See https://github.com/twitter/secureheaders/pull/215

This should allow use of Rails 5 without deprecations from before_filter. before_filter is deprecated and will be removed in Rails 5.1.

This change is not applicable for the 3.x series.

secure_headers 3.0.0 is a near-complete, not-entirely-backward-compatible rewrite. Please see the upgrade guide for an in-depth explanation of the changes and the suggested upgrade path.

secure_headers 3.0 is a near-complete rewrite that is not entirely backwards compatible. See the upgrade guide for migrating between 2.x and 3.x: https://github.com/twitter/secureheaders/blob/master/upgrading-to-3-0.md

See https://github.com/twitter/secureheaders/issues/203 and https://github.com/twitter/secureheaders/commit/cfad0e52285353b88e46fe384e7cd60bf2a01735

Upon upgrading to secure_headers 2.5.0, I get a flood of these deprecations when running my tests:
[DEPRECATION] secure_header_options_for will not be supported in secure_headers

/cc @bquorning

This release contains deprecation warnings for those wishing to upgrade to the 3.x series. With this release, fixing all deprecation warnings will make your configuration compatible when you decide to upgrade to the soon-to-be-released 3.x series (currently in pre-release stage).

No changes to functionality should be observed unless you were using procs as CSP config values.

If you use the header_hash method for setting your headers in middleware and you opted out of a header (via setting the value to false), you would run into an exception as described in https://github.com/twitter/secureheaders/pull/193

     NoMethodError:
       undefined method `name' for nil:NilClass
     # ./lib/secure_headers.rb:63:in `block in header_hash'
     # ./lib/secure_headers.rb:54:in `each'
     # ./lib/secure_headers.rb:54:in `inject'
     # ./lib/secure_headers.rb:54:in `header_hash'

@igrep reported an anti-patter in use regarding UserAgentParser. This caused UserAgentParser to reload it's entire configuration set twice* per request. Moving this to a cached constant prevents the constant reinstantiation and will improve performance.

https://github.com/twitter/secureheaders/issues/187

A nasty regression meant that many CSP configuration values were "reset" after the first request, one of these being the "enforce" flag. See https://github.com/twitter/secureheaders/pull/184 for the full list of fields that were affected. Thanks to @spdawson for reporting this https://github.com/twitter/secureheaders/issues/183

This release may change the output of headers based on per browser support. Unsupported directives will be omitted based on the user agent per request. See https://github.com/twitter/secureheaders/pull/179

p.s. this will likely be the last non-bugfix release for the 2.x line. 3.x will be a major change. Sneak preview: https://github.com/twitter/secureheaders/pull/181

If you leveraged secure_headers automatic filling of empty directives, the header value will change but it should not affect how the browser applies the policy. The content of CSP reports may change if you do not update your policy.

before

  config.csp = {
    :default_src => "'self'"
  }

would produce default-src 'self'; connect-src 'self'; frame-src 'self' ... etc.

after

  config.csp = {
    :default_src => "'self'"
  }

will produce default-src 'self'

The reason for this is that a default-src violation was basically impossible to handle. Chrome sends an effective-directive which helps indicate what kind of violation occurred even if it fell back to default-src. This is part of the CSP Level 2 spec so hopefully other browsers will implement this soon.

Workaround

Just set the values yourself, but really a default-src of anything other than 'none' implies the policy can be tightened dramatically. "ZOMG don't you work for github and doesn't github send a default-src of *???" Yes, this is true. I disagree with this but at the same time, github defines every single known directive that a browser supports so default-src will only apply if a new directive is introduced, and we'd rather fail open. For now.

  config.csp = {
    :default_src => "'self'",
    :connect_src => "'self'",
    :frame_src => "'self'"
    ... etc.
  }

Besides, relying on default-src is often not what you want and encourages an overly permissive policy. I've seen it. Seriously. default-src 'unsafe-inline' 'unsafe-eval' https: http:; That's terrible.

See https://github.com/twitter/secureheaders/issues/167 and https://github.com/twitter/secureheaders/pull/168

tl;dr is that there is a class method SecureHeaders::header_hash that will return a hash of header name => value pairs useful for merging with the rack header hash in middleware.

As discussed in https://github.com/twitter/secureheaders/issues/154

See https://github.com/twitter/secureheaders/pull/161

See https://github.com/twitter/secureheaders/pull/147

Allows you to override a controller method that returns a config in the context of the executing action.

See https://github.com/twitter/secureheaders/pull/150

Safari will generate a warning that it doesn't support nonces. Safari will fall back to the unsafe-inline. Things will still work, but an ugly message is printed to the console.

This opts out safari and IE users from the inline script protection. I haven't verified any IE behavior yet, so I'm just assuming it doesn't work.

https://github.com/twitter/secureheaders/pull/148

Facilitates better per-request config:

:enforce => lambda { |controller| controller.current_user.beta_testing? }

NOTE if you used lambda config values, this will raise an exception until you add the controller reference:

bad:

lambda { true }

good:

lambda { |controller| true }
proc { true }
proc { |controller| true }

Includes https://github.com/twitter/secureheaders/pull/143 (which is really just https://github.com/twitter/secureheaders/pull/132) from @thirstscolr

Just a small change that adds a constant that was missing as reported in https://github.com/twitter/secureheaders/issues/141

Fixes an issue where view helpers (for nonces, hashes, etc) weren't available in views.

This release contains support for more csp level 2 features such as the new directives, the script hash integration, and more.

It also sets a new header by default: X-Permitted-Cross-Domain-Policies

Support for hpkp is not included in this release as the implementations are still very unstable.

🚀