We can now create and jump in iptables. For example:

$ iptables -N foochain
$ iptables -A INPUT -j foochain

10 files changed, 420 insertions, 94 deletions

diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go
index bbc4df74c..bd2e13ba1 100644
--- a/pkg/abi/linux/netfilter.go
+++ b/pkg/abi/linux/netfilter.go
@@ -225,11 +225,14 @@ type XTEntryTarget struct {
 // SizeOfXTEntryTarget is the size of an XTEntryTarget.
 const SizeOfXTEntryTarget = 32
 
-// XTStandardTarget is a builtin target, one of ACCEPT, DROP, JUMP, QUEUE, or
-// RETURN. It corresponds to struct xt_standard_target in
+// XTStandardTarget is a built-in target, one of ACCEPT, DROP, JUMP, QUEUE,
+// RETURN, or jump. It corresponds to struct xt_standard_target in
 // include/uapi/linux/netfilter/x_tables.h.
 type XTStandardTarget struct {
-	Target  XTEntryTarget
+	Target XTEntryTarget
+	// A positive verdict indicates a jump, and is the offset from the
+	// start of the table to jump to. A negative value means one of the
+	// other built-in targets.
 	Verdict int32
 	_       [4]byte
 }
diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD
index c91ec7494..7cd2ce55b 100644
--- a/pkg/sentry/socket/netfilter/BUILD
+++ b/pkg/sentry/socket/netfilter/BUILD
@@ -7,6 +7,7 @@ go_library(
     srcs = [
         "extensions.go",
         "netfilter.go",
+        "targets.go",
         "tcp_matcher.go",
         "udp_matcher.go",
     ],
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index 3fc80e0de..d322e4144 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -240,13 +240,15 @@ func marshalTarget(target iptables.Target) []byte {
 		return marshalErrorTarget(tg.Name)
 	case iptables.ReturnTarget:
 		return marshalStandardTarget(iptables.RuleReturn)
+	case JumpTarget:
+		return marshalJumpTarget(tg)
 	default:
 		panic(fmt.Errorf("unknown target of type %T", target))
 	}
 }
 
 func marshalStandardTarget(verdict iptables.RuleVerdict) []byte {
-	nflog("convert to binary: marshalling standard target with size %d", linux.SizeOfXTStandardTarget)
+	nflog("convert to binary: marshalling standard target")
 
 	// The target's name will be the empty string.
 	target := linux.XTStandardTarget{
@@ -274,6 +276,23 @@ func marshalErrorTarget(errorName string) []byte {
 	return binary.Marshal(ret, usermem.ByteOrder, target)
 }
 
+func marshalJumpTarget(jt JumpTarget) []byte {
+	nflog("convert to binary: marshalling jump target")
+
+	// The target's name will be the empty string.
+	target := linux.XTStandardTarget{
+		Target: linux.XTEntryTarget{
+			TargetSize: linux.SizeOfXTStandardTarget,
+		},
+		// Verdict is overloaded by the ABI. When positive, it holds
+		// the jump offset from the start of the table.
+		Verdict: int32(jt.Offset),
+	}
+
+	ret := make([]byte, 0, linux.SizeOfXTStandardTarget)
+	return binary.Marshal(ret, usermem.ByteOrder, target)
+}
+
 // translateFromStandardVerdict translates verdicts the same way as the iptables
 // tool.
 func translateFromStandardVerdict(verdict iptables.RuleVerdict) int32 {
@@ -335,7 +354,8 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 
 	// Convert input into a list of rules and their offsets.
 	var offset uint32
-	var offsets []uint32
+	// offsets maps rule byte offsets to their position in table.Rules.
+	offsets := map[uint32]int{}
 	for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ {
 		nflog("set entries: processing entry at offset %d", offset)
 
