Oskar Andreasson - Iptables Tutorial 1.2.2

Within iptables, packets can be related to tracked connections in four different so called states. These are known as NEW, ESTABLISHED, RELATED and INVALID. We will discuss each of these in more depth later. With the --state match we can easily control who or what is allowed to initiate new sessions.

All of the connection tracking is done by special framework within the kernel called conntrack. conntrack may be loaded either as a module, or as an internal part of the kernel itself. Most of the time, we need and want more specific connection tracking than the default conntrack engine can maintain. Because of this, there are also more specific parts of conntrack that handles the TCP, UDP or ICMP protocols among others. These modules grab specific, unique, information from the packets, so that they may keep track of each stream of data. The information that conntrack gathers is then used to tell conntrack in which state the stream is currently in. For example, UDP streams are, generally, uniquely identified by their destination IP address, source IP address, destination port and source port.

In previous kernels, we had the possibility to turn on and off defragmentation. However, since iptables and Netfilter were introduced and connection tracking in particular, this option was gotten rid of. The reason for this is that connection tracking can not work properly without defragmenting packets, and hence defragmenting has been incorporated into conntrack and is carried out automatically. It can not be turned off, except by turning off connection tracking. Defragmentation is always carried out if connection tracking is turned on.

All connection tracking is handled in the PREROUTING chain, except locally generated packets which are handled in the OUTPUT chain. What this means is that iptables will do all recalculation of states and so on within the PREROUTING chain. If we send the initial packet in a stream, the state gets set to NEW within the OUTPUT chain, and when we receive a return packet, the state gets changed in the PREROUTING chain to ESTABLISHED, and so on. If the first packet is not originated by ourself, the NEW state is set within the PREROUTING chain of course. So, all state changes and calculations are done within the PREROUTING and OUTPUT chains of the nat table.

Let's take a brief look at a conntrack entry and how to read them in /proc/net/ip_conntrack. This gives a list of all the current entries in your conntrack database. If you have the ip_conntrack module loaded, a cat of /proc/net/ip_conntrack might look like:

tcp 6 117 SYN_SENT src=192.168.1.6 dst=192.168.1.9 sport=32775 \

dport=22 [UNREPLIED] src=192.168.1.9 dst=192.168.1.6 sport=22 \

dport=32775 [ASSURED] use=2

This example contains all the information that the conntrack module maintains to know which state a specific connection is in. First of all, we have a protocol, which in this case is tcp. Next, the same value in normal decimal coding. After this, we see how long this conntrack entry has to live. This value is set to 117 seconds right now and is decremented regularly until we see more traffic. This value is then reset to the default value for the specific state that it is in at that relevant point of time. Next comes the actual state that this entry is in at the present point of time. In the above mentioned case we are looking at a packet that is in the SYN_SENT state. The internal value of a connection is slightly different from the ones used externally with iptables. The value SYN_SENT tells us that we are looking at a connection that has only seen a TCP SYN packet in one direction. Next, we see the source IP address, destination IP address, source port and destination port. At this point we see a specific keyword that tells us that we have seen no return traffic for this connection. Lastly, we see what we expect of return packets. The information details the source IP address and destination IP address (which are both inverted, since the packet is to be directed back to us). The same thing goes for the source port and destination port of the connection. These are the values that should be of any interest to us.

The connection tracking entries may take on a series of different values, all specified in the conntrack headers available in linux/include/netfilter-ipv4/ip_conntrack*.h files. These values are dependent on which sub-protocol of IP we use. TCP, UDP or ICMP protocols take specific default values as specified in linux/include/netfilter-ipv4/ip_conntrack.h. We will look closer at this when we look at each of the protocols; however, we will not use them extensively through this chapter, since they are not used outside of the conntrack internals. Also, depending on how this state changes, the default value of the time until the connection is destroyed will also change.

NoteRecently there was a new patch made available in iptables patch-o-matic, called tcp-window-tracking. This patch adds, among other things, all of the above timeouts to special sysctl variables, which means that they can be changed on the fly, while the system is still running. Hence, this makes it unnecessary to recompile the kernel every time you want to change the timeouts.

These can be altered via using specific system calls available in the /proc/sys/net/ipv4/netfilter directory. You should in particular look at the /proc/sys/net/ipv4/netfilter/ip_ct_* variables.

When a connection has seen traffic in both directions, the conntrack entry will erase the [UNREPLIED] flag, and then reset it. The entry that tells us that the connection has not seen any traffic in both directions, will be replaced by the [ASSURED] flag, to be found close to the end of the entry. The [ASSURED] flag tells us that this connection is assured and that it will not be erased if we reach the maximum possible tracked connections. Thus, connections marked as [ASSURED] will not be erased, contrary to the non-assured connections (those not marked as [ASSURED]). How many connections that the connection tracking table can hold depends upon a variable that can be set through the ip-sysctl functions in recent kernels. The default value held by this entry varies heavily depending on how much memory you have. On 128 MB of RAM you will get 8192 possible entries, and at 256 MB of RAM, you will get 16376 entries. You can read and set your settings through the /proc/sys/net/ipv4/ip_conntrack_max setting.

A different way of doing this, that is more efficient, is to set the hashsize option to the ip_conntrack module once this is loaded. Under normal circumstances ip_conntrack_max equals 8 * hashsize. In other words, setting the hashsize to 4096 will result in ip_conntrack_max being set to 32768 conntrack entries. An example of this would be:

work3:/home/blueflux# modprobe ip_conntrack hashsize=4096

work3:/home/blueflux# cat /proc/sys/net/ipv4/ip_conntrack_max

32768

work3:/home/blueflux#

As you have seen, packets may take on several different states within the kernel itself, depending on what protocol we are talking about. However, outside the kernel, we only have the 4 states as described previously. These states can mainly be used in conjunction with the state match which will then be able to match packets based on their current connection tracking state. The valid states are NEW, ESTABLISHED, RELATED and INVALID. The following table will briefly explain each possible state.

Table 7-1. User-land states

These states can be used together with the --state match to match packets based on their connection tracking state. This is what makes the state machine so incredibly strong and efficient for our firewall. Previously, we often had to open up all ports above 1024 to let all traffic back into our local networks again. With the state machine in place this is not necessary any longer, since we can now just open up the firewall for return traffic and not for all kinds of other traffic.

In this section and the upcoming ones, we will take a closer look at the states and how they are handled for each of the three basic protocols TCP, UDP and ICMP. Also, we will take a closer look at how connections are handled per default, if they can not be classified as either of these three protocols. We have chosen to start out with the TCP protocol since it is a stateful protocol in itself, and has a lot of interesting details with regard to the state machine in iptables.

A TCP connection is always initiated with the 3-way handshake, which establishes and negotiates the actual connection over which data will be sent. The whole session is begun with a SYN packet, then a SYN/ACK packet and finally an ACK packet to acknowledge the whole session establishment. At this point the connection is established and able to start sending data. The big problem is, how does connection tracking hook up into this? Quite simply really.