MFH: sync the netmap code with the one in HEAD

(enhanced VALE switch, netmap pipes, emulated netmap mode).
See details in the log for svn 261909.

2 files changed, 1045 insertions, 264 deletions

diff --git a/sys/net/netmap.h b/sys/net/netmap.h
index b5ab6d5..f0b4c56 100644
--- a/sys/net/netmap.h
+++ b/sys/net/netmap.h
@@ -1,33 +1,27 @@
 /*
- * Copyright (C) 2011-2013 Matteo Landi, Luigi Rizzo. All rights reserved.
- * 
+ * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo. All rights reserved.
+ *
  * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- * 
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
  *   1. Redistributions of source code must retain the above copyright
  *      notice, this list of conditions and the following disclaimer.
- * 
  *   2. Redistributions in binary form must reproduce the above copyright
  *      notice, this list of conditions and the following disclaimer in the
- *      documentation and/or other materials provided with the
- *      distribution.
- * 
- *   3. Neither the name of the authors nor the names of their contributors
- *      may be used to endorse or promote products derived from this
- *      software without specific prior written permission.
- * 
- * THIS SOFTWARE IS PROVIDED BY MATTEO LANDI AND CONTRIBUTORS "AS IS" AND
+ *      documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``S IS''AND
  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
- * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL MATTEO LANDI OR CONTRIBUTORS
- * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
- * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
- * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
- * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
- * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
- * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
- * THE POSSIBILITY OF SUCH DAMAGE.
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
  */
 
 /*
@@ -36,215 +30,249 @@
  * Definitions of constants and the structures used by the netmap
  * framework, for the part visible to both kernel and userspace.
  * Detailed info on netmap is available with "man netmap" or at
- * 
+ *
  *	http://info.iet.unipi.it/~luigi/netmap/
+ *
+ * This API is also used to communicate with the VALE software switch
  */
 
 #ifndef _NET_NETMAP_H_
 #define _NET_NETMAP_H_
 
+#define	NETMAP_API	11		/* current API version */
+
+#define	NETMAP_MIN_API	11		/* min and max versions accepted */
+#define	NETMAP_MAX_API	15
+/*
+ * Some fields should be cache-aligned to reduce contention.
+ * The alignment is architecture and OS dependent, but rather than
+ * digging into OS headers to find the exact value we use an estimate
+ * that should cover most architectures.
+ */
+#define NM_CACHE_ALIGN	128
+
 /*
  * --- Netmap data structures ---
  *
- * The data structures used by netmap are shown below. Those in
- * capital letters are in an mmapp()ed area shared with userspace,
- * while others are private to the kernel.
- * Shared structures do not contain pointers but only memory
- * offsets, so that addressing is portable between kernel and userspace.
-
-
- softc
-+----------------+
-| standard fields|
-| if_pspare[0] ----------+
-+----------------+       |
-                         |
-+----------------+<------+
-|(netmap_adapter)|
-|                |                             netmap_kring
-| tx_rings *--------------------------------->+---------------+
-|                |       netmap_kring         | ring    *---------.
-| rx_rings *--------->+---------------+       | nr_hwcur      |   |
-+----------------+    | ring    *--------.    | nr_hwavail    |   V
-                      | nr_hwcur      |  |    | selinfo       |   |
-                      | nr_hwavail    |  |    +---------------+   .
-                      | selinfo       |  |    |     ...       |   .
-                      +---------------+  |    |(ntx+1 entries)|
-                      |    ....       |  |    |               |
-                      |(nrx+1 entries)|  |    +---------------+
-                      |               |  |
-   KERNEL             +---------------+  |
-                                         |
+ * The userspace data structures used by netmap are shown below.
+ * They are allocated by the kernel and mmap()ed by userspace threads.
+ * Pointers are implemented as memory offsets or indexes,
+ * so that they can be easily dereferenced in kernel and userspace.
+
+   KERNEL (opaque, obviously)
+
   ====================================================================
                                          |
-   USERSPACE                             |      NETMAP_RING
-                                         +---->+-------------+
-                                             / | cur         |
-   NETMAP_IF  (nifp, one per file desc.)    /  | avail       |
-    +---------------+                      /   | buf_ofs     |
-    | ni_tx_rings   |                     /    +=============+
-    | ni_rx_rings   |                    /     | buf_idx     | slot[0]
-    |               |                   /      | len, flags  |
-    |               |                  /       +-------------+
-    +===============+                 /        | buf_idx     | slot[1]
-    | txring_ofs[0] | (rel.to nifp)--'         | len, flags  |
-    | txring_ofs[1] |                          +-------------+
-  (num_rings+1 entries)                     (nr_num_slots entries)
-    | txring_ofs[n] |                          | buf_idx     | slot[n-1]
-    +---------------+                          | len, flags  |
-    | rxring_ofs[0] |                          +-------------+
+   USERSPACE                             |      struct netmap_ring
+                                         +---->+---------------+
+                                             / | head,cur,tail |
+   struct netmap_if (nifp, 1 per fd)        /  | buf_ofs       |
+    +---------------+                      /   | other fields  |
+    | ni_tx_rings   |                     /    +===============+
+    | ni_rx_rings   |                    /     | buf_idx, len  | slot[0]
+    |               |                   /      | flags, ptr    |
+    |               |                  /       +---------------+
+    +===============+                 /        | buf_idx, len  | slot[1]
+    | txring_ofs[0] | (rel.to nifp)--'         | flags, ptr    |
+    | txring_ofs[1] |                          +---------------+
+     (tx+1 entries)                           (num_slots entries)
+    | txring_ofs[t] |                          | buf_idx, len  | slot[n-1]
+    +---------------+                          | flags, ptr    |
+    | rxring_ofs[0] |                          +---------------+
     | rxring_ofs[1] |
-  (num_rings+1 entries)
-    | txring_ofs[n] |
+     (rx+1 entries)
+    | rxring_ofs[r] |
     +---------------+
 
