Merge release-20200115.0-48-g5f82f09 (automated)

12 files changed, 143 insertions, 38 deletions

diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index a9cfc1749..507a77483 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -25,7 +25,6 @@ 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"
 )
@@ -45,7 +44,7 @@ type metadata struct {
 }
 
 // GetInfo returns information about iptables.
-func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) {
+func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) {
 	// Read in the struct and table name.
 	var info linux.IPTGetinfo
 	if _, err := t.CopyIn(outPtr, &info); err != nil {
@@ -53,7 +52,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG
 	}
 
 	// Find the appropriate table.
-	table, err := findTable(ep, info.Name)
+	table, err := findTable(stack, info.Name)
 	if err != nil {
 		return linux.IPTGetinfo{}, err
 	}
@@ -76,7 +75,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG
 }
 
 // GetEntries returns netstack's iptables rules encoded for the iptables tool.
-func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) {
+func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) {
 	// Read in the struct and table name.
 	var userEntries linux.IPTGetEntries
 	if _, err := t.CopyIn(outPtr, &userEntries); err != nil {
@@ -84,7 +83,7 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i
 	}
 
 	// Find the appropriate table.
-	table, err := findTable(ep, userEntries.Name)
+	table, err := findTable(stack, userEntries.Name)
 	if err != nil {
 		return linux.KernelIPTGetEntries{}, err
 	}
@@ -103,11 +102,8 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i
 	return entries, nil
 }
 
-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)
-	}
+func findTable(stack *stack.Stack, tablename linux.TableName) (iptables.Table, *syserr.Error) {
+	ipt := stack.IPTables()
 	table, ok := ipt.Tables[tablename.String()]
 	if !ok {
 		return iptables.Table{}, syserr.ErrInvalidArgument
@@ -348,7 +344,7 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 	// 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 {
+		if table.ValidHooks()&(1<<hook) != 0 {
 			hk := hookFromLinux(hook)
 			for ruleIdx, offset := range offsets {
 				if offset == replace.HookEntry[hook] {
@@ -369,6 +365,23 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error {
 		}
 	}
 
+	// 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.
+	for hook, ruleIdx := range table.BuiltinChains {
+		if hook != iptables.Input {
+			if _, ok := table.Rules[ruleIdx].Target.(iptables.UnconditionalAcceptTarget); !ok {
+				log.Warningf("Hook %d is unsupported.", hook)
+				return syserr.ErrInvalidArgument
+			}
+		}
+	}
+
+	// TODO(gvisor.dev/issue/170): Check the following conditions:
+	// - There are no loops.
+	// - There are no chains without an unconditional final rule.
+	// - There are no chains without an unconditional underflow rule.
+
 	ipt := stack.IPTables()
 	table.SetMetadata(metadata{
 		HookEntry:  replace.HookEntry,
@@ -411,10 +424,7 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) {
 		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
+			return iptables.UnconditionalDropTarget{}, linux.SizeOfXTStandardTarget, nil
 		default:
 			panic(fmt.Sprintf("Unknown verdict: %v", verdict))
 		}
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index fec575357..2662fbc0f 100755
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -868,7 +868,11 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
 				return nil, syserr.ErrInvalidArgument
 			}
 
-			info, err := netfilter.GetInfo(t, s.Endpoint, outPtr)
+			stack := inet.StackFromContext(t)
+			if stack == nil {
+				return nil, syserr.ErrNoDevice
+			}
+			info, err := netfilter.GetInfo(t, stack.(*Stack).Stack, outPtr)
 			if err != nil {
 				return nil, err
 			}
@@ -879,7 +883,11 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
 				return nil, syserr.ErrInvalidArgument
 			}
 
