ipf(5)

NAME

ipf, ipf.conf, ipf6.conf - IP packet filter rule syntax

DESCRIPTION

A rule file for ipf may have any name or even be stdin.
As ipfstat produces parsable rules as output when displaying the internal kernel filter lists, it is quite plausible to use its output to feed back into ipf. Thus, to
remove all filters on input packets, the following could
be done:

# ipfstat -i | ipf -rf

GRAMMAR

The format used by ipf for construction of filtering rules
can be described using the following grammar in BNF:

filter-rule = [ insert ] action in-out [ options ] [ tos ]

[ ttl ]

[ proto ] ip [ group ].

insert = "@" decnumber .
action = block | "pass" | log | "count" | skip | auth

call .
in-out = "in" | "out" .
options = [ log ] [ tag ] [ "quick" ] [ "on" interface

name [ dup ]

[ froute ] [ replyto ] ] .

tos = "tos" decnumber | "tos" hexnumber .
ttl = "ttl" decnumber .
proto = "proto" protocol .
ip = srcdst [ flags ] [ with withopt ] [ icmp ] [ keep ]
group = [ "head" decnumber ] [ "group" decnumber ] .

block = "block" [ return-icmp[return-code] | "return

rst" ] .
log = "log" [ "body" ] [ "first" ] [ "or-block" ] [ "lev

el" loglevel ] .
tag = "tag" tagid .
skip = "skip" decnumber .
auth = "auth" | "preauth" .
call = "call" [ "now" ] function-name .
dup = "dup-to" interface-name [ ":" ipaddr ] .
froute = "fastroute" | "to" interface-name [ ":" ipaddr

] .
replyto = "reply-to" interface-name [ ":" ipaddr ] .
protocol = "tcp/udp" | "udp" | "tcp" | "icmp" | decnumber
srcdst = "all" | fromto .
fromto = "from" [ "!" ] object "to" [ "!" ] object .

return-icmp = "return-icmp" | "return-icmp-as-dest" .
return-code = "(" icmp-code ")" .
object = addr [ port-comp | port-range ] .
addr = "any" | nummask | host-name [ "mask" ipaddr

"mask" hexnumber ] .
addr = "any" | "<thishost>" | nummask

host-name [ "mask" ipaddr | "mask" hexnumber ] .

port-comp = "port" compare port-num .
port-range = "port" port-num range port-num .
flags = "flags" flag { flag } [ "/" flag { flag } ] .
with = "with" | "and" .
icmp = "icmp-type" icmp-type [ "code" decnumber ] .
return-code = "("icmp-code")" .
keep = "keep" "state" | "keep" "frags" .
loglevel = facility"."priority | priority .

nummask = host-name [ "/" decnumber ] .
host-name = ipaddr | hostname | "any" .
ipaddr = host-num "." host-num "." host-num "." host

num .
host-num = digit [ digit [ digit ] ] .
port-num = service-name | decnumber .

withopt = [ "not" | "no" ] opttype [ withopt ] .
opttype = "ipopts" | "short" | "frag" | "opt" optname .
optname = ipopts [ "," optname ] .
ipopts = optlist | "sec-class" [ secname ] .
secname = seclvl [ "," secname ] .
seclvl = "unclass" | "confid" | "reserv-1" | "reserv-2"

"reserv-3"

"reserv-4" | "secret" | "topsecret" .

icmp-type = "unreach" | "echo" | "echorep" | "squench"

"redir"

"timex" | "paramprob" | "timest" | "timestrep"

"inforeq"
"inforep" | "maskreq" | "maskrep" | decnumber .

icmp-code = decumber | "net-unr" | "host-unr" | "proto

unr" | "port-unr"

"needfrag" | "srcfail" | "net-unk" | "host-unk"

"isolate"
"net-prohib" | "host-prohib" | "net-tos" | "host

tos"
"filter-prohib" | "host-preced" | "cutoff-preced"

optlist = "nop" | "rr" | "zsu" | "mtup" | "mtur" | "en

code" | "ts"

"tr" | "sec" | "lsrr" | "e-sec" | "cipso" | "satid"

| "ssrr"
"addext" | "visa" | "imitd" | "eip" | "finn" .

facility = "kern" | "user" | "mail" | "daemon" | "auth"

"syslog"

"lpr" | "news" | "uucp" | "cron" | "ftp" | "auth

priv"
"audit" | "logalert" | "local0" | "local1" | "lo

cal2"
"local3" | "local4" | "local5" | "local6" | "lo

cal7" .

priority = "emerg" | "alert" | "crit" | "err" | "warn"

"notice"

"info" | "debug" .

hexnumber = "0" "x" hexstring .
hexstring = hexdigit [ hexstring ] .
decnumber = digit [ decnumber ] .

compare = "=" | "!=" | "<" | ">" | "<=" | ">=" | "eq"

"ne" | "lt"

"gt" | "le" | "ge" .

range = "<>" | "><" .
hexdigit = digit | "a" | "b" | "c" | "d" | "e" | "f" .
digit = "0" | "1" | "2" | "3" | "4" | "5" | "6" | "7"

| "8" | "9" .
flag = "F" | "S" | "R" | "P" | "A" | "U" .

