18 files changed, 469 insertions, 122 deletions

diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go
index 4301b697c..1684dfc24 100644
--- a/pkg/sentry/socket/control/control.go
+++ b/pkg/sentry/socket/control/control.go
@@ -327,7 +327,7 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte {
 }
 
 // PackTOS packs an IP_TOS socket control message.
-func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte {
+func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte {
 	return putCmsgStruct(
 		buf,
 		linux.SOL_IP,
diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD
index 5eb06bbf4..b70047d81 100644
--- a/pkg/sentry/socket/netfilter/BUILD
+++ b/pkg/sentry/socket/netfilter/BUILD
@@ -14,6 +14,7 @@ go_library(
     deps = [
         "//pkg/abi/linux",
         "//pkg/binary",
+        "//pkg/log",
         "//pkg/sentry/kernel",
         "//pkg/sentry/usermem",
         "//pkg/syserr",
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index 9f87c32f1..a9cfc1749 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -21,6 +21,7 @@ import (
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/binary"
+	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/pkg/sentry/kernel"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
 	"gvisor.dev/gvisor/pkg/syserr"
@@ -35,6 +36,7 @@ const errorTargetName = "ERROR"
 
 // metadata is opaque to netstack. It holds data that we need to translate
 // between Linux's and netstack's iptables representations.
+// TODO(gvisor.dev/issue/170): This might be removable.
 type metadata struct {
 	HookEntry  [linux.NF_INET_NUMHOOKS]uint32
 	Underflow  [linux.NF_INET_NUMHOOKS]uint32
@@ -51,7 +53,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG
 	}
 
 	// Find the appropriate table.
-	table, err := findTable(ep, info.TableName())
+	table, err := findTable(ep, info.Name)
 	if err != nil {
 		return linux.IPTGetinfo{}, err
 	}
@@ -82,30 +84,31 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i
 	}
 
 	// Find the appropriate table.
-	table, err := findTable(ep, userEntries.TableName())
+	table, err := findTable(ep, userEntries.Name)
 	if err != nil {
 		return linux.KernelIPTGetEntries{}, err
 	}
 
 	// Convert netstack's iptables rules to something that the iptables
 	// tool can understand.
-	entries, _, err := convertNetstackToBinary(userEntries.TableName(), table)
+	entries, _, err := convertNetstackToBinary(userEntries.Name.String(), table)
 	if err != nil {
 		return linux.KernelIPTGetEntries{}, err
 	}
 	if binary.Size(entries) > uintptr(outLen) {
+		log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen))
 		return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument
 	}
 
 	return entries, nil
 }
 
-func findTable(ep tcpip.Endpoint, tableName string) (iptables.Table, *syserr.Error) {
+func findTable(ep tcpip.Endpoint, tablename linux.TableName) (iptables.Table, *syserr.Error) {
 	ipt, err := ep.IPTables()
 	if err != nil {
 		return iptables.Table{}, syserr.FromError(err)
 	}
-	table, ok := ipt.Tables[tableName]
+	table, ok := ipt.Tables[tablename.String()]
 	if !ok {
 		return iptables.Table{}, syserr.ErrInvalidArgument
 	}
@@ -135,110 +138,68 @@ func FillDefaultIPTables(stack *stack.Stack) {
 // format expected by the iptables tool. Linux stores each table as a binary
 // blob that can only be traversed by parsing a bit, reading some offsets,
 // jumping to those offsets, parsing again, etc.
-func convertNetstackToBinary(name string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) {
+func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) {
 	// Return values.
 	var entries linux.KernelIPTGetEntries
 	var meta metadata
 
 	// The table name has to fit in the struct.
-	if linux.XT_TABLE_MAXNAMELEN < len(name) {
+	if linux.XT_TABLE_MAXNAMELEN < len(tablename) {
+		log.Warningf("Table name %q too long.", tablename)
 		return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument
 	}
-	copy(entries.Name[:], name)
-
-	// Deal with the built in chains first (INPUT, OUTPUT, etc.). Each of
-	// these chains ends with an unconditional policy entry.
-	for hook := iptables.Prerouting; hook < iptables.NumHooks; hook++ {
-		chain, ok := table.BuiltinChains[hook]
-		if !ok {
-			// This table doesn't support this hook.
-			continue
-		}
-
-		// Sanity check.
-		if len(chain.Rules) < 1 {
-			return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument
-		}
+	copy(entries.Name[:], tablename)
 
