Content Security Policy (CSP) Bypass
Last updated
Last updated
Content Security Policy (CSP) is recognized as a browser technology, primarily aimed at shielding against attacks such as cross-site scripting (XSS). It functions by defining and detailing paths and sources from which resources can be securely loaded by the browser. These resources encompass a range of elements such as images, frames, and JavaScript. For instance, a policy might permit the loading and execution of resources from the same domain (self), including inline resources and the execution of string code through functions like eval, setTimeout, or setInterval.
Implementation of CSP is conducted through response headers or by incorporating meta elements into the HTML page. Following this policy, browsers proactively enforce these stipulations and immediately block any detected violations.
Implemented via response header:
Implemented via meta tag:
xml
CSP can be enforced or monitored using these headers:
Content-Security-Policy: Enforces the CSP; the browser blocks any violations.
Content-Security-Policy-Report-Only: Used for monitoring; reports violations without blocking them. Ideal for testing in pre-production environments.
CSP restricts the origins for loading both active and passive content, controlling aspects like inline JavaScript execution and the use of eval(). An example policy is:
bash
script-src: Allows specific sources for JavaScript, including URLs, inline scripts, and scripts triggered by event handlers or XSLT stylesheets.
default-src: Sets a default policy for fetching resources when specific fetch directives are absent.
child-src: Specifies allowed resources for web workers and embedded frame contents.
connect-src: Restricts URLs which can be loaded using interfaces like fetch, WebSocket, XMLHttpRequest.
frame-src: Restricts URLs for frames.
frame-ancestors: Specifies which sources can embed the current page, applicable to elements like <frame>, <iframe>, <object>, <embed>, and <applet>.
img-src: Defines allowed sources for images.
font-src: Specifies valid sources for fonts loaded using @font-face.
manifest-src: Defines allowed sources of application manifest files.
media-src: Defines allowed sources for loading media objects.
object-src: Defines allowed sources for <object>, <embed>, and <applet> elements.
base-uri: Specifies allowed URLs for loading using <base> elements.
form-action: Lists valid endpoints for form submissions.
plugin-types: Restricts mime types that a page may invoke.
upgrade-insecure-requests: Instructs browsers to rewrite HTTP URLs to HTTPS.
sandbox: Applies restrictions similar to the sandbox attribute of an <iframe>.
report-to: Specifies a group to which a report will be sent if the policy is violated.
worker-src: Specifies valid sources for Worker, SharedWorker, or ServiceWorker scripts.
prefetch-src: Specifies valid sources for resources that will be fetched or prefetched.
navigate-to: Restricts the URLs to which a document can navigate by any means (a, form, window.location, window.open, etc.)
*: Allows all URLs except those with data:, blob:, filesystem: schemes.
'self': Allows loading from the same domain.
'data': Allows resources to be loaded via the data scheme (e.g., Base64 encoded images).
'none': Blocks loading from any source.
'unsafe-eval': Allows the use of eval() and similar methods, not recommended for security reasons.
'unsafe-hashes': Enables specific inline event handlers.
'unsafe-inline': Allows the use of inline resources like inline <script> or <style>, not recommended for security reasons.
'nonce': A whitelist for specific inline scripts using a cryptographic nonce (number used once).
If you have JS limited execution it's possible to get a used nonce inside the page with doc.defaultView.top.document.querySelector("[nonce]") and then reuse it to load a malicious script (if strict-dynamic is used, any allowed source can load new sources so this isn't needed), like in:
'sha256-<hash>': Whitelists scripts with a specific sha256 hash.
'strict-dynamic': Allows loading scripts from any source if it has been whitelisted by a nonce or hash.
'host': Specifies a specific host, like example.com.
https:: Restricts URLs to those that use HTTPS.
blob:: Allows resources to be loaded from Blob URLs (e.g., Blob URLs created via JavaScript).
filesystem:: Allows resources to be loaded from the filesystem.
'report-sample': Includes a sample of the violating code in the violation report (useful for debugging).
'strict-origin': Similar to 'self' but ensures the protocol security level of the sources matches the document (only secure origins can load resources from secure origins).
'strict-origin-when-cross-origin': Sends full URLs when making same-origin requests but only sends the origin when the request is cross-origin.
