summaryrefslogtreecommitdiffhomepage
path: root/pkg/tcpip
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/tcpip')
-rw-r--r--pkg/tcpip/network/internal/ip/stats.go46
-rw-r--r--pkg/tcpip/network/ipv4/ipv4.go26
-rw-r--r--pkg/tcpip/network/ipv6/ipv6.go26
-rw-r--r--pkg/tcpip/stack/conntrack.go234
-rw-r--r--pkg/tcpip/stack/hook_string.go41
-rw-r--r--pkg/tcpip/stack/iptables.go7
-rw-r--r--pkg/tcpip/stack/iptables_targets.go78
-rw-r--r--pkg/tcpip/stack/packet_buffer.go11
-rw-r--r--pkg/tcpip/tcpip.go4
9 files changed, 325 insertions, 148 deletions
diff --git a/pkg/tcpip/network/internal/ip/stats.go b/pkg/tcpip/network/internal/ip/stats.go
index b6f39ddb1..d06b26309 100644
--- a/pkg/tcpip/network/internal/ip/stats.go
+++ b/pkg/tcpip/network/internal/ip/stats.go
@@ -21,53 +21,56 @@ import "gvisor.dev/gvisor/pkg/tcpip"
// MultiCounterIPStats holds IP statistics, each counter may have several
// versions.
type MultiCounterIPStats struct {
- // PacketsReceived is the total number of IP packets received from the link
- // layer.
+ // PacketsReceived is the number of IP packets received from the link layer.
PacketsReceived tcpip.MultiCounterStat
- // DisabledPacketsReceived is the total number of IP packets received from the
- // link layer when the IP layer is disabled.
+ // DisabledPacketsReceived is the number of IP packets received from the link
+ // layer when the IP layer is disabled.
DisabledPacketsReceived tcpip.MultiCounterStat
- // InvalidDestinationAddressesReceived is the total number of IP packets
- // received with an unknown or invalid destination address.
+ // InvalidDestinationAddressesReceived is the number of IP packets received
+ // with an unknown or invalid destination address.
InvalidDestinationAddressesReceived tcpip.MultiCounterStat
- // InvalidSourceAddressesReceived is the total number of IP packets received
- // with a source address that should never have been received on the wire.
+ // InvalidSourceAddressesReceived is the number of IP packets received with a
+ // source address that should never have been received on the wire.
InvalidSourceAddressesReceived tcpip.MultiCounterStat
- // PacketsDelivered is the total number of incoming IP packets that are
- // successfully delivered to the transport layer.
+ // PacketsDelivered is the number of incoming IP packets that are successfully
+ // delivered to the transport layer.
PacketsDelivered tcpip.MultiCounterStat
- // PacketsSent is the total number of IP packets sent via WritePacket.
+ // PacketsSent is the number of IP packets sent via WritePacket.
PacketsSent tcpip.MultiCounterStat
- // OutgoingPacketErrors is the total number of IP packets which failed to
- // write to a link-layer endpoint.
+ // OutgoingPacketErrors is the number of IP packets which failed to write to a
+ // link-layer endpoint.
OutgoingPacketErrors tcpip.MultiCounterStat
- // MalformedPacketsReceived is the total number of IP Packets that were
- // dropped due to the IP packet header failing validation checks.
+ // MalformedPacketsReceived is the number of IP Packets that were dropped due
+ // to the IP packet header failing validation checks.
MalformedPacketsReceived tcpip.MultiCounterStat
- // MalformedFragmentsReceived is the total number of IP Fragments that were
- // dropped due to the fragment failing validation checks.
+ // MalformedFragmentsReceived is the number of IP Fragments that were dropped
+ // due to the fragment failing validation checks.
MalformedFragmentsReceived tcpip.MultiCounterStat
- // IPTablesPreroutingDropped is the total number of IP packets dropped in the
+ // IPTablesPreroutingDropped is the number of IP packets dropped in the
// Prerouting chain.
