Merge remote-tracking branch 'origin/main' into revert/component-types

Signed-off-by: Dmitri Dolguikh <dmitri.external@netbird.io>

client/internal/debug/debug.go

@@ -51,6 +51,7 @@ nftables.txt: Anonymized nftables rules with packet counters across all families
 sysctls.txt: Forwarding, reverse-path filter, source-validation, and conntrack accounting sysctl values that the NetBird client may read or modify, if --system-info flag was provided (Linux only).
 resolv.conf: DNS resolver configuration from /etc/resolv.conf (Unix systems only), if --system-info flag was provided.
 scutil_dns.txt: DNS configuration from scutil --dns (macOS only), if --system-info flag was provided.
+dns_windows.txt: Anonymized NRPT rules and policy table in effect, DNS client policy, and per-interface and per-adapter DNS configuration (Windows only), if --system-info flag was provided.
 resolved_domains.txt: Anonymized resolved domain IP addresses from the status recorder.
 config.txt: Anonymized configuration information of the NetBird client.
 network_map.json: Anonymized sync response containing peer configurations, routes, DNS settings, and firewall rules.
@@ -237,6 +238,13 @@ scutil_dns.txt (macOS only):
 - Shows DNS configuration for all network interfaces
 - Includes search domains, nameservers, and DNS resolver settings
 - All IP addresses and domain names are anonymized
+
+dns_windows.txt (Windows only):
+- Lists the NRPT rules of both policy stores, the local one and the group policy one, marking the rules the client created
+- Follows them with the policy table the resolver has loaded, which differs from the rules while a change has not been picked up yet
+- Includes the DNS client group policy, the global TCP/IP and Dnscache parameters, and the DNS values of every interface that has any
+- Ends with the resolver configuration in effect per adapter, from GetAdaptersAddresses
+- All IP addresses and domain names are anonymized
 `
 
