netstack: move SO_SNDBUF/RCVBUF clamping logic out of //pkg/tcpip

- Keeps Linux-specific behavior out of //pkg/tcpip
- Makes it clearer that clamping is done only for setsockopt calls from users
- Removes code duplication

PiperOrigin-RevId: 384389809

pkg/sentry/socket/netstack/netstack.go

@@ -1682,6 +1682,26 @@ func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int
 	return nil
 }
 
+func clampBufSize(newSz, min, max int64) int64 {
+	// packetOverheadFactor is used to multiply the value provided by the user on
+	// a setsockopt(2) for setting the send/receive buffer sizes sockets.
+	const packetOverheadFactor = 2
+
+	if newSz > max {
+		newSz = max
+	}
+
+	if newSz < math.MaxInt32/packetOverheadFactor {
+		newSz *= packetOverheadFactor
+		if newSz < min {
+			newSz = min
+		}
+	} else {
+		newSz = math.MaxInt32
+	}
+	return newSz
+}
+
 // setSockOptSocket implements SetSockOpt when level is SOL_SOCKET.
 func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name int, optVal []byte) *syserr.Error {
 	switch name {
@@ -1691,7 +1711,9 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam
 		}
 
 		v := hostarch.ByteOrder.Uint32(optVal)
-		ep.SocketOptions().SetSendBufferSize(int64(v), true /* notify */)
+		min, max := ep.SocketOptions().SendBufferLimits()
+		clamped := clampBufSize(int64(v), min, max)
+		ep.SocketOptions().SetSendBufferSize(clamped, true /* notify */)
 		return nil
 
 	case linux.SO_RCVBUF:
@@ -1700,7 +1722,9 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam
 		}
 
 		v := hostarch.ByteOrder.Uint32(optVal)
-		ep.SocketOptions().SetReceiveBufferSize(int64(v), true /* notify */)
+		min, max := ep.SocketOptions().ReceiveBufferLimits()
+		clamped := clampBufSize(int64(v), min, max)
+		ep.SocketOptions().SetReceiveBufferSize(clamped, true /* notify */)
 		return nil
 
 	case linux.SO_REUSEADDR:

pkg/tcpip/socketops.go

@@ -15,17 +15,12 @@
 package tcpip
 
 import (
-	"math"
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/atomicbitops"
 	"gvisor.dev/gvisor/pkg/sync"
 )
 
-// PacketOverheadFactor is used to multiply the value provided by the user on a
-// SetSockOpt for setting the send/receive buffer sizes sockets.
-const PacketOverheadFactor = 2
-
 // SocketOptionsHandler holds methods that help define endpoint specific
 // behavior for socket level socket options. These must be implemented by
 // endpoints to get notified when socket level options are set.
@@ -617,39 +612,20 @@ func (so *SocketOptions) GetSendBufferSize() int64 {
 	return so.sendBufferSize.Load()
 }
 
+// SendBufferLimits returns the [min, max) range of allowable send buffer
+// sizes.
+func (so *SocketOptions) SendBufferLimits() (min, max int64) {
+	limits := so.getSendBufferLimits(so.stackHandler)
+	return int64(limits.Min), int64(limits.Max)
+}
+
 // SetSendBufferSize sets value for SO_SNDBUF option. notify indicates if the
 // stack handler should be invoked to set the send buffer size.
 func (so *SocketOptions) SetSendBufferSize(sendBufferSize int64, notify bool) {
-	v := sendBufferSize
-
-	if !notify {
-		so.sendBufferSize.Store(v)
-		return
+	if notify {
+		sendBufferSize = so.handler.OnSetSendBufferSize(sendBufferSize)
 	}
-
-	// Make sure the send buffer size is within the min and max
-	// allowed.
-	ss := so.getSendBufferLimits(so.stackHandler)
-	min := int64(ss.Min)
-	max := int64(ss.Max)
-	// Validate the send buffer size with min and max values.
-	// Multiply it by factor of 2.
-	if v > max {
-		v = max
-	}
-
-	if v < math.MaxInt32/PacketOverheadFactor {
-		v *= PacketOverheadFactor
-		if v < min {
-			v = min
-		}
-	} else {
-		v = math.MaxInt32
-	}
-
-	// Notify endpoint about change in buffer size.
-	newSz := so.handler.OnSetSendBufferSize(v)
-	so.sendBufferSize.Store(newSz)
+	so.sendBufferSize.Store(sendBufferSize)
 }
 
 // GetReceiveBufferSize gets value for SO_RCVBUF option.
@@ -657,36 +633,19 @@ func (so *SocketOptions) GetReceiveBufferSize() int64 {
 	return so.receiveBufferSize.Load()
 }
 