'unsafe-allow-redirects': Allows resources to be loaded that will immediately redirect to another resource. Not recommended as it weakens security.
yaml
Working payload: "/><script>alert(1);</script>
yaml
Working payload:
html
If you can somehow make an allowed JS code created a new script tag in the DOM with your JS code, because an allowed script is creating it, the new script tag will be allowed to be executed.
yaml
Working payload:
html
[!CAUTION] > It looks like this is not longer working
yaml
Working payloads:
html
yaml
If you can upload a JS file you can bypass this CSP:
Working payload:
html
However, it's highly probable that the server is validating the uploaded file and will only allow you to upload determined type of files.
Moreover, even if you could upload a JS code inside a file using an extension accepted by the server (like: script.png) this won't be enough because some servers like apache server select MIME type of the file based on the extension and browsers like Chrome will reject to execute Javascript code inside something that should be an image. "Hopefully", there are mistakes. For example, from a CTF I learnt that Apache doesn't know the .wave extension, therefore it doesn't serve it with a MIME type like audio/*.
For some of the following payload unsafe-eval is not even needed.
yaml
Load a vulnerable version of angular and execute arbitrary JS:
xml
The post shows that you could load all libraries from cdn.cloudflare.com (or any other allowed JS libraries repo), execute all added functions from each library, and check which functions from which libraries return the window object.
html
Angular XSS from a class name:
html
html
html
Abusing *.google.com/script.google.com
http
Scenarios like this where script-src is set to self and a particular domain which is whitelisted can be bypassed using JSONP. JSONP endpoints allow insecure callback methods which allow an attacker to perform XSS, working payload:
html
html
The same vulnerability will occur if the trusted endpoint contains an Open Redirect because if the initial endpoint is trusted, redirects are trusted.
www.facebook.com, *.facebook.com
Exfil
Hotjar
*.hotjar.com, ask.hotjar.io
Exfil
Jsdelivr
*.jsdelivr.com, cdn.jsdelivr.net
Exec
Amazon CloudFront
*.cloudfront.net
Exfil, Exec
Amazon AWS
*.amazonaws.com
Exfil, Exec
Azure Websites
*.azurewebsites.net, *.azurestaticapps.net
Exfil, Exec
Salesforce Heroku
*.herokuapp.com
Exfil, Exec
Google Firebase
*.firebaseapp.com
Exfil, Exec
If you find any of the allowed domains in the CSP of your target, chances are that you might be able to bypass the CSP by registering on the third-party service and, either exfiltrate data to that service or to execute code.
For example, if you find the following CSP:
or
Create a Facebook Developer account here.
Create a new "Facebook Login" app and select "Website".
Go to "Settings -> Basic" and get your "App ID"
In the target site you want to exfiltrate data from, you can exfiltrate data by directly using the Facebook SDK gadget "fbq" through a "customEvent" and the data payload.
Go to your App "Event Manager" and select the application you created (note the event manager could be found in an URL similar to this: https://www.facebook.com/events_manager2/list/pixel/[app-id]/test_events
Select the tab "Test Events" to see the events being sent out by "your" web site.
Then, on the victim side, you execute the following code to initialize the Facebook tracking pixel to point to the attacker's Facebook developer account app-id and issue a custom event like this:
JavaScript
In addition to the aforementioned redirection to bypass path restrictions, there is another technique called Relative Path Overwrite (RPO) that can be used on some servers.
For example, if CSP allows the path https://example.com/scripts/react/, it can be bypassed as follows:
html
The browser will ultimately load https://example.com/scripts/angular/angular.js.
This works because for the browser, you are loading a file named ..%2fangular%2fangular.js located under https://example.com/scripts/react/, which is compliant with CSP.
∑, they will decode it, effectively requesting https://example.com/scripts/react/../angular/angular.js, which is equivalent to https://example.com/scripts/angular/angular.js.
By exploiting this inconsistency in URL interpretation between the browser and the server, the path rules can be bypassed.
The solution is to not treat %2f as / on the server-side, ensuring consistent interpretation between the browser and the server to avoid this issue.
Moreover, if the page is loading a script using a relative path (like <script src="/js/app.js">) using a Nonce, you can abuse the base tag to make it load the script from your own server achieving a XSS.
If the vulnerable page is loaded with httpS, make use an httpS url in the base.
html
A specific policy known as Content Security Policy (CSP) may restrict JavaScript events. Nonetheless, AngularJS introduces custom events as an alternative. Within an event, AngularJS provides a unique object $event, referencing the native browser event object. This $event object can be exploited to circumvent the CSP. Notably, in Chrome, the $event/event object possesses a path attribute, holding an object array implicated in the event's execution chain, with the window object invariably positioned at the end. This structure is pivotal for sandbox escape tactics.
