riched20: Avoid releasing a non-existent interface.

[Nikolay Sivov]

On 4/17/2014 11:59, Jactry Zeng wrote:
>
> 2014-04-17 11:54 GMT+08:00 Nikolay Sivov <bunglehead at gmail.com 
> <mailto:bunglehead at gmail.com>>:
> >
> > What you're doing is a violation of refcount handling. The rule is 
> to release what you got, without relying on internals like that. In 
> this case GetSelection() returns interface pointer and you're 
> responsible in exactly one Release() on it.
> >
> > Interesting thing to test would be to check if GetSelection() 
> returns new instance every time it's called. If this is a case it will 
> justify some code changes to support this, right now patch is wrong.
> >
> > If it actually returns same interface pointer you can't protect from 
> use-after-free because I can grab multiple references with several 
> GetSelection() calls, and when I'll try to release them it will be 
> freed already by a loop like that.
>
> Yes, you are right. GetSelection() will not return a new pointer every 
> time it's called, I have checked it by some tests on Windows.
Did you actually change selection range in between of these calls? It's 
possible that new instance will be returned after selection changed. 
This way you'll end up with two alive instances.

Another thing to test if ITextDocument responds to IID_ITextSelection query.
> I will try some other ways to fix this bug.
First thing to do is to figure out what relations are between these 
interfaces.
>
> Thanks again!
>
> -- 
> Regards,
> Jactry Zeng
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.winehq.org/pipermail/wine-devel/attachments/20140417/8eb11ef7/attachment.html>