@@ -396,11 +416,12 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 			Target:   target,
 			Matchers: matchers,
 		})
-		offsets = append(offsets, offset)
+		offsets[offset] = int(entryIdx)
 		offset += uint32(entry.NextOffset)
 
 		if initialOptValLen-len(optVal) != int(entry.NextOffset) {
 			nflog("entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal))
+			return syserr.ErrInvalidArgument
 		}
 	}
 
@@ -409,13 +430,13 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 	for hook, _ := range replace.HookEntry {
 		if table.ValidHooks()&(1<<hook) != 0 {
 			hk := hookFromLinux(hook)
-			for ruleIdx, offset := range offsets {
+			for offset, ruleIdx := range offsets {
 				if offset == replace.HookEntry[hook] {
 					table.BuiltinChains[hk] = ruleIdx
 				}
 				if offset == replace.Underflow[hook] {
 					if !validUnderflow(table.Rules[ruleIdx]) {
-						nflog("underflow for hook %d isn't an unconditional ACCEPT or DROP.")
+						nflog("underflow for hook %d isn't an unconditional ACCEPT or DROP")
 						return syserr.ErrInvalidArgument
 					}
 					table.Underflows[hk] = ruleIdx
@@ -444,16 +465,35 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 		// - There's some other rule after it.
 		// - There are no matchers.
 		if ruleIdx == len(table.Rules)-1 {
-			nflog("user chain must have a rule or default policy.")
+			nflog("user chain must have a rule or default policy")
 			return syserr.ErrInvalidArgument
 		}
 		if len(table.Rules[ruleIdx].Matchers) != 0 {
-			nflog("user chain's first node must have no matcheres.")
+			nflog("user chain's first node must have no matchers")
 			return syserr.ErrInvalidArgument
 		}
 		table.UserChains[target.Name] = ruleIdx + 1
 	}
 
+	// Set each jump to point to the appropriate rule. Right now they hold byte
+	// offsets.
+	for ruleIdx, rule := range table.Rules {
+		jump, ok := rule.Target.(JumpTarget)
+		if !ok {
+			continue
+		}
+
+		// Find the rule corresponding to the jump rule offset.
+		jumpTo, ok := offsets[jump.Offset]
+		if !ok {
+			nflog("failed to find a rule to jump to")
+			return syserr.ErrInvalidArgument
+		}
+		jump.RuleNum = jumpTo
+		rule.Target = jump
+		table.Rules[ruleIdx] = rule
+	}
+
 	// TODO(gvisor.dev/issue/170): Support other chains.
 	// Since we only support modifying the INPUT chain right now, make sure
 	// all other chains point to ACCEPT rules.
@@ -548,7 +588,13 @@ func parseTarget(optVal []byte) (iptables.Target, error) {
 		buf = optVal[:linux.SizeOfXTStandardTarget]
 		binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget)
 
-		return translateToStandardTarget(standardTarget.Verdict)
+		if standardTarget.Verdict < 0 {
+			// A Verdict < 0 indicates a non-jump verdict.
+			return translateToStandardTarget(standardTarget.Verdict)
+		} else {
+			// A verdict >= 0 indicates a jump.
+			return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil
+		}
 
 	case errorTargetName:
 		// Error target.
diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go
new file mode 100644
index 000000000..c421b87cf
--- /dev/null
+++ b/pkg/sentry/socket/netfilter/targets.go
@@ -0,0 +1,35 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package netfilter
+
+import (
+	"gvisor.dev/gvisor/pkg/tcpip"
+	"gvisor.dev/gvisor/pkg/tcpip/iptables"
+)
+
+// JumpTarget implements iptables.Target.
+type JumpTarget struct {
+	// Offset is the byte offset of the rule to jump to. It is used for
+	// marshaling and unmarshaling.
+	Offset uint32
+
+	// RuleNum is the rule to jump to.
+	RuleNum int
+}
+
+// Action implements iptables.Target.Action.
+func (jt JumpTarget) Action(tcpip.PacketBuffer) (iptables.RuleVerdict, int) {
+	return iptables.RuleJump, jt.RuleNum
+}
diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go
index 75a433a3b..dbaccbb36 100644
--- a/pkg/tcpip/iptables/iptables.go
+++ b/pkg/tcpip/iptables/iptables.go
@@ -135,25 +135,53 @@ func EmptyFilterTable() Table {
 	}
 }
 
