tendrils/snmp.go

package tendrils

import (
	"fmt"
	"log"
	"net"
	"regexp"
	"strings"
	"sync"
	"time"

	"github.com/gosnmp/gosnmp"
)

var portNameRewrites = []struct {
	re   *regexp.Regexp
	repl string
}{
	{regexp.MustCompile(`.*Slot: (\d+) Port: (\d+).*`), "$1/$2"},
}

func rewritePortName(name string) string {
	for _, rw := range portNameRewrites {
		if rw.re.MatchString(name) {
			return rw.re.ReplaceAllString(name, rw.repl)
		}
	}
	return name
}

type snmpConfig struct {
	username  string
	authKey   string
	privKey   string
	authProto gosnmp.SnmpV3AuthProtocol
	privProto gosnmp.SnmpV3PrivProtocol
	secLevel  gosnmp.SnmpV3MsgFlags
	timeout   time.Duration
	retries   int
}

func defaultSNMPConfig() *snmpConfig {
	return &snmpConfig{
		username:  "tendrils",
		authKey:   "tendrils",
		privKey:   "tendrils",
		authProto: gosnmp.SHA512,
		privProto: gosnmp.AES,
		secLevel:  gosnmp.AuthPriv,
		timeout:   5 * time.Second,
		retries:   1,
	}
}

func snmpToInt(val interface{}) (int, bool) {
	switch v := val.(type) {
	case int:
		return v, true
	case int32:
		return int(v), true
	case int64:
		return int(v), true
	case uint:
		return int(v), true
	case uint32:
		return int(v), true
	case uint64:
		return int(v), true
	default:
		return 0, false
	}
}

func (t *Tendrils) connectSNMP(ip net.IP) (*gosnmp.GoSNMP, error) {
	cfg := defaultSNMPConfig()

	snmp := &gosnmp.GoSNMP{
		Target:        ip.String(),
		Port:          161,
		Version:       gosnmp.Version3,
		Timeout:       cfg.timeout,
		Retries:       cfg.retries,
		SecurityModel: gosnmp.UserSecurityModel,
		MsgFlags:      cfg.secLevel,
		SecurityParameters: &gosnmp.UsmSecurityParameters{
			UserName:                 cfg.username,
			AuthenticationProtocol:   cfg.authProto,
			AuthenticationPassphrase: cfg.authKey,
			PrivacyProtocol:          cfg.privProto,
			PrivacyPassphrase:        cfg.privKey,
		},
	}

	err := snmp.Connect()
	if err != nil {
		return nil, err
	}

	return snmp, nil
}

func (t *Tendrils) pollSNMP(node *Node, ip net.IP) {
	var wg sync.WaitGroup
	wg.Add(2)

	go func() {
		defer wg.Done()
		t.queryWirelessAP(node, ip)
	}()

	go func() {
		defer wg.Done()
		t.querySNMPDevice(node, ip)
	}()

	wg.Wait()
}

func (t *Tendrils) querySNMPDevice(node *Node, ip net.IP) {
	snmp, err := t.connectSNMP(ip)
	if err != nil {
		if t.DebugSNMP {
			log.Printf("[snmp] %s: connect failed: %v", ip, err)
		}
		return
	}
	defer snmp.Conn.Close()

	ifNames := t.getInterfaceNames(snmp)

	t.querySysName(snmp, node)
	t.queryInterfaceMACs(snmp, node, ifNames)
	sysUpTime := t.getSysUpTime(snmp)
	t.queryInterfaceStats(snmp, node, ifNames, sysUpTime)
	t.queryPoEBudget(snmp, node)
	t.queryBridgeMIB(snmp, node, ifNames)
	t.queryDHCPBindings(snmp)
}

func (t *Tendrils) getSysUpTime(snmp *gosnmp.GoSNMP) uint64 {
	oid := "1.3.6.1.2.1.1.3.0"

	result, err := snmp.Get([]string{oid})
	if err != nil {
		return 0
	}

	if len(result.Variables) == 0 {
		return 0
	}

	v, ok := snmpToInt(result.Variables[0].Value)
	if !ok {
		log.Printf("[ERROR] failed to parse sysUpTime: type=%T value=%v", result.Variables[0].Value, result.Variables[0].Value)
		return 0
	}
	return uint64(v)
}

func (t *Tendrils) querySysName(snmp *gosnmp.GoSNMP, node *Node) {
	oid := "1.3.6.1.2.1.1.5.0"

	result, err := snmp.Get([]string{oid})
	if err != nil {
		return
	}

	if len(result.Variables) == 0 {
		return
	}

	variable := result.Variables[0]
	if variable.Type != gosnmp.OctetString {
		return
	}

	sysName := string(variable.Value.([]byte))
	if sysName == "" {
		return
	}

	t.nodes.Update(node, nil, nil, "", sysName, "snmp-sysname")
}

func (t *Tendrils) queryInterfaceMACs(snmp *gosnmp.GoSNMP, node *Node, ifNames map[int]string) {
	oid := "1.3.6.1.2.1.2.2.1.6"

