To view license terms, attribution, and copyright for IP Filter, the
default path is /usr/lib/ipf/IPFILTER.LICENCE. If the Illumos operating
environment has been installed anywhere other than the default, modify the
given path to access the file at the installed location.
Portions Copyright (c) 2015, Joyent, Inc.
ipmap :: = mapblock | redir | map . map ::= mapit ifname lhs "->" dstipmask [ mapicmp | mapport | mapproxy ] mapoptions . mapblock ::= "map-block" ifname lhs "->" ipmask [ ports ] mapoptions . redir ::= "rdr" ifname rlhs "->" ip [ "," ip ] rdrport rdroptions . lhs ::= ipmask | fromto . rlhs ::= ipmask dport | fromto . dport ::= "port" portnum [ "-" portnum ] . ports ::= "ports" numports | "auto" . rdrport ::= "port" portnum . mapit ::= "map" | "bimap" . fromto ::= "from" object "to" object . ipmask ::= ip "/" bits | ip "/" mask | ip "netmask" mask . dstipmask ::= ipmask | "range" ip "-" ip . mapicmp ::= "icmpidmap" "icmp" number ":" number . mapport ::= "portmap" tcpudp portspec . mapoptions ::= [ tcpudp ] [ "frag" ] [ age ] [ clamp ] . rdroptions ::= rdrproto [ rr ] [ "frag" ] [ age ] [ clamp ] [ rdrproxy ] . object :: = addr [ port-comp | port-range ] . addr :: = "any" | nummask | host-name [ "mask" ipaddr | "mask" hexnumber ] . port-comp :: = "port" compare port-num . port-range :: = "port" port-num range port-num . rdrproto ::= tcpudp | protocol . rr ::= "round-robin" . age ::= "age" decnumber [ "/" decnumber ] . clamp ::= "mssclamp" decnumber . tcpudp ::= "tcp/udp" | protocol . mapproxy ::= "proxy" "port" port proxy-name '/' protocol rdrproxy ::= "proxy" proxy-name . protocol ::= protocol-name | decnumber . nummask ::= host-name [ "/" decnumber ] . portspec ::= "auto" | portnumber ":" portnumber . port ::= portnumber | port-name . portnumber ::= number { numbers } . ifname ::= 'A' - 'Z' { 'A' - 'Z' } numbers . numbers ::= '0' | '1' | '2' | '3' | '4' | '5' | '6' | '7' | '8' | '9' .
For standard NAT functionality, a rule should start with map and then proceeds to specify the interface for which outgoing packets will have their source address rewritten.
Packets which will be rewritten can only be selected by matching the original source address. A netmask must be specified with the IP address.
The address selected for replacing the original is chosen from an IP#/netmask pair. A netmask of all 1's indicating a hostname is valid. A netmask of 31 1's (255.255.255.254) is considered invalid as there is no space for allocating host IP#'s after consideration for broadcast and network addresses.
When remapping TCP and UDP packets, it is also possible to change the source port number. Either TCP or UDP or both can be selected by each rule, with a range of port numbers to remap into given as port-number:port-number.
map that is used for mapping one address or network to another in an unregulated round robin fashion;
rdr that is used for redirecting packets to one IP address and port pair to another;
bimap for setting up bidirectional NAT between an external IP address and an internal IP address and
map-block which sets up static IP address based translation, based on a algorithm to squeeze the addresses to be translated into the destination range.
For basic NAT and redirection of packets, the address subject to change is used along with its protocol to check if a packet should be altered. The packet matching part of the rule is to the left of the "->" in each rule.
Matching of packets has now been extended to allow more complex compares. In place of the address which is to be translated, an IP address and port number comparison can be made using the same expressions available with ipf. A simple NAT rule could be written as:
map de0 10.1.0.0/16 -> 201.2.3.4/32
or as
map de0 from 10.1.0.0/16 to any -> 201.2.3.4/32
Only IP address and port numbers can be compared against. This is available with all NAT rules.
To the right of the "->" is the address and port specification which will be written into the packet providing it has already successfully matched the prior constraints. The case of redirections (rdr) is the simplest: the new destination address is that specified in the rule. For map rules, the destination address will be one for which the tuple combining the new source and destination is known to be unique. If the packet is either a TCP or UDP packet, the destination and source ports come into the equation too. If the tuple already exists, IP Filter will increment the port number first, within the available range specified with portmap and if there exists no unique tuple, the source address will be incremented within the specified netmask. If a unique tuple cannot be determined, then the packet will not be translated. The map-block is more limited in how it searches for a new, free and unique tuple, in that it will used an algorithm to determine what the new source address should be, along with the range of available ports - the IP address is never changed and nor does the port number ever exceed its allotted range.
ICMP messages can be divided into two groups: "errors" and "queries". ICMP errors are generated as a response of another IP packet. IP Filter will take care that ICMP errors that are the response of a NAT-ed IP packet are handled properly.
