Support owner matching for iptables.

This feature will match UID and GID of the packet creator, for locally
generated packets. This match is only valid in the OUTPUT and POSTROUTING
chains. Forwarded packets do not have any socket associated with them.
Packets from kernel threads do have a socket, but usually no owner.

13 files changed, 86 insertions, 13 deletions

diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index b3ee6000e..a7d9a8b25 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -244,6 +244,14 @@ func (e *endpoint) WritePacket(r *stack.Route, gso *stack.GSO, params stack.Netw
 	ip := e.addIPHeader(r, &pkt.Header, pkt.Data.Size(), params)
 	pkt.NetworkHeader = buffer.View(ip)
 
+	// iptables filtering. All packets that reach here are locally
+	// generated.
+	ipt := e.stack.IPTables()
+	if ok := ipt.Check(stack.Output, pkt); !ok {
+		// iptables is telling us to drop the packet.
+		return nil
+	}
+
 	if r.Loop&stack.PacketLoop != 0 {
 		// The inbound path expects the network header to still be in
 		// the PacketBuffer's Data field.
@@ -280,7 +288,14 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts []stack.Pac
 		return len(pkts), nil
 	}
 
+	// iptables filtering. All packets that reach here are locally
+	// generated.
+	ipt := e.stack.IPTables()
 	for i := range pkts {
+		if ok := ipt.Check(stack.Output, pkts[i]); !ok {
+			// iptables is telling us to drop the packet.
+			continue
+		}
 		ip := e.addIPHeader(r, &pkts[i].Header, pkts[i].DataSize, params)
 		pkts[i].NetworkHeader = buffer.View(ip)
 	}
diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go
index 9505a4e92..9367de180 100644
--- a/pkg/tcpip/stack/packet_buffer.go
+++ b/pkg/tcpip/stack/packet_buffer.go
@@ -13,7 +13,10 @@
 
 package stack
 
-import "gvisor.dev/gvisor/pkg/tcpip/buffer"
+import (
+	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+)
 
 // A PacketBuffer contains all the data of a network packet.
 //
@@ -59,6 +62,10 @@ type PacketBuffer struct {
 	// Hash is the transport layer hash of this packet. A value of zero
 	// indicates no valid hash has been set.
 	Hash uint32
+
+	// Owner is implemented by task to get the uid and gid.
+	// Only set for locally generated packets.
+	Owner tcpip.PacketOwner
 }
 
 // Clone makes a copy of pk. It clones the Data field, which creates a new
diff --git a/pkg/tcpip/stack/transport_test.go b/pkg/tcpip/stack/transport_test.go
index 8ca9ac3cf..3084e6593 100644
--- a/pkg/tcpip/stack/transport_test.go
+++ b/pkg/tcpip/stack/transport_test.go
@@ -56,6 +56,8 @@ func (f *fakeTransportEndpoint) Stats() tcpip.EndpointStats {
 	return nil
 }
 
+func (f *fakeTransportEndpoint) SetOwner(owner tcpip.PacketOwner) {}
+
 func newFakeTransportEndpoint(s *stack.Stack, proto *fakeTransportProtocol, netProto tcpip.NetworkProtocolNumber, uniqueID uint64) tcpip.Endpoint {
 	return &fakeTransportEndpoint{stack: s, TransportEndpointInfo: stack.TransportEndpointInfo{NetProto: netProto}, proto: proto, uniqueID: uniqueID}
 }
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 3dc5d87d6..2ef3271f1 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -336,6 +336,15 @@ type ControlMessages struct {
 	PacketInfo IPPacketInfo
 }
 
+// PacketOwner is used to get UID and GID of the packet.
+type PacketOwner interface {
+	// UID returns UID of the packet.
+	UID() uint32
+
+	// GID returns GID of the packet.
+	GID() uint32
+}
+
 // Endpoint is the interface implemented by transport protocols (e.g., tcp, udp)
 // that exposes functionality like read, write, connect, etc. to users of the
 // networking stack.
@@ -470,6 +479,9 @@ type Endpoint interface {
 
 	// Stats returns a reference to the endpoint stats.
 	Stats() EndpointStats
+
+	// SetOwner sets the task owner to the endpoint owner.
+	SetOwner(owner PacketOwner)
 }
 