- * The private descriptor ('softc' or 'adapter') of each interface
- * is extended with a "struct netmap_adapter" containing netmap-related
- * info (see description in dev/netmap/netmap_kernel.h.
- * Among other things, tx_rings and rx_rings point to the arrays of
- * "struct netmap_kring" which in turn reache the various
- * "struct netmap_ring", shared with userspace.
-
- * The NETMAP_RING is the userspace-visible replica of the NIC ring.
- * Each slot has the index of a buffer, its length and some flags.
+ * For each "interface" (NIC, host stack, PIPE, VALE switch port) bound to
+ * a file descriptor, the mmap()ed region contains a (logically readonly)
+ * struct netmap_if pointing to struct netmap_ring's.
+ *
+ * There is one netmap_ring per physical NIC ring, plus one tx/rx ring
+ * pair attached to the host stack (this pair is unused for non-NIC ports).
+ *
+ * All physical/host stack ports share the same memory region,
+ * so that zero-copy can be implemented between them.
+ * VALE switch ports instead have separate memory regions.
+ *
+ * The netmap_ring is the userspace-visible replica of the NIC ring.
+ * Each slot has the index of a buffer (MTU-sized and residing in the
+ * mmapped region), its length and some flags. An extra 64-bit pointer
+ * is provided for user-supplied buffers in the tx path.
+ *
  * In user space, the buffer address is computed as
- *	(char *)ring + buf_ofs + index*NETMAP_BUF_SIZE
- * In the kernel, buffers do not necessarily need to be contiguous,
- * and the virtual and physical addresses are derived through
- * a lookup table.
- *
- * struct netmap_slot:
- *
- * buf_idx	is the index of the buffer associated to the slot.
- * len		is the length of the payload
- * NS_BUF_CHANGED	must be set whenever userspace wants
- *		to change buf_idx (it might be necessary to
- *		reprogram the NIC slot)
- * NS_REPORT	must be set if we want the NIC to generate an interrupt
- *		when this slot is used. Leaving it to 0 improves
- *		performance.
- * NS_FORWARD	if set on a receive ring, and the device is in
- *		transparent mode, buffers released with the flag set
- *		will be forwarded to the 'other' side (host stack
- *		or NIC, respectively) on the next select() or ioctl()
- *
- *		The following will be supported from NETMAP_API = 5
- * NS_NO_LEARN	on a VALE switch, do not 'learn' the source port for
- *		this packet.
- * NS_INDIRECT	the netmap buffer contains a 64-bit pointer to
- *		the actual userspace buffer. This may be useful
- *		to reduce copies in a VM environment.
- * NS_MOREFRAG	Part of a multi-segment frame. The last (or only)
- *		segment must not have this flag.
- * NS_PORT_MASK	the high 8 bits of the flag, if not zero, indicate the
- *		destination port for the VALE switch, overriding
- *		the lookup table.
+ *	(char *)ring + buf_ofs + index * NETMAP_BUF_SIZE
+ *
+ * Added in NETMAP_API 11:
+ *
+ * + NIOCREGIF can request the allocation of extra spare buffers from
+ *   the same memory pool. The desired number of buffers must be in
+ *   nr_arg3. The ioctl may return fewer buffers, depending on memory
+ *   availability. nr_arg3 will return the actual value, and, once
+ *   mapped, nifp->ni_bufs_head will be the index of the first buffer.
+ *
+ *   The buffers are linked to each other using the first uint32_t
+ *   as the index. On close, ni_bufs_head must point to the list of
+ *   buffers to be released.
+ *
+ * + NIOCREGIF can request space for extra rings (and buffers)
+ *   allocated in the same memory space. The number of extra rings
+ *   is in nr_arg1, and is advisory. This is a no-op on NICs where
+ *   the size of the memory space is fixed.
+ *
+ * + NIOCREGIF can attach to PIPE rings sharing the same memory
+ *   space with a parent device. The ifname indicates the parent device,
+ *   which must already exist. Flags in nr_flags indicate if we want to
+ *   bind the master or slave side, the index (from nr_ringid)
+ *   is just a cookie and does need to be sequential.
+ *
+ * + NIOCREGIF can also attach to 'monitor' rings that replicate
+ *   the content of specific rings, also from the same memory space.
+ *
+ *   Extra flags in nr_flags support the above functions.
+ *   Application libraries may use the following naming scheme:
+ *	netmap:foo			all NIC ring pairs
+ *	netmap:foo^			only host ring pair
+ *	netmap:foo+			all NIC ring + host ring pairs
+ *	netmap:foo-k			the k-th NIC ring pair
+ *	netmap:foo{k			PIPE ring pair k, master side
+ *	netmap:foo}k			PIPE ring pair k, slave side
  */
 