 const (

client/internal/debug/debug_nonunix.go

@@ -1,4 +1,4 @@
-//go:build !unix
+//go:build !unix && !windows
 
 package debug
 

client/internal/debug/debug_windows.go

@@ -0,0 +1,443 @@
+//go:build windows
+
+package debug
+
+import (
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"net/netip"
+	"strings"
+	"unsafe"
+
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
+	"golang.org/x/sys/windows/registry"
+
+	nbdns "github.com/netbirdio/netbird/client/internal/dns"
+)
+
+const dnsInfoFileName = "dns_windows.txt"
+
+const (
+	gpoDNSClientRoot = `SOFTWARE\Policies\Microsoft\Windows NT\DNSClient`
+	tcpipParamsPath  = `SYSTEM\CurrentControlSet\Services\Tcpip\Parameters`
+	dnscacheParams   = `SYSTEM\CurrentControlSet\Services\Dnscache\Parameters`
+)
+
+// interfaceDNSValues are the per-interface values that decide how a name is
+// resolved and registered. Everything the DNS host manager writes is in here,
+// so a bundle shows both what we set and what it replaced.
+var interfaceDNSValues = []string{
+	"NameServer",
+	"DhcpNameServer",
+	"Domain",
+	"DhcpDomain",
+	"SearchList",
+	"RegistrationEnabled",
+	"DisableDynamicUpdate",
+	"MaxNumberOfAddressesToRegister",
+	"EnableDHCP",
+}
+
+// addDNSInfo collects and adds DNS configuration information to the archive
+func (g *BundleGenerator) addDNSInfo() error {
+	if err := g.addFileToZip(strings.NewReader(g.collectDNSInfo()), dnsInfoFileName); err != nil {
+		return fmt.Errorf("add DNS info to zip: %w", err)
+	}
+
+	return nil
+}
+
+// collectDNSInfo renders the report. Everything below it reaches the platform
+// through COM and through lazily resolved procedures, which panic when a
+// procedure is missing rather than returning an error, and a debug bundle is not
+// allowed to take the daemon down. The panic is contained here, and whatever was
+// collected before it is kept and reported with it.
+func (g *BundleGenerator) collectDNSInfo() (content string) {
+	var sb strings.Builder
+
+	defer func() {
+		if r := recover(); r != nil {
+			log.Errorf("collecting Windows DNS configuration panicked: %v", r)
+			fmt.Fprintf(&sb, "\nerror: collection stopped: %v\n", r)
+		}
+		content = sb.String()
+	}()
+
+	sb.WriteString("Windows DNS configuration\n")
+	sb.WriteString("=========================\n")
+
+	adapters, adaptersErr := adapterAddresses()
+
+	g.writeNRPTRules(&sb, "NRPT rules, local policy store", nbdns.DNSPolicyConfigRoot)
+	g.writeNRPTRules(&sb, "NRPT rules, group policy store", nbdns.GPODNSPolicyConfigRoot)
+	g.writeEffectiveNRPTPolicies(&sb)
+	g.writeRegistryKey(&sb, "DNS client group policy", gpoDNSClientRoot)
+	g.writeRegistryKey(&sb, "Global TCP/IP parameters", tcpipParamsPath)
+	g.writeRegistryKey(&sb, "Dnscache parameters", dnscacheParams)
+	g.writeInterfaceDNS(&sb, "Per-interface DNS, IPv4", nbdns.InterfaceConfigPath, adapterNames(adapters))
+	g.writeInterfaceDNS(&sb, "Per-interface DNS, IPv6", nbdns.InterfaceConfigPathV6, adapterNames(adapters))
+	g.writeAdapterDNS(&sb, adapters, adaptersErr)
+
+	return sb.String()
+}
+
+// writeNRPTRules lists every rule in a policy store, ours and any other
+// product's, since a foreign rule for the same namespace decides resolution
+// just as ours does. Rules the client wrote are marked.
+func (g *BundleGenerator) writeNRPTRules(sb *strings.Builder, title, root string) {
+	writeSection(sb, title, root)
+
+	names, err := subKeyNames(root)
+	if err != nil {
+		fmt.Fprintf(sb, "error: %v\n", err)
+		return
+	}
+
+	if len(names) == 0 {
+		sb.WriteString("no rules\n")
+		return
+	}
+
+	for _, name := range names {
+		owner := ""
+		if strings.HasPrefix(strings.ToLower(name), strings.ToLower(nbdns.NRPTKeyPrefix)) {
+			owner = "  (netbird)"
+		}
+		fmt.Fprintf(sb, "%s%s\n", name, owner)
+		g.writeValues(sb, root+`\`+name, nil, "  ")
+	}
+}
+
+// writeEffectiveNRPTPolicies reports the table the resolver answers from, which
+// the registry cannot show: a rule is written before it is loaded, and it keeps
+// being enforced after its key is gone until the resolver reloads its policy.
+func (g *BundleGenerator) writeEffectiveNRPTPolicies(sb *strings.Builder) {
+	writeSection(sb, "NRPT policy table in effect", nrptPolicyClass+"."+nrptPolicyMethod+" in "+nrptPolicyNamespace)
+
+	entries, err := effectiveNRPTPolicies()
+	if err != nil {
+		fmt.Fprintf(sb, "error: %v\n", err)
+		return
+	}
+
+	if len(entries) == 0 {
+		sb.WriteString("no policies\n")
+		return
+	}
+
+	for _, entry := range entries {
+		fmt.Fprintf(sb, "%s\n", g.anonymizeValue("Namespace", entry.namespace))
+		for _, value := range entry.values {
+			fmt.Fprintf(sb, "  %s: %s\n", value.name, g.anonymizeValue(value.name, value.value))
+		}
+	}
+}
+
+// writeInterfaceDNS reports the DNS values of every interface that has any, so
+// the netbird interface can be compared against the physical ones. The registry
+// keys the values by GUID, so each is named from the adapter list; a GUID with
+// no adapter is a leftover key of an interface that no longer exists.
+func (g *BundleGenerator) writeInterfaceDNS(sb *strings.Builder, title, root string, names map[string]string) {
+	writeSection(sb, title, root)
+
+	guids, err := subKeyNames(root)
+	if err != nil {
+		fmt.Fprintf(sb, "error: %v\n", err)
+		return
+	}
+
+	var reported int
+	for _, guid := range guids {
+		var iface strings.Builder
+		g.writeValues(&iface, root+`\`+guid, interfaceDNSValues, "  ")
+		if iface.Len() == 0 {
+			continue
+		}
+
+		name, ok := names[strings.ToLower(guid)]
+		if !ok {
+			name = "no adapter with this GUID"
+		}
+
+		reported++
+		fmt.Fprintf(sb, "%s (%s)\n%s", guid, name, iface.String())
+	}
+
+	if reported == 0 {
+		sb.WriteString("no interface holds DNS values\n")
+	}
+}
+
+// writeRegistryKey reports the values of a single key, without its subkeys.
+func (g *BundleGenerator) writeRegistryKey(sb *strings.Builder, title, path string) {
+	writeSection(sb, title, path)
+
+	var values strings.Builder
+	g.writeValues(&values, path, nil, "")
+	if values.Len() == 0 {
+		sb.WriteString("no values\n")
+		return
+	}
+
+	sb.WriteString(values.String())
+}
+
+// writeValues renders the values of a key. A nil names list reports every
+// value, otherwise only those named and present.
+func (g *BundleGenerator) writeValues(sb *strings.Builder, path string, names []string, indent string) {
+	k, err := registry.OpenKey(registry.LOCAL_MACHINE, path, registry.QUERY_VALUE)
+	switch {
+	case errors.Is(err, registry.ErrNotExist), errors.Is(err, windows.ERROR_PATH_NOT_FOUND):
+		// an absent key is the normal state for the GPO store and for
+		// interfaces without DNS settings
+		log.Debugf("HKEY_LOCAL_MACHINE\\%s does not exist", path)
+		return
+	case err != nil:
+		fmt.Fprintf(sb, "%serror: open HKEY_LOCAL_MACHINE\\%s: %v\n", indent, path, err)
+		return
+	}
+	defer closeKey(k)
+
+	if names == nil {
+		names, err = k.ReadValueNames(-1)
+		if err != nil {
+			fmt.Fprintf(sb, "%serror: read value names: %v\n", indent, err)
+			return
+		}
+	}
+
+	for _, name := range names {
+		value, err := readRegistryValue(k, name)
+		switch {
+		case errors.Is(err, registry.ErrNotExist):
+			// the caller asks for a fixed set of values, most of which a
+			// given interface does not carry
+			continue
+		case err != nil:
+			// report rather than omit: a value that is there but cannot be
+			// read reads as unset otherwise
+			fmt.Fprintf(sb, "%s%s: error: %v\n", indent, name, err)
+			continue
