Compare commits

..

No commits in common. "b49d67c87d3e92fcd3c3e7787e2faf761606cd2f" and "03f781f7e106185c3a8a9bbadce60b87e8282720" have entirely different histories.

4 changed files with 53 additions and 72 deletions

View File

@ -30,17 +30,15 @@ type LLDP struct {
advertiseOptions *AdvertiseOptions advertiseOptions *AdvertiseOptions
onChange func(neighbors []Neighbor) onChange func(neighbors []Neighbor)
neighbors *ttlcache.Cache[neighborCacheKey, Neighbor] neighbors *ttlcache.Cache[string, Neighbor]
// State tracking // State tracking
rxRunning bool
txRunning bool txRunning bool
txCtx context.Context txCtx context.Context
txCancel context.CancelFunc txCancel context.CancelFunc
rxCtx context.Context
rxRunning bool rxCancel context.CancelFunc
rxWaitGroup *sync.WaitGroup
rxCtx context.Context
rxCancel context.CancelFunc
} }
type AdvertiseOptions struct { type AdvertiseOptions struct {
@ -74,9 +72,8 @@ func NewLLDP(opts *Options) *LLDP {
advertiseOptions: opts.AdvertiseOptions, advertiseOptions: opts.AdvertiseOptions,
enableRx: opts.EnableRx, enableRx: opts.EnableRx,
enableTx: opts.EnableTx, enableTx: opts.EnableTx,
rxWaitGroup: &sync.WaitGroup{},
l: opts.Logger, l: opts.Logger,
neighbors: ttlcache.New(ttlcache.WithTTL[neighborCacheKey, Neighbor](1 * time.Hour)), neighbors: ttlcache.New(ttlcache.WithTTL[string, Neighbor](1 * time.Hour)),
onChange: opts.OnChange, onChange: opts.OnChange,
} }
} }

View File

@ -1,6 +1,7 @@
package lldp package lldp
import ( import (
"fmt"
"time" "time"
) )
@ -26,28 +27,22 @@ type Neighbor struct {
Values map[string]string `json:"values"` Values map[string]string `json:"values"`
} }
type neighborCacheKey struct { func (n *Neighbor) cacheKey() string {
mac string return fmt.Sprintf("%s-%s", n.Mac, n.Source)
source string
}
func (n *Neighbor) cacheKey() neighborCacheKey {
return neighborCacheKey{mac: n.Mac, source: n.Source}
} }
func (l *LLDP) addNeighbor(neighbor *Neighbor, ttl time.Duration) { func (l *LLDP) addNeighbor(neighbor *Neighbor, ttl time.Duration) {
logger := l.l.With(). logger := l.l.With().
Str("source", neighbor.Source).
Str("mac", neighbor.Mac). Str("mac", neighbor.Mac).
Interface("neighbor", neighbor). Interface("neighbor", neighbor).
Logger() Logger()
key := neighbor.cacheKey() key := neighbor.cacheKey()
currentNeighbor := l.neighbors.Get(key) current_neigh := l.neighbors.Get(key)
if currentNeighbor != nil { if current_neigh != nil {
currentSource := currentNeighbor.Value().Source current_source := current_neigh.Value().Source
if currentSource == "lldp" && neighbor.Source != "lldp" { if current_source == "lldp" && neighbor.Source != "lldp" {
logger.Info().Msg("skip updating neighbor, as LLDP has higher priority") logger.Info().Msg("skip updating neighbor, as LLDP has higher priority")
return return
} }
@ -61,7 +56,6 @@ func (l *LLDP) addNeighbor(neighbor *Neighbor, ttl time.Duration) {
func (l *LLDP) deleteNeighbor(neighbor *Neighbor) { func (l *LLDP) deleteNeighbor(neighbor *Neighbor) {
logger := l.l.With(). logger := l.l.With().
Str("source", neighbor.Source).
Str("mac", neighbor.Mac). Str("mac", neighbor.Mac).
Logger() Logger()

