Add a default Content-Security-Policy and clickjacking defence to core

I guess the X-Frame-Options header that core adds via FinishResponseSubscriber should also be removed when this is done. The "frame-ancestors" directive in the CSP header is well supported and obsoletes X-Frame-Options.

With IE no longer being supported, core could replace its default X-Frame-Options: SAMEORIGIN with Content-Security-Policy: frame-ancestors 'self'. Starting to always add a default CSP header could cause problems for a site that sets X-Frame-Options: DENY though. (It looks like IE was the only browser to support ALLOW-FROM).

A possible transition is:
[11.1] Add a late-subscriber that translates x-frame-options to a CSP frame-ancestors if the response does not have a CSP header (assume if a CSP header is set but does not include frame-ancestors, then it was intended to allow any frame parent). Issue a deprecation warning if the response has a value other than X-Frame-Options: SAMEORIGIN to notify users to use CSP instead.
[12.0] Remove the late subscriber, and replace X-Frame-Options with a static CSP policy that includes frame-ancestors 'self'. A site that uses the Content-Security-Policy module, SecKit, or a custom listener to set the CSP header will override this default value.

My suggestion for a minimal static policy that core can add by default:

  Content-Security-Policy: script-src * 'unsafe-inline'; object-src 'none'; frame-ancestors 'self'

The simplest way to allow modifying the default CSP (or adding a CSP-Report-Only), would be to have a container parameter. Users wanting a more flexible implementation can use the CSP module.

A slightly rough MR:
- Introduce a new service parameter for setting static CSP values
- A late acting response subscriber will translate X-Frame-Options to Content-Secutity-Policy: frame-ancestors
- If a CSP policy is set (via service parameter, or another module like CSP or seckit), then that value is not changed.
- The X-Frame-Options header is always removed - either it's being replaced by core with an equivalent CSP policy, or we assume that the CSP policy set by the user has their desired frame-ancestors value (including the option of being omitted).
- If X-Frame-Options is not set to SAMEORIGIN, then a deprecation warning is issued to use CSP to set the value. (If the value is still the default SAMEORIGIN, then a future version of Drupal changing to frame-ancestors 'self' will not change browser behaviour).
- ALLOW-FROM is ignored by modern browsers (and equivalent to not sending the X-Frame-Options header), but is translated to a CSP value if used.

In a future version of Drupal:
- The service parameter can be set to a default enforced CSP value
- The late acting event subscriber can be completely removed
- Core can stop setting X-Frame-Options
- Modules (such as CSP or seckit) will still override core's default (now static) value.

Why the CSP policy value:
- script-src * 'unsafe-inline' will only block eval(). Not including 'unsafe-inline' would have the potential to break existing sites (and hopefully its presence leads people to explore making sure it's not required...)
- object-src 'none' is recommended to block legacy HTML elements
- frame-ancestors replaces X-Frame-Options
- I don't think there's other values that can safely be added as a default enforced policy directive without a reasonable risk of negatively affecting some sites. (maybe base-uri 'self'?)

The Needs Review Queue Bot → tested this issue. It fails the Drupal core commit checks. Therefore, this issue status is now "Needs work".

This does not mean that the patch necessarily needs to be re-rolled or the MR rebased. Read the Issue Summary, the issue tags and the latest discussion here to determine what needs to be done.

Consult the Drupal Contributor Guide → to find step-by-step guides for working with issues.

Update issue summary

Rebased and fixed the bot's previous issue.

In my previous comment I wasn't sure of the core dev timeline, but this looks like it's something that could be put in 10.4, with the compatibility later removed in 11.0.0.

Added some comments to MR.

Can be rebased so tests appear green please.
Also is it meant to be draft?

For the CR you can actually go ahead and create it just don't check published yet till the issue is merged.