+// A chainVerdict is what a table decides should be done with a packet.
+type chainVerdict int
+
+const (
+	// chainAccept indicates the packet should continue through netstack.
+	chainAccept chainVerdict = iota
+
+	// chainAccept indicates the packet should be dropped.
+	chainDrop
+
+	// chainReturn indicates the packet should return to the calling chain
+	// or the underflow rule of a builtin chain.
+	chainReturn
+)
+
 // Check runs pkt through the rules for hook. It returns true when the packet
 // should continue traversing the network stack and false when it should be
 // dropped.
 //
 // Precondition: pkt.NetworkHeader is set.
 func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool {
-	// TODO(gvisor.dev/issue/170): A lot of this is uncomplicated because
-	// we're missing features. Jumps, the call stack, etc. aren't checked
-	// for yet because we're yet to support them.
-
 	// Go through each table containing the hook.
 	for _, tablename := range it.Priorities[hook] {
-		switch verdict := it.checkTable(hook, pkt, tablename); verdict {
+		table := it.Tables[tablename]
+		ruleIdx := table.BuiltinChains[hook]
+		switch verdict := it.checkChain(hook, pkt, table, ruleIdx); verdict {
 		// If the table returns Accept, move on to the next table.
-		case TableAccept:
+		case chainAccept:
 			continue
 		// The Drop verdict is final.
-		case TableDrop:
+		case chainDrop:
 			return false
+		case chainReturn:
+			// Any Return from a built-in chain means we have to
+			// call the underflow.
+			underflow := table.Rules[table.Underflows[hook]]
+			switch v, _ := underflow.Target.Action(pkt); v {
+			case RuleAccept:
+				continue
+			case RuleDrop:
+				return false
+			case RuleJump, RuleReturn:
+				panic("Underflows should only return RuleAccept or RuleDrop.")
+			default:
+				panic(fmt.Sprintf("Unknown verdict: %d", v))
+			}
+
 		default:
 			panic(fmt.Sprintf("Unknown verdict %v.", verdict))
 		}
@@ -164,37 +192,37 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool {
 }
 
 // Precondition: pkt.NetworkHeader is set.
-func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) TableVerdict {
+func (it *IPTables) checkChain(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) chainVerdict {
 	// Start from ruleIdx and walk the list of rules until a rule gives us
 	// a verdict.
-	table := it.Tables[tablename]
-	for ruleIdx := table.BuiltinChains[hook]; ruleIdx < len(table.Rules); ruleIdx++ {
-		switch verdict := it.checkRule(hook, pkt, table, ruleIdx); verdict {
+	for ruleIdx < len(table.Rules) {
+		switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx); verdict {
 		case RuleAccept:
-			return TableAccept
+			return chainAccept
 
 		case RuleDrop:
-			return TableDrop
-
-		case RuleContinue:
-			continue
+			return chainDrop
 
 		case RuleReturn:
-			// TODO(gvisor.dev/issue/170): We don't implement jump
-			// yet, so any Return is from a built-in chain. That
-			// means we have to to call the underflow.
-			underflow := table.Rules[table.Underflows[hook]]
-			// Underflow is guaranteed to be an unconditional
-			// ACCEPT or DROP.
-			switch v, _ := underflow.Target.Action(pkt); v {
-			case RuleAccept:
-				return TableAccept
-			case RuleDrop:
-				return TableDrop
-			case RuleContinue, RuleReturn:
-				panic("Underflows should only return RuleAccept or RuleDrop.")
+			return chainReturn
+
+		case RuleJump:
+			// "Jumping" to the next rule just means we're
+			// continuing on down the list.
+			if jumpTo == ruleIdx+1 {
+				ruleIdx++
+				continue
+			}
+			switch verdict := it.checkChain(hook, pkt, table, jumpTo); verdict {
+			case chainAccept:
+				return chainAccept
+			case chainDrop:
+				return chainDrop
+			case chainReturn:
+				ruleIdx++
+				continue
 			default:
-				panic(fmt.Sprintf("Unknown verdict: %d", v))
+				panic(fmt.Sprintf("Unknown verdict: %d", verdict))
 			}
 
 		default:
@@ -205,17 +233,18 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri
 