	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return
	}

	for _, result := range results {
		if result.Type != gosnmp.OctetString {
			continue
		}

		macBytes := result.Value.([]byte)
		if len(macBytes) != 6 {
			continue
		}

		mac := net.HardwareAddr(macBytes)
		if isBroadcastOrZero(mac) {
			continue
		}

		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
		var ifIndex int
		if _, err := fmt.Sscanf(oidSuffix, "%d", &ifIndex); err != nil {
			continue
		}

		name := rewritePortName(ifNames[ifIndex])
		if t.DebugSNMP {
			log.Printf("[snmp] %s: interface %d mac=%s name=%s", snmp.Target, ifIndex, mac, name)
		}

		t.nodes.Update(node, mac, nil, name, "", "snmp-ifmac")
	}
}

func (t *Tendrils) queryInterfaceStats(snmp *gosnmp.GoSNMP, node *Node, ifNames map[int]string, sysUpTime uint64) {
	ifOperStatus := t.getInterfaceTable(snmp, "1.3.6.1.2.1.2.2.1.8")
	ifLastChange := t.getInterfaceTable(snmp, "1.3.6.1.2.1.2.2.1.9")
	ifHighSpeed := t.getInterfaceTable(snmp, "1.3.6.1.2.1.31.1.1.1.15")
	ifInErrors := t.getInterfaceTable(snmp, "1.3.6.1.2.1.2.2.1.14")
	ifOutErrors := t.getInterfaceTable(snmp, "1.3.6.1.2.1.2.2.1.20")

	ifHCInOctets := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.6")
	ifHCOutOctets := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.10")
	ifHCInUcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.7")
	ifHCInMcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.8")
	ifHCInBcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.9")
	ifHCOutUcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.11")
	ifHCOutMcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.12")
	ifHCOutBcastPkts := t.getInterfaceTable64(snmp, "1.3.6.1.2.1.31.1.1.1.13")

	poeStats := t.getPoEStats(snmp, ifNames)
	now := time.Now()

	t.nodes.mu.Lock()
	defer t.nodes.mu.Unlock()

	for ifIndex, name := range ifNames {
		name = rewritePortName(name)
		iface, exists := node.Interfaces[name]
		if !exists {
			continue
		}

		status, hasStatus := ifOperStatus[ifIndex]
		isUp := hasStatus && status == 1
		if !isUp {
			if iface.Up {
				log.Printf("[ERROR] port down on %s %s", node.DisplayName(), name)
				t.errors.AddPortDown(node, name)
			}
			iface.Up = false
			iface.Stats = nil
			continue
		}
		iface.Up = true

		stats := &InterfaceStats{}

		if speed, ok := ifHighSpeed[ifIndex]; ok {
			stats.Speed = uint64(speed) * 1000000
		}

		if lastChange, ok := ifLastChange[ifIndex]; ok && sysUpTime > 0 {
			if uint64(lastChange) <= sysUpTime {
				stats.Uptime = (sysUpTime - uint64(lastChange)) / 100
			}
		}

		if inErr, ok := ifInErrors[ifIndex]; ok {
			stats.InErrors = uint64(inErr)
		}

		if outErr, ok := ifOutErrors[ifIndex]; ok {
			stats.OutErrors = uint64(outErr)
		}

		inBytes, hasInBytes := ifHCInOctets[ifIndex]
		outBytes, hasOutBytes := ifHCOutOctets[ifIndex]

		inPkts := ifHCInUcastPkts[ifIndex] + ifHCInMcastPkts[ifIndex] + ifHCInBcastPkts[ifIndex]
		outPkts := ifHCOutUcastPkts[ifIndex] + ifHCOutMcastPkts[ifIndex] + ifHCOutBcastPkts[ifIndex]