For 4 types of ICMP queries (echo request, timestamp request, information request and address mask request) IP Filter supports an additional mapping called "ICMP id mapping". All these 4 types of ICMP queries use a unique identifier called the ICMP id. This id is set by the process sending the ICMP query and it is usually equal to the process id. The receiver of the ICMP query will use the same id in its response, thus enabling the sender to recognize that the incoming ICMP reply is intended for him and is an answer to a query that he made. The "ICMP id mapping" feature modifies these ICMP id in a way identical to portmap for TCP or UDP.
The reason that you might want this, is that using this feature you don't need an IP address per host behind the NAT box, that wants to do ICMP queries. The two numbers behind the icmpidmap keyword are the first and the last icmp id number that can be used. There is one important caveat: if you map to an IP address that belongs to the NAT box itself (notably if you have only a single public IP address), then you must ensure that the NAT box does not use the icmpidmap range that you specified in the map rule.
IP Filter comes with a few, simple, proxies built into the code that is loaded into the kernel to allow secondary channels to be opened without forcing the packets through a user program. The current state of the proxies is listed below, as one of three states:
Aging - protocol is roughly understood from the time at which the proxy was written but it is not well tested or maintained;
Developmental - basic functionality exists, works most of the time but may be problematic in extended real use;
Experimental - rough support for the protocol at best, may or may not work as testing has been at best sporadic, possible large scale changes to the code in order to properly support the protocol.
Mature - well tested, protocol is properly understood by the proxy;
The currently compiled in proxy list is as follows:
FTP - Mature
IRC - Experimental
rpcbind - Experimental
H.323 - Experimental
Real Audio (PNA) - Aging
IPsec - Developmental
netbios - Experimental
R-command - Mature
True transparent proxying should be performed using the redirect (rdr) rules directing ports to localhost (127.0.0.1) with the proxy program doing a lookup through /dev/ipnat to determine the real source and address of the connection.
Two options for use with rdr are available to support primitive, round-robin based load balancing. The first option allows for a rdr to specify a second destination, as follows:
rdr le0 203.1.2.3/32 port 80 -> 203.1.2.3,203.1.2.4 port 80 tcp
This would send alternate connections to either 203.1.2.3 or 203.1.2.4. In scenarios where the load is being spread amongst a larger set of servers, you can use:
rdr le0 203.1.2.3/32 port 80 -> 203.1.2.3,203.1.2.4 port 80 tcp round-robin rdr le0 203.1.2.3/32 port 80 -> 203.1.2.5 port 80 tcp round-robin
In this case, a connection will be redirected to 203.1.2.3, then 203.1.2.4 and then 203.1.2.5 before going back to 203.1.2.3. In accomplishing this, the rule is removed from the top of the list and added to the end, automatically, as required. This will not effect the display of rules using "ipnat -l", only the internal application order.
This section deals with the map command and its variations.
To change IP#'s used internally from network 10 into an ISP provided 8 bit subnet at 209.1.2.0 through the ppp0 interface, the following would be used:
map ppp0 10.0.0.0/8 -> 209.1.2.0/24
The obvious problem here is we're trying to squeeze over 16,000,000 IP addresses into a 254 address space. To increase the scope, remapping for TCP and/or UDP, port remapping can be used;
map ppp0 10.0.0.0/8 -> 209.1.2.0/24 portmap tcp/udp 1025:65000
which falls only 527,566 `addresses' short of the space available in network 10. If we were to combine these rules, they would need to be specified as follows:
map ppp0 10.0.0.0/8 -> 209.1.2.0/24 portmap tcp/udp 1025:65000 map ppp0 10.0.0.0/8 -> 209.1.2.0/24
so that all TCP/UDP packets were port mapped and only other protocols, such as ICMP, only have their IP# changed. In some instances, it is more appropriate to use the keyword auto in place of an actual range of port numbers if you want to guarantee simultaneous access to all within the given range. However, in the above case, it would default to 1 port per IP address, since we need to squeeze 24 bits of address space into 8. A good example of how this is used might be:
map ppp0 172.192.0.0/16 -> 209.1.2.0/24 portmap tcp/udp auto
which would result in each IP address being given a small range of ports to use (252). In all cases, the new port number that is used is deterministic. That is, port X will always map to port Y. WARNING: It is not advisable to use the auto feature if you are map'ing to a /32 (i.e. 0/32) because the NAT code will try to map multiple hosts to the same port number, outgoing and ultimately this will only succeed for one of them. The problem here is that the map directive tells the NAT code to use the next address/port pair available for an outgoing connection, resulting in no easily discernible relation between external addresses/ports and internal ones. This is overcome by using map-block as follows:
map-block ppp0 172.192.0.0/16 -> 209.1.2.0/24 ports auto
For example, this would result in 172.192.0.0/24 being mapped to 209.1.2.0/32 with each address, from 172.192.0.0 to 172.192.0.255 having 252 ports of its own. As opposed to the above use of map, if for some reason the user of (say) 172.192.0.2 wanted 260 simultaneous connections going out, they would be limited to 252 with map-block but would just move on to the next IP address with the map command. /dev/ipnat