4. IP Firewalling Chains

This section describes all you really need to know to build a packet filter that meets your needs.

4.1 How Packets Traverse The Filters

The kernel starts with three lists of rules; these lists are called firewall chains or just chains. The three chains are called input, output and forward. When a packet comes in (say, through the Ethernet card) the kernel uses the input chain to decide its fate. If it survives that step, then the kernel decides where to send the packet next (this is called routing). If it is destined for another machine, it consults the forward chain. Finally, just before a packet is to go out, the kernel consults the output chain.

A chain is a checklist of rules. Each rule says `if the packet header looks like this, then here's what to do with the packet'. If the rule doesn't match the packet, then the next rule in the chain is consulted. Finally, if there are no more rules to consult, then the kernel looks at the chain policy to decide what to do. In a security-conscious system, this policy usually tells the kernel to reject or deny the packet.

For ASCII-art fans, this shown the complete path of a packet coming into a machine.

        ----------------------------------------------------------------
        |            ACCEPT/                              lo interface |
        v           REDIRECT                  _______                  |
--> C --> S --> ______ --> D --> ~~~~~~~~ -->|forward|----> _______ --> 
    h     a    |input |    e    {Routing }   |Chain  |     |output |ACCEPT
    e     n    |Chain |    m    {Decision}   |_______| --->|Chain  |
    c     i    |______|    a     ~~~~~~~~        |     | ->|_______|
    k     t       |        s       |             |     | |     |    
    s     y       |        q       |             v     | |     |    
    u     |       v        e       v            DENY/  | |     v    
    m     |     DENY/      r   Local Process   REJECT  | |   DENY/
    |     v    REJECT      a       |                   | |  REJECT
    |   DENY               d       --------------------- | 
    v                      e -----------------------------
   DENY                              

Checksum:

This is a test that the packet hasn't been corrupted in some way. If it has, it is denied.

Sanity:

There is actually one of these sanity checks before each firewall chain, but the input chain's is the most important. Some malformed packets might confuse the rule-checking code, and these are denied here (a message is printed to the syslog if this happens).

input chain:

This is the first firewall chain against which the packet will be tested. If the verdict of the chain is not DENY or REJECT, the packet continues on.

Demasquerade:

If the packet is a reply to a previously masqueraded packet, it is demasqueraded, and skips straight to the output chain. If you don't use IP Masquerading, you can mentally erase this from the diagram.

Routing decision:

The destination field is examined by the routing code, to decide if this packet should go to a local process (see Local process below) or forwarded to a remote machine (see forward chain below).

Local process:

A process running on the machine can receive packets after the Routing Decision step, and can send packets (which go through the Routing Decision step, then traverse the output chain).

lo interface:

If packets from a local process are destined for a local process, they will go through the output chain with interface set to `lo', then return through the input chain with interface also `lo'. The lo interface is usually called the loopback interface.

local:

If the packet was not created by a local process, then the forward chain is checked, otherwise the packet goes to the output chain.

forward chain:

This chain is traversed for any packets which are attempting to pass through this machine to another.

output chain:

This chain is traversed for all packets just before they are sent out.

Using ipchains

First, check that you have the version of ipchains that this document refers to:

$ ipchains --version
ipchains 1.3.9, 17-Mar-1999

Note that I recommend 1.3.4 (which has no long options, like `--sport'), or 1.3.8 or above; these are very stable.

ipchains has a fairly detailed manual page (man ipchains), and if you need more detail on particulars, you can check out the programming interface (man 4 ipfw), or the file net/ipv4/ip_fw.c in the 2.1.x kernel source, which is (obviously) authoritative.

There is also an excellent quick reference card by Scott Bronson in the source package, in both A4 and US Letter PostScript(TM).

There are several different things you can do with ipchains. First the operations to manage whole chains. You start with three built-in chains input, output and forward which you can't delete.

1. Create a new chain (-N).
2. Delete an empty chain (-X).
3. Change the policy for a built-in chain. (-P).
4. List the rules in a chain (-L).
5. Flush the rules out of a chain (-F).
6. Zero the packet and byte counters on all rules in a chain (-Z).

There are several ways to manipulate rules inside a chain:

1. Append a new rule to a chain (-A).
2. Insert a new rule at some position in a chain (-I).
3. Replace a rule at some position in a chain (-R).
4. Delete a rule at some position in a chain (-D).
5. Delete the first rule that matches in a chain (-D).

There are a few operations for masquerading, which are in ipchains for want of a good place to put them:

1. List the currently masqueraded connections (-M -L).
2. Set masquerading timeout values (-M -S). (But see I can't set masquerading timeouts!).

The final (and perhaps the most useful) function allows you to check what would happen to a given packet if it were to traverse a given chain.

What You'll See When Your Computer Starts Up

Before any ipchains commands have been run (be careful: some distributions run ipchains in their initialization scripts), there will be no rules in any of the built-in chains (`input', `forward' and `output'), and each of the chains will have a policy of ACCEPT. This is as wide-open as you can get.