-		for ruleIdx, rule := range chain.Rules {
-			// If this is the first rule of a builtin chain, set
-			// the metadata hook entry point.
-			if ruleIdx == 0 {
+	for ruleIdx, rule := range table.Rules {
+		// Is this a chain entry point?
+		for hook, hookRuleIdx := range table.BuiltinChains {
+			if hookRuleIdx == ruleIdx {
 				meta.HookEntry[hook] = entries.Size
 			}
-
-			// Each rule corresponds to an entry.
-			entry := linux.KernelIPTEntry{
-				IPTEntry: linux.IPTEntry{
-					NextOffset:   linux.SizeOfIPTEntry,
-					TargetOffset: linux.SizeOfIPTEntry,
-				},
+		}
+		// Is this a chain underflow point?
+		for underflow, underflowRuleIdx := range table.Underflows {
+			if underflowRuleIdx == ruleIdx {
+				meta.Underflow[underflow] = entries.Size
 			}
+		}
 
-			for _, matcher := range rule.Matchers {
-				// Serialize the matcher and add it to the
-				// entry.
-				serialized := marshalMatcher(matcher)
-				entry.Elems = append(entry.Elems, serialized...)
-				entry.NextOffset += uint16(len(serialized))
-				entry.TargetOffset += uint16(len(serialized))
-			}
+		// Each rule corresponds to an entry.
+		entry := linux.KernelIPTEntry{
+			IPTEntry: linux.IPTEntry{
+				NextOffset:   linux.SizeOfIPTEntry,
+				TargetOffset: linux.SizeOfIPTEntry,
+			},
+		}
 
-			// Serialize and append the target.
-			serialized := marshalTarget(rule.Target)
+		for _, matcher := range rule.Matchers {
+			// Serialize the matcher and add it to the
+			// entry.
+			serialized := marshalMatcher(matcher)
 			entry.Elems = append(entry.Elems, serialized...)
 			entry.NextOffset += uint16(len(serialized))
-
-			// The underflow rule is the last rule in the chain,
-			// and is an unconditional rule (i.e. it matches any
-			// packet). This is enforced when saving iptables.
-			if ruleIdx == len(chain.Rules)-1 {
-				meta.Underflow[hook] = entries.Size
-			}
-
-			entries.Size += uint32(entry.NextOffset)
-			entries.Entrytable = append(entries.Entrytable, entry)
-			meta.NumEntries++
+			entry.TargetOffset += uint16(len(serialized))
 		}
 
-	}
-
-	// TODO(gvisor.dev/issue/170): Deal with the user chains here. Each of
-	// these starts with an error node holding the chain's name and ends
-	// with an unconditional return.
+		// Serialize and append the target.
+		serialized := marshalTarget(rule.Target)
+		entry.Elems = append(entry.Elems, serialized...)
+		entry.NextOffset += uint16(len(serialized))
 
-	// Lastly, each table ends with an unconditional error target rule as
-	// its final entry.
-	errorEntry := linux.KernelIPTEntry{
-		IPTEntry: linux.IPTEntry{
-			NextOffset:   linux.SizeOfIPTEntry,
-			TargetOffset: linux.SizeOfIPTEntry,
-		},
+		entries.Size += uint32(entry.NextOffset)
+		entries.Entrytable = append(entries.Entrytable, entry)
+		meta.NumEntries++
 	}
-	var errorTarget linux.XTErrorTarget
-	errorTarget.Target.TargetSize = linux.SizeOfXTErrorTarget
-	copy(errorTarget.ErrorName[:], errorTargetName)
-	copy(errorTarget.Target.Name[:], errorTargetName)
-
-	// Serialize and add it to the list of entries.
-	errorTargetBuf := make([]byte, 0, linux.SizeOfXTErrorTarget)
-	serializedErrorTarget := binary.Marshal(errorTargetBuf, usermem.ByteOrder, errorTarget)
-	errorEntry.Elems = append(errorEntry.Elems, serializedErrorTarget...)
-	errorEntry.NextOffset += uint16(len(serializedErrorTarget))
-
-	entries.Size += uint32(errorEntry.NextOffset)
-	entries.Entrytable = append(entries.Entrytable, errorEntry)
-	meta.NumEntries++
-	meta.Size = entries.Size
 
+	meta.Size = entries.Size
 	return entries, meta, nil
 }
 
 func marshalMatcher(matcher iptables.Matcher) []byte {
 	switch matcher.(type) {
 	default:
-		// TODO(gvisor.dev/issue/170): We don't support any matchers yet, so
-		// any call to marshalMatcher will panic.
+		// TODO(gvisor.dev/issue/170): We don't support any matchers
+		// yet, so any call to marshalMatcher will panic.
 		panic(fmt.Errorf("unknown matcher of type %T", matcher))
 	}
 }
@@ -246,28 +207,46 @@ func marshalMatcher(matcher iptables.Matcher) []byte {
 func marshalTarget(target iptables.Target) []byte {
 	switch target.(type) {
 	case iptables.UnconditionalAcceptTarget:
-		return marshalUnconditionalAcceptTarget()
+		return marshalStandardTarget(iptables.Accept)
+	case iptables.UnconditionalDropTarget:
+		return marshalStandardTarget(iptables.Drop)
+	case iptables.ErrorTarget:
+		return marshalErrorTarget()
 	default:
 		panic(fmt.Errorf("unknown target of type %T", target))
 	}
 }
 
