Group renderer processes by privilige when we hit the max process count....

Group renderer processes by privilige when we hit the max process count.

BUG=12128
TEST=Create a bunch of tabs (40+) and make sure New Tab pages are always
grouped in a process with other chrome internal pages, extensions are
always grouped together, and regular web pages are never in a process with
extensions or New Tab pages.

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=18428

[brettw, 2009-06-11 20:57:15]

This is great. Do you think it's possible to write a unittest?

[Matt Perry, 2009-06-11 22:44:59]

On 2009/06/11 20:57:15, brettw wrote:
> This is great. Do you think it's possible to write a unittest?

Added a test.

[brettw, 2009-06-11 22:52:51]

Thanks.

My question for Charlie: Is it possible for a site to open a page with a chrome:
scheme and get a site instance with that in such a way that it tricks this code?

[Charlie Reis, 2009-06-12 17:30:27]

On 2009/06/11 22:52:51, brettw wrote:
> Thanks.
> 
> My question for Charlie: Is it possible for a site to open a page with a
chrome:
> scheme and get a site instance with that in such a way that it tricks this
code?

Sorry for the delay-- I've got family in town for graduation, so I'm a bit out
of touch at the moment.

I'm very excited about this changelist, though.  At a quick glance, I think it
looks good.  The thing to check is whether a new SiteInstance is properly
created if you click on a link to a normal page in either a DOM UI page or an
extension page.  (If you click on a cross-site link in a normal page, we do not
yet create a new SiteInstance for it.  If that's true for extensions, for
example, it may leak these privileges to a normal page.)

If a new SiteInstance is always created when you navigate away from a DOM UI
page or extensions page (even via renderer-initiated navigations like links and
script redirects), then I think you should be fine.

I can follow up more next week if you have more questions about it.  Hope that
helps!

[Matt Perry, 2009-06-12 18:45:01]

On 2009/06/12 17:30:27, creis wrote:
> On 2009/06/11 22:52:51, brettw wrote:
> > Thanks.
> > 
> > My question for Charlie: Is it possible for a site to open a page with a
> chrome:
> > scheme and get a site instance with that in such a way that it tricks this
> code?
> 
> Sorry for the delay-- I've got family in town for graduation, so I'm a bit out
> of touch at the moment.
> 
> I'm very excited about this changelist, though.  At a quick glance, I think it
> looks good.  The thing to check is whether a new SiteInstance is properly
> created if you click on a link to a normal page in either a DOM UI page or an
> extension page.  (If you click on a cross-site link in a normal page, we do
not
> yet create a new SiteInstance for it.  If that's true for extensions, for
> example, it may leak these privileges to a normal page.)

We don't create a new SiteInstance, or even a new RenderViewHost, when you click
a cross-site link in an extension.  That's a separate bug though.  This CL is
only concerned with what happens when we try to reuse existing processes after
we hit our limit.

Thanks for bringing the link clicking issue to my attention though.  NTP handles
that case fine.  We probably just need to do the same for extensions.