-			entries, err := netfilter.GetEntries(t, s.Endpoint, outPtr, outLen)
+			stack := inet.StackFromContext(t)
+			if stack == nil {
+				return nil, syserr.ErrNoDevice
+			}
+			entries, err := netfilter.GetEntries(t, stack.(*Stack).Stack, outPtr, outLen)
 			if err != nil {
 				return nil, err
 			}
diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go
index 647970133..605a71679 100644
--- a/pkg/tcpip/iptables/iptables.go
+++ b/pkg/tcpip/iptables/iptables.go
@@ -16,6 +16,12 @@
 // tool.
 package iptables
 
+import (
+	"fmt"
+
+	"gvisor.dev/gvisor/pkg/tcpip"
+)
+
 // Table names.
 const (
 	TablenameNat    = "nat"
@@ -127,3 +133,72 @@ func EmptyFilterTable() Table {
 		UserChains: map[string]int{},
 	}
 }
+
+// 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.
+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 {
+		// If the table returns Accept, move on to the next table.
+		case Accept:
+			continue
+		// The Drop verdict is final.
+		case Drop:
+			return false
+		case Stolen, Queue, Repeat, None, Jump, Return, Continue:
+			panic(fmt.Sprintf("Unimplemented verdict %v.", verdict))
+		default:
+			panic(fmt.Sprintf("Unknown verdict %v.", verdict))
+		}
+	}
+
+	// Every table returned Accept.
+	return true
+}
+
+func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename string) Verdict {
+	// 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 {
+		// In either of these cases, this table is done with the packet.
+		case Accept, Drop:
+			return verdict
+		// Continue traversing the rules of the table.
+		case Continue:
+			continue
+		case Stolen, Queue, Repeat, None, Jump, Return:
+			panic(fmt.Sprintf("Unimplemented verdict %v.", verdict))
+		default:
+			panic(fmt.Sprintf("Unknown verdict %v.", verdict))
+		}
+	}
+
+	panic(fmt.Sprintf("Traversed past the entire list of iptables rules in table %q.", tablename))
+}
+
+func (it *IPTables) checkRule(hook Hook, pkt tcpip.PacketBuffer, table Table, ruleIdx int) Verdict {
+	rule := table.Rules[ruleIdx]
+	// Go through each rule matcher. If they all match, run
+	// the rule target.
+	for _, matcher := range rule.Matchers {
+		matches, hotdrop := matcher.Match(hook, pkt, "")
+		if hotdrop {
+			return Drop
+		}
+		if !matches {
+			return Continue
+		}
+	}
+
+	// All the matchers matched, so run the target.
+	verdict, _ := rule.Target.Action(pkt)
+	return verdict
+}
diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go
index b94a4c941..4dd281371 100644
--- a/pkg/tcpip/iptables/targets.go
+++ b/pkg/tcpip/iptables/targets.go
@@ -18,14 +18,14 @@ package iptables
 
 import (
 	"gvisor.dev/gvisor/pkg/log"
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
+	"gvisor.dev/gvisor/pkg/tcpip"
 )
 
 // UnconditionalAcceptTarget accepts all packets.
 type UnconditionalAcceptTarget struct{}
 
 // Action implements Target.Action.
-func (UnconditionalAcceptTarget) Action(packet buffer.VectorisedView) (Verdict, string) {
+func (UnconditionalAcceptTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) {
 	return Accept, ""
 }
 
@@ -33,7 +33,7 @@ func (UnconditionalAcceptTarget) Action(packet buffer.VectorisedView) (Verdict,
 type UnconditionalDropTarget struct{}
 
 // Action implements Target.Action.
-func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, string) {
+func (UnconditionalDropTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) {
 	return Drop, ""
 }
 
@@ -42,8 +42,7 @@ func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, st
 type ErrorTarget struct{}
 
