CSRF referer checking too strict

[adam]

I get this error:

Forbidden (403)
CSRF verification failed. Request aborted.

Reason given for failure:

Referer checking failed - ​https://sub.domain.com does not match ​https://sum.domain.com/.

Using IE6 on my site. In the apache log the request looks like:

86.24.194.171 - - [15/Mar/2011:15:07:06 +0000] "POST / HTTP/1.1" 403 1030 "​https://sub.domain.com" "Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; SV1; .NET CLR 1.1.4322; .NET CLR 2.0.50727; .NET CLR 3.0.4506.2152; .NET CLR 3.5.30729; .NET4.0C; .NET4.0E)"

So it looks like the referer should not be required to start with a url including a trailing slash. That is a change to make:

good_referer = '​https://%s' % request.get_host()

Happy to provide a patch if people agree with my conclusions.

[anonymous]

The above urls should all be the same except for the trailing slash. I've obfuscated the url because the site is not yet public.

[Luke Plant]

This is definitely a bug. However, the fix is not so simple, because we are using 'startswith' to check the referer, and with the proposed change, https://example.com could be matched by https://example.com.evil.com . That is almost certainly why I added that slash.

The correct way to do with is to compare the (protocol, domain, port) triple, as specified here: ​http://www.w3.org/Security/wiki/Same_Origin_Policy

We could do with a 'same_origin' utility function that is properly tested and implements the above check using urlparse, and then is used by the middleware.

[Luke Plant]

In [15840]:

Fixed #15617 - CSRF referer checking too strict

Thanks to adam for the report.

[Luke Plant]

In [15844]:

[1.2.X] Fixed #15617 - CSRF referer checking too strict

Thanks to adam for the report.

Backport of [15840] from trunk.