summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip/stack/iptables.go
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip/stack/iptables.go')
-rw-r--r--pkg/tcpip/stack/iptables.go216
1 files changed, 155 insertions, 61 deletions
diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go
index f152c0d83..fd61387bf 100644
--- a/pkg/tcpip/stack/iptables.go
+++ b/pkg/tcpip/stack/iptables.go
@@ -42,7 +42,7 @@ const reaperDelay = 5 * time.Second
// DefaultTables returns a default set of tables. Each chain is set to accept
// all packets.
-func DefaultTables(seed uint32) *IPTables {
+func DefaultTables(seed uint32, clock tcpip.Clock) *IPTables {
return &IPTables{
v4Tables: [NumTables]Table{
NATID: {
@@ -182,7 +182,8 @@ func DefaultTables(seed uint32) *IPTables {
Postrouting: {MangleID, NATID},
},
connections: ConnTrack{
- seed: seed,
+ seed: seed,
+ clock: clock,
},
reaperDone: make(chan struct{}, 1),
}
@@ -264,33 +265,125 @@ const (
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.
+// CheckPrerouting performs the prerouting hook on the packet.
+//
+// Returns true iff the packet may continue traversing the stack; the packet
+// must be dropped if false is returned.
+//
+// Precondition: The packet's network and transport header must be set.
+func (it *IPTables) CheckPrerouting(pkt *PacketBuffer, addressEP AddressableEndpoint, inNicName string) bool {
+ const hook = Prerouting
+
+ if it.shouldSkip(pkt.NetworkProtocolNumber) {
+ return true
+ }
+
+ pkt.tuple = it.connections.getConnOrMaybeInsertNoop(pkt)
+
+ return it.check(hook, pkt, nil /* route */, addressEP, inNicName, "" /* outNicName */)
+}
+
+// CheckInput performs the input hook on the packet.
+//
+// Returns true iff the packet may continue traversing the stack; the packet
+// must be dropped if false is returned.
+//
+// Precondition: The packet's network and transport header must be set.
+func (it *IPTables) CheckInput(pkt *PacketBuffer, inNicName string) bool {
+ const hook = Input
+
+ if it.shouldSkip(pkt.NetworkProtocolNumber) {
+ return true
+ }
+
+ ret := it.check(hook, pkt, nil /* route */, nil /* addressEP */, inNicName, "" /* outNicName */)
+ if t := pkt.tuple; t != nil {
+ t.conn.finalize()
+ }
+ pkt.tuple = nil
+ return ret
+}
+
+// CheckForward performs the forward hook on the packet.
+//
+// Returns true iff the packet may continue traversing the stack; the packet
+// must be dropped if false is returned.
+//
+// Precondition: The packet's network and transport header must be set.
+func (it *IPTables) CheckForward(pkt *PacketBuffer, inNicName, outNicName string) bool {
+ if it.shouldSkip(pkt.NetworkProtocolNumber) {
+ return true
+ }
+ return it.check(Forward, pkt, nil /* route */, nil /* addressEP */, inNicName, outNicName)
+}
+
+// CheckOutput performs the output hook on the packet.
+//
+// Returns true iff the packet may continue traversing the stack; the packet
+// must be dropped if false is returned.
+//
+// Precondition: The packet's network and transport header must be set.
+func (it *IPTables) CheckOutput(pkt *PacketBuffer, r *Route, outNicName string) bool {
+ const hook = Output
+
+ if it.shouldSkip(pkt.NetworkProtocolNumber) {
+ return true
+ }
+
+ pkt.tuple = it.connections.getConnOrMaybeInsertNoop(pkt)
+
+ return it.check(hook, pkt, r, nil /* addressEP */, "" /* inNicName */, outNicName)
+}
+
+// CheckPostrouting performs the postrouting hook on the packet.
//
-// Precondition: pkt.NetworkHeader is set.
-func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, r *Route, preroutingAddr tcpip.Address, inNicName, outNicName string) bool {
- if pkt.NetworkProtocolNumber != header.IPv4ProtocolNumber && pkt.NetworkProtocolNumber != header.IPv6ProtocolNumber {
+// Returns true iff the packet may continue traversing the stack; the packet
+// must be dropped if false is returned.
+//
+// Precondition: The packet's network and transport header must be set.