-// SetReceiveBufferSize sets value for SO_RCVBUF option.
+// ReceiveBufferLimits returns the [min, max) range of allowable receive buffer
+// sizes.
+func (so *SocketOptions) ReceiveBufferLimits() (min, max int64) {
+	limits := so.getReceiveBufferLimits(so.stackHandler)
+	return int64(limits.Min), int64(limits.Max)
+}
+
+// SetReceiveBufferSize sets the value of the SO_RCVBUF option, optionally
+// notifying the owning endpoint.
 func (so *SocketOptions) SetReceiveBufferSize(receiveBufferSize int64, notify bool) {
-	if !notify {
-		so.receiveBufferSize.Store(receiveBufferSize)
-		return
-	}
-
-	// Make sure the send buffer size is within the min and max
-	// allowed.
-	v := receiveBufferSize
-	ss := so.getReceiveBufferLimits(so.stackHandler)
-	min := int64(ss.Min)
-	max := int64(ss.Max)
-	// Validate the send buffer size with min and max values.
-	if v > max {
-		v = max
+	if notify {
+		oldSz := so.receiveBufferSize.Load()
+		receiveBufferSize = so.handler.OnSetReceiveBufferSize(receiveBufferSize, oldSz)
 	}
-
-	// Multiply it by factor of 2.
-	if v < math.MaxInt32/PacketOverheadFactor {
-		v *= PacketOverheadFactor
-		if v < min {
-			v = min
-		}
-	} else {
-		v = math.MaxInt32
-	}
-
-	oldSz := so.receiveBufferSize.Load()
-	// Notify endpoint about change in buffer size.
-	newSz := so.handler.OnSetReceiveBufferSize(v, oldSz)
-	so.receiveBufferSize.Store(newSz)
+	so.receiveBufferSize.Store(receiveBufferSize)
 }

pkg/tcpip/transport/raw/endpoint.go

@@ -132,7 +132,7 @@ func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProt
 	// headers included. Because they're write-only, We don't need to
 	// register with the stack.
 	if !associated {
-		e.ops.SetReceiveBufferSize(0, false)
+		e.ops.SetReceiveBufferSize(0, false /* notify */)
 		e.waiterQueue = nil
 		return e, nil
 	}

pkg/tcpip/transport/tcp/tcp_test.go

@@ -2147,7 +2147,7 @@ func TestSmallSegReceiveWindowAdvertisement(t *testing.T) {
 
 	// Bump up the receive buffer size such that, when the receive window grows,
 	// the scaled window exceeds maxUint16.
-	c.EP.SocketOptions().SetReceiveBufferSize(int64(opt.Max), true)
+	c.EP.SocketOptions().SetReceiveBufferSize(int64(opt.Max)*2, true /* notify */)
 
 	// Keep the payload size < segment overhead and such that it is a multiple
 	// of the window scaled value. This enables the test to perform equality
@@ -2267,7 +2267,7 @@ func TestNoWindowShrinking(t *testing.T) {
 	initialWnd := header.TCP(header.IPv4(pkt).Payload()).WindowSize() << c.RcvdWindowScale
 	initialLastAcceptableSeq := iss.Add(seqnum.Size(initialWnd))
 	// Now shrink the receive buffer to half its original size.
-	c.EP.SocketOptions().SetReceiveBufferSize(int64(rcvBufSize/2), true)
+	c.EP.SocketOptions().SetReceiveBufferSize(int64(rcvBufSize), true /* notify */)
 
 	data := generateRandomPayload(t, rcvBufSize)
 	// Send a payload of half the size of rcvBufSize.
@@ -2523,7 +2523,7 @@ func TestScaledWindowAccept(t *testing.T) {
 	defer ep.Close()
 
 	// Set the window size greater than the maximum non-scaled window.
-	ep.SocketOptions().SetReceiveBufferSize(65535*3, true)
+	ep.SocketOptions().SetReceiveBufferSize(65535*6, true /* notify */)
 
 	if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil {
 		t.Fatalf("Bind failed: %s", err)
@@ -2595,7 +2595,7 @@ func TestNonScaledWindowAccept(t *testing.T) {
 	defer ep.Close()
 
