XMLHttpRequest cannot load XXX No 'Access-Control-Allow-Origin' header

ghz 1years ago ⋅ 9128 views

Question

tl;dr; About the Same Origin Policy

I have a Grunt process which initiates an instance of express.js server. This was working absolutely fine up until just now when it started serving a blank page with the following appearing in the error log in the developer's console in Chrome (latest version):

XMLHttpRequest cannot load https://www.example.com/ No 'Access-Control- Allow-Origin' header is present on the requested resource. Origin 'http://localhost:4300' is therefore not allowed access.

What is stopping me from accessing the page?


Answer

tl;dr — When you want to read data, (mostly) using client-side JS, from a different server you need the server with the data to grant explicit permission to the code that wants the data.

There's a summary at the end and headings in the answer to make it easier to find the relevant parts. Reading everything is recommended though as it provides useful background for understanding the why that makes seeing how the how applies in different circumstances easier.

About the Same Origin Policy

This is the [Same Origin Policy](https://developer.mozilla.org/en- US/docs/Web/Security/Same-origin_policy). It is a security feature implemented by browsers.

Your particular case is showing how it is implemented for XMLHttpRequest (and you'll get identical results if you were to use fetch), but it also applies to other things (such as images loaded onto a <canvas> or documents loaded into an <iframe>), just with slightly different implementations.

The standard scenario that demonstrates the need for the SOP can be demonstrated with three characters:

  • Alice is a person with a web browser
  • Bob runs a website (https://www.example.com/ in your example)
  • Mallory runs a website (http://localhost:4300 in your example)

Alice is logged into Bob's site and has some confidential data there. Perhaps it is a company intranet (accessible only to browsers on the LAN), or her online banking (accessible only with a cookie you get after entering a username and password).

Alice visits Mallory's website which has some JavaScript that causes Alice's browser to make an HTTP request to Bob's website (from her IP address with her cookies, etc). This could be as simple as using XMLHttpRequest and reading the responseText.

The browser's Same Origin Policy prevents that JavaScript from reading the data returned by Bob's website (which Bob and Alice don't want Mallory to access). (Note that you can, for example, display an image using an <img> element across origins because the content of the image is not exposed to JavaScript (or Mallory) … unless you throw canvas into the mix in which case you will generate a same-origin violation error).


Why the Same Origin Policy applies when you don't think it should

For any given URL it is possible that the SOP is not needed. A couple of common scenarios where this is the case are:

  • Alice, Bob, and Mallory are the same person.
  • Bob is providing entirely public information

… but the browser has no way of knowing if either of the above is true, so trust is not automatic and the SOP is applied. Permission has to be granted explicitly before the browser will give the data it has received from Bob to some other website.


Why the Same Origin Policy applies to JavaScript in a web page but little

else

Outside the web page

Browser extensions*, the Network tab in browser developer tools, and applications like Postman are installed software. They aren't passing data from one website to the JavaScript belonging to a different website just because you visited that different website. Installing software usually takes a more conscious choice.

There isn't a third party (Mallory) who is considered a risk.

* Browser extensions do need to be written carefully to avoid cross-origin issues. [See the Chrome documentation for example](https://www.chromium.org/Home/chromium-security/extension-content- script-fetches).

Inside the webpage

Most of the time, there isn't a great deal of information leakage when just showing something on a webpage.

If you use an <img> element to load an image, then it gets shown on the page, but very little information is exposed to Mallory. JavaScript can't read the image (unless you use a crossOrigin attribute to explicitly enable request permission with CORS) and then copy it to her server.

That said, some information does leak so, to quote Domenic Denicola (of Google):

The web's fundamental security model is the same origin policy. We have several legacy exceptions to that rule from before that security model was in place, with script tags being one of the most egregious and most dangerous. (See the various "JSONP" attacks.)

Many years ago, perhaps with the introduction of XHR or web fonts (I can't recall precisely), we drew a line in the sand, and said no new web platform features would break the same origin policy. The existing features need to be grandfathered in and subject to carefully-honed and oft-exploited exceptions, for the sake of not breaking the web, but we certainly can't add any more holes to our security policy.

