GRE(4) MidnightBSD Kernel Interfaces Manual GRE(4)


gre — encapsulating network device


device gre


The gre network interface pseudo device encapsulates datagrams into IP. These encapsulated datagrams are routed to a destination host, where they are decapsulated and further routed to their final destination. The ‘‘tunnel’’ appears to the inner datagrams as one hop.

gre interfaces are dynamically created and destroyed with the ifconfig(8) create and destroy subcommands.

This driver currently supports the following modes of operation:

GRE encapsulation (IP protocol number 47)

Encapsulated datagrams are prepended an outer datagram and a GRE header. The GRE header specifies the type of the encapsulated datagram and thus allows for tunneling other protocols than IP like e.g. AppleTalk. GRE mode is also the default tunnel mode on Cisco routers. This is also the default mode of operation of the gre interfaces. As part of the GRE mode, gre also supports Cisco WCCP protocol, both version 1 and version 2. Since there is no reliable way to distinguish between WCCP versions, it should be configured manually using the link2 flag. If the link2 flag is not set (default), then WCCP version 1 is selected.

MOBILE encapsulation (IP protocol number 55)

Datagrams are encapsulated into IP, but with a shorter encapsulation. The original IP header is modified and the modifications are inserted between the so modified header and the original payload. Like gif(4), only for IP-in-IP encapsulation.

The gre interfaces support a number of ioctl(2)s, such as:


Set the IP address of the local tunnel end. This is the source address set by or displayed by ifconfig(8) for the gre interface.


Set the IP address of the remote tunnel end. This is the destination address set by or displayed by ifconfig(8) for the gre interface.


Query the IP address that is set for the local tunnel end. This is the address the encapsulation header carries as local address (i.e., the real address of the tunnel start point).


Query the IP address that is set for the remote tunnel end. This is the address the encapsulated packets are sent to (i.e., the real address of the remote tunnel endpoint).


Set the operation mode to the specified IP protocol value. The protocol is passed to the interface in (

struct ifreq )->ifr_flags. The operation mode can also be given as





to ifconfig(8).

The link1 flag is not used to choose encapsulation, but to modify the internal route search for the remote tunnel endpoint, see the BUGS section below.


Query operation mode.

Note that the IP addresses of the tunnel endpoints may be the same as the ones defined with ifconfig(8) for the interface (as if IP is encapsulated), but need not be, as e.g. when encapsulating AppleTalk.


Configuration example:

Host X-- Host A ----------------tunnel---------- Cisco D------Host E
\ |
\ /
+------Host B----------Host C----------+

On host A (FreeBSD):

route add default B
ifconfig greN create
ifconfig greN A D netmask 0xffffffff linkX up
ifconfig greN tunnel A D
route add E D

On Host D (Cisco):

Interface TunnelX
ip unnumbered D ! e.g. address from Ethernet interface
tunnel source D ! e.g. address from Ethernet interface
tunnel destination A
ip route C <some interface and mask>
ip route A mask C
ip route X mask tunnelX


On Host D (FreeBSD):

route add default C
ifconfig greN create
ifconfig greN D A
ifconfig tunnel greN D A

If all goes well, you should see packets flowing ;-)

If you want to reach Host A over the tunnel (from Host D (Cisco)), then you have to have an alias on Host A for e.g. the Ethernet interface like:

ifconfig <etherif> alias Y

and on the Cisco:

ip route Y mask tunnelX

A similar setup can be used to create a link between two private networks (for example in the 192.168 subnet) over the Internet:

192.168.1.* --- Router A -------tunnel-------- Router B --- 192.168.2.*
\ /
\ /
+------ the Internet ------+

Assuming router A has the (external) IP address A and the internal address, while router B has external address B and internal address, the following commands will configure the tunnel:

On router A:

ifconfig greN create
ifconfig greN link1
ifconfig greN tunnel A B
route add -net 192.168.2 -netmask

On router B:

ifconfig greN create
ifconfig greN link1
ifconfig greN tunnel B A
route add -net 192.168.1 -netmask

Note that this is a safe situation where the link1 flag (as discussed in the BUGS section below) may (and probably should) be set.


The MTU of gre interfaces is set to 1476 by default, to match the value used by Cisco routers. This may not be an optimal value, depending on the link between the two tunnel endpoints. It can be adjusted via ifconfig(8).

For correct operation, the gre device needs a route to the destination that is less specific than the one over the tunnel. (Basically, there needs to be a route to the decapsulating host that does not run over the tunnel, as this would be a loop.) If the addresses are ambiguous, doing the ifconfig tunnel step before the ifconfig(8) call to set the gre IP addresses will help to find a route outside the tunnel.

In order to tell ifconfig(8) to actually mark the interface as ‘‘up’’, the keyword up must be given last on its command line.

The kernel must be set to forward datagrams by setting the ip.forwarding sysctl(8) variable to non-zero.


gif(4), inet(4), ip(4), netintro(4), protocols(5), ifconfig(8), sysctl(8)

A description of GRE encapsulation can be found in RFC 1701 and RFC 1702.

A description of MOBILE encapsulation can be found in RFC 2004.


Heiko W.Rupp 〈〉


The compute_route() code in if_gre.c toggles the last bit of the IP-address to provoke the search for a less specific route than the one directly over the tunnel to prevent loops. This is possibly not the best solution.

To avoid the address munging described above, turn on the link1 flag on the ifconfig(8) command line. This implies that the GRE packet destination and the ifconfig remote host are not the same IP addresses, and that the GRE destination does not route over the gre interface itself.

The GRE RFCs are not yet fully implemented (no GRE options).

MidnightBSD 0.3 June 9, 2002 MidnightBSD 0.3