 	// We got through the entire table without a decision. Default to DROP
 	// for safety.
-	return TableDrop
+	return chainDrop
 }
 
 // Precondition: pk.NetworkHeader is set.
-func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) RuleVerdict {
+func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) (RuleVerdict, int) {
 	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 RuleContinue
+		// Continue on to the next rule.
+		return RuleJump, ruleIdx + 1
 	}
 
 	// Go through each rule matcher. If they all match, run
@@ -223,14 +252,14 @@ func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ru
 	for _, matcher := range rule.Matchers {
 		matches, hotdrop := matcher.Match(hook, pkt, "")
 		if hotdrop {
-			return RuleDrop
+			return RuleDrop, 0
 		}
 		if !matches {
-			return RuleContinue
+			// Continue on to the next rule.
+			return RuleJump, ruleIdx + 1
 		}
 	}
 
 	// All the matchers matched, so run the target.
-	verdict, _ := rule.Target.Action(pkt)
-	return verdict
+	return rule.Target.Action(pkt)
 }
diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go
index 9fc60cfad..81a2e39a2 100644
--- a/pkg/tcpip/iptables/targets.go
+++ b/pkg/tcpip/iptables/targets.go
@@ -12,8 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// This file contains various Targets.
-
 package iptables
 
 import (
@@ -25,16 +23,16 @@ import (
 type AcceptTarget struct{}
 
 // Action implements Target.Action.
-func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) {
-	return RuleAccept, ""
+func (AcceptTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) {
+	return RuleAccept, 0
 }
 
 // DropTarget drops packets.
 type DropTarget struct{}
 
 // Action implements Target.Action.
-func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) {
-	return RuleDrop, ""
+func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) {
+	return RuleDrop, 0
 }
 
 // ErrorTarget logs an error and drops the packet. It represents a target that
@@ -42,9 +40,9 @@ func (DropTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) {
 type ErrorTarget struct{}
 
 // Action implements Target.Action.
-func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, string) {
+func (ErrorTarget) Action(packet tcpip.PacketBuffer) (RuleVerdict, int) {
 	log.Debugf("ErrorTarget triggered.")
-	return RuleDrop, ""
+	return RuleDrop, 0
 }
 
 // UserChainTarget marks a rule as the beginning of a user chain.
@@ -53,7 +51,7 @@ type UserChainTarget struct {
 }
 
 // Action implements Target.Action.
-func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) {
+func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) {
 	panic("UserChainTarget should never be called.")
 }
 
@@ -62,6 +60,6 @@ func (UserChainTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) {
 type ReturnTarget struct{}
 
 // Action implements Target.Action.
-func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, string) {
-	return RuleReturn, ""
+func (ReturnTarget) Action(tcpip.PacketBuffer) (RuleVerdict, int) {
+	return RuleReturn, 0
 }
diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go
index 5735d001b..7d032fd23 100644
--- a/pkg/tcpip/iptables/types.go
+++ b/pkg/tcpip/iptables/types.go
@@ -56,17 +56,6 @@ const (
 	NumHooks
 )
 
-// A TableVerdict is what a table decides should be done with a packet.
-type TableVerdict int
-
-const (
-	// TableAccept indicates the packet should continue through netstack.
-	TableAccept TableVerdict = iota
-
-	// TableAccept indicates the packet should be dropped.
-	TableDrop
-)
-
 // A RuleVerdict is what a rule decides should be done with a packet.
 type RuleVerdict int
 
