fix: bad rule xargs, cgroup v2 mkdir, amnvpnrt rt_tables, restart flush

fix: solve infinite reconnect NM down

client/daemon/daemon.cpp

@@ -637,7 +637,7 @@ void Daemon::checkHandshake() {
       pendingHandshakes++;
     }
   }
-
+  
   // Check again if there were connections that haven't completed a handshake.
   if (pendingHandshakes > 0) {
     m_handshakeTimer.start(HANDSHAKE_POLL_MSEC);

client/platforms/linux/daemon/dnsutilslinux.cpp

@@ -7,10 +7,14 @@
 #include <net/if.h>
 
 #include <QDBusVariant>
+#include <QNetworkInterface>
+#include <QTimer>
 #include <QtDBus/QtDBus>
 
+#include "core/utils/networkUtilities.h"
 #include "leakdetector.h"
 #include "logger.h"
+#include "router_linux.h"
 
 constexpr const char* DBUS_RESOLVE_SERVICE = "org.freedesktop.resolve1";
 constexpr const char* DBUS_RESOLVE_PATH = "/org/freedesktop/resolve1";
@@ -27,24 +31,56 @@ DnsUtilsLinux::DnsUtilsLinux(QObject* parent) : DnsUtils(parent) {
   logger.debug() << "DnsUtilsLinux created.";
 
   QDBusConnection conn = QDBusConnection::systemBus();
-  m_resolver = new QDBusInterface(DBUS_RESOLVE_SERVICE, DBUS_RESOLVE_PATH,
-                                  DBUS_RESOLVE_MANAGER, conn, this);
+  auto* watcher = new QDBusServiceWatcher(
+      DBUS_RESOLVE_SERVICE, conn,
+      QDBusServiceWatcher::WatchForRegistration |
+      QDBusServiceWatcher::WatchForUnregistration, this);
+
+  connect(watcher, &QDBusServiceWatcher::serviceRegistered,
+          this, &DnsUtilsLinux::onResolverRegistered);
+  connect(watcher, &QDBusServiceWatcher::serviceUnregistered,
+          this, &DnsUtilsLinux::onResolverUnregistered);
+
+  if (conn.interface()->isServiceRegistered(DBUS_RESOLVE_SERVICE)) {
+    onResolverRegistered();
+  }
+}
+
+void DnsUtilsLinux::onResolverRegistered() {
+  m_resolver.reset(new QDBusInterface(DBUS_RESOLVE_SERVICE, DBUS_RESOLVE_PATH,
+                                      DBUS_RESOLVE_MANAGER,
+                                      QDBusConnection::systemBus()));
+  logger.debug() << "systemd-resolved available, DNS resolver initialized";
+
+  if (!m_pendingIfname.isEmpty()) {
+    logger.debug() << "Re-applying DNS configuration for" << m_pendingIfname;
+    updateResolvers(m_pendingIfname, m_pendingResolvers);
+  }
+}
+
+void DnsUtilsLinux::onResolverUnregistered() {
+  logger.debug() << "systemd-resolved disappeared, dropping DNS resolver";
+  m_resolver.reset();
 }
 
 DnsUtilsLinux::~DnsUtilsLinux() {
   MZ_COUNT_DTOR(DnsUtilsLinux);
 
-  for (auto iterator = m_linkDomains.constBegin();
-       iterator != m_linkDomains.constEnd(); ++iterator) {
-    QList<QVariant> argumentList;
-    argumentList << QVariant::fromValue(iterator.key());
-    argumentList << QVariant::fromValue(iterator.value());
-    m_resolver->asyncCallWithArgumentList(QStringLiteral("SetLinkDomains"),
-                                          argumentList);
-  }
+  if (m_resolver) {
+    if (m_gatewayIfindex > 0)
+      setLinkDefaultRoute(m_gatewayIfindex, true);
 
-  if (m_ifindex > 0) {
-    m_resolver->asyncCall(QStringLiteral("RevertLink"), m_ifindex);
+    for (auto iterator = m_linkDomains.constBegin();
+         iterator != m_linkDomains.constEnd(); ++iterator) {
+      QList<QVariant> argumentList;
+      argumentList << QVariant::fromValue(iterator.key());
+      argumentList << QVariant::fromValue(iterator.value());
+      m_resolver->asyncCallWithArgumentList(QStringLiteral("SetLinkDomains"),
+                                            argumentList);
+    }
+    if (m_ifindex > 0) {
+      m_resolver->asyncCall(QStringLiteral("RevertLink"), m_ifindex);
+    }
   }
 
   logger.debug() << "DnsUtilsLinux destroyed.";
@@ -52,19 +88,52 @@ DnsUtilsLinux::~DnsUtilsLinux() {
 
 bool DnsUtilsLinux::updateResolvers(const QString& ifname,
                                     const QList<QHostAddress>& resolvers) {
+  if (m_gatewayIfindex > 0) {
+    setLinkDefaultRoute(m_gatewayIfindex, true);
+    m_gatewayIfindex = 0;
+  }
+
+  const int previousIfindex = m_ifindex;
   m_ifindex = if_nametoindex(qPrintable(ifname));
   if (m_ifindex <= 0) {
     logger.error() << "Unable to resolve ifindex for" << ifname;
     return false;
   }
 