-func marshalUnconditionalAcceptTarget() []byte {
+func marshalStandardTarget(verdict iptables.Verdict) []byte {
 	// The target's name will be the empty string.
 	target := linux.XTStandardTarget{
 		Target: linux.XTEntryTarget{
 			TargetSize: linux.SizeOfXTStandardTarget,
 		},
-		Verdict: translateStandardVerdict(iptables.Accept),
+		Verdict: translateFromStandardVerdict(verdict),
 	}
 
 	ret := make([]byte, 0, linux.SizeOfXTStandardTarget)
 	return binary.Marshal(ret, usermem.ByteOrder, target)
 }
 
-// translateStandardVerdict translates verdicts the same way as the iptables
+func marshalErrorTarget() []byte {
+	// This is an error target named error
+	target := linux.XTErrorTarget{
+		Target: linux.XTEntryTarget{
+			TargetSize: linux.SizeOfXTErrorTarget,
+		},
+	}
+	copy(target.Name[:], errorTargetName)
+	copy(target.Target.Name[:], errorTargetName)
+
+	ret := make([]byte, 0, linux.SizeOfXTErrorTarget)
+	return binary.Marshal(ret, usermem.ByteOrder, target)
+}
+
+// translateFromStandardVerdict translates verdicts the same way as the iptables
 // tool.
-func translateStandardVerdict(verdict iptables.Verdict) int32 {
+func translateFromStandardVerdict(verdict iptables.Verdict) int32 {
 	switch verdict {
 	case iptables.Accept:
 		return -linux.NF_ACCEPT - 1
@@ -280,7 +259,258 @@ func translateStandardVerdict(verdict iptables.Verdict) int32 {
 	case iptables.Jump:
 		// TODO(gvisor.dev/issue/170): Support Jump.
 		panic("Jump isn't supported yet")
+	}
+	panic(fmt.Sprintf("unknown standard verdict: %d", verdict))
+}
+
+// translateToStandardVerdict translates from the value in a
+// linux.XTStandardTarget to an iptables.Verdict.
+func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) {
+	// TODO(gvisor.dev/issue/170): Support other verdicts.
+	switch val {
+	case -linux.NF_ACCEPT - 1:
+		return iptables.Accept, nil
+	case -linux.NF_DROP - 1:
+		return iptables.Drop, nil
+	case -linux.NF_QUEUE - 1:
+		log.Warningf("Unsupported iptables verdict QUEUE.")
+	case linux.NF_RETURN:
+		log.Warningf("Unsupported iptables verdict RETURN.")
+	default:
+		log.Warningf("Unknown iptables verdict %d.", val)
+	}
+	return iptables.Invalid, syserr.ErrInvalidArgument
+}
+
+// SetEntries sets iptables rules for a single table. See
+// net/ipv4/netfilter/ip_tables.c:translate_table for reference.
+func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
+	printReplace(optVal)
+
+	// Get the basic rules data (struct ipt_replace).
+	if len(optVal) < linux.SizeOfIPTReplace {
+		log.Warningf("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal))
+		return syserr.ErrInvalidArgument
+	}
+	var replace linux.IPTReplace
+	replaceBuf := optVal[:linux.SizeOfIPTReplace]
+	optVal = optVal[linux.SizeOfIPTReplace:]
+	binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace)
+
+	// TODO(gvisor.dev/issue/170): Support other tables.
+	var table iptables.Table
+	switch replace.Name.String() {
+	case iptables.TablenameFilter:
+		table = iptables.EmptyFilterTable()
 	default:
-		panic(fmt.Sprintf("unknown standard verdict: %d", verdict))
+		log.Warningf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String())
+		return syserr.ErrInvalidArgument
+	}
+
+	// Convert input into a list of rules and their offsets.
+	var offset uint32
+	var offsets []uint32
+	for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ {
+		// Get the struct ipt_entry.
+		if len(optVal) < linux.SizeOfIPTEntry {
+			log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal))
+			return syserr.ErrInvalidArgument
+		}
+		var entry linux.IPTEntry
+		buf := optVal[:linux.SizeOfIPTEntry]
+		optVal = optVal[linux.SizeOfIPTEntry:]
+		binary.Unmarshal(buf, usermem.ByteOrder, &entry)
+		if entry.TargetOffset != linux.SizeOfIPTEntry {
+			// TODO(gvisor.dev/issue/170): Support matchers.
+			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
+		}
+
+		// Get the target of the rule.
+		target, consumed, err := parseTarget(optVal)
+		if err != nil {
+			return err
+		}
+		optVal = optVal[consumed:]
+
+		table.Rules = append(table.Rules, iptables.Rule{Target: target})
+		offsets = append(offsets, offset)
+		offset += linux.SizeOfIPTEntry + consumed
+	}
+
+	// Go through the list of supported hooks for this table and, for each
+	// one, set the rule it corresponds to.
+	for hook, _ := range replace.HookEntry {
+		if table.ValidHooks()&uint32(hook) != 0 {
+			hk := hookFromLinux(hook)
+			for ruleIdx, offset := range offsets {
+				if offset == replace.HookEntry[hook] {
+					table.BuiltinChains[hk] = ruleIdx
+				}
+				if offset == replace.Underflow[hook] {
+					table.Underflows[hk] = ruleIdx
+				}
+			}
+			if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset {
+				log.Warningf("Hook %v is unset.", hk)
+				return syserr.ErrInvalidArgument
+			}
+			if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset {
+				log.Warningf("Underflow %v is unset.", hk)
+				return syserr.ErrInvalidArgument
+			}
+		}
+	}
+
+	ipt := stack.IPTables()
+	table.SetMetadata(metadata{
+		HookEntry:  replace.HookEntry,
+		Underflow:  replace.Underflow,
+		NumEntries: replace.NumEntries,
+		Size:       replace.Size,
+	})
+	ipt.Tables[replace.Name.String()] = table
+	stack.SetIPTables(ipt)
+
+	return nil
+}
+
+// parseTarget parses a target from the start of optVal and returns the target
+// along with the number of bytes it occupies in optVal.
+func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) {
