Interception caching involves some network trickery, so it is helpful for you to understand what happens between the client and Squid. I'll use Figure 9-1 and the following sample tcpdump output to explain how the packets are intercepted as they flow through your network.
The user-agent wants to request a resource, say /index.html from an origin server, say www.oreilly.com. It needs the origin server's IP address, so it makes a DNS request:
Packet 1 TIME: 19:54:41.317310 UDP: 220.127.116.11.2459 -> 18.104.22.168.53 DATA: .d...........www.oreilly.com..... --------------------------------------------------------------------------- Packet 2 TIME: 19:54:41.317707 (0.000397) UDP: 22.214.171.124.53 -> 126.96.36.199.2459 DATA: .d...........www.oreilly.com.............PR.....%........PR. ....$........PR...ns1.sonic.net.........PR...ns2.Q........PR ...ns...M...............h.............!.z.......b......
Now that it has the IP address, the user-agent initiates a TCP connection to the origin server on port 80:
Packet 3 TIME: 19:54:41.320652 (0.002945) TCP: 188.8.131.52.3897 -> 184.108.40.206.80 Syn DATA: <No data>
The switch/router notices a TCP SYN packet with destination port 80. What happens next depends on the particular interception technology. In the case of layer four switches and policy routing, the device simply forwards the TCP packet to Squid's datalink layer (Ethernet) address. This works only when Squid is directly attached to the network device. For WCCP, the router encapsulates the TCP packet into a GRE packet. Because the GRE packet has its own IP address, it can be routed through multiple subnets. In other words, WCCP doesn't require Squid to be directly attached to the router.
The Squid host's operating system receives the intercepted packet. For layer four switches, the TCP/IP packet is unchanged from the earlier explanation.
If the packet is encapsulated with GRE, the host removes the outer IP and GRE headers and places the original TCP/IP packet on the input queue.
Note that the Squid host receives an IP packet for a foreign address (the origin server's). Normally this packet is dropped because its destination address doesn't match any of the local interface addresses. To make the host accept the foreign packet, you must enable IP forwarding on most operating systems.
The client's TCP/IP packet is processed by the packet filtering code. The packet matches a rule that instructs the kernel to forward or divert this packet to Squid. Without this rule, the kernel simply forwards this packet on its way to the origin server, which isn't what you want.
Note that the SYN packet's destination port is 80, but Squid may be listening on a different port, such as 3128. The packet filtering rules allow you to change the port number. You don't need to make Squid listen on port 80. You can't see this step with tcpdump because the diverted packet doesn't flow through the network interface code again.
The packet filter's redirection rule is still necessary even if you have Squid listen on port 80. Simply making the port numbers match doesn't allow Squid to receive the intercepted packets. The redirection rule is the magic that delivers foreign packets to Squid.
Squid receives notification of the new connection, which it accepts. The kernel sends a SYN/ACK packet back to the client:
Packet 4 TIME: 19:54:41.320735 (0.000083) TCP: 220.127.116.11.80 -> 18.104.22.168.3897 SynAck DATA: <No data>
As you can see, the source address is the origin server's, even though this packet didn't reach the origin. The operating system simply copies and swaps the source and destination IP addresses from the SYN packet into the reply.
The user-agent receives the SYN/ACK packet, fully establishing the TCP connection. The user-agent now believes it is connected to the origin server, so it writes the HTTP request:
Packet 5 TIME: 19:54:41.323080 (0.002345) TCP: 22.214.171.124.3897 -> 126.96.36.199.80 Ack DATA: <No data> --------------------------------------------------------------------------- Packet 6 TIME: 19:54:41.323482 (0.000402) TCP: 188.8.131.52.3897 -> 184.108.40.206.80 AckPsh DATA: GET / HTTP/1.0 User-Agent: Wget/1.8.2 Host: www.oreilly.com Accept: */* Connection: Keep-Alive
Squid receives the HTTP request. It uses the HTTP Host header to convert the partial URL into a full URL. In this case, you'll see http://www.oreilly.com/ in the access.log file.
From this point on, Squid treats the request normally. As usual, cache hits are returned immediately. Cache misses are forwarded to the origin server.
Lastly, here is the response that Squid receives from the origin server:
Packet 8 TIME: 19:54:41.448391 (0.030030) TCP: 220.127.116.11.80 -> 18.104.22.168.3897 AckPsh DATA: HTTP/1.0 200 OK Date: Mon, 29 Sep 2003 01:54:41 GMT Server: Apache/1.3.26 (Unix) PHP/4.2.1 mod_gzip/22.214.171.124a mo d_perl/1.27 P3P: policyref="http://www.oreillynet.com/w3c/p3p.xml",CP="C AO DSP COR CURa ADMa DEVa TAIa PSAa PSDa IVAa IVDa CONo OUR DELa PUBi OTRa IND PHY ONL UNI PUR COM NAV INT DEM CNT STA P RE" Last-Modified: Sun, 28 Sep 2003 23:54:44 GMT ETag: "1b76bf-b910-3ede86c4" Accept-Ranges: bytes Content-Length: 47376 Content-Type: text/html X-Cache: MISS from www.oreilly.com X-Cache: MISS from 10.0.0.1 Connection: keep-alive
You don't want your switch/router to intercept the connections that Squid makes to origin servers. If that happens, Squid ends up talking to itself and can't satisfy any cache misses. The best way to avoid forwarding loops like this is to make sure that your users and Squid connect to separate interfaces on the switch/router. Whenever feasible, you should apply the interception rules to specific interfaces. Obviously, you should not enable interception on the interface that Squid uses.