 // Action implements Target.Action.
-func (ErrorTarget) Action(packet buffer.VectorisedView) (Verdict, string) {
+func (ErrorTarget) Action(packet tcpip.PacketBuffer) (Verdict, string) {
 	log.Warningf("ErrorTarget triggered.")
 	return Drop, ""
-
 }
diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go
index 540f8c0b4..9f6906100 100644
--- a/pkg/tcpip/iptables/types.go
+++ b/pkg/tcpip/iptables/types.go
@@ -14,9 +14,7 @@
 
 package iptables
 
-import (
-	"gvisor.dev/gvisor/pkg/tcpip/buffer"
-)
+import "gvisor.dev/gvisor/pkg/tcpip"
 
 // A Hook specifies one of the hooks built into the network stack.
 //
@@ -165,7 +163,7 @@ type Matcher interface {
 	// Match returns whether the packet matches and whether the packet
 	// should be "hotdropped", i.e. dropped immediately. This is usually
 	// used for suspicious packets.
-	Match(hook Hook, packet buffer.VectorisedView, interfaceName string) (matches bool, hotdrop bool)
+	Match(hook Hook, packet tcpip.PacketBuffer, interfaceName string) (matches bool, hotdrop bool)
 }
 
 // A Target is the interface for taking an action for a packet.
@@ -173,5 +171,5 @@ 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 buffer.VectorisedView) (Verdict, string)
+	Action(packet tcpip.PacketBuffer) (Verdict, string)
 }
diff --git a/pkg/tcpip/network/arp/arp.go b/pkg/tcpip/network/arp/arp.go
index 42cacb8a6..1ceaebfbd 100644
--- a/pkg/tcpip/network/arp/arp.go
+++ b/pkg/tcpip/network/arp/arp.go
@@ -137,7 +137,7 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
 	return tcpip.Address(h.ProtocolAddressSender()), ProtocolAddress
 }
 
-func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) {
+func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, sender stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) {
 	if addrWithPrefix.Address != ProtocolAddress {
 		return nil, tcpip.ErrBadLocalAddress
 	}
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 4ee3d5b45..0a1453b31 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -26,6 +26,7 @@ import (
 	"gvisor.dev/gvisor/pkg/tcpip"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
 	"gvisor.dev/gvisor/pkg/tcpip/header"
+	"gvisor.dev/gvisor/pkg/tcpip/iptables"
 	"gvisor.dev/gvisor/pkg/tcpip/network/fragmentation"
 	"gvisor.dev/gvisor/pkg/tcpip/network/hash"
 	"gvisor.dev/gvisor/pkg/tcpip/stack"
@@ -54,10 +55,11 @@ type endpoint struct {
 	dispatcher    stack.TransportDispatcher
 	fragmentation *fragmentation.Fragmentation
 	protocol      *protocol
+	stack         *stack.Stack
 }
 
 // NewEndpoint creates a new ipv4 endpoint.
-func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) {
+func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) {
 	e := &endpoint{
 		nicID:         nicID,
 		id:            stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address},
@@ -66,6 +68,7 @@ func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWi
 		dispatcher:    dispatcher,
 		fragmentation: fragmentation.NewFragmentation(fragmentation.HighFragThreshold, fragmentation.LowFragThreshold, fragmentation.DefaultReassembleTimeout),
 		protocol:      p,
+		stack:         st,
 	}
 
 	return e, nil
@@ -350,6 +353,13 @@ func (e *endpoint) HandlePacket(r *stack.Route, pkt tcpip.PacketBuffer) {
 	}
 	pkt.NetworkHeader = headerView[:h.HeaderLength()]
 
+	// iptables filtering.
+	ipt := e.stack.IPTables()
+	if ok := ipt.Check(iptables.Input, pkt); !ok {
+		// iptables is telling us to drop the packet.
+		return
+	}
+
 	hlen := int(h.HeaderLength())
 	tlen := int(h.TotalLength())
 	pkt.Data.TrimFront(hlen)
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index 58c3c79b9..180a480fd 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -221,7 +221,7 @@ func (*protocol) ParseAddresses(v buffer.View) (src, dst tcpip.Address) {
 }
 