+	if len(optVal) < linux.SizeOfXTEntryTarget {
+		log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal))
+		return nil, 0, syserr.ErrInvalidArgument
+	}
+	var target linux.XTEntryTarget
+	buf := optVal[:linux.SizeOfXTEntryTarget]
+	binary.Unmarshal(buf, usermem.ByteOrder, &target)
+	switch target.Name.String() {
+	case "":
+		// Standard target.
+		if len(optVal) < linux.SizeOfXTStandardTarget {
+			log.Warningf("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal))
+			return nil, 0, syserr.ErrInvalidArgument
+		}
+		var standardTarget linux.XTStandardTarget
+		buf = optVal[:linux.SizeOfXTStandardTarget]
+		binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget)
+
+		verdict, err := translateToStandardVerdict(standardTarget.Verdict)
+		if err != nil {
+			return nil, 0, err
+		}
+		switch verdict {
+		case iptables.Accept:
+			return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil
+		case iptables.Drop:
+			// TODO(gvisor.dev/issue/170): Return an
+			// iptables.UnconditionalDropTarget to support DROP.
+			log.Infof("netfilter DROP is not supported yet.")
+			return nil, 0, syserr.ErrInvalidArgument
+		default:
+			panic(fmt.Sprintf("Unknown verdict: %v", verdict))
+		}
+
+	case errorTargetName:
+		// Error target.
+		if len(optVal) < linux.SizeOfXTErrorTarget {
+			log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal))
+			return nil, 0, syserr.ErrInvalidArgument
+		}
+		var errorTarget linux.XTErrorTarget
+		buf = optVal[:linux.SizeOfXTErrorTarget]
+		binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget)
+
+		// Error targets are used in 2 cases:
+		// * An actual error case. These rules have an error
+		//   named errorTargetName. The last entry of the table
+		//   is usually an error case to catch any packets that
+		//   somehow fall through every rule.
+		// * To mark the start of a user defined chain. These
+		//   rules have an error with the name of the chain.
+		switch errorTarget.Name.String() {
+		case errorTargetName:
+			return iptables.ErrorTarget{}, linux.SizeOfXTErrorTarget, nil
+		default:
+			log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String())
+			return nil, 0, syserr.ErrInvalidArgument
+		}
+	}
+
+	// Unknown target.
+	log.Infof("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String())
+	return nil, 0, syserr.ErrInvalidArgument
+}
+
+func hookFromLinux(hook int) iptables.Hook {
+	switch hook {
+	case linux.NF_INET_PRE_ROUTING:
+		return iptables.Prerouting
+	case linux.NF_INET_LOCAL_IN:
+		return iptables.Input
+	case linux.NF_INET_FORWARD:
+		return iptables.Forward
+	case linux.NF_INET_LOCAL_OUT:
+		return iptables.Output
+	case linux.NF_INET_POST_ROUTING:
+		return iptables.Postrouting
+	}
+	panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook))
+}
+
+// printReplace prints information about the struct ipt_replace in optVal. It
+// is only for debugging.
+func printReplace(optVal []byte) {
+	// Basic replace info.
+	var replace linux.IPTReplace
+	replaceBuf := optVal[:linux.SizeOfIPTReplace]
+	optVal = optVal[linux.SizeOfIPTReplace:]
+	binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace)
+	log.Infof("Replacing table %q: %+v", replace.Name.String(), replace)
+
+	// Read in the list of entries at the end of replace.
+	var totalOffset uint16
+	for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ {
+		var entry linux.IPTEntry
+		entryBuf := optVal[:linux.SizeOfIPTEntry]
+		binary.Unmarshal(entryBuf, usermem.ByteOrder, &entry)
+		log.Infof("Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry)
+
+		totalOffset += entry.NextOffset
+		if entry.TargetOffset == linux.SizeOfIPTEntry {
+			log.Infof("Entry has no matches.")
+		} else {
+			log.Infof("Entry has matches.")
+		}
+
+		var target linux.XTEntryTarget
+		targetBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTEntryTarget]
+		binary.Unmarshal(targetBuf, usermem.ByteOrder, &target)
+		log.Infof("Target named %q: %+v", target.Name.String(), target)
+
+		switch target.Name.String() {
+		case "":
+			var standardTarget linux.XTStandardTarget
+			stBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTStandardTarget]
+			binary.Unmarshal(stBuf, usermem.ByteOrder, &standardTarget)
+			log.Infof("Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict)
+		case errorTargetName:
+			var errorTarget linux.XTErrorTarget
+			etBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTErrorTarget]
+			binary.Unmarshal(etBuf, usermem.ByteOrder, &errorTarget)
+			log.Infof("Error target with name %q.", errorTarget.Name.String())
+		default:
+			log.Infof("Unknown target type.")
+		}
+
+		optVal = optVal[entry.NextOffset:]
 	}
 }
diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD
index 136821963..103933144 100644
--- a/pkg/sentry/socket/netlink/BUILD
+++ b/pkg/sentry/socket/netlink/BUILD
@@ -27,6 +27,7 @@ go_library(
         "//pkg/sentry/socket/unix",
         "//pkg/sentry/socket/unix/transport",
         "//pkg/sentry/usermem",
+        "//pkg/sync",
         "//pkg/syserr",
         "//pkg/syserror",
         "//pkg/tcpip",
diff --git a/pkg/sentry/socket/netlink/port/BUILD b/pkg/sentry/socket/netlink/port/BUILD
index 463544c1a..2d9f4ba9b 100644
--- a/pkg/sentry/socket/netlink/port/BUILD
+++ b/pkg/sentry/socket/netlink/port/BUILD
@@ -8,6 +8,7 @@ go_library(
     srcs = ["port.go"],
     importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port",
     visibility = ["//pkg/sentry:internal"],
+    deps = ["//pkg/sync"],
 )
 