Operations on a Single Rule

This is the bread-and-butter of ipchains; manipulating rules. Most commonly, you will probably use the append (-A) and delete (-D) commands. The others (-I for insert and -R for replace) are simple extensions of these concepts.

Each rule specifies a set of conditions the packet must meet, and what to do if it meets them (a `target'). For example, you might want to deny all ICMP packets coming from the IP address 127.0.0.1. So in this case our conditions are that the protocol must be ICMP and that the source address must be 127.0.0.1. Our target is `DENY'.

127.0.0.1 is the `loopback' interface, which you will have even if you have no real network connection. You can use the `ping' program to generate such packets (it simply sends an ICMP type 8 (echo request) which all cooperative hosts should obligingly respond to with an ICMP type 0 (echo reply) packet). This makes it useful for testing.

# ping -c 1 127.0.0.1
PING 127.0.0.1 (127.0.0.1): 56 data bytes
64 bytes from 127.0.0.1: icmp_seq=0 ttl=64 time=0.2 ms

--- 127.0.0.1 ping statistics ---
1 packets transmitted, 1 packets received, 0% packet loss
round-trip min/avg/max = 0.2/0.2/0.2 ms
# ipchains -A input -s 127.0.0.1 -p icmp -j DENY
# ping -c 1 127.0.0.1
PING 127.0.0.1 (127.0.0.1): 56 data bytes

--- 127.0.0.1 ping statistics ---
1 packets transmitted, 0 packets received, 100% packet loss
# 

You can see here that the first ping succeeds (the `-c 1' tells ping to only send a single packet).

Then we append (-A) to the `input' chain, a rule specifying that for packets from 127.0.0.1 (`-s 127.0.0.1') with protocol ICMP (`-p ICMP') we should jump to DENY (`-j DENY').

Then we test our rule, using the second ping. There will be a pause before the program gives up waiting for a response that will never come.

We can delete the rule in one of two ways. Firstly, since we know that it is the only rule in the input chain, we can use a numbered delete, as in:

        # ipchains -D input 1
        #

The second way is to mirror the -A command, but replacing the -A with -D. This is useful when you have a complex chain of rules and you don't want to have to count them to figure out that it's rule 37 that you want to get rid of. In this case, we would use:

        # ipchains -D input -s 127.0.0.1 -p icmp -j DENY
        #

Filtering Specifications

We have seen the use of `-p' to specify protocol, and `-s' to specify source address, but there are other options we can use to specify packet characteristics. What follows is an exhaustive compendium.

Specifying Source and Destination IP Addresses

Source (-s) and destination (-d) IP addresses can be specified in four ways. The most common way is to use the full name, such as `localhost' or `www.linuxhq.com'. The second way is to specify the IP address such as `127.0.0.1'.

The third and fourth ways allow specification of a group of IP addresses, such as `199.95.207.0/24' or `199.95.207.0/255.255.255.0'. These both specify any IP address from 199.95.207.0 to 199.95.207.255 inclusive; the digits after the `/' tell which parts of the IP address are significant. `/32' or `/255.255.255.255' is the default (match all of the IP address). To specify any IP address at all `/0' can be used, like so:

        # ipchains -A input -s 0/0 -j DENY
        #

This is rarely used, as the effect above is the same as not specifying the `-s' option at all.

Specifying Inversion

Many flags, including the `-s' and `-d' flags can have their arguments preceded by `!' (pronounced `not') to match addresses NOT equal to the ones given. For example. `-s ! localhost' matches any packet not coming from localhost.

Don't forget the spaces around the `!': they really are needed.

Specifying Protocol

The protocol can be specified with the `-p' flag. Protocol can be a number (if you know the numeric protocol values for IP) or a name for the special cases of `TCP', `UDP' or `ICMP'. Case doesn't matter, so `tcp' works as well as `TCP'.

The protocol name can be prefixed by a `!', to invert it, such as `-p ! TCP'.

Specifying UDP and TCP Ports

For the special case where a protocol of TCP or UDP is specified, there can be an extra argument indicating the TCP or UDP port, or an (inclusive) range of ports (but see Handling Fragments below). A range is represented using a `:' character, such as `6000:6010', which covers 11 port numbers, from 6000 to 6010 inclusive. If the lower bound is omitted, it defaults to 0. If the upper bound is omitted, it defaults to 65535. So to specify TCP connections coming from ports under 1024, the syntax would be as `-p TCP -s 0.0.0.0/0 :1023'. Port numbers can be specified by name, eg. `www'.

Note that the port specification can be preceded by a `!', which inverts it. So to specify every TCP packet BUT a WWW packet, you would specify

-p TCP -d 0.0.0.0/0 ! www

It is important to realize that the specification

-p TCP -d ! 192.168.1.1 www

is very different from

-p TCP -d 192.168.1.1 ! www

The first specifies any TCP packet to the WWW port on any machine but 192.168.1.1. The second specifies any TCP connection to any port on 192.168.1.1 but the WWW port.

