netlink/conntrack_test.go

408 lines
12 KiB
Go

// +build linux
package netlink
import (
"fmt"
"net"
"runtime"
"testing"
"github.com/vishvananda/netns"
"golang.org/x/sys/unix"
)
func CheckErrorFail(t *testing.T, err error) {
if err != nil {
t.Fatalf("Fatal Error: %s", err)
}
}
func CheckError(t *testing.T, err error) {
if err != nil {
t.Errorf("Error: %s", err)
}
}
func udpFlowCreateProg(t *testing.T, flows, srcPort int, dstIP string, dstPort int) {
for i := 0; i < flows; i++ {
ServerAddr, err := net.ResolveUDPAddr("udp", fmt.Sprintf("%s:%d", dstIP, dstPort))
CheckError(t, err)
LocalAddr, err := net.ResolveUDPAddr("udp", fmt.Sprintf("127.0.0.1:%d", srcPort+i))
CheckError(t, err)
Conn, err := net.DialUDP("udp", LocalAddr, ServerAddr)
CheckError(t, err)
Conn.Write([]byte("Hello World"))
Conn.Close()
}
}
func nsCreateAndEnter(t *testing.T) (*netns.NsHandle, *netns.NsHandle, *Handle) {
// Lock the OS Thread so we don't accidentally switch namespaces
runtime.LockOSThread()
// Save the current network namespace
origns, _ := netns.Get()
ns, err := netns.New()
CheckErrorFail(t, err)
h, err := NewHandleAt(ns)
CheckErrorFail(t, err)
// Enter the new namespace
netns.Set(ns)
// Bing up loopback
link, _ := h.LinkByName("lo")
h.LinkSetUp(link)
return &origns, &ns, h
}
func applyFilter(flowList []ConntrackFlow, ipv4Filter *ConntrackFilter, ipv6Filter *ConntrackFilter) (ipv4Match, ipv6Match uint) {
for _, flow := range flowList {
if ipv4Filter.MatchConntrackFlow(&flow) == true {
ipv4Match++
}
if ipv6Filter.MatchConntrackFlow(&flow) == true {
ipv6Match++
}
}
return ipv4Match, ipv6Match
}
// TestConntrackSocket test the opening of a NETFILTER family socket
func TestConntrackSocket(t *testing.T) {
skipUnlessRoot(t)
setUpNetlinkTestWithKModule(t, "nf_conntrack")
setUpNetlinkTestWithKModule(t, "nf_conntrack_netlink")
h, err := NewHandle(unix.NETLINK_NETFILTER)
CheckErrorFail(t, err)
if h.SupportsNetlinkFamily(unix.NETLINK_NETFILTER) != true {
t.Fatal("ERROR not supporting the NETFILTER family")
}
}
// TestConntrackTableList test the conntrack table list
// Creates some flows and checks that they are correctly fetched from the conntrack table
func TestConntrackTableList(t *testing.T) {
skipUnlessRoot(t)
setUpNetlinkTestWithKModule(t, "nf_conntrack")
setUpNetlinkTestWithKModule(t, "nf_conntrack_netlink")
setUpNetlinkTestWithKModule(t, "nf_conntrack_ipv4")
setUpNetlinkTestWithKModule(t, "nf_conntrack_ipv6")
// Creates a new namespace and bring up the loopback interface
origns, ns, h := nsCreateAndEnter(t)
defer netns.Set(*origns)
defer origns.Close()
defer ns.Close()
defer runtime.UnlockOSThread()
setUpF(t, "/proc/sys/net/netfilter/nf_conntrack_acct", "1")
// Flush the table to start fresh
err := h.ConntrackTableFlush(ConntrackTable)
CheckErrorFail(t, err)
// Create 5 udp
udpFlowCreateProg(t, 5, 2000, "127.0.0.10", 3000)
// Fetch the conntrack table
flows, err := h.ConntrackTableList(ConntrackTable, unix.AF_INET)
CheckErrorFail(t, err)
// Check that it is able to find the 5 flows created
var found int
for _, flow := range flows {
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.10")) &&
flow.Forward.DstPort == 3000 &&
(flow.Forward.SrcPort >= 2000 && flow.Forward.SrcPort <= 2005) {
found++
}
if flow.Forward.Bytes == 0 && flow.Forward.Packets == 0 && flow.Reverse.Bytes == 0 && flow.Reverse.Packets == 0 {
t.Error("No traffic statistics are collected")
}
}
if found != 5 {
t.Fatalf("Found only %d flows over 5", found)
}
// Give a try also to the IPv6 version
_, err = h.ConntrackTableList(ConntrackTable, unix.AF_INET6)
CheckErrorFail(t, err)
// Switch back to the original namespace
netns.Set(*origns)
}
// TestConntrackTableFlush test the conntrack table flushing
