fix(network): prevent ping-pong switching in multi-NIC scenario

config/org.deepin.dde.network.json

@@ -349,6 +349,17 @@
             "description":"if check the network is unavailable and then switch to the available network",
             "permissions":"readwrite",
             "visibility":"private"
+         },
+         "reapplyFlags":{
+            "value":2,
+            "serial":0,
+            "flags":["global"],
+            "name":"reapplyFlags",
+            "name[zh_CN]":"reapplyConnection标志位",
+            "description[zh_CN]":"reapplyConnection的标志位，只能是0、1、2三个值之一。0：无标志；1：保留外部配置(PRESERVE_EXTERNAL)；2：非阻塞(NONBLOCKING)",
+            "description":"Flags for reapplyConnection, must be 0, 1 or 2. 0: none; 1: PRESERVE_EXTERNAL; 2: NONBLOCKING",
+            "permissions":"readwrite",
+            "visibility":"private"
          }
     }
 }

network-service-plugin/src/system/connectivitychecker.cpp

@@ -94,8 +94,10 @@ void LocalConnectionvityChecker::onConnectivityChanged(network::service::Connect
     if (m_internetChecker && (connectivity == network::service::Connectivity::Limited
                               || connectivity == network::service::Connectivity::Noconnectivity
                               || connectivity == network::service::Connectivity::Unknownconnectivity)) {
-        // 在开启网络检测的情况下，如果当前网络不可用，则启动切换主连接的工作流程，尝试恢复网络连通性。
-        QMetaObject::invokeMethod(m_internetChecker, &InternetChecker::switchInternetAccess, Qt::QueuedConnection);
+        bool checkPrimary = m_statusChecker->primaryConnectionChanged();
+        QMetaObject::invokeMethod(m_internetChecker, [this, checkPrimary]() {
+            m_internetChecker->switchInternetAccess(checkPrimary);
+        }, Qt::QueuedConnection);
     }
     qCInfo(DSM) << "Connectivity changed, incomming: " << static_cast<int>(connectivity) << ", current: " << static_cast<int>(m_connectivity);
     if (m_connectivity == connectivity)
@@ -115,6 +117,7 @@ StatusChecker::StatusChecker(QObject *parent)
     , m_connectivity(network::service::Connectivity::Unknownconnectivity)
     , m_checkCount(0)
     , m_isStop(false)
+    , m_primaryConnectionChanged(false)
 {
     qRegisterMetaType<NetworkManager::ActiveConnection::State>("NetworkManager::ActiveConnection::State");
     qRegisterMetaType<NetworkManager::Device::State>("NetworkManager::Device::State");
@@ -179,6 +182,11 @@ QString StatusChecker::detectionConnectionId() const
     return m_primaryId;
 }
 
+bool StatusChecker::primaryConnectionChanged() const
+{
+    return m_primaryConnectionChanged;
+}
+
 void StatusChecker::initDeviceConnect(const QList<QSharedPointer<NetworkManager::Device>> &devices)
 {
     for (const QSharedPointer<NetworkManager::Device> &device : devices) {
@@ -234,6 +242,12 @@ void StatusChecker::initConnectivityChecker()
 void StatusChecker::stop()
 {
     m_isStop = true;
+    if (m_checkTimer && m_checkTimer->isActive())
+        m_checkTimer->stop();
+    if (m_timer && m_timer->isActive())
+        m_timer->stop();
+    if (m_pendingCheckTimer && m_pendingCheckTimer->isActive())
+        m_pendingCheckTimer->stop();
 }
 
 void StatusChecker::onUpdataActiveState(const QSharedPointer<NetworkManager::ActiveConnection> &networks)
@@ -261,17 +275,11 @@ void StatusChecker::realStartCheck()
         return;
     }
 