Finally, this case means not the WWW port and not 192.168.1.1:

-p TCP -d ! 192.168.1.1 ! www

Specifying ICMP Type and Code

ICMP also allows an optional argument, but as ICMP doesn't have ports, (ICMP has a type and a code) they have a different meaning.

You can specify them as ICMP names (use ipchains -h icmp to list the names) after the `-s' option, or as a numeric ICMP type and code, where the type follows the `-s' option and the code follows the `-d' option.

The ICMP names are fairly long: you only need use enough letters to make the name distinct from any other.

Here is a small table of some of the most common ICMP packets:

Number  Name                     Required by

0       echo-reply               ping
3       destination-unreachable  Any TCP/UDP traffic.
5       redirect                 routing if not running routing daemon
8       echo-request             ping
11      time-exceeded            traceroute

Note that the ICMP names cannot be preceeded by `!' at the moment.

DO NOT DO NOT DO NOT block all ICMP type 3 messages! (See ICMP Packets below).

Specifying an Interface

The `-i' option specifies the name of an interface to match. An interface is the physical device the packet came in on, or is going out on. You can use the ifconfig command to list the interfaces which are `up' (ie. working at the moment).

The interface for incoming packets (ie. packets traversing the input chain) is considered to be the interface they came in on. Logically, the interface for outgoing packets (packets traversing the output chain) is the interface they will go out on. The interface for packets traversing the forward chain is also the interface they will go out on; a fairly arbitrary decision it seems to me.

It is perfectly legal to specify an interface that currently does not exist; the rule will not match anything until the interface comes up. This is extremely useful for dial-up PPP links (usually interface ppp0) and the like.

As a special case, an interface name ending with a `+' will match all interfaces (whether they currently exist or not) which begin with that string. For example, to specify a rule which matches all PPP interfaces, the -i ppp+ option would be used.

The interface name can be preceded by a `!' to match a packet which does NOT match the specified interface(s).

Specifying TCP SYN Packets Only

It is sometimes useful to allow TCP connections in one direction, but not the other. For example, you might want to allow connections to an external WWW server, but not connections from that server.

The naive approach would be to block TCP packets coming from the server. Unfortunately, TCP connections require packets going in both directions to work at all.

The solution is to block only the packets used to request a connection. These packets are called SYN packets (ok, technically they're packets with the SYN flag set, and the FIN and ACK flags cleared, but we call them SYN packets). By disallowing only these packets, we can stop attempted connections in their tracks.

The `-y' flag is used for this: it is only valid for rules which specify TCP as their protocol. For example, to specify TCP connection attempts from 192.168.1.1:

-p TCP -s 192.168.1.1 -y

Once again, this flag can be inverted by preceding it with a `!', which means every packet other than the connection initiation.

Handling Fragments

Sometimes a packet is too large to fit down a wire all at once. When this happens, the packet is divided into fragments, and sent as multiple packets. The other end reassembles the fragments to reconstruct the whole packet.

The problem with fragments is that some of the specifications listed above (in particular, source port, destinations port, ICMP type, ICMP code, or TCP SYN flag) require the kernel to peek at the start of the packet, which is only contained in the first fragment.

If your machine is the only connection to an external network, then you can tell the Linux kernel to reassemble all fragments which pass through it, by compiling the kernel with IP: always defragment set to `Y'. This sidesteps the issue neatly.

Otherwise, it is important to understand how fragments get treated by the filtering rules. Any filtering rule that asks for information we don't have will not match. This means that the first fragment is treated like any other packet. Second and further fragments won't be. Thus a rule -p TCP -s 192.168.1.1 www (specifying a source port of `www') will never match a fragment (other than the first fragment). Neither will the opposite rule -p TCP -s 192.168.1.1 ! www.

However, you can specify a rule specifically for second and further fragments, using the `-f' flag. Obviously, it is illegal to specify a TCP or UDP port, ICMP type, ICMP code or TCP SYN flag in such a fragment rule.

It is also legal to specify that a rule does not apply to second and further fragments, by preceding the `-f' with `!'.

Usually it is regarded as safe to let second and further fragments through, since filtering will effect the first fragment, and thus prevent reassembly on the target host, however, bugs have been known to allow crashing of machines simply by sending fragments. Your call.

Note for network-heads: malformed packets (TCP, UDP and ICMP packets too short for the firewalling code to read the ports or ICMP code and type) are treated as fragments as well. Only TCP fragments starting at position 8 are explicitly dropped by the firewall code (a message should appear in the syslog if this occurs).

