Download the PHP package aidantwoods/secureheaders without Composer

SecureHeaders

A PHP class aiming to make the use of browser security features more accessible.

For full documentation, please see the Wiki.

A demonstration with a sample configuration is also available.

What is a 'secure header'?

Secure headers, are a set of headers that configure browser security features. All of these headers can be used in any web application, and most can be deployed without any, or very minor code changes. However some of the most effective ones do require code changes – especially to implement well.

Features

• Add/remove and manage headers easily
• Build a Content Security Policy, or combine multiple together
• Content Security Policy analysis
• Easy integeration with arbitrary frameworks (take a look at the HttpAdapter)
• Protect incorrectly set cookies
• Strict mode
• Safe mode prevents accidental long-term self-DOS when using HSTS, or HPKP
• Receive warnings about missing, or misconfigured security headers

Methodology and Philosophy

Error messages are often a great way for a program to tell the programmer that something is wrong. Whether it's calling a variable that's not yet been assigned, or causing a fatal error by exhausting the memory allocation limit.

Both of these situations can usually be rectified very quickly by the programmer. The effort required to do so is greatly reduced because the program communicated exactly what the problem was, as soon as the programmer introduced the bug. SecureHeaders aims to apply this concept to browser security features.

Utilising the error reporting level set within PHP configuration, SecureHeaders will generate E_USER_WARNING and E_USER_NOTICE level error messages to inform the programmer about either misconfigurations or lack of configuration.

In addition to error reporting, SecureHeaders will make some safe proactive changes to certain headers, or even add new ones if they're missing.

Installation

Via Composer

Other

Download SecureHeaders.phar, then

Sounds good, but let's see some of the code...

Here is a good implementation example

These few lines of code will take an application from a grade F, to a grade A on Scott Helme's https://securityheaders.io/

Woah, that was easy! Tell me what it did...

Let's break down the example above.

'Out of the box', SecureHeaders will already do quite a lot (by running the following code)

Automatic Headers and Errors

With such code, the following will occur:

• Warnings will be issued (E_USER_WARNING)

  Warning: Missing security header: 'Strict-Transport-Security'

  Warning: Missing security header: 'Content-Security-Policy'

• The following headers will be automatically added

• The following header will also be removed (SecureHeaders will also attempt to remove the Server header, though it is unlikely this header will be under PHP jurisdiction)

Cookies

Additionally, if any cookies have been set (at any time before ->apply() is called) e.g.

Even though in the current PHP configuration, cookie flags Secure and HTTPOnly do not default to on, and despite the fact that PHP does not support the SameSite cookie attribute, the end result of the Set-Cookie header will be

These flags were inserted by SecureHeaders because the cookie name contained the substring auth. Of course if that was a bad assumption, you can correct SecureHeaders' behaviour, or conversely you can tell SecureHeaders about some of your cookies that have less obvious names – but may need protecting in case of accidental missing flags.

If you enable ->strictMode() then the SameSite setting will be set to strict (you can also upgrade this without using strict mode).

Strict Mode

Strict mode will enable settings that you should be using. It is highly advisable to adjust your application to work with strict mode enabled.

When enabled, strict mode will:

• Auto-enable HSTS with a 1 year duration, and the includeSubDomains and preload flags set. Note that this HSTS policy is made as a header proposal, and can thus be removed or modified.

• The source keyword 'strict-dynamic' will also be added to the first of the following directives that exist: script-src, default-src; only if that directive also contains a nonce or hash source value, and not otherwise.

  This will disable the source whitelist in script-src in CSP3 compliant browsers. The use of whitelists in script-src is [considered not to be an ideal practice][1], because they are often trivial to bypass.

  [1]: https://research.google.com/pubs/pub45542.html "The Insecurity of Whitelists and the Future of Content Security Policy"

  Don't forget to manually submit your domain to the HSTS preload list if you are using this option.

• The default SameSite value injected into ->protectedCookie will be changed from SameSite=Lax to SameSite=Strict. See documentation on ->auto to enable/disable injection of SameSite and documentation on ->sameSiteCookies for more on specific behaviour and to explicitly define this value manually, to override the default.

• Auto-enable Expect-CT with a 1 year duration, and the enforce flag set. Note that this Expect-CT policy is made as a header proposal, and can thus be removed or modified.

Back to the example

Let's take a look at those other three lines, the first of which was

This enabled HSTS (Strict-Transport-Security) on the application for a duration of 1 year.

