nebula/firewall.go

package nebula

import (
	"encoding/binary"
	"encoding/json"
	"fmt"
	"net"
	"sync"
	"time"

	"crypto/sha256"
	"encoding/hex"
	"errors"
	"reflect"
	"strconv"
	"strings"

	"github.com/rcrowley/go-metrics"
	"github.com/slackhq/nebula/cert"
)

const (
	fwProtoAny  = 0 // When we want to handle HOPOPT (0) we can change this, if ever
	fwProtoTCP  = 6
	fwProtoUDP  = 17
	fwProtoICMP = 1

	fwPortAny      = 0  // Special value for matching `port: any`
	fwPortFragment = -1 // Special value for matching `port: fragment`
)

const tcpACK = 0x10
const tcpFIN = 0x01

type FirewallInterface interface {
	AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, ip *net.IPNet, caName string, caSha string) error
}

type conn struct {
	Expires time.Time // Time when this conntrack entry will expire
	Seq     uint32    // If tcp rtt tracking is enabled this will be the seq we are looking for an ack
	Sent    time.Time // If tcp rtt tracking is enabled this will be when Seq was last set
}

// TODO: need conntrack max tracked connections handling
type Firewall struct {
	Conns map[FirewallPacket]*conn

	InRules  *FirewallTable
	OutRules *FirewallTable

	//TODO: we should have many more options for TCP, an option for ICMP, and mimic the kernel a bit better
	// https://www.kernel.org/doc/Documentation/networking/nf_conntrack-sysctl.txt
	TCPTimeout     time.Duration //linux: 5 days max
	UDPTimeout     time.Duration //linux: 180s max
	DefaultTimeout time.Duration //linux: 600s

	TimerWheel *TimerWheel

	// Used to ensure we don't emit local packets for ips we don't own
	localIps *CIDRTree

	connMutex sync.Mutex
	rules     string

	trackTCPRTT  bool
	metricTCPRTT metrics.Histogram
}

type FirewallTable struct {
	TCP      firewallPort
	UDP      firewallPort
	ICMP     firewallPort
	AnyProto firewallPort
}

func newFirewallTable() *FirewallTable {
	return &FirewallTable{
		TCP:      firewallPort{},
		UDP:      firewallPort{},
		ICMP:     firewallPort{},
		AnyProto: firewallPort{},
	}
}

type FirewallRule struct {
	// Any makes Hosts, Groups, and CIDR irrelevant. CAName and CASha still need to be checked
	Any     bool
	Hosts   map[string]struct{}
	Groups  [][]string
	CIDR    *CIDRTree
	CANames map[string]struct{}
	CAShas  map[string]struct{}
}

// Even though ports are uint16, int32 maps are faster for lookup
// Plus we can use `-1` for fragment rules
type firewallPort map[int32]*FirewallRule

type FirewallPacket struct {
	LocalIP    uint32
	RemoteIP   uint32
	LocalPort  uint16
	RemotePort uint16
	Protocol   uint8
	Fragment   bool
}

func (fp *FirewallPacket) Copy() *FirewallPacket {
	return &FirewallPacket{
		LocalIP:    fp.LocalIP,
		RemoteIP:   fp.RemoteIP,
		LocalPort:  fp.LocalPort,
		RemotePort: fp.RemotePort,
		Protocol:   fp.Protocol,
		Fragment:   fp.Fragment,
	}
}

func (fp FirewallPacket) MarshalJSON() ([]byte, error) {
	var proto string
	switch fp.Protocol {
	case fwProtoTCP:
		proto = "tcp"
	case fwProtoICMP:
		proto = "icmp"
	case fwProtoUDP:
		proto = "udp"
	default:
		proto = fmt.Sprintf("unknown %v", fp.Protocol)
	}
	return json.Marshal(m{
		"LocalIP":    int2ip(fp.LocalIP).String(),
		"RemoteIP":   int2ip(fp.RemoteIP).String(),
		"LocalPort":  fp.LocalPort,
		"RemotePort": fp.RemotePort,
		"Protocol":   proto,
		"Fragment":   fp.Fragment,
	})
}

