various: add resolver fallback for captive portalsandrew/dns-fallback

If enabled via a nodeAttr, we can fall back to querying the system
resolvers instead of any configured MagicDNS resolvers if a captive
portal is detected.

TODO: more docs

Signed-off-by: Andrew Dunham <andrew@du.nham.ca>
Change-Id: I1f0d29d3a066cd4673d794c72d5b69790fb81def

5 files changed, 79 insertions, 6 deletions

diff --git a/ipn/ipnlocal/local.go b/ipn/ipnlocal/local.go
index 9fea49870..e9de193c3 100644
--- a/ipn/ipnlocal/local.go
+++ b/ipn/ipnlocal/local.go
@@ -3627,7 +3627,14 @@ func (b *LocalBackend) authReconfig() {
 	disableSubnetsIfPAC := nm.HasCap(tailcfg.NodeAttrDisableSubnetsIfPAC)
 	userDialUseRoutes := nm.HasCap(tailcfg.NodeAttrUserDialUseRoutes)
 	dohURL, dohURLOK := exitNodeCanProxyDNS(nm, b.peers, prefs.ExitNodeID())
-	dcfg := dnsConfigForNetmap(nm, b.peers, prefs, b.logf, version.OS())
+	haveCaptivePortal := func() bool {
+		lastReport := b.MagicConn().GetLastNetcheckReport(b.ctx)
+		if lastReport == nil {
+			return false
+		}
+		return lastReport.CaptivePortal.EqualBool(true)
+	}
+	dcfg := dnsConfigForNetmap(nm, b.peers, prefs, b.logf, version.OS(), haveCaptivePortal)
 	// If the current node is an app connector, ensure the app connector machine is started
 	b.reconfigAppConnectorLocked(nm, prefs)
 	b.mu.Unlock()
@@ -3725,12 +3732,14 @@ func shouldUseOneCGNATRoute(logf logger.Logf, controlKnobs *controlknobs.Knobs,
 	return false
 }
 
+var dnsAllowFallback = envknob.RegisterBool("TS_DNS_ALLOW_FALLBACK")
+
 // dnsConfigForNetmap returns a *dns.Config for the given netmap,
 // prefs, client OS version, and cloud hosting environment.
 //
 // The versionOS is a Tailscale-style version ("iOS", "macOS") and not
 // a runtime.GOOS.
-func dnsConfigForNetmap(nm *netmap.NetworkMap, peers map[tailcfg.NodeID]tailcfg.NodeView, prefs ipn.PrefsView, logf logger.Logf, versionOS string) *dns.Config {
+func dnsConfigForNetmap(nm *netmap.NetworkMap, peers map[tailcfg.NodeID]tailcfg.NodeView, prefs ipn.PrefsView, logf logger.Logf, versionOS string, haveCaptivePortal func() bool) *dns.Config {
 	if nm == nil {
 		return nil
 	}
@@ -3739,6 +3748,17 @@ func dnsConfigForNetmap(nm *netmap.NetworkMap, peers map[tailcfg.NodeID]tailcfg.
 		Hosts:  map[dnsname.FQDN][]netip.Addr{},
 	}
 
+	// TODO(andrew-d): this isn't enough; we should probably healthcheck
+	// the resolver and decide based on that, instead of just using the
+	// netcheck 'captive portal' bool.
+	if dnsAllowFallback() {
+		logf("[v1] dnsConfigForNetmap: allowing system resolver fallback due to envknob")
+		dcfg.AllowFallback = true
+	} else if nm.HasCap(tailcfg.NodeAttrDNSFallbackToSystemResolver) && haveCaptivePortal() {
+		logf("[v1] dnsConfigForNetmap: allowing system resolver fallback due to nodeAttr and captive portal")
+		dcfg.AllowFallback = true
+	}
+
 	// selfV6Only is whether we only have IPv6 addresses ourselves.
 	selfV6Only := nm.GetAddresses().ContainsFunc(tsaddr.PrefixIs6) &&
 		!nm.GetAddresses().ContainsFunc(tsaddr.PrefixIs4)
diff --git a/net/dns/config.go b/net/dns/config.go
index 9c55f6d73..eac4eaa1d 100644
--- a/net/dns/config.go
+++ b/net/dns/config.go
@@ -44,6 +44,8 @@ type Config struct {
 	// OnlyIPv6, if true, uses the IPv6 service IP (for MagicDNS)
 	// instead of the IPv4 version (100.100.100.100).
 	OnlyIPv6 bool
+	// AllowFallback TODO
+	AllowFallback bool
 }
 
 func (c *Config) serviceIP() netip.Addr {
diff --git a/net/dns/manager.go b/net/dns/manager.go
index 6810d5a6b..9711918aa 100644
--- a/net/dns/manager.go
+++ b/net/dns/manager.go
@@ -94,6 +94,11 @@ func (m *Manager) Set(cfg Config) error {
 		ocfg.WriteToBufioWriter(w)
 	}))
 