By directing this array to the orderBy filter, it's possible to iterate over it, harnessing the terminal element (the window object) to trigger a global function like alert(). The demonstrated code snippet below elucidates this process:
xml
This snippet highlights the usage of the ng-focus directive to trigger the event, employing $event.path|orderBy to manipulate the path array, and leveraging the window object to execute the alert() function, thereby revealing document.cookie.
Working payloads:
html
What happens when CSP encounters server-side redirection? If the redirection leads to a different origin that is not allowed, it will still fail.
Here's an example:
html
If CSP is set to https://www.google.com/a/b/c/d, since the path is considered, both /test and /a/test scripts will be blocked by CSP.
However, the final http://localhost:5555/301 will be redirected on the server-side to https://www.google.com/complete/search?client=chrome&q=123&jsonp=alert(1)//. Since it is a redirection, the path is not considered, and the script can be loaded, thus bypassing the path restriction.
With this redirection, even if the path is specified completely, it will still be bypassed.
Therefore, the best solution is to ensure that the website does not have any open redirect vulnerabilities and that there are no domains that can be exploited in the CSP rules.
'unsafe-inline' means that you can execute any script inside the code (XSS can execute code) and img-src * means that you can use in the webpage any image from any resource.
You can bypass this CSP by exfiltrating the data via images (in this occasion the XSS abuses a CSRF where a page accessible by the bot contains an SQLi, and extract the flag via an image):
javascript
Service workers importScripts function isn't limited by CSP:
If a parameter sent by you is being pasted inside the declaration of the policy, then you could alter the policy in some way that makes it useless. You could allow script 'unsafe-inline' with any of these bypasses:
bash
Notice the lack of the directive 'unsafe-inline'
This time you can make the victim load a page in your control via XSS with a <iframe. This time you are going to make the victim access the page from where you want to extract information (CSRF). You cannot access the content of the page, but if somehow you can control the time the page needs to load you can extract the information you need.
This time a flag is going to be extracted, whenever a char is correctly guessed via SQLi the response takes more time due to the sleep function. Then, you will be able to extract the flag:
html
This attack would imply some social engineering where the attacker convinces the user to drag and drop a link over the bookmarklet of the browser. This bookmarklet would contain malicious javascript code that when drag&dropped or clicked would be executed in the context of the current web window, bypassing CSP and allowing to steal sensitive information such as cookies or tokens.
You can restrict a CSP of an Iframe with the csp attribute:
html
html
If you can manage to make the server responds with the header Content-Security-Policy-Report-Only with a value controlled by you (maybe because of a CRLF), you could make it point your server and if you wraps the JS content you want to exfiltrate with <script> and because highly probable unsafe-inline isn't allowed by the CSP, this will trigger a CSP error and part of the script (containing the sensitive info) will be sent to the server from Content-Security-Policy-Report-Only.
javascript
An iframe is created that points to a URL (let's call it https://example.redirect.com) which is permitted by CSP.
This URL then redirects to a secret URL (e.g., https://usersecret.example2.com) that is not allowed by CSP.
By listening to the securitypolicyviolation event, one can capture the blockedURI property. This property reveals the domain of the blocked URI, leaking the secret domain to which the initial URL redirected.
It's interesting to note that browsers like Chrome and Firefox have different behaviors in handling iframes with respect to CSP, leading to potential leakage of sensitive information due to undefined behavior.
Another technique involves exploiting the CSP itself to deduce the secret subdomain. This method relies on a binary search algorithm and adjusting the CSP to include specific domains that are deliberately blocked. For example, if the secret subdomain is composed of unknown characters, you can iteratively test different subdomains by modifying the CSP directive to block or allow these subdomains. Here’s a snippet showing how the CSP might be set up to facilitate this method:
markdown
By monitoring which requests are blocked or allowed by the CSP, one can narrow down the possible characters in the secret subdomain, eventually uncovering the full URL.
Both methods exploit the nuances of CSP implementation and behavior in browsers, demonstrating how seemingly secure policies can inadvertently leak sensitive information.
PHP is known for buffering the response to 4096 bytes by default. Therefore, if PHP is showing a warning, by providing enough data inside warnings, the response will be sent before the CSP header, causing the header to be ignored. Then, the technique consists basically in filling the response buffer with warnings so the CSP header isn't sent.
javascript
SOME is a technique that abuses an XSS (or highly limited XSS) in an endpoint of a page to abuse other endpoints of the same origin. This is done by loading the vulnerable endpoint from an attacker page and then refreshing the attacker page to the real endpoint in the same origin you want to abuse. This way the vulnerable endpoint can use the opener object in the payload to access the DOM of the real endpoint to abuse. For more information check:
Moreover, wordpress has a JSONP endpoint in /wp-json/wp/v2/users/1?_jsonp=data that will reflect the data sent in the output (with the limitation of only letter, numbers and dots).