As an example, the following rule will drop any fragments going to 192.168.1.1:

 
# ipchains -A output -f -d 192.168.1.1 -j DENY
#

Filtering Side Effects

OK, so now we know all the ways we can match a packet using a rule. If a packet matches a rule, the following things happen:

1. The byte counter for that rule is increased by the size of the packet (header and all).
2. The packet counter for that rule is incremented.
3. If the rule requests it, the packet is logged.
4. If the rule requests it, the packet's Type Of Service field is changed.
5. If the rule requests it, the packet is marked (not in 2.0 kernel series).
6. The rule target is examined to decide what to do to the packet next.

For variety, I'll address these in order of importance.

Specifying a Target

A target tells the kernel what to do with a packet that matches a rule. ipchains uses `-j' (think `jump-to') for the target specification. The target name must be less than 8 letters, and case matters: "RETURN" and "return" are completely different.

The simplest case is when there is no target specified. This type of rule (often called an `accounting' rule) is useful for simply counting a certain type of packet. Whether this rule matches or not, the kernel simply examines the next rule in the chain. For example, to count the number of packets from 192.168.1.1, we could do this:

# ipchains -A input -s 192.168.1.1
#

(Using `ipchains -L -v' we can see the byte and packet counters associated with each rule).

There are six special targets. The first three, ACCEPT, REJECT and DENY are fairly simple. ACCEPT allows the packet through. DENY drops the packet as if it had never been received. REJECT drops the packet, but (if it's not an ICMP packet) generates an ICMP reply to the source to tell it that the destination was unreachable.

The next one, MASQ tells the kernel to masquerade the packet. For this to work, your kernel needs to be compiled with IP Masquerading enabled. For details on this, see the Masquerading-HOWTO and the Appendix Differences between ipchains and ipfwadm. This target is only valid for packets traversing the forward chain.

The other major special target is REDIRECT which tells the kernel to send a packet to a local port instead of wherever it was heading. This can only be specified for rules specifying TCP or UDP as their protocol. Optionally, a port (name or number) can be specified following `-j REDIRECT' which will cause the packet to be redirected to that particular port, even if it was addressed to another port. This target is only valid for packets traversing the input chain.

The final special target is RETURN which is identical to falling off the end of the chain immediately. (See Setting Policy below).

Any other target indicates a user-defined chain (as described in Operations on an Entire Chain below). The packet will begin traversing the rules in that chain. If that chain doesn't decide the fate of the packet, then once traversal on that chain has finished, traversal resumes on the next rule in the current chain.

Time for more ASCII art. Consider two (silly) chains: input (the built-in chain) and Test (a user-defined chain).

         `input'                         `Test'
        ----------------------------    ----------------------------
        | Rule1: -p ICMP -j REJECT |    | Rule1: -s 192.168.1.1    |
        |--------------------------|    |--------------------------|
        | Rule2: -p TCP -j Test    |    | Rule2: -d 192.168.1.1    |
        |--------------------------|    ----------------------------
        | Rule3: -p UDP -j DENY    |
        ----------------------------

Consider a TCP packet coming from 192.168.1.1, going to 1.2.3.4. It enters the input chain, and gets tested against Rule1 - no match. Rule2 matches, and its target is Test, so the next rule examined is the start of Test. Rule1 in Test matches, but doesn't specify a target, so the next rule is examined, Rule2. This doesn't match, so we have reached the end of the chain. We return to the input chain, where we had just examined Rule2, so we now examine Rule3, which doesn't match either.

So the packet path is:

                                v    __________________________
         `input'                |   /    `Test'                v
        ------------------------|--/    -----------------------|----
        | Rule1                 | /|    | Rule1                |   |
        |-----------------------|/-|    |----------------------|---|
        | Rule2                 /  |    | Rule2                |   |
        |--------------------------|    -----------------------v----
        | Rule3                 /--+___________________________/
        ------------------------|---
                                v

See the section How to Organise Your Firewall Rules for ways to use user-defined chains effectively.

Logging Packets

This is a side effect that matching a rule can have; you can have the matching packet logged using the `-l' flag. You will usually not want this for routine packets, but it is a useful feature if you want to look for exceptional events.

The kernel logs this information looking like:

Packet log: input DENY eth0 PROTO=17 192.168.2.1:53 192.168.1.1:1025
  L=34 S=0x00 I=18 F=0x0000 T=254

This log message is designed to be terse, and contain technical information useful only to networking gurus, but it can be useful to the rest of us. It breaks down like so:

