| ROUTED(8) | System Manager's Manual | ROUTED(8) |
| routed | [-sqdghmAtv] [-T tracefile] [-F net[/mask[,metric]]] [-P parms] |
It listens on the udp(4) socket for the route(8) service (see services(5)) for Routing Information Protocol packets. It also sends and receives multicast Router Discovery ICMP messages. If the host is a router, routed periodically supplies copies of its routing tables to any directly connected hosts and networks. It also advertises or solicits default routes using Router Discovery ICMP messages.
When started (or when a network interface is later turned on), routed uses an AF_ROUTE address family facility to find those directly connected interfaces configured into the system and marked "up". It adds necessary routes for the interfaces to the kernel routing table. Soon after being first started, and provided there is at least one interface on which RIP has not been disabled, routed deletes all pre-existing non-static routes in kernel table. Static routes in the kernel table are preserved and included in RIP responses if they have a valid RIP metric (see route(8)).
If more than one interface is present (not counting the loopback interface), it is assumed that the host should forward packets among the connected networks. After transmitting a RIP request and Router Discovery Advertisements or Solicitations on a new interface, the daemon enters a loop, listening for RIP request and response and Router Discovery packets from other hosts.
When a request packet is received, routed formulates a reply based on the information maintained in its internal tables. The response packet generated contains a list of known routes, each marked with a "hop count" metric (a count of 16 or greater is considered "infinite"). The advertised metric for a route reflects the metrics associated with interfaces (see ifconfig(8)) through which it is received and sent, so setting the metric on an interface is an effective way to steer traffic. See also adj_inmetric and adj_outmetric parameters below.
Responses do not include routes with a first hop on the requesting network to implement in part split-horizon. Requests from query programs such as rtquery(8) are answered with the complete table.
The routing table maintained by the daemon includes space for several gateways for each destination to speed recovery from a failing router. RIP response packets received are used to update the routing tables provided they are from one of the several currently recognized gateways or advertise a better metric than at least one of the existing gateways.
When an update is applied, routed records the change in its own tables and updates the kernel routing table if the best route to the destination changes. The change in the kernel routing table is reflected in the next batch of response packets sent. If the next response is not scheduled for a while, a flash update response containing only recently changed routes is sent.
In addition to processing incoming packets, routed also periodically checks the routing table entries. If an entry has not been updated for 3 minutes, the entry's metric is set to infinity and marked for deletion. Deletions are delayed until the route has been advertised with an infinite metric to ensure the invalidation is propagated throughout the local internet. This is a form of poison reverse.
Routes in the kernel table that are added or changed as a result of ICMP Redirect messages are deleted after a while to minimize black-holes. When a TCP connection suffers a timeout, the kernel tells routed, which deletes all redirected routes through the gateway involved, advances the age of all RIP routes through the gateway to allow an alternate to be chosen, and advances of the age of any relevant Router Discovery Protocol default routes.
Hosts acting as internetwork routers gratuitously supply their routing tables every 30 seconds to all directly connected hosts and networks. These RIP responses are sent to the broadcast address on nets that support broadcasting, to the destination address on point-to-point links, and to the router's own address on other networks. If RIPv2 is enabled, multicast packets are sent on interfaces that support multicasting.
If no response is received on a remote interface, if there are errors while sending responses, or if there are more errors than input or output (see netstat(1)), then the cable or some other part of the interface is assumed to be disconnected or broken, and routes are adjusted appropriately.
The Internet Router Discovery Protocol is handled similarly. When the daemon is supplying RIP routes, it also listens for Router Discovery Solicitations and sends Advertisements. When it is quiet and listening to other RIP routers, it sends Solicitations and listens for Advertisements. If it receives a good Advertisement and it is not multi-homed, it stops listening for broadcast or multicast RIP responses. It tracks several advertising routers to speed recovery when the currently chosen router dies. If all discovered routers disappear, the daemon resumes listening to RIP responses. It continues listening to RIP while using Router Discovery if multi-homed to ensure all interfaces are used.
The Router Discovery standard requires that advertisements have a default "lifetime" of 30 minutes. That means should something happen, a client can be without a good route for 30 minutes. It is a good idea to reduce the default to 45 seconds using -P rdisc_interval=45 on the command line or rdisc_interval=45 in the /etc/gateways file.
While using Router Discovery (which happens by default when the system has a single network interface and a Router Discover Advertisement is received), there is a single default route and a variable number of redirected host routes in the kernel table. On a host with more than one network interface, this default route will be via only one of the interfaces. Thus, multi-homed hosts running with -q might need no_rdisc described below.
See the pm_rdisc facility described below to support "legacy" systems that can handle neither RIPv2 nor Router Discovery.
By default, neither Router Discovery advertisements nor solicitations are sent over point to point links (e.g. PPP). The netmask associated with point-to-point links (such as SLIP or PPP, with the IFF_POINTOPOINT flag) is used by routed to infer the netmask used by the remote system when RIPv1 is used.
The following options are available:
Any other argument supplied is interpreted as the name of a file in which the actions of routed should be logged. It is better to use -T instead of appending the name of the trace file to the command.
routed also supports the notion of "distant" passive or active gateways. When routed is started, it reads the file /etc/gateways to find such distant gateways which may not be located using only information from a routing socket, to discover if some of the local gateways are passive, and to obtain other parameters. Gateways specified in this manner should be marked passive if they are not expected to exchange routing information, while gateways marked active should be willing to exchange RIP packets. Routes through passive gateways are installed in the kernel's routing tables once upon startup and are not included in transmitted RIP responses.
Distant active gateways are treated like network interfaces. RIP responses are sent to the distant active gateway. If no responses are received, the associated route is deleted from the kernel table and RIP responses advertised via other interfaces. If the distant gateway resumes sending RIP responses, the associated route is restored.
Such gateways can be useful on media that do not support broadcasts or multicasts but otherwise act like classic shared media like Ethernets such as some ATM networks. One can list all RIP routers reachable on the HIPPI or ATM network in /etc/gateways with a series of "host" lines. Note that it is usually desirable to use RIPv2 in such situations to avoid generating lists of inferred host routes.
Gateways marked external are also passive, but are not placed in the kernel routing table nor are they included in routing updates. The function of external entries is to indicate that another routing process will install such a route if necessary, and that other routes to that destination should not be installed by routed. Such entries are only required when both routers may learn of routes to the same destination.
The /etc/gateways file is comprised of a series of lines, each in one of the following two formats or consist of parameters described later. Blank lines and lines starting with '#' are comments.
Nname or Hname is the name of the destination network or host. It may be a symbolic network name or an Internet address specified in "dot" notation (see inet(3)). (If it is a name, then it must either be defined in /etc/networks or /etc/hosts, or named(8), must have been started before routed.)
Mask is an optional number between 1 and 32 indicating the netmask associated with Nname.
Gname is the name or address of the gateway to which RIP responses should be forwarded.
Value is the hop count to the destination host or network. Host hname is equivalent to net nname/32 .
One of the keywords passive, active or external must be present to indicate whether the gateway should be treated as passive or active (as described above), or whether the gateway is external to the scope of the RIP protocol.
As can be seen when debugging is turned on with -t, such lines create pseudo-interfaces. To set parameters for remote or external interfaces, a line starting with if=alias(Hname), if=remote(Hname), etc. should be used.
Do not use this feature unless necessary. It is dangerous.
Note that turning off RIP without explicitly turning on router discovery advertisements with rdisc_adv or -s causes routed to act as a client router discovery daemon, not advertising.
Internet Transport Protocols, XSIS 028112, Xerox System Integration Standard.
| May 17, 2004 | NetBSD 6.1 |