 // EndpointInfo is the interface implemented by each endpoint info struct.
diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go
index 613b12ead..b007302fb 100644
--- a/pkg/tcpip/transport/icmp/endpoint.go
+++ b/pkg/tcpip/transport/icmp/endpoint.go
@@ -73,6 +73,9 @@ type endpoint struct {
 	route         stack.Route `state:"manual"`
 	ttl           uint8
 	stats         tcpip.TransportEndpointStats `state:"nosave"`
+
+	// owner is used to get uid and gid of the packet.
+	owner tcpip.PacketOwner
 }
 
 func newEndpoint(s *stack.Stack, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, waiterQueue *waiter.Queue) (tcpip.Endpoint, *tcpip.Error) {
@@ -133,6 +136,10 @@ func (e *endpoint) Close() {
 // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf.
 func (e *endpoint) ModerateRecvBuf(copied int) {}
 
+func (e *endpoint) SetOwner(owner tcpip.PacketOwner) {
+	e.owner = owner
+}
+
 // IPTables implements tcpip.Endpoint.IPTables.
 func (e *endpoint) IPTables() (stack.IPTables, error) {
 	return e.stack.IPTables(), nil
@@ -321,7 +328,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
 
 	switch e.NetProto {
 	case header.IPv4ProtocolNumber:
-		err = send4(route, e.ID.LocalPort, v, e.ttl)
+		err = send4(route, e.ID.LocalPort, v, e.ttl, e.owner)
 
 	case header.IPv6ProtocolNumber:
 		err = send6(route, e.ID.LocalPort, v, e.ttl)
@@ -415,7 +422,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 	}
 }
 
-func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Error {
+func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8, owner tcpip.PacketOwner) *tcpip.Error {
 	if len(data) < header.ICMPv4MinimumSize {
 		return tcpip.ErrInvalidEndpointState
 	}
@@ -444,6 +451,7 @@ func send4(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err
 		Header:          hdr,
 		Data:            data.ToVectorisedView(),
 		TransportHeader: buffer.View(icmpv4),
+		Owner:           owner,
 	})
 }
 
diff --git a/pkg/tcpip/transport/packet/endpoint.go b/pkg/tcpip/transport/packet/endpoint.go
index df49d0995..23158173d 100644
--- a/pkg/tcpip/transport/packet/endpoint.go
+++ b/pkg/tcpip/transport/packet/endpoint.go
@@ -392,3 +392,5 @@ func (ep *endpoint) Info() tcpip.EndpointInfo {
 func (ep *endpoint) Stats() tcpip.EndpointStats {
 	return &ep.stats
 }
+
+func (ep *endpoint) SetOwner(owner tcpip.PacketOwner) {}
diff --git a/pkg/tcpip/transport/raw/endpoint.go b/pkg/tcpip/transport/raw/endpoint.go
index 536dafd1e..337bc1c71 100644
--- a/pkg/tcpip/transport/raw/endpoint.go
+++ b/pkg/tcpip/transport/raw/endpoint.go
@@ -80,6 +80,9 @@ type endpoint struct {
 	// Connect(), and is valid only when conneted is true.
 	route stack.Route                  `state:"manual"`
 	stats tcpip.TransportEndpointStats `state:"nosave"`
+
+	// owner is used to get uid and gid of the packet.
+	owner tcpip.PacketOwner
 }
 
 // NewEndpoint returns a raw  endpoint for the given protocols.
@@ -159,6 +162,10 @@ func (e *endpoint) Close() {
 // ModerateRecvBuf implements tcpip.Endpoint.ModerateRecvBuf.
 func (e *endpoint) ModerateRecvBuf(copied int) {}
 