 go_test(
diff --git a/pkg/sentry/socket/netlink/port/port.go b/pkg/sentry/socket/netlink/port/port.go
index e9d3275b1..2cd3afc22 100644
--- a/pkg/sentry/socket/netlink/port/port.go
+++ b/pkg/sentry/socket/netlink/port/port.go
@@ -24,7 +24,8 @@ import (
 	"fmt"
 	"math"
 	"math/rand"
-	"sync"
+
+	"gvisor.dev/gvisor/pkg/sync"
 )
 
 // maxPorts is a sanity limit on the maximum number of ports to allocate per
diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go
index d2e3644a6..cea56f4ed 100644
--- a/pkg/sentry/socket/netlink/socket.go
+++ b/pkg/sentry/socket/netlink/socket.go
@@ -17,7 +17,6 @@ package netlink
 
 import (
 	"math"
-	"sync"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/binary"
@@ -34,6 +33,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/socket/unix"
 	"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/tcpip"
diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD
index e414d8055..f78784569 100644
--- a/pkg/sentry/socket/netstack/BUILD
+++ b/pkg/sentry/socket/netstack/BUILD
@@ -34,6 +34,7 @@ go_library(
         "//pkg/sentry/socket/netfilter",
         "//pkg/sentry/unimpl",
         "//pkg/sentry/usermem",
+        "//pkg/sync",
         "//pkg/syserr",
         "//pkg/syserror",
         "//pkg/tcpip",
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index 9e0d69046..d2f7e987d 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -29,7 +29,6 @@ import (
 	"io"
 	"math"
 	"reflect"
-	"sync"
 	"syscall"
 	"time"
 