IPTablesPreroutingDropped tcpip.MultiCounterStat
- // IPTablesInputDropped is the total number of IP packets dropped in the Input
+ // IPTablesInputDropped is the number of IP packets dropped in the Input
// chain.
IPTablesInputDropped tcpip.MultiCounterStat
- // IPTablesOutputDropped is the total number of IP packets dropped in the
- // Output chain.
+ // IPTablesOutputDropped is the number of IP packets dropped in the Output
+ // chain.
IPTablesOutputDropped tcpip.MultiCounterStat
+ // IPTablesPostroutingDropped is the number of IP packets dropped in the
+ // Postrouting chain.
+ IPTablesPostroutingDropped tcpip.MultiCounterStat
+
// TODO(https://gvisor.dev/issues/5529): Move the IPv4-only option stats out
// of IPStats.
@@ -98,6 +101,7 @@ func (m *MultiCounterIPStats) Init(a, b *tcpip.IPStats) {
m.IPTablesPreroutingDropped.Init(a.IPTablesPreroutingDropped, b.IPTablesPreroutingDropped)
m.IPTablesInputDropped.Init(a.IPTablesInputDropped, b.IPTablesInputDropped)
m.IPTablesOutputDropped.Init(a.IPTablesOutputDropped, b.IPTablesOutputDropped)
+ m.IPTablesPostroutingDropped.Init(a.IPTablesPostroutingDropped, b.IPTablesPostroutingDropped)
m.OptionTimestampReceived.Init(a.OptionTimestampReceived, b.OptionTimestampReceived)
m.OptionRecordRouteReceived.Init(a.OptionRecordRouteReceived, b.OptionRecordRouteReceived)
m.OptionRouterAlertReceived.Init(a.OptionRouterAlertReceived, b.OptionRouterAlertReceived)
diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go
index 2e44f8523..9a3dc78cb 100644
--- a/pkg/tcpip/network/ipv4/ipv4.go
+++ b/pkg/tcpip/network/ipv4/ipv4.go
@@ -415,6 +415,15 @@ func (e *endpoint) writePacket(r *stack.Route, gso *stack.GSO, pkt *stack.Packet
return nil
}
+ // Postrouting NAT can only change the source address, and does not alter the
+ // route or outgoing interface of the packet.
+ outNicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
+ if ok := e.protocol.stack.IPTables().Check(stack.Postrouting, pkt, gso, r, "" /* preroutingAddr */, "" /* inNicName */, outNicName); !ok {
+ // iptables is telling us to drop the packet.
+ e.stats.ip.IPTablesPostroutingDropped.Increment()
+ return nil
+ }
+
stats := e.stats.ip
networkMTU, err := calculateNetworkMTU(e.nic.MTU(), uint32(pkt.NetworkHeader().View().Size()))
@@ -486,9 +495,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
outNicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
// iptables filtering. All packets that reach here are locally
// generated.
- dropped, natPkts := e.protocol.stack.IPTables().CheckPackets(stack.Output, pkts, gso, r, "", outNicName)
- stats.IPTablesOutputDropped.IncrementBy(uint64(len(dropped)))
- for pkt := range dropped {
+ outputDropped, natPkts := e.protocol.stack.IPTables().CheckPackets(stack.Output, pkts, gso, r, "" /* inNicName */, outNicName)
+ stats.IPTablesOutputDropped.IncrementBy(uint64(len(outputDropped)))
+ for pkt := range outputDropped {
pkts.Remove(pkt)
}
@@ -510,6 +519,15 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
}
+ // We ignore the list of NAT-ed packets here because Postrouting NAT can only
+ // change the source address, and does not alter the route or outgoing
+ // interface of the packet.
+ postroutingDropped, _ := e.protocol.stack.IPTables().CheckPackets(stack.Postrouting, pkts, gso, r, "" /* inNicName */, outNicName)
+ stats.IPTablesPostroutingDropped.IncrementBy(uint64(len(postroutingDropped)))
+ for pkt := range postroutingDropped {
+ pkts.Remove(pkt)
+ }
+
// The rest of the packets can be delivered to the NIC as a batch.
pktsLen := pkts.Len()
written, err := e.nic.WritePackets(r, gso, pkts, ProtocolNumber)
@@ -517,7 +535,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
stats.OutgoingPacketErrors.IncrementBy(uint64(pktsLen - written))
// Dropped packets aren't errors, so include them in the return value.
