oracular (5) filter_syntax.5.gz
NAME
rules.filter - Input format for filtergen packet filter compiler
INTRO
This file describes the input syntax accepted by filtergen(8).
BASICS
In general form, a filter rule is described by a direction, an interface, a target and (possibly empty)
sets of matches and options.
Simple rules will look like:
direction interface match0 .. matchN target;
for example:
input eth0 source host1 dest host2 proto tcp dport http accept;
Note that the elements of the rule can be placed in any order, with the exception that the interface must
immediately follow the direction. Thus, this rule is equivalent to the above (though perhaps less
readable):
proto tcp source host1 dport http accept dest host2 input eth0;
The semicolon separates rules. It is optional before a closing brace or the end of a file. Whitespace
is not significant. Anything after a hash ("#") on a line is ignored.
DIRECTION
A direction merely specifies whether to match packets being sent or received. The only two directions
available are "input" and "output". Forwarded packets will pass through both,
INTERFACE
This specifies which real or virtual network device to filter. As far as filtergen is concerned, this is
just a text string. It must be the same as the device name on the target system. Common names on Linux
are "eth0", "eth1", ..., "ppp0", etc. Other systems will have different naming rules.
If you want to match all interfaces, you can specify "*" (without the quotes).
TARGET
A target notes what we do with a matching packet. Universal options are accept and drop which,
respectively, state that the packet should be allowed as normal, or thrown away. Some backends support
reject to throw away a packet, but send notification to the sender that it was denied, masq (on output
rules only) to "masquerade" a packet - alter it so that it appears to come from the address of the
sending interface - and proxy (and its deprecated alias redirect) to divert a connection via the local
system.
MATCHES
The matches are the meat of the rule. They apply a set of tests to a packet and decide if this rule will
be used to process it. Available matches are:
source addr-range dest addr-range
proto {tcp|udp|icmp|...} sport port-range dport port-range icmptype icmp-type
Matches can be negated by prefixing them with a "!":
input eth0 ! dest 10.0.0.3 reject;
(note than not all backends can support this).
OPTIONS
Options affect the behaviour of the matcher or the target. Currently implemented are log, which logs
packets, local, which means only to check packets to or from this interface, forward which means the
opposite of local, and oneway which causes the backend to omit rules which permit return packets.
The log option can optionally specify a message to log matching packets with, where the backend supports
it:
input eth0 source {
10.0.0.0/8 192.168.0.0/16
} log text "private addresses" drop;
Note that the oneway option make have no effect when used with the -l command-line flag on backends which
support it.
GROUPING
Because it can get tedious to type:
input eth0 source foo dest bar proto tcp dport http accept;
input eth0 source foo dest bar proto tcp dport https accept;
input eth0 source foo dest bar proto tcp dport nntp accept;
input eth0 source foo dest bar proto tcp sport 1:1023 dport ssh accept;
...
filter allows you to group rules with a set syntax:
input eth0 source foo dest bar proto tcp {
dport http;
dport https;
dport nntp;
sport 1:1023 dport ssh;
} accept;
Matches which accept arguments can also be grouped:
input eth0 source foo dest bar proto tcp {
dport {http https nntp};
sport 1:1023 dport ssh;
} accept;
OUT-OF-LINE GROUPS
It is commonly the case that both hosts and routers have long lists of similar looking rules to allow
traffic between groups of hosts, as above. What if we had another pair of hosts which needed a variety
of services? We could simply put the rule groups one after the other:
input eth0 source foo dest bar proto tcp {
dport {http https nntp};
sport 1:1023 dport ssh;
} accept;
input eth0 source baz dest quux proto tcp {
dport {1264 1521 1984 8008 8080 26000};
} accept;
The above generates 11 rules, and every additional port adds another rule through which packets will pass
(well, ones which don't match any of the above). The first four output rules have the same source and
destination hosts and protocol, and we know that if it doesn't match those on the first rule, it won't on
the next three, either. Out-of-line groups use this fact to streamline things somewhat:
input eth0 source foo dest bar [
proto tcp {
dport {http https nntp};
sport 1:1023 dport ssh;
} accept;
];
input eth0 source baz dest quux [
proto tcp { dport {1264 1521 1984 8008 8080 26000}; } accept;
];
Where the underlying system supports it, everything inside the square brackets is moved into a separate
"chain" (in ipchains and iptables-speak) or "group" (in ipfilter-speak). Thus, any packet not matching
"source foo dest bar" or "source baz dest quux" above will be checked against only two rules, not eleven.
Note that matches which must appear together, like "proto tcp" and "sport 12345" need to be either both
in the group, or both out of it.
INCLUDING OTHER FILES
You can, if necessary, include other files containing filtergen statements in your filter files, with the
include directive. This is legal anywhere in the file, and acts as if you literally read the included
file into the position where the include is placed. This allows you to, for example, specify a common
set of addresses (say, your monitoring system) and reference them in multiple locations:
input eth0 {
proto tcp dport ssh source {
include monitoring-hosts.acl
} accept;
...
# NRPE
proto tcp dport 5666 source {
include monitoring-hosts.acl
} accept;
}
output eth0 {
proto tcp sport ssh dest {
include monitoring-hosts.acl
} accept;
...
# NRPE
proto tcp sport 5666 source {
include monitoring-hosts.acl
} accept;
}
Whilst you could improve this particular example without an include by grouping your ssh and NRPE ports
together, you'd still have two places to edit when you changed your set of monitoring hosts. Using
include, you can have a single place to change when you change your monitoring hosts.
You can also include a glob of files, rather than a single file, by using one of the shell globbing
metacharacters '*', '?', or '['. This will cause filtergen to expand that glob and parse the files as if
you had included them one-by-one, in your locale's sorted order. This is very handy if you configure
your systems with an automated system, because you can dump a series of files into a directory depending
on what classes are defined, and filtergen will pick them all up. The sorting is to ensure that your
rules appear in the order you want them, rather than some higgledy-piggledy mess. Note that any
directories that match your glob will not be recursively included in this mode.
Finally, you can just provide a directory name, and filtergen will include all the files and directories
under there, recursively (excluding hidden files, starting with a '.'). This method of including files
is strongly deprecated. There is no defined sort order, and if your editor leaves backup or temporary
files around, or your configuration management system puts it's backups in the same directory as the
original file (cfengine, I'm looking at you), you can find yourself including files you really didn't
intend -- hence why it is a bad idea. Instead, use a glob (above) with an explicit extension (we like
*.fg) to perform your directory inclusions with minimal chance of accidents.
EXAMPLE
Here's a fairly complete example, for a single-interface machine:
#
# Example filter for (for example) a mail server
#
# Unfortunately, we don't have time to audit the
# communications which go on locally
{input lo; output lo} accept;
# But we want to be a bit more careful when speaking
# to the outside world
input eth0 {
# Sadly, we share a DMZ with Windows machines.
# Don't log their netbios noise
proto {tcp udp} source ournet/24 dport 137:139 drop;
proto tcp {
dport { smtp pop-3 } accept;
dport ssh source ournet/24 accept;
# We don't answer this, but don't want to
# cause timeouts by blocking it
dport auth reject;
log drop;
};
# We don't run any UDP (or other non-TCP)
# services
log drop;
};
output eth0 {
proto tcp {
dport { smtp auth } accept;
log drop;
};
# Outbound DNS is OK
proto udp dport domain dest { ns0 ns1 } accept;
log drop;
};
SEE ALSO
filtergen(8), filter_backends(7) January 7, 2004 FILTER SYNTAX(5)