+  m_pendingIfname = ifname;
+  m_pendingResolvers = resolvers;
+
+  if (!m_resolver) {
+    logger.debug() << "systemd-resolved not ready, queuing DNS configuration";
+    return true;
+  }
+
+  const int gwIdx = NetworkUtilities::getGatewayAndIface().second.index();
+  if (gwIdx > 0 && gwIdx != m_ifindex && gwIdx != m_gatewayIfindex) {
+    m_gatewayIfindex = gwIdx;
+    setLinkDefaultRoute(gwIdx, false);
+  }
+
   setLinkDNS(m_ifindex, resolvers);
   setLinkDefaultRoute(m_ifindex, true);
   updateLinkDomains();
+
+  if (previousIfindex > 0 && previousIfindex != m_ifindex) {
+    m_resolver->callWithArgumentList(QDBus::Block, QStringLiteral("RevertLink"),
+                                     {QVariant::fromValue(previousIfindex)});
+  }
   return true;
 }
 
 bool DnsUtilsLinux::restoreResolvers() {
+  m_pendingIfname.clear();
+  m_pendingResolvers.clear();
+
+  if (m_gatewayIfindex > 0) {
+    setLinkDefaultRoute(m_gatewayIfindex, true);
+    m_gatewayIfindex = 0;
+  }
+
   for (auto iterator = m_linkDomains.constBegin();
        iterator != m_linkDomains.constEnd(); ++iterator) {
     setLinkDomains(iterator.key(), iterator.value());
@@ -72,7 +141,7 @@ bool DnsUtilsLinux::restoreResolvers() {
   m_linkDomains.clear();
 
   /* Revert the VPN interface's DNS configuration */
-  if (m_ifindex > 0) {
+  if (m_ifindex > 0 && m_resolver) {
     QList<QVariant> argumentList = {QVariant::fromValue(m_ifindex)};
     QDBusPendingReply<> reply = m_resolver->asyncCallWithArgumentList(
         QStringLiteral("RevertLink"), argumentList);
@@ -90,13 +159,17 @@ bool DnsUtilsLinux::restoreResolvers() {
 void DnsUtilsLinux::dnsCallCompleted(QDBusPendingCallWatcher* call) {
   QDBusPendingReply<> reply = *call;
   if (reply.isError()) {
-    logger.error() << "Error received from the DBus service";
+    logger.debug() << "DBus call failed (may be transient after systemd-resolved restart)";
+    logger.debug() << "Restarting resolved to clear its query backlog";
+    RouterLinux::Instance().flushDns();
+    scheduleRetry();
   }
   delete call;
 }
 
 void DnsUtilsLinux::setLinkDNS(int ifindex,
                                const QList<QHostAddress>& resolvers) {
+  if (!m_resolver) return;
   QList<DnsResolver> resolverList;
   char ifnamebuf[IF_NAMESIZE];
   const char* ifname = if_indextoname(ifindex, ifnamebuf);
@@ -121,6 +194,7 @@ void DnsUtilsLinux::setLinkDNS(int ifindex,
 
 void DnsUtilsLinux::setLinkDomains(int ifindex,
                                    const QList<DnsLinkDomain>& domains) {
+  if (!m_resolver) return;
   char ifnamebuf[IF_NAMESIZE];
   const char* ifname = if_indextoname(ifindex, ifnamebuf);
   if (ifname) {
@@ -144,6 +218,7 @@ void DnsUtilsLinux::setLinkDomains(int ifindex,
 }
 
 void DnsUtilsLinux::setLinkDefaultRoute(int ifindex, bool enable) {
+  if (!m_resolver) return;
   QList<QVariant> argumentList;
   argumentList << QVariant::fromValue(ifindex);
   argumentList << QVariant::fromValue(enable);
@@ -156,6 +231,7 @@ void DnsUtilsLinux::setLinkDefaultRoute(int ifindex, bool enable) {
 }
 
 void DnsUtilsLinux::updateLinkDomains() {
+  if (!m_resolver) return;
   /* Get the list of search domains, and remove any others that might conspire
    * to satisfy DNS resolution. Unfortunately, this is a pain because Qt doesn't
    * seem to be able to demarshall complex property types.
@@ -174,11 +250,20 @@ void DnsUtilsLinux::updateLinkDomains() {
 