That sounds like something that might break things – I wouldn't want to accidentally enable that.

Safe Mode

Okay, SecureHeaders has got you covered – use $headers->safeMode(); to prevent headers being sent that will cause lasting effects.

So for example, if the following code was run (safe mode can be called at any point before ->apply() to be effective)

HSTS would still be enabled (as asked), but would be limited to lasting 24 hours.

SecureHeaders would also generate the following notice

Notice: HSTS settings were overridden because Safe-Mode is enabled. Read about some common mistakes when setting HSTS via copy/paste, and ensure you understand the details and possible side effects of this security feature before using it.

What if I set it via a method not related to SecureHeaders? Can SecureHeaders still enforce safe mode?

Yup! SecureHeaders will look at the names and values of headers independently of its own built in functions that can be used to generate them.

For example, if I use PHPs built in header function to set HSTS for 1 year, for all subdomains, and indicate consent to preload that rule into major browsers, and then (before or after setting that header) enable safe-mode...

The same above notice will be generated, max-age will be modified to 1 day, and the preload and includesubdomains flags will be removed.

Content Security Policy

The final two lines to cover from the initial example are as follows

These tell SecureHeaders that it should build a CSP (Content Security Policy) that allows default assets to be loaded from the current domain (self), and that scripts should be allowed from https://my.cdn.org.

Note that if we had said http://my.cdn.org instead, then the following would have been generated

Warning: Content Security Policy contains the insecure protocol HTTP in a source value http://my.cdn.org; this can allow anyone to insert elements covered by the script-src directive into the page.

Similarly, if wildcards such as 'unsafe-inline', https:, or * are included – SecureHeaders will generate warnings to highlight these CSP bypasses.

Note that the ->csp function is very diverse in what it will accept, to see some more on that take a look at Using CSP

Sending the headers

In order to apply anything added through SecureHeaders, you'll need to call ->apply(). By design, SecureHeaders doesn't have a construct function – so everything up until ->apply() is called is just configuration. However, if you don't want to have to remember to call this function, you can call ->applyOnOutput() instead, at any time. This will utilise PHP's ob_start() function to start output buffering. This lets SecureHeaders attatch itself to the first instance of any piece of code that generates output – and prior to actually sending that output to the user, make sure all headers are sent, by calling ->apply() for you.

Because SecureHeaders doesn't have a construct function, you can easily implement your own, via a simple class extension, e.g.

The above would implement the example discussed above, and would automatically apply to any page that ran just one line of code

Of course, pages could add additional configuration too, and headers would only be applied when the page started generating output.

Another Example

If the following CSP is created (note this probably isn't the best way to define a CSP of this size, see the array syntax that is available in the section on Using CSP)

The following messages will be issued with regard to CSP: (level E_USER_WARNING and level E_USER_NOTICE)

• The default-src directive contains a wildcard (so is a CSP bypass)

  Warning: Content Security Policy contains a wildcard * as a source value in default-src; this can allow anyone to insert elements covered by the default-src directive into the page.

• The script-src directive contains an a flag that allows inline script (so is a CSP bypass)

  Warning: Content Security Policy contains the 'unsafe-inline' keyword in script-src, which prevents CSP protecting against the injection of arbitrary code into the page.

• The script-src directive contains an insecure resource as a source value (HTTP responses can be trivially spoofed – spoofing allows a bypass)

  Warning: Content Security Policy contains the insecure protocol HTTP in a source value http://insecure.cdn.org; this can allow anyone to insert elements covered by the script-src directive into the page.

• The style-src directive contains two wildcards (so is a CSP bypass) – both wildcards are listed

  Warning: Content Security Policy contains the following wildcards https:, * as a source value in style-src; this can allow anyone to insert elements covered by the style-src directive into the page.

• The report only header was sent, but no/an invalid reporting address was given – preventing the report only header from doing anything useful in the wild

  Notice: Content Security Policy Report Only header was sent, but an invalid, or no reporting address was given. This header will not enforce violations, and with no reporting address specified, the browser can only report them locally in its console. Consider adding a reporting address to make full use of this header.

Using CSP

If you're new to Content-Security-Policy then running your proposed policy through Google's CSP Evaluator may be a good idea.

Let's take a look at a few ways of declaring the following CSP (or parts of it). Newlines and indentation added here for readability

CSP as an array

In the above, we've specified the policy using an array in the way it makes the most sense (bar some slight variation to demonstrate supported syntax). We then passed our policy array to the csp function.