1. `input' is the chain which contained the rule which matched the packet, causing the log message.
2. `DENY' is what the rule said to do to the packet. If this is `-' then the rule didn't effect the packet at all (an accounting rule).
3. `eth0' is the interface name. Because this was the input chain, it means that the packet came in `eth0'.
4. `PROTO=17' means that the packet was protocol 17. A list of protocol numbers is given in `/etc/protocols'. The most common are 1 (ICMP), 6 (TCP) and 17 (UDP).
5. `192.168.2.1' means that the packet's source IP address was 192.168.2.1.
6. `:53' means that the source port was port 53. Looking in `/etc/services' shows that this is the `domain' port (ie. this is probably an DNS reply). For UDP and TCP, this number is the source port. For ICMP, it's the ICMP type. For others, it will be 65535.
7. `192.168.1.1' is the destination IP address.
8. `:1025' means that the destination port was 1025. For UDP and TCP, this number is the destination port. For ICMP, it's the ICMP code. For others, it will be 65535.
9. `L=34' means that packet was a total of 34 bytes long.
10. `S=0x00' means the Type of Service field (divide by 4 to get the Type of Service as used by ipchains).
11. `I=18' is the IP ID.
12. `F=0x0000' is the 16-bit fragment offset plus flags. A value starting with `0x4' or `0x5' means that the Don't Fragment bit is set. `0x2' or `0x3' means the `More Fragments' bit is set; expect more fragments after this. The rest of the number is the offset of this fragment, divided by 8.
13. `T=254' is the Time To Live of the packet. One is subtracted from this value for every hop, and it usually starts at 15 or 255.
14. `(#5)' there may be a final number in brackets on more recent kernels (perhaps after 2.2.9). This is the rule number which caused the packet log.

On standard Linux systems, this kernel output is captured by klogd (the kernel logging daemon) which hands it to syslogd (the system logging daemon). The `/etc/syslog.conf' controls the behaviour of syslogd, by specifying a destination for each `facility' (in our case, the facility is "kernel") and `level' (for ipchains, the level used is "info").

For example, my (Debian) /etc/syslog.conf contains two lines which match `kern.info':

kern.*                          -/var/log/kern.log
*.=info;*.=notice;*.=warn;\
        auth,authpriv.none;\
        cron,daemon.none;\
        mail,news.none          -/var/log/messages

These mean that the messags are duplicated in `/var/log/kern.log' and `/var/log/messages'. For more details, see `man syslog.conf'.

Manipulating the Type Of Service

There are four seldom-used bits in the IP header, called the Type of Service (TOS) bits. They effect the way packets are treated; the four bits are "Minimum Delay", "Maximum Throughput", "Maximum Reliability" and "Minimum Cost". Only one of these bits is allowed to be set. Rob van Nieuwkerk, the author of the TOS-mangling code, puts it as follows:

Especially the "Minimum Delay" is important for me. I switch it on for "interactive" packets in my upstream (Linux) router. I'm behind a 33k6 modem link. Linux prioritizes packets in 3 queues. This way I get acceptable interactive performance while doing bulk downloads at the same time. (It could even be better if there wasn't such a big queue in the serial driver, but latency is kept down 1.5 seconds now).

Note: obviously, you have no control over incoming packets; you can only control the priority of packets leaving your box. To negotiate priorities with the other end, a protocol like RSVP (which I know nothing about, so don't ask me) must be used.

The most common use is to set telnet & ftp control connections to "Minimum Delay" and FTP data to "Maximum Throughput". This would be done as follows:

ipchains -A output -p tcp -d 0.0.0.0/0 telnet -t 0x01 0x10
ipchains -A output -p tcp -d 0.0.0.0/0 ftp -t 0x01 0x10
ipchains -A output -p tcp -s 0.0.0.0/0 ftp-data -t 0x01 0x08

The `-t' flag takes two extra parameters, both in hexadecimal. These allow complex twiddling of the TOS bits: the first mask is ANDed with the packet's current TOS, and then the second mask is XORed with it. If this is too confusing, just use the following table:

TOS Name                Value           Typical Uses

Minimum Delay           0x01 0x10       ftp, telnet
Maximum Throughput      0x01 0x08       ftp-data
Maximum Reliability     0x01 0x04       snmp
Minimum Cost            0x01 0x02       nntp

Andi Kleen goes on to point out the following (mildly edited for posterity):

Maybe it would be useful to add an reference to the txqueuelen parameter of ifconfig to the discussion of TOS bits. The default device queue length is tuned for ethernet cards, on modems it is too long and makes the 3 band scheduler (which queues based on TOS) work suboptimally. It is a good idea to set it to a value between 4-10 on modem or single b channel ISDN links: on bundled devices a longer queue is needed. This is a 2.0 and 2.1 problem, but in 2.1 it is a ifconfig flag (with recent nettools), while in 2.0 it requires source patches in the device drivers to change.

So, to see maximal benifits of TOS manipulation for modem PPP links, do `ifconfig $1 txqueuelen' in your /etc/ppp/ip-up script. The number to use depends on the modem speed and the amount of buffering in the modem; here's Andi setting me straight again:

The best value for a given configuration needs experiment. If the queues are too short on a router then packets will get dropped. Also of course one gets benefits even without TOS rewriting, just that TOS rewriting helps to give the benefits to non cooperating programs (but all standard linux programs are cooperating).

Marking a Packet

This allows complex and powerful interactions with Alexey Kuznetsov's new Quality of Service implementation, as well as the mark-based forwarding in later 2.1 series kernels. More news as it comes to hand. This option is ignored altogether in the 2.0 kernel series.

