Skip site navigation (1)Skip section navigation (2)

FreeBSD Manual Pages


home | help
NETMAP(4)		 BSD Kernel Interfaces Manual		     NETMAP(4)

     netmap -- a framework for fast packet I/O

     device netmap

     netmap is a framework for fast and	safe access to network devices (reach-
     ing 14.88 Mpps at less than 1 GHz).  netmap uses memory mapped buffers
     and metadata (buffer indexes and lengths) to communicate with the kernel,
     which is in charge	of validating information through ioctl() and
     select()/poll().  netmap can exploit the parallelism in multiqueue	de-
     vices and multicore systems.

     netmap requires explicit support in device	drivers.  For a	list of	sup-
     ported devices, see the end of this manual	page.

     netmap clients must first open the	open("/dev/netmap"), and then issue an
     ioctl(...,NIOCREGIF,...) to bind the file descriptor to a network device.

     When a device is put in netmap mode, its data path	is disconnected	from
     the host stack.  The processes owning the file descriptor can exchange
     packets with the device, or with the host stack, through an mmapped mem-
     ory region	that contains pre-allocated buffers and	metadata.

     Non blocking I/O is done with special ioctl()'s, whereas the file de-
     scriptor can be passed to select()/poll() to be notified about incoming
     packet or available transmit buffers.

   Data	structures
     All data structures for all devices in netmap mode	are in a memory	region
     shared by the kernel and all processes who	open /dev/netmap (NOTE:	visi-
     bility may	be restricted in future	implementations).  All references be-
     tween the shared data structure are relative (offsets or indexes).	Some
     macros help converting them into actual pointers.

     The data structures in shared memory are the following:

     struct netmap_if (one per interface)
	  indicates the	number of rings	supported by an	interface, their
	  sizes, and the offsets of the	netmap_rings associated	to the inter-
	  face.	 The offset of a struct	netmap_if in the shared	memory region
	  is indicated by the nr_offset	field in the structure returned	by the
	  NIOCREGIF (see below).

	  struct netmap_if {
	      char ni_name[IFNAMSIZ]; /* name of the interface.	*/
	      const u_int ni_num_queues; /* number of hw ring pairs */
	      const ssize_t   ring_ofs[]; /* offset of tx and rx rings */

     struct netmap_ring	(one per ring)
	  contains the index of	the current read or write slot (cur), the num-
	  ber of slots available for reception or transmission (avail),	and an
	  array	of slots describing the	buffers.  There	is one ring pair for
	  each of the N	hardware ring pairs supported by the card (numbered
	  0..N-1), plus	one ring pair (numbered	N) for packets from/to the
	  host stack.

	  struct netmap_ring {
	      const ssize_t buf_ofs;
	      const uint32_t num_slots;	/* number of slots in the ring.	*/
	      uint32_t avail; /* number	of usable slots	*/
	      uint32_t cur; /* 'current' index for the user side */

	      const uint16_t nr_buf_size;
	      uint16_t flags;
	      struct netmap_slot slot[0]; /* array of slots. */

     struct netmap_slot	(one per packet)
	  contains the metadata	for a packet: a	buffer index (buf_idx),	a buf-
	  fer length (len), and	some flags.

	  struct netmap_slot {
	      uint32_t buf_idx;	/* buffer index	*/
	      uint16_t len;   /* packet	length */
	      uint16_t flags; /* buf changed, etc. */
	  #define NS_BUF_CHANGED  0x0001  /* must resync, buffer changed */
	  #define NS_REPORT	  0x0002  /* tell hw to	report results
					   * e.g. by generating	an interrupt

     packet buffers
	  are fixed size (approximately	2k) buffers allocated by the kernel
	  that contain packet data. Buffers addresses are computed through