// NewFirewall creates a new Firewall object. A TimerWheel is created for you from the provided timeouts.
func NewFirewall(tcpTimeout, UDPTimeout, defaultTimeout time.Duration, c *cert.NebulaCertificate) *Firewall {
	//TODO: error on 0 duration
	var min, max time.Duration

	if tcpTimeout < UDPTimeout {
		min = tcpTimeout
		max = UDPTimeout
	} else {
		min = UDPTimeout
		max = tcpTimeout
	}

	if defaultTimeout < min {
		min = defaultTimeout
	} else if defaultTimeout > max {
		max = defaultTimeout
	}

	localIps := NewCIDRTree()
	for _, ip := range c.Details.Ips {
		localIps.AddCIDR(&net.IPNet{IP: ip.IP, Mask: net.IPMask{255, 255, 255, 255}}, struct{}{})
	}

	for _, n := range c.Details.Subnets {
		localIps.AddCIDR(n, struct{}{})
	}

	return &Firewall{
		Conns:          make(map[FirewallPacket]*conn),
		InRules:        newFirewallTable(),
		OutRules:       newFirewallTable(),
		TimerWheel:     NewTimerWheel(min, max),
		TCPTimeout:     tcpTimeout,
		UDPTimeout:     UDPTimeout,
		DefaultTimeout: defaultTimeout,
		localIps:       localIps,
		metricTCPRTT:   metrics.GetOrRegisterHistogram("network.tcp.rtt", nil, metrics.NewExpDecaySample(1028, 0.015)),
	}
}

func NewFirewallFromConfig(nc *cert.NebulaCertificate, c *Config) (*Firewall, error) {
	fw := NewFirewall(
		c.GetDuration("firewall.conntrack.tcp_timeout", time.Duration(time.Minute*12)),
		c.GetDuration("firewall.conntrack.udp_timeout", time.Duration(time.Minute*3)),
		c.GetDuration("firewall.conntrack.default_timeout", time.Duration(time.Minute*10)),
		nc,
		//TODO: max_connections
	)

	err := AddFirewallRulesFromConfig(false, c, fw)
	if err != nil {
		return nil, err
	}

	err = AddFirewallRulesFromConfig(true, c, fw)
	if err != nil {
		return nil, err
	}

	return fw, nil
}

// AddRule properly creates the in memory rule structure for a firewall table.
func (f *Firewall) AddRule(incoming bool, proto uint8, startPort int32, endPort int32, groups []string, host string, ip *net.IPNet, caName string, caSha string) error {

	// We need this rule string because we generate a hash. Removing this will break firewall reload.
	ruleString := fmt.Sprintf(
		"incoming: %v, proto: %v, startPort: %v, endPort: %v, groups: %v, host: %v, ip: %v, caName: %v, caSha: %s",
		incoming, proto, startPort, endPort, groups, host, ip, caName, caSha,
	)
	f.rules += ruleString + "\n"

	direction := "incoming"
	if !incoming {
		direction = "outgoing"
	}
	l.WithField("firewallRule", m{"direction": direction, "proto": proto, "startPort": startPort, "endPort": endPort, "groups": groups, "host": host, "ip": ip, "caName": caName, "caSha": caSha}).
		Info("Firewall rule added")

	var (
		ft *FirewallTable
		fp firewallPort
	)

	if incoming {
		ft = f.InRules
	} else {
		ft = f.OutRules
	}

	switch proto {
	case fwProtoTCP:
		fp = ft.TCP
	case fwProtoUDP:
		fp = ft.UDP
	case fwProtoICMP:
		fp = ft.ICMP
	case fwProtoAny:
		fp = ft.AnyProto
	default:
		return fmt.Errorf("unknown protocol %v", proto)
	}

	return fp.addRule(startPort, endPort, groups, host, ip, caName, caSha)
}

// GetRuleHash returns a hash representation of all inbound and outbound rules
func (f *Firewall) GetRuleHash() string {
	sum := sha256.Sum256([]byte(f.rules))
	return hex.EncodeToString(sum[:])
}

func AddFirewallRulesFromConfig(inbound bool, config *Config, fw FirewallInterface) error {
	var table string
	if inbound {
		table = "firewall.inbound"
	} else {
		table = "firewall.outbound"
	}

	r := config.Get(table)
	if r == nil {
		return nil
	}

	rs, ok := r.([]interface{})
	if !ok {
		return fmt.Errorf("%s failed to parse, should be an array of rules", table)
	}

	for i, t := range rs {
		var groups []string
		r, err := convertRule(t)
		if err != nil {
			return fmt.Errorf("%s rule #%v; %s", table, i, err)
		}

		if r.Code != "" && r.Port != "" {
			return fmt.Errorf("%s rule #%v; only one of port or code should be provided", table, i)
		}