View File

@ -87,47 +87,51 @@ func (l *LLDP) setUpCapture() error {
return nil return nil
} }
func (l *LLDP) doCapture(logger *zerolog.Logger) { func (l *LLDP) doCapture(logger *zerolog.Logger, rxCtx context.Context) {
l.rxWaitGroup.Add(1) defer func() {
defer l.rxWaitGroup.Done() l.mu.Lock()
l.rxRunning = false
l.mu.Unlock()
}()
// TODO: use a channel to handle the packets // TODO: use a channel to handle the packets
// PacketSource.Packets() is not reliable and can cause panics and the upstream hasn't fixed it yet // PacketSource.Packets() is not reliable and can cause panics and the upstream hasn't fixed it yet
for { for rxCtx.Err() == nil {
// check if the context is done before blocking call if l.pktSourceRx == nil || l.tPacketRx == nil {
select { logger.Error().Msg("packet source or TPacketRx not initialized")
case <-l.rxCtx.Done(): break
logger.Info().Msg("RX context cancelled")
return
default:
} }
logger.Trace().Msg("waiting for next packet")
packet, err := l.pktSourceRx.NextPacket() packet, err := l.pktSourceRx.NextPacket()
logger.Trace().Interface("packet", packet).Err(err).Msg("got next packet") if err == nil {
if handleErr := l.handlePacket(packet, logger); handleErr != nil {
if err != nil { logger.Error().
logger.Error(). Err(handleErr).
Err(err). Msg("error handling packet")
Msg("error getting next packet")
// Immediately break for known unrecoverable errors
if err == io.EOF || err == io.ErrUnexpectedEOF ||
err == io.ErrNoProgress || err == io.ErrClosedPipe || err == io.ErrShortBuffer ||
err == syscall.EBADF ||
strings.Contains(err.Error(), "use of closed file") {
return
} }
continue continue
} }
if err := l.handlePacket(packet, logger); err != nil { // Immediately retry for temporary network errors and EAGAIN
logger.Error(). // temporary has been deprecated and most cases are timeouts
Err(err). if nerr, ok := err.(net.Error); ok && nerr.Timeout() {
Msg("error handling packet")
continue continue
} }
if err == syscall.EAGAIN {
continue
}
// Immediately break for known unrecoverable errors
if err == io.EOF || err == io.ErrUnexpectedEOF ||
err == io.ErrNoProgress || err == io.ErrClosedPipe || err == io.ErrShortBuffer ||
err == syscall.EBADF ||
strings.Contains(err.Error(), "use of closed file") {
break
}
logger.Error().
Err(err).
Msg("error receiving LLDP packet")
} }
} }
@ -154,7 +158,9 @@ func (l *LLDP) startCapture() error {
l.rxCtx, l.rxCancel = context.WithCancel(context.Background()) l.rxCtx, l.rxCancel = context.WithCancel(context.Background())
l.rxRunning = true l.rxRunning = true
go l.doCapture(&logger) // Capture context in closure
rxCtx := l.rxCtx
go l.doCapture(&logger, rxCtx)
return nil return nil
} }
@ -363,24 +369,8 @@ func (l *LLDP) stopCapture() error {
logger.Info().Msg("cancelled RX context, waiting for goroutine to finish") logger.Info().Msg("cancelled RX context, waiting for goroutine to finish")
} }
// stop the TPacketRx // Wait a bit for goroutine to finish
go func() { time.Sleep(500 * time.Millisecond)
if l.tPacketRx == nil {
return
}
// write an empty packet to the TPacketRx to interrupt the blocking read
// it's a shitty workaround until https://github.com/google/gopacket/pull/777 is merged,
// or we have a better solution, see https://github.com/google/gopacket/issues/1064
l.tPacketRx.WritePacketData([]byte{})
}()
// wait for the goroutine to finish
start := time.Now()
l.rxWaitGroup.Wait()
logger.Info().Dur("duration", time.Since(start)).Msg("RX goroutine finished")
l.rxRunning = false
if l.tPacketRx != nil { if l.tPacketRx != nil {
logger.Info().Msg("closing TPacketRx") logger.Info().Msg("closing TPacketRx")

View File

@ -60,10 +60,10 @@ var (
func toLLDPCapabilitiesBytes(capabilities []string) uint16 { func toLLDPCapabilitiesBytes(capabilities []string) uint16 {
r := uint16(0) r := uint16(0)
for _, capability := range capabilities { for _, capability := range capabilities {
mask, ok := capabilityMap[capability] if _, ok := capabilityMap[capability]; !ok {
if ok { continue
r |= mask
} }
r |= capabilityMap[capability]
} }
return r return r
} }