Merge pull request #1617 from kevinGC:iptables-write-filter-proto

PiperOrigin-RevId: 291249314

11 files changed, 124 insertions, 31 deletions

diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD
index 2e581e9d2..b70047d81 100644
--- a/pkg/sentry/socket/netfilter/BUILD
+++ b/pkg/sentry/socket/netfilter/BUILD
@@ -18,6 +18,7 @@ go_library(
         "//pkg/sentry/kernel",
         "//pkg/sentry/usermem",
         "//pkg/syserr",
+        "//pkg/tcpip",
         "//pkg/tcpip/iptables",
         "//pkg/tcpip/stack",
     ],
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index 507a77483..c65c36081 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -25,6 +25,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserr"
+	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/iptables"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
 )
@@ -163,6 +164,9 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern
 		// Each rule corresponds to an entry.
 		entry := linux.KernelIPTEntry{
 			IPTEntry: linux.IPTEntry{
+				IP: linux.IPTIP{
+					Protocol: uint16(rule.Filter.Protocol),
+				},
 				NextOffset:   linux.SizeOfIPTEntry,
 				TargetOffset: linux.SizeOfIPTEntry,
 			},
@@ -321,12 +325,11 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 			return syserr.ErrInvalidArgument
 		}
 
-		// TODO(gvisor.dev/issue/170): We should support IPTIP
-		// filtering. We reject any nonzero IPTIP values for now.
-		emptyIPTIP := linux.IPTIP{}
-		if entry.IP != emptyIPTIP {
-			log.Warningf("netfilter: non-empty struct iptip found")
-			return syserr.ErrInvalidArgument
+		// TODO(gvisor.dev/issue/170): We should support more IPTIP
+		// filtering fields.
+		filter, err := filterFromIPTIP(entry.IP)
+		if err != nil {
+			return err
 		}
 
 		// Get the target of the rule.
@@ -336,7 +339,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 		}
 		optVal = optVal[consumed:]
 
-		table.Rules = append(table.Rules, iptables.Rule{Target: target})
+		table.Rules = append(table.Rules, iptables.Rule{
+			Filter: filter,
+			Target: target,
+		})
 		offsets = append(offsets, offset)
 		offset += linux.SizeOfIPTEntry + consumed
 	}
@@ -460,6 +466,32 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) {
 	return nil, 0, syserr.ErrInvalidArgument
 }
 
+func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) {
+	if containsUnsupportedFields(iptip) {
+		log.Warningf("netfilter: unsupported fields in struct iptip: %+v", iptip)
+		return iptables.IPHeaderFilter{}, syserr.ErrInvalidArgument
+	}
+	return iptables.IPHeaderFilter{
+		Protocol: tcpip.TransportProtocolNumber(iptip.Protocol),
+	}, nil
+}
+
+func containsUnsupportedFields(iptip linux.IPTIP) bool {
+	// Currently we check that everything except protocol is zeroed.
+	var emptyInetAddr = linux.InetAddr{}
+	var emptyInterface = [linux.IFNAMSIZ]byte{}
+	return iptip.Dst != emptyInetAddr ||
+		iptip.Src != emptyInetAddr ||
+		iptip.SrcMask != emptyInetAddr ||
+		iptip.DstMask != emptyInetAddr ||
+		iptip.InputInterface != emptyInterface ||
+		iptip.OutputInterface != emptyInterface ||
+		iptip.InputInterfaceMask != emptyInterface ||
+		iptip.OutputInterfaceMask != emptyInterface ||
+		iptip.Flags != 0 ||
+		iptip.InverseFlags != 0
+}
+
 func hookFromLinux(hook int) iptables.Hook {
 	switch hook {
 	case linux.NF_INET_PRE_ROUTING:
diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD
index 2893c80cd..297eaccaf 100644
--- a/pkg/tcpip/iptables/BUILD
+++ b/pkg/tcpip/iptables/BUILD
@@ -14,5 +14,6 @@ go_library(
     deps = [
         "//pkg/log",
         "//pkg/tcpip",
+        "//pkg/tcpip/header",
     ],
 )
diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go
index 605a71679..fc06b5b87 100644
--- a/pkg/tcpip/iptables/iptables.go
+++ b/pkg/tcpip/iptables/iptables.go
@@ -20,6 +20,7 @@ import (
 	"fmt"
 