@@ -74,12 +63,12 @@ const (
 	// RuleAccept indicates the packet should continue through netstack.
 	RuleAccept RuleVerdict = iota
 
-	// RuleContinue indicates the packet should continue to the next rule.
-	RuleContinue
-
 	// RuleDrop indicates the packet should be dropped.
 	RuleDrop
 
+	// RuleJump indicates the packet should jump to another chain.
+	RuleJump
+
 	// RuleReturn indicates the packet should return to the previous chain.
 	RuleReturn
 )
@@ -174,6 +163,6 @@ type Matcher interface {
 type Target interface {
 	// Action takes an action on the packet and returns a verdict on how
 	// traversal should (or should not) continue. If the return value is
-	// Jump, it also returns the name of the chain to jump to.
-	Action(packet tcpip.PacketBuffer) (RuleVerdict, string)
+	// Jump, it also returns the index of the rule to jump to.
+	Action(packet tcpip.PacketBuffer) (RuleVerdict, int)
 }
diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go
index e26d6a7d2..706c09cea 100644
--- a/test/iptables/filter_input.go
+++ b/test/iptables/filter_input.go
@@ -26,6 +26,7 @@ const (
 	acceptPort       = 2402
 	sendloopDuration = 2 * time.Second
 	network          = "udp4"
+	chainName        = "foochain"
 )
 
 func init() {
@@ -40,6 +41,12 @@ func init() {
 	RegisterTestCase(FilterInputDefaultPolicyAccept{})
 	RegisterTestCase(FilterInputDefaultPolicyDrop{})
 	RegisterTestCase(FilterInputReturnUnderflow{})
+	RegisterTestCase(FilterInputSerializeJump{})
+	RegisterTestCase(FilterInputJumpBasic{})
+	RegisterTestCase(FilterInputJumpReturn{})
+	RegisterTestCase(FilterInputJumpReturnDrop{})
+	RegisterTestCase(FilterInputJumpBuiltin{})
+	RegisterTestCase(FilterInputJumpTwice{})
 }
 
 // FilterInputDropUDP tests that we can drop UDP traffic.
@@ -267,13 +274,12 @@ func (FilterInputMultiUDPRules) Name() string {
 
 // ContainerAction implements TestCase.ContainerAction.
 func (FilterInputMultiUDPRules) ContainerAction(ip net.IP) error {
-	if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"); err != nil {
-		return err
-	}
-	if err := filterTable("-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"); err != nil {
-		return err
+	rules := [][]string{
+		{"-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", dropPort), "-j", "DROP"},
+		{"-A", "INPUT", "-p", "udp", "-m", "udp", "--destination-port", fmt.Sprintf("%d", acceptPort), "-j", "ACCEPT"},
+		{"-L"},
 	}
-	return filterTable("-L")
+	return filterTableRules(rules)
 }
 
 // LocalAction implements TestCase.LocalAction.
@@ -314,14 +320,13 @@ func (FilterInputCreateUserChain) Name() string {
 
 // ContainerAction implements TestCase.ContainerAction.
 func (FilterInputCreateUserChain) ContainerAction(ip net.IP) error {
-	// Create a chain.
-	const chainName = "foochain"
-	if err := filterTable("-N", chainName); err != nil {
-		return err
+	rules := [][]string{
+		// Create a chain.
+		{"-N", chainName},
+		// Add a simple rule to the chain.
+		{"-A", chainName, "-j", "DROP"},
 	}
-
-	// Add a simple rule to the chain.
-	return filterTable("-A", chainName, "-j", "DROP")
+	return filterTableRules(rules)
 }
 