- return locallyDelivered + written + len(dropped), err
+ return locallyDelivered + written + len(outputDropped) + len(postroutingDropped), err
}
// WriteHeaderIncludedPacket implements stack.NetworkEndpoint.
diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go
index d36cefcd0..2e515379c 100644
--- a/pkg/tcpip/network/ipv6/ipv6.go
+++ b/pkg/tcpip/network/ipv6/ipv6.go
@@ -769,6 +769,15 @@ func (e *endpoint) writePacket(r *stack.Route, gso *stack.GSO, pkt *stack.Packet
return nil
}
+ // Postrouting NAT can only change the source address, and does not alter the
+ // route or outgoing interface of the packet.
+ outNicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
+ if ok := e.protocol.stack.IPTables().Check(stack.Postrouting, pkt, gso, r, "" /* preroutingAddr */, "" /* inNicName */, outNicName); !ok {
+ // iptables is telling us to drop the packet.
+ e.stats.ip.IPTablesPostroutingDropped.Increment()
+ return nil
+ }
+
stats := e.stats.ip
networkMTU, err := calculateNetworkMTU(e.nic.MTU(), uint32(pkt.NetworkHeader().View().Size()))
if err != nil {
@@ -840,9 +849,9 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
// iptables filtering. All packets that reach here are locally
// generated.
outNicName := e.protocol.stack.FindNICNameFromID(e.nic.ID())
- dropped, natPkts := e.protocol.stack.IPTables().CheckPackets(stack.Output, pkts, gso, r, "" /* inNicName */, outNicName)
- stats.IPTablesOutputDropped.IncrementBy(uint64(len(dropped)))
- for pkt := range dropped {
+ outputDropped, natPkts := e.protocol.stack.IPTables().CheckPackets(stack.Output, pkts, gso, r, "" /* inNicName */, outNicName)
+ stats.IPTablesOutputDropped.IncrementBy(uint64(len(outputDropped)))
+ for pkt := range outputDropped {
pkts.Remove(pkt)
}
@@ -863,6 +872,15 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
locallyDelivered++
}
+ // We ignore the list of NAT-ed packets here because Postrouting NAT can only
+ // change the source address, and does not alter the route or outgoing
+ // interface of the packet.
+ postroutingDropped, _ := e.protocol.stack.IPTables().CheckPackets(stack.Postrouting, pkts, gso, r, "" /* inNicName */, outNicName)
+ stats.IPTablesPostroutingDropped.IncrementBy(uint64(len(postroutingDropped)))
+ for pkt := range postroutingDropped {
+ pkts.Remove(pkt)
+ }
+
// The rest of the packets can be delivered to the NIC as a batch.
pktsLen := pkts.Len()
written, err := e.nic.WritePackets(r, gso, pkts, ProtocolNumber)
@@ -870,7 +888,7 @@ func (e *endpoint) WritePackets(r *stack.Route, gso *stack.GSO, pkts stack.Packe
stats.OutgoingPacketErrors.IncrementBy(uint64(pktsLen - written))
// Dropped packets aren't errors, so include them in the return value.
- return locallyDelivered + written + len(dropped), err
+ return locallyDelivered + written + len(outputDropped) + len(postroutingDropped), err
}
// WriteHeaderIncludedPacket implements stack.NetworkEndpoint.
diff --git a/pkg/tcpip/stack/conntrack.go b/pkg/tcpip/stack/conntrack.go
index 3f083928f..41e964cf3 100644
--- a/pkg/tcpip/stack/conntrack.go
+++ b/pkg/tcpip/stack/conntrack.go
@@ -16,6 +16,7 @@ package stack
import (
"encoding/binary"
+ "fmt"
"sync"
"time"
@@ -29,7 +30,7 @@ import (
// The connection is created for a packet if it does not exist. Every
// connection contains two tuples (original and reply). The tuples are
// manipulated if there is a matching NAT rule. The packet is modified by
-// looking at the tuples in the Prerouting and Output hooks.