 	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/header"
 )
 
 // Table names.
@@ -184,8 +185,16 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri
 	panic(fmt.Sprintf("Traversed past the entire list of iptables rules in table %q.", tablename))
 }
 
+// Precondition: pk.NetworkHeader is set.
 func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict {
 	rule := table.Rules[ruleIdx]
+
+	// First check whether the packet matches the IP header filter.
+	// TODO(gvisor.dev/issue/170): Support other fields of the filter.
+	if rule.Filter.Protocol != 0 && rule.Filter.Protocol != header.IPv4(pkt.NetworkHeader).TransportProtocol() {
+		return Continue
+	}
+
 	// Go through each rule matcher. If they all match, run
 	// the rule target.
 	for _, matcher := range rule.Matchers {
diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go
index 9f6906100..a8b972f1b 100644
--- a/pkg/tcpip/iptables/types.go
+++ b/pkg/tcpip/iptables/types.go
@@ -14,7 +14,9 @@
 
 package iptables
 
-import "gvisor.dev/gvisor/pkg/tcpip"
+import (
+	"gvisor.dev/gvisor/pkg/tcpip"
+)
 
 // A Hook specifies one of the hooks built into the network stack.
 //
@@ -151,6 +153,9 @@ func (table *Table) SetMetadata(metadata interface{}) {
 // packets this rule applies to. If there are no matchers in the rule, it
 // applies to any packet.
 type Rule struct {
+	// Filter holds basic IP filtering fields common to every rule.
+	Filter IPHeaderFilter
+
 	// Matchers is the list of matchers for this rule.
 	Matchers []Matcher
 
@@ -158,6 +163,12 @@ type Rule struct {
 	Target Target
 }
 
+// IPHeaderFilter holds basic IP filtering data common to every rule.
+type IPHeaderFilter struct {
+	// Protocol matches the transport protocol.
+	Protocol tcpip.TransportProtocolNumber
+}
+
 // A Matcher is the interface for matching packets.
 type Matcher interface {
 	// Match returns whether the packet matches and whether the packet
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 0a1453b31..85512f9b2 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -353,7 +353,8 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) {
 	}
 	pkt.NetworkHeader = headerView[:h.HeaderLength()]
 
-	// iptables filtering.
+	// iptables filtering. All packets that reach here are intended for
+	// this machine and will not be forwarded.
 	ipt := e.stack.IPTables()
 	if ok := ipt.Check(iptables.Input, pkt); !ok {
 		// iptables is telling us to drop the packet.
diff --git a/test/iptables/BUILD b/test/iptables/BUILD
index 372ba7abf..22f470092 100644
--- a/test/iptables/BUILD
+++ b/test/iptables/BUILD
@@ -4,6 +4,7 @@ package(licenses = ["notice"])
 
 go_library(
     name = "iptables",
+    testonly = 1,
     srcs = [
         "filter_input.go",
         "filter_output.go",
@@ -13,6 +14,9 @@ go_library(
     ],
     importpath = "gvisor.dev/gvisor/test/iptables",
     visibility = ["//test/iptables:__subpackages__"],
+    deps = [
+        "//runsc/testutil",
+    ],
 )
 
 go_test(
diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go
index 03e4a1d72..fd02ff2ff 100644
--- a/test/iptables/filter_input.go
+++ b/test/iptables/filter_input.go
@@ -30,6 +30,7 @@ const (
 func init() {
 	RegisterTestCase(FilterInputDropAll{})
 	RegisterTestCase(FilterInputDropDifferentUDPPort{})
+	RegisterTestCase(FilterInputDropOnlyUDP{})
 	RegisterTestCase(FilterInputDropTCPDestPort{})
 	RegisterTestCase(FilterInputDropTCPSrcPort{})
 	RegisterTestCase(FilterInputDropUDPPort{})
@@ -67,6 +68,35 @@ func (FilterInputDropUDP) LocalAction(ip net.IP) error {
 	return sendUDPLoop(ip, dropPort, sendloopDuration)
 }
 