+		}
+
+		fmt.Fprintf(sb, "%s%s: %s\n", indent, name, g.anonymizeValue(name, value))
+	}
+}
+
+// anonymizeValue redacts a registry value according to what its name says it
+// holds. Domains and addresses are handled per entry rather than by the string
+// pass: the pass only replaces domains something else in the bundle already
+// seeded, and its address regex would eat the digit labels of a reverse zone.
+func (g *BundleGenerator) anonymizeValue(name, value string) string {
+	if !g.anonymize || value == "" {
+		return value
+	}
+
+	switch {
+	case holdsDomains(name):
+		return joinValueEntries(splitValueEntries(value), g.anonymizeDomain)
+	case holdsAddresses(name):
+		return joinValueEntries(splitValueEntries(value), g.anonymizer.AnonymizeIPString)
+	default:
+		return g.anonymizer.AnonymizeString(value)
+	}
+}
+
+// holdsDomains reports whether a value name holds domains: the domain list of
+// an NRPT rule (Name) or of the policy table (Namespace), a search list, the
+// DNS suffix values of the TCP/IP and policy keys, which all end in "Domain"
+// (Domain, DhcpDomain, NV Domain, ICSDomain), and a proxy host name.
+func holdsDomains(name string) bool {
+	lower := strings.ToLower(name)
+	return lower == "name" || lower == "namespace" || lower == "searchlist" ||
+		strings.HasSuffix(lower, "domain") || strings.HasSuffix(lower, "proxyname")
+}
+
+// holdsAddresses reports whether a value name holds DNS server addresses
+// (NameServer, DhcpNameServer, GenericDNSServers, NameServers).
+func holdsAddresses(name string) bool {
+	lower := strings.ToLower(name)
+	return strings.Contains(lower, "nameserver") || strings.Contains(lower, "dnsserver")
+}
+
+// adapterNames maps adapter GUIDs, as the registry keys the interfaces, to the
+// names an operator sees.
+func adapterNames(adapters []*windows.IpAdapterAddresses) map[string]string {
+	names := make(map[string]string, len(adapters))
+	for _, adapter := range adapters {
+		guid := windows.BytePtrToString(adapter.AdapterName)
+		names[strings.ToLower(guid)] = windows.UTF16PtrToString(adapter.FriendlyName)
+	}
+	return names
+}
+
+// writeAdapterDNS reports the resolver configuration in effect per adapter,
+// which is what the resolver uses for a name no NRPT rule matches.
+func (g *BundleGenerator) writeAdapterDNS(sb *strings.Builder, adapters []*windows.IpAdapterAddresses, err error) {
+	writeSection(sb, "Adapter DNS configuration", "GetAdaptersAddresses")
+
+	if err != nil {
+		fmt.Fprintf(sb, "error: %v\n", err)
+		return
+	}
+
+	for _, adapter := range adapters {
+		name := windows.UTF16PtrToString(adapter.FriendlyName)
+		suffix := g.anonymizeDomain(windows.UTF16PtrToString(adapter.DnsSuffix))
+
+		fmt.Fprintf(sb, "%s (index %d, oper status %d)\n", name, adapter.IfIndex, adapter.OperStatus)
+		fmt.Fprintf(sb, "  DNS suffix: %s\n", suffix)
+
+		var servers []string
+		for server := adapter.FirstDnsServerAddress; server != nil; server = server.Next {
+			addr, ok := netip.AddrFromSlice(server.Address.IP())
+			if !ok {
+				continue
+			}
+
+			addr = addr.Unmap()
+			if g.anonymize {
+				addr = g.anonymizer.AnonymizeIP(addr)
+			}
+			servers = append(servers, addr.String())
+		}
+
+		fmt.Fprintf(sb, "  DNS servers: %s\n", strings.Join(servers, ", "))
+	}
+}
+
+// anonymizeDomain anonymizes a single domain, keeping the leading dot an NRPT
+// match domain carries.
+func (g *BundleGenerator) anonymizeDomain(entry string) string {
+	if !g.anonymize {
+		return entry
+	}
+
+	domain, dot := strings.CutPrefix(entry, ".")
+	if domain == "" {
+		return entry
+	}
+
+	anonymized := g.anonymizer.AnonymizeDomain(domain)
+	if dot {
+		anonymized = "." + anonymized
+	}
+	return anonymized
+}
+
+// splitValueEntries splits a registry value that holds a list. The separator
+// differs per value: a REG_MULTI_SZ arrives joined with ", ", a SearchList is
+// comma separated and a NameServer may use commas or spaces.
+func splitValueEntries(value string) []string {
+	return strings.FieldsFunc(value, func(r rune) bool {
+		return r == ',' || r == ';' || r == ' ' || r == '\t'
+	})
+}
+
+func joinValueEntries(entries []string, anonymize func(string) string) string {
+	for i, entry := range entries {
+		entries[i] = anonymize(entry)
+	}
+	return strings.Join(entries, ", ")
+}
+
+func writeSection(sb *strings.Builder, title, source string) {
+	fmt.Fprintf(sb, "\n%s\n%s\n%s\n", title, strings.Repeat("-", len(title)), source)
+}
+
+func subKeyNames(root string) ([]string, error) {
+	k, err := registry.OpenKey(registry.LOCAL_MACHINE, root, registry.ENUMERATE_SUB_KEYS)
+	if err != nil {
+		return nil, fmt.Errorf("open HKEY_LOCAL_MACHINE\\%s: %w", root, err)
+	}
+	defer closeKey(k)
+
+	names, err := k.ReadSubKeyNames(-1)
+	if err != nil {
+		return nil, fmt.Errorf("read subkey names: %w", err)
+	}
+
+	return names, nil
+}
+
+// readRegistryValue renders a value as text regardless of its type, so an
+// unexpected type in a policy key still shows up instead of being dropped.
+func readRegistryValue(k registry.Key, name string) (string, error) {
+	_, valueType, err := k.GetValue(name, nil)
+	if err != nil {
+		return "", fmt.Errorf("get value %s: %w", name, err)
+	}
+
+	switch valueType {
+	case registry.SZ, registry.EXPAND_SZ:
+		value, _, err := k.GetStringValue(name)
+		if err != nil {
+			return "", fmt.Errorf("get string value %s: %w", name, err)
+		}
+		return value, nil
+	case registry.MULTI_SZ:
+		values, _, err := k.GetStringsValue(name)
+		if err != nil {
+			return "", fmt.Errorf("get strings value %s: %w", name, err)
+		}
+		return strings.Join(values, ", "), nil
+	case registry.DWORD, registry.QWORD:
+		value, _, err := k.GetIntegerValue(name)
+		if err != nil {
+			return "", fmt.Errorf("get integer value %s: %w", name, err)
+		}
+		return fmt.Sprintf("%d (0x%x)", value, value), nil
+	case registry.BINARY:
+		value, _, err := k.GetBinaryValue(name)
+		if err != nil {
+			return "", fmt.Errorf("get binary value %s: %w", name, err)
+		}
+		return hex.EncodeToString(value), nil
+	default:
+		return fmt.Sprintf("<unhandled registry type %d>", valueType), nil
+	}
+}
+
+// adapterAddresses returns the adapter list including DNS servers. The call
+// reports the size it needs, so grow the buffer and retry until it fits.
+func adapterAddresses() (adapters []*windows.IpAdapterAddresses, err error) {
+	// GetAdaptersAddresses is resolved on first use and panics when it is
+	// missing, so this reports it as an error and leaves the rest of the
+	// report intact.
+	defer func() {
+		if r := recover(); r != nil {
+			adapters, err = nil, fmt.Errorf("GetAdaptersAddresses: %v", r)
+		}
+	}()
+
+	const flags = windows.GAA_FLAG_SKIP_ANYCAST | windows.GAA_FLAG_SKIP_MULTICAST
+
+	size := uint32(15000)
+	for range 3 {
+		buf := make([]byte, size)
+		first := (*windows.IpAdapterAddresses)(unsafe.Pointer(&buf[0]))
+
+		err := windows.GetAdaptersAddresses(windows.AF_UNSPEC, flags, 0, first, &size)
+		if errors.Is(err, windows.ERROR_BUFFER_OVERFLOW) {
+			continue
+		}
+		if err != nil {
+			return nil, fmt.Errorf("GetAdaptersAddresses: %w", err)
+		}
+
+		for adapter := first; adapter != nil; adapter = adapter.Next {
+			adapters = append(adapters, adapter)
+		}
+		return adapters, nil
+	}
+
+	return nil, fmt.Errorf("GetAdaptersAddresses: buffer kept growing")
+}
+
+func closeKey(k registry.Key) {
+	if err := k.Close(); err != nil {
+		log.Debugf("close registry key: %v", err)
+	}
+}

