Update DNS resolver, reset resolvers if all failed

intel/resolve.go

@@ -9,7 +9,6 @@ import (
 	"net"
 	"strings"
 	"sync"
-	"sync/atomic"
 	"time"
 
 	"github.com/miekg/dns"
@@ -248,11 +247,12 @@ func intelligentResolve(ctx context.Context, fqdn string, qtype dns.Type, securi
 	}
 
 	log.Tracer(ctx).Warningf("intel: failed to resolve %s%s: all resolvers failed (or were skipped to fulfill the security level)", fqdn, qtype.String())
-	log.Criticalf("intel: failed to resolve %s%s: all resolvers failed (or were skipped to fulfill the security level)", fqdn, qtype.String())
+	log.Criticalf("intel: failed to resolve %s%s: all resolvers failed (or were skipped to fulfill the security level), resetting servers...", fqdn, qtype.String())
+	go resetResolverFailStatus()
+
 	return nil
 
 	// TODO: check if there would be resolvers available in lower security modes and alert user
-
 }
 
 func tryResolver(ctx context.Context, resolver *Resolver, lastFailBoundary int64, fqdn string, qtype dns.Type, securityLevel uint8) (*RRCache, bool) {
@@ -270,7 +270,7 @@ func tryResolver(ctx context.Context, resolver *Resolver, lastFailBoundary int64
 		return nil, false
 	}
 	// check if failed recently
-	if atomic.LoadInt64(resolver.LastFail) > lastFailBoundary {
+	if resolver.LastFail() > lastFailBoundary {
 		log.Tracer(ctx).Tracef("intel: skipping resolver %s, because it failed recently", resolver)
 		return nil, false
 	}
@@ -281,10 +281,10 @@ func tryResolver(ctx context.Context, resolver *Resolver, lastFailBoundary int64
 		return nil, false
 	}
 	// check if resolver is already initialized
-	if !resolver.Initialized.IsSet() {
+	if !resolver.Initialized() {
 		// first should init, others wait
 		resolver.InitLock.Lock()
-		if resolver.Initialized.IsSet() {
+		if resolver.Initialized() {
 			// unlock immediately if resolver was initialized
 			resolver.InitLock.Unlock()
 		} else {
@@ -292,7 +292,7 @@ func tryResolver(ctx context.Context, resolver *Resolver, lastFailBoundary int64
 			defer resolver.InitLock.Unlock()
 		}
 		// check if previous init failed
-		if atomic.LoadInt64(resolver.LastFail) > lastFailBoundary {
+		if resolver.LastFail() > lastFailBoundary {
 			return nil, false
 		}
 	}
@@ -300,21 +300,23 @@ func tryResolver(ctx context.Context, resolver *Resolver, lastFailBoundary int64
 	rrCache, err := query(ctx, resolver, fqdn, qtype)
 	if err != nil {
 		// check if failing is disabled
-		if atomic.LoadInt64(resolver.LastFail) == -1 {
+		if resolver.LastFail() == -1 {
 			log.Tracer(ctx).Tracef("intel: non-failing resolver %s failed, moving to next: %s", resolver, err)
 			// log.Tracef("intel: non-failing resolver %s failed (%s), moving to next", resolver, err)
 			return nil, false
 		}
 		log.Tracer(ctx).Warningf("intel: resolver %s failed, moving to next: %s", resolver, err)
 		log.Warningf("intel: resolver %s failed, moving to next: %s", resolver, err)
-		resolver.LockReason.Lock()
-		resolver.FailReason = err.Error()
-		resolver.LockReason.Unlock()
-		atomic.StoreInt64(resolver.LastFail, time.Now().Unix())
-		resolver.Initialized.UnSet()
+		resolver.Lock()
+		resolver.failReason = err.Error()
+		resolver.lastFail = time.Now().Unix()
+		resolver.initialized = false
+		resolver.Unlock()
 		return nil, false
 	}
-	resolver.Initialized.SetToIf(false, true)
+	resolver.Lock()
+	resolver.initialized = true
+	resolver.Unlock()
 
 	return rrCache, true
 }

intel/resolver.go

@@ -10,7 +10,6 @@ import (
 	"sync"
 