@@ -49,6 +48,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/socket/netfilter"
 	"gvisor.dev/gvisor/pkg/sentry/unimpl"
 	"gvisor.dev/gvisor/pkg/sentry/usermem"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/tcpip"
@@ -324,22 +324,15 @@ func bytesToIPAddress(addr []byte) tcpip.Address {
 // converts it to the FullAddress format. It supports AF_UNIX, AF_INET,
 // AF_INET6, and AF_PACKET addresses.
 //
-// strict indicates whether addresses with the AF_UNSPEC family are accepted of not.
-//
 // AddressAndFamily returns an address and its family.
-func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, uint16, *syserr.Error) {
+func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) {
 	// Make sure we have at least 2 bytes for the address family.
 	if len(addr) < 2 {
 		return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument
 	}
 
-	family := usermem.ByteOrder.Uint16(addr)
-	if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) {
-		return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported
-	}
-
 	// Get the rest of the fields based on the address family.
-	switch family {
+	switch family := usermem.ByteOrder.Uint16(addr); family {
 	case linux.AF_UNIX:
 		path := addr[2:]
 		if len(path) > linux.UnixPathMax {
@@ -638,10 +631,40 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask {
 	return r
 }
 
+func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error {
+	if family == uint16(s.family) {
+		return nil
+	}
+	if !exact && family == linux.AF_INET && s.family == linux.AF_INET6 {
+		v, err := s.Endpoint.GetSockOptBool(tcpip.V6OnlyOption)
+		if err != nil {
+			return syserr.TranslateNetstackError(err)
+		}
+		if !v {
+			return nil
+		}
+	}
+	return syserr.ErrInvalidArgument
+}
+
+// mapFamily maps the AF_INET ANY address to the IPv4-mapped IPv6 ANY if the
+// receiver's family is AF_INET6.
+//
+// This is a hack to work around the fact that both IPv4 and IPv6 ANY are
+// represented by the empty string.
+//
+// TODO(gvisor.dev/issues/1556): remove this function.
+func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress {
+	if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET {
+		addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00"
+	}
+	return addr
+}
+
 // Connect implements the linux syscall connect(2) for sockets backed by
 // tpcip.Endpoint.
 func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error {
-	addr, family, err := AddressAndFamily(s.family, sockaddr, false /* strict */)
+	addr, family, err := AddressAndFamily(sockaddr)
 	if err != nil {
 		return err
 	}
@@ -653,6 +676,12 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo
 		}
 		return syserr.TranslateNetstackError(err)
 	}
+
+	if err := s.checkFamily(family, false /* exact */); err != nil {
+		return err
+	}
+	addr = s.mapFamily(addr, family)
+
 	// Always return right away in the non-blocking case.
 	if !blocking {
 		return syserr.TranslateNetstackError(s.Endpoint.Connect(addr))
@@ -681,10 +710,14 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo
 // Bind implements the linux syscall bind(2) for sockets backed by
 // tcpip.Endpoint.
 func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error {
-	addr, _, err := AddressAndFamily(s.family, sockaddr, true /* strict */)
+	addr, family, err := AddressAndFamily(sockaddr)
 	if err != nil {
 		return err
 	}
+	if err := s.checkFamily(family, true /* exact */); err != nil {
+		return err
+	}
+	addr = s.mapFamily(addr, family)
 
 	// Issue the bind request to the endpoint.
 	return syserr.TranslateNetstackError(s.Endpoint.Bind(addr))
@@ -985,13 +1018,23 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family
 		if err := ep.GetSockOpt(&v); err != nil {
 			return nil, syserr.TranslateNetstackError(err)
 		}
-		if len(v) == 0 {
+		if v == 0 {
 			return []byte{}, nil
 		}
 		if outLen < linux.IFNAMSIZ {
 			return nil, syserr.ErrInvalidArgument
 		}
-		return append([]byte(v), 0), nil
+		s := t.NetworkContext()
+		if s == nil {
+			return nil, syserr.ErrNoDevice
+		}
+		nic, ok := s.Interfaces()[int32(v)]
+		if !ok {
+			// The NICID no longer indicates a valid interface, probably because that
+			// interface was removed.
+			return nil, syserr.ErrUnknownDevice
+		}
+		return append([]byte(nic.Name), 0), nil
 
 	case linux.SO_BROADCAST:
 		if outLen < sizeOfInt32 {
@@ -1225,11 +1268,11 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
 		if err != nil {
 			return nil, syserr.TranslateNetstackError(err)
 		}
-		var o uint32
+		var o int32
 		if v {
 			o = 1
 		}
-		return int32(o), nil
+		return o, nil
 
 	case linux.IPV6_PATHMTU:
 		t.Kernel().EmitUnimplementedEvent(t)
@@ -1334,6 +1377,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
 		}
 		return int32(v), nil
 