		if r.Host == "" && len(r.Groups) == 0 && r.Group == "" && r.Cidr == "" && r.CAName == "" && r.CASha == "" {
			return fmt.Errorf("%s rule #%v; at least one of host, group, cidr, ca_name, or ca_sha must be provided", table, i)
		}

		if len(r.Groups) > 0 {
			groups = r.Groups
		}

		if r.Group != "" {
			// Check if we have both groups and group provided in the rule config
			if len(groups) > 0 {
				return fmt.Errorf("%s rule #%v; only one of group or groups should be defined, both provided", table, i)
			}

			groups = []string{r.Group}
		}

		var sPort, errPort string
		if r.Code != "" {
			errPort = "code"
			sPort = r.Code
		} else {
			errPort = "port"
			sPort = r.Port
		}

		startPort, endPort, err := parsePort(sPort)
		if err != nil {
			return fmt.Errorf("%s rule #%v; %s %s", table, i, errPort, err)
		}

		var proto uint8
		switch r.Proto {
		case "any":
			proto = fwProtoAny
		case "tcp":
			proto = fwProtoTCP
		case "udp":
			proto = fwProtoUDP
		case "icmp":
			proto = fwProtoICMP
		default:
			return fmt.Errorf("%s rule #%v; proto was not understood; `%s`", table, i, r.Proto)
		}

		var cidr *net.IPNet
		if r.Cidr != "" {
			_, cidr, err = net.ParseCIDR(r.Cidr)
			if err != nil {
				return fmt.Errorf("%s rule #%v; cidr did not parse; %s", table, i, err)
			}
		}

		err = fw.AddRule(inbound, proto, startPort, endPort, groups, r.Host, cidr, r.CAName, r.CASha)
		if err != nil {
			return fmt.Errorf("%s rule #%v; `%s`", table, i, err)
		}
	}

	return nil
}

func (f *Firewall) Drop(packet []byte, fp FirewallPacket, incoming bool, c *cert.NebulaCertificate, caPool *cert.NebulaCAPool) bool {
	// Check if we spoke to this tuple, if we did then allow this packet
	if f.inConns(packet, fp, incoming) {
		return false
	}

	// Make sure we are supposed to be handling this local ip address
	if f.localIps.Contains(fp.LocalIP) == nil {
		return true
	}

	table := f.OutRules
	if incoming {
		table = f.InRules
	}

	// We now know which firewall table to check against
	if !table.match(fp, incoming, c, caPool) {
		return true
	}

	// We always want to conntrack since it is a faster operation
	f.addConn(packet, fp, incoming)

	return false
}

// Destroy cleans up any known cyclical references so the object can be free'd my GC. This should be called if a new
// firewall object is created
func (f *Firewall) Destroy() {
	//TODO: clean references if/when needed
}

func (f *Firewall) EmitStats() {
	conntrackCount := len(f.Conns)
	metrics.GetOrRegisterGauge("firewall.conntrack.count", nil).Update(int64(conntrackCount))
}

func (f *Firewall) inConns(packet []byte, fp FirewallPacket, incoming bool) bool {
	f.connMutex.Lock()

	// Purge every time we test
	ep, has := f.TimerWheel.Purge()
	if has {
		f.evict(ep)
	}

	c, ok := f.Conns[fp]

	if !ok {
		f.connMutex.Unlock()
		return false
	}

	switch fp.Protocol {
	case fwProtoTCP:
		c.Expires = time.Now().Add(f.TCPTimeout)
		if incoming {
			f.checkTCPRTT(c, packet)
		} else {
			setTCPRTTTracking(c, packet)
		}
	case fwProtoUDP:
		c.Expires = time.Now().Add(f.UDPTimeout)
	default:
		c.Expires = time.Now().Add(f.DefaultTimeout)
	}

	f.connMutex.Unlock()

	return true
}

func (f *Firewall) addConn(packet []byte, fp FirewallPacket, incoming bool) {
	var timeout time.Duration
	c := &conn{}

	switch fp.Protocol {
	case fwProtoTCP:
		timeout = f.TCPTimeout
		if !incoming {
			setTCPRTTTracking(c, packet)
		}
	case fwProtoUDP:
		timeout = f.UDPTimeout
	default:
		timeout = f.DefaultTimeout
	}

	f.connMutex.Lock()
	if _, ok := f.Conns[fp]; !ok {
		f.TimerWheel.Add(fp, timeout)
	}

	c.Expires = time.Now().Add(timeout)
	f.Conns[fp] = c
	f.connMutex.Unlock()
}

// Evict checks if a conntrack entry has expired, if so it is removed, if not it is re-added to the wheel
// Caller must own the connMutex lock!
func (f *Firewall) evict(p FirewallPacket) {
	//TODO: report a stat if the tcp rtt tracking was never resolved?
	// Are we still tracking this conn?
	t, ok := f.Conns[p]
	if !ok {
		return
	}

	newT := t.Expires.Sub(time.Now())

