Bug 32111 – Geolocation bug causes crash when watch is cleared from some callbacks

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Steve Block]

In Geolocation::GeoNotifier::timerFired, a JS callback may be invoked if a fatal error was registered on this GeoNotifier or if this request has timed out. If the request is a watch, and is cleared by a call to clearWatch in the callback, the GeoNotifier object will be destroyed. This then causes a crash when the m_geolocation member is accessed to call Geolocation::fatalErrorOccurred or Geolocation::requestTimedOut.

Note that use of the GeoNotifier this pointer as a parameter for the Geolocation methods is safe, even if the GeoNotifier object has been destroyed, as the this pointer is used only as a look-up in a map.

[Steve Block]

Created attachment 44241 [details]
Patch 1 for Bug 32111

[WebKit Review Bot]

style-queue ran check-webkit-style on attachment 44241 [details] without any errors.

[Darin Adler]

Comment on attachment 44241 [details]
Patch 1 for Bug 32111

> +    // Cache our pointer to the Geolocation object, as this GeoNotifier object
> +    // could be deleted by a call to clearWatch in a callback.
> +    Geolocation* geolocation = m_geolocation;

Since the Geolocation object is reference counted, shouldn't this be a RefPtr? If not, then why is the Geolocation object reference counted? In other words, what guarantees it will not be destroyed?

[Steve Block]

> Since the Geolocation object is reference counted, shouldn't this be a RefPtr?
> If not, then why is the Geolocation object reference counted? In other words,
> what guarantees it will not be destroyed?
The GeoNotifier objects are owned by the Geolocation object and their lifetime is controlled with RefPtrs. Each GeoNotifier has a raw pointer back to the Geolocation object which it uses to make callbacks. Since the GeoNotifier objects are owned by the Geolocation object, the Geolocation object is guaranteed to have a longer lifespan, so the raw Geolocation pointers held by the GeoNotifier objects will always be valid.

[WebKit Commit Bot]

Comment on attachment 44241 [details]
Patch 1 for Bug 32111

Clearing flags on attachment: 44241

Committed r51692: <http://trac.webkit.org/changeset/51692>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.