		hasPrev := !iface.prevTimestamp.IsZero()
		if hasPrev {
			if iface.prevUptime > 0 && stats.Uptime > 0 && stats.Uptime < iface.prevUptime {
				log.Printf("[ERROR] port flap on %s %s: uptime dropped from %d to %d seconds", node.DisplayName(), name, iface.prevUptime, stats.Uptime)
				t.errors.AddPortFlap(node, name)
			}
			if hasInBytes && hasOutBytes {
				elapsed := now.Sub(iface.prevTimestamp).Seconds()
				if elapsed > 0 {
					stats.InPktsRate = float64(inPkts-iface.prevInPkts) / elapsed
					stats.OutPktsRate = float64(outPkts-iface.prevOutPkts) / elapsed
					stats.InBytesRate = float64(inBytes-iface.prevInBytes) / elapsed
					stats.OutBytesRate = float64(outBytes-iface.prevOutBytes) / elapsed

					maxBytesRate := float64(stats.Speed) / 8 * 2
					if stats.InBytesRate < 0 || stats.InBytesRate > maxBytesRate {
						stats.InPktsRate = 0
						stats.InBytesRate = 0
					}
					if stats.OutBytesRate < 0 || stats.OutBytesRate > maxBytesRate {
						stats.OutPktsRate = 0
						stats.OutBytesRate = 0
					}
				}
			}
		}

		iface.prevInPkts = inPkts
		iface.prevOutPkts = outPkts
		iface.prevInBytes = inBytes
		iface.prevOutBytes = outBytes
		if stats.Uptime > 0 {
			iface.prevUptime = stats.Uptime
		}
		iface.prevTimestamp = now

		if poe, ok := poeStats[name]; ok {
			stats.PoE = poe
		}

		node.SetInterfaceStats(name, stats)
	}
}

func (t *Tendrils) queryPoEBudget(snmp *gosnmp.GoSNMP, node *Node) {
	maxPowerOID := "1.3.6.1.2.1.105.1.3.1.1.2.1"
	powerOID := "1.3.6.1.2.1.105.1.3.1.1.4.1"

	result, err := snmp.Get([]string{maxPowerOID, powerOID})
	if err != nil {
		return
	}

	var power, maxPower float64
	for _, v := range result.Variables {
		val, ok := snmpToInt(v.Value)
		if !ok {
			continue
		}
		if v.Name == "."+powerOID {
			power = float64(val)
		} else if v.Name == "."+maxPowerOID {
			maxPower = float64(val)
		}
	}

	if maxPower > 0 {
		t.nodes.mu.Lock()
		node.PoEBudget = &PoEBudget{Power: power, MaxPower: maxPower}
		node.Type = NodeTypeSwitch
		t.nodes.mu.Unlock()
	}
}

func (t *Tendrils) getInterfaceTable(snmp *gosnmp.GoSNMP, oid string) map[int]int {
	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return nil
	}

	table := map[int]int{}
	for _, result := range results {
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
		var ifIndex int
		if _, err := fmt.Sscanf(oidSuffix, "%d", &ifIndex); err != nil {
			continue
		}

		value, ok := snmpToInt(result.Value)
		if !ok {
			continue
		}
		table[ifIndex] = value
	}
	return table
}

func (t *Tendrils) getInterfaceTable64(snmp *gosnmp.GoSNMP, oid string) map[int]uint64 {
	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return nil
	}

	table := map[int]uint64{}
	for _, result := range results {
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
		var ifIndex int
		if _, err := fmt.Sscanf(oidSuffix, "%d", &ifIndex); err != nil {
			continue
		}

		switch v := result.Value.(type) {
		case uint64:
			table[ifIndex] = v
		case uint:
			table[ifIndex] = uint64(v)
		case int:
			table[ifIndex] = uint64(v)
		case int64:
			table[ifIndex] = uint64(v)
		}
	}
	return table
}

func (t *Tendrils) getPoEStats(snmp *gosnmp.GoSNMP, ifNames map[int]string) map[string]*PoEStats {
	statusOID := "1.3.6.1.2.1.105.1.1.1.6"

	statusResults, err := snmp.BulkWalkAll(statusOID)
	if err != nil {
		return nil
	}

	powerTable := t.getPoETable(snmp, "1.3.6.1.4.1.4526.10.15.1.1.1.2")
	maxPowerTable := t.getPoETable(snmp, "1.3.6.1.4.1.4526.10.15.1.1.1.1")

	stats := map[string]*PoEStats{}
	for _, result := range statusResults {
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+statusOID), ".")
		parts := strings.Split(oidSuffix, ".")
		if len(parts) < 2 {
			continue
		}

		var portIndex int
		if _, err := fmt.Sscanf(parts[1], "%d", &portIndex); err != nil {
			continue
		}

		status, ok := snmpToInt(result.Value)
		if !ok {
			continue
		}

		ifName := rewritePortName(ifNames[portIndex])
		if ifName == "" {
			continue
		}

