Hey all,

Here is a problem I'm trying to solve for hours, with no luck:

My debian host has a single ethernet card and is configured with 3 virtual interfaces, each with it's own IP address, let's say a.a.a.a, b.b.b.b, and c.c.c.c

A squid server is proxying http requests for authenticated users. Squid is configured to listen to all interfaces, not bound to any single one.

What I need to do is make the squid http requests originate from the *same* IP address that the client request comes in. For example if a client uses a.a.a.a:3128 to make a request to the proxy, squid will use a.a.a.a in its corresponding http request. If another client request comes in b.b.b.b:3128, squid will use b.b.b.b for it's http request.
In other words, I need the source IP of the squid http requests leaving my box to be the same as the destination IP of the client requests that come in.
I haven't found any way to do this with squid itself, so I tried in vain to implement it in iptables by trying to MARK/SNAT traffic. As it seems, squid http requests are not RELATED to the client requests to squid, so there is no way to do this with usual connection tracking. Is there any other way?
Thanks for any help!

I don't quite understand the scenario.

I'm not an HTTP expert, but I think when you say "http request" in this context, you really mean the response + message body that is returned by the remote web server, right?

Also, is your squid host providing NAT to your client networks or not?

-------

Finally, what OS / version / squid version?

Thank you for your reply anomie. Sorry for not being very clear.
It also turns out I should have RTFMed better before asking for help! I found that squid is perfectly capable of providing what I wanted. I just added these to my squid.conf file, and it worked:

acl incoming1 myip a.a.a.a
acl incoming2 myip b.b.b.b
acl incoming3 myip c.c.c.c

http_access allow incoming1 incoming2 incoming3

tcp_outgoing_address a.a.a.a incoming1
tcp_outgoing_address b.b.b.b incoming2
tcp outgoing_address c.c.c.c incoming3


Again, what I wanted was that squid would use the same IP address to make any requests to remote servers that the client uses in his requests.
For example if a client makes a request to squid on IP a.a.a.a, squid will use the same IP to connect to the remote server.

For the reference, the OS is Debian 5.0.3 (stable), Squid is 2.7


One last question though: how can I, in addition to the above, demand that the clients are authenticated? I mean something similar to:

acl auth proxy_auth REQUIRED

Is it possible to "chain" acl statements? Sorry, I'm a complete squid noob at the moment!

Replying to myself, the http_access statement above is of course wrong, and incidentally the answer to my second question. The correct configuration is:

Case solved.

You already solved your own problem, but yes.

Note that named acl directives are cumulative. So:

... means that "foo" consists of both .some.place and .other.place.

On the other hand, http_access directives are treated like logical ANDs. So:

... means that both "foo" and "goodguys" must be satisfied to match the rule.