+/*
+ * struct netmap_slot is a buffer descriptor
+ */
 struct netmap_slot {
-	uint32_t buf_idx; /* buffer index */
-	uint16_t len;	/* packet length, to be copied to/from the hw ring */
-	uint16_t flags;	/* buf changed, etc. */
-#define	NS_BUF_CHANGED	0x0001	/* must resync the map, buffer changed */
-#define	NS_REPORT	0x0002	/* ask the hardware to report results
-				 * e.g. by generating an interrupt
-				 */
-#define	NS_FORWARD	0x0004	/* pass packet to the other endpoint
-				 * (host stack or device)
-				 */
-#define	NS_NO_LEARN	0x0008
-#define	NS_INDIRECT	0x0010
-#define	NS_MOREFRAG	0x0020
+	uint32_t buf_idx;	/* buffer index */
+	uint16_t len;		/* length for this slot */
+	uint16_t flags;		/* buf changed, etc. */
+	uint64_t ptr;		/* pointer for indirect buffers */
+};
+
+/*
+ * The following flags control how the slot is used
+ */
+
+#define	NS_BUF_CHANGED	0x0001	/* buf_idx changed */
+	/*
+	 * must be set whenever buf_idx is changed (as it might be
+	 * necessary to recompute the physical address and mapping)
+	 */
+
+#define	NS_REPORT	0x0002	/* ask the hardware to report results */
+	/*
+	 * Request notification when slot is used by the hardware.
+	 * Normally transmit completions are handled lazily and
+	 * may be unreported. This flag lets us know when a slot
+	 * has been sent (e.g. to terminate the sender).
+	 */
+
+#define	NS_FORWARD	0x0004	/* pass packet 'forward' */
+	/*
+	 * (Only for physical ports, rx rings with NR_FORWARD set).
+	 * Slot released to the kernel (i.e. before ring->head) with
+	 * this flag set are passed to the peer ring (host/NIC),
+	 * thus restoring the host-NIC connection for these slots.
+	 * This supports efficient traffic monitoring or firewalling.
+	 */
+
+#define	NS_NO_LEARN	0x0008	/* disable bridge learning */
+ 	/*
+	 * On a VALE switch, do not 'learn' the source port for
+ 	 * this buffer.
+	 */
+
+#define	NS_INDIRECT	0x0010	/* userspace buffer */
+ 	/*
+	 * (VALE tx rings only) data is in a userspace buffer,
+	 * whose address is in the 'ptr' field in the slot.
+	 */
+
+#define	NS_MOREFRAG	0x0020	/* packet has more fragments */
+ 	/*
+	 * (VALE ports only)
+	 * Set on all but the last slot of a multi-segment packet.
+	 * The 'len' field refers to the individual fragment.
+	 */
+
 #define	NS_PORT_SHIFT	8
 #define	NS_PORT_MASK	(0xff << NS_PORT_SHIFT)
-};
+	/*
+ 	 * The high 8 bits of the flag, if not zero, indicate the
+	 * destination port for the VALE switch, overriding
+ 	 * the lookup table.
+ 	 */
+
+#define	NS_RFRAGS(_slot)	( ((_slot)->flags >> 8) & 0xff)
+	/*
+	 * (VALE rx rings only) the high 8 bits
+	 *  are the number of fragments.
+	 */
+
 