+// looking at the tuples in each hook.
//
// Currently, only TCP tracking is supported.
@@ -46,12 +47,14 @@ const (
)
// Manipulation type for the connection.
+// TODO(gvisor.dev/issue/5696): Define this as a bit set and support SNAT and
+// DNAT at the same time.
type manipType int
const (
manipNone manipType = iota
- manipDstPrerouting
- manipDstOutput
+ manipSource
+ manipDestination
)
// tuple holds a connection's identifying and manipulating data in one
@@ -108,6 +111,7 @@ type conn struct {
reply tuple
// manip indicates if the packet should be manipulated. It is immutable.
+ // TODO(gvisor.dev/issue/5696): Support updating manipulation type.
manip manipType
// tcbHook indicates if the packet is inbound or outbound to
@@ -124,6 +128,18 @@ type conn struct {
lastUsed time.Time `state:".(unixTime)"`
}
+// newConn creates new connection.
+func newConn(orig, reply tupleID, manip manipType, hook Hook) *conn {
+ conn := conn{
+ manip: manip,
+ tcbHook: hook,
+ lastUsed: time.Now(),
+ }
+ conn.original = tuple{conn: &conn, tupleID: orig}
+ conn.reply = tuple{conn: &conn, tupleID: reply, direction: dirReply}
+ return &conn
+}
+
// timedOut returns whether the connection timed out based on its state.
func (cn *conn) timedOut(now time.Time) bool {
const establishedTimeout = 5 * 24 * time.Hour
@@ -219,18 +235,6 @@ func packetToTupleID(pkt *PacketBuffer) (tupleID, tcpip.Error) {
}, nil
}
-// newConn creates new connection.
-func newConn(orig, reply tupleID, manip manipType, hook Hook) *conn {
- conn := conn{
- manip: manip,
- tcbHook: hook,
- lastUsed: time.Now(),
- }
- conn.original = tuple{conn: &conn, tupleID: orig}
- conn.reply = tuple{conn: &conn, tupleID: reply, direction: dirReply}
- return &conn
-}
-
func (ct *ConnTrack) init() {
ct.mu.Lock()
defer ct.mu.Unlock()
@@ -284,20 +288,41 @@ func (ct *ConnTrack) insertRedirectConn(pkt *PacketBuffer, hook Hook, port uint1
return nil
}
- // Create a new connection and change the port as per the iptables
- // rule. This tuple will be used to manipulate the packet in
- // handlePacket.
replyTID := tid.reply()
replyTID.srcAddr = address
replyTID.srcPort = port
- var manip manipType
- switch hook {
- case Prerouting:
- manip = manipDstPrerouting
- case Output:
- manip = manipDstOutput
+
+ conn, _ := ct.connForTID(tid)
+ if conn != nil {
+ // The connection is already tracked.
+ // TODO(gvisor.dev/issue/5696): Support updating an existing connection.
+ return nil
}
- conn := newConn(tid, replyTID, manip, hook)
+ conn = newConn(tid, replyTID, manipDestination, hook)
+ ct.insertConn(conn)
+ return conn
+}
+
+func (ct *ConnTrack) insertSNATConn(pkt *PacketBuffer, hook Hook, port uint16, address tcpip.Address) *conn {
+ tid, err := packetToTupleID(pkt)
+ if err != nil {
+ return nil
+ }
+ if hook != Input && hook != Postrouting {
+ return nil
+ }
+
+ replyTID := tid.reply()
+ replyTID.dstAddr = address
+ replyTID.dstPort = port
+
+ conn, _ := ct.connForTID(tid)
+ if conn != nil {
+ // The connection is already tracked.
+ // TODO(gvisor.dev/issue/5696): Support updating an existing connection.