-    NetworkManager::ActiveConnection::Ptr pConnection = NetworkManager::primaryConnection();
-    if (!pConnection.isNull()) {
-        m_primaryId = pConnection->connection()->uuid();
-        m_checkStartPrimaryId = m_primaryId;
-    } else {
-        m_checkStartPrimaryId.clear();
-    }
+    NetworkManager::ActiveConnection::Ptr prePrimary = NetworkManager::primaryConnection();
+    QString prePrimaryId = !prePrimary.isNull() ? prePrimary->connection()->uuid() : QString();
+
     int httpTimeout = SettingConfig::instance()->httpRequestTimeout();
     bool networkIsOk = false;
-    network::service::Connectivity detectedConnectivity = m_connectivity;
-    QString detectedPortalUrl;
     for (const QString &url : m_checkUrls) {
         network::service::HttpManager http;
         network::service::HttpReply *httpReply = http.get(url, httpTimeout);
@@ -280,41 +288,35 @@ void StatusChecker::realStartCheck()
             break;
         }
         if (httpReply->isTimeout()) {
-            qCWarning(DSM) << "check network time out, network is unavaibled ";
-            detectedConnectivity = network::service::Connectivity::Limited;
+            // 如果是读取超时了，则无需进行第二次检测，一般超时的情况下基本上都是网络不通
+            qCWarning(DSM) << "request network timeout";
             break;
         }
         int httpCode = httpReply->httpCode();
-        QString portalUrl = httpReply->portal();
         if (httpCode == 0) {
             qCWarning(DSM) << "Nework is unreachabel:" << url << httpReply->errorMessage();
-            detectedConnectivity = network::service::Connectivity::Limited;
             continue;
         }
+
+        QString portalUrl = httpReply->portal();
         networkIsOk = true;
         qCDebug(DSM) << "Http reply code:" << httpCode << ", portal url:" << portalUrl;
-        detectedPortalUrl = portalUrl;
-        detectedConnectivity = portalUrl.isEmpty() ? network::service::Connectivity::Full : network::service::Connectivity::Portal;
+        if (portalUrl.isEmpty()) {
+            // if the portal is empty, I think it ok
+            setConnectivity(network::service::Connectivity::Full);
+        } else {
+            setConnectivity(network::service::Connectivity::Portal);
+        }
+        setPortalUrl(portalUrl);
         break;
     }
-    if (m_isStop)
-        return;
-
-    // 主连接在检查期间可能已被 InternetChecker 切换，此时检测结果基于旧路由，必须丢弃
-    NetworkManager::ActiveConnection::Ptr currentPrimary = NetworkManager::primaryConnection();
-    QString currentPrimaryId;
-    if (!currentPrimary.isNull())
-        currentPrimaryId = currentPrimary->connection()->uuid();
-    if (m_checkStartPrimaryId != currentPrimaryId) {
-        qCInfo(DSM) << "Primary connection changed during check, discarding stale result"
-                    << "check:" << m_checkStartPrimaryId << "current:" << currentPrimaryId;
-        return;
+    // HTTP 检查完成后，再获取当前主连接，以检测检查期间是否发生了连接切换
+    NetworkManager::ActiveConnection::Ptr pConnection = NetworkManager::primaryConnection();
+    if (!pConnection.isNull()) {
+        m_primaryId = pConnection->connection()->uuid();
     }
-
-    if (networkIsOk) {
-        setConnectivity(detectedConnectivity);
-        setPortalUrl(detectedPortalUrl);
-    } else {
+    m_primaryConnectionChanged = (prePrimaryId != m_primaryId);
+    if (!m_isStop && !networkIsOk) {
         NetworkManager::Device::List devices = NetworkManager::networkInterfaces();
         int disconnectCount = 0;
         for (NetworkManager::Device::Ptr device : devices) {
@@ -324,13 +326,12 @@ void StatusChecker::realStartCheck()
             }
         }
         qCDebug(DSM) << "Network is unreachabel, disconnect count:" << disconnectCount;
+        setPortalUrl(QString());
         if (disconnectCount == devices.size()) {
-            // 如果所有的网络设备都断开了，就默认让其变为断开的状态
             setConnectivity(network::service::Connectivity::Noconnectivity);
         } else {
             setConnectivity(network::service::Connectivity::Limited);
         }
-        setPortalUrl(QString());
     }
 }
 

network-service-plugin/src/system/connectivitychecker.h

@@ -83,6 +83,7 @@ class StatusChecker : public QObject
     QString portalUrl() const;
     void checkConnectivity();
     QString detectionConnectionId() const;
+    bool primaryConnectionChanged() const;
 
 signals:
     void portalDetected(const QString &);
@@ -113,7 +114,7 @@ private slots:
     QStringList m_checkUrls;
     bool m_isStop;
     QString m_primaryId;
-    QString m_checkStartPrimaryId;
+    bool m_primaryConnectionChanged;
 };
 
 class NMConnectionvityChecker : public ConnectivityChecker

network-service-plugin/src/system/internetchecker.cpp

@@ -8,13 +8,13 @@
 #include "settingconfig.h"
 #include "constants.h"
 
 #include <QDebug>
 #include <QTimer>
-#include <QThread>
 #include <QElapsedTimer>
+#include <QEventLoop>
 
