1. Pike Module

Bogdan Iancu

FhG FOKUS
Revision History
Revision $Revision$ $Date$

1.1. Overview
1.2. Parameters
1.2.1. sampling_time_unit (integer)
1.2.2. reqs_density_per_unit (integer)
1.2.3. remove_latency (integer)
1.3. Functions
1.3.1. pike_check_req()
1.4. Developer's Guide
1.5. RPC calls
1.5.1. pike.top

1.1. Overview

The module keeps trace of all (or selected ones) incoming request's IP source and blocks the ones that exceeded some limit. Works simultaneous for IPv4 and IPv6 addresses.

1.2. Parameters

Revision History
Revision $Revision$ $Date$

1.2.1. sampling_time_unit (integer)

Time period used for sampling (or the sampling accuracy ;-) ). The smaller the better, but slower. If you want to detect peeks, use a small one. To limit the access (like total number of requests on a long period of time) to a proxy resource (a gateway for ex), use a bigger value of this parameter.

Default value is 2.

Example 1. Set sampling_time_unit parameter

...
modparam("pike", "sampling_time_unit", 10)
...

1.2.2. reqs_density_per_unit (integer)

How many requests should be allowed per sampling_time_unit before blocking all the incoming request from that IP. Practically, the blocking limit is between ( let's have x=reqs_density_per_unit) x and 3*x for IPv4 addresses and between x and 8*x for ipv6 addresses.

Default value is 30.

Example 2. Set reqs_density_per_unit parameter

...
modparam("pike", "reqs_density_per_unit", 30)
...

1.2.3. remove_latency (integer)

For how long the IP address will be kept in memory after the last request from that IP address. It's a sort of timeout value.

Default value is 120.

Example 3. Set remove_latency parameter

...
modparam("pike", "remove_latency", 130)
...

1.3. Functions

Revision History
Revision $Revision$ $Date$

1.3.1.  pike_check_req()

Process the source IP of the current request and returns false if the IP was exceeded the blocking limit.

Meaning of the parameters is as follows:

Example 4. pike_check_req usage

...
if (!pike_check_req()) { break; };
...

1.4. Developer's Guide

Revision History
Revision $Revision$ $Date$

One single tree (for both IPv4 and IPv6) is used. Each node contains a byte, the IP addresses stretching from root to the leafs.

Example 5. Tree of IP addresses

	   / 193 - 175 - 132 - 164
tree root /                  \ 142
	  \ 195 - 37 - 78 - 163
	             \ 79 - 134

To detect the whole address, step by step, from the root to the leafs, the nodes corresponding to each byte of the ip address are expanded. In order to be expended a node has to be hit for a given number of times (possible by different addresses; in the previous example, the node "37" was expanded by the 195.37.78.163 and 195.37.79.134 hits).

For 193.175.132.164 with x= reqs_density_per_unit:

  • After first req hits -> the "193" node is built.

  • After x more hits, the "175" node is build; the hits of "193" node are split between itself and its child--both of them gone have x/2.

  • And so on for node "132" and "164".

  • Once "164" build the entire address can be found in the tree. "164" becomes a leaf. After it will be hit as a leaf for x times, it will become "RED" (further request from this address will be blocked).

So, to build and block this address were needed 3*x hits. Now, if reqs start coming from 193.175.132.142, the first 3 bytes are already in the tree (they are shared with the previous address), so I will need only x hits (to build node "142" and to make it "RED") to make this address also to be blocked. This is the reason for the variable number of hits necessary to block an IP.

The maximum number of hits to turn an address red are (n is the address's number of bytes):

1 (first byte) + x (second byte) + (x / 2) * (n - 2) (for the rest of the bytes) + (n - 1) (to turn the node to red).

So, for IPv4 (n = 4) will be 3x and for IPv6 (n = 16) will be 9x. The minimum number of hits to turn an address red is x.

1.5. RPC calls

Revision History
Revision $Revision$ $Date$

1.5.1.  pike.top

Pike.top behaves like a 'top' command and shows source IP addresses of incoming requestes to pike_check_req() function.

The IP list is sorted by sum of leaf hits (prev and curr) descending and in second level by hits.

Some IPs could be marked as HOT depending on theirs request rates.

pike.top command takes one string parameter which specifies what kind of nodes you are interested in. Possible values are HOT or ALL. If no argument is given, it behaves as HOT was used.

Marking nodes HOT is done on server side, client only presents given data and make some postprocessing like sorting.

Output of this command is a simple dump of ip_tree nodes marked as ip-leafs.