iptables: replace maps with arrays

For iptables users, Check() is a hot path called for every packet one or more
times. Let's avoid a bunch of map lookups.

PiperOrigin-RevId: 322678699

pkg/sentry/socket/netfilter/netfilter.go

@@ -342,10 +342,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 	// TODO(gvisor.dev/issue/170): Support other tables.
 	var table stack.Table
 	switch replace.Name.String() {
-	case stack.TablenameFilter:
+	case stack.FilterTable:
 		table = stack.EmptyFilterTable()
-	case stack.TablenameNat:
-		table = stack.EmptyNatTable()
+	case stack.NATTable:
+		table = stack.EmptyNATTable()
 	default:
 		nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String())
 		return syserr.ErrInvalidArgument
@@ -431,6 +431,8 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 	for hook, _ := range replace.HookEntry {
 		if table.ValidHooks()&(1<<hook) != 0 {
 			hk := hookFromLinux(hook)
+			table.BuiltinChains[hk] = stack.HookUnset
+			table.Underflows[hk] = stack.HookUnset
 			for offset, ruleIdx := range offsets {
 				if offset == replace.HookEntry[hook] {
 					table.BuiltinChains[hk] = ruleIdx
@@ -456,8 +458,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 
 	// Add the user chains.
 	for ruleIdx, rule := range table.Rules {
-		target, ok := rule.Target.(stack.UserChainTarget)
-		if !ok {
+		if _, ok := rule.Target.(stack.UserChainTarget); !ok {
 			continue
 		}
 
@@ -473,7 +474,6 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 			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
@@ -499,7 +499,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 	// Since we only support modifying the INPUT, PREROUTING and OUTPUT chain right now,
 	// make sure all other chains point to ACCEPT rules.
 	for hook, ruleIdx := range table.BuiltinChains {
-		if hook == stack.Forward || hook == stack.Postrouting {
+		if hook := stack.Hook(hook); hook == stack.Forward || hook == stack.Postrouting {
+			if ruleIdx == stack.HookUnset {
+				continue
+			}
 			if !isUnconditionalAccept(table.Rules[ruleIdx]) {
 				nflog("hook %d is unsupported.", hook)
 				return syserr.ErrInvalidArgument
@@ -512,9 +515,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
 	// - There are no chains without an unconditional final rule.
 	// - There are no chains without an unconditional underflow rule.
 
-	stk.IPTables().ReplaceTable(replace.Name.String(), table)
-
-	return nil
+	return syserr.TranslateNetstackError(stk.IPTables().ReplaceTable(replace.Name.String(), table))
 }
 
 // parseMatchers parses 0 or more matchers from optVal. optVal should contain

pkg/tcpip/stack/iptables.go

@@ -22,22 +22,30 @@ import (
 	"gvisor.dev/gvisor/pkg/tcpip/header"
 )
 
-// Table names.
+// tableID is an index into IPTables.tables.
+type tableID int
+
 const (
-	TablenameNat    = "nat"
-	TablenameMangle = "mangle"
-	TablenameFilter = "filter"
+	natID tableID = iota
+	mangleID
+	filterID
+	numTables
 )
 
-// Chain names as defined by net/ipv4/netfilter/ip_tables.c.
+// Table names.
 const (
-	ChainNamePrerouting  = "PREROUTING"
-	ChainNameInput       = "INPUT"
-	ChainNameForward     = "FORWARD"
-	ChainNameOutput      = "OUTPUT"
-	ChainNamePostrouting = "POSTROUTING"
+	NATTable    = "nat"
+	MangleTable = "mangle"
+	FilterTable = "filter"
 )
 