client/internal/debug/debug_windows_test.go

@@ -0,0 +1,146 @@
+//go:build windows
+
+package debug
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/netbirdio/netbird/client/anonymize"
+)
+
+func newDNSValueGenerator(level anonymize.Level) *BundleGenerator {
+	anonymizer := anonymize.NewAnonymizer(anonymize.DefaultAddresses())
+	anonymizer.SetLevel(level)
+
+	return &BundleGenerator{
+		anonymize:      true,
+		anonymizeLevel: level,
+		anonymizer:     anonymizer,
+	}
+}
+
+// TestAnonymizeValueByName covers the value kinds of the DNS registry keys. The
+// names decide the treatment, because the string pass alone replaces only
+// domains another part of the bundle already seeded.
+func TestAnonymizeValueByName(t *testing.T) {
+	tests := []struct {
+		name      string
+		valueName string
+		value     string
+		assert    func(t *testing.T, got string)
+	}{
+		{
+			name:      "NRPT match domains keep the leading dot",
+			valueName: "Name",
+			value:     ".internal.example.com, .corp.example.org",
+			assert: func(t *testing.T, got string) {
+				t.Helper()
+				for _, entry := range strings.Split(got, ", ") {
+					assert.True(t, strings.HasPrefix(entry, "."), "entry %q should keep its leading dot", entry)
+					assert.NotContains(t, entry, "example", "entry %q should not keep the original domain", entry)
+				}
+			},
+		},
+		{
+			name:      "any value name ending in Domain is treated as a domain",
+			valueName: "ICSDomain",
+			value:     "mshome.net",
+			assert: func(t *testing.T, got string) {
+				t.Helper()
+				assert.NotContains(t, got, "mshome", "should anonymize a domain suffix value")
+			},
+		},
+		{
+			name:      "search list is a comma separated domain list",
+			valueName: "SearchList",
+			value:     "corp.example.com,branch.example.com",
+			assert: func(t *testing.T, got string) {
+				t.Helper()
+				assert.NotContains(t, got, "example", "should anonymize every search domain")
+				assert.Len(t, strings.Split(got, ", "), 2, "should keep both search domains")
+			},
+		},
+		{
+			name:      "name servers are anonymized as addresses",
+			valueName: "DhcpNameServer",
+			value:     "203.0.113.10 8.8.8.8",
+			assert: func(t *testing.T, got string) {
+				t.Helper()
+				assert.NotContains(t, got, "203.0.113.10", "should anonymize a public resolver address")
+				// well-known resolvers stay readable at every level
+				assert.Contains(t, got, "8.8.8.8", "should keep a well-known resolver address")
+			},
+		},
+		{
+			name:      "opaque values are left to the string pass",
+			valueName: "DataBasePath",
+			value:     `%SystemRoot%\System32\drivers\etc`,
+			assert: func(t *testing.T, got string) {
+				t.Helper()
+				assert.Equal(t, `%SystemRoot%\System32\drivers\etc`, got, "should not alter a path")
+			},
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			g := newDNSValueGenerator(anonymize.LevelDefault)
+			tc.assert(t, g.anonymizeValue(tc.valueName, tc.value))
+		})
+	}
+}
+
+// TestParseNRPTPolicyTable parses the MOF text of the policy table out
+// parameters, as the provider on a client with one NRPT rule renders it.
+func TestParseNRPTPolicyTable(t *testing.T) {
+	const text = `[abstract]
+class __PARAMETERS
+{
+	[Out, EmbeddedInstance("DnsClientPolicyConfiguration"): ToSubClass, ID(2): DisableOverride ToInstance] DnsClientPolicyConfiguration cmdletOutput[] = {
+instance of DnsClientPolicyConfiguration
+{
+	DirectAccessProxyType = "NoProxy";
+	DirectAccessQueryIPsecRequired = FALSE;
+	NameEncoding = "Utf8WithoutMapping";
+	Namespace = ".0.100.in-addr.arpa";
+}, 
+instance of DnsClientPolicyConfiguration
+{
+	DirectAccessProxyType = "NoProxy";
+	NameEncoding = "Utf8WithoutMapping";
+	NameServers = {"100.0.255.254", "100.0.255.253"};
+	Namespace = ".nb.internal";
+}};
+	[in] boolean Effective;
+	[out] uint32 ReturnValue = 0;
+};
+`
+
+	entries := parseNRPTPolicyTable(text)
+	require.Len(t, entries, 2, "should parse both embedded instances")
+
+	assert.Equal(t, ".0.100.in-addr.arpa", entries[0].namespace, "should read the namespace of the first instance")
+	assert.Equal(t, ".nb.internal", entries[1].namespace, "should read the namespace of the second instance")
+
+	assert.Equal(t, []registryValue{
+		{name: "DirectAccessProxyType", value: "NoProxy"},
+		{name: "DirectAccessQueryIPsecRequired", value: "FALSE"},
+		{name: "NameEncoding", value: "Utf8WithoutMapping"},
+	}, entries[0].values, "should keep the remaining values in order")
+
+	assert.Contains(t, entries[1].values, registryValue{name: "NameServers", value: "100.0.255.254, 100.0.255.253"},
+		"should flatten a MOF array")
+
+	for _, value := range entries[1].values {
+		assert.NotContains(t, value.name, "ReturnValue", "should not read the class level parameters as values")
+	}
+}
+
+func TestParseNRPTPolicyTableEmpty(t *testing.T) {
+	assert.Empty(t, parseNRPTPolicyTable(""), "should parse no entries from empty text")
+	assert.Empty(t, parseNRPTPolicyTable("class __PARAMETERS\n{\n};\n"), "should parse no entries from a table with no instances")
+}

