Bug 187607 – Make sure WebProcess::ensureNetworkProcessConnection() is always called on the main thread

• Component: WebKit2
• Status: RESOLVED FIXED

[Chris Dumez]

Make sure WebProcess::ensureNetworkProcessConnection() is always called on the main thread.

[Chris Dumez]

Created attachment 344853 [details]
Patch

[EWS Watchlist]

Comment on attachment 344853 [details]
Patch

Attachment 344853 [details] did not pass mac-wk2-ews (mac-wk2):
Output: https://webkit-queues.webkit.org/results/8517302

New failing tests:
inspector/worker/runtime-basic.html
http/tests/inspector/worker/blob-script-with-cross-domain-imported-scripts.html
inspector/page/searchInResources.html

[EWS Watchlist]

Created attachment 344863 [details]
Archive of layout-test-results from ews106 for mac-sierra-wk2

The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews.
Bot: ews106  Port: mac-sierra-wk2  Platform: Mac OS X 10.12.6

[Chris Dumez]

Well, it is already finding bugs:
Thread 7 Crashed:: WebCore: Worker
0   com.apple.WebKit              	0x000000010bebb85b WebKit::WebProcess::ensureNetworkProcessConnection() + 265 (WebProcess.cpp:1117)
1   com.apple.WebKit              	0x000000010be0e3ad WebKit::WebLoaderStrategy::setCaptureExtraNetworkLoadMetricsEnabled(bool) + 21 (DumbPtrTraits.h:41)
2   com.apple.WebCore             	0x000000010de38e27 WebCore::WorkerInspectorController::disconnectFrontend(Inspector::DisconnectReason) + 39 (WorkerInspectorController.cpp:132)
3   com.apple.WebCore             	0x000000010e4de8aa WTF::Function<void (WebCore::ScriptExecutionContext&)>::CallableWrapper<WebCore::WorkerInspectorProxy::disconnectFromWorkerInspectorController()::$_6>::call(WebCore::ScriptExecutionContext&) + 26 (WorkerGlobalScope.h:81)
4   com.apple.WebCore             	0x000000010e4e0769 WebCore::WorkerRunLoop::runInMode(WebCore::WorkerGlobalScope*, WebCore::ModePredicate const&, WebCore::WorkerRunLoop::WaitMode) + 441 (Function.h:56)
5   com.apple.WebCore             	0x000000010e4e0550 WebCore::WorkerRunLoop::run(WebCore::WorkerGlobalScope*) + 96 (WorkerRunLoop.cpp:138)
6   com.apple.WebCore             	0x000000010e4e3361 WebCore::WorkerThread::workerThread() + 1025 (RefPtr.h:59)
7   com.apple.JavaScriptCore      	0x0000000110f46184 WTF::Thread::entryPoint(WTF::Thread::NewThreadContext*) + 212 (memory:2735)
8   com.apple.JavaScriptCore      	0x0000000110f47d09 WTF::wtfThreadEntryPoint(void*) + 9 (ThreadingPthreads.cpp:228)
9   libsystem_pthread.dylib       	0x00007fffb3ad293b _pthread_body + 180
10  libsystem_pthread.dylib       	0x00007fffb3ad2887 _pthread_start + 286
11  libsystem_pthread.dylib       	0x00007fffb3ad208d thread_start + 13

[Chris Dumez]

+Joe who added this code.

[Chris Dumez]

(In reply to Chris Dumez from comment #5)
> +Joe who added this code.

We could callOnMainThread() before using the WebLoaderStrategy. However, the WebInspector code relies on a s_frontendCounter counter which is a static int and is now incremented / decremented from various threads.

[youenn fablet]

Cannot we do all these incrementing/decrementing in the main thread?
Maybe we need callOnMainThreadAndWait.

[Chris Dumez]

Created attachment 344875 [details]
Patch

[Joseph Pecoraro]

Comment on attachment 344875 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=344875&action=review

> Source/WebCore/inspector/InspectorInstrumentation.h:1466
> +    if (!isMainThread()) {
> +        callOnMainThread([] {
> +            InspectorInstrumentation::frontendCreated();
> +        });
> +        return;
> +    }

The s_frontendCounter is consulted for the fast return on most InspectorInstrumentation functions. If this is async, we might end up ignoring some InspectorInstrumentation hooks for a short period of time that we ordinarily would not have ignored. I don't know if I have a concrete example of what would go wrong though. I imagine the only case that could be affected by this is a Service Worker inspector (since Web Inspectors's Dedicated Workers would have had this already happen on the main thread already). I think its reasonable to give this a shot.

You could avoid the isMainThreadCheck if you put the callOnMainThread inside of WorkerInspectorController. InspectorController's should always be on the main thread.

[Alex Christensen]

callOnMainThreadAndWait?

[Chris Dumez]

Created attachment 344880 [details]
Patch

[EWS Watchlist]

Comment on attachment 344880 [details]
Patch

Attachment 344880 [details] did not pass win-ews (win):
Output: https://webkit-queues.webkit.org/results/8519486

New failing tests:
http/tests/security/canvas-remote-read-remote-video-blocked-no-crossorigin.html

[EWS Watchlist]

Created attachment 344890 [details]
Archive of layout-test-results from ews201 for win-future

The attached test failures were seen while running run-webkit-tests on the win-ews.
Bot: ews201  Port: win-future  Platform: CYGWIN_NT-6.1-2.9.0-0.318-5-3-x86_64-64bit

[WebKit Commit Bot]

Comment on attachment 344880 [details]
Patch

Clearing flags on attachment: 344880

Committed r233784: <https://trac.webkit.org/changeset/233784>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/42140266>