     Some macros support the access to objects in the shared memory region. In

     struct netmap_if *nifp;
     struct netmap_ring	*txring	= NETMAP_TXRING(nifp, i);
     struct netmap_ring	*rxring	= NETMAP_RXRING(nifp, i);
     int i = txring->slot[txring->cur].buf_idx;
     char *buf = NETMAP_BUF(txring, i);

     netmap supports some ioctl() to synchronize the state of the rings	be-
     tween the kernel and the user processes, plus some	to query and configure
     the interface.  The former	do not require any argument, whereas the lat-
     ter use a struct netmap_req defined as follows:

     struct nmreq {
	     char      nr_name[IFNAMSIZ];
	     uint32_t  nr_offset;      /* nifp offset in the shared region */
	     uint32_t  nr_memsize;     /* size of the shared region */
	     uint32_t  nr_numdescs;    /* descriptors per queue	*/
	     uint16_t  nr_numqueues;
	     uint16_t  nr_ringid;      /* ring(s) we care about	*/
     #define NETMAP_HW_RING  0x4000    /* low bits indicate one	hw ring	*/
     #define NETMAP_SW_RING  0x2000    /* we process the sw ring */
     #define NETMAP_NO_TX_POLL 0x1000  /* no gratuitous	txsync on poll */
     #define NETMAP_RING_MASK 0xfff    /* the actual ring number */

     A device descriptor obtained through /dev/netmap also supports the	ioctl
     supported by network devices.

     The netmap-specific ioctl(2) command codes	below are defined in
     <net/netmap.h> and	are:

	   returns information about the interface named in nr_name.  On re-
	   turn, nr_memsize indicates the size of the shared netmap memory re-
	   gion	(this is device-independent), nr_numslots indicates how	many
	   buffers are in a ring, nr_numrings indicates	the number of rings
	   supported by	the hardware.

	   If the device does not support netmap, the ioctl returns EINVAL.

	   puts	the interface named in nr_name into netmap mode, disconnecting
	   it from the host stack, and/or defines which	rings are controlled
	   through this	file descriptor.  On return, it	gives the same info as
	   NIOCGINFO, and nr_ringid indicates the identity of the rings	con-
	   trolled through the file descriptor.

	   Possible values for nr_ringid are

	   0	  default, all hardware	rings

		  the ``host rings'' connecting	to the host stack

		  the i-th hardware ring
	   By default, a poll or select	call pushes out	any pending packets on
	   the transmit	ring, even if no write events are specified.  The fea-
	   ture	can be disabled	by or-ing NETMAP_NO_TX_SYNC to nr_ringid.  But
	   normally you	should keep this feature unless	you are	using separate
	   file	descriptors for	the send and receive rings, because otherwise
	   packets are pushed out only if NETMAP_TXSYNC	is called, or the send
	   queue is full.

	   NIOCREGIF can be used multiple times	to change the association of a
	   file	descriptor to a	ring pair, always within the same device.

	   brings an interface back to normal mode.

	   tells the hardware of new packets to	transmit, and updates the num-
	   ber of slots	available for transmission.

	   tells the hardware of consumed packets, and asks for	newly avail-
	   able	packets.

     netmap uses select	and poll to wake up processes when significant events

     The following code	implements a traffic generator

     #include <net/netmap.h>
     #include <net/netmap_user.h>
     struct netmap_if *nifp;
     struct netmap_ring	*ring;
     struct netmap_request nmr;

     fd	= open("/dev/netmap", O_RDWR);
     bzero(&nmr, sizeof(nmr));
     strcpy(nmr.nm_name, "ix0");
     ioctl(fd, NIOCREG,	&nmr);
     p = mmap(0, nmr.memsize, fd);
     nifp = NETMAP_IF(p, nmr.offset);
     ring = NETMAP_TXRING(nifp,	0);
     fds.fd = fd;	= POLLOUT;
     for (;;) {
	 poll(list, 1, -1);
	 while (ring->avail-- >	0) {
	     i = ring->cur;
	     buf = NETMAP_BUF(ring, ring->slot[i].buf_index);
	     ... prepare packet	in buf ...
	     ring->slot[i].len = ... packet length ...
	     ring->cur = NETMAP_RING_NEXT(ring,	i);

     netmap supports the following interfaces: em(4), ixgbe(4),	re(4),

     The netmap	framework has been designed and	implemented by Luigi Rizzo and
     Matteo Landi in 2011 at the Universita` di	Pisa.

BSD			       November	16, 2011			   BSD


Want to link to this manual page? Use this URL:

home | help