+// nameToID is immutable.
+var nameToID = map[string]tableID{
+	NATTable:    natID,
+	MangleTable: mangleID,
+	FilterTable: filterID,
+}
+
 // HookUnset indicates that there is no hook set for an entrypoint or
 // underflow.
 const HookUnset = -1
@@ -48,11 +56,10 @@ const reaperDelay = 5 * time.Second
 // DefaultTables returns a default set of tables. Each chain is set to accept
 // all packets.
 func DefaultTables() *IPTables {
-	// TODO(gvisor.dev/issue/170): We may be able to swap out some strings for
-	// iotas.
 	return &IPTables{
-		tables: map[string]Table{
-			TablenameNat: Table{
+		tables: [numTables]Table{
+			// NAT table.
+			Table{
 				Rules: []Rule{
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: AcceptTarget{}},
@@ -60,60 +67,70 @@ func DefaultTables() *IPTables {
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: ErrorTarget{}},
 				},
-				BuiltinChains: map[Hook]int{
-					Prerouting:  0,
-					Input:       1,
-					Output:      2,
-					Postrouting: 3,
+				BuiltinChains: [NumHooks]int{
+					0,         // Prerouting.
+					1,         // Input.
+					HookUnset, // Forward.
+					2,         // Output.
+					3,         // Postrouting.
 				},
-				Underflows: map[Hook]int{
-					Prerouting:  0,
-					Input:       1,
-					Output:      2,
-					Postrouting: 3,
+				Underflows: [NumHooks]int{
+					0,         // Prerouting.
+					1,         // Input.
+					HookUnset, // Forward.
+					2,         // Output.
+					3,         // Postrouting.
 				},
-				UserChains: map[string]int{},
 			},
-			TablenameMangle: Table{
+			// Mangle table.
+			Table{
 				Rules: []Rule{
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: ErrorTarget{}},
 				},
-				BuiltinChains: map[Hook]int{
+				BuiltinChains: [NumHooks]int{
 					Prerouting: 0,
 					Output:     1,
 				},
-				Underflows: map[Hook]int{
-					Prerouting: 0,
-					Output:     1,
+				Underflows: [NumHooks]int{
+					0,         // Prerouting.
+					HookUnset, // Input.
+					HookUnset, // Forward.
+					1,         // Output.
+					HookUnset, // Postrouting.
 				},
-				UserChains: map[string]int{},
 			},
-			TablenameFilter: Table{
+			// Filter table.
+			Table{
 				Rules: []Rule{
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: AcceptTarget{}},
 					Rule{Target: ErrorTarget{}},
 				},
-				BuiltinChains: map[Hook]int{
-					Input:   0,
-					Forward: 1,
-					Output:  2,
+				BuiltinChains: [NumHooks]int{
+					HookUnset, // Prerouting.
+					Input:     0, // Input.
+					Forward:   1, // Forward.
+					Output:    2, // Output.
+					HookUnset, // Postrouting.
 				},
-				Underflows: map[Hook]int{
-					Input:   0,
-					Forward: 1,
-					Output:  2,
+				Underflows: [NumHooks]int{
+					HookUnset, // Prerouting.
+					0,         // Input.
+					1,         // Forward.
+					2,         // Output.
+					HookUnset, // Postrouting.
 				},
-				UserChains: map[string]int{},
 			},
 		},
-		priorities: map[Hook][]string{
-			Input:      []string{TablenameNat, TablenameFilter},
-			Prerouting: []string{TablenameMangle, TablenameNat},
-			Output:     []string{TablenameMangle, TablenameNat, TablenameFilter},
+		priorities: [NumHooks][]tableID{
+			[]tableID{mangleID, natID},           // Prerouting.
+			[]tableID{natID, filterID},           // Input.
+			[]tableID{},                          // Forward.
+			[]tableID{mangleID, natID, filterID}, // Output.
+			[]tableID{},                          // Postrouting.
 		},
 		connections: ConnTrack{
 			seed: generateRandUint32(),
@@ -127,51 +144,62 @@ func DefaultTables() *IPTables {
 func EmptyFilterTable() Table {
 	return Table{
 		Rules: []Rule{},
-		BuiltinChains: map[Hook]int{
-			Input:   HookUnset,
-			Forward: HookUnset,
-			Output:  HookUnset,
+		BuiltinChains: [NumHooks]int{
+			HookUnset,
+			0,
+			0,
+			0,
+			HookUnset,
 		},
-		Underflows: map[Hook]int{
-			Input:   HookUnset,
-			Forward: HookUnset,
-			Output:  HookUnset,
+		Underflows: [NumHooks]int{
+			HookUnset,
+			0,
+			0,
+			0,
+			HookUnset,
 		},
-		UserChains: map[string]int{},
 	}
 }
 
-// EmptyNatTable returns a Table with no rules and the filter table chains
+// EmptyNATTable returns a Table with no rules and the filter table chains
 // mapped to HookUnset.
-func EmptyNatTable() Table {
+func EmptyNATTable() Table {
 	return Table{
 		Rules: []Rule{},
-		BuiltinChains: map[Hook]int{
-			Prerouting:  HookUnset,
-			Input:       HookUnset,
-			Output:      HookUnset,
-			Postrouting: HookUnset,
+		BuiltinChains: [NumHooks]int{
+			0,
+			0,
+			HookUnset,
+			0,
+			0,
 		},
-		Underflows: map[Hook]int{
-			Prerouting:  HookUnset,
-			Input:       HookUnset,
-			Output:      HookUnset,
-			Postrouting: HookUnset,
+		Underflows: [NumHooks]int{
+			0,
+			0,
+			HookUnset,
+			0,
+			0,
 		},
-		UserChains: map[string]int{},
 	}
 }
 