+// FilterInputDropOnlyUDP tests that "-p udp -j DROP" only affects UDP traffic.
+type FilterInputDropOnlyUDP struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputDropOnlyUDP) Name() string {
+	return "FilterInputDropOnlyUDP"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputDropOnlyUDP) ContainerAction(ip net.IP) error {
+	if err := filterTable("-A", "INPUT", "-p", "udp", "-j", "DROP"); err != nil {
+		return err
+	}
+
+	// Listen for a TCP connection, which should be allowed.
+	if err := listenTCP(acceptPort, sendloopDuration); err != nil {
+		return fmt.Errorf("failed to establish a connection %v", err)
+	}
+
+	return nil
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputDropOnlyUDP) LocalAction(ip net.IP) error {
+	// Try to establish a TCP connection with the container, which should
+	// succeed.
+	return connectTCP(ip, acceptPort, dropPort, sendloopDuration)
+}
+
 // FilterInputDropUDPPort tests that we can drop UDP traffic by port.
 type FilterInputDropUDPPort struct{}
 
diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go
index 1cda10365..679a29bef 100644
--- a/test/iptables/iptables_test.go
+++ b/test/iptables/iptables_test.go
@@ -15,6 +15,7 @@
 package iptables
 
 import (
+	"flag"
 	"fmt"
 	"net"
 	"os"
@@ -22,7 +23,6 @@ import (
 	"testing"
 	"time"
 
-	"flag"
 	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/runsc/dockerutil"
 	"gvisor.dev/gvisor/runsc/testutil"
@@ -184,6 +184,12 @@ func TestFilterInputDropAll(t *testing.T) {
 	}
 }
 
+func TestFilterInputDropOnlyUDP(t *testing.T) {
+	if err := singleTest(FilterInputDropOnlyUDP{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
 func TestNATRedirectUDPPort(t *testing.T) {
 	if err := singleTest(NATRedirectUDPPort{}); err != nil {
 		t.Fatal(err)
diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go
index 1c4f4f665..043114c78 100644
--- a/test/iptables/iptables_util.go
+++ b/test/iptables/iptables_util.go
@@ -19,6 +19,8 @@ import (
 	"net"
 	"os/exec"
 	"time"
+
+	"gvisor.dev/gvisor/runsc/testutil"
 )
 
 const iptablesBinary = "iptables"
@@ -105,31 +107,26 @@ func listenTCP(port int, timeout time.Duration) error {
 }
 
 // connectTCP connects the TCP server over specified local port, server IP and remote/server port.
-func connectTCP(ip net.IP, remotePort, localPort int, duration time.Duration) error {
-	remote := net.TCPAddr{
+func connectTCP(ip net.IP, remotePort, localPort int, timeout time.Duration) error {
+	contAddr := net.TCPAddr{
 		IP:   ip,
 		Port: remotePort,
 	}
-
-	local := net.TCPAddr{
-		Port: localPort,
-	}
-
-	// Container may not be up. Retry DialTCP over a duration.
-	to := time.After(duration)
-	for {
-		conn, err := net.DialTCP("tcp4", &local, &remote)
-		if err == nil {
-			conn.Close()
-			return nil
+	// The container may not be listening when we first connect, so retry
+	// upon error.
+	callback := func() error {
+		localAddr := net.TCPAddr{
+			Port: localPort,
 		}
-		select {
-		// Timed out waiting for connection to be accepted.
-		case <-to:
-			return err
-		default:
-			time.Sleep(200 * time.Millisecond)
+		conn, err := net.DialTCP("tcp4", &localAddr, &contAddr)
+		if conn != nil {
+			conn.Close()
 		}
+		return err
 	}
-	return fmt.Errorf("Failed to establish connection on port %d", localPort)
+	if err := testutil.Poll(callback, timeout); err != nil {
+		return fmt.Errorf("timed out waiting to send IP, most recent error: %v", err)
+	}
+
+	return nil
 }
diff --git a/test/iptables/runner/BUILD b/test/iptables/runner/BUILD
index c6c42d870..a5b6f082c 100644
--- a/test/iptables/runner/BUILD
+++ b/test/iptables/runner/BUILD
@@ -10,6 +10,7 @@ container_image(
 
 go_image(
     name = "runner",
+    testonly = 1,
     srcs = ["main.go"],
     base = ":iptables-base",
     deps = ["//test/iptables"],