client/internal/debug/nrpt_windows.go

@@ -0,0 +1,317 @@
+//go:build windows
+
+package debug
+
+import (
+	"errors"
+	"fmt"
+	"runtime"
+	"strings"
+	"time"
+
+	"github.com/go-ole/go-ole"
+	"github.com/go-ole/go-ole/oleutil"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	// The NRPT policy table is reachable through the CIM class that backs
+	// Get-DnsClientNrptPolicy. Unlike the rules in the registry, the table is
+	// what the resolver currently has loaded, which is the only way to tell an
+	// applied rule from one that is merely written, in either direction.
+	nrptPolicyNamespace = `root\Microsoft\Windows\DNS`
+	nrptPolicyClass     = "PS_DnsClientNrptPolicy"
+	nrptPolicyMethod    = "Get"
+
+	// The class has no instances, so the table comes from the out parameters
+	// of a static method call, rendered as MOF text: the embedded instances
+	// arrive as a safe array of objects, which cannot be read back through the
+	// COM bindings, and the text form carries all of them.
+	nrptPolicyInstanceKeyword = "instance of DnsClientPolicyConfiguration"
+
+	nrptPolicyTimeout = 15 * time.Second
+)
+
+// COM initialization results that leave the calling thread usable: S_FALSE for
+// a thread this process already initialized, RPC_E_CHANGED_MODE for one that
+// belongs to another apartment.
+const (
+	sFalse          = 0x00000001
+	rpcEChangedMode = 0x80010106
+)
+
+// nrptQueryInFlight admits one read of the policy table at a time. A provider
+// that stops answering keeps its goroutine and the OS thread that goroutine
+// pinned, so a later bundle reports that instead of pinning another one.
+var nrptQueryInFlight = make(chan struct{}, 1)
+
+// nrptPolicyEntry is one namespace of the effective policy table, holding the
+// values of an embedded DnsClientPolicyConfiguration instance in the order the
+// provider reported them.
+type nrptPolicyEntry struct {
+	namespace string
+	values    []registryValue
+}
+
+// registryValue is a name and its rendered value, shared by the registry and
+// policy table readers so both anonymize by value name the same way.
+type registryValue struct {
+	name  string
+	value string
+}
+
+// effectiveNRPTPolicies reads the effective NRPT table. The call is bounded
+// because a WMI provider can block indefinitely and a debug bundle must not.
+func effectiveNRPTPolicies() ([]nrptPolicyEntry, error) {
+	type result struct {
+		text string
+		err  error
+	}
+
+	select {
+	case nrptQueryInFlight <- struct{}{}:
+	default:
+		return nil, errors.New("an earlier read of the policy table has not returned")
+	}
+
+	done := make(chan result, 1)
+	go func() {
+		// the slot is released here rather than by the caller, so a read that
+		// outlives the timeout holds it until the provider answers
+		defer func() { <-nrptQueryInFlight }()
+
+		text, err := nrptPolicyTableText()
+		done <- result{text: text, err: err}
+	}()
+
+	select {
+	case res := <-done:
+		if res.err != nil {
+			return nil, res.err
+		}
+		return parseNRPTPolicyTable(res.text), nil
+	case <-time.After(nrptPolicyTimeout):
+		return nil, errors.New("read of the policy table timed out")
+	}
+}
+
+// nrptPolicyTableText calls the policy table method and returns the MOF text of
+// its out parameters.
+func nrptPolicyTableText() (text string, err error) {
+	// COM is per thread, and the collection is short lived, so the thread is
+	// pinned for the duration rather than initialized for the process.
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	defer func() {
+		// The COM call chain is dynamically typed, so a provider that answers
+		// with an unexpected shape must not take the daemon down with it.
+		if r := recover(); r != nil {
+			err = fmt.Errorf("read NRPT policy table: %v", r)
+		}
+	}()
+
+	owns, err := coInitialize()
+	if err != nil {
+		return "", err
+	}
+	if owns {
+		defer ole.CoUninitialize()
+	}
+
+	locator, err := oleutil.CreateObject("WbemScripting.SWbemLocator")
+	if err != nil {
+		return "", fmt.Errorf("create WMI locator: %w", err)
+	}
+	defer locator.Release()
+
+	dispatch, err := locator.QueryInterface(ole.IID_IDispatch)
+	if err != nil {
+		return "", fmt.Errorf("query WMI locator interface: %w", err)
+	}
+	defer dispatch.Release()
+
+	service, err := dispatchCall(dispatch, "ConnectServer", nil, nrptPolicyNamespace)
+	if err != nil {
+		return "", fmt.Errorf("connect to %s: %w", nrptPolicyNamespace, err)
+	}
+	defer service.Release()
+
+	inParams, err := spawnMethodInParams(service)
+	if err != nil {
+		return "", err
+	}
+	defer inParams.Release()
+
+	// The effective table is the merge of the local and the group policy
+	// store, which is what the resolver answers from.
+	if _, err := oleutil.PutProperty(inParams, "Effective", true); err != nil {
+		return "", fmt.Errorf("set Effective parameter: %w", err)
+	}
+
+	outParams, err := dispatchCall(service, "ExecMethod", nrptPolicyClass, nrptPolicyMethod, inParams)
+	if err != nil {
+		return "", fmt.Errorf("call %s.%s: %w", nrptPolicyClass, nrptPolicyMethod, err)
+	}
+	defer outParams.Release()
+
+	textVariant, err := oleutil.CallMethod(outParams, "GetObjectText_")
+	if err != nil {
+		return "", fmt.Errorf("render policy table: %w", err)
+	}
+	defer func() {
+		if err := textVariant.Clear(); err != nil {
+			log.Debugf("clear policy table variant: %v", err)
+		}
+	}()
+
+	return textVariant.ToString(), nil
+}
+
+// spawnMethodInParams builds the in parameters instance the method needs. The
+// provider rejects the call without one, even when every parameter is optional.
+func spawnMethodInParams(service *ole.IDispatch) (*ole.IDispatch, error) {
+	class, err := dispatchCall(service, "Get", nrptPolicyClass)
+	if err != nil {
+		return nil, fmt.Errorf("get class %s: %w", nrptPolicyClass, err)
+	}
+	defer class.Release()
+
+	methods, err := dispatchProperty(class, "Methods_")
+	if err != nil {
+		return nil, fmt.Errorf("get class methods: %w", err)
+	}
+	defer methods.Release()
+
+	method, err := dispatchCall(methods, "Item", nrptPolicyMethod)
+	if err != nil {
+		return nil, fmt.Errorf("get method %s: %w", nrptPolicyMethod, err)
+	}
+	defer method.Release()
+
+	params, err := dispatchProperty(method, "InParameters")
+	if err != nil {
+		return nil, fmt.Errorf("get method parameters: %w", err)
+	}
+	defer params.Release()
+
+	inParams, err := dispatchCall(params, "SpawnInstance_")
+	if err != nil {
+		return nil, fmt.Errorf("spawn parameter instance: %w", err)
+	}
+
+	return inParams, nil
+}
+
+// parseNRPTPolicyTable pulls the embedded instances out of the MOF text. Each
+// instance is a namespace of the table, with one name and value per line.
+func parseNRPTPolicyTable(text string) []nrptPolicyEntry {
+	var entries []nrptPolicyEntry
+	var current *nrptPolicyEntry
+
+	for _, line := range strings.Split(text, "\n") {
+		line = strings.TrimSpace(strings.TrimSuffix(strings.TrimSpace(line), ";"))
+
+		switch {
+		case strings.HasPrefix(line, nrptPolicyInstanceKeyword):
+			entries = append(entries, nrptPolicyEntry{})
+			current = &entries[len(entries)-1]
+			continue
+		case strings.HasPrefix(line, "}"):
+			// closes an instance, and the array with the last one, so the
+			// class level parameters that follow are not read as values
+			current = nil
+			continue
+		case current == nil, line == "{":
+			continue
+		}
+
+		name, value, ok := strings.Cut(line, " = ")
+		if !ok {
+			continue
+		}
+
+		value = unquoteMOFValue(value)
+		if name == "Namespace" {
+			current.namespace = value
+			continue
+		}
+
+		current.values = append(current.values, registryValue{name: name, value: value})
+	}
+
+	return entries
+}
+
+// unquoteMOFValue renders a MOF scalar or array as plain text: "a" becomes a,
+// and {"a", "b"} becomes a, b.
+func unquoteMOFValue(value string) string {
+	value = strings.TrimSpace(value)
+
+	if inner, ok := strings.CutPrefix(value, "{"); ok {
+		value = strings.TrimSuffix(inner, "}")
+
+		entries := strings.Split(value, ",")
+		for i, entry := range entries {
+			entries[i] = strings.Trim(strings.TrimSpace(entry), `"`)
+		}
+		return strings.Join(entries, ", ")
+	}
+
+	return strings.Trim(value, `"`)
+}
+
+// coInitialize prepares the calling thread for COM and reports whether this
+// call owns the initialization, which decides whether it may be balanced with
+// CoUninitialize. S_FALSE took a reference on a thread this process had already
+// initialized and so has to be released, while RPC_E_CHANGED_MODE took none:
+// the thread belongs to another apartment, which is usable but is not ours to
+// uninitialize.
+func coInitialize() (bool, error) {
+	err := ole.CoInitializeEx(0, ole.COINIT_MULTITHREADED)
+	if err == nil {
+		return true, nil
+	}
+
+	var oleErr *ole.OleError
+	if errors.As(err, &oleErr) {
+		switch oleErr.Code() {
+		case sFalse:
+			return true, nil
+		case rpcEChangedMode:
+			return false, nil
+		}
+	}
+
+	return false, fmt.Errorf("initialize COM: %w", err)
+}
+
+// dispatchCall calls a COM method that returns an object.
+func dispatchCall(dispatch *ole.IDispatch, method string, params ...any) (*ole.IDispatch, error) {
+	variant, err := oleutil.CallMethod(dispatch, method, params...)
+	if err != nil {
+		return nil, err
+	}
+
+	object := variant.ToIDispatch()
+	if object == nil {
+		return nil, fmt.Errorf("%s returned no object", method)
+	}
+
+	return object, nil
+}
+
+// dispatchProperty reads a COM property that holds an object.
+func dispatchProperty(dispatch *ole.IDispatch, property string) (*ole.IDispatch, error) {
+	variant, err := oleutil.GetProperty(dispatch, property)
+	if err != nil {
+		return nil, err
+	}
+
+	object := variant.ToIDispatch()
+	if object == nil {
+		return nil, fmt.Errorf("property %s holds no object", property)
+	}
+
+	return object, nil
+}