+ return nil
+ }
+ conn = newConn(tid, replyTID, manipSource, hook)
ct.insertConn(conn)
return conn
}
@@ -322,6 +347,7 @@ func (ct *ConnTrack) insertConn(conn *conn) {
// Now that we hold the locks, ensure the tuple hasn't been inserted by
// another thread.
+ // TODO(gvisor.dev/issue/5773): Should check conn.reply.tupleID, too?
alreadyInserted := false
for other := ct.buckets[tupleBucket].tuples.Front(); other != nil; other = other.Next() {
if other.tupleID == conn.original.tupleID {
@@ -343,86 +369,6 @@ func (ct *ConnTrack) insertConn(conn *conn) {
}
}
-// handlePacketPrerouting manipulates ports for packets in Prerouting hook.
-// TODO(gvisor.dev/issue/170): Change address for Prerouting hook.
-func handlePacketPrerouting(pkt *PacketBuffer, conn *conn, dir direction) {
- // If this is a noop entry, don't do anything.
- if conn.manip == manipNone {
- return
- }
-
- netHeader := pkt.Network()
- tcpHeader := header.TCP(pkt.TransportHeader().View())
-
- // For prerouting redirection, packets going in the original direction
- // have their destinations modified and replies have their sources
- // modified.
- switch dir {
- case dirOriginal:
- port := conn.reply.srcPort
- tcpHeader.SetDestinationPort(port)
- netHeader.SetDestinationAddress(conn.reply.srcAddr)
- case dirReply:
- port := conn.original.dstPort
- tcpHeader.SetSourcePort(port)
- netHeader.SetSourceAddress(conn.original.dstAddr)
- }
-
- // TODO(gvisor.dev/issue/170): TCP checksums aren't usually validated
- // on inbound packets, so we don't recalculate them. However, we should
- // support cases when they are validated, e.g. when we can't offload
- // receive checksumming.
-
- // After modification, IPv4 packets need a valid checksum.
- if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
- netHeader := header.IPv4(pkt.NetworkHeader().View())
- netHeader.SetChecksum(0)
- netHeader.SetChecksum(^netHeader.CalculateChecksum())
- }
-}
-
-// handlePacketOutput manipulates ports for packets in Output hook.
-func handlePacketOutput(pkt *PacketBuffer, conn *conn, gso *GSO, r *Route, dir direction) {
- // If this is a noop entry, don't do anything.
- if conn.manip == manipNone {
- return
- }
-
- netHeader := pkt.Network()
- tcpHeader := header.TCP(pkt.TransportHeader().View())
-
- // For output redirection, packets going in the original direction
- // have their destinations modified and replies have their sources
- // modified. For prerouting redirection, we only reach this point
- // when replying, so packet sources are modified.
- if conn.manip == manipDstOutput && dir == dirOriginal {
- port := conn.reply.srcPort
- tcpHeader.SetDestinationPort(port)
- netHeader.SetDestinationAddress(conn.reply.srcAddr)
- } else {
- port := conn.original.dstPort
- tcpHeader.SetSourcePort(port)
- netHeader.SetSourceAddress(conn.original.dstAddr)
- }
-
- // Calculate the TCP checksum and set it.
- tcpHeader.SetChecksum(0)
- length := uint16(len(tcpHeader) + pkt.Data().Size())
- xsum := header.PseudoHeaderChecksum(header.TCPProtocolNumber, netHeader.SourceAddress(), netHeader.DestinationAddress(), length)
- if gso != nil && gso.NeedsCsum {
- tcpHeader.SetChecksum(xsum)
- } else if r.RequiresTXTransportChecksum() {
- xsum = header.ChecksumCombine(xsum, pkt.Data().AsRange().Checksum())
- tcpHeader.SetChecksum(^tcpHeader.CalculateChecksum(xsum))
- }
-
- if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
- netHeader := header.IPv4(pkt.NetworkHeader().View())
- netHeader.SetChecksum(0)
- netHeader.SetChecksum(^netHeader.CalculateChecksum())
- }
-}
-
// handlePacket will manipulate the port and address of the packet if the
// connection exists. Returns whether, after the packet traverses the tables,
// it should create a new entry in the table.
