Reduce overhead in Form Signing

[Ciaro]

Hello guys

The current implementation of CSRF protection in Lithium seems to have a lot of unneeded overhead. It uses sha512 by default, but even if you define md5 as a type, you are still getting tokens generated by 'php::crypt'. This can add up if you use a couple of forms one a page.

Could we not accomplish the same functionality by just using plain-old md5 tokens?

Please do correct me if I'm wrong about this :)

Best regards,

Ciaro

[nateabele]

I would need to do more research into (a) what the real hashing needs are for CSRF tokens, and (b) the overall time difference of each hashing function.

If someone wants to help with a breakdown of these, that would be great.

[Ciaro]

(a) in http://shiflett.org/articles/cross-site-request-forgeries, the author uses md5 in his example.
(b) on a default nginx setup (dev and production) it's between 50ms and 100ms per token generated by 'php::crypt'. Using password::hash with an md5 salt is 16ms for the first, then around 7-8 for anything after it, so acceptable?

Hope this helps.

[nateabele]

Thanks for mentioning that link. It's pretty old, so it's possible that the information is outdated, but I'll ask Chris about it the next time I talk to him.

• Nate

On Saturday, April 20, 2013 at 7:01 PM, Ciaro Vermeire wrote:

(a) in http://shiflett.org/articles/cross-site-request-forgeries, the author uses md5 in his example.
(b) on a default nginx setup (dev and production) it's between 50ms and 100ms per token generated by 'php::crypt'. Using password::hash with an md5 salt is 16ms for the first, then around 7-8 for anything after it, so acceptable?

Hope this helps.

—
Reply to this email directly or view it on GitHub (#839 (comment)).

[Ciaro]

@nateabele Where you able to talk to Chris about this yet? Thanks.

[mariuswilms]

@Ciaro I think the performance problem is not coming from using sha512 for the request token but from using blowfish for the form signing in:
https://github.com/UnionOfRAD/lithium/blob/master/security/validation/FormSignature.php#L39

Also see:
http://php.net/manual/en/function.crypt.php

$2a$10$ is blowfish with 10 rounds and $6$ would be sha512. The question now is if we really need BF which is meant to be slow?

[mariuswilms]

Refs 60bbe47

[Ciaro]

@DavidPersson I don't see why plain md5 shouldn't work for CSRF?

[mariuswilms]

@Ciaro I'd say the tiny performance improvement switching sha512 to md5 isn't worth what we eventually lose in security. Also as I mentioned the real problem here is the usage of blowfish.

[Ciaro]

@DavidPersson I was suggesting to bypass 'php::crypt' completely for the CSRF tokens and just use plain md5 for those.

[gwoo]

Can we use hmac?

On Dec 21, 2013, at 8:35 AM, Ciaro Vermeire notifications@github.com wrote:

@DavidPersson I was suggesting to bypass 'php::crypt' completely for the CSRF tokens and just use plain md5 for those.

—
Reply to this email directly or view it on GitHub.

[nateabele]

Fixed.