client/internal/dns/host_windows.go

@@ -31,10 +31,28 @@ var (
 	dnsFlushResolverCacheFn = dnsapi.NewProc("DnsFlushResolverCache")
 )
 
+// Registry locations of the host DNS configuration this package programs,
+// exported so a diagnostic reader reports the same locations that are written.
 const (
-	dnsPolicyConfigMatchPath    = `SYSTEM\CurrentControlSet\Services\Dnscache\Parameters\DnsPolicyConfig\NetBird-Match`
-	gpoDnsPolicyRoot            = `SOFTWARE\Policies\Microsoft\Windows NT\DNSClient\DnsPolicyConfig`
-	gpoDnsPolicyConfigMatchPath = gpoDnsPolicyRoot + `\NetBird-Match`
+	// NRPTKeyPrefix starts the name of every NRPT rule key this client creates.
+	NRPTKeyPrefix = "NetBird-Match"
+
+	// DNSPolicyConfigRoot holds the NRPT rules of the local policy store.
+	DNSPolicyConfigRoot = `SYSTEM\CurrentControlSet\Services\Dnscache\Parameters\DnsPolicyConfig`
+
+	// GPODNSPolicyConfigRoot holds the NRPT rules of the group policy store,
+	// which takes precedence over the local one when it is present.
+	GPODNSPolicyConfigRoot = `SOFTWARE\Policies\Microsoft\Windows NT\DNSClient\DnsPolicyConfig`
+
+	// InterfaceConfigPath and InterfaceConfigPathV6 hold the per-interface DNS
+	// settings, keyed by interface GUID, in separate hives per address family.
+	InterfaceConfigPath   = `SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\Interfaces`
+	InterfaceConfigPathV6 = `SYSTEM\CurrentControlSet\Services\Tcpip6\Parameters\Interfaces`
+)
+
+const (
+	dnsPolicyConfigMatchPath    = DNSPolicyConfigRoot + `\` + NRPTKeyPrefix
+	gpoDnsPolicyConfigMatchPath = GPODNSPolicyConfigRoot + `\` + NRPTKeyPrefix
 