This syntax is somewhat simplified for readability, some
combinations that match this grammar are disallowed by the
software because they do not make sense (such as tcp flags
for non-TCP packets).

FILTER RULES

The "briefest" valid rules are (currently) no-ops and are
of the form:

block in all
pass in all
log out all
count in all

Filter rules are checked in order, with the last matching
rule determining the fate of the packet (but see the quick
option, below).

Filters are installed by default at the end of the kernel's filter lists, prepending the rule with @n will cause
it to be inserted as the n'th entry in the current list.
This is especially useful when modifying and testing
active filter rulesets. See ipf(8) for more information.

ACTIONS

The action indicates what to do with the packet if it
matches the rest of the filter rule. Each rule MUST have
an action. The following actions are recognised:

block indicates that the packet should be flagged to be

dropped. In response to blocking a packet, the filter may be instructed to send a reply packet,
either an ICMP packet (return-icmp), an ICMP packet masquerading as being from the original packet's
destination (return-icmp-as-dest), or a TCP "reset" (return-rst). An ICMP packet may be generated in response to any IP packet, and its type may optionally be specified, but a TCP reset may only be used
with a rule which is being applied to TCP packets.
When using return-icmp or return-icmp-as-dest, it is possible to specify the actual unreachable
`type'. That is, whether it is a network unreachable, port unreachable or even administratively
prohibited. This is done by enclosing the ICMP code
associated with it in parenthesis directly following return-icmp or return-icmp-as-dest as follows:

block return-icmp(11) ...

Would return a Type-Of-Service (TOS) ICMP unreachable
error.

pass will flag the packet to be let through the filter.

log causes the packet to be logged (as described in the

LOGGING section below) and has no effect on whether
the packet will be allowed through the filter.

count causes the packet to be included in the accounting

statistics kept by the filter, and has no effect on
whether the packet will be allowed through the filter. These statistics are viewable with ipfstat(8).

call this action is used to invoke the named function in

the kernel, which must conform to a specific calling interface. Customised actions and semantics can
thus be implemented to supplement those available.
This feature is for use by knowledgeable hackers,
and is not currently documented.

skip <n>

causes the filter to skip over the next n filter
rules. If a rule is inserted or deleted inside the
region being skipped over, then the value of n is
adjusted appropriately.

auth this allows authentication to be performed by a

user-space program running and waiting for packet
information to validate. The packet is held for a
period of time in an internal buffer whilst it
waits for the program to return to the kernel the
real flags for whether it should be allowed through
or not. Such a program might look at the source
address and request some sort of authentication
from the user (such as a password) before allowing
the packet through or telling the kernel to drop it
if from an unrecognised source.

preauth

tells the filter that for packets of this class, it
should look in the pre-authenticated list for further clarification. If no further matching rule is
found, the packet will be dropped (the FR_PREAUTH
is not the same as FR_PASS). If a further matching
rule is found, the result from that is used in its
instead. This might be used in a situation where a
person logs in to the firewall and it sets up some temporary rules defining the access for that person.

The next word must be either in or out. Each packet moving through the kernel is either inbound (just been
received on an interface, and moving towards the kernel's
protocol processing) or outbound (transmitted or forwarded
by the stack, and on its way to an interface). There is a
requirement that each filter rule explicitly state which
side of the I/O it is to be used on.

OPTIONS

The list of options is brief, and all are indeed optional.
Where options are used, they must be present in the order
shown here. These are the currently supported options:

log indicates that, should this be the last matching

rule, the packet header will be written to the ipl
log (as described in the LOGGING section below).

tag tagid

indicates that, if this rule causes the packet to
be logged or entered in the state table, the tagid
will be logged as part of the log entry. This can
be used to quickly match "similar" rules in scripts
that post process the log files for e.g. generation
of security reports or accounting purposes. The
tagid is a 32 bit unsigned integer.

quick allows "short-cut" rules in order to speed up the

filter or override later rules. If a packet
matches a filter rule which is marked as quick,
this rule will be the last rule checked, allowing a
"short-circuit" path to avoid processing later
rules for this packet. The current status of the
packet (after any effects of the current rule) will
determine whether it is passed or blocked.

If this option is missing, the rule is taken to be
a "fall-through" rule, meaning that the result of
the match (block/pass) is saved and that processing
will continue to see if there are any more matches.

on allows an interface name to be incorporated into

the matching procedure. Interface names are as
printed by "netstat -i". If this option is used,
the rule will only match if the packet is going
through that interface in the specified direction
(in/out). If this option is absent, the rule is
taken to be applied to a packet regardless of the
interface it is present on (i.e. on all interfaces). Filter rulesets are common to all interfaces, rather than having a filter list for each
interface.

