297
votes

I'm getting a bunch of errors in the developer console:

Refused to evaluate a string

Refused to execute inline script because it violates the following Content Security Policy directive

Refused to load the script

Refused to load the stylesheet

What's this all about? How does Content Security Policy (CSP) work? How do I use the Content-Security-Policy HTTP header?

Specifically, how to...

  1. ...allow multiple sources?
  2. ...use different directives?
  3. ...use multiple directives?
  4. ...handle ports?
  5. ...handle different protocols?
  6. ...allow file:// protocol?
  7. ...use inline styles, scripts, and tags <style> and <script>?
  8. ...allow eval()?

And finally:

  1. What exactly does 'self' mean?
3

3 Answers

671
votes

The Content-Security-Policy meta-tag allows you to reduce the risk of XSS attacks by allowing you to define where resources can be loaded from, preventing browsers from loading data from any other locations. This makes it harder for an attacker to inject malicious code into your site.

I banged my head against a brick wall trying to figure out why I was getting CSP errors one after another, and there didn't seem to be any concise, clear instructions on just how does it work. So here's my attempt at explaining some points of CSP briefly, mostly concentrating on the things I found hard to solve.

For brevity I won’t write the full tag in each sample. Instead I'll only show the content property, so a sample that says content="default-src 'self'" means this:

<meta http-equiv="Content-Security-Policy" content="default-src 'self'">

1. How can I allow multiple sources?

You can simply list your sources after a directive as a space-separated list:

content="default-src 'self' https://example.com/js/"

Note that there are no quotes around parameters other than the special ones, like 'self'. Also, there's no colon (:) after the directive. Just the directive, then a space-separated list of parameters.

Everything below the specified parameters is implicitly allowed. That means that in the example above these would be valid sources:

https://example.com/js/file.js
https://example.com/js/subdir/anotherfile.js

These, however, would not be valid:

http://example.com/js/file.js
^^^^ wrong protocol

https://example.com/file.js
                   ^^ above the specified path

2. How can I use different directives? What do they each do?

The most common directives are:

  • default-src the default policy for loading javascript, images, CSS, fonts, AJAX requests, etc
  • script-src defines valid sources for javascript files
  • style-src defines valid sources for css files
  • img-src defines valid sources for images
  • connect-src defines valid targets for to XMLHttpRequest (AJAX), WebSockets or EventSource. If a connection attempt is made to a host that's not allowed here, the browser will emulate a 400 error

There are others, but these are the ones you're most likely to need.

3. How can I use multiple directives?

You define all your directives inside one meta-tag by terminating them with a semicolon (;):

content="default-src 'self' https://example.com/js/; style-src 'self'"

4. How can I handle ports?

Everything but the default ports needs to be allowed explicitly by adding the port number or an asterisk after the allowed domain:

content="default-src 'self' https://ajax.googleapis.com http://example.com:123/free/stuff/"

The above would result in:

https://ajax.googleapis.com:123
                           ^^^^ Not ok, wrong port

https://ajax.googleapis.com - OK

http://example.com/free/stuff/file.js
                 ^^ Not ok, only the port 123 is allowed

http://example.com:123/free/stuff/file.js - OK

As I mentioned, you can also use an asterisk to explicitly allow all ports:

content="default-src example.com:*"

5. How can I handle different protocols?

By default, only standard protocols are allowed. For example to allow WebSockets ws:// you will have to allow it explicitly:

content="default-src 'self'; connect-src ws:; style-src 'self'"
                                         ^^^ web Sockets are now allowed on all domains and ports.

6. How can I allow the file protocol file://?

If you'll try to define it as such it won’t work. Instead, you'll allow it with the filesystem parameter:

content="default-src filesystem"

7. How can I use inline scripts and style definitions?

Unless explicitly allowed, you can't use inline style definitions, code inside <script> tags or in tag properties like onclick. You allow them like so:

content="script-src 'unsafe-inline'; style-src 'unsafe-inline'"

You'll also have to explicitly allow inline, base64 encoded images:

content="img-src data:"

8. How can I allow eval()?

I'm sure many people would say that you don't, since 'eval is evil' and the most likely cause for the impending end of the world. Those people would be wrong. Sure, you can definitely punch major holes into your site's security with eval, but it has perfectly valid use cases. You just have to be smart about using it. You allow it like so:

content="script-src 'unsafe-eval'"

9. What exactly does 'self' mean?

You might take 'self' to mean localhost, local filesystem, or anything on the same host. It doesn't mean any of those. It means sources that have the same scheme (protocol), same host, and same port as the file the content policy is defined in. Serving your site over HTTP? No https for you then, unless you define it explicitly.

I've used 'self' in most examples as it usually makes sense to include it, but it's by no means mandatory. Leave it out if you don't need it.

But hang on a minute! Can't I just use content="default-src *" and be done with it?