 /*
+ * struct netmap_ring
+ *
  * Netmap representation of a TX or RX ring (also known as "queue").
  * This is a queue implemented as a fixed-size circular array.
- * At the software level, two fields are important: avail and cur.
+ * At the software level the important fields are: head, cur, tail.
  *
  * In TX rings:
- *	avail	indicates the number of slots available for transmission.
- *		It is updated by the kernel after every netmap system call.
- *		It MUST BE decremented by the application when it appends a
- *		packet.
- *	cur	indicates the slot to use for the next packet
- *		to send (i.e. the "tail" of the queue).
- *		It MUST BE incremented by the application before
- *		netmap system calls to reflect the number of newly
- *		sent packets.
- *		It is checked by the kernel on netmap system calls
- *		(normally unmodified by the kernel unless invalid).
- *
- *   The kernel side of netmap uses two additional fields in its own
- *   private ring structure, netmap_kring:
- *	nr_hwcur is a copy of nr_cur on an NIOCTXSYNC.
- *	nr_hwavail is the number of slots known as available by the
- *		hardware. It is updated on an INTR (inc by the
- *		number of packets sent) and on a NIOCTXSYNC
- *		(decrease by nr_cur - nr_hwcur)
- *		A special case, nr_hwavail is -1 if the transmit
- *		side is idle (no pending transmits).
+ *
+ *	head	first slot available for transmission.
+ *	cur	wakeup point. select() and poll() will unblock
+ *		when 'tail' moves past 'cur'
+ *	tail	(readonly) first slot reserved to the kernel
+ *
+ *	[head .. tail-1] can be used for new packets to send;
+ *	'head' and 'cur' must be incremented as slots are filled
+ *	    with new packets to be sent;
+ *	'cur' can be moved further ahead if we need more space
+ *	for new transmissions.
  *
  * In RX rings:
- *	avail	is the number of packets available (possibly 0).
- *		It MUST BE decremented by the application when it consumes
- *		a packet, and it is updated to nr_hwavail on a NIOCRXSYNC
- *	cur	indicates the first slot that contains a packet not
- *		processed yet (the "head" of the queue).
- *		It MUST BE incremented by the software when it consumes
- *		a packet.
- *	reserved	indicates the number of buffers before 'cur'
- *		that the application has still in use. Normally 0,
- *		it MUST BE incremented by the application when it
- *		does not return the buffer immediately, and decremented
- *		when the buffer is finally freed.
- *
- *   The kernel side of netmap uses two additional fields in the kring:
- *	nr_hwcur is a copy of nr_cur on an NIOCRXSYNC
- *	nr_hwavail is the number of packets available. It is updated
- *		on INTR (inc by the number of new packets arrived)
- *		and on NIOCRXSYNC (decreased by nr_cur - nr_hwcur).
+ *
+ *	head	first valid received packet
+ *	cur	wakeup point. select() and poll() will unblock
+ *		when 'tail' moves past 'cur'
+ *	tail	(readonly) first slot reserved to the kernel
+ *
+ *	[head .. tail-1] contain received packets;
+ *	'head' and 'cur' must be incremented as slots are consumed
+ *		and can be returned to the kernel;
+ *	'cur' can be moved further ahead if we want to wait for
+ *		new packets without returning the previous ones.
  *
  * DATA OWNERSHIP/LOCKING:
- *	The netmap_ring is owned by the user program and it is only
- *	accessed or modified in the upper half of the kernel during
- *	a system call.
- *
- *	The netmap_kring is only modified by the upper half of the kernel.
- *
- * FLAGS
- *	NR_TIMESTAMP	updates the 'ts' field on each syscall. This is
- *			a global timestamp for all packets.
- *	NR_RX_TSTMP	if set, the last 64 byte in each buffer will
- *			contain a timestamp for the frame supplied by
- *			the hardware (if supported)
- *	NR_FORWARD	if set, the NS_FORWARD flag in each slot of the
- *			RX ring is checked, and if set the packet is
- *			passed to the other side (host stack or device,
- *			respectively). This permits bpf-like behaviour
- *			or transparency for selected packets.
+ *	The netmap_ring, and all slots and buffers in the range
+ *	[head .. tail-1] are owned by the user program;
+ *	the kernel only accesses them during a netmap system call
+ *	and in the user thread context.
+ *
+ *	Other slots and buffers are reserved for use by the kernel
  */
 struct netmap_ring {
 	/*
-	 * nr_buf_base_ofs is meant to be used through macros.
+	 * buf_ofs is meant to be used through macros.
 	 * It contains the offset of the buffer region from this
 	 * descriptor.
 	 */
-	const ssize_t	buf_ofs;
+	const int64_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' r/w position */
-	uint32_t	reserved;	/* not refilled before current */
+	const uint32_t	nr_buf_size;
+	const uint16_t	ringid;
+	const uint16_t	dir;		/* 0: tx, 1: rx */
 
-	const uint16_t	nr_buf_size;
-	uint16_t	flags;
-#define	NR_TIMESTAMP	0x0002		/* set timestamp on *sync() */
-#define	NR_FORWARD	0x0004		/* enable NS_FORWARD for ring */
-#define	NR_RX_TSTMP	0x0008		/* set rx timestamp in slots */
+	uint32_t        head;		/* (u) first user slot */
+	uint32_t        cur;		/* (u) wakeup point */
+	uint32_t	tail;		/* (k) first kernel slot */
+
+	uint32_t	flags;
 
-	struct timeval	ts;		/* time of last *sync() */
+	struct timeval	ts;		/* (k) time of last *sync() */
+
+	/* opaque room for a mutex or similar object */
+	uint8_t		sem[128] __attribute__((__aligned__(NM_CACHE_ALIGN)));
 
 	/* the slots follow. This struct has variable size */
 	struct netmap_slot slot[0];	/* array of slots. */
@@ -252,88 +280,246 @@ struct netmap_ring {
 