 void DnsUtilsLinux::dnsDomainsReceived(QDBusPendingCallWatcher* call) {
   QDBusPendingReply<QVariant> reply = *call;
+  call->deleteLater();
   if (reply.isError()) {
-    logger.error() << "Error retrieving the DNS  domains from the DBus service";
-    delete call;
+    // systemd-resolved may still be starting up after a restart — retry a few times
+    if (m_ifindex > 0 && m_domainRetries++ < 5) {
+      logger.debug() << "systemd-resolved not ready yet, retrying DNS setup ("
+                     << m_domainRetries << "/5)";
+      QTimer::singleShot(500, this, &DnsUtilsLinux::updateLinkDomains);
+    } else {
+      logger.warning() << "Failed to configure DNS after 5 retries";
+      m_domainRetries = 0;
+    }
     return;
   }
+  m_domainRetries = 0;
 
   /* Update the state of the DNS domains */
   m_linkDomains.clear();
@@ -204,9 +289,30 @@ void DnsUtilsLinux::dnsDomainsReceived(QDBusPendingCallWatcher* call) {
   }
 
   /* Add a root search domain for the new interface. */
-  QList<DnsLinkDomain> newlist = {root};
-  setLinkDomains(m_ifindex, newlist);
-  delete call;
+  if (m_ifindex > 0) {
+    setLinkDomains(m_ifindex, {root});
+
+    /* Disable DefaultRoute on the physical gateway so systemd-resolved
+     * routes all DNS through the VPN interface. */
+    const int gwIdx = NetworkUtilities::getGatewayAndIface().second.index();
+    if (gwIdx > 0 && gwIdx != m_ifindex && gwIdx != m_gatewayIfindex) {
+      m_gatewayIfindex = gwIdx;
+      setLinkDefaultRoute(gwIdx, false);
+    }
+  }
+}
+
+void DnsUtilsLinux::scheduleRetry() {
+  if (m_pendingIfname.isEmpty() || m_retryPending || m_domainRetries >= 5)
+    return;
+  m_retryPending = true;
+  ++m_domainRetries;
+  logger.debug() << "Retrying full DNS setup (" << m_domainRetries << "/5)";
+  QTimer::singleShot(1000, this, [this]() {
+    m_retryPending = false;
+    if (!m_pendingIfname.isEmpty())
+      updateResolvers(m_pendingIfname, m_pendingResolvers);
+  });
 }
 
 static DnsMetatypeRegistrationProxy s_dnsMetatypeProxy;

client/platforms/linux/daemon/dnsutilslinux.h

@@ -6,7 +6,12 @@
 #define DNSUTILSLINUX_H
 
 #include <QDBusInterface>
+#include <QScopedPointer>
 #include <QDBusPendingCallWatcher>
+#include <QDBusServiceWatcher>
+#include <QHostAddress>
+#include <QList>
+#include <QString>
 
 #include "daemon/dnsutils.h"
 #include "dbustypeslinux.h"
@@ -29,13 +34,22 @@ class DnsUtilsLinux final : public DnsUtils {
   void updateLinkDomains();
 
  private slots:
+  void onResolverRegistered();
+  void onResolverUnregistered();
   void dnsCallCompleted(QDBusPendingCallWatcher*);
   void dnsDomainsReceived(QDBusPendingCallWatcher*);
 
  private:
+  void scheduleRetry();
+
   int m_ifindex = 0;
+  int m_gatewayIfindex = 0;
+  int m_domainRetries = 0;
+  bool m_retryPending = false;
   QMap<int, DnsLinkDomainList> m_linkDomains;
-  QDBusInterface* m_resolver = nullptr;
+  QScopedPointer<QDBusInterface> m_resolver;
+  QString m_pendingIfname;
+  QList<QHostAddress> m_pendingResolvers;
 };
 
 #endif  // DNSUTILSLINUX_H

client/platforms/linux/daemon/linuxfirewall.cpp

@@ -33,6 +33,7 @@
 #include "linuxfirewall.h"
 #include "logger.h"
 #include "xray_defs.h"
+#include <QFileInfo>
 #include <QProcess>
 