@@ -431,7 +377,9 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou
return false
}
- if hook != Prerouting && hook != Output {
+ switch hook {
+ case Prerouting, Input, Output, Postrouting:
+ default:
return false
}
@@ -441,23 +389,79 @@ func (ct *ConnTrack) handlePacket(pkt *PacketBuffer, hook Hook, gso *GSO, r *Rou
}
conn, dir := ct.connFor(pkt)
- // Connection or Rule not found for the packet.
+ // Connection not found for the packet.
if conn == nil {
- return true
+ // If this is the last hook in the data path for this packet (Input if
+ // incoming, Postrouting if outgoing), indicate that a connection should be
+ // inserted by the end of this hook.
+ return hook == Input || hook == Postrouting
}
+ netHeader := pkt.Network()
tcpHeader := header.TCP(pkt.TransportHeader().View())
if len(tcpHeader) < header.TCPMinimumSize {
return false
}
+ // TODO(gvisor.dev/issue/5748): TCP checksums on inbound packets should be
+ // validated if checksum offloading is off. It may require IP defrag if the
+ // packets are fragmented.
+
+ switch hook {
+ case Prerouting, Output:
+ if conn.manip == manipDestination {
+ switch dir {
+ case dirOriginal:
+ tcpHeader.SetDestinationPort(conn.reply.srcPort)
+ netHeader.SetDestinationAddress(conn.reply.srcAddr)
+ case dirReply:
+ tcpHeader.SetSourcePort(conn.original.dstPort)
+ netHeader.SetSourceAddress(conn.original.dstAddr)
+ }
+ pkt.NatDone = true
+ }
+ case Input, Postrouting:
+ if conn.manip == manipSource {
+ switch dir {
+ case dirOriginal:
+ tcpHeader.SetSourcePort(conn.reply.dstPort)
+ netHeader.SetSourceAddress(conn.reply.dstAddr)
+ case dirReply:
+ tcpHeader.SetDestinationPort(conn.original.srcPort)
+ netHeader.SetDestinationAddress(conn.original.srcAddr)
+ }
+ pkt.NatDone = true
+ }
+ default:
+ panic(fmt.Sprintf("unrecognized hook = %s", hook))
+ }
+ if !pkt.NatDone {
+ return false
+ }
+
switch hook {
- case Prerouting:
- handlePacketPrerouting(pkt, conn, dir)
- case Output:
- handlePacketOutput(pkt, conn, gso, r, dir)
+ case Prerouting, Input:
+ case Output, Postrouting:
+ // Calculate the TCP checksum and set it.
+ tcpHeader.SetChecksum(0)
+ length := uint16(len(tcpHeader) + pkt.Data().Size())
+ xsum := header.PseudoHeaderChecksum(header.TCPProtocolNumber, netHeader.SourceAddress(), netHeader.DestinationAddress(), length)
+ if gso != nil && gso.NeedsCsum {
+ tcpHeader.SetChecksum(xsum)
+ } else if r.RequiresTXTransportChecksum() {
+ xsum = header.ChecksumCombine(xsum, pkt.Data().AsRange().Checksum())
+ tcpHeader.SetChecksum(^tcpHeader.CalculateChecksum(xsum))
+ }
+ default:
+ panic(fmt.Sprintf("unrecognized hook = %s", hook))
+ }
+
+ // After modification, IPv4 packets need a valid checksum.
+ if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ netHeader.SetChecksum(0)
+ netHeader.SetChecksum(^netHeader.CalculateChecksum())
}
- pkt.NatDone = true
// Update the state of tcb.
// TODO(gvisor.dev/issue/170): Add support in tcpcontrack to handle
@@ -638,8 +642,8 @@ func (ct *ConnTrack) originalDst(epID TransportEndpointID, netProto tcpip.Networ
if conn == nil {
// Not a tracked connection.
return "", 0, &tcpip.ErrNotConnected{}
- } else if conn.manip == manipNone {
- // Unmanipulated connection.