 	// Set the window size greater than the maximum non-scaled window.
-	ep.SocketOptions().SetReceiveBufferSize(65535*3, true)
+	ep.SocketOptions().SetReceiveBufferSize(65535*6, true /* notify */)
 
 	if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil {
 		t.Fatalf("Bind failed: %s", err)
@@ -3188,7 +3188,7 @@ func TestPassiveSendMSSLessThanMTU(t *testing.T) {
 	// Set the buffer size to a deterministic size so that we can check the
 	// window scaling option.
 	const rcvBufferSize = 0x20000
-	ep.SocketOptions().SetReceiveBufferSize(rcvBufferSize, true)
+	ep.SocketOptions().SetReceiveBufferSize(rcvBufferSize*2, true /* notify */)
 
 	if err := ep.Bind(tcpip.FullAddress{Port: context.StackPort}); err != nil {
 		t.Fatalf("Bind failed: %s", err)
@@ -3327,7 +3327,7 @@ func TestSynOptionsOnActiveConnect(t *testing.T) {
 	// window scaling option.
 	const rcvBufferSize = 0x20000
 	const wndScale = 3
-	c.EP.SocketOptions().SetReceiveBufferSize(rcvBufferSize, true)
+	c.EP.SocketOptions().SetReceiveBufferSize(rcvBufferSize*2, true /* notify */)
 
 	// Start connection attempt.
 	we, ch := waiter.NewChannelEntry(nil)
@@ -4669,52 +4669,6 @@ func TestDefaultBufferSizes(t *testing.T) {
 	checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*3)
 }
 
-func TestMinMaxBufferSizes(t *testing.T) {
-	s := stack.New(stack.Options{
-		NetworkProtocols:   []stack.NetworkProtocolFactory{ipv4.NewProtocol},
-		TransportProtocols: []stack.TransportProtocolFactory{tcp.NewProtocol},
-	})
-
-	// Check the default values.
-	ep, err := s.NewEndpoint(tcp.ProtocolNumber, ipv4.ProtocolNumber, &waiter.Queue{})
-	if err != nil {
-		t.Fatalf("NewEndpoint failed; %s", err)
-	}
-	defer ep.Close()
-
-	// Change the min/max values for send/receive
-	{
-		opt := tcpip.TCPReceiveBufferSizeRangeOption{Min: 200, Default: tcp.DefaultReceiveBufferSize * 2, Max: tcp.DefaultReceiveBufferSize * 20}
-		if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, &opt); err != nil {
-			t.Fatalf("SetTransportProtocolOption(%d, &%#v): %s", tcp.ProtocolNumber, opt, err)
-		}
-	}
-
-	{
-		opt := tcpip.TCPSendBufferSizeRangeOption{Min: 300, Default: tcp.DefaultSendBufferSize * 3, Max: tcp.DefaultSendBufferSize * 30}
-		if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, &opt); err != nil {
-			t.Fatalf("SetTransportProtocolOption(%d, &%#v): %s", tcp.ProtocolNumber, opt, err)
-		}
-	}
-
-	// Set values below the min/2.
-	ep.SocketOptions().SetReceiveBufferSize(99, true)
-	checkRecvBufferSize(t, ep, 200)
-
-	ep.SocketOptions().SetSendBufferSize(149, true)
-
-	checkSendBufferSize(t, ep, 300)
-
-	// Set values above the max.
-	ep.SocketOptions().SetReceiveBufferSize(1+tcp.DefaultReceiveBufferSize*20, true)
-	// Values above max are capped at max and then doubled.
-	checkRecvBufferSize(t, ep, tcp.DefaultReceiveBufferSize*20*2)
-
-	ep.SocketOptions().SetSendBufferSize(1+tcp.DefaultSendBufferSize*30, true)
-	// Values above max are capped at max and then doubled.
-	checkSendBufferSize(t, ep, tcp.DefaultSendBufferSize*30*2)
-}
-
 func TestBindToDeviceOption(t *testing.T) {
 	s := stack.New(stack.Options{
 		NetworkProtocols:   []stack.NetworkProtocolFactory{ipv4.NewProtocol},
@@ -7752,7 +7706,7 @@ func TestIncreaseWindowOnBufferResize(t *testing.T) {
 
 	// Increasing the buffer from should generate an ACK,
 	// since window grew from small value to larger equal MSS
-	c.EP.SocketOptions().SetReceiveBufferSize(rcvBuf*2, true)
+	c.EP.SocketOptions().SetReceiveBufferSize(rcvBuf*4, true /* notify */)
 	checker.IPv4(t, c.GetPacket(),
 		checker.PayloadLen(header.TCPMinimumSize),
 		checker.TCP(

pkg/tcpip/transport/tcp/testing/context/context.go

@@ -757,7 +757,7 @@ func (c *Context) Create(epRcvBuf int) {
 	}
 
 	if epRcvBuf != -1 {
-		c.EP.SocketOptions().SetReceiveBufferSize(int64(epRcvBuf), true /* notify */)
+		c.EP.SocketOptions().SetReceiveBufferSize(int64(epRcvBuf)*2, true /* notify */)
 	}
 }