	// Timeout is in the future, re-add the timer
	if newT > 0 {
		f.TimerWheel.Add(p, newT)
		return
	}

	// This conn is done
	delete(f.Conns, p)
}

func (ft *FirewallTable) match(p FirewallPacket, incoming bool, c *cert.NebulaCertificate, caPool *cert.NebulaCAPool) bool {
	if ft.AnyProto.match(p, incoming, c, caPool) {
		return true
	}

	switch p.Protocol {
	case fwProtoTCP:
		if ft.TCP.match(p, incoming, c, caPool) {
			return true
		}
	case fwProtoUDP:
		if ft.UDP.match(p, incoming, c, caPool) {
			return true
		}
	case fwProtoICMP:
		if ft.ICMP.match(p, incoming, c, caPool) {
			return true
		}
	}

	return false
}

func (fp firewallPort) addRule(startPort int32, endPort int32, groups []string, host string, ip *net.IPNet, caName string, caSha string) error {
	if startPort > endPort {
		return fmt.Errorf("start port was lower than end port")
	}

	for i := startPort; i <= endPort; i++ {
		if _, ok := fp[i]; !ok {
			fp[i] = &FirewallRule{
				Groups:  make([][]string, 0),
				Hosts:   make(map[string]struct{}),
				CIDR:    NewCIDRTree(),
				CANames: make(map[string]struct{}),
				CAShas:  make(map[string]struct{}),
			}
		}

		if err := fp[i].addRule(groups, host, ip, caName, caSha); err != nil {
			return err
		}
	}

	return nil
}

func (fp firewallPort) match(p FirewallPacket, incoming bool, c *cert.NebulaCertificate, caPool *cert.NebulaCAPool) bool {
	// We don't have any allowed ports, bail
	if fp == nil {
		return false
	}

	var port int32

	if p.Fragment {
		port = fwPortFragment
	} else if incoming {
		port = int32(p.LocalPort)
	} else {
		port = int32(p.RemotePort)
	}

	if fp[port].match(p, c, caPool) {
		return true
	}

	return fp[fwPortAny].match(p, c, caPool)
}

func (fr *FirewallRule) addRule(groups []string, host string, ip *net.IPNet, caName string, caSha string) error {
	if caName != "" {
		fr.CANames[caName] = struct{}{}
	}

	if caSha != "" {
		fr.CAShas[caSha] = struct{}{}
	}

	if fr.Any {
		return nil
	}

	if fr.isAny(groups, host, ip) {
		fr.Any = true
		// If it's any we need to wipe out any pre-existing rules to save on memory
		fr.Groups = make([][]string, 0)
		fr.Hosts = make(map[string]struct{})
		fr.CIDR = NewCIDRTree()
	} else {
		if len(groups) > 0 {
			fr.Groups = append(fr.Groups, groups)
		}

		if host != "" {
			fr.Hosts[host] = struct{}{}
		}

		if ip != nil {
			fr.CIDR.AddCIDR(ip, struct{}{})
		}
	}

	return nil
}

func (fr *FirewallRule) isAny(groups []string, host string, ip *net.IPNet) bool {
	for _, group := range groups {
		if group == "any" {
			return true
		}
	}

	if host == "any" {
		return true
	}

	if ip != nil && ip.Contains(net.IPv4(0, 0, 0, 0)) {
		return true
	}

	return false
}

func (fr *FirewallRule) match(p FirewallPacket, c *cert.NebulaCertificate, caPool *cert.NebulaCAPool) bool {
	if fr == nil {
		return false
	}

	// CASha and CAName always need to be checked
	if len(fr.CAShas) > 0 {
		if _, ok := fr.CAShas[c.Details.Issuer]; !ok {
			return false
		}
	}

	if len(fr.CANames) > 0 {
		s, err := caPool.GetCAForCert(c)
		if err != nil {
			return false
		}
		if _, ok := fr.CANames[s.Details.Name]; !ok {
			return false
		}
	}

	// Shortcut path for if groups, hosts, or cidr contained an `any`
	if fr.Any {
		return true
	}