+ } else if conn.manip != manipDestination {
+ // Unmanipulated destination.
return "", 0, &tcpip.ErrInvalidOptionValue{}
}
diff --git a/pkg/tcpip/stack/hook_string.go b/pkg/tcpip/stack/hook_string.go
new file mode 100644
index 000000000..3dc8a7b02
--- /dev/null
+++ b/pkg/tcpip/stack/hook_string.go
@@ -0,0 +1,41 @@
+// Copyright 2021 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at //
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Code generated by "stringer -type Hook ."; DO NOT EDIT.
+
+package stack
+
+import "strconv"
+
+func _() {
+ // An "invalid array index" compiler error signifies that the constant values have changed.
+ // Re-run the stringer command to generate them again.
+ var x [1]struct{}
+ _ = x[Prerouting-0]
+ _ = x[Input-1]
+ _ = x[Forward-2]
+ _ = x[Output-3]
+ _ = x[Postrouting-4]
+ _ = x[NumHooks-5]
+}
+
+const _Hook_name = "PreroutingInputForwardOutputPostroutingNumHooks"
+
+var _Hook_index = [...]uint8{0, 10, 15, 22, 28, 39, 47}
+
+func (i Hook) String() string {
+ if i >= Hook(len(_Hook_index)-1) {
+ return "Hook(" + strconv.FormatInt(int64(i), 10) + ")"
+ }
+ return _Hook_name[_Hook_index[i]:_Hook_index[i+1]]
+}
diff --git a/pkg/tcpip/stack/iptables.go b/pkg/tcpip/stack/iptables.go
index 52890f6eb..7ea87d325 100644
--- a/pkg/tcpip/stack/iptables.go
+++ b/pkg/tcpip/stack/iptables.go
@@ -175,9 +175,10 @@ func DefaultTables() *IPTables {
},
},
priorities: [NumHooks][]TableID{
- Prerouting: {MangleID, NATID},
- Input: {NATID, FilterID},
- Output: {MangleID, NATID, FilterID},
+ Prerouting: {MangleID, NATID},
+ Input: {NATID, FilterID},
+ Output: {MangleID, NATID, FilterID},
+ Postrouting: {MangleID, NATID},
},
connections: ConnTrack{
seed: generateRandUint32(),
diff --git a/pkg/tcpip/stack/iptables_targets.go b/pkg/tcpip/stack/iptables_targets.go
index 0e8b90c9b..317efe754 100644
--- a/pkg/tcpip/stack/iptables_targets.go
+++ b/pkg/tcpip/stack/iptables_targets.go
@@ -182,3 +182,81 @@ func (rt *RedirectTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gs
return RuleAccept, 0
}
+
+// SNATTarget modifies the source port/IP in the outgoing packets.
+type SNATTarget struct {
+ Addr tcpip.Address
+ Port uint16
+
+ // NetworkProtocol is the network protocol the target is used with. It
+ // is immutable.
+ NetworkProtocol tcpip.NetworkProtocolNumber
+}
+
+// Action implements Target.Action.
+func (st *SNATTarget) Action(pkt *PacketBuffer, ct *ConnTrack, hook Hook, gso *GSO, r *Route, address tcpip.Address) (RuleVerdict, int) {
+ // Sanity check.
+ if st.NetworkProtocol != pkt.NetworkProtocolNumber {
+ panic(fmt.Sprintf(
+ "SNATTarget.Action with NetworkProtocol %d called on packet with NetworkProtocolNumber %d",
+ st.NetworkProtocol, pkt.NetworkProtocolNumber))
+ }
+
+ // Packet is already manipulated.
+ if pkt.NatDone {
+ return RuleAccept, 0
+ }
+
+ // Drop the packet if network and transport header are not set.