+	if cfg.AllowFallback && len(rcfg.Routes["."]) > 0 {
+		// TODO: this isn't exactly right
+		rcfg.FallbackResolvers = slices.Clone(rcfg.Routes["."])
+	}
+
 	if err := m.resolver.SetConfig(rcfg); err != nil {
 		return err
 	}
diff --git a/net/dns/resolver/tsdns.go b/net/dns/resolver/tsdns.go
index c44565ee1..b71c40b36 100644
--- a/net/dns/resolver/tsdns.go
+++ b/net/dns/resolver/tsdns.go
@@ -64,6 +64,7 @@ type packet struct {
 // If the query is an exact match for an entry in LocalHosts, return that.
 // Else if the query suffix matches an entry in LocalDomains, return NXDOMAIN.
 // Else forward the query to the most specific matching entry in Routes.
+// Else if any FallbackResolvers are set, forward the query to them.
 // Else return SERVFAIL.
 type Config struct {
 	// Routes is a map of DNS name suffix to the resolvers to use for
@@ -76,6 +77,9 @@ type Config struct {
 	// LocalDomains is a list of DNS name suffixes that should not be
 	// routed to upstream resolvers.
 	LocalDomains []dnsname.FQDN
+	// FallbackResolvers is a list of resolvers to use if the resolver(s)
+	// in Routes do not respond. It can be empty.
+	FallbackResolvers []*dnstype.Resolver
 }
 
 // WriteToBufioWriter write a debug version of c for logs to w, omitting
@@ -191,10 +195,11 @@ type Resolver struct {
 	closed chan struct{}
 
 	// mu guards the following fields from being updated while used.
-	mu           sync.Mutex
-	localDomains []dnsname.FQDN
-	hostToIP     map[dnsname.FQDN][]netip.Addr
-	ipToHost     map[netip.Addr]dnsname.FQDN
+	mu                sync.Mutex
+	localDomains      []dnsname.FQDN
+	hostToIP          map[dnsname.FQDN][]netip.Addr
+	ipToHost          map[netip.Addr]dnsname.FQDN
+	fallbackResolvers []*dnstype.Resolver
 }
 
 type ForwardLinkSelector interface {
@@ -247,6 +252,7 @@ func (r *Resolver) SetConfig(cfg Config) error {
 	r.localDomains = cfg.LocalDomains
 	r.hostToIP = cfg.Hosts
 	r.ipToHost = reverse
+	r.fallbackResolvers = cfg.FallbackResolvers
 	return nil
 }
 
@@ -286,6 +292,13 @@ func (r *Resolver) Query(ctx context.Context, bs []byte, family string, from net
 		defer cancel()
 		err = r.forwarder.forwardWithDestChan(ctx, packet{bs, family, from}, responses)
 		if err != nil {
+			// If we have any fallbacks, try them.
+			// TODO(andrew-d): don't use a new context?
+			if resp, err := r.resolveWithFallbacks(context.Background(), bs, family, from); err == nil {
+				r.logf("resolved with fallback resolver")
+				return resp, nil
+			}
+
 			select {
 			// Best effort: use any error response sent by forwardWithDestChan.
 			// This is present in some errors paths, such as when all upstream
@@ -302,6 +315,35 @@ func (r *Resolver) Query(ctx context.Context, bs []byte, family string, from net
 	return out, err
 }
 
+func (r *Resolver) resolveWithFallbacks(ctx context.Context, bs []byte, family string, from netip.AddrPort) ([]byte, error) {
+	r.mu.Lock()
+	fallbacks := r.fallbackResolvers
+	r.mu.Unlock()
+
+	r.logf("resolveWithFallbacks: have %d fallback resolvers", len(fallbacks))
+	if len(fallbacks) == 0 {
+		return nil, errNotOurName // any error is fine here
+	}
+
+	var resolvers []resolverAndDelay
+	for _, resolver := range fallbacks {
+		resolvers = append(resolvers, resolverAndDelay{
+			name: resolver,
+		})
+	}
+
+	responses := make(chan packet, 1)
+	ctx, cancel := context.WithTimeout(ctx, 5*time.Second) // TODO?
+	defer close(responses)
+	defer cancel()
+
+	err := r.forwarder.forwardWithDestChan(ctx, packet{bs, family, from}, responses, resolvers...)
+	if err == nil {
+		return (<-responses).bs, nil
+	}
+	return nil, errNotOurName // any error is fine here
+}
+
 // parseExitNodeQuery parses a DNS request packet.
 // It returns nil if it's malformed or lacking a question.
 func parseExitNodeQuery(q []byte) *response {
diff --git a/tailcfg/tailcfg.go b/tailcfg/tailcfg.go
index c842b88d3..4e0992307 100644
--- a/tailcfg/tailcfg.go
+++ b/tailcfg/tailcfg.go
@@ -2274,6 +2274,10 @@ const (
 	// depending on the destination address and the configured routes. When present, it also makes
 	// the DNS forwarder use UserDial instead of SystemDial when dialing resolvers.
 	NodeAttrUserDialUseRoutes NodeCapability = "user-dial-routes"
+
+	// NodeAttrDNSFallbackToSystemResolver makes the DNS forwarder fall
+	// back to the system resolver if all configured DNS servers fail.
+	NodeAttrDNSFallbackToSystemResolver NodeCapability = "dns-fallback-to-system-resolver"
 )
 
 // SetDNSRequest is a request to add a DNS record.