 #define BRAND_CODE "amn"
@@ -109,7 +110,7 @@ int LinuxFirewall::linkChain(LinuxFirewall::IPVersion ip, const QString& chain,
         //    (we can't safely delete all rules at once since rule numbers change)
         // TODO: occasionally this script results in warnings in logs "Bad rule (does a matching rule exist in the chain?)" - this happens when
         // the e.g OUTPUT chain is empty but this script attempts to delete things from it anyway. It doesn't cause any problems, but we should still fix at some point..
-        return execute(QStringLiteral("if ! %1 -L %2 -n --line-numbers -t %4 2> /dev/null | awk 'int($1) == 1 && $2 == \"%3\" { found=1 } END { if(found==1) { exit 0 } else { exit 1 } }' ; then %1 -I %2 -j %3 -t %4 && %1 -L %2 -n --line-numbers -t %4 2> /dev/null | awk 'int($1) > 1 && $2 == \"%3\" { print $1; exit }' | xargs %1 -t %4 -D %2 ; fi").arg(cmd, parent, chain, tableName));
+        return execute(QStringLiteral("if ! %1 -L %2 -n --line-numbers -t %4 2> /dev/null | awk 'int($1) == 1 && $2 == \"%3\" { found=1 } END { if(found==1) { exit 0 } else { exit 1 } }' ; then %1 -I %2 -j %3 -t %4 && %1 -L %2 -n --line-numbers -t %4 2> /dev/null | awk 'int($1) > 1 && $2 == \"%3\" { print $1; exit }' | xargs -r %1 -t %4 -D %2 ; fi").arg(cmd, parent, chain, tableName));
     }
     else
         return execute(QStringLiteral("if ! %1 -C %2 -j %3 -t %4 2> /dev/null ; then %1 -A %2 -j %3 -t %4; fi").arg(cmd, parent, chain, tableName));
@@ -501,13 +502,22 @@ int LinuxFirewall::execute(const QString &command, bool ignoreErrors)
         logger.debug() << "(" << exitCode << ") $ " << command;
     if (!out.isEmpty())
         logger.info() << out;
-    if (!err.isEmpty())
+    if (!err.isEmpty() && !ignoreErrors)
         logger.warning() << err;
     return exitCode;
 }
 
 void LinuxFirewall::setupTrafficSplitting()
 {
+    const QString cgroupBase = QStringLiteral("/sys/fs/cgroup/net_cls");
+    if (!QFileInfo::exists(cgroupBase)) {
+        logger.warning() << "net_cls cgroup v1 not available, traffic splitting disabled";
+        return;
+    }
+    execute(QStringLiteral(
+        "if ! grep -qE '^[0-9]+[[:space:]]+%1$' /etc/iproute2/rt_tables 2>/dev/null ; then "
+        "echo '200 %1' >> /etc/iproute2/rt_tables ; fi"
+    ).arg(kRtableName));
     auto cGroupDir = "/sys/fs/cgroup/net_cls/" BRAND_CODE "vpnexclusions/";
     logger.info() << "Should be setting up cgroup in" << cGroupDir << "for traffic splitting";
     execute(QStringLiteral("if [ ! -d %1 ] ; then mkdir %1 ; sleep 0.1 ; echo %2 > %1/net_cls.classid ; fi").arg(cGroupDir).arg(kCGroupId));
@@ -519,6 +529,7 @@ void LinuxFirewall::teardownTrafficSplitting()
 {
     logger.info() << "Tearing down cgroup and routing rules";
     execute(QStringLiteral("if ip rule list | grep -q %1; then ip rule del from all fwmark %1 lookup %2 2> /dev/null ; fi").arg(kPacketTag, kRtableName));
-    execute(QStringLiteral("ip route flush table %1").arg(kRtableName));
+    execute(QStringLiteral("ip route flush table %1").arg(kRtableName), true);
     execute(QStringLiteral("ip route flush cache"));
+    execute(QStringLiteral("sed -i '/%1/d' /etc/iproute2/rt_tables").arg(kRtableName));
 }

client/platforms/linux/daemon/wireguardutilslinux.cpp

@@ -263,7 +263,11 @@ bool WireguardUtilsLinux::updatePeer(const InterfaceConfig& config) {
     // Exclude the server address, except for multihop exit servers.
     if ((config.m_hopType != InterfaceConfig::MultiHopExit) &&
         (m_rtmonitor != nullptr)) {
-        m_rtmonitor->addExclusionRoute(IPAddress(config.m_serverIpv4AddrIn));
+        if (!config.m_serverIpv4AddrIn.isEmpty() &&
+            !m_rtmonitor->addExclusionRoute(IPAddress(config.m_serverIpv4AddrIn))) {
+            logger.error() << "No gateway — cannot add server exclusion route";
+            return false;
+        }
         m_rtmonitor->addExclusionRoute(IPAddress(config.m_serverIpv6AddrIn));
     }
 

service/server/router_linux.cpp

@@ -178,11 +178,16 @@ bool RouterLinux::flushDns()
     }
 
     p.waitForFinished();
-    QByteArray output(p.readAll());
+    QByteArray output = p.readAll();
+    if ((p.exitStatus() != QProcess::NormalExit) || (p.exitCode() != 0)) {
+        qDebug().noquote() << "Failed to flush DNS: " + output;
+        return false;
+    }
+
     if (output.isEmpty())
         qDebug().noquote() << "Flush dns completed";
     else
-        qDebug().noquote() << "OUTPUT systemctl restart nscd/systemd-resolved: " + output;
+        qDebug().noquote() << "OUTPUT systemctl restart: " + output;
 
     return true;
 }