 
 /*
+ * RING FLAGS
+ */
+#define	NR_TIMESTAMP	0x0002		/* set timestamp on *sync() */
+	/*
+	 * updates the 'ts' field on each netmap syscall. This saves
+	 * saves a separate gettimeofday(), and is not much worse than
+	 * software timestamps generated in the interrupt handler.
+	 */
+
+#define	NR_FORWARD	0x0004		/* enable NS_FORWARD for ring */
+ 	/*
+	 * Enables the NS_FORWARD slot flag for the ring.
+	 */
+
+
+/*
  * Netmap representation of an interface and its queue(s).
+ * This is initialized by the kernel when binding a file
+ * descriptor to a port, and should be considered as readonly
+ * by user programs. The kernel never uses it.
+ *
  * There is one netmap_if for each file descriptor on which we want
- * to select/poll.  We assume that on each interface has the same number
- * of receive and transmit queues.
+ * to select/poll.
  * select/poll operates on one or all pairs depending on the value of
  * nmr_queueid passed on the ioctl.
  */
 struct netmap_if {
 	char		ni_name[IFNAMSIZ]; /* name of the interface. */
-	const u_int	ni_version;	/* API version, currently unused */
-	const u_int	ni_rx_rings;	/* number of rx rings */
-	const u_int	ni_tx_rings;	/* if zero, same as ni_rx_rings */
+	const uint32_t	ni_version;	/* API version, currently unused */
+	const uint32_t	ni_flags;	/* properties */
+#define	NI_PRIV_MEM	0x1		/* private memory region */
+
+	/*
+	 * The number of packet rings available in netmap mode.
+	 * Physical NICs can have different numbers of tx and rx rings.
+	 * Physical NICs also have a 'host' ring pair.
+	 * Additionally, clients can request additional ring pairs to
+	 * be used for internal communication.
+	 */
+	const uint32_t	ni_tx_rings;	/* number of HW tx rings */
+	const uint32_t	ni_rx_rings;	/* number of HW rx rings */
+
+	uint32_t	ni_bufs_head;	/* head index for extra bufs */
+	uint32_t	ni_spare1[5];
 	/*
 	 * The following array contains the offset of each netmap ring
-	 * from this structure. The first ni_tx_queues+1 entries refer
-	 * to the tx rings, the next ni_rx_queues+1 refer to the rx rings
-	 * (the last entry in each block refers to the host stack rings).
-	 * The area is filled up by the kernel on NIOCREG,
+	 * from this structure, in the following order:
+	 * NIC tx rings (ni_tx_rings); host tx ring (1); extra tx rings;
+	 * NIC rx rings (ni_rx_rings); host tx ring (1); extra rx rings.
+	 *
+	 * The area is filled up by the kernel on NIOCREGIF,
 	 * and then only read by userspace code.
 	 */
 	const ssize_t	ring_ofs[0];
 };
 