 #include <NetworkManagerQt/Manager>
 #include <NetworkManagerQt/ActiveConnection>
 #include <NetworkManagerQt/Ipv4Setting>
 #include <NetworkManagerQt/Ipv6Setting>
 #include <NetworkManagerQt/Settings>
@@ -79,8 +79,9 @@
     }
     if (changed) {
         conn->updateUnsaved(settingsMap);
-        if (!device.isNull())
-            device->reapplyConnection(conn->settings()->toMap(), 0, 0);
+        if (!device.isNull()) {
+            device->reapplyConnection(conn->settings()->toMap(), 0, SettingConfig::instance()->reapplyFlags());
+        }
         qCInfo(DSM) << "Set connection" << conn->name() << "never-default to" << neverDefault;
     }
     return changed;
@@ -132,7 +133,11 @@
     int httpTimeout = SettingConfig::instance()->httpRequestTimeout();
     for (int i = 0; i < maxRetry; ++i) {
         // 每次检测前等待1秒，让NM路由表和DHCP先稳定
-        QThread::sleep(1);
+        {
+            QEventLoop loop;
+            QTimer::singleShot(1000, &loop, &QEventLoop::quit);
+            loop.exec();
+        }
         bool timedOut = false;
         // 首次使用20秒超时，避免在无网线路由器的环境中长时间阻塞
         QElapsedTimer timer;
@@ -216,14 +221,27 @@
     }
 }
 
-void InternetChecker::switchInternetAccess()
+void InternetChecker::switchInternetAccess(bool checkPrimaryConnection)
 {
     // 正在切换中，忽略新的切换请求
     if (m_isSwitching) {
         qCDebug(DSM) << "current is switching";
         return;
     }
 
+    qCDebug(DSM) << "need check primary connection" << checkPrimaryConnection;
+    // 主连接在检测期间发生变化，需要重新检查当前主连接是否可上网
+    NetworkManager::ActiveConnection::Ptr primaryConnection = NetworkManager::primaryConnection();
+    if (checkPrimaryConnection && !primaryConnection.isNull()) {
+        int httpTimeout = SettingConfig::instance()->httpRequestTimeout();
+        bool isTimeout = false;
+        if (checkInternetAccessible(httpTimeout, isTimeout)) {
+            qCInfo(DSM) << "Primary connection is accessible, skip switching";
+            emit switchSuccess();
+            return;
+        }
+    }
+
     m_isSwitching = true;
     NetworkManager::Device::List availableDevice;
     NetworkManager::Device::List devices = NetworkManager::networkInterfaces();
@@ -244,7 +262,6 @@
         return;
     }
 
-    NetworkManager::ActiveConnection::Ptr primaryConnection = NetworkManager::primaryConnection();
     if (primaryConnection.isNull()) {
         qCWarning(DSM) << "primary connection is null, it will reset all never default";
         resetAllNeverDefault();

network-service-plugin/src/system/internetchecker.h

@@ -21,7 +21,7 @@ class InternetChecker : public QObject
 public:
     explicit InternetChecker(QObject *parent = nullptr);
     ~InternetChecker() override;
-    void switchInternetAccess();
+    void switchInternetAccess(bool checkPrimaryConnection = false);
 
 signals:
     void switchSuccess();

network-service-plugin/src/utils/httpmanager.cpp

@@ -8,10 +8,11 @@
 
 #include <QRegularExpression>
 #include <QDebug>
 #include <QUrl>
 
 #include <mutex>
+#include <atomic>
 #include <curl/curl.h>
 #include <sys/socket.h>
 #include <sys/time.h>
 #include <netdb.h>
@@ -23,34 +24,40 @@
 namespace {
 
 struct GetAddrInfoParams {
-    const char *node;
-    const char *service;
+    std::string node;
+    std::string service;
     const addrinfo *hints;
     addrinfo *result = nullptr;
     int ret = 0;
+    std::atomic<bool> delete_by_thread{false};
 };
 
 static void *getaddrinfo_thread(void *arg)
 {
-    auto *params = static_cast<GetAddrInfoParams *>(arg);
-    params->ret = getaddrinfo(params->node, params->service, params->hints, &params->result);
+    GetAddrInfoParams *params = static_cast<GetAddrInfoParams *>(arg);
+    params->ret = getaddrinfo(params->node.c_str(), params->service.c_str(), params->hints, &params->result);
+    // deleteByThread 为 true 说明主线程已 detach，由本线程释放 params
+    if (params->delete_by_thread.load())
+        delete params;
     return nullptr;
 }
 
