Opened 4 years ago
Last modified 2 years ago
#32800 closed Cleanup/optimization
CsrfViewMiddleware unnecessarily masks CSRF cookie — at Version 1
Reported by: | Chris Jerdonek | Owned by: | nobody |
---|---|---|---|
Component: | CSRF | Version: | dev |
Severity: | Normal | Keywords: | |
Cc: | Shai Berger, Florian Apolloner | Triage Stage: | Ready for checkin |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
I noticed that CsrfViewMiddleware
unnecessarily masks the CSRF cookie. See, for example:
https://github.com/django/django/blob/d270dd584e0af12fe6229fb712d0704c232dc7e5/django/middleware/csrf.py#L91
My understanding of the BREACH attack is that the vulnerability comes from not masking the CSRF token in the response body (e.g. what is included in the HTML). Masking the cookie itself doesn't help with this. (Django also doesn't change the mask of the cookie with each request, so the mask wouldn't help in this regard anyways.)
Some advantages of not masking the cookie are: It would simplify the code in CsrfViewMiddleware
because it would remove some complexity and operations that aren't needed for security. Currently, masking the CSRF cookie is a red herring for someone wanting to understand the various security features. Also, not masking the cookie would make requests and responses smaller when CSRF_USE_SESSIONS
is false or when true and cookie-based sessions are used. This is because masking doubles the length of the string.
This can be changed fairly easily while (1) continuing to respect masked cookie values, and (2) not forcing session stores to update their cookie in its unmasked form if they are currently storing it masked.