From mboxrd@z Thu Jan  1 00:00:00 1970
From: john at keeping.me.uk (John Keeping)
Date: Sat, 16 Jun 2018 17:28:27 +0100
Subject: [PATCH] Encode value and field before calculating cookie digest, 
 the same way secure_value() does
In-Reply-To: <20180412175431.33587-1-thevlad@gmail.com>
References: <20180412175431.33587-1-thevlad@gmail.com>
Message-ID: <20180616162827.GE1922@john.keeping.me.uk>

On Thu, Apr 12, 2018 at 08:54:31PM +0300, thevlad at gmail.com wrote:
> From: Vlad Safronov <thevlad at gmail.com>
> 
> Bugfix: Encode value and field before calculating cookie digest, the same way as secure_value() does
> so validating will work correctly on encoded values.

Missing sign-off (see [1] for what this means).

[1] https://elinux.org/Developer_Certificate_Of_Origin

However, I don't think this change is correct.  secure_value() places
the encoded strings into the authstr, which is what validate_value() is
reading, so field and value are already URL encoded here.

This is why field is url_decode()'d later in this function before
comparing with the expected value.

> ---
>  filters/simple-authentication.lua | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/filters/simple-authentication.lua b/filters/simple-authentication.lua
> index de34d09..b40a819 100644
> --- a/filters/simple-authentication.lua
> +++ b/filters/simple-authentication.lua
> @@ -230,6 +230,8 @@ function validate_value(expected_field, cookie)
>  		return nil
>  	end
>  
> +	value = url_encode(value)
> +	field = url_encode(field)
>  	-- Lua hashes strings, so these comparisons are time invariant.
>  	if hmac ~= crypto.hmac.digest("sha1", field .. "|" .. value .. "|" .. tostring(expiration) .. "|" .. salt, secret) then
>  		return nil
> -- 
> 2.17.0