 // LocalAction implements TestCase.LocalAction.
@@ -396,13 +401,12 @@ func (FilterInputReturnUnderflow) Name() string {
 func (FilterInputReturnUnderflow) ContainerAction(ip net.IP) error {
 	// Add a RETURN rule followed by an unconditional accept, and set the
 	// default policy to DROP.
-	if err := filterTable("-A", "INPUT", "-j", "RETURN"); err != nil {
-		return err
+	rules := [][]string{
+		{"-A", "INPUT", "-j", "RETURN"},
+		{"-A", "INPUT", "-j", "DROP"},
+		{"-P", "INPUT", "ACCEPT"},
 	}
-	if err := filterTable("-A", "INPUT", "-j", "DROP"); err != nil {
-		return err
-	}
-	if err := filterTable("-P", "INPUT", "ACCEPT"); err != nil {
+	if err := filterTableRules(rules); err != nil {
 		return err
 	}
 
@@ -415,3 +419,178 @@ func (FilterInputReturnUnderflow) ContainerAction(ip net.IP) error {
 func (FilterInputReturnUnderflow) LocalAction(ip net.IP) error {
 	return sendUDPLoop(ip, acceptPort, sendloopDuration)
 }
+
+// Verify that we can serialize jumps.
+type FilterInputSerializeJump struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputSerializeJump) Name() string {
+	return "FilterInputSerializeJump"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputSerializeJump) ContainerAction(ip net.IP) error {
+	// Write a JUMP rule, the serialize it with `-L`.
+	rules := [][]string{
+		{"-N", chainName},
+		{"-A", "INPUT", "-j", chainName},
+		{"-L"},
+	}
+	return filterTableRules(rules)
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputSerializeJump) LocalAction(ip net.IP) error {
+	// No-op.
+	return nil
+}
+
+// Jump to a chain and execute a rule there.
+type FilterInputJumpBasic struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputJumpBasic) Name() string {
+	return "FilterInputJumpBasic"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputJumpBasic) ContainerAction(ip net.IP) error {
+	rules := [][]string{
+		{"-P", "INPUT", "DROP"},
+		{"-N", chainName},
+		{"-A", "INPUT", "-j", chainName},
+		{"-A", chainName, "-j", "ACCEPT"},
+	}
+	if err := filterTableRules(rules); err != nil {
+		return err
+	}
+
+	// Listen for UDP packets on acceptPort.
+	return listenUDP(acceptPort, sendloopDuration)
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputJumpBasic) LocalAction(ip net.IP) error {
+	return sendUDPLoop(ip, acceptPort, sendloopDuration)
+}
+
+// Jump, return, and execute a rule.
+type FilterInputJumpReturn struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputJumpReturn) Name() string {
+	return "FilterInputJumpReturn"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputJumpReturn) ContainerAction(ip net.IP) error {
+	rules := [][]string{
+		{"-N", chainName},
+		{"-P", "INPUT", "ACCEPT"},
+		{"-A", "INPUT", "-j", chainName},
+		{"-A", chainName, "-j", "RETURN"},
+		{"-A", chainName, "-j", "DROP"},
+	}
+	if err := filterTableRules(rules); err != nil {
+		return err
+	}
+
+	// Listen for UDP packets on acceptPort.
+	return listenUDP(acceptPort, sendloopDuration)
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputJumpReturn) LocalAction(ip net.IP) error {
+	return sendUDPLoop(ip, acceptPort, sendloopDuration)
+}
+
+type FilterInputJumpReturnDrop struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputJumpReturnDrop) Name() string {
+	return "FilterInputJumpReturnDrop"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputJumpReturnDrop) ContainerAction(ip net.IP) error {
+	rules := [][]string{
+		{"-N", chainName},
+		{"-A", "INPUT", "-j", chainName},
+		{"-A", "INPUT", "-j", "DROP"},
+		{"-A", chainName, "-j", "RETURN"},
+	}
+	if err := filterTableRules(rules); err != nil {
+		return err
+	}
+
+	// Listen for UDP packets on dropPort.
+	if err := listenUDP(dropPort, sendloopDuration); err == nil {
