Spec descendant frames' ongoing navigations block disableUntrustedNetwork promise from being resolved

[xiaochen-z]

fix: #205

This PR only implements the change in algorithm "recalculate the untrusted network status of all fenced frame descendants".

The part that this algorithm should be called whenever a navigation commits will be done in a follow-up PR. I'm still investigating how to spec this. See previous comment on this.

---

Preview | Diff

[blu25]

LGTM, thanks!

[xiaochen-z]

+@domfarolino for review, thank you.

[domfarolino]

Pretty much lgtm. This covers traversals as well btw, which I think is the goal too? Does that match the implementation? And are there WPTs?

[domfarolino]

Suggested change

	1. If |networkCutoffReady| is false or |ongoingNavigation| is not null:
	1. If |ongoingNavigation| is not null:

Isn't this the same, since at this point we already know that |networkCuttoffReady| is false?

[xiaochen-z]

The above if branch does not early return. Also the line Otherwise: is removed. So at this point |networkCuttoffReady| can be true or false here.

Even if we add the early return above, the condition |networkCutoffReady| is false is still needed. As the suggested edit, IIUC when |networkCutoffReady| is false but |ongoingNavigation| is null, the branch will not be entered, however it should.

[xiaochen-z]

Thanks!

This covers traversals as well btw, which I think is the goal too? Does that match the implementation?

From what I read from the spec, when ongoing navigation is "traversal", it means the navigable is traversing its history. For example, FF going back to a previously navigated URL. The FF may end up with a different page and having its untrusted network status reset to enabled.

This is the race condition that the PR targets: When a parent frame tries to disable untrusted networks, it checks the child frame and finds it has untrusted network disabled. Without checking descendant frames' ongoing navigations, the parent frame goes ahead marking its untrusted network also disabled. But if the child frame has an ongoing navigation, including the history traversal, it resets the child frame's untrusted network to enabled as the frame navigates to a different page.

The implementation checks this via FrameTreeNode::HasNavigation(), which is set during normal navigation as well as history traversal. So the spec matches the implementation.

And are there WPTs?

No. I'm adding WPTs for this. Tracking in bug: crbug/397377177.