Bug 1240481 - Limit PR_Close calls during shutdown. r=mcmanus

modules/libpref/init/all.js

@@ -1481,6 +1481,10 @@ pref("network.ftp.control.qos", 0);
 
 // The max time to spend on xpcom events between two polls in ms.
 pref("network.sts.max_time_for_events_between_two_polls", 100);
+
+// During shutdown we limit PR_Close calls. If time exceeds this pref (in ms)
+// let sockets just leak.
+pref("network.sts.max_time_for_pr_close_during_shutdown", 5000);
 // </http>
 
 // 2147483647 == PR_INT32_MAX == ~2 GB

netwerk/base/nsIOService.cpp

@@ -190,6 +190,7 @@ nsIOService::nsIOService()
     , mLastOfflineStateChange(PR_IntervalNow())
     , mLastConnectivityChange(PR_IntervalNow())
     , mLastNetworkLinkChange(PR_IntervalNow())
+    , mNetTearingDownStarted(0)
 {
 }
 
@@ -1419,6 +1420,7 @@ nsIOService::Observe(nsISupports *subject,
     } else if (!strcmp(topic, kProfileChangeNetTeardownTopic)) {
         if (!mOffline) {
             mOfflineForProfileChange = true;
+            mNetTearingDownStarted = PR_IntervalNow();
             if (gHttpHandler) {
                 gHttpHandler->ShutdownConnectionManager();
             }
@@ -1447,7 +1449,9 @@ nsIOService::Observe(nsISupports *subject,
         // changes of the offline status from now. We must not allow going
         // online after this point.
         mShutdown = true;
-
+        if (!mOfflineForProfileChange) {
+          mNetTearingDownStarted = PR_IntervalNow();
+        }
         if (gHttpHandler) {
             gHttpHandler->ShutdownConnectionManager();
         }

netwerk/base/nsIOService.h

@@ -85,6 +85,7 @@ public:
     PRIntervalTime LastConnectivityChange() { return mLastConnectivityChange; }
     PRIntervalTime LastNetworkLinkChange() { return mLastNetworkLinkChange; }
     bool IsNetTearingDown() { return mShutdown || mOfflineForProfileChange; }
+    PRIntervalTime NetTearingDownStarted() { return mNetTearingDownStarted; }
     bool IsLinkUp();
 
     // Should only be called from NeckoChild. Use SetAppOffline instead.
@@ -180,9 +181,12 @@ private:
     // PR_ConnectContinue and PR_Close blocking time.  If we spend very long
     // time in any of these functions we want to know if and what network
     // change has happened shortly before.
-    mozilla::Atomic<PRIntervalTime>  mLastOfflineStateChange;
-    mozilla::Atomic<PRIntervalTime>  mLastConnectivityChange;
-    mozilla::Atomic<PRIntervalTime>  mLastNetworkLinkChange;
+    mozilla::Atomic<PRIntervalTime> mLastOfflineStateChange;
+    mozilla::Atomic<PRIntervalTime> mLastConnectivityChange;
+    mozilla::Atomic<PRIntervalTime> mLastNetworkLinkChange;
+
+    // Time a network tearing down started.
+    mozilla::Atomic<PRIntervalTime> mNetTearingDownStarted;
 public:
     // Used for all default buffer sizes that necko allocates.
     static uint32_t   gDefaultSegmentSize;

netwerk/base/nsSocketTransport2.cpp

@@ -1748,7 +1748,12 @@ nsSocketTransport::ReleaseFD_Locked(PRFileDesc *fd)
     SOCKET_LOG(("JIMB: ReleaseFD_Locked: mFDref = %d\n", mFDref));
 
     if (--mFDref == 0) {
-        if (PR_GetCurrentThread() == gSocketThread) {
+        if (gIOService->IsNetTearingDown() &&
+            ((PR_IntervalNow() - gIOService->NetTearingDownStarted()) >
+             gSocketTransportService->MaxTimeForPrClosePref())) {
+          // If shutdown last to long, let the socket leak and do not close it.
+          SOCKET_LOG(("Intentional leak"));
+        } else if (PR_GetCurrentThread() == gSocketThread) {
             SOCKET_LOG(("nsSocketTransport: calling PR_Close [this=%p]\n", this));
             PR_Close(mFD);
         } else {

netwerk/base/nsSocketTransportService2.cpp

@@ -47,6 +47,7 @@ Atomic<PRThread*, Relaxed> gSocketThread;
 #define BLIP_INTERVAL_PREF "network.activity.blipIntervalMilliseconds"
 #define MAX_TIME_BETWEEN_TWO_POLLS "network.sts.max_time_for_events_between_two_polls"
 #define TELEMETRY_PREF "toolkit.telemetry.enabled"
+#define MAX_TIME_FOR_PR_CLOSE_DURING_SHUTDOWN "network.sts.max_time_for_pr_close_during_shutdown"
 
 uint32_t nsSocketTransportService::gMaxCount;
 PRCallOnceType nsSocketTransportService::gMaxCountInitOnce;
@@ -111,6 +112,7 @@ nsSocketTransportService::nsSocketTransportService()
     , mServingPendingQueue(false)
     , mMaxTimePerPollIter(100)
     , mTelemetryEnabledPref(false)
+    , mMaxTimeForPrClosePref(PR_SecondsToInterval(5))
     , mProbedMaxCount(false)
 {
     NS_ASSERTION(NS_IsMainThread(), "wrong thread");
@@ -550,6 +552,7 @@ nsSocketTransportService::Init()
         tmpPrefService->AddObserver(KEEPALIVE_PROBE_COUNT_PREF, this, false);
         tmpPrefService->AddObserver(MAX_TIME_BETWEEN_TWO_POLLS, this, false);
         tmpPrefService->AddObserver(TELEMETRY_PREF, this, false);
+        tmpPrefService->AddObserver(MAX_TIME_FOR_PR_CLOSE_DURING_SHUTDOWN, this, false);
     }
     UpdatePrefs();
 
@@ -1218,6 +1221,13 @@ nsSocketTransportService::UpdatePrefs()
         if (NS_SUCCEEDED(rv)) {
             mTelemetryEnabledPref = telemetryPref;
         }
+
+        int32_t maxTimeForPrClosePref;
+        rv = tmpPrefService->GetIntPref(MAX_TIME_FOR_PR_CLOSE_DURING_SHUTDOWN,
+                                        &maxTimeForPrClosePref);
+        if (NS_SUCCEEDED(rv) && maxTimeForPrClosePref >=0) {
+            mMaxTimeForPrClosePref = PR_MillisecondsToInterval(maxTimeForPrClosePref);
+        }
     }
     
     return NS_OK;

netwerk/base/nsSocketTransportService2.h

@@ -112,6 +112,7 @@ public:
     bool IsKeepaliveEnabled() { return mKeepaliveEnabledPref; }
 
     bool IsTelemetryEnabled() { return mTelemetryEnabledPref; }
+    PRIntervalTime MaxTimeForPrClosePref() {return mMaxTimeForPrClosePref; }
 protected:
 
     virtual ~nsSocketTransportService();
@@ -237,6 +238,7 @@ private:
     mozilla::Atomic<bool>  mServingPendingQueue;
     mozilla::Atomic<int32_t, mozilla::Relaxed> mMaxTimePerPollIter;
     mozilla::Atomic<bool, mozilla::Relaxed>  mTelemetryEnabledPref;
+    mozilla::Atomic<PRIntervalTime, mozilla::Relaxed> mMaxTimeForPrClosePref;
 
     void OnKeepaliveEnabledPrefChange();
     void NotifyKeepaliveEnabledPrefChange(SocketContext *sock);