// Creates some flows and then call the table flush
func TestConntrackTableFlush(t *testing.T) {
skipUnlessRoot(t)
setUpNetlinkTestWithKModule(t, "nf_conntrack")
setUpNetlinkTestWithKModule(t, "nf_conntrack_netlink")
setUpNetlinkTestWithKModule(t, "nf_conntrack_ipv4")
// Creates a new namespace and bring up the loopback interface
origns, ns, h := nsCreateAndEnter(t)
defer netns.Set(*origns)
defer origns.Close()
defer ns.Close()
defer runtime.UnlockOSThread()
// Create 5 udp flows using netcat
udpFlowCreateProg(t, 5, 3000, "127.0.0.10", 4000)
// Fetch the conntrack table
flows, err := h.ConntrackTableList(ConntrackTable, unix.AF_INET)
CheckErrorFail(t, err)
// Check that it is able to find the 5 flows created
var found int
for _, flow := range flows {
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.10")) &&
flow.Forward.DstPort == 4000 &&
(flow.Forward.SrcPort >= 3000 && flow.Forward.SrcPort <= 3005) {
found++
}
}
if found != 5 {
t.Fatalf("Found only %d flows over 5", found)
}
// Flush the table
err = h.ConntrackTableFlush(ConntrackTable)
CheckErrorFail(t, err)
// Fetch again the flows to validate the flush
flows, err = h.ConntrackTableList(ConntrackTable, unix.AF_INET)
CheckErrorFail(t, err)
// Check if it is still able to find the 5 flows created
found = 0
for _, flow := range flows {
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.10")) &&
flow.Forward.DstPort == 4000 &&
(flow.Forward.SrcPort >= 3000 && flow.Forward.SrcPort <= 3005) {
found++
}
}
if found > 0 {
t.Fatalf("Found %d flows, they should had been flushed", found)
}
// Switch back to the original namespace
netns.Set(*origns)
}
// TestConntrackTableDelete tests the deletion with filter
// Creates 2 group of flows then deletes only one group and validates the result
func TestConntrackTableDelete(t *testing.T) {
skipUnlessRoot(t)
setUpNetlinkTestWithKModule(t, "nf_conntrack")
setUpNetlinkTestWithKModule(t, "nf_conntrack_netlink")
setUpNetlinkTestWithKModule(t, "nf_conntrack_ipv4")
// Creates a new namespace and bring up the loopback interface
origns, ns, h := nsCreateAndEnter(t)
defer netns.Set(*origns)
defer origns.Close()
defer ns.Close()
defer runtime.UnlockOSThread()
// Create 10 udp flows
udpFlowCreateProg(t, 5, 5000, "127.0.0.10", 6000)
udpFlowCreateProg(t, 5, 7000, "127.0.0.20", 8000)
// Fetch the conntrack table
flows, err := h.ConntrackTableList(ConntrackTable, unix.AF_INET)
CheckErrorFail(t, err)
// Check that it is able to find the 5 flows created for each group
var groupA int
var groupB int
for _, flow := range flows {
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.10")) &&
flow.Forward.DstPort == 6000 &&
(flow.Forward.SrcPort >= 5000 && flow.Forward.SrcPort <= 5005) {
groupA++
}
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.20")) &&
flow.Forward.DstPort == 8000 &&
(flow.Forward.SrcPort >= 7000 && flow.Forward.SrcPort <= 7005) {
groupB++
}
}
if groupA != 5 || groupB != 5 {
t.Fatalf("Flow creation issue groupA:%d, groupB:%d", groupA, groupB)
}
// Create a filter to erase groupB flows
filter := &ConntrackFilter{}
filter.AddIP(ConntrackOrigDstIP, net.ParseIP("127.0.0.20"))
// Flush entries of groupB
var deleted uint
if deleted, err = h.ConntrackDeleteFilter(ConntrackTable, unix.AF_INET, filter); err != nil {
t.Fatalf("Error during the erase: %s", err)
}
if deleted != 5 {
t.Fatalf("Error deleted a wrong number of flows:%d instead of 5", deleted)
}
// Check again the table to verify that are gone
flows, err = h.ConntrackTableList(ConntrackTable, unix.AF_INET)
CheckErrorFail(t, err)
// Check if it is able to find the 5 flows of groupA but none of groupB
groupA = 0
groupB = 0
for _, flow := range flows {
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.10")) &&
flow.Forward.DstPort == 6000 &&