+		return fmt.Errorf("packets on port %d should have been dropped, but got a packet", dropPort)
+	} else if netErr, ok := err.(net.Error); !ok || !netErr.Timeout() {
+		return fmt.Errorf("error reading: %v", err)
+	}
+
+	// At this point we know that reading timed out and never received a
+	// packet.
+	return nil
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputJumpReturnDrop) LocalAction(ip net.IP) error {
+	return sendUDPLoop(ip, dropPort, sendloopDuration)
+}
+
+// Jumping to a top-levl chain is illegal.
+type FilterInputJumpBuiltin struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputJumpBuiltin) Name() string {
+	return "FilterInputJumpBuiltin"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputJumpBuiltin) ContainerAction(ip net.IP) error {
+	if err := filterTable("-A", "INPUT", "-j", "OUTPUT"); err == nil {
+		return fmt.Errorf("iptables should be unable to jump to a built-in chain")
+	}
+	return nil
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputJumpBuiltin) LocalAction(ip net.IP) error {
+	// No-op.
+	return nil
+}
+
+// Jump twice, then return twice and execute a rule.
+type FilterInputJumpTwice struct{}
+
+// Name implements TestCase.Name.
+func (FilterInputJumpTwice) Name() string {
+	return "FilterInputJumpTwice"
+}
+
+// ContainerAction implements TestCase.ContainerAction.
+func (FilterInputJumpTwice) ContainerAction(ip net.IP) error {
+	const chainName2 = chainName + "2"
+	rules := [][]string{
+		{"-P", "INPUT", "DROP"},
+		{"-N", chainName},
+		{"-N", chainName2},
+		{"-A", "INPUT", "-j", chainName},
+		{"-A", chainName, "-j", chainName2},
+		{"-A", "INPUT", "-j", "ACCEPT"},
+	}
+	if err := filterTableRules(rules); err != nil {
+		return err
+	}
+
+	// UDP packets should jump and return twice, eventually hitting the
+	// ACCEPT rule.
+	return listenUDP(acceptPort, sendloopDuration)
+}
+
+// LocalAction implements TestCase.LocalAction.
+func (FilterInputJumpTwice) LocalAction(ip net.IP) error {
+	return sendUDPLoop(ip, acceptPort, sendloopDuration)
+}
diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go
index 46a7c99b0..0621861eb 100644
--- a/test/iptables/iptables_test.go
+++ b/test/iptables/iptables_test.go
@@ -249,3 +249,39 @@ func TestFilterOutputDropTCPSrcPort(t *testing.T) {
 		t.Fatal(err)
 	}
 }
+
+func TestJumpSerialize(t *testing.T) {
+	if err := singleTest(FilterInputSerializeJump{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestJumpBasic(t *testing.T) {
+	if err := singleTest(FilterInputJumpBasic{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestJumpReturn(t *testing.T) {
+	if err := singleTest(FilterInputJumpReturn{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestJumpReturnDrop(t *testing.T) {
+	if err := singleTest(FilterInputJumpReturnDrop{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestJumpBuiltin(t *testing.T) {
+	if err := singleTest(FilterInputJumpBuiltin{}); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestJumpTwice(t *testing.T) {
+	if err := singleTest(FilterInputJumpTwice{}); err != nil {
+		t.Fatal(err)
+	}
+}
diff --git a/test/iptables/iptables_util.go b/test/iptables/iptables_util.go
index 043114c78..293c4e6ed 100644
--- a/test/iptables/iptables_util.go
+++ b/test/iptables/iptables_util.go
@@ -35,6 +35,16 @@ func filterTable(args ...string) error {
 	return nil
 }
 
+// filterTableRules is like filterTable, but runs multiple iptables commands.
+func filterTableRules(argsList [][]string) error {
+	for _, args := range argsList {
+		if err := filterTable(args...); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
 // listenUDP listens on a UDP port and returns the value of net.Conn.Read() for
 // the first read on that port.
 func listenUDP(port int, timeout time.Duration) error {