Merge pull request #279 from safing/fix/resolver-unfailing

Improve resolver unfailing

resolver/config.go

@@ -174,7 +174,7 @@ The format is: "protocol://ip:port?parameter=value&parameter=value"
 	if err != nil {
 		return err
 	}
-	nameserverRetryRate = config.Concurrent.GetAsInt(CfgOptionNameserverRetryRateKey, 600)
+	nameserverRetryRate = config.Concurrent.GetAsInt(CfgOptionNameserverRetryRateKey, 300)
 
 	err = config.Register(&config.Option{
 		Name:           "Ignore System/Network Servers",

resolver/resolve.go

@@ -377,6 +377,10 @@ resolveLoop:
 			if rrCache.RCode != dns.RcodeSuccess && tryAll {
 				continue
 			}
+
+			// Report a successful connection.
+			resolver.Conn.ResetFailure()
+
 			break resolveLoop
 		}
 	}

resolver/resolver-env.go

@@ -124,3 +124,5 @@ func (er *envResolverConn) ReportFailure() {}
 func (er *envResolverConn) IsFailing() bool {
 	return false
 }
+
+func (er *envResolverConn) ResetFailure() {}

resolver/resolver-mdns.go

@@ -54,6 +54,8 @@ func (mrc *mDNSResolverConn) IsFailing() bool {
 	return false
 }
 
+func (mrc *mDNSResolverConn) ResetFailure() {}
+
 type savedQuestion struct {
 	question dns.Question
 	expires  time.Time

resolver/resolver-plain.go

@@ -24,11 +24,13 @@ type PlainResolver struct {
 
 // NewPlainResolver returns a new TPCResolver.
 func NewPlainResolver(resolver *Resolver) *PlainResolver {
-	return &PlainResolver{
+	newResolver := &PlainResolver{
 		BasicResolverConn: BasicResolverConn{
 			resolver: resolver,
 		},
 	}
+	newResolver.BasicResolverConn.init()
+	return newResolver
 }
 
 // Query executes the given query against the resolver.

resolver/resolver-tcp.go

@@ -62,7 +62,7 @@ func (ifq *InFlightQuery) MakeCacheRecord(reply *dns.Msg) *RRCache {
 // NewTCPResolver returns a new TPCResolver.
 func NewTCPResolver(resolver *Resolver) *TCPResolver {
 	var instanceID uint32
-	return &TCPResolver{
+	newResolver := &TCPResolver{
 		BasicResolverConn: BasicResolverConn{
 			resolver: resolver,
 		},
@@ -79,6 +79,8 @@ func NewTCPResolver(resolver *Resolver) *TCPResolver {
 		queries:         make(chan *dns.Msg, 1000),
 		inFlightQueries: make(map[uint16]*InFlightQuery),
 	}
+	newResolver.BasicResolverConn.init()
+	return newResolver
 }
 
 // UseTLS enabled TLS for the TCPResolver. TLS settings must be correctly configured in the Resolver.

resolver/resolver.go

@@ -8,6 +8,9 @@ import (
 	"time"
 
 	"github.com/miekg/dns"
+	"github.com/tevino/abool"
+
+	"github.com/safing/portbase/utils"
 	"github.com/safing/portmaster/netenv"
 	"github.com/safing/portmaster/network/netutils"
 )
@@ -179,16 +182,27 @@ type ResolverConn interface { //nolint:go-lint // TODO
 	Query(ctx context.Context, q *Query) (*RRCache, error)
 	ReportFailure()
 	IsFailing() bool
+	ResetFailure()
 }
 
 // BasicResolverConn implements ResolverConn for standard dns clients.
 type BasicResolverConn struct {
-	sync.Mutex // for lastFail
+	sync.Mutex // Also used by inheriting structs.
 
 	resolver *Resolver
 
+	failing      *abool.AtomicBool
 	failingUntil time.Time
 	fails        int
+	failLock     sync.Mutex
+
+	networkChangedFlag *utils.Flag
+}
+
+// init initializes the basic resolver connection.
+func (brc *BasicResolverConn) init() {
+	brc.failing = abool.New()
+	brc.networkChangedFlag = netenv.GetNetworkChangedFlag()
 }
 
 // ReportFailure reports that an error occurred with this resolver.
@@ -198,20 +212,47 @@ func (brc *BasicResolverConn) ReportFailure() {
 		return
 	}
 
-	brc.Lock()
-	defer brc.Unlock()
+	brc.failLock.Lock()
+	defer brc.failLock.Unlock()
 
 	brc.fails++
 	if brc.fails > FailThreshold {
+		brc.failing.Set()
 		brc.failingUntil = time.Now().Add(time.Duration(nameserverRetryRate()) * time.Second)
 		brc.fails = 0
+
+		// Refresh the network changed flag in order to only regard changes after
+		// the fail.
+		brc.networkChangedFlag.Refresh()
 	}
 }
 
 // IsFailing returns if this resolver is currently failing.
 func (brc *BasicResolverConn) IsFailing() bool {
-	brc.Lock()
-	defer brc.Unlock()
+	// Check if not failing.
+	if !brc.failing.IsSet() {
+		return false
+	}
 
+	brc.failLock.Lock()
+	defer brc.failLock.Unlock()
+
+	// Reset failure status if the network changed since the last query.
+	if brc.networkChangedFlag.IsSet() {
+		brc.networkChangedFlag.Refresh()
+		brc.ResetFailure()
+		return false
+	}
+
+	// Check if we are still
 	return time.Now().Before(brc.failingUntil)
 }
+
+// ResetFailure resets the failure status.
+func (brc *BasicResolverConn) ResetFailure() {
+	if brc.failing.SetToIf(true, false) {
+		brc.failLock.Lock()
+		defer brc.failLock.Unlock()
+		brc.fails = 0
+	}
+}