Operations on an Entire Chain

A very useful feature of ipchains is the ability to group related rules into chains. You can call the chains whatever you want, as long as the names don't clash with the built-in chains (input, output and forward) or the targets (MASQ, REDIRECT, ACCEPT, DENY, REJECT or RETURN). I suggest avoiding upper-case labels entirely, since I may use these for future extensions. The chain name can be up to 8 characters long.

Creating a New Chain

Let's create a new chain. Because I am such an imaginative fellow, I'll call it test.

# ipchains -N test
#

It's that simple. Now you can put rules in it as detailed above.

Deleting a Chain

Deleting a chain is simple as well.

# ipchains -X test
# 

Why `-X'? Well, all the good letters were taken.

There are a couple of restrictions to deleting chains: they must be empty (see Flushing a Chain below) and they must not be the target of any rule. You can't delete any of the three built-in chains.

Flushing a Chain

There is a simple way of emptying all rules out of a chain, using the `-F' command.

        # ipchains -F forward
        # 

If you don't specify a chain, then all chains will be flushed.

Listing a Chain

You can list all the rules in a chain by using the `-L' command.

# ipchains -L input
Chain input (refcnt = 1): (policy ACCEPT)
target     prot opt    source                destination           ports
ACCEPT     icmp -----  anywhere              anywhere              any
# ipchains -L test
Chain test (refcnt = 0):
target     prot opt    source                destination           ports
DENY       icmp -----  localnet/24           anywhere              any
#

The `refcnt' listed for test is the number of rules which have test as their target. This must be zero (and the chain be empty) before this chain can be deleted.

If the chain name is omitted, all chains are listed, even empty ones.

There are three options which can accompany `-L'. The `-n' (numeric) option is very useful as it prevents ipchains from trying to lookup the IP addresses, which (if you are using DNS like most people) will cause large delays if your DNS is not set up properly, or you have filtered out DNS requests. It also causes ports to be printed out as numbers rather than names.

The `-v' options shows you all the details of the rules, such as the the packet and byte counters, the TOS masks, the interface, and the packet mark. Otherwise these values are omitted. For example:

# ipchains -v -L input
Chain input (refcnt = 1): (policy ACCEPT)
 pkts bytes target     prot opt   tosa tosx  ifname    mark        source                destination           ports
   10   840 ACCEPT     icmp ----- 0xFF 0x00  lo                    anywhere              anywhere              any

Note that the packet and byte counters are printed out using the suffixes `K', `M' or `G' for 1000, 1,000,000 and 1,000,000,000 respectively. Using the `-x' (expand numbers) flag as well prints the full numbers, no matter how large they are.

Resetting (Zeroing) Counters

It is useful to be able to reset the counters. This can be done with the `-Z' (zero counters) option. For example:

# ipchains -v -L input
Chain input (refcnt = 1): (policy ACCEPT)
 pkts bytes target     prot opt   tosa tosx  ifname    mark        source                destination           ports
   10   840 ACCEPT     icmp ----- 0xFF 0x00  lo                    anywhere              anywhere              any
# ipchains -Z input
# ipchains -v -L input
Chain input (refcnt = 1): (policy ACCEPT)
 pkts bytes target     prot opt   tosa tosx  ifname    mark        source                destination           ports
    0     0 ACCEPT     icmp ----- 0xFF 0x00  lo                    anywhere              anywhere              any
#

The problem with this approach is that sometimes you need to know the counter values immediately before they are reset. In the above example, some packets could pass through between the `-L' and `-Z' commands. For this reason, you can use the `-L' and `-Z' together, to reset the counters while reading them. Unfortunately, if you do this, you can't operate on a single chain: you have to list and zero all the chains at once.

# ipchains -L -v -Z
Chain input (policy ACCEPT):
 pkts bytes target     prot opt   tosa tosx  ifname    mark        source                destination           ports
   10   840 ACCEPT     icmp ----- 0xFF 0x00  lo                    anywhere              anywhere              any

Chain forward (refcnt = 1): (policy ACCEPT)
Chain output (refcnt = 1): (policy ACCEPT)
Chain test (refcnt = 0):
    0     0 DENY       icmp ----- 0xFF 0x00  ppp0                  localnet/24           anywhere              any
# ipchains -L -v
Chain input (policy ACCEPT):
 pkts bytes target     prot opt   tosa tosx  ifname    mark        source                destination           ports
   10   840 ACCEPT     icmp ----- 0xFF 0x00  lo                    anywhere              anywhere              any

Chain forward (refcnt = 1): (policy ACCEPT)
Chain output (refcnt = 1): (policy ACCEPT)
Chain test (refcnt = 0):
    0     0 DENY       icmp ----- 0xFF 0x00  ppp0                  localnet/24           anywhere              any
#

Setting Policy

We glossed over what happens when a packet hits the end of a built-in chain when we discussed how a packet walks through chains in Specifying a Target above. In this case, the policy of the chain determines the fate of the packet. Only built-in chains (input, output and forward) have policies, because if a packet falls off the end of a user-defined chain, traversal resumes at the previous chain.