+	case linux.IP_RECVTOS:
+		if outLen < sizeOfInt32 {
+			return nil, syserr.ErrInvalidArgument
+		}
+
+		v, err := ep.GetSockOptBool(tcpip.ReceiveTOSOption)
+		if err != nil {
+			return nil, syserr.TranslateNetstackError(err)
+		}
+		var o int32
+		if v {
+			o = 1
+		}
+		return o, nil
+
 	default:
 		emitUnimplementedEventIP(t, name)
 	}
@@ -1367,6 +1425,26 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa
 		return nil
 	}
 
+	if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP {
+		switch name {
+		case linux.IPT_SO_SET_REPLACE:
+			if len(optVal) < linux.SizeOfIPTReplace {
+				return syserr.ErrInvalidArgument
+			}
+
+			stack := inet.StackFromContext(t)
+			if stack == nil {
+				return syserr.ErrNoDevice
+			}
+			// Stack must be a netstack stack.
+			return netfilter.SetEntries(stack.(*Stack).Stack, optVal)
+
+		case linux.IPT_SO_SET_ADD_COUNTERS:
+			// TODO(gvisor.dev/issue/170): Counter support.
+			return nil
+		}
+	}
+
 	return SetSockOpt(t, s, s.Endpoint, level, name, optVal)
 }
 
@@ -1438,7 +1516,20 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i
 		if n == -1 {
 			n = len(optVal)
 		}
-		return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(optVal[:n])))
+		name := string(optVal[:n])
+		if name == "" {
+			return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(0)))
+		}
+		s := t.NetworkContext()
+		if s == nil {
+			return syserr.ErrNoDevice
+		}
+		for nicID, nic := range s.Interfaces() {
+			if nic.Name == name {
+				return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(nicID)))
+			}
+		}
+		return syserr.ErrUnknownDevice
 
 	case linux.SO_BROADCAST:
 		if len(optVal) < sizeOfInt32 {
@@ -1819,6 +1910,13 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
 		}
 		return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v)))
 
+	case linux.IP_RECVTOS:
+		v, err := parseIntOrChar(optVal)
+		if err != nil {
+			return err
+		}
+		return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0))
+
 	case linux.IP_ADD_SOURCE_MEMBERSHIP,
 		linux.IP_BIND_ADDRESS_NO_PORT,
 		linux.IP_BLOCK_SOURCE,
@@ -1839,7 +1937,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
 		linux.IP_RECVFRAGSIZE,
 		linux.IP_RECVOPTS,
 		linux.IP_RECVORIGDSTADDR,
-		linux.IP_RECVTOS,
 		linux.IP_RECVTTL,
 		linux.IP_RETOPTS,
 		linux.IP_TRANSPARENT,
@@ -2037,8 +2134,8 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32)
 
 	case linux.AF_INET6:
 		var out linux.SockAddrInet6
-		if len(addr.Addr) == 4 {
-			// Copy address is v4-mapped format.
+		if len(addr.Addr) == header.IPv4AddressSize {
+			// Copy address in v4-mapped format.
 			copy(out.Addr[12:], addr.Addr)
 			out.Addr[10] = 0xff
 			out.Addr[11] = 0xff
@@ -2259,7 +2356,14 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe
 }
 
 func (s *SocketOperations) controlMessages() socket.ControlMessages {
-	return socket.ControlMessages{IP: tcpip.ControlMessages{HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, Timestamp: s.readCM.Timestamp}}
+	return socket.ControlMessages{
+		IP: tcpip.ControlMessages{
+			HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp,
+			Timestamp:    s.readCM.Timestamp,
+			HasTOS:       s.readCM.HasTOS,
+			TOS:          s.readCM.TOS,
+		},
+	}
 }
 