This option is especially useful for simple IPspoofing protection: packets should only be allowed
to pass inbound on the interface from which the
specified source address would be expected, others
may be logged and/or dropped.

dup-to causes the packet to be copied, and the duplicate

packet to be sent outbound on the specified interface, optionally with the destination IP address
changed to that specified. This is useful for offhost logging, using a network sniffer.

to causes the packet to be moved to the outbound queue

on the specified interface. This can be used to
circumvent kernel routing decisions, and even to
bypass the rest of the kernel processing of the
packet (if applied to an inbound rule). It is thus
possible to construct a firewall that behaves
transparently, like a filtering hub or switch,
rather than a router. The fastroute keyword is a synonym for this option.

MATCHING PARAMETERS

The keywords described in this section are used to
describe attributes of the packet to be used when determining whether rules match or don't match. The following
general-purpose attributes are provided for matching, and
must be used in this order:

tos packets with different Type-Of-Service values can

be filtered. Individual service levels or combinations can be filtered upon. The value for the TOS
mask can either be represented as a hex number or a
decimal integer value.

ttl packets may also be selected by their Time-To-Live

value. The value given in the filter rule must
exactly match that in the packet for a match to
occur. This value can only be given as a decimal
integer value.

proto allows a specific protocol to be matched against.

All protocol names found in /etc/protocols are recognised and may be used. However, the protocol
may also be given as a DECIMAL number, allowing for
rules to match your own protocols, or new ones
which would out-date any attempted listing.

The special protocol keyword tcp/udp may be used to match either a TCP or a UDP packet, and has been
added as a convenience to save duplication of otherwise-identical rules.

The from and to keywords are used to match against IP addresses (and optionally port numbers). Rules must specify BOTH source and destination parameters.

IP addresses may be specified in one of two ways: as a
numerical address/mask, or as a hostname mask netmask.
The hostname may either be a valid hostname, from either
the hosts file or DNS (depending on your configuration and
library) or of the dotted numeric form. There is no special designation for networks but network names are recognised. Note that having your filter rules depend on DNS
results can introduce an avenue of attack, and is discouraged.

There is a special case for the hostname any which is
taken to be 0.0.0.0/0 (see below for mask syntax) and
matches all IP addresses. Only the presence of "any" has
an implied mask, in all other situations, a hostname MUST
be accompanied by a mask. It is possible to give "any" a
hostmask, but in the context of this language, it is nonsensical.

The numerical format "x/y" indicates that a mask of y
consecutive 1 bits set is generated, starting with the
MSB, so a y value of 16 would give 0xffff0000. The symbolic "x mask y" indicates that the mask y is in dotted IP
notation or a hexadecimal number of the form 0x12345678.
Note that all the bits of the IP address indicated by the
bitmask must match the address on the packet exactly;
there isn't currently a way to invert the sense of the
match, or to match ranges of IP addresses which do not
express themselves easily as bitmasks (anthropomorphization; it's not just for breakfast anymore).

If a port match is included, for either or both of source
and destination, then it is only applied to TCP and UDP
packets. If there is no proto match parameter, packets
from both protocols are compared. This is equivalent to
"proto tcp/udp". When composing port comparisons, either
the service name or an integer port number may be used.
Port comparisons may be done in a number of forms, with a
number of comparison operators, or port ranges may be
specified. When the port appears as part of the from
object, it matches the source port number, when it appears
as part of the to object, it matches the destination port
number. See the examples for more information.

The all keyword is essentially a synonym for "from any to
any" with no other match parameters.

Following the source and destination matching parameters,
the following additional parameters may be used:

with is used to match irregular attributes that some

packets may have associated with them. To match
the presence of IP options in general, use with
ipopts. To match packets that are too short to contain a complete header, use with short. To match fragmented packets, use with frag. For more specific filtering on IP options, individual options
can be listed.

Before any parameter used after the with keyword,
the word not or no may be inserted to cause the
filter rule to only match if the option(s) is not
present.

Multiple consecutive with clauses are allowed.
Alternatively, the keyword and may be used in place
of with, this is provided purely to make the rules
more readable ("with ... and ..."). When multiple
clauses are listed, all those must match to cause a
match of the rule.

flags is only effective for TCP filtering. Each of the

letters possible represents one of the possible
flags that can be set in the TCP header. The association is as follows:

F - FIN
S - SYN
R - RST
P - PUSH
A - ACK
U - URG

The various flag symbols may be used in combination, so that "SA" would represent a SYN-ACK combination present in a packet. There is nothing
preventing the specification of combinations, such
as "SFR", that would not normally be generated by
law-abiding TCP implementations. However, to guard
against weird aberrations, it is necessary to state
which flags you are filtering against. To allow
this, it is possible to set a mask indicating which
TCP flags you wish to compare (i.e., those you deem
significant). This is done by appending "/<flags>"
to the set of TCP flags you wish to match against,
e.g.:

... flags S

# becomes "flags S/AUPRFS" and will match
# packets with ONLY the SYN flag set.

... flags SA

# becomes "flags SA/AUPRFS" and will match

any
# packet with only the SYN and ACK flags

set.