Within the array, take a look at default-src. This is the full directive name (the key of the array), and its source list is specified as an array containing source values. In this case, the directive only has one source value, 'self', which is spelled out in full (note the single quotes within the string).

In this case, we've actually written a lot more than necessary – see the directive base for comparison. The actual CSP directive here is base-uri, but base is a supported shorthand by SecureHeaders. Secondly, we've omitted the array syntax from the descending source list entirely – we only wanted to declare one valid source, so SecureHeaders supports foregoing the array structure if its not useful. Additionally, we've made use of a shorthand within the source value too – omitting the single quotes from the string's value (i.e. self is a shorthand for 'self').

There are two CSP 'flags' included also in this policy, namely upgrade-insecure-requests and block-all-mixed-content. These do not hold any source values (and would not be valid in CSP if they did). You can specify these by either giving a source value of null (either as above, or an array containing only null as a source), or forgoing any mention of decedents entirely (as shown in block-all-mixed-content, which is written as-is). Once a flag has been set, no sources may be added. Similarly once a directive has been set, it may not become a flag. (This to prevent accidental loss of the entire source list).

The csp function also supports combining these CSP arrays, so the following would combine the csp defined in $myCSP, and $myOtherCSP. You can combine as many csp arrays as you like by adding additional arguments.

CSP as ordered pairs

Using the same csp function as above, you can add sources to directives as follows

or if you prefer to do this all in one line

Note that directives and sources are specified as ordered pairs here.

If you wanted to add a CSP flag in this way, simply use one of the following.

Note that the second way is necessary if embedded in a list of ordered pairs – otherwise SecureHeaders can't tell what is a directive name or a source value. e.g. this would set block-all-mixed-content as a CSP flag, and https://my.cdn.org as a script-src source value.

However, the csp function also supports mixing these ordered pairs with the array structure, and a string without a source at the end of the argument list will also be treated as a flag. You could, in perhaps an abuse of notation, use the following to set two CSP flags and the policy contained in the $csp array structure.

CSP as, uhh..

The CSP function aims to be as tolerant as possible, a CSP should be able to be communicated in whatever way is easiest to you.

That said, please use responsibly – the following is quite hard to read

Behaviour when a CSP header has already been set

The above code will perform a merge the set CSP header, and the additional script-src value set in the final line. Producing the following merged CSP header

Content-Security-Policy-Report-Only

All of the above is applicable to report only policies in exactly the same way. To tell SecureHeaders that you're creating a report only policy, simply use ->cspro in place of ->csp.

As an alternate method, you can also include the boolean true, or a non zero integer (loosely compares to true) in the regular ->csp function's argument list. The boolean false or the integer zero will signify enforced CSP (already the default). The left-most of these booleans or intgers will be taken as the mode. So to force enforced CSP (in-case you are unsure of the eventual variable types in the CSP argument list), use ->csp(false, arg1[, arg2[, ...]]) etc... or use zero in place of false. Similarly, to force report-only (in-case you are unsure of the eventual variable types in the CSP argument list) you can use either ->cspro(arg1[, arg2[, ...]]) or ->csp(true, arg1[, arg2[, ...]]).

Note that while ->csp supports having its mode changed to report-only, ->cspro does not (since is an alias for ->csp with report-only forced on). ->csp and ->cspro are identical in their interpretation of the various structures a Content-Security-Policy can be communicated in.

More on Usage

For full documentation, please see the Wiki

Versioning

The SecureHeaders project will follow Semantic Versioning 2, with the following declared public API:

Any method baring the @api phpdoc tag.

Roughtly speaking

• Every public method in Aidantwoods\SecureHeaders\SecureHeaders (except Aidantwoods\SecureHeaders\SecureHeaders::returnBuffer)
• Every public method in Aidantwoods\SecureHeaders\Http
• Every public method in Aidantwoods\SecureHeaders\HeaderBag

This allows the main SecureHeaders class to be used as expected by semver, and also the HttpAdapter interface/implementation (for integration with anything) to be used as expected by semver.

All other methods and properties are therefore non-public for the purposes of semver. That means that, e.g. methods with public visibility that are not in the above scope are subject to change in a backwards incompatible way, without a major version bump.

ChangeLog

The SecureHeaders project will follow Keep a CHANGELOG principles

Check out the ChangeLogs/ folder, to see these.