpoint) sendBatch(batchFD int, batch []*stack.PacketBuffer) (int, *tc
 				}
 			}
 			vnetHdrBuf = binary.Marshal(make([]byte, 0, virtioNetHdrSize), binary.LittleEndian, vnetHdr)
-			iovLen++
 		}
 
-		iovecs := make([]syscall.Iovec, iovLen+1+len(pkt.Data.Views()))
+		var builder iovec.Builder
+		builder.Add(vnetHdrBuf)
+		builder.Add(eth)
+		builder.Add(pkt.Header.View())
+		for _, v := range pkt.Data.Views() {
+			builder.Add(v)
+		}
+		iovecs := builder.Build()
+
 		var mmsgHdr rawfile.MMsgHdr
 		mmsgHdr.Msg.Iov = &iovecs[0]
-		iovecIdx := 0
-		if vnetHdrBuf != nil {
-			v := &iovecs[iovecIdx]
-			v.Base = &vnetHdrBuf[0]
-			v.Len = uint64(len(vnetHdrBuf))
-			iovecIdx++
-		}
-		if ethHdrBuf != nil {
-			v := &iovecs[iovecIdx]
-			v.Base = &ethHdrBuf[0]
-			v.Len = uint64(len(ethHdrBuf))
-			iovecIdx++
-		}
-		pktSize := uint64(0)
-		// Encode L3 Header
-		v := &iovecs[iovecIdx]
-		hdr := &pkt.Header
-		hdrView := hdr.View()
-		v.Base = &hdrView[0]
-		v.Len = uint64(len(hdrView))
-		pktSize += v.Len
-		iovecIdx++
-
-		// Now encode the Transport Payload.
-		pktViews := pkt.Data.Views()
-		for i := range pktViews {
-			vec := &iovecs[iovecIdx]
-			iovecIdx++
-			vec.Base = &pktViews[i][0]
-			vec.Len = uint64(len(pktViews[i]))
-			pktSize += vec.Len
-		}
-		mmsgHdr.Msg.Iovlen = uint64(iovecIdx)
+		mmsgHdr.Msg.Iovlen = uint64(len(iovecs))
 		mmsgHdrs = append(mmsgHdrs, mmsgHdr)
 	}
 
diff --git a/pkg/tcpip/link/rawfile/rawfile_unsafe.go b/pkg/tcpip/link/rawfile/rawfile_unsafe.go
index 69de6eb3e..f4c32c2da 100644
--- a/pkg/tcpip/link/rawfile/rawfile_unsafe.go
+++ b/pkg/tcpip/link/rawfile/rawfile_unsafe.go
@@ -66,38 +66,14 @@ func NonBlockingWrite(fd int, buf []byte) *tcpip.Error {
 	return nil
 }
 
-// NonBlockingWrite3 writes up to three byte slices to a file descriptor in a
-// single syscall. It fails if partial data is written.
-func NonBlockingWrite3(fd int, b1, b2, b3 []byte) *tcpip.Error {
-	// If there is no second and third buffer, issue a regular write.
-	if len(b2) == 0 && len(b3) == 0 {
-		return NonBlockingWrite(fd, b1)
-	}
-
-	// Build the iovec that represents them and issue a writev syscall.
-	iovec := [3]syscall.Iovec{
-		{
-			Base: &b1[0],
-			Len:  uint64(len(b1)),
-		},
-		{
-			Base: &b2[0],
-			Len:  uint64(len(b2)),
-		},
-	}
-	iovecLen := uintptr(2)
-
-	if len(b3) > 0 {
-		iovecLen++
-		iovec[2].Base = &b3[0]
-		iovec[2].Len = uint64(len(b3))
-	}
-
+// NonBlockingWriteIovec writes iovec to a file descriptor in a single syscall.
+// It fails if partial data is written.
+func NonBlockingWriteIovec(fd int, iovec []syscall.Iovec) *tcpip.Error {
+	iovecLen := uintptr(len(iovec))
 	_, _, e := syscall.RawSyscall(syscall.SYS_WRITEV, uintptr(fd), uintptr(unsafe.Pointer(&iovec[0])), iovecLen)
 	if e != 0 {
 		return TranslateErrno(e)
 	}
-
 	return nil
 }