	// Need any of group, host, or cidr to match
	for _, sg := range fr.Groups {
		found := false

		for _, g := range sg {
			if _, ok := c.Details.InvertedGroups[g]; !ok {
				found = false
				break
			}

			found = true
		}

		if found {
			return true
		}
	}

	if fr.Hosts != nil {
		if _, ok := fr.Hosts[c.Details.Name]; ok {
			return true
		}
	}

	if fr.CIDR != nil && fr.CIDR.Contains(p.RemoteIP) != nil {
		return true
	}

	// No host, group, or cidr matched, bye bye
	return false
}

type rule struct {
	Port   string
	Code   string
	Proto  string
	Host   string
	Group  string
	Groups []string
	Cidr   string
	CAName string
	CASha  string
}

func convertRule(p interface{}) (rule, error) {
	r := rule{}

	m, ok := p.(map[interface{}]interface{})
	if !ok {
		return r, errors.New("could not parse rule")
	}

	toString := func(k string, m map[interface{}]interface{}) string {
		v, ok := m[k]
		if !ok {
			return ""
		}
		return fmt.Sprintf("%v", v)
	}

	r.Port = toString("port", m)
	r.Code = toString("code", m)
	r.Proto = toString("proto", m)
	r.Host = toString("host", m)
	r.Group = toString("group", m)
	r.Cidr = toString("cidr", m)
	r.CAName = toString("ca_name", m)
	r.CASha = toString("ca_sha", m)

	if rg, ok := m["groups"]; ok {
		switch reflect.TypeOf(rg).Kind() {
		case reflect.Slice:
			v := reflect.ValueOf(rg)
			r.Groups = make([]string, v.Len())
			for i := 0; i < v.Len(); i++ {
				r.Groups[i] = v.Index(i).Interface().(string)
			}
		case reflect.String:
			r.Groups = []string{rg.(string)}
		default:
			r.Groups = []string{fmt.Sprintf("%v", rg)}
		}
	}

	return r, nil
}

func parsePort(s string) (startPort, endPort int32, err error) {
	if s == "any" {
		startPort = fwPortAny
		endPort = fwPortAny

	} else if s == "fragment" {
		startPort = fwPortFragment
		endPort = fwPortFragment

	} else if strings.Contains(s, `-`) {
		sPorts := strings.SplitN(s, `-`, 2)
		sPorts[0] = strings.Trim(sPorts[0], " ")
		sPorts[1] = strings.Trim(sPorts[1], " ")

		if len(sPorts) != 2 || sPorts[0] == "" || sPorts[1] == "" {
			return 0, 0, fmt.Errorf("appears to be a range but could not be parsed; `%s`", s)
		}

		rStartPort, err := strconv.Atoi(sPorts[0])
		if err != nil {
			return 0, 0, fmt.Errorf("beginning range was not a number; `%s`", sPorts[0])
		}

		rEndPort, err := strconv.Atoi(sPorts[1])
		if err != nil {
			return 0, 0, fmt.Errorf("ending range was not a number; `%s`", sPorts[1])
		}

		startPort = int32(rStartPort)
		endPort = int32(rEndPort)

		if startPort == fwPortAny {
			endPort = fwPortAny
		}

	} else {
		rPort, err := strconv.Atoi(s)
		if err != nil {
			return 0, 0, fmt.Errorf("was not a number; `%s`", s)
		}
		startPort = int32(rPort)
		endPort = startPort
	}

	return
}

//TODO: write tests for these
func setTCPRTTTracking(c *conn, p []byte) {
	if c.Seq != 0 {
		return
	}

	ihl := int(p[0]&0x0f) << 2

	// Don't track FIN packets
	if uint8(p[ihl+13])&tcpFIN != 0 {
		return
	}

	c.Seq = binary.BigEndian.Uint32(p[ihl+4 : ihl+8])
	c.Sent = time.Now()
}

func (f *Firewall) checkTCPRTT(c *conn, p []byte) bool {
	if c.Seq == 0 {
		return false
	}

	ihl := int(p[0]&0x0f) << 2
	if uint8(p[ihl+13])&tcpACK == 0 {
		return false
	}

	// Deal with wrap around, signed int cuts the ack window in half
	// 0 is a bad ack, no data acknowledged
	// positive number is a bad ack, ack is over half the window away
	if int32(c.Seq-binary.BigEndian.Uint32(p[ihl+8:ihl+12])) >= 0 {
		return false
	}

	f.metricTCPRTT.Update(time.Since(c.Sent).Nanoseconds())
	c.Seq = 0
	return true
}