 	"github.com/miekg/dns"
-	"github.com/tevino/abool"
 
 	"github.com/Safing/portbase/log"
 
@@ -20,6 +19,8 @@ import (
 
 // Resolver holds information about an active resolver.
 type Resolver struct {
+	sync.Mutex
+
 	// static
 	Server        string
 	ServerType    string
@@ -34,20 +35,51 @@ type Resolver struct {
 	Search             *[]string
 	SkipFqdnBeforeInit string
 
-	// atomic
-	Initialized *abool.AtomicBool
 	InitLock sync.Mutex
-	LastFail    *int64
-	Expires     *int64
 
 	// must be locked
-	LockReason sync.Mutex
-	FailReason string
+	initialized bool
+	lastFail    int64
+	failReason  string
+	fails       int
+	expires     int64
 
-	// TODO: add:
-	// Expiration (for server got from DHCP / ICMPv6)
-	// bootstrapping (first query is already sent, wait for it to either succeed or fail - think about http bootstrapping here!)
-	// expanded server info: type, server address, server port, options - so we do not have to parse this every time!
+	// TODO: add Expiration (for server got from DHCP / ICMPv6)
+}
+
+// Initialized returns the internal initialized value while locking the Resolver.
+func (r *Resolver) Initialized() bool {
+	r.Lock()
+	defer r.Unlock()
+	return r.initialized
+}
+
+// LastFail returns the internal lastfail value while locking the Resolver.
+func (r *Resolver) LastFail() int64 {
+	r.Lock()
+	defer r.Unlock()
+	return r.lastFail
+}
+
+// FailReason returns the internal failreason value while locking the Resolver.
+func (r *Resolver) FailReason() string {
+	r.Lock()
+	defer r.Unlock()
+	return r.failReason
+}
+
+// Fails returns the internal fails value while locking the Resolver.
+func (r *Resolver) Fails() int {
+	r.Lock()
+	defer r.Unlock()
+	return r.fails
+}
+
+// Expires returns the internal expires value while locking the Resolver.
+func (r *Resolver) Expires() int64 {
+	r.Lock()
+	defer r.Unlock()
+	return r.expires
 }
 
 func (r *Resolver) String() string {
@@ -144,7 +176,6 @@ configuredServersLoop:
 				continue configuredServersLoop
 			}
 
-			var lastFail int64
 			new := &Resolver{
 				Server:        server,
 				ServerType:    strings.ToLower(parts[0]),
@@ -152,9 +183,7 @@ configuredServersLoop:
 				ServerIP:      ip,
 				ServerIPScope: netutils.ClassifyIP(ip),
 				ServerPort:    port,
-				LastFail:      &lastFail,
 				Source:        "config",
-				Initialized:   abool.NewBool(false),
 			}
 
 			switch new.ServerType {
@@ -197,7 +226,6 @@ assignedServersLoop:
 		key = indexOfResolver(server, globalResolvers)
 		if resetResolvers || key == -1 {
 
-			var lastFail int64
 			new := &Resolver{
 				Server:        server,
 				ServerType:    "dns",
@@ -205,9 +233,7 @@ assignedServersLoop:
 				ServerIP:      nameserver.IP,
 				ServerIPScope: netutils.ClassifyIP(nameserver.IP),
 				ServerPort:    53,
-				LastFail:      &lastFail,
 				Source:        "dhcp",
-				Initialized:   abool.NewBool(false),
 			}
 			new.clientManager = newDNSClientManager(new)
 
@@ -288,3 +314,18 @@ assignedServersLoop:
 	}
 
 }
+
+// resetResolverFailStatus resets all resolver failures.
+func resetResolverFailStatus() {
+	resolversLock.Lock()
+	defer resolversLock.Unlock()
+
+	log.Tracef("old: %+v %+v, ", globalResolvers, localResolvers)
+	for _, resolver := range append(globalResolvers, localResolvers...) {
+		resolver.Lock()
+		resolver.failReason = ""
+		resolver.lastFail = 0
+		resolver.Unlock()
+	}
+	log.Tracef("new: %+v %+v, ", globalResolvers, localResolvers)
+}