+func (e *endpoint) SetOwner(owner tcpip.PacketOwner) {
+	e.owner = owner
+}
+
 // IPTables implements tcpip.Endpoint.IPTables.
 func (e *endpoint) IPTables() (stack.IPTables, error) {
 	return e.stack.IPTables(), nil
@@ -348,10 +355,12 @@ func (e *endpoint) finishWrite(payloadBytes []byte, route *stack.Route) (int64,
 			}
 			break
 		}
+
 		hdr := buffer.NewPrependable(len(payloadBytes) + int(route.MaxHeaderLength()))
 		if err := route.WritePacket(nil /* gso */, stack.NetworkHeaderParams{Protocol: e.TransProto, TTL: route.DefaultTTL(), TOS: stack.DefaultTOS}, stack.PacketBuffer{
 			Header: hdr,
 			Data:   buffer.View(payloadBytes).ToVectorisedView(),
+			Owner:  e.owner,
 		}); err != nil {
 			return 0, nil, err
 		}
diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go
index 375ca21f6..7a9dea4ac 100644
--- a/pkg/tcpip/transport/tcp/accept.go
+++ b/pkg/tcpip/transport/tcp/accept.go
@@ -276,7 +276,7 @@ func (l *listenContext) createConnectingEndpoint(s *segment, iss seqnum.Value, i
 // and then performs the TCP 3-way handshake.
 //
 // The new endpoint is returned with e.mu held.
-func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue) (*endpoint, *tcpip.Error) {
+func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *header.TCPSynOptions, queue *waiter.Queue, owner tcpip.PacketOwner) (*endpoint, *tcpip.Error) {
 	// Create new endpoint.
 	irs := s.sequenceNumber
 	isn := generateSecureISN(s.id, l.stack.Seed())
@@ -284,6 +284,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head
 	if err != nil {
 		return nil, err
 	}
+	ep.owner = owner
 
 	// listenEP is nil when listenContext is used by tcp.Forwarder.
 	deferAccept := time.Duration(0)
@@ -414,7 +415,7 @@ func (e *endpoint) handleSynSegment(ctx *listenContext, s *segment, opts *header
 	}()
 	defer s.decRef()
 
-	n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{})
+	n, err := ctx.createEndpointAndPerformHandshake(s, opts, &waiter.Queue{}, e.owner)
 	if err != nil {
 		e.stack.Stats().TCP.FailedConnectionAttempts.Increment()
 		e.stats.FailedConnectionAttempts.Increment()
diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go
index 1d245c2c6..3239a5911 100644
--- a/pkg/tcpip/transport/tcp/connect.go
+++ b/pkg/tcpip/transport/tcp/connect.go
@@ -745,7 +745,7 @@ func (e *endpoint) sendSynTCP(r *stack.Route, tf tcpFields, opts header.TCPSynOp
 
 func (e *endpoint) sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error {
 	tf.txHash = e.txHash
-	if err := sendTCP(r, tf, data, gso); err != nil {
+	if err := sendTCP(r, tf, data, gso, e.owner); err != nil {
 		e.stats.SendErrors.SegmentSendToNetworkFailed.Increment()
 		return err
 	}
@@ -787,7 +787,7 @@ func buildTCPHdr(r *stack.Route, tf tcpFields, pkt *stack.PacketBuffer, gso *sta
 	}
 }
 
-func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error {
+func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error {
 	optLen := len(tf.opts)
 	if tf.rcvWnd > 0xffff {
 		tf.rcvWnd = 0xffff
@@ -816,6 +816,7 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso
 		pkts[i].DataSize = packetSize
 		pkts[i].Data = data
 		pkts[i].Hash = tf.txHash
+		pkts[i].Owner = owner
 		buildTCPHdr(r, tf, &pkts[i], gso)
 		off += packetSize
 		tf.seq = tf.seq.Add(seqnum.Size(packetSize))
@@ -833,14 +834,14 @@ func sendTCPBatch(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso
 
 // sendTCP sends a TCP segment with the provided options via the provided
 // network endpoint and under the provided identity.
-func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO) *tcpip.Error {
+func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stack.GSO, owner tcpip.PacketOwner) *tcpip.Error {
 	optLen := len(tf.opts)
 	if tf.rcvWnd > 0xffff {
 		tf.rcvWnd = 0xffff
 	}
 
 	if r.Loop&stack.PacketLoop == 0 && gso != nil && gso.Type == stack.GSOSW && int(gso.MSS) < data.Size() {
-		return sendTCPBatch(r, tf, data, gso)
+		return sendTCPBatch(r, tf, data, gso, owner)
 	}
 
 	pkt := stack.PacketBuffer{
@@ -849,6 +850,7 @@ func sendTCP(r *stack.Route, tf tcpFields, data buffer.VectorisedView, gso *stac
 		DataSize:   data.Size(),
 		Data:       data,
 		Hash:       tf.txHash,
+		Owner:      owner,
 	}
 	buildTCPHdr(r, tf, &pkt, gso)
 
diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go
index 1ebee0cfe..9b123e968 100644
--- a/pkg/tcpip/transport/tcp/endpoint.go
+++ b/pkg/tcpip/transport/tcp/endpoint.go
@@ -603,6 +603,9 @@ type endpoint struct {
 	// txHash is the transport layer hash to be set on outbound packets
 	// emitted by this endpoint.
 	txHash uint32
+
+	// owner is used to get uid and gid of the packet.
+	owner tcpip.PacketOwner
 }
 
