Manual Page Search Parameters
TUN(4) Device Drivers Manual TUN(4)

tun, tap — network tunnel pseudo-device

pseudo-device tun


#include <sys/types.h>
#include <net/if_tun.h>

The tun pseudo-device driver provides character special devices for communicating with the kernel network stack via the tun and tap network interfaces. Packets sent to these interfaces can be read from the device special file by a userland process and processed as desired. Packets written to the device special file by the userland process are injected back into the kernel networking subsystem.

tun and tap interfaces can be created at runtime using the ifconfig ifaceN create command, or by opening the character special devices /dev/tunN or /dev/tapN respectively.

The minor number of the device special files are associated with the unit number of the network interfaces.

Each device has an exclusive open property: it cannot be opened if it is already open and in use by another process. On the last close of the device all queued packets are discarded. If the device was created by opening a device special file it will be automatically destroyed. The last close of a device special file associated with an interface created via ifconfig(8) will be marked as not running and traffic sent out the will be dropped.

Each read returns at most one packet; if insufficient buffer space is provided, the packet is truncated. Each write supplies exactly one packet. Writes never block. If the protocol queue is full, the packet is dropped, a “collision” is counted, and ENOBUFS is returned.

The following ioctl(2) special commands defined in <net/if_tun.h> are supported:

struct tuninfo *
 
struct tuninfo *
Get or set the interface characteristics. 
/* iface info */
struct tuninfo {
        u_int   mtu;
        u_short type;
        u_short flags;
        u_int   baudrate;
};

flags and type are set by the kernel when an interface is created, and must be set to the same values that the kernel provided.

int *
Provided for backwards compatibility. The flags set must match what the kernel initialised them to.

struct tun_capabilities *
Enable the prepending of network packets with a struct tun_hdr offload header, and set which interface offload capabilities userland can handle on behalf of the kernel. 
struct tun_capabilities {
	uint32_t	tun_if_capabilities;
};

The tun_if_capabilities field is set with IFCAP values from <net/if.h>. It is acceptable to enable the offload header without enabling any interface offload capabilities.

The prepending of network packets with a struct tun_hdr offload header is disabled, and the interface offload capabilities are reset when the device special file is closed.

struct tun_capabilities *
Get which interface offload capabilities are currently enabled. If the interface has not been configured with TUNSCAP this command will fail with ENODEV.

Disable the prepending of network packets with the struct tun_hdr offload header, and clear interface offload capabilities.

The generic ioctls FIONREAD, FIONBIO, FIOASYNC, FIOSETOWN, FIOGETOWN are also supported.

Each packet read from or written to a tun interface is prefixed with a tunnel header consisting of a 4-byte network byte order integer containing the address family of the packet. tun supports AF_INET, AF_INET6, and AF_MPLS packets.

Each packet read from or written to a tap interface is an Ethernet packet. The Ethernet CRC at the end of the frame is not required.

The device special files for tap interfaces support the following additional ioctl(2) commands:

uint8_t[ETHER_ADDR_LEN]
 
uint8_t[ETHER_ADDR_LEN]
Get or set the link layer address (MAC address) of the interface.

When network offload support has been enabled with the TUNSCAP ioctl(2) command, struct tun_hdr is prepended to packets read from and written to the device special file.

The kernel will populate the offload header for reads from the device special file depending on which interface offload capabilities are enabled. A program may use any offload feature when writing to the device special file regardless of which interface offload capabilities are enabled. The offload header can be read from one device special file and written directly to another without interpretation or modification.

struct tun_hdr {
	uint16_t	th_flags;
	uint16_t	th_pad;
	uint16_t	th_vtag;
	uint16_t	th_mss;
};

The fields in the struct tun_hdr are in host native byte order.

The following flags can be set in th_flags to specify which offloads are requested for the current packet:

th_vtag is set.
TCP segmentation offload is requested for the current packet. The maximum segment size is specified in th_mss.
IPv4 header checksum calculation requested.
TCP checksum calculation requested.
UDP checksum calculation requested.
ICMP checksum calculation requested.

Only one of TUN_H_TCP_CSUM, TUN_H_UDP_CSUM, and TUN_H_ICMP_CSUM can be specified at a time. TUN_H_VTAG can only be used with tap interfaces.

th_pad should be set to 0 when writing to the device special file, and ignored when reading from the device special file.

/dev/tun*
 
/dev/tap*
 

If open fails, errno(2) may be set to one of:

[]
Not that many devices configured.
[]
Device was already open.

If a write(2) call fails, errno(2) is set to one of:

[]
The packet supplied was too small or too large. The maximum sized packet allowed is currently 16384 bytes.
[]
There were no mbufs.

ioctl(2) commands may fail with:

[]
Invalid parameters were specified.
[]
Unrecognized ioctl command.

A read(2) call may fail because of:

[]
The associated interface has been destroyed.
[]
Non-blocking I/O was selected and no packets were available.

ioctl(2), inet(4), intro(4), netintro(4), hostname.if(5), ifconfig(8), netstart(8)

tun interfaces originally supported both Layer 3 and Ethernet packets by reconfiguring the type of interface with TUNSIFINFO. Ethernet packet support was split into the separate tap interface in OpenBSD 5.9.

tun was written by Julian Onions <Julian.Onions@nexor.co.uk> at Nottingham University.

The tap interface functionality was written by
Claudio Jeker <claudio@openbsd.org>.

November 18, 2024 openbsd