The policy can be any of the first four special targets: ACCEPT, DENY, REJECT or MASQ. MASQ is only valid for the `forward' chain.

It is also important to note that a RETURN target in a rule in one of the built-in chains is useful to explicitly target the chain policy when a packet matches a rule.

Operations on Masquerading

There are several parameters you can tweak for IP Masquerading. They are bundled with ipchains because it's not worth writing a separate tool for them (although this will change).

The IP Masquerading command is `-M', and it can be combined with `-L' to list currently masqueraded connections, or `-S' to set the masquerading parameters.

The `-L' command can be accompanied by `-n' (show numbers instead of hostnames and port names) or `-v' (show deltas in sequence numbers for masqueraded connection, just in case you care).

The `-S' command should be followed by three timeout values, each in seconds: for TCP sessions, for TCP sessions after a FIN packet, and for UDP packets. If you don't want to change one of these values, simply give a value of `0'.

The default values are listed in `/usr/src/linux/include/net/ip_masq.h', currently 15 minutes, 2 minutes and 5 minutes respectively.

The most common value to change is the first one, for FTP (see FTP Nightmares below).

Note the problems with setting timeouts listed in I can't set masquerading timeouts!.

Checking a Packet

Sometimes you want to see what happens when a certain packet enters your machine, such as for debugging your firewall chains. ipchains has the `-C' command to allow this, using the exact same routines that the kernel uses to diagnose real packets.

You specify which chain to test the packet on by following the `-C' argument with its name. Whereas the kernel always starts traversing on the input, output or forward chains, you are allowed to begin traversing on any chain for testing purposes.

The details of the `packet' are specified using the same syntax used to specify firewall rules. In particular, a protocol (`-p'), source address (`-s'), destination address (`-d') and interface (`-i') are compulsory. If the protocol is TCP or UDP, then a single source and a single destination port must be specified, and a ICMP type and code must be specified for the ICMP protocol (unless the `-f' flag is specified to indicate a fragment rule, in which case these options are illegal).

If the protocol is TCP (and the `-f' flag is not specified), the `-y' flag may be specified, to indicate that the test packet should have the SYN bit set.

Here is an example of testing a TCP SYN packet from 192.168.1.1 port 60000 to 192.168.1.2 port www, coming in the eth0 interface, entering the `input' chain. (This is a classic incoming WWW connection initiation):

# ipchains -C input -p tcp -y -i eth0 -s 192.168.1.1 60000 -d 192.168.1.2 www
packet accepted
# 

Multiple Rules at Once and Watching What Happens

Sometimes a single command line can result in multiple rules being effected. This is done in two ways. Firstly, if you specify a hostname which resolves (using DNS) to multiple IP addresses, ipchains will act as if you had typed multiple commands with each combination of addresses.

So if the hostname `www.foo.com' resolves to three IP addresses, and the hostname `www.bar.com' resolves to two IP addresses, then the command `ipchains -A input -j reject -s www.bar.com -d www.foo.com' would append six rules to the input chain.

The other way to have ipchains perform multiple actions is to use the bidirectional flag (`-b'). This flag makes ipchains behave as if you had typed the command twice, the second time with the `-s' and `-d' arguments reversed. So, to avoid forwarding either to or from 192.168.1.1, you could do the following:

# ipchains -b -A forward -j reject -s 192.168.1.1
# 

Personally, I don't like the `-b' option much; if you want convenience, see Using ipchains-save below.

The -b option can be used with the insert (`-I'), delete (`-D') (but not the variation which takes a rule number), append (`-A') and check (`-C') commands.

Another useful flag is `-v' (verbose) which prints out exactly what ipchains is doing with your commands. This is useful if you are dealing with commands that may effect multiple rules. For example, here we check the behaviour of fragments between 192.168.1.1 and 192.168.1.2.

# ipchains -v -b -C input -p tcp -f -s 192.168.1.1 -d 192.168.1.2 -i lo
  tcp opt   ---f- tos 0xFF 0x00  via lo    192.168.1.1  -> 192.168.1.2    * ->   *
packet accepted
  tcp opt   ---f- tos 0xFF 0x00  via lo    192.168.1.2  -> 192.168.1.1    * ->   *
packet accepted
# 

4.2 Useful Examples

I have a dialup PPP connection (-i ppp0). I grab news (-p TCP -s news.virtual.net.au nntp) and mail (-p TCP -s mail.virtual.net.au pop-3) every time I dial up. I use Debian's FTP method to update my machine regularly (-p TCP -y -s ftp.debian.org.au ftp-data). I surf the web through my ISP's proxy while this is going on (-p TCP -d proxy.virtual.net.au 8080), but hate the ads from doubleclick.net on the Dilbert Archive (-p TCP -y -d 199.95.207.0/24 and -p TCP -y -d 199.95.208.0/24).

