tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From André Warnier ...@ice-sa.com>
Subject Re: Filter by HTTP_REFERER
Date Wed, 02 Nov 2011 12:51:09 GMT
Casper Wandahl Schmidt wrote:
> 
> 
> Den 02-11-2011 10:50, André Warnier skrev:
>> André Warnier wrote:
>>> Casper Wandahl Schmidt wrote:
>>>>
>>>>
>>>> Den 02-11-2011 10:17, André Warnier skrev:
>>>>> Pid * wrote:
>>>>>> On 31 Oct 2011, at 18:25, Christopher Schultz
>>>>>> <chris@christopherschultz.net> wrote:
>>>>>>
>>>>>>> -----BEGIN PGP SIGNED MESSAGE-----
>>>>>>> Hash: SHA1
>>>>>>>
>>>>>>> Richardo,
>>>>>>>
>>>>>>> On 10/31/2011 12:33 PM, Ricardo Bayley wrote:
>>>>>>>> You are right. What I intend to do is prevent hot linking.
>>>>>>> We get what you are trying to do: you'll just have to write your
own
>>>>>>> code to do it. Tomcat ships with a Filter called RequestFilter
that
>>>>>>> you can subclass if you need that kind of flexibility. If you
don't
>>>>>>> need such flexibility, just write it yourself: it's pretty much
a
>>>>>>> one-liner.
>>>>>>
>>>>>> It'll still be fragile and open to exploitation. An AJAX call can
set
>>>>>> any request headers it likes. You be better off using authentication
>>>>>> if you want anything more than a casual defence.
>>>>>>
>>>>>>
>>>>>> p
>>>>>>
>>>>>>
>>>>>>>> My webapp, is working as a REST webservice.
>>>>>>>>
>>>>>>>> So I would like to have tomcat reply only when requests come
from
>>>>>>>> specific sites.
>>>>>>> You mean when the requests are referred from specific sites,
right?
>>>>>>>
>>>>>>> - -chris
>>>>>
>>>>> I think that a bit of clarification is in order now.
>>>>>
>>>>> 1) When receiving a request, Tomcat "knows" from which client IP 
>>>>> this request is coming.
>>>>> This is because there is a TCP connection nbetween the client and 
>>>>> Tomcat, and the TCP/IP stack on the Tomcat machine "knows" the IP 
>>>>> address and TCP port from which the remote client is making this 
>>>>> connection.
>>>>>
>>>>> Tomcat can allow/block requests originating from specific IP 
>>>>> addresses, using the RemoteAddress filter (See 
>>>>> http://tomcat.apache.org/tomcat-7.0-doc/config/valve.html#Remote_Address_Filter)

>>>>> or the Remote IP Valve (also mentioned there).
>>>>> This is fairly efficient, because Tomcat already has the 
>>>>> information needed to make the decision.
>>>>>
>>>>> 2) Because Tomcat has the client's remote IP address, it can also 
>>>>> make a "reverse DNS lookup", to find out which domain name 
>>>>> corresponds to this IP address, and then allow/deny the request 
>>>>> based on the remote host's domain name.  This is done via the 
>>>>> RemoteHost filter 
>>>>> (http://tomcat.apache.org/tomcat-7.0-doc/config/valve.html#Remote_Host_Filter).