This is why you need CORS permission to load fonts across origins.


Why you can display data on the page without reading it with JS

There are a number of circumstances where Mallory's site can cause a browser to fetch data from a third party and display it (e.g. by adding an <img> element to display an image). It isn't possible for Mallory's JavaScript to read the data in that resource though, only Alice's browser and Bob's server can do that, so it is still secure.


CORS

The Access-Control-Allow-Origin HTTP response header referred to in the error message is part of the [CORS](https://developer.mozilla.org/en- US/docs/Web/HTTP/Access_control_CORS) standard which allows Bob to explicitly grant permission to Mallory's site to access the data via Alice's browser.

A basic implementation would just include:

Access-Control-Allow-Origin: *

… in the response headers to permit any website to read the data.

Access-Control-Allow-Origin: http://example.com

… would allow only a specific site to access it, and Bob can dynamically generate that based on the Origin request header to permit multiple, but not all, sites to access it.

The specifics of how Bob sets that response header depend on Bob's HTTP server and/or server-side programming language. Users of Node.js/Express.js should use the well-documented CORS middleware. Users of other platforms should take a look at this collection of guides for various common configurations that might help.

Model of where CORS rules are
applied

NB: Some requests are complex and send a [preflight](https://developer.mozilla.org/en- US/docs/Web/HTTP/Access_control_CORS#Preflighted_requests) OPTIONS request that the server will have to respond to before the browser will send the GET/POST/PUT/Whatever request that the JS wants to make. Implementations of CORS that only add Access-Control-Allow-Origin to specific URLs often get tripped up by this.


Obviously granting permission via CORS is something Bob would only do only if either:

  • The data was not private or
  • Mallory was trusted

How do I add these headers?

It depends on your server-side environment.

If you can, use a library designed to handle CORS as they will present you with simple options instead of having to deal with everything manually.

Enable-Cors.org has a list of documentation for specific platforms and frameworks that you might find useful.

But I'm not Bob!

There is no standard mechanism for Mallory to add this header because it has to come from Bob's website, which she does not control.

If Bob is running a public API then there might be a mechanism to turn on CORS (perhaps by formatting the request in a certain way, or a config option after logging into a Developer Portal site for Bob's site). This will have to be a mechanism implemented by Bob though. Mallory could read the documentation on Bob's site to see if something is available, or she could talk to Bob and ask him to implement CORS.


Error messages which mention "Response for preflight"

Some cross-origin requests are [preflighted](https://developer.mozilla.org/en- US/docs/Web/HTTP/Access_control_CORS#Preflighted_requests).

This happens when (roughly speaking) you try to make a cross-origin request that:

  • Includes credentials like cookies
  • Couldn't be generated with a regular HTML form (e.g. sets a Content-Type that you couldn't use in a form's enctype or some other request header(s)).

If you are correctly doing something that needs a preflight

In these cases then the rest of this answer still applies but you also need to make sure that the server can listen for the preflight request (which will be OPTIONS (and not GET, POST, or whatever you were trying to send) and respond to it with the right Access-Control-Allow-Origin header but also Access-Control-Allow-Methods and Access-Control-Allow-Headers to allow your specific HTTP methods or headers.

If you are triggering a preflight by mistake

Sometimes people make mistakes when trying to construct Ajax requests, and sometimes these trigger the need for a preflight. If the API is designed to allow cross-origin requests but doesn't require anything that would need a preflight, then this can break access.

Common mistakes that trigger this include:

  • trying to put Access-Control-Allow-Origin and other CORS response headers on the request. These don't belong on the request, don't do anything helpful (what would be the point of a permissions system where you could grant yourself permission?), and must appear only on the response.
  • trying to put a Content-Type: application/json header on a GET request that has no request body the content of which to describe (typically when the author confuses Content-Type and Accept).

In either of these cases, removing the extra request header will often be enough to avoid the need for a preflight (which will solve the problem when communicating with APIs that support simple requests but not preflighted requests).


Opaque responses (no-cors mode)

Sometimes you need to make an HTTP request, but you don't need to read the response. e.g. if you are posting a log message to the server for recording.