(flow.Forward.SrcPort >= 5000 && flow.Forward.SrcPort <= 5005) {
groupA++
}
if flow.Forward.Protocol == 17 &&
flow.Forward.DstIP.Equal(net.ParseIP("127.0.0.20")) &&
flow.Forward.DstPort == 8000 &&
(flow.Forward.SrcPort >= 7000 && flow.Forward.SrcPort <= 7005) {
groupB++
}
}
if groupA != 5 || groupB > 0 {
t.Fatalf("Error during the erase groupA:%d, groupB:%d", groupA, groupB)
}
// Switch back to the original namespace
netns.Set(*origns)
}
func TestConntrackFilter(t *testing.T) {
var flowList []ConntrackFlow
flowList = append(flowList, ConntrackFlow{
FamilyType: unix.AF_INET,
Forward: ipTuple{
SrcIP: net.ParseIP("10.0.0.1"),
DstIP: net.ParseIP("20.0.0.1"),
SrcPort: 1000,
DstPort: 2000,
},
Reverse: ipTuple{
SrcIP: net.ParseIP("20.0.0.1"),
DstIP: net.ParseIP("192.168.1.1"),
SrcPort: 2000,
DstPort: 1000,
},
},
ConntrackFlow{
FamilyType: unix.AF_INET,
Forward: ipTuple{
SrcIP: net.ParseIP("10.0.0.2"),
DstIP: net.ParseIP("20.0.0.2"),
SrcPort: 5000,
DstPort: 6000,
},
Reverse: ipTuple{
SrcIP: net.ParseIP("20.0.0.2"),
DstIP: net.ParseIP("192.168.1.1"),
SrcPort: 6000,
DstPort: 5000,
},
},
ConntrackFlow{
FamilyType: unix.AF_INET6,
Forward: ipTuple{
SrcIP: net.ParseIP("eeee:eeee:eeee:eeee:eeee:eeee:eeee:eeee"),
DstIP: net.ParseIP("dddd:dddd:dddd:dddd:dddd:dddd:dddd:dddd"),
SrcPort: 1000,
DstPort: 2000,
},
Reverse: ipTuple{
SrcIP: net.ParseIP("dddd:dddd:dddd:dddd:dddd:dddd:dddd:dddd"),
DstIP: net.ParseIP("eeee:eeee:eeee:eeee:eeee:eeee:eeee:eeee"),
SrcPort: 2000,
DstPort: 1000,
},
})
// Empty filter
v4Match, v6Match := applyFilter(flowList, &ConntrackFilter{}, &ConntrackFilter{})
if v4Match > 0 || v6Match > 0 {
t.Fatalf("Error, empty filter cannot match, v4:%d, v6:%d", v4Match, v6Match)
}
// SrcIP filter
filterV4 := &ConntrackFilter{}
filterV4.AddIP(ConntrackOrigSrcIP, net.ParseIP("10.0.0.1"))
filterV6 := &ConntrackFilter{}
filterV6.AddIP(ConntrackOrigSrcIP, net.ParseIP("eeee:eeee:eeee:eeee:eeee:eeee:eeee:eeee"))
v4Match, v6Match = applyFilter(flowList, filterV4, filterV6)
if v4Match != 1 || v6Match != 1 {
t.Fatalf("Error, there should be only 1 match, v4:%d, v6:%d", v4Match, v6Match)
}
// DstIp filter
filterV4 = &ConntrackFilter{}
filterV4.AddIP(ConntrackOrigDstIP, net.ParseIP("20.0.0.1"))
filterV6 = &ConntrackFilter{}
filterV6.AddIP(ConntrackOrigDstIP, net.ParseIP("dddd:dddd:dddd:dddd:dddd:dddd:dddd:dddd"))
v4Match, v6Match = applyFilter(flowList, filterV4, filterV6)
if v4Match != 1 || v6Match != 1 {
t.Fatalf("Error, there should be only 1 match, v4:%d, v6:%d", v4Match, v6Match)
}
// SrcIP for NAT
filterV4 = &ConntrackFilter{}
filterV4.AddIP(ConntrackNatSrcIP, net.ParseIP("20.0.0.1"))
filterV6 = &ConntrackFilter{}
filterV6.AddIP(ConntrackNatSrcIP, net.ParseIP("dddd:dddd:dddd:dddd:dddd:dddd:dddd:dddd"))
v4Match, v6Match = applyFilter(flowList, filterV4, filterV6)
if v4Match != 1 || v6Match != 1 {
t.Fatalf("Error, there should be only 1 match, v4:%d, v6:%d", v4Match, v6Match)
}
// DstIP for NAT
filterV4 = &ConntrackFilter{}
filterV4.AddIP(ConntrackNatDstIP, net.ParseIP("192.168.1.1"))
filterV6 = &ConntrackFilter{}
filterV6.AddIP(ConntrackNatDstIP, net.ParseIP("dddd:dddd:dddd:dddd:dddd:dddd:dddd:dddd"))
v4Match, v6Match = applyFilter(flowList, filterV4, filterV6)
if v4Match != 2 || v6Match != 0 {
t.Fatalf("Error, there should be an exact match, v4:%d, v6:%d", v4Match, v6Match)
}
// AnyIp for Nat
filterV4 = &ConntrackFilter{}
filterV4.AddIP(ConntrackNatAnyIP, net.ParseIP("192.168.1.1"))
filterV6 = &ConntrackFilter{}
filterV6.AddIP(ConntrackNatAnyIP, net.ParseIP("eeee:eeee:eeee:eeee:eeee:eeee:eeee:eeee"))
v4Match, v6Match = applyFilter(flowList, filterV4, filterV6)
if v4Match != 2 || v6Match != 1 {
t.Fatalf("Error, there should be an exact match, v4:%d, v6:%d", v4Match, v6Match)
}
}