Title: [183259] trunk
Revision
183259
Author
[email protected]
Date
2015-04-24 04:05:40 -0700 (Fri, 24 Apr 2015)

Log Message

Source/WebCore:
CrashTracer: WebProcess at com.apple.WebCore: WebCore::toScriptElementIfPossible + 4
https://bugs.webkit.org/show_bug.cgi?id=144050

Reviewed by Chris Dumez.

We are seeing null Element pointer crashes with this stack:

47 com.apple.WebCore:  WebCore::toScriptElementIfPossible + 4 <==
47 com.apple.WebCore:  WebCore::ScriptRunner::timerFired + 452
47 com.apple.WebCore:  WebCore::ThreadTimers::sharedTimerFiredInternal + 175

The most likely cause seems to be that this code

    ASSERT(m_pendingAsyncScripts.contains(scriptElement));
    m_scriptsToExecuteSoon.append(m_pendingAsyncScripts.take(scriptElement));

in ScriptRunner::notifyScriptReady fails to find scriptElement and we are left with null entry in
m_scriptsToExecuteSoon. However I haven't managed to repro this or find the exact path how this
could happen. The related code is fragile with lot of state (especially in ScriptElement class)
and involves many opportunities for re-entry via scripts.

No repro, no test case.

* dom/ScriptRunner.cpp:
(WebCore::ScriptRunner::timerFired):

    Paper this over by adding a null check. We could check m_pendingAsyncScripts.take() above
    but this also covers possibility this is caused by something else.

LayoutTests:
Memory cache live resources repeatedly purged during painting
https://bugs.webkit.org/show_bug.cgi?id=144104

Reviewed by Chris Dumez.

* http/tests/cache/disk-cache/disk-cache-vary-cookie.html:

    These clearMemoryCache calls are now done by cache-test.js.

* http/tests/cache/disk-cache/resources/cache-test.js:
(loadResources):

    Make sure to clear the memory cache explicitly in the beginning so we always hit the disk cache.

(runTests):

Modified Paths

Diff

Modified: trunk/LayoutTests/ChangeLog (183258 => 183259)


--- trunk/LayoutTests/ChangeLog	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/LayoutTests/ChangeLog	2015-04-24 11:05:40 UTC (rev 183259)
@@ -1,3 +1,21 @@
+2015-04-24  Antti Koivisto  <[email protected]>
+
+        Memory cache live resources repeatedly purged during painting
+        https://bugs.webkit.org/show_bug.cgi?id=144104
+
+        Reviewed by Chris Dumez.
+
+        * http/tests/cache/disk-cache/disk-cache-vary-cookie.html:
+
+            These clearMemoryCache calls are now done by cache-test.js.
+
+        * http/tests/cache/disk-cache/resources/cache-test.js:
+        (loadResources):
+
+            Make sure to clear the memory cache explicitly in the beginning so we always hit the disk cache.
+
+        (runTests):
+
 2015-04-23  Basile Clement  <[email protected]>
 
         Allow function allocation sinking

Modified: trunk/LayoutTests/http/tests/cache/disk-cache/disk-cache-vary-cookie.html (183258 => 183259)


--- trunk/LayoutTests/http/tests/cache/disk-cache/disk-cache-vary-cookie.html	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/LayoutTests/http/tests/cache/disk-cache/disk-cache-vary-cookie.html	2015-04-24 11:05:40 UTC (rev 183259)
@@ -15,16 +15,13 @@
 document.cookie = "cookie=value";
 loadResources(tests, function () {
     printResults(tests);
-    internals.clearMemoryCache();
     debug("Loading again");
     loadResources(tests, function () {
         printResults(tests);
-        internals.clearMemoryCache();
         debug("Changing cookie and loading");
         document.cookie = "cookie=othervalue";
         loadResources(tests, function () {
             printResults(tests);
-            internals.clearMemoryCache()
             debug("Loading again");
             loadResources(tests, function () {
                printResults(tests);

Modified: trunk/LayoutTests/http/tests/cache/disk-cache/resources/cache-test.js (183258 => 183259)


--- trunk/LayoutTests/http/tests/cache/disk-cache/resources/cache-test.js	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/LayoutTests/http/tests/cache/disk-cache/resources/cache-test.js	2015-04-24 11:05:40 UTC (rev 183259)
@@ -69,6 +69,9 @@
 
 function loadResources(tests, completetion)
 {
+    // Otherwise we just get responses from the memory cache.
+    internals.clearMemoryCache();
+    
     var pendingCount = tests.length;
     for (var i = 0; i < tests.length; ++i) {
         loadResource(tests[i], function (ev) {
@@ -97,8 +100,6 @@
 function runTests(tests, completionHandler)
 {
     loadResources(tests, function () {
-        // Otherwise we just get responses from the memory cache.
-        internals.clearMemoryCache();
         // Wait a bit so things settle down in the disk cache.
         setTimeout(function () {
             loadResources(tests, function () {

Modified: trunk/Source/WebCore/ChangeLog (183258 => 183259)