If you are using [the fetch API](https://developer.mozilla.org/en- US/docs/Web/API/Fetch_API/Using_Fetch) (rather than XMLHttpRequest), then you can configure it to not try to use CORS.

Note that this won't let you do anything that you require CORS to do. You will not be able to read the response. You will not be able to make a request that requires a preflight.

It will let you make a simple request, not see the response, and not fill the Developer Console with error messages.

How to do it is explained by the Chrome error message given when you make a request using fetch and don't get permission to view the response with CORS:

Access to fetch at 'https://example.com/' from origin 'https://example.net' has been blocked by CORS policy: No 'Access- Control-Allow-Origin' header is present on the requested resource. If an opaque response serves your needs, set the request's mode to 'no-cors' to fetch the resource with CORS disabled.

Thus:

fetch("http://example.com", { mode: "no-cors" });

Alternatives to CORS

JSONP

Bob could also provide the data using a hack like JSONP which is how people did cross-origin Ajax before CORS came along.

It works by presenting the data in the form of a JavaScript program that injects the data into Mallory's page.

It requires that Mallory trust Bob not to provide malicious code.

Note the common theme: The site providing the data has to tell the browser that it is OK for a third-party site to access the data it is sending to the browser.

Since JSONP works by appending a <script> element to load the data in the form of a JavaScript program that calls a function already in the page, attempting to use the JSONP technique on a URL that returns JSON will fail — typically with a CORB error — because JSON is not JavaScript.

Move the two resources to a single Origin

If the HTML document the JS runs in and the URL being requested are on the same origin (sharing the same scheme, hostname, and port) then the Same Origin Policy grants permission by default. CORS is not needed.

A Proxy

Mallory could use server-side code to fetch the data (which she could then pass from her server to Alice's browser through HTTP as usual).

It will either:

  • add CORS headers
  • convert the response to JSONP
  • exist on the same origin as the HTML document

That server-side code could be written & hosted by a third party (such as CORS Anywhere). Note the privacy implications of this: The third party can monitor who proxies what across their servers.

Bob wouldn't need to grant any permissions for that to happen.

There are no security implications here since that is just between Mallory and Bob. There is no way for Bob to think that Mallory is Alice and to provide Mallory with data that should be kept confidential between Alice and Bob.

Consequently, Mallory can only use this technique to read public data.

Do note, however, that taking content from someone else's website and displaying it on your own might be a violation of copyright and open you up to legal action.

Writing something other than a web app

As noted in the section "Why the Same Origin Policy only applies to JavaScript in a web page", you can avoid the SOP by not writing JavaScript in a webpage.

That doesn't mean you can't continue to use JavaScript and HTML, but you could distribute it using some other mechanism, such as Node-WebKit or PhoneGap.

Browser extensions

It is possible for a browser extension to inject the CORS headers in the response before the Same Origin Policy is applied.

These can be useful for development but are not practical for a production site (asking every user of your site to install a browser extension that disables a security feature of their browser is unreasonable).

They also tend to work only with simple requests (failing when handling preflight OPTIONS requests).

Having a proper development environment with a local development server is usually a better approach.


Other security risks

Note that SOP / CORS do not mitigate XSS, CSRF, or SQL Injection attacks which need to be handled independently.


Summary

  • There is nothing you can do in your client-side code that will enable CORS access to someone else's server.
  • If you control the server the request is being made to: Add CORS permissions to it.
  • If you are friendly with the person who controls it: Get them to add CORS permissions to it.
  • If it is a public service then (keeping in mind that most third-party APIs are designed to be interacted with only by your server-side code and do not support CORS but for those that are):
    • Read their API documentation to see what they say about accessing it with client-side JavaScript:
      • They might tell you to use specific URLs
      • They might support JSONP instead of CORS
      • They might not support cross-origin access from client-side code at all (this might be a deliberate decision on security grounds, especially if you have to pass a personalized API Key in each request).
    • Make sure you aren't triggering a preflight request you don't need. The API might grant permission for simple requests but not preflighted requests.
  • If none of the above apply: Get the browser to talk to your server instead, and then have your server fetch the data from the other server and pass it on. (There are also third-party hosted services that attach CORS headers to publically accessible resources that you could use).