Bug 108861 – WebProcess crashes handling repeated NetworkProcess crashes

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKit2

[Brady Eidson]

WebProcess crashes handling repeated NetworkProcess crashes

1 - The NetworkProcess crashed once while lots of loads were in progress.
2 - The WebProcess saw this, and started failing out all of the resource loaders, which involved messaging back to the new NetworkProcess in a tight loop.
3 - The new NetworkProcess crashed (ASSERTed) trying to remove a loader that didn't exist.  This was fixed in http://trac.webkit.org/changeset/140852 but some other symptom could arise causing similar behavior.
4 - The WebProcess was still trying to fail out these resource loaders in a tight loop, and in the middle of this loop the connection once again became invalid from the crash in step 3.
5 - WebProcess crashed as a result.

In radar as <rdar://problem/13049867>

[Brady Eidson]

Created attachment 186464 [details]
Patch v1

[Alexey Proskuryakov]

Comment on attachment 186464 [details]
Patch v1

View in context: https://bugs.webkit.org/attachment.cgi?id=186464&action=review

> Source/WebKit2/WebProcess/Network/WebResourceLoadScheduler.cpp:135
> +    ASSERT(m_interallyFailedResourceLoaders.isEmpty());

This does not look certain to me. Can't didFail() calls execute JavaScript, which will start new loaders that will fail?

> Source/WebKit2/WebProcess/Network/WebResourceLoadScheduler.h:71
> +    HashSet<RefPtr<WebCore::ResourceLoader> > m_interallyFailedResourceLoaders;
> +    WebCore::RunLoop::Timer<WebResourceLoadScheduler> m_interallyFailedLoadTimer;

Typos: should be "internally".

[Brady Eidson]

(In reply to comment #2)
> (From update of attachment 186464 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=186464&action=review
> 
> > Source/WebKit2/WebProcess/Network/WebResourceLoadScheduler.cpp:135
> > +    ASSERT(m_interallyFailedResourceLoaders.isEmpty());
> 
> This does not look certain to me. Can't didFail() calls execute JavaScript, which will start new loaders that will fail?

That's a very good point.  I'll remove the ASSERT.

[Brady Eidson]

http://trac.webkit.org/changeset/141824