On 12/07/2014 8:15 p.m., James Harper wrote:
>>
>> On 12/07/2014 5:21 p.m., James Harper wrote:
>>> The docs says that ident doesn't work with intercept proxying, and it
>>> doesn't, but I think it wouldn't be too hard to make it work. In fact
>>> maybe as simple as setting COMM_TRANSPARENT on the ident socket.
>>
>> COMM_TRANSPARENT is a Squid inernal flag telling Squid to use TPROXY
>> binding on the outgoing connection. If you use this you will be sending
>> IDENT requests to the original destination *server*, using the from-IP
>> as the one you were trying to contact.
> 
> Setting COMM_TRANSPARENT actually does work (but maybe unwanted side 
> effects?). I've just tested it. The ident connection appears to come from the 
> destination server so the client handles them correctly and the correct 
> username is logged for intercepted connections.
> 
> But you're saying I should find another way of setting IP_TRANSPARENT on the 
> ident socket?
> 

Which OS are you using?
 what are your http_port settings?
 and what Comm::Connection IP address details are being passed to comm
to setup the IDENT connection?


>> The problem is that the TCP source-port details are used by IDENT
>> protocol. Source-NAT operations in the network before reaching Squid can
>> remove/obscure them completely.
>>
> 
> Ah. Squid is actually running on my gateway so there is no NAT before it 
> reaches squid (and from memory, there is a way of redirecting packets over a 
> GRE tunnel or something to preserve that info... was it WCCP?)
> 

It's not that the information is preserved by the routing technique. It
is that the SNAT operation removes it completely, and some kernel lookup
APIs only present the IP alone. A "works for you+me but nobody else"
type scenario.


>>> Does that sound plausible? What I've found is that not only doesn't
>>> ident not work on an intercepted connection, the connection just
>>> hangs forever (or at least for the 10 minutes that I waited) if any
>>> acl's are encountered that would require an ident lookup.
>>
>> The hang is a separate bug which has now been resolved:
>> http://bugs.squid-cache.org/show_bug.cgi?id=4080
>>
> 
> Excellent. Applying now.
> 
> Thanks
> 
> James
> 

Reply via email to