--- trunk/Source/WebCore/ChangeLog	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/Source/WebCore/ChangeLog	2015-04-24 11:05:40 UTC (rev 183259)
@@ -1,3 +1,34 @@
+2015-04-24  Antti Koivisto  <[email protected]>
+
+        CrashTracer: WebProcess at com.apple.WebCore: WebCore::toScriptElementIfPossible + 4
+        https://bugs.webkit.org/show_bug.cgi?id=144050
+
+        Reviewed by Chris Dumez.
+
+        We are seeing null Element pointer crashes with this stack:
+
+        47 com.apple.WebCore:  WebCore::toScriptElementIfPossible + 4 <==
+        47 com.apple.WebCore:  WebCore::ScriptRunner::timerFired + 452
+        47 com.apple.WebCore:  WebCore::ThreadTimers::sharedTimerFiredInternal + 175
+
+        The most likely cause seems to be that this code
+
+            ASSERT(m_pendingAsyncScripts.contains(scriptElement));
+            m_scriptsToExecuteSoon.append(m_pendingAsyncScripts.take(scriptElement));
+
+        in ScriptRunner::notifyScriptReady fails to find scriptElement and we are left with null entry in
+        m_scriptsToExecuteSoon. However I haven't managed to repro this or find the exact path how this
+        could happen. The related code is fragile with lot of state (especially in ScriptElement class)
+        and involves many opportunities for re-entry via scripts.
+
+        No repro, no test case.
+
+        * dom/ScriptRunner.cpp:
+        (WebCore::ScriptRunner::timerFired):
+
+            Paper this over by adding a null check. We could check m_pendingAsyncScripts.take() above
+            but this also covers possibility this is caused by something else.
+
 2015-04-24  Carlos Garcia Campos  <[email protected]>
 
         [SOUP] Use a webkit subdirectory for the disk cache

Modified: trunk/Source/WebCore/loader/cache/CachedResource.cpp (183258 => 183259)


--- trunk/Source/WebCore/loader/cache/CachedResource.cpp	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/Source/WebCore/loader/cache/CachedResource.cpp	2015-04-24 11:05:40 UTC (rev 183259)
@@ -460,7 +460,7 @@
             // We allow non-secure content to be reused in history, but we do not allow secure content to be reused.
             memoryCache.remove(*this);
         }
-        memoryCache.prune();
+        memoryCache.pruneSoon();
     }
     // This object may be dead here.
 }
@@ -557,7 +557,7 @@
             memoryCache.removeFromLiveDecodedResourcesList(*this);
             memoryCache.insertInLiveDecodedResourcesList(*this);
         }
-        memoryCache.prune();
+        memoryCache.pruneSoon();
     }
 }
     

Modified: trunk/Source/WebCore/loader/cache/MemoryCache.cpp (183258 => 183259)


--- trunk/Source/WebCore/loader/cache/MemoryCache.cpp	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/Source/WebCore/loader/cache/MemoryCache.cpp	2015-04-24 11:05:40 UTC (rev 183259)
@@ -69,6 +69,7 @@
     , m_deadDecodedDataDeletionInterval(defaultDecodedDataDeletionInterval)
     , m_liveSize(0)
     , m_deadSize(0)
+    , m_pruneTimer(*this, &MemoryCache::pruneTimerFired)
 {
 }
 
@@ -745,15 +746,34 @@
     ASSERT(!m_sessionResources.contains(sessionID));
 }
 
+bool MemoryCache::needsPruning() const
+{
+    return m_liveSize + m_deadSize > m_capacity || m_deadSize > m_maxDeadCapacity;
+}
+
 void MemoryCache::prune()
 {
-    if (m_liveSize + m_deadSize <= m_capacity && m_deadSize <= m_maxDeadCapacity) // Fast path.
+    if (!needsPruning())
         return;
         
     pruneDeadResources(); // Prune dead first, in case it was "borrowing" capacity from live.
     pruneLiveResources();
 }
 
+void MemoryCache::pruneTimerFired()
+{
+    prune();
+}
+
+void MemoryCache::pruneSoon()
+{
+     if (m_pruneTimer.isActive())
+        return;
+     if (!needsPruning())
+         return;
+     m_pruneTimer.startOneShot(0);
+}
+
 #ifndef NDEBUG
 void MemoryCache::dumpStats()
 {

Modified: trunk/Source/WebCore/loader/cache/MemoryCache.h (183258 => 183259)


--- trunk/Source/WebCore/loader/cache/MemoryCache.h	2015-04-24 10:21:13 UTC (rev 183258)
+++ trunk/Source/WebCore/loader/cache/MemoryCache.h	2015-04-24 11:05:40 UTC (rev 183259)
@@ -28,6 +28,7 @@
 #include "NativeImagePtr.h"
 #include "SecurityOriginHash.h"
 #include "SessionID.h"
+#include "Timer.h"
 #include <wtf/Forward.h>
 #include <wtf/HashMap.h>
 #include <wtf/HashSet.h>
@@ -117,6 +118,7 @@
     WEBCORE_EXPORT void evictResources(SessionID);
     
     void prune();
+    void pruneSoon();
     unsigned size() const { return m_liveSize + m_deadSize; }
 
     void setDeadDecodedDataDeletionInterval(std::chrono::milliseconds interval) { m_deadDecodedDataDeletionInterval = interval; }
@@ -184,6 +186,8 @@
 
     unsigned liveCapacity() const;
     unsigned deadCapacity() const;
+    bool needsPruning() const;
+    void pruneTimerFired();
 
     CachedResource* resourceForRequestImpl(const ResourceRequest&, CachedResourceMap&);
 
@@ -213,6 +217,8 @@
     // referenced by a Web page).
     typedef HashMap<SessionID, std::unique_ptr<CachedResourceMap>> SessionCachedResourceMap;
     SessionCachedResourceMap m_sessionResources;
+
+    Timer m_pruneTimer;
 };
 
 }
_______________________________________________
webkit-changes mailing list
[email protected]
https://lists.webkit.org/mailman/listinfo/webkit-changes

Reply via email to