 // updateTimestamp sets the timestamp for SIOCGSTAMP. It should be called after
@@ -2352,10 +2456,14 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []
 
 	var addr *tcpip.FullAddress
 	if len(to) > 0 {
-		addrBuf, _, err := AddressAndFamily(s.family, to, true /* strict */)
+		addrBuf, family, err := AddressAndFamily(to)
 		if err != nil {
 			return 0, err
 		}
+		if err := s.checkFamily(family, false /* exact */); err != nil {
+			return 0, err
+		}
+		addrBuf = s.mapFamily(addrBuf, family)
 
 		addr = &addrBuf
 	}
diff --git a/pkg/sentry/socket/rpcinet/conn/BUILD b/pkg/sentry/socket/rpcinet/conn/BUILD
index 23eadcb1b..b2677c659 100644
--- a/pkg/sentry/socket/rpcinet/conn/BUILD
+++ b/pkg/sentry/socket/rpcinet/conn/BUILD
@@ -10,6 +10,7 @@ go_library(
     deps = [
         "//pkg/binary",
         "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto",
+        "//pkg/sync",
         "//pkg/syserr",
         "//pkg/unet",
         "@com_github_golang_protobuf//proto:go_default_library",
diff --git a/pkg/sentry/socket/rpcinet/conn/conn.go b/pkg/sentry/socket/rpcinet/conn/conn.go
index 356adad99..02f39c767 100644
--- a/pkg/sentry/socket/rpcinet/conn/conn.go
+++ b/pkg/sentry/socket/rpcinet/conn/conn.go
@@ -17,12 +17,12 @@ package conn
 
 import (
 	"fmt"
-	"sync"
 	"sync/atomic"
 	"syscall"
 
 	"github.com/golang/protobuf/proto"
 	"gvisor.dev/gvisor/pkg/binary"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/unet"
 
diff --git a/pkg/sentry/socket/rpcinet/notifier/BUILD b/pkg/sentry/socket/rpcinet/notifier/BUILD
index a3585e10d..a5954f22b 100644
--- a/pkg/sentry/socket/rpcinet/notifier/BUILD
+++ b/pkg/sentry/socket/rpcinet/notifier/BUILD
@@ -10,6 +10,7 @@ go_library(
     deps = [
         "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto",
         "//pkg/sentry/socket/rpcinet/conn",
+        "//pkg/sync",
         "//pkg/waiter",
         "@org_golang_x_sys//unix:go_default_library",
     ],
diff --git a/pkg/sentry/socket/rpcinet/notifier/notifier.go b/pkg/sentry/socket/rpcinet/notifier/notifier.go
index 7efe4301f..82b75d6dd 100644
--- a/pkg/sentry/socket/rpcinet/notifier/notifier.go
+++ b/pkg/sentry/socket/rpcinet/notifier/notifier.go
@@ -17,12 +17,12 @@ package notifier
 
 import (
 	"fmt"
-	"sync"
 	"syscall"
 
 	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn"
 	pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
 
diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD
index 788ad70d2..d7ba95dff 100644
--- a/pkg/sentry/socket/unix/transport/BUILD
+++ b/pkg/sentry/socket/unix/transport/BUILD
@@ -32,6 +32,7 @@ go_library(
         "//pkg/ilist",
         "//pkg/refs",
         "//pkg/sentry/context",
+        "//pkg/sync",
         "//pkg/syserr",
         "//pkg/tcpip",
         "//pkg/tcpip/buffer",
diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go
index dea11e253..9e6fbc111 100644
--- a/pkg/sentry/socket/unix/transport/connectioned.go
+++ b/pkg/sentry/socket/unix/transport/connectioned.go
@@ -15,10 +15,9 @@
 package transport
 
 import (
-	"sync"
-
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/context"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/waiter"
diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go
index e27b1c714..5dcd3d95e 100644
--- a/pkg/sentry/socket/unix/transport/queue.go
+++ b/pkg/sentry/socket/unix/transport/queue.go
@@ -15,9 +15,8 @@
 package transport
 
 import (
-	"sync"
-
 	"gvisor.dev/gvisor/pkg/refs"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go
index 37c7ac3c1..fcc0da332 100644
--- a/pkg/sentry/socket/unix/transport/unix.go
+++ b/pkg/sentry/socket/unix/transport/unix.go
@@ -16,11 +16,11 @@
 package transport
 
 import (
-	"sync"
 	"sync/atomic"
 
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/sentry/context"
+	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserr"
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 91effe89a..7f49ba864 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -116,13 +116,16 @@ func (s *SocketOperations) Endpoint() transport.Endpoint {
 
 // extractPath extracts and validates the address.
 func extractPath(sockaddr []byte) (string, *syserr.Error) {
-	addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */)
+	addr, family, err := netstack.AddressAndFamily(sockaddr)
 	if err != nil {
 		if err == syserr.ErrAddressFamilyNotSupported {
 			err = syserr.ErrInvalidArgument
 		}
 		return "", err
 	}
+	if family != linux.AF_UNIX {
+		return "", syserr.ErrInvalidArgument
+	}
 
 	// The address is trimmed by GetAddress.
 	p := string(addr.Addr)