diff options
-rw-r--r-- | pkg/sentry/socket/netfilter/netfilter.go | 27 | ||||
-rw-r--r-- | pkg/sentry/socket/netstack/netstack.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/iptables/BUILD | 2 | ||||
-rw-r--r-- | pkg/tcpip/iptables/iptables.go | 71 | ||||
-rw-r--r-- | pkg/tcpip/iptables/targets.go | 9 | ||||
-rw-r--r-- | pkg/tcpip/iptables/types.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/network/arp/arp.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/BUILD | 1 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 12 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ipv6.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/stack/nic.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/stack/registration.go | 2 | ||||
-rw-r--r-- | pkg/tcpip/tcpip.go | 4 | ||||
-rw-r--r-- | test/iptables/filter_input.go | 32 | ||||
-rw-r--r-- | test/iptables/iptables_test.go | 6 |
16 files changed, 155 insertions, 38 deletions
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index a9cfc1749..37f726295 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 +365,10 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } } + // TODO(gvisor.dev/issue/170): Check the following conditions: + // - There are no loops. + // - There are no chains without an unconditional final rule. + ipt := stack.IPTables() table.SetMetadata(metadata{ HookEntry: replace.HookEntry, @@ -411,10 +411,7 @@ 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 099319327..f7b069eed 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -834,7 +834,11 @@ 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 } @@ -845,7 +849,11 @@ 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/BUILD b/pkg/tcpip/BUILD index ebc8d0209..23e4b09e7 100644 --- a/pkg/tcpip/BUILD +++ b/pkg/tcpip/BUILD @@ -17,7 +17,6 @@ go_library( deps = [ "//pkg/sync", "//pkg/tcpip/buffer", - "//pkg/tcpip/iptables", "//pkg/waiter", ], ) diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index 64769c333..2893c80cd 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -13,6 +13,6 @@ go_library( visibility = ["//visibility:public"], deps = [ "//pkg/log", - "//pkg/tcpip/buffer", + "//pkg/tcpip", ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 647970133..83d807a4d 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 +133,68 @@ 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)) + } + } + + // 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)) + } + } + + 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/BUILD b/pkg/tcpip/network/ipv4/BUILD index aeddfcdd4..4e2aae9a3 100644 --- a/pkg/tcpip/network/ipv4/BUILD +++ b/pkg/tcpip/network/ipv4/BUILD @@ -15,6 +15,7 @@ go_library( "//pkg/tcpip", "//pkg/tcpip/buffer", "//pkg/tcpip/header", + "//pkg/tcpip/iptables", "//pkg/tcpip/network/fragmentation", "//pkg/tcpip/network/hash", "//pkg/tcpip/stack", 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 fe557ccbd..d50693634 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -469,7 +469,7 @@ 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/tcpip.go b/pkg/tcpip/tcpip.go index 4a090ac86..a0a3d1e53 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 diff --git a/test/iptables/filter_input.go b/test/iptables/filter_input.go index 923f44e68..7c4d469fa 100644 --- a/test/iptables/filter_input.go +++ b/test/iptables/filter_input.go @@ -31,6 +31,7 @@ func init() { RegisterTestCase(FilterInputDropUDP{}) RegisterTestCase(FilterInputDropUDPPort{}) RegisterTestCase(FilterInputDropDifferentUDPPort{}) + RegisterTestCase(FilterInputDropAll{}) } // FilterInputDropUDP tests that we can drop UDP traffic. @@ -122,3 +123,34 @@ func (FilterInputDropDifferentUDPPort) ContainerAction(ip net.IP) error { func (FilterInputDropDifferentUDPPort) LocalAction(ip net.IP) error { return sendUDPLoop(ip, acceptPort, sendloopDuration) } + +// FilterInputDropAll tests that we can drop all traffic to the INPUT chain. +type FilterInputDropAll struct{} + +// Name implements TestCase.Name. +func (FilterInputDropAll) Name() string { + return "FilterInputDropAll" +} + +// ContainerAction implements TestCase.ContainerAction. +func (FilterInputDropAll) ContainerAction(ip net.IP) error { + if err := filterTable("-A", "INPUT", "-j", "DROP"); err != nil { + return err + } + + // Listen for all packets on dropPort. + if err := listenUDP(dropPort, sendloopDuration); err == nil { + return fmt.Errorf("packets should have been dropped, but got a packet") + } 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 (FilterInputDropAll) LocalAction(ip net.IP) error { + return sendUDPLoop(ip, dropPort, sendloopDuration) +} diff --git a/test/iptables/iptables_test.go b/test/iptables/iptables_test.go index bfbf1bb87..0c9dee8fe 100644 --- a/test/iptables/iptables_test.go +++ b/test/iptables/iptables_test.go @@ -177,3 +177,9 @@ func TestFilterInputDropDifferentUDPPort(t *testing.T) { t.Fatal(err) } } + +func TestFilterInputDropAll(t *testing.T) { + if err := singleTest(FilterInputDropAll{}); err != nil { + t.Fatal(err) + } +} |