I don't mind people trying to ftp to my machine while I'm online (-p TCP -d $LOCALIP ftp), but don't want anyone outside pretending to have an IP address of my internal network (-s 192.168.1.0/24). This is commonly called IP spoofing, and there is a better way to protect yourself from it in the 2.1.x kernels and above: see How do I set up IP spoof protection?.

This setup is fairly simple, because there are currently no other boxes on my internal network.

I don't want any local process (ie. Netscape, lynx etc.) to connect to doubleclick.net:

# ipchains -A output -d 199.95.207.0/24 -j REJECT
# ipchains -A output -d 199.95.208.0/24 -j REJECT
# 

Now I want to set priorities on various outgoing packets (there isn't much point in doing it on incoming packets). Since I have a fair number of these rules, it makes sense to put them all in a single chain, called ppp-out.

# ipchains -N ppp-out
# ipchains -A output -i ppp0 -j ppp-out
# 

Minimum delay for web traffic & telnet.

# ipchains -A ppp-out -p TCP -d proxy.virtual.net.au 8080 -t 0x01 0x10
# ipchains -A ppp-out -p TCP -d 0.0.0.0/0 telnet -t 0x01 0x10
# 

Low cost for ftp data, nntp, pop-3:

# ipchains -A ppp-out -p TCP -d 0.0.0.0/0 ftp-data -t 0x01 0x02
# ipchains -A ppp-out -p TCP -d 0.0.0.0/0 nntp -t 0x01 0x02
# ipchains -A ppp-out -p TCP -d 0.0.0.0/0 pop-3 -t 0x01 0x02
# 

There are a few restrictions on packets coming in the ppp0 interface: let's create a chain called `ppp-in':

# ipchains -N ppp-in
# ipchains -A input -i ppp0 -j ppp-in
# 

Now, no packets coming in ppp0 should be claiming a source address of 192.168.1.*, so we log and deny them:

# ipchains -A ppp-in -s 192.168.1.0/24 -l -j DENY
#

I allow UDP packets in for DNS (I run a caching nameserver which forwards all requests to 203.29.16.1, so I expect DNS replies from them only), incoming ftp, and return ftp-data only (which should only be going to a port above 1023, and not the X11 ports around 6000).

# ipchains -A ppp-in -p UDP -s 203.29.16.1 -d $LOCALIP dns -j ACCEPT
# ipchains -A ppp-in -p TCP -s 0.0.0.0/0 ftp-data -d $LOCALIP 1024:5999 -j ACCEPT
# ipchains -A ppp-in -p TCP -s 0.0.0.0/0 ftp-data -d $LOCALIP 6010: -j ACCEPT
# ipchains -A ppp-in -p TCP -d $LOCALIP ftp -j ACCEPT
#

I allow TCP reply packets back in

# ipchains -A ppp-in -p TCP ! -y -j ACCEPT
#

Finally, local-to-local packets are OK:

# ipchains -A input -i lo -j ACCEPT
# 

Now, my default policy on the input chain is DENY, so everything else gets dropped:

# ipchains -P input DENY
# 

NOTE: I wouldn't set up my chains in this order, as packets might get through while I'm setting up. Safest is usually to set the policy to DENY first, then insert the rules. Of course, if your rules require DNS lookups to resolve hostnames, you could be in trouble.

Using ipchains-save

Setting up firewall chains just the way you want them, and then trying to remember the commands you used so you can do them next time is a pain.

So, ipchains-save is a script which reads your current chains setup and saves it to a file. For the moment I'll keep you in suspense with regards to what ipchains-restore does.

ipchains-save can save a single chain, or all chains (if no chain name is specified). The only option currently permitted is `-v' which prints the rules (to stderr) as they are saved. The policy of the chain is also saved for input, output and forward chains.

# ipchains-save > my_firewall
Saving `input'.
Saving `output'.
Saving `forward'.
Saving `ppp-in'.
Saving `ppp-out'.
# 

Using ipchains-restore

ipchains-restore restores chains as saved with ipchains-save. It can take two options: `-v' which describes each rule as it is added, and `-f' which forces flushing of user-defined chains if they exist, as described below.

If a user-defined chain is found in the input, ipchains-restore checks if that chain already exists. If it does, then you will be prompted whether the chains should be flushed (cleared of all rules) or whether restoring this chain should be skipped. If you specified `-f' on the command line, you will not be prompted; the chain will be flushed.

For example:

# ipchains-restore < my_firewall
Restoring `input'.
Restoring `output'.
Restoring `forward'.
Restoring `ppp-in'.
Chain `ppp-in' already exists. Skip or flush? [S/f]? s
Skipping `ppp-in'.
Restoring `ppp-out'.
Chain `ppp-out' already exists. Skip or flush? [S/f]? f
Flushing `ppp-out'.
#