Opened 11 years ago

Closed 9 years ago

#20825 closed New feature (wontfix)

Make cookies use the HTTPOnly flag by default

Reported by: Julien Phalip Owned by: Julien Phalip
Component: HTTP handling Version: 1.5
Severity: Normal Keywords:
Cc: Triage Stage: Accepted
Has patch: no Needs documentation: no
Needs tests: no Patch needs improvement: no
Easy pickings: no UI/UX: no

Description

The HttpOnly cookie flag helps prevent the most common types of XSS attacks (see https://www.owasp.org/index.php/HttpOnly). Thus it would be preferable for Django to use it by default.

This change would be backwards-incompatible, in particular for applications that rely on cookies being accessible from genuine client-side Javascript code. Therefore this change should be clearly explained in the release notes.

Some information about the benefits and caveats of using this flag should also be added to the security section of the documentation.

This change is somewhat dependent on a resolution being found for #20755, as the Python standard library's cookies module's management of HttpOnly is inconsistent.

Change History (3)

comment:1 by Sasha Romijn, 11 years ago

The current default for SESSION_COOKIE_HTTPONLY in current master is True. So, this concerns only the CSRF cookie, and custom cookies set though set_cookie() on a response. Are you referring to one of these specifically, or both?

comment:2 by Raphael Michel, 9 years ago

Actually, accessing the CSRF cookie via JavaScript is the recommended way to use the CSRF protection with AJAX (see: https://docs.djangoproject.com/en/1.8/ref/csrf/#ajax). So, as far as the CSRF cookie is concerned, I'd close this as "wontfix".

As for set_cookie, this would break backwards compatibility and as the most important cookie, the sessison cookie, is protected, I'd also vote for wontfix here. Any other opinions?

comment:3 by Sasha Romijn, 9 years ago

Resolution: wontfix
Status: newclosed

I completely concur. For the set_cookie default, I do see a security benefit, but I feel it's so small that it's not worth breaking the backwards compatibility. Especially when considering that this is likely to break things silently, and not be noticed in unit tests - only by vague error reports from users.

Note: See TracTickets for help on using tickets.
Back to Top