No. In addition to the obvious security vulnerabilities, this also won’t work as you'd expect. Even though some docs claim it allows anything, that's not true. It doesn't allow inlining or evals, so to really, really make your site extra vulnerable, you would use this:

content="default-src * 'unsafe-inline' 'unsafe-eval'"

... but I trust you won’t.

Further reading:

http://content-security-policy.com

http://en.wikipedia.org/wiki/Content_Security_Policy

18
votes

Apache 2 mod_headers

You could also enable Apache 2 mod_headers. On Fedora it's already enabled by default. If you use Ubuntu/Debian, enable it like this:

# First enable headers module for Apache 2,
# and then restart the Apache2 service
a2enmod headers
apache2 -k graceful

On Ubuntu/Debian you can configure headers in the file /etc/apache2/conf-enabled/security.conf

#
# Setting this header will prevent MSIE from interpreting files as something
# else than declared by the content type in the HTTP headers.
# Requires mod_headers to be enabled.
#
#Header set X-Content-Type-Options: "nosniff"

#
# Setting this header will prevent other sites from embedding pages from this
# site as frames. This defends against clickjacking attacks.
# Requires mod_headers to be enabled.
#
Header always set X-Frame-Options: "sameorigin"
Header always set X-Content-Type-Options nosniff
Header always set X-XSS-Protection "1; mode=block"
Header always set X-Permitted-Cross-Domain-Policies "master-only"
Header always set Cache-Control "no-cache, no-store, must-revalidate"
Header always set Pragma "no-cache"
Header always set Expires "-1"
Header always set Content-Security-Policy: "default-src 'none';"
Header always set Content-Security-Policy: "script-src 'self' www.google-analytics.com adserver.example.com www.example.com;"
Header always set Content-Security-Policy: "style-src 'self' www.example.com;"

Note: This is the bottom part of the file. Only the last three entries are CSP settings.

The first parameter is the directive, the second is the sources to be white-listed. I've added Google analytics and an adserver, which you might have. Furthermore, I found that if you have aliases, e.g, www.example.com and example.com configured in Apache 2 you should add them to the white-list as well.

Inline code is considered harmful, and you should avoid it. Copy all the JavaScript code and CSS to separate files and add them to the white-list.

While you're at it you could take a look at the other header settings and install mod_security

Further reading:

https://developers.google.com/web/fundamentals/security/csp/

https://www.w3.org/TR/CSP/

1
votes

The information from historical threads is outdated, so I want to refresh...

Content-Security-Policy Best Practices in 2021:

1. Always start out with a strict Report-Only policy to map-out your site

Most developers directly set the Content-Security-Policy http header, breaking your entire site. Start by using Content-Security-Policy-Report-Only which will simulate blocking, and send violation reports to your report-uri endpoint.

2. Never use the <head> tag to set the policy

Serving the CSP through an html meta header is considered legacy and has some drawbacks with multiple browser edge-cases. Setting the CSP via the HTTP headers of the request only.

More info on <head> bugs

3. Don't try to reverse-engineer 3rd party service dependencies - Instead use Crowdsourced CSP packages

The majority of CSP reports sent to your report-uri, are actually not a part of your code, but of services used by your site. This can be impossible to reverse-engineer yourself, so use Crowdsourced CSP packages, which are actively maintained.

Adding the directives needed by your 3rd party will:

  1. Clear out the noise, leaving you to decisions only on the actual custom code in your site.
  2. Save you a lot of money on useless requests, as most report-uri service providers charge "per request"...

4. Maintain versions of your CSP.

Without maintaining proper versions, it's really hard to control the CSP at scale, and understanding which CSP, created which report. Think maintaining code without proper version control...

5. Enforce gradually, with two separate policies !

It is a common best practice to run multiple policies:

  1. A more strict policy running in Report-Only mode (so that you get reports of suspicious violations, and act accordingly.
  2. A less strict policy that is enforced, and has less chances of breaking legitimate parts of your site.

6. Use a good CSP Generator

Use a strong Content Security Policy Generator like RapidSec, that supports the above requirements, and also:

  1. Has automatic deployment plugins. These will allow deploying the CSP from a UI, and passing on the copy-paste. Example Wordpress CSP plugin, Node.js CSP module
  2. Automatic detection of packages. So that you don't have to manually select them
  3. Threat Detection / Analytics. Understand what of the blocked content is noise, what is interesting, and what is critical.

Full disclosure: I am associated with RapidSec, which is why I know all this.

7. Level up to Nonces.

Once you're happy with your allowlist CSP, you can invest in an additional nonce based Content-Security-Policy, that should run as a separate header. Note that this will require code refactoring, and sometimes custom serving infrastructure (to generate the nonces), so it's suggested to start with nonces once you already have some decent running policies.

Google's security team has some great talks on nonce-based CSP.

Good luck on your CSP journey - don't despair.