		if status != 3 {
			continue
		}

		poe := &PoEStats{}
		if power, ok := powerTable[portIndex]; ok {
			poe.Power = float64(power) / 1000.0
		}
		if maxPower, ok := maxPowerTable[portIndex]; ok {
			poe.MaxPower = float64(maxPower) / 1000.0
		}

		if t.DebugSNMP {
			log.Printf("[snmp] %s: poe port %d (%s) power=%v maxpower=%v", snmp.Target, portIndex, ifName, powerTable[portIndex], maxPowerTable[portIndex])
		}

		if poe.Power > 0 || poe.MaxPower > 0 {
			stats[ifName] = poe
		}
	}
	return stats
}

func (t *Tendrils) getPoETable(snmp *gosnmp.GoSNMP, oid string) map[int]int {
	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return nil
	}

	table := map[int]int{}
	for _, result := range results {
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
		parts := strings.Split(oidSuffix, ".")
		if len(parts) < 2 {
			continue
		}

		var portIndex int
		if _, err := fmt.Sscanf(parts[1], "%d", &portIndex); err != nil {
			continue
		}

		value, ok := snmpToInt(result.Value)
		if !ok {
			continue
		}
		table[portIndex] = value
	}
	return table
}

func (t *Tendrils) queryBridgeMIB(snmp *gosnmp.GoSNMP, node *Node, ifNames map[int]string) {
	portOID := "1.3.6.1.2.1.17.7.1.2.2.1.2"

	portResults, err := snmp.BulkWalkAll(portOID)
	if err != nil {
		return
	}

	type macPortEntry struct {
		mac        net.HardwareAddr
		bridgePort int
	}
	var macPorts []macPortEntry

	for _, result := range portResults {
		if result.Type == gosnmp.Integer {
			oidSuffix := strings.TrimPrefix(result.Name[len(portOID):], ".")
			parts := strings.Split(oidSuffix, ".")

			if len(parts) >= 8 {
				var macBytes []byte
				for j := 2; j <= 7; j++ {
					var b int
					fmt.Sscanf(parts[j], "%d", &b)
					macBytes = append(macBytes, byte(b))
				}

				if len(macBytes) == 6 {
					mac := net.HardwareAddr(macBytes)
					bridgePort := result.Value.(int)
					macPorts = append(macPorts, macPortEntry{mac: mac, bridgePort: bridgePort})
				}
			}
		}
	}

	bridgePortToIfIndex := t.getBridgePortMapping(snmp)

	t.nodes.ClearMACTable(node)

	for _, entry := range macPorts {
		mac := entry.mac

		if isBroadcastOrZero(mac) {
			continue
		}

		ifIndex, exists := bridgePortToIfIndex[entry.bridgePort]
		if !exists {
			ifIndex = entry.bridgePort
		}
		ifName := rewritePortName(ifNames[ifIndex])

		if t.DebugSNMP {
			log.Printf("[snmp] %s: mac=%s port=%s", snmp.Target, mac, ifName)
		}

		t.nodes.Update(nil, mac, nil, "", "", "snmp")
		t.nodes.UpdateMACTable(node, mac, ifName)
	}
}

func (t *Tendrils) getBridgePortMapping(snmp *gosnmp.GoSNMP) map[int]int {
	oid := "1.3.6.1.2.1.17.1.4.1.2"

	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return nil
	}

	mapping := make(map[int]int)
	for _, result := range results {
		if result.Type == gosnmp.Integer {
			oidParts := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
			var bridgePort int
			_, err := fmt.Sscanf(oidParts, "%d", &bridgePort)
			if err != nil {
				continue
			}
			ifIndex := result.Value.(int)
			mapping[bridgePort] = ifIndex
		}
	}

	return mapping
}

func (t *Tendrils) getInterfaceNames(snmp *gosnmp.GoSNMP) map[int]string {
	oid := "1.3.6.1.2.1.2.2.1.2"

	results, err := snmp.BulkWalkAll(oid)
	if err != nil {
		return nil
	}

	names := make(map[int]string)
	for _, result := range results {
		if result.Type == gosnmp.OctetString {
			oidParts := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+oid), ".")
			var ifIndex int
			_, err := fmt.Sscanf(oidParts, "%d", &ifIndex)
			if err != nil {
				continue
			}
			name := string(result.Value.([]byte))
			names[ifIndex] = name
		}
	}

	return names
}

func (t *Tendrils) queryDHCPBindings(snmp *gosnmp.GoSNMP) {
	baseOID := "1.3.6.1.4.1.4526.10.12.3.2.1"
	ipOID := baseOID + ".1"
	macOID := baseOID + ".3"