 // NewEndpoint creates a new ipv6 endpoint.
-func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint) (stack.NetworkEndpoint, *tcpip.Error) {
+func (p *protocol) NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache stack.LinkAddressCache, dispatcher stack.TransportDispatcher, linkEP stack.LinkEndpoint, st *stack.Stack) (stack.NetworkEndpoint, *tcpip.Error) {
 	return &endpoint{
 		nicID:         nicID,
 		id:            stack.NetworkEndpointID{LocalAddress: addrWithPrefix.Address},
diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go
index 4afe7b744..de88c0bfa 100644
--- a/pkg/tcpip/stack/nic.go
+++ b/pkg/tcpip/stack/nic.go
@@ -598,7 +598,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar
 	}
 
 	// Create the new network endpoint.
-	ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP)
+	ep, err := netProto.NewEndpoint(n.id, protocolAddress.AddressWithPrefix, n.stack, n, n.linkEP, n.stack)
 	if err != nil {
 		return nil, err
 	}
diff --git a/pkg/tcpip/stack/registration.go b/pkg/tcpip/stack/registration.go
index 2b8751d49..ec91f60dd 100644
--- a/pkg/tcpip/stack/registration.go
+++ b/pkg/tcpip/stack/registration.go
@@ -282,7 +282,7 @@ type NetworkProtocol interface {
 	ParseAddresses(v buffer.View) (src, dst tcpip.Address)
 
 	// NewEndpoint creates a new endpoint of this protocol.
-	NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint) (NetworkEndpoint, *tcpip.Error)
+	NewEndpoint(nicID tcpip.NICID, addrWithPrefix tcpip.AddressWithPrefix, linkAddrCache LinkAddressCache, dispatcher TransportDispatcher, sender LinkEndpoint, st *Stack) (NetworkEndpoint, *tcpip.Error)
 
 	// SetOption allows enabling/disabling protocol specific features.
 	// SetOption returns an error if the option is not supported or the
diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go
index fc56a6d79..7057b110e 100644
--- a/pkg/tcpip/stack/stack.go
+++ b/pkg/tcpip/stack/stack.go
@@ -423,7 +423,11 @@ type Stack struct {
 	// handleLocal allows non-loopback interfaces to loop packets.
 	handleLocal bool
 
-	// tables are the iptables packet filtering and manipulation rules.
+	// tablesMu protects iptables.
+	tablesMu sync.RWMutex
+
+	// tables are the iptables packet filtering and manipulation rules. The are
+	// protected by tablesMu.`
 	tables iptables.IPTables
 
 	// resumableEndpoints is a list of endpoints that need to be resumed if the
@@ -1594,12 +1598,17 @@ func (s *Stack) LeaveGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NIC
 
 // IPTables returns the stack's iptables.
 func (s *Stack) IPTables() iptables.IPTables {
-	return s.tables
+	s.tablesMu.RLock()
+	t := s.tables
+	s.tablesMu.RUnlock()
+	return t
 }
 
 // SetIPTables sets the stack's iptables.
 func (s *Stack) SetIPTables(ipt iptables.IPTables) {
+	s.tablesMu.Lock()
 	s.tables = ipt
+	s.tablesMu.Unlock()
 }
 
 // ICMPLimit returns the maximum number of ICMP messages that can be sent
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 6243762e3..3fc823a36 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -40,7 +40,6 @@ import (
 
 	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/tcpip/buffer"
-	"gvisor.dev/gvisor/pkg/tcpip/iptables"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
 
@@ -454,9 +453,6 @@ type Endpoint interface {
 	// NOTE: This method is a no-op for sockets other than TCP.
 	ModerateRecvBuf(copied int)
 
-	// IPTables returns the iptables for this endpoint's stack.
-	IPTables() (iptables.IPTables, error)
-
 	// Info returns a copy to the transport endpoint info.
 	Info() EndpointInfo