 	dnsPolicyConfigVersionKey           = "Version"
 	dnsPolicyConfigVersionValue         = 2
@@ -45,8 +63,6 @@ const (
 
 	nrptMaxDomainsPerRule = 50
 
-	interfaceConfigPath           = `SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\Interfaces`
-	interfaceConfigPathV6         = `SYSTEM\CurrentControlSet\Services\Tcpip6\Parameters\Interfaces`
 	interfaceConfigNameServerKey  = "NameServer"
 	interfaceConfigDhcpNameSrvKey = "DhcpNameServer"
 	interfaceConfigSearchListKey  = "SearchList"
@@ -84,7 +100,7 @@ func newHostManager(wgInterface WGIface) (*registryConfigurator, error) {
 	}
 
 	var useGPO bool
-	k, err := registry.OpenKey(registry.LOCAL_MACHINE, gpoDnsPolicyRoot, registry.QUERY_VALUE)
+	k, err := registry.OpenKey(registry.LOCAL_MACHINE, GPODNSPolicyConfigRoot, registry.QUERY_VALUE)
 	if err != nil {
 		log.Debugf("failed to open GPO DNS policy root: %v", err)
 	} else {
@@ -123,7 +139,7 @@ func (r *registryConfigurator) captureOriginalNameservers() ([]netip.Addr, error
 	seen := make(map[netip.Addr]struct{})
 	var out []netip.Addr
 	var merr *multierror.Error
-	for _, root := range []string{interfaceConfigPath, interfaceConfigPathV6} {
+	for _, root := range []string{InterfaceConfigPath, InterfaceConfigPathV6} {
 		addrs, err := r.captureFromTcpipRoot(root)
 		if err != nil {
 			merr = multierror.Append(merr, fmt.Errorf("%s: %w", root, err))
@@ -496,7 +512,7 @@ func (r *registryConfigurator) deleteInterfaceRegistryKeyProperty(propertyKey st
 }
 
 func (r *registryConfigurator) getInterfaceRegistryKey() (registry.Key, error) {
-	regKeyPath := interfaceConfigPath + "\\" + r.guid
+	regKeyPath := InterfaceConfigPath + "\\" + r.guid
 	regKey, err := registry.OpenKey(registry.LOCAL_MACHINE, regKeyPath, registry.SET_VALUE)
 	if err != nil {
 		return regKey, fmt.Errorf("open HKEY_LOCAL_MACHINE\\%s: %w", regKeyPath, err)

client/internal/profilemanager/state.go

@@ -87,9 +87,10 @@ func (pm *ProfileManager) SetActiveProfileState(state *ProfileState) error {
 
 // RemoveProfileState deletes the per-profile state file (which holds the
 // account email used for the SSO login hint and the UI display). Called after
-// a successful logout so a logged-out profile no longer shows a stale account
-// email. The state file only stores the email, so deleting it is equivalent to
-// clearing it; the next SSO login recreates it. A missing file is not an error.
+// profile removal; logout keeps the file so the next login can pass the email
+// as the login_hint. The state file only stores the email, so deleting it is
+// equivalent to clearing it; the next SSO login recreates it. A missing file
+// is not an error.
 func (pm *ProfileManager) RemoveProfileState(profileName string) error {
 	configDir, err := getConfigDir()
 	if err != nil {

client/internal/routemanager/systemops/routeselection_windows_test.go

@@ -0,0 +1,82 @@
+//go:build windows
+
+package systemops
+
+import (
+	"math"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestSortRouteCandidates(t *testing.T) {
+	tests := []struct {
+		name       string
+		candidates []candidateRoute
+		wantOrder  []uint32
+	}{
+		{
+			name: "longest prefix wins over metrics",
+			candidates: []candidateRoute{
+				{interfaceIndex: 1, prefixLength: 0, routeMetric: 0, interfaceMetric: 5},
+				{interfaceIndex: 2, prefixLength: 24, routeMetric: 100, interfaceMetric: 50},
+			},
+			wantOrder: []uint32{2, 1},
+		},
+		{
+			// Windows ranks equal-length prefixes by route metric + interface metric,
+			// so a higher route metric on a low metric interface can still win.
+			name: "combined metric beats route metric alone",
+			candidates: []candidateRoute{
+				{interfaceIndex: 8, prefixLength: 0, routeMetric: 0, interfaceMetric: 100},
+				{interfaceIndex: 5, prefixLength: 0, routeMetric: 10, interfaceMetric: 5},
+			},
+			wantOrder: []uint32{5, 8},
+		},
+		{
+			name: "lower combined metric wins",
+			candidates: []candidateRoute{
+				{interfaceIndex: 5, prefixLength: 0, routeMetric: 300, interfaceMetric: 5},
+				{interfaceIndex: 8, prefixLength: 0, routeMetric: 0, interfaceMetric: 100},
+			},
+			wantOrder: []uint32{8, 5},
+		},
+		{
+			name: "equal combined metric falls back to route metric",
+			candidates: []candidateRoute{
+				{interfaceIndex: 1, prefixLength: 0, routeMetric: 20, interfaceMetric: 10},
+				{interfaceIndex: 2, prefixLength: 0, routeMetric: 5, interfaceMetric: 25},
+			},
+			wantOrder: []uint32{2, 1},
+		},
+		{
+			// The metrics are uint32 on the Windows side, so the sum must not wrap.
+			name: "combined metric beyond the uint32 range",
+			candidates: []candidateRoute{
+				{interfaceIndex: 1, prefixLength: 0, routeMetric: math.MaxUint32, interfaceMetric: 5},
+				{interfaceIndex: 2, prefixLength: 0, routeMetric: math.MaxUint32 - 10, interfaceMetric: 5},
+			},
+			wantOrder: []uint32{2, 1},
+		},
+		{
+			name: "unknown interface metric ranks on route metric only",
+			candidates: []candidateRoute{
+				{interfaceIndex: 1, prefixLength: 0, routeMetric: 30, interfaceMetric: -1},
+				{interfaceIndex: 2, prefixLength: 0, routeMetric: 5, interfaceMetric: 10},
+			},
+			wantOrder: []uint32{2, 1},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			sortRouteCandidates(tt.candidates)
+
+			got := make([]uint32, 0, len(tt.candidates))
+			for _, c := range tt.candidates {
+				got = append(got, c.interfaceIndex)
+			}
+			assert.Equal(t, tt.wantOrder, got)
+		})
+	}
+}