 // 在独立线程中执行 getaddrinfo，主线程等待 timeoutMs 毫秒
 // 如果超时返回 ETIMEDOUT，成功返回 0
-static int getaddrinfo_with_timeout(const char *node, const char *service,
+static int getaddrinfo_with_timeout(const std::string &node, const std::string &service,
                                     const addrinfo *hints, addrinfo **result,
                                     int timeoutMs)
 {
-    GetAddrInfoParams params = {node, service, hints, nullptr, 0};
+    // 在堆上分配 params，防止 detach 的线程在超时后写入已释放的栈内存
+    GetAddrInfoParams *params = new GetAddrInfoParams{node, service, hints, nullptr, 0};
     pthread_t thread;
     pthread_attr_t attr;
     pthread_attr_init(&attr);
     pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
 
-    if (pthread_create(&thread, &attr, getaddrinfo_thread, &params) != 0) {
+    if (pthread_create(&thread, &attr, getaddrinfo_thread, params) != 0) {
         pthread_attr_destroy(&attr);
+        delete params;
         return EAI_SYSTEM;
     }
     pthread_attr_destroy(&attr);
@@ -67,13 +74,16 @@
     void *threadRet = nullptr;
     int joinRet = pthread_timedjoin_np(thread, &threadRet, &ts);
     if (joinRet == ETIMEDOUT) {
-        // 线程超时，detach 让其自行结束（getaddrinfo 最终会返回）
+        // 通知子线程由其负责释放 params
+        params->delete_by_thread.store(true);
         pthread_detach(thread);
         return ETIMEDOUT;
     }
 
-    *result = params.result;
-    return params.ret;
+    *result = params->result;
+    int ret = params->ret;
+    delete params;
+    return ret;
 }
 
 } // anonymous namespace
@@ -196,11 +206,10 @@
 
         // getaddrinfo 本身会调用系统 DNS，如果路由器不通，glibc 会等待很久
         // 用独立线程包装，只等待 timeoutSec 秒
-        // 注意：必须用局部变量持有 std::string，确保 getaddrinfo_with_timeout
-        // 返回前字符串生命周期有效，否则子线程会使用悬垂指针
+        // std::string 存入 GetAddrInfoParams 后独立持有数据，不依赖栈生命周期
         std::string hostStd = host.toStdString();
-        int ret = getaddrinfo_with_timeout(hostStd.c_str(),
-                                            nullptr,
+        int ret = getaddrinfo_with_timeout(hostStd,
+                                            {},
                                             &hints,
                                             &result,
                                             timeoutSec * 1000);

network-service-plugin/src/utils/settingconfig.cpp

@@ -89,6 +89,11 @@ bool SettingConfig::needCheckNetwork() const
     return m_needCheckNetwork;
 }
 
+int SettingConfig::reapplyFlags() const
+{
+    return m_reapplyFlags;
+}
+
 void SettingConfig::onValueChanged(const QString &key)
 {
     if (key == "reconnectIfIpConflicted") {
@@ -120,6 +125,9 @@ void SettingConfig::onValueChanged(const QString &key)
         m_protalProcessMode = dConfig->value("portalProcessMode").toString();
     } else if (key == QString("needCheckNetwork")) {
         m_needCheckNetwork = dConfig->value("needCheckNetwork").toBool();
+    } else if (key == QString("reapplyFlags")) {
+        int val = dConfig->value("reapplyFlags").toInt();
+        m_reapplyFlags = (val >= 0 && val <= 2) ? val : 2;
     }
 }
 
@@ -134,6 +142,7 @@ SettingConfig::SettingConfig(QObject *parent)
     , m_disableFailureNotify(false)
     , m_resetWifiOSDEnableTimeout(300)
     , m_needCheckNetwork(true)
+    , m_reapplyFlags(2)
 {
     if (!dConfig)
         dConfig = Dtk::Core::DConfig::create("org.deepin.dde.network", "org.deepin.dde.network");
@@ -179,6 +188,14 @@ SettingConfig::SettingConfig(QObject *parent)
             m_needCheckNetwork = dConfig->value("needCheckNetwork").toBool();
 
         m_disableFailureNotify = dConfig->value("disableFailureNotify", false).toBool();
+
+        if (keys.contains("reapplyFlags")) {
+            int val = dConfig->value("reapplyFlags").toInt();
+            if (val >= 0 && val <= 2)
+                m_reapplyFlags = val;
+            else
+                m_reapplyFlags = 2;
+        }
     }
     if (m_networkUrls.isEmpty())
         m_networkUrls = CheckUrls;