... flags S/SA

# will match any packet with just the SYN

flag set
# out of the SYN-ACK pair; the common "es

tablish"
# keyword action. "S/SA" will NOT match a

packet
# with BOTH SYN and ACK set, but WILL match

"SFP".

icmp-type

is only effective when used with proto icmp and must NOT be used in conjunction with flags. There
are a number of types, which can be referred to by
an abbreviation recognised by this language, or the
numbers with which they are associated can be used.
The most important from a security point of view is
the ICMP redirect.

KEEP HISTORY

The second last parameter which can be set for a filter
rule is whether or not to record historical information
for that packet, and what sort to keep. The following
information can be kept:

state keeps information about the flow of a communication

session. State can be kept for TCP, UDP, and ICMP
packets.

frags keeps information on fragmented packets, to be

applied to later fragments.

allowing packets which match these to flow straight
through, rather than going through the access control
list.

GROUPS

The last pair of parameters control filter rule "grouping". By default, all filter rules are placed in group 0
if no other group is specified. To add a rule to a nondefault group, the group must first be started by creating
a group head. If a packet matches a rule which is the
head of a group, the filter processing then switches to
the group, using that rule as the default for the group.
If quick is used with a head rule, rule processing isn't stopped until it has returned from processing the group.

A rule may be both the head for a new group and a member
of a non-default group (head and group may be used together in a rule).

head <n>

indicates that a new group (number n) should be
created.

group <n>

indicates that the rule should be put in group
(number n) rather than group 0.

LOGGING

When a packet is logged, with either the log action or
option, the headers of the packet are written to the ipl
packet logging pseudo-device. Immediately following the
log keyword, the following qualifiers may be used (in
order):

body indicates that the first 128 bytes of the packet

contents will be logged after the headers.

first If log is being used in conjunction with a "keep"

option, it is recommended that this option is also
applied so that only the triggering packet is
logged and not every packet which thereafter
matches state information.

or-block

indicates that, if for some reason the filter is
unable to log the packet (such as the log reader
being too slow) then the rule should be interpreted
as if the action was block for this packet.

level <loglevel>

indicates what logging facility and priority, or
just priority with the default facility being used,
will be used to log information about this packet
using ipmon's -s option.

See ipl(4) for the format of records written to this
device. The ipmon(8) program can be used to read and format this log.

EXAMPLES

The quick option is good for rules such as:

block in quick from any to any with ipopts

which will match any packet with a non-standard header
length (IP options present) and abort further processing
of later rules, recording a match and also that the packet
should be blocked.

The "fall-through" rule parsing allows for effects such as
this:

block in from any to any port < 6000
pass in from any to any port >= 6000
block in from any to any port > 6003

which sets up the range 6000-6003 as being permitted and
all others being denied. Note that the effect of the
first rule is overridden by subsequent rules. Another
(easier) way to do the same is:

block in from any to any port 6000 <> 6003
pass in from any to any port 5999 >< 6004

Note that both the "block" and "pass" are needed here to
effect a result as a failed match on the "block" action
does not imply a pass, only that the rule hasn't taken
effect. To then allow ports < 1024, a rule such as:

pass in quick from any to any port < 1024

would be needed before the first block. To create a new
group for processing all inbound packets on le0/le1/lo0,
with the default being to block all inbound packets, we
would do something like:

block in all
block in quick on le0 all head 100
block in quick on le1 all head 200
block in quick on lo0 all head 300

and to then allow ICMP packets in on le0, only, we would
do:

pass in proto icmp all group 100

Note that because only inbound packets on le0 are used
processed by group 100, there is no need to respecify the
interface name. Likewise, we could further breakup processing of TCP, etc, as follows:

block in proto tcp all head 110 group 100
pass in from any to any port = 23 group 110

and so on. The last line, if written without the groups
would be:

pass in on le0 proto tcp from any to any port =

telnet

Note, that if we wanted to say "port = telnet", "proto
tcp" would need to be specified as the parser interprets
each rule on its own and qualifies all service/port names
with the protocol specified.

FILES

/dev/ipauth
/dev/ipl
/dev/ipstate
/etc/hosts
/etc/services

SEE ALSO

ipftest(1), iptest(1), mkfilters(1), ipf(4), ipnat(5),
ipf(8), ipfstat(8)

IPF(5)