+ if pkt.NetworkHeader().View().IsEmpty() || pkt.TransportHeader().View().IsEmpty() {
+ return RuleDrop, 0
+ }
+
+ switch hook {
+ case Postrouting, Input:
+ case Prerouting, Output, Forward:
+ panic(fmt.Sprintf("%s not supported", hook))
+ default:
+ panic(fmt.Sprintf("%s unrecognized", hook))
+ }
+
+ switch protocol := pkt.TransportProtocolNumber; protocol {
+ case header.UDPProtocolNumber:
+ udpHeader := header.UDP(pkt.TransportHeader().View())
+ udpHeader.SetChecksum(0)
+ udpHeader.SetSourcePort(st.Port)
+ netHeader := pkt.Network()
+ netHeader.SetSourceAddress(st.Addr)
+
+ // Only calculate the checksum if offloading isn't supported.
+ if r.RequiresTXTransportChecksum() {
+ length := uint16(pkt.Size()) - uint16(len(pkt.NetworkHeader().View()))
+ xsum := header.PseudoHeaderChecksum(protocol, netHeader.SourceAddress(), netHeader.DestinationAddress(), length)
+ xsum = header.ChecksumCombine(xsum, pkt.Data().AsRange().Checksum())
+ udpHeader.SetChecksum(^udpHeader.CalculateChecksum(xsum))
+ }
+
+ // After modification, IPv4 packets need a valid checksum.
+ if pkt.NetworkProtocolNumber == header.IPv4ProtocolNumber {
+ netHeader := header.IPv4(pkt.NetworkHeader().View())
+ netHeader.SetChecksum(0)
+ netHeader.SetChecksum(^netHeader.CalculateChecksum())
+ }
+ pkt.NatDone = true
+ case header.TCPProtocolNumber:
+ if ct == nil {
+ return RuleAccept, 0
+ }
+
+ // Set up conection for matching NAT rule. Only the first
+ // packet of the connection comes here. Other packets will be
+ // manipulated in connection tracking.
+ if conn := ct.insertSNATConn(pkt, hook, st.Port, st.Addr); conn != nil {
+ ct.handlePacket(pkt, hook, gso, r)
+ }
+ default:
+ return RuleDrop, 0
+ }
+
+ return RuleAccept, 0
+}
diff --git a/pkg/tcpip/stack/packet_buffer.go b/pkg/tcpip/stack/packet_buffer.go
index 8f288675d..c10304d5f 100644
--- a/pkg/tcpip/stack/packet_buffer.go
+++ b/pkg/tcpip/stack/packet_buffer.go
@@ -299,9 +299,18 @@ func (pk *PacketBuffer) Network() header.Network {
// See PacketBuffer.Data for details about how a packet buffer holds an inbound
// packet.
func (pk *PacketBuffer) CloneToInbound() *PacketBuffer {
- return NewPacketBuffer(PacketBufferOptions{
+ newPk := NewPacketBuffer(PacketBufferOptions{
Data: buffer.NewVectorisedView(pk.Size(), pk.Views()),
})
+ // TODO(gvisor.dev/issue/5696): reimplement conntrack so that no need to
+ // maintain this flag in the packet. Currently conntrack needs this flag to
+ // tell if a noop connection should be inserted at Input hook. Once conntrack
+ // redefines the manipulation field as mutable, we won't need the special noop
+ // connection.
+ if pk.NatDone {
+ newPk.NatDone = true
+ }
+ return newPk
}
// headerInfo stores metadata about a header in a packet.
diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go
index 60de16579..2b6e6a89f 100644
--- a/pkg/tcpip/tcpip.go
+++ b/pkg/tcpip/tcpip.go
@@ -1556,6 +1556,10 @@ type IPStats struct {
// chain.
IPTablesOutputDropped *StatCounter
+ // IPTablesPostroutingDropped is the number of IP packets dropped in the
+ // Postrouting chain.
+ IPTablesPostroutingDropped *StatCounter
+
// TODO(https://gvisor.dev/issues/5529): Move the IPv4-only option stats out
// of IPStats.
// OptionTimestampReceived is the number of Timestamp options seen.
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-27 03:14:35 +0000