-// GetTable returns table by name.
+// GetTable returns a table by name.
 func (it *IPTables) GetTable(name string) (Table, bool) {
+	id, ok := nameToID[name]
+	if !ok {
+		return Table{}, false
+	}
 	it.mu.RLock()
 	defer it.mu.RUnlock()
-	t, ok := it.tables[name]
-	return t, ok
+	return it.tables[id], true
 }
 
 // ReplaceTable replaces or inserts table by name.
-func (it *IPTables) ReplaceTable(name string, table Table) {
+func (it *IPTables) ReplaceTable(name string, table Table) *tcpip.Error {
+	id, ok := nameToID[name]
+	if !ok {
+		return tcpip.ErrInvalidOptionValue
+	}
 	it.mu.Lock()
 	defer it.mu.Unlock()
 	// If iptables is being enabled, initialize the conntrack table and
@@ -181,14 +209,8 @@ func (it *IPTables) ReplaceTable(name string, table Table) {
 		it.startReaper(reaperDelay)
 	}
 	it.modified = true
-	it.tables[name] = table
-}
-
-// GetPriorities returns slice of priorities associated with hook.
-func (it *IPTables) GetPriorities(hook Hook) []string {
-	it.mu.RLock()
-	defer it.mu.RUnlock()
-	return it.priorities[hook]
+	it.tables[id] = table
+	return nil
 }
 
 // A chainVerdict is what a table decides should be done with a packet.
@@ -226,8 +248,11 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, gso *GSO, r *Route, addr
 	it.connections.handlePacket(pkt, hook, gso, r)
 
 	// Go through each table containing the hook.
-	for _, tablename := range it.GetPriorities(hook) {
-		table, _ := it.GetTable(tablename)
+	it.mu.RLock()
+	defer it.mu.RUnlock()
+	priorities := it.priorities[hook]
+	for _, tableID := range priorities {
+		table := it.tables[tableID]
 		ruleIdx := table.BuiltinChains[hook]
 		switch verdict := it.checkChain(hook, pkt, table, ruleIdx, gso, r, address, nicName); verdict {
 		// If the table returns Accept, move on to the next table.

pkg/tcpip/stack/iptables_types.go

@@ -84,14 +84,14 @@ type IPTables struct {
 	// mu protects tables, priorities, and modified.
 	mu sync.RWMutex
 
-	// tables maps table names to tables. User tables have arbitrary names.
-	// mu needs to be locked for accessing.
-	tables map[string]Table
+	// tables maps tableIDs to tables. Holds builtin tables only, not user
+	// tables. mu must be locked for accessing.
+	tables [numTables]Table
 
 	// priorities maps each hook to a list of table names. The order of the
 	// list is the order in which each table should be visited for that
 	// hook. mu needs to be locked for accessing.
-	priorities map[Hook][]string
+	priorities [NumHooks][]tableID
 
 	// modified is whether tables have been modified at least once. It is
 	// used to elide the iptables performance overhead for workloads that
@@ -113,22 +113,20 @@ type Table struct {
 	Rules []Rule
 
 	// BuiltinChains maps builtin chains to their entrypoint rule in Rules.
-	BuiltinChains map[Hook]int
+	BuiltinChains [NumHooks]int
 
 	// Underflows maps builtin chains to their underflow rule in Rules
 	// (i.e. the rule to execute if the chain returns without a verdict).
-	Underflows map[Hook]int
-
-	// UserChains holds user-defined chains for the keyed by name. Users
-	// can give their chains arbitrary names.
-	UserChains map[string]int
+	Underflows [NumHooks]int
 }
 
 // ValidHooks returns a bitmap of the builtin hooks for the given table.
 func (table *Table) ValidHooks() uint32 {
 	hooks := uint32(0)
-	for hook := range table.BuiltinChains {
-		hooks |= 1 << hook
+	for hook, ruleIdx := range table.BuiltinChains {
+		if ruleIdx != HookUnset {
+			hooks |= 1 << hook
+		}
 	}
 	return hooks
 }