Is it safe to send Content-Security-Policy for dynamically generated pages with text/html and other hypertext content-types only or do I need to send this header for all files including static assets - images, JS and CSS files?
2 Answers
There are, cases where users can influence MIME type based on different factors. For example, IE can be fooled to render text/plain as text/html within certain circumstances. And, again there are various other MIME types which are rendered and can exfiltrate data. For example, even pdf files can execute JavaScript and so can Flash, SVG, XML or any other plug-in handled content types.
Therefore, it's best to apply CSP using configuration file on all rendered contents.
On a side note, always return correct content type with correct charset attribute along with X-Content-Type-Options: nosniff header.
-
I have
nosniffalready applied. For user uploaded files I can apply a strict CSP policy which doesn't need to change. But the files I have in mind are static assets of the web application which are managed by developers.– AlexDCommented May 25, 2018 at 17:34 -
Also, I understand that CSP can limit plugins used on the page but is the CSP honored by browser plugins?– AlexDCommented May 25, 2018 at 17:54
-
No, plug-ins can do whatever they want. And, please remember, on same origin, any MIME type can be sniffed unless instructed not to. Commented May 26, 2018 at 4:44
-
So, what is the point of applying CSP for a PDF if it is ignored by a plugin used to render it?– AlexDCommented May 26, 2018 at 16:24
-
Well, PDFs no longer require a separate plug-in. It is handled by the pdf.js library. Commented May 26, 2018 at 18:52
In addition to text/html responses, web worker responses (text/javascript) should also have a Content-Security-Policy header:
WebWorkers | Content security policy
Workers are considered to have their own execution context, distinct from the document that created them. For this reason they are, in general, not governed by the content security policy of the document (or parent worker) that created them.
There are some other excellent answers on this topic here. In short, CSP governs what can be done inside a document, and web pages are the normal document, though it's conceivable a browser could support other types of documents (SVG? PDF?). CSP is applied at the root of the execution context, so it doesn't make sense on accessory requests.
text/xmlcontent which can be rendered as (x)HTML, but I do not see the point in configuring the server to send different security headers based on mimetype.pngshould be quite safe, but I would be not that sure forsvgfor instance or folders for user uploads and similar.svgbut I'm not sure about other types.