If there is a strict CSP that doesn't allow you to interact with external servers, there are some things you can always do to exfiltrate the information.
You could just update the location to send to the attacker's server the secret information:
javascript
You could redirect by injecting a meta tag (this is just a redirect, this won't leak content)
html
To load pages faster, browsers are going to pre-resolve hostnames into IP addresses and cache them for later usage.
You can indicate a browser to pre-resolve a hostname with: <link rel="dns-prefetch" href="something.com">
You could abuse this behaviour to exfiltrate sensitive information via DNS requests:
javascript
Another way:
javascript
In order to avoid this from happening the server can send the HTTP header:
Apparently, this technique doesn't work in headless browsers (bots)
On several pages you can read that WebRTC doesn't check the connect-src policy of the CSP.
Actually you can leak informations using a DNS request. Check out this code:
javascript
Another option:
javascript
The credential popup sends a DNS request to the iconURL without being restricted by the page. It only works in a secure context (HTTPS) or on localhost.
javascript
This is not working, for more info .
From here, if you find a XSS and a file upload, and you manage to find a misinterpreted extension, you could try to upload a file with that extension and the Content of the script. Or, if the server is checking the correct format of the uploaded file, create a polyglot ().
If not possible to inject JS, you could still try to exfiltrate for example credentials injecting a form action (and maybe expecting password managers to auto-fill passwords). You can find an . Also, notice that default-src does not cover form actions.
):
According to you can abuse inside a CSP to execute arbitrary JS code bypassing the CSP:
More :
The following URL redirects to example.com (from ):
It's possible to abuse Google Apps Script to receive information in a page inside script.google.com. Like it's .
contains ready to use JSONP endpoints to CSP bypass of different websites.
As described in the , there are many third party domains, that might be allowed somewhere in the CSP, can be abused to either exfiltrate data or execute JavaScript code. Some of these third-parties are:
You should be able to exfiltrate data, similarly as it has always be done with /. In this case, you follow these general steps:
As for the other seven third-party domains specified in the previous table, there are many other ways you can abuse them. Refer to the previously for additional explanations about other third-party abuses.
Online Example:
If the base-uri directive is missing you can abuse it to perform a .
Find other Angular bypasses in
A CSP policy that whitelists domains for script loading in an Angular JS application can be bypassed through the invocation of callback functions and certain vulnerable classes. Further information on this technique can be found in a detailed guide available on this .
Other JSONP arbitrary execution endpoints can be found in (some of them were deleted or fixed)
However, according to the description in , if the redirection leads to a different path, it can bypass the original restrictions.
Read .
From:
You could also abuse this configuration to load javascript code inserted inside an image. If for example, the page allows loading images from Twitter. You could craft an special image, upload it to Twitter and abuse the "unsafe-inline" to execute a JS code (as a regular XSS) that will load the image, extract the JS from it and execute it:
Research:
Because this directive will overwrite existing script-src directives. You can find an example here:
In Edge is much simpler. If you can add in the CSP just this: ;_ Edge would drop the entire policy.
Example:
For more information .
In , CSP is bypassed by injecting inside an allowed iframe a more restrictive CSP that disallowed to load a specific JS file that, then, via prototype pollution or dom clobbering allowed to abuse a different script to load an arbitrary script.
In , it was possible via HTML injection to restrict more a CSP so a script preventing CSTI was disabled and therefore the vulnerability became exploitable. CSP can be made more restrictive using HTML meta tags and inline scripts can disabled removing the entry allowing their nonce and enable specific inline script via sha:
For an example .
Trick from .
According to the , sending too many parameters (1001 GET parameters although you can also do it with POST params and more that 20 files). Any defined header() in the PHP web code won't be sent because of the error that this will trigger.
Idea from .
From it looks like it was possible to bypass a CSP protection by loading an error page (potentially without CSP) and rewriting its content.
An attacker can abuse that endpoint to generate a SOME attack against WordPress and embed it inside <script src=/wp-json/wp/v2/users/1?_jsonp=some_attack></script> note that this script will be loaded because it's allowed by 'self'. Moreover, and because WordPress is installed, an attacker might abuse the SOME attack through the vulnerable callback endpoint that bypasses the CSP to give more privileges to a user, install a new plugin...
For more information about how to perform this attack check