+func (it *IPTables) CheckPostrouting(pkt *PacketBuffer, r *Route, addressEP AddressableEndpoint, outNicName string) bool {
+ const hook = Postrouting
+
+ if it.shouldSkip(pkt.NetworkProtocolNumber) {
return true
}
+
+ ret := it.check(hook, pkt, r, addressEP, "" /* inNicName */, outNicName)
+ if t := pkt.tuple; t != nil {
+ t.conn.finalize()
+ }
+ pkt.tuple = nil
+ return ret
+}
+
+func (it *IPTables) shouldSkip(netProto tcpip.NetworkProtocolNumber) bool {
+ switch netProto {
+ case header.IPv4ProtocolNumber, header.IPv6ProtocolNumber:
+ default:
+ // IPTables only supports IPv4/IPv6.
+ return true
+ }
+
+ it.mu.RLock()
+ defer it.mu.RUnlock()
// Many users never configure iptables. Spare them the cost of rule
// traversal if rules have never been set.
+ return !it.modified
+}
+
+// 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: The packet's network and transport header must be set.
+func (it *IPTables) check(hook Hook, pkt *PacketBuffer, r *Route, addressEP AddressableEndpoint, inNicName, outNicName string) bool {
it.mu.RLock()
defer it.mu.RUnlock()
- if !it.modified {
- return true
- }
-
- // Packets are manipulated only if connection and matching
- // NAT rule exists.
- shouldTrack := it.connections.handlePacket(pkt, hook, r)
// Go through each table containing the hook.
priorities := it.priorities[hook]
for _, tableID := range priorities {
- // If handlePacket already NATed the packet, we don't need to
- // check the NAT table.
- if tableID == NATID && pkt.NatDone {
+ if t := pkt.tuple; t != nil && tableID == NATID && t.conn.handlePacket(pkt, hook, r) {
continue
}
var table Table
@@ -300,7 +393,7 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, r *Route, preroutingAddr
table = it.v4Tables[tableID]
}
ruleIdx := table.BuiltinChains[hook]
- switch verdict := it.checkChain(hook, pkt, table, ruleIdx, r, preroutingAddr, inNicName, outNicName); verdict {
+ switch verdict := it.checkChain(hook, pkt, table, ruleIdx, r, addressEP, inNicName, outNicName); verdict {
// If the table returns Accept, move on to the next table.
case chainAccept:
continue
@@ -311,7 +404,7 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, r *Route, preroutingAddr
// 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, &it.connections, hook, r, preroutingAddr); v {
+ switch v, _ := underflow.Target.Action(pkt, hook, r, addressEP); v {
case RuleAccept:
continue
case RuleDrop:
@@ -327,21 +420,6 @@ func (it *IPTables) Check(hook Hook, pkt *PacketBuffer, r *Route, preroutingAddr
}
}
- // If this connection should be tracked, try to add an entry for it. If
- // traversing the nat table didn't end in adding an entry,
- // maybeInsertNoop will add a no-op entry for the connection. This is
- // needeed when establishing connections so that the SYN/ACK reply to an
- // outgoing SYN is delivered to the correct endpoint rather than being
- // redirected by a prerouting rule.
- //
- // From the iptables documentation: "If there is no rule, a `null'
- // binding is created: this usually does not map the packet, but exists
- // to ensure we don't map another stream over an existing one."
- if shouldTrack {
- it.connections.maybeInsertNoop(pkt, hook)
- }
-
- // Every table returned Accept.
return true
}
@@ -375,30 +453,46 @@ func (it *IPTables) startReaper(interval time.Duration) {
}()
}
-// CheckPackets runs pkts through the rules for hook and returns a map of packets that
-// should not go forward.
+// CheckOutputPackets performs the output hook on the packets.
//
-// Preconditions:
-// * pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
-// * pkt.NetworkHeader is not nil.
+// Returns a map of packets that must be dropped.
//
-// NOTE: unlike the Check API the returned map contains packets that should be
-// dropped.
-func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, r *Route, inNicName, outNicName string) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) {
+// Precondition: The packets' network and transport header must be set.
+func (it *IPTables) CheckOutputPackets(pkts PacketBufferList, r *Route, outNicName string) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) {
+ return checkPackets(pkts, func(pkt *PacketBuffer) bool {
+ return it.CheckOutput(pkt, r, outNicName)
+ }, true /* dnat */)
+}
+
+// CheckPostroutingPackets performs the postrouting hook on the packets.
+//
+// Returns a map of packets that must be dropped.
+//
+// Precondition: The packets' network and transport header must be set.