	ipResults, err := snmp.BulkWalkAll(ipOID)
	if err != nil {
		return
	}

	macResults, err := snmp.BulkWalkAll(macOID)
	if err != nil {
		return
	}

	ips := map[int]net.IP{}
	for _, result := range ipResults {
		if result.Type != gosnmp.IPAddress {
			continue
		}
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+ipOID), ".")
		var idx int
		if _, err := fmt.Sscanf(oidSuffix, "%d", &idx); err != nil {
			continue
		}
		ips[idx] = net.ParseIP(result.Value.(string))
	}

	macs := map[int]net.HardwareAddr{}
	for _, result := range macResults {
		if result.Type != gosnmp.OctetString {
			continue
		}
		macBytes := result.Value.([]byte)
		if len(macBytes) != 6 {
			continue
		}
		oidSuffix := strings.TrimPrefix(strings.TrimPrefix(result.Name, "."+macOID), ".")
		var idx int
		if _, err := fmt.Sscanf(oidSuffix, "%d", &idx); err != nil {
			continue
		}
		macs[idx] = net.HardwareAddr(macBytes)
	}

	for idx, ip := range ips {
		mac, ok := macs[idx]
		if !ok {
			continue
		}
		if isBroadcastOrZero(mac) {
			continue
		}
		if t.DebugSNMP {
			log.Printf("[snmp] %s: dhcp binding mac=%s ip=%s", snmp.Target, mac, ip)
		}
		t.nodes.Update(nil, mac, []net.IP{ip}, "", "", "dhcp")
	}
}

func (t *Tendrils) connectSNMPv2c(ip net.IP, community string) (*gosnmp.GoSNMP, error) {
	snmp := &gosnmp.GoSNMP{
		Target:    ip.String(),
		Port:      161,
		Version:   gosnmp.Version2c,
		Community: community,
		Timeout:   5 * time.Second,
		Retries:   1,
	}

	err := snmp.Connect()
	if err != nil {
		return nil, err
	}

	return snmp, nil
}

func (t *Tendrils) queryWirelessAP(node *Node, ip net.IP) bool {
	snmp, err := t.connectSNMPv2c(ip, "tendrils12!")
	if err != nil {
		return false
	}
	defer snmp.Conn.Close()

	sysObjID := t.getSysObjectID(snmp)
	if !strings.HasPrefix(sysObjID, "1.3.6.1.4.1.11863.") {
		return false
	}

	t.nodes.mu.Lock()
	node.Type = NodeTypeAP
	t.nodes.mu.Unlock()

	ifNames := t.getInterfaceNames(snmp)
	t.querySysName(snmp, node)
	t.queryInterfaceMACs(snmp, node, ifNames)
	t.queryTPLinkWirelessClients(snmp, node)

	return true
}

func (t *Tendrils) getSysObjectID(snmp *gosnmp.GoSNMP) string {
	oid := "1.3.6.1.2.1.1.2.0"

	result, err := snmp.Get([]string{oid})
	if err != nil {
		return ""
	}

	if len(result.Variables) == 0 {
		return ""
	}

	variable := result.Variables[0]
	if variable.Type != gosnmp.ObjectIdentifier {
		return ""
	}

	return strings.TrimPrefix(variable.Value.(string), ".")
}

func (t *Tendrils) queryTPLinkWirelessClients(snmp *gosnmp.GoSNMP, node *Node) {
	clientMACOID := "1.3.6.1.4.1.11863.10.1.1.2.1.2"

	results, err := snmp.BulkWalkAll(clientMACOID)
	if err != nil {
		if t.DebugSNMP {
			log.Printf("[snmp] %s: tp-link client walk failed: %v", snmp.Target, err)
		}
		return
	}

	t.nodes.ClearMACTable(node)

	for _, result := range results {
		if result.Type != gosnmp.OctetString {
			continue
		}

		macText := string(result.Value.([]byte))
		macText = strings.TrimRight(macText, "\x00")
		macText = strings.ReplaceAll(macText, "-", ":")
		mac, err := net.ParseMAC(macText)
		if err != nil {
			if t.DebugSNMP {
				log.Printf("[snmp] %s: invalid wireless client mac %q: %v", snmp.Target, macText, err)
			}
			continue
		}

		t.nodes.UpdateMACTable(node, mac, "wifi")

		if t.DebugSNMP {
			log.Printf("[snmp] %s: wireless client mac=%s", snmp.Target, mac.String())
		}
	}

	t.NotifyUpdate()
}