-#ifndef NIOCREGIF	
+
+#ifndef NIOCREGIF
 /*
  * ioctl names and related fields
  *
+ * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
+ *	whose identity is set in NIOCREGIF through nr_ringid.
+ *	These are non blocking and take no argument.
+ *
  * NIOCGINFO takes a struct ifreq, the interface name is the input,
  *	the outputs are number of queues and number of descriptor
  *	for each queue (useful to set number of threads etc.).
+ *	The info returned is only advisory and may change before
+ *	the interface is bound to a file descriptor.
  *
- * NIOCREGIF takes an interface name within a struct ifreq,
+ * NIOCREGIF takes an interface name within a struct nmre,
  *	and activates netmap mode on the interface (if possible).
  *
- *	For vale ports, starting with NETMAP_API = 5,
- *	nr_tx_rings and nr_rx_rings specify how many software rings
- *	are created (0 means 1).
+ * The argument to NIOCGINFO/NIOCREGIF overlays struct ifreq so we
+ * can pass it down to other NIC-related ioctls.
  *
- *	NIOCREGIF is also used to attach a NIC to a VALE switch.
- *	In this case the name is vale*:ifname, and "nr_cmd"
- *	is set to 'NETMAP_BDG_ATTACH' or 'NETMAP_BDG_DETACH'.
- *	nr_ringid specifies which rings should be attached, 0 means all,
- *	NETMAP_HW_RING + n means only the n-th ring.
- *	The process can terminate after the interface has been attached.
+ * The actual argument (struct nmreq) has a number of options to request
+ * different functions.
+ * The following are used in NIOCREGIF when nr_cmd == 0:
  *
- * NIOCUNREGIF unregisters the interface associated to the fd.
- *	this is deprecated and will go away.
+ * nr_name	(in)
+ *	The name of the port (em0, valeXXX:YYY, etc.)
+ *	limited to IFNAMSIZ for backward compatibility.
  *
- * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
- *	whose identity is set in NIOCREGIF through nr_ringid
+ * nr_version	(in/out)
+ *	Must match NETMAP_API as used in the kernel, error otherwise.
+ *	Always returns the desired value on output.
+ *
+ * nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings (in/out)
+ *	On input, non-zero values may be used to reconfigure the port
+ *	according to the requested values, but this is not guaranteed.
+ *	On output the actual values in use are reported.
+ *
+ * nr_ringid (in)
+ *	Indicates how rings should be bound to the file descriptors.
+ *	If nr_flags != 0, then the low bits (in NETMAP_RING_MASK)
+ *	are used to indicate the ring number, and nr_flags specifies
+ *	the actual rings to bind. NETMAP_NO_TX_POLL is unaffected.
+ *
+ *	NOTE: THE FOLLOWING (nr_flags == 0) IS DEPRECATED:
+ *	If nr_flags == 0, NETMAP_HW_RING and NETMAP_SW_RING control
+ *	the binding as follows:
+ *	0 (default)			binds all physical rings
+ *	NETMAP_HW_RING | ring number	binds a single ring pair
+ *	NETMAP_SW_RING			binds only the host tx/rx rings
+ *
+ *	NETMAP_NO_TX_POLL can be OR-ed to make select()/poll() push
+ *		packets on tx rings only if POLLOUT is set.
+ *		The default is to push any pending packet.
+ *
+ *	NETMAP_DO_RX_POLL can be OR-ed to make select()/poll() release
+ *		packets on rx rings also when POLLIN is NOT set.
+ *		The default is to touch the rx ring only with POLLIN.
+ *		Note that this is the opposite of TX because it
+ *		reflects the common usage.
+ *
+ *	NOTE: NETMAP_PRIV_MEM IS DEPRECATED, use nr_arg2 instead.
+ *	NETMAP_PRIV_MEM is set on return for ports that do not use
+ *		the global memory allocator.
+ *		This information is not significant and applications
+ *		should look at the region id in nr_arg2
+ *
+ * nr_flags	is the recommended mode to indicate which rings should
+ *		be bound to a file descriptor. Values are NR_REG_*
+ *
+ * nr_arg1 (in)	The number of extra rings to be reserved.
+ *		Especially when allocating a VALE port the system only
+ *		allocates the amount of memory needed for the port.
+ *		If more shared memory rings are desired (e.g. for pipes),
+ *		the first invocation for the same basename/allocator
+ *		should specify a suitable number. Memory cannot be
+ *		extended after the first allocation without closing
+ *		all ports on the same region.
+ *
+ * nr_arg2 (in/out) The identity of the memory region used.
+ *		On input, 0 means the system decides autonomously,
+ *		other values may try to select a specific region.
+ *		On return the actual value is reported.
+ *		Region '1' is the global allocator, normally shared
+ *		by all interfaces. Other values are private regions.
+ *		If two ports the same region zero-copy is possible.
+ *
+ * nr_arg3 (in/out)	number of extra buffers to be allocated.
+ *
+ *
+ *
+ * nr_cmd (in)	if non-zero indicates a special command:
+ *	NETMAP_BDG_ATTACH	 and nr_name = vale*:ifname
+ *		attaches the NIC to the switch; nr_ringid specifies
+ *		which rings to use. Used by vale-ctl -a ...
+ *	    nr_arg1 = NETMAP_BDG_HOST also attaches the host port
+ *		as in vale-ctl -h ...
+ *
+ *	NETMAP_BDG_DETACH	and nr_name = vale*:ifname
+ *		disconnects a previously attached NIC.
+ *		Used by vale-ctl -d ...
+ *
+ *	NETMAP_BDG_LIST
+ *		list the configuration of VALE switches.
+ *
+ *	NETMAP_BDG_VNET_HDR
+ *		Set the virtio-net header length used by the client
+ *		of a VALE switch port.
+ *
+ * nr_arg1, nr_arg2, nr_arg3  (in/out)		command specific
+ *
+ *	
  *
- * NETMAP_API is the API version.
  */
 