+func (it *IPTables) CheckPostroutingPackets(pkts PacketBufferList, r *Route, addressEP AddressableEndpoint, outNicName string) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) {
+ return checkPackets(pkts, func(pkt *PacketBuffer) bool {
+ return it.CheckPostrouting(pkt, r, addressEP, outNicName)
+ }, false /* dnat */)
+}
+
+func checkPackets(pkts PacketBufferList, f func(*PacketBuffer) bool, dnat bool) (drop map[*PacketBuffer]struct{}, natPkts map[*PacketBuffer]struct{}) {
for pkt := pkts.Front(); pkt != nil; pkt = pkt.Next() {
- if !pkt.NatDone {
- if ok := it.Check(hook, pkt, r, "", inNicName, outNicName); !ok {
- if drop == nil {
- drop = make(map[*PacketBuffer]struct{})
- }
- drop[pkt] = struct{}{}
+ natDone := &pkt.SNATDone
+ if dnat {
+ natDone = &pkt.DNATDone
+ }
+
+ if ok := f(pkt); !ok {
+ if drop == nil {
+ drop = make(map[*PacketBuffer]struct{})
}
- if pkt.NatDone {
- if natPkts == nil {
- natPkts = make(map[*PacketBuffer]struct{})
- }
- natPkts[pkt] = struct{}{}
+ drop[pkt] = struct{}{}
+ }
+ if *natDone {
+ if natPkts == nil {
+ natPkts = make(map[*PacketBuffer]struct{})
}
+ natPkts[pkt] = struct{}{}
}
}
return drop, natPkts
@@ -407,11 +501,11 @@ func (it *IPTables) CheckPackets(hook Hook, pkts PacketBufferList, r *Route, inN
// Preconditions:
// * pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
// * pkt.NetworkHeader is not nil.
-func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, r *Route, preroutingAddr tcpip.Address, inNicName, outNicName string) chainVerdict {
+func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, r *Route, addressEP AddressableEndpoint, inNicName, outNicName string) chainVerdict {
// Start from ruleIdx and walk the list of rules until a rule gives us
// a verdict.
for ruleIdx < len(table.Rules) {
- switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, r, preroutingAddr, inNicName, outNicName); verdict {
+ switch verdict, jumpTo := it.checkRule(hook, pkt, table, ruleIdx, r, addressEP, inNicName, outNicName); verdict {
case RuleAccept:
return chainAccept
@@ -428,7 +522,7 @@ func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleId
ruleIdx++
continue
}
- switch verdict := it.checkChain(hook, pkt, table, jumpTo, r, preroutingAddr, inNicName, outNicName); verdict {
+ switch verdict := it.checkChain(hook, pkt, table, jumpTo, r, addressEP, inNicName, outNicName); verdict {
case chainAccept:
return chainAccept
case chainDrop:
@@ -454,7 +548,7 @@ func (it *IPTables) checkChain(hook Hook, pkt *PacketBuffer, table Table, ruleId
// Preconditions:
// * pkt is a IPv4 packet of at least length header.IPv4MinimumSize.
// * pkt.NetworkHeader is not nil.
-func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, r *Route, preroutingAddr tcpip.Address, inNicName, outNicName string) (RuleVerdict, int) {
+func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx int, r *Route, addressEP AddressableEndpoint, inNicName, outNicName string) (RuleVerdict, int) {
rule := table.Rules[ruleIdx]
// Check whether the packet matches the IP header filter.
@@ -477,16 +571,16 @@ func (it *IPTables) checkRule(hook Hook, pkt *PacketBuffer, table Table, ruleIdx
}
// All the matchers matched, so run the target.
- return rule.Target.Action(pkt, &it.connections, hook, r, preroutingAddr)
+ return rule.Target.Action(pkt, hook, r, addressEP)
}
// OriginalDst returns the original destination of redirected connections. It
// returns an error if the connection doesn't exist or isn't redirected.
-func (it *IPTables) OriginalDst(epID TransportEndpointID, netProto tcpip.NetworkProtocolNumber) (tcpip.Address, uint16, tcpip.Error) {
+func (it *IPTables) OriginalDst(epID TransportEndpointID, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber) (tcpip.Address, uint16, tcpip.Error) {
it.mu.RLock()
defer it.mu.RUnlock()
if !it.modified {
return "", 0, &tcpip.ErrNotConnected{}
}
- return it.connections.originalDst(epID, netProto)
+ return it.connections.originalDst(epID, netProto, transProto)
}