GitLab

make SiteIsolationPolicy public.

The problem was that --site-per-process disabled extension
process sharing, but the site-per-process base::Feature (which
we've been field trialing) did not. This was due to the
extensions code checking only for the flag, and not considering
the field trial state as well.

components/printing actually got the logic right, but only by
reproducing a lot of business logic. Thus, it seems
appropriate to move SiteIsolationPolicy to content/public,
so that we can centralize the "what kind of oopifs are there"
logic. For printing, this change adds a new getter function
specific to oopif compositor, since that's basically a
derived policy of the process model.

For extensions, we've decided to disable LockToOrigin in
--site-per-process (rather than to enable it in the feature),
since origin-locking extensions doesn't help with the spectre
threat, and --site-per-process is about spectre these days.
[Charlie suggests we develop some kind of "extension isolation v2"
proposal, maybe reviving the --isolate-extension flag for that
purpose!]

Bug: 824966, 766267

Change-Id: Ibf7592c9d522fd0c99057358bcc34b5881780db8
Reviewed-on: https://chromium-review.googlesource.com/949966
Commit-Queue: Nick Carter <nick@chromium.org>
Reviewed-by: Devlin <rdevlin.cronin@chromium.org>
Reviewed-by: Wei Li <weili@chromium.org>
Reviewed-by: Lei Zhang <thestig@chromium.org>
Reviewed-by: Charlie Reis <creis@chromium.org>
Reviewed-by: Łukasz Anforowicz <lukasza@chromium.org>
Cr-Commit-Position: refs/heads/master@{#548645}