 // UniqueID implements stack.TransportEndpoint.UniqueID.
@@ -1132,6 +1135,10 @@ func (e *endpoint) ModerateRecvBuf(copied int) {
 	e.rcvListMu.Unlock()
 }
 
+func (e *endpoint) SetOwner(owner tcpip.PacketOwner) {
+	e.owner = owner
+}
+
 // IPTables implements tcpip.Endpoint.IPTables.
 func (e *endpoint) IPTables() (stack.IPTables, error) {
 	return e.stack.IPTables(), nil
diff --git a/pkg/tcpip/transport/tcp/forwarder.go b/pkg/tcpip/transport/tcp/forwarder.go
index a094471b8..808410c92 100644
--- a/pkg/tcpip/transport/tcp/forwarder.go
+++ b/pkg/tcpip/transport/tcp/forwarder.go
@@ -157,7 +157,7 @@ func (r *ForwarderRequest) CreateEndpoint(queue *waiter.Queue) (tcpip.Endpoint,
 		TSVal:         r.synOptions.TSVal,
 		TSEcr:         r.synOptions.TSEcr,
 		SACKPermitted: r.synOptions.SACKPermitted,
-	}, queue)
+	}, queue, nil)
 	if err != nil {
 		return nil, err
 	}
diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go
index 1377107ca..dce9a1652 100644
--- a/pkg/tcpip/transport/tcp/protocol.go
+++ b/pkg/tcpip/transport/tcp/protocol.go
@@ -199,7 +199,7 @@ func replyWithReset(s *segment) {
 		seq:    seq,
 		ack:    ack,
 		rcvWnd: 0,
-	}, buffer.VectorisedView{}, nil /* gso */)
+	}, buffer.VectorisedView{}, nil /* gso */, nil /* PacketOwner */)
 }
 
 // SetOption implements stack.TransportProtocol.SetOption.
diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go
index a3372ac58..120d3baa3 100644
--- a/pkg/tcpip/transport/udp/endpoint.go
+++ b/pkg/tcpip/transport/udp/endpoint.go
@@ -143,6 +143,9 @@ type endpoint struct {
 
 	// TODO(b/142022063): Add ability to save and restore per endpoint stats.
 	stats tcpip.TransportEndpointStats `state:"nosave"`
+
+	// owner is used to get uid and gid of the packet.
+	owner tcpip.PacketOwner
 }
 
 // +stateify savable
@@ -484,7 +487,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c
 		useDefaultTTL = false
 	}
 
-	if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS); err != nil {
+	if err := sendUDP(route, buffer.View(v).ToVectorisedView(), e.ID.LocalPort, dstPort, ttl, useDefaultTTL, e.sendTOS, e.owner); err != nil {
 		return 0, nil, err
 	}
 	return int64(len(v)), nil, nil
@@ -886,7 +889,7 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error {
 
 // sendUDP sends a UDP segment via the provided network endpoint and under the
 // provided identity.
-func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8) *tcpip.Error {
+func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort uint16, ttl uint8, useDefaultTTL bool, tos uint8, owner tcpip.PacketOwner) *tcpip.Error {
 	// Allocate a buffer for the UDP header.
 	hdr := buffer.NewPrependable(header.UDPMinimumSize + int(r.MaxHeaderLength()))
 
@@ -916,6 +919,7 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u
 		Header:          hdr,
 		Data:            data,
 		TransportHeader: buffer.View(udp),
+		Owner:           owner,
 	}); err != nil {
 		r.Stats().UDP.PacketSendErrors.Increment()
 		return err
@@ -1356,3 +1360,7 @@ func (*endpoint) Wait() {}
 func isBroadcastOrMulticast(a tcpip.Address) bool {
 	return a == header.IPv4Broadcast || header.IsV4MulticastAddress(a) || header.IsV6MulticastAddress(a)
 }
+
+func (e *endpoint) SetOwner(owner tcpip.PacketOwner) {
+	e.owner = owner
+}