client/internal/routemanager/systemops/systemops_windows.go

@@ -882,26 +882,40 @@ func getInterfaceMetric(interfaceIndex uint32, family int16) int {
 	return int(ipInterfaceRow.Metric)
 }
 
-// sortRouteCandidates sorts route candidates by priority: prefix length -> route metric -> interface metric
+// sortRouteCandidates sorts route candidates by priority: prefix length -> combined metric -> route metric.
+// Windows prefers the longest matching prefix and, among prefixes of the same length, the lowest metric, see
+// https://learn.microsoft.com/en-us/windows-hardware/customize/desktop/unattend/microsoft-windows-tcpip-interfaces-interface-routes-route-metric
 func sortRouteCandidates(candidates []candidateRoute) {
 	sort.Slice(candidates, func(i, j int) bool {
 		if candidates[i].prefixLength != candidates[j].prefixLength {
 			return candidates[i].prefixLength > candidates[j].prefixLength
 		}
-		if candidates[i].routeMetric != candidates[j].routeMetric {
-			return candidates[i].routeMetric < candidates[j].routeMetric
+		mi, mj := combinedMetric(candidates[i]), combinedMetric(candidates[j])
+		if mi != mj {
+			return mi < mj
 		}
-		return candidates[i].interfaceMetric < candidates[j].interfaceMetric
+		return candidates[i].routeMetric < candidates[j].routeMetric
 	})
 }
 
+// combinedMetric returns the effective metric Windows uses to rank routes with an equal prefix length:
+// the sum of the route metric and the metric of the interface the route is on, see
+// https://learn.microsoft.com/en-us/windows-server/networking/technologies/network-subsystem/net-sub-interface-metric
+// An unknown interface metric contributes nothing.
+func combinedMetric(candidate candidateRoute) uint64 {
+	if candidate.interfaceMetric < 0 {
+		return uint64(candidate.routeMetric)
+	}
+	return uint64(candidate.routeMetric) + uint64(candidate.interfaceMetric)
+}
+
 // GetBestInterface finds the best interface for reaching a destination,
 // excluding the VPN interface to avoid routing loops.
 //
 // Route selection priority:
 // 1. Longest prefix match (most specific route)
-// 2. Lowest route metric
-// 3. Lowest interface metric
+// 2. Lowest combined metric (route metric + interface metric)
+// 3. Lowest route metric.
 func GetBestInterface(dest netip.Addr, vpnIntf string) (*net.Interface, error) {
 	var skipInterfaceIndex int
 	if vpnIntf != "" {
@@ -925,7 +939,6 @@ func GetBestInterface(dest netip.Addr, vpnIntf string) (*net.Interface, error) {
 		return nil, fmt.Errorf("no route to %s", dest)
 	}
 
-	// Sort routes: prefix length -> route metric -> interface metric
 	sortRouteCandidates(candidates)
 
 	for _, candidate := range candidates {

client/internal/routemanager/systemops/v6route_linux_test.go

@@ -5,6 +5,7 @@ package systemops
 import (
 	"errors"
 	"net"
+	"net/netip"
 	"syscall"
 	"testing"
 
@@ -29,6 +30,7 @@ func ensureIPv6DefaultRoute(t *testing.T) {
 	}
 	if err := netlink.RouteAdd(route); err != nil {
 		if errors.Is(err, syscall.EEXIST) {
+			requireUsableIPv6Nexthop(t)
 			return
 		}
 		t.Skipf("install IPv6 fallback default route: %v", err)
@@ -38,4 +40,36 @@ func ensureIPv6DefaultRoute(t *testing.T) {
 			t.Logf("delete IPv6 fallback default route: %v", err)
 		}
 	})
+
+	requireUsableIPv6Nexthop(t)
+}
+
+// requireUsableIPv6Nexthop skips the test unless the resolved IPv6 default
+// nexthop can actually carry a route. Installing the default route succeeding
+// does not imply the kernel accepts it as a nexthop for a concrete prefix.
+func requireUsableIPv6Nexthop(t *testing.T) {
+	t.Helper()
+
+	nexthop, err := GetNextHop(netip.IPv6Unspecified())
+	if err != nil {
+		t.Skipf("resolve IPv6 default nexthop: %v", err)
+	}
+
+	probe := &netlink.Route{
+		Scope:  netlink.SCOPE_UNIVERSE,
+		Table:  syscall.RT_TABLE_MAIN,
+		Family: netlink.FAMILY_V6,
+		Dst:    &net.IPNet{IP: net.ParseIP("100::64"), Mask: net.CIDRMask(128, 128)},
+	}
+	require.NoError(t, addNextHop(nexthop, probe), "build IPv6 probe route")
+
+	switch err := netlink.RouteAdd(probe); {
+	case err == nil:
+		if err := netlink.RouteDel(probe); err != nil && !errors.Is(err, syscall.ESRCH) {
+			t.Logf("delete IPv6 probe route: %v", err)
+		}
+	case errors.Is(err, syscall.EEXIST):
+	default:
+		t.Skipf("IPv6 nexthop %s unusable for route installation: %v", nexthop, err)
+	}
 }