>>>>>
>>>>> It is less efficient than option (1), because a DNS lookup has to 
>>>>> take place.
>>>>> Furthermore, this DNS lookup is not guaranteed to succeed, because 
>>>>> not all IP addresses are mapped in reverse to a hostname.
>>>>> I also wonder what happens exactly when the remote host's IP 
>>>>> address corresponds to multiple DNS names, as can happen.
>>>>>
>>>>> 3) the HTTP "Referer" header in a HTTP request is a different animal.
>>>>>
>>>>> First, as pointed out by Pid, it can be easily faked by a client.
>>>>> (So anything based on this should certainly /not/ be considered as 
>>>>> any kind of security-enhancing feature).
>>>>>
>>>>> Second, it works as follows :
>>>>>
>>>>> Imagine a browser B, some website W, and the Tomcat web server T.
>>>>> The browser B gets a web page from site W. This is now the 
>>>>> browser's "current page", and its origin is (to the browser) 
>>>>> something like : "http://website-W/some-document.html".
>>>>>
>>>>> In that page, there is a URL link to the Tomcat website on T, for 
>>>>> example something like :
>>>>> <a href="http://website-T/some-REST-url">click here</a>.
>>>>> When the user clicks on that link, the browser will send a HTTP 
>>>>> request to Tomcat on server T.  And the browser, in this request, 
>>>>> will add a "Referer:" header containing 
>>>>> "http://website-W/some-document.html".
>>>>> That is because the request originates from the current page, which 
>>>>> had been obtained from "http://website-W/some-document.html".  That 
>>>>> is what "referrer" means.
>>>>>
>>>>> As you can see, this is different from cases (1) and (2), because 
>>>>> (1) and (2) refer to the client's own IP address or name, while (3) 
>>>>> refers to another server altogether.
>>>>>
>>>>> Now the point is to know exactly what Ricardo wants, as per this 
>>>>> phrase :
>>>>> >>> So I would like to have tomcat reply only when requests
come from
>>>>> >>> specific sites.
>>>>>
>>>>> Does that mean :
>>>>> - Tomcat should reject requests coming from workstations that are 
>>>>> not themselves within a particular range of IP addresses ? (for 
>>>>> example, only from some customer LANs L1, L2, L3,..) of which the 
>>>>> IP addresses are known in advance ?
>>>>> OR
>>>>> - Tomcat should reject requests coming from clients whose own IP 
>>>>> addresses cannot be resolved to hostnames that are members of some 
>>>>> DNS domain (like "*.customer1.com" or "*.customer2.org") ?
>>>>> OR
>>>>> - Tomcat should reject requests (from any client), unless these 
>>>>> requests come from a link which itself appears only on some websites
?
>>>>> (and the possibility of a malicious client "faking" such a 
>>>>> "Referer" can be ignored).
>>>> According to OP he want's to avoid hot-linking which would be this 
>>>> last case or have I misunderstood something?
>>>
>>> Well, "hot-linking" is not a term he himself mentioned, it is a term 
>>> someone else mentioned, after interpreting the above phrase that he 
>>> wrote.  It is not so clear to me what he really wants.
>>>
>> Sorry, I stand corrected. He did use "hot-linking".  But then, the 
>> next phrase kind of contradicts that, because "request comes from a 
>> specific site" means something else (to me at least).  When you get a 
>> page from google.com, and on that page is a link to your website, 
>> clicking it does not mean that the request "comes from google.com".  
>> It comes from your browser.
>>
> I guess that depends how you read the word "site" (I believe you read it 
> as like a building site, a psysical location - or well semi-psysical due 
> to the fact that LAN's can be spread to multiple psysical locations - 
> whereas I read it as website and then simply ignore the fact that he 
> does not have the full understanding of who and how requests are made) :)
>>

Right. I was thinking of the following type of scenario, which we encounter frequently :
A large organisation has a large internal LAN, spread over multiple physical locations; 
all accesses to the WWW from workstations within that LAN, go over one of 3 HTTP 
firewalls/proxies. So to an external webserver, all accesses from workstations in that LAN

appear as coming from only 3 distinct IP addresses (the external IP addresses of the 
proxies).  So it makes it easy to filter using the RemoteAddress filter.
(Not preventing hot-linking, but allowing only requests "coming from one site", with a 
certain meaning of "site").

One more thing, if Ricardo is still listening : no matter which filtering you want to do 
in the end, and in addition to the filters mentioned above, you may want to have a look at

the UrlRewriteFilter, at http://www.tuckey.org.
It resembles, and is inspired by, the Apache httpd mod_rewrite module, and it provides /a

lot/ of possibilities in terms of filtering requests.
It works as a servlet filter, not as a Valve. So it will not filter requests as soon as 
they come in Tomcat, but it will filter them at the level of one web application.
Depending on your needs, this may be an advantage or an inconvenient.



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@tomcat.apache.org
For additional commands, e-mail: users-help@tomcat.apache.org


Mime
View raw message