+
 /*
- * struct nmreq overlays a struct ifreq
+ * struct nmreq overlays a struct ifreq (just the name)
+ *
+ * On input, nr_ringid indicates which rings we are requesting,
+ * with the low flags for the specific ring number.
+ * selection			FLAGS	RING INDEX
+ *
+ *	all the NIC rings	0x0000	-
+ *	only HOST ring		0x2000	ring index
+ *	single NIC ring		0x4000	-
+ *	all the NIC+HOST rings	0x6000	-
+ *	one pipe ring, master	0x8000	ring index
+ *	*** INVALID		0xA000
+ *	one pipe ring, slave	0xC000	ring index
+ *	*** INVALID		0xE000
+ * 
  */
 struct nmreq {
 	char		nr_name[IFNAMSIZ];
 	uint32_t	nr_version;	/* API version */
-#define	NETMAP_API	4		/* current version */
 	uint32_t	nr_offset;	/* nifp offset in the shared region */
 	uint32_t	nr_memsize;	/* size of the shared region */
 	uint32_t	nr_tx_slots;	/* slots in tx rings */
 	uint32_t	nr_rx_slots;	/* slots in rx rings */
 	uint16_t	nr_tx_rings;	/* number of tx rings */
 	uint16_t	nr_rx_rings;	/* number of rx rings */
+
 	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		/* process the sw ring */
+#define NETMAP_HW_RING		0x4000	/* single NIC ring pair */
+#define NETMAP_SW_RING		0x2000	/* only host ring pair */
+
+#define NETMAP_RING_MASK	0x0fff	/* the ring number */
+
 #define NETMAP_NO_TX_POLL	0x1000	/* no automatic txsync on poll */
-#define NETMAP_RING_MASK 0xfff		/* the ring number */
+
+#define NETMAP_DO_RX_POLL	0x8000	/* DO automatic rxsync on poll */
+
 	uint16_t	nr_cmd;
 #define NETMAP_BDG_ATTACH	1	/* attach the NIC */
 #define NETMAP_BDG_DETACH	2	/* detach the NIC */
 #define NETMAP_BDG_LOOKUP_REG	3	/* register lookup function */
 #define NETMAP_BDG_LIST		4	/* get bridge's info */
-	uint16_t	nr_arg1;
+#define NETMAP_BDG_VNET_HDR     5       /* set the port virtio-net-hdr length */
+#define NETMAP_BDG_OFFSET	NETMAP_BDG_VNET_HDR	/* deprecated alias */
+
+	uint16_t	nr_arg1;	/* reserve extra rings in NIOCREGIF */
 #define NETMAP_BDG_HOST		1	/* attach the host stack on ATTACH */
+
 	uint16_t	nr_arg2;
-	uint32_t	spare2[3];
+	uint32_t	nr_arg3;	/* req. extra buffers in NIOCREGIF */
+	uint32_t	nr_flags;
+	/* various modes, extends nr_ringid */
+	uint32_t	spare2[1];
+};
+
+#define NR_REG_MASK		0xf /* values for nr_flags */
+enum {	NR_REG_DEFAULT	= 0,	/* backward compat, should not be used. */
+	NR_REG_ALL_NIC	= 1,
+	NR_REG_SW	= 2,
+	NR_REG_NIC_SW	= 3,
+	NR_REG_ONE_NIC	= 4,
+	NR_REG_PIPE_MASTER = 5,
+	NR_REG_PIPE_SLAVE = 6,
 };
+/* monitor uses the NR_REG to select the rings to monitor */
+#define NR_MONITOR_TX	0x100
+#define NR_MONITOR_RX	0x200
+
 
 /*
  * FreeBSD uses the size value embedded in the _IOWR to determine
@@ -343,9 +529,22 @@ struct nmreq {
  */
 #define NIOCGINFO	_IOWR('i', 145, struct nmreq) /* return IF info */
 #define NIOCREGIF	_IOWR('i', 146, struct nmreq) /* interface register */
-#define NIOCUNREGIF	_IO('i', 147) /* interface unregister */
 #define NIOCTXSYNC	_IO('i', 148) /* sync tx queues */
 #define NIOCRXSYNC	_IO('i', 149) /* sync rx queues */
 #endif /* !NIOCREGIF */
 
+
+/*
+ * Helper functions for kernel and userspace
+ */
+
+/*
+ * check if space is available in the ring.
+ */
+static inline int
+nm_ring_empty(struct netmap_ring *ring)
+{
+	return (ring->cur == ring->tail);
+}
+
 #endif /* _NET_NETMAP_H_ */
diff --git a/sys/net/netmap_user.h b/sys/net/netmap_user.h
index fcb5cb3..9c3a4c1 100644
--- a/sys/net/netmap_user.h
+++ b/sys/net/netmap_user.h
@@ -1,40 +1,34 @@
 /*
- * Copyright (C) 2011 Matteo Landi, Luigi Rizzo. All rights reserved.
- * 
+ * Copyright (C) 2011-2014 Universita` di Pisa. All rights reserved.
+ *
  * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions are
- * met:
- * 
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
  *   1. Redistributions of source code must retain the above copyright
  *      notice, this list of conditions and the following disclaimer.
- * 
  *   2. Redistributions in binary form must reproduce the above copyright
  *      notice, this list of conditions and the following disclaimer in the
- *      documentation and/or other materials provided with the
- *      distribution.
- * 
- *   3. Neither the name of the authors nor the names of their contributors
- *      may be used to endorse or promote products derived from this
- *      software without specific prior written permission.
- * 
- * THIS SOFTWARE IS PROVIDED BY MATTEO LANDI AND CONTRIBUTORS "AS IS" AND
+ *      documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
- * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL MATTEO LANDI OR CONTRIBUTORS
- * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
- * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
- * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
- * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
- * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
- * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
- * THE POSSIBILITY OF SUCH DAMAGE.
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
  */
 
 /*
  * $FreeBSD$
  *
- * This header contains the macros used to manipulate netmap structures
- * and packets in userspace. See netmap(4) for more information.
+ * Functions and macros to manipulate netmap structures and packets
+ * in userspace. See netmap(4) for more information.
  *
  * The address of the struct netmap_if, say nifp, is computed from the
  * value returned from ioctl(.., NIOCREG, ...) and the mmap region:
@@ -49,22 +43,44 @@
  *		we can access ring->nr_cur, ring->nr_avail, ring->nr_flags
  *
  *	ring->slot[i] gives us the i-th slot (we can access
- *		directly plen, flags, bufindex)
+ *		directly len, flags, buf_idx)
  *
  *	char *buf = NETMAP_BUF(ring, x) returns a pointer to
  *		the buffer numbered x
  *
- * Since rings are circular, we have macros to compute the next index
- *	i = NETMAP_RING_NEXT(ring, i);
+ * All ring indexes (head, cur, tail) should always move forward.
+ * To compute the next index in a circular ring you can use
+ *	i = nm_ring_next(ring, i);
+ *
+ * To ease porting apps from pcap to netmap we supply a few fuctions
+ * that can be called to open, close, read and write on netmap in a way
+ * similar to libpcap. Note that the read/write function depend on
+ * an ioctl()/select()/poll() being issued to refill rings or push
+ * packets out.
+ *
+ * In order to use these, include #define NETMAP_WITH_LIBS
+ * in the source file that invokes these functions.
  */
 
 #ifndef _NET_NETMAP_USER_H_
 #define _NET_NETMAP_USER_H_
 
+#include <stdint.h>
+#include <sys/socket.h>		/* apple needs sockaddr */
+#include <net/if.h>		/* IFNAMSIZ */
+
+#ifndef likely
+#define likely(x)	__builtin_expect(!!(x), 1)
+#define unlikely(x)	__builtin_expect(!!(x), 0)
+#endif /* likely and unlikely */
+
+#include <net/netmap.h>
+
+/* helper macro */
 #define _NETMAP_OFFSET(type, ptr, offset) \
 	((type)(void *)((char *)(ptr) + (offset)))
 
-#define NETMAP_IF(b, o)	_NETMAP_OFFSET(struct netmap_if *, b, o)
+#define NETMAP_IF(_base, _ofs)	_NETMAP_OFFSET(struct netmap_if *, _base, _ofs)
 
 #define NETMAP_TXRING(nifp, index) _NETMAP_OFFSET(struct netmap_ring *, \
 	nifp, (nifp)->ring_ofs[index] )
@@ -77,19 +93,585 @@
 
 #define NETMAP_BUF_IDX(ring, buf)			\
 	( ((char *)(buf) - ((char *)(ring) + (ring)->buf_ofs) ) / \
-		(ring)->nr_buf_size ) 
+		(ring)->nr_buf_size )
+
+
+static inline uint32_t
+nm_ring_next(struct netmap_ring *r, uint32_t i)
+{
+	return ( unlikely(i + 1 == r->num_slots) ? 0 : i + 1);
+}
+
+
+/*
+ * Return 1 if we have pending transmissions in the tx ring.
+ * When everything is complete ring->head = ring->tail + 1 (modulo ring size)
+ */
+static inline int
+nm_tx_pending(struct netmap_ring *r)
+{
+	return nm_ring_next(r, r->tail) != r->head;
+}
+
 
-#define	NETMAP_RING_NEXT(r, i)				\
-	((i)+1 == (r)->num_slots ? 0 : (i) + 1 )
+static inline uint32_t
+nm_ring_space(struct netmap_ring *ring)
+{
+        int ret = ring->tail - ring->cur;
+        if (ret < 0)
+                ret += ring->num_slots;
+        return ret;
+}
 
-#define	NETMAP_RING_FIRST_RESERVED(r)			\
-	( (r)->cur < (r)->reserved ?			\
-	  (r)->cur + (r)->num_slots - (r)->reserved :	\
-	  (r)->cur - (r)->reserved )
 
+#ifdef NETMAP_WITH_LIBS
 /*
- * Return 1 if the given tx ring is empty.
+ * Support for simple I/O libraries.
+ * Include other system headers required for compiling this.
  */
-#define NETMAP_TX_RING_EMPTY(r)	((r)->avail >= (r)->num_slots - 1)
+
+#ifndef HAVE_NETMAP_WITH_LIBS
+#define HAVE_NETMAP_WITH_LIBS
+
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <string.h>	/* memset */
+#include <sys/ioctl.h>
+#include <sys/errno.h>	/* EINVAL */
+#include <fcntl.h>	/* O_RDWR */
+#include <unistd.h>	/* close() */
+#include <signal.h>
+#include <stdlib.h>
+
+#ifndef ND /* debug macros */
+/* debug support */
+#define ND(_fmt, ...) do {} while(0)
+#define D(_fmt, ...)						\
+	do {							\
+		struct timeval t0;				\
+		gettimeofday(&t0, NULL);			\
+		fprintf(stderr, "%03d.%06d %s [%d] " _fmt "\n",	\
+		    (int)(t0.tv_sec % 1000), (int)t0.tv_usec,	\
+		    __FUNCTION__, __LINE__, ##__VA_ARGS__);	\
+        } while (0)
+
+/* Rate limited version of "D", lps indicates how many per second */
+#define RD(lps, format, ...)                                    \
+    do {                                                        \
+        static int t0, __cnt;                                   \
+        struct timeval __xxts;                                  \
+        gettimeofday(&__xxts, NULL);                            \
+        if (t0 != __xxts.tv_sec) {                              \
+            t0 = __xxts.tv_sec;                                 \
+            __cnt = 0;                                          \
+        }                                                       \
+        if (__cnt++ < lps) {                                    \
+            D(format, ##__VA_ARGS__);                           \
+        }                                                       \
+    } while (0)
+#endif
+
+struct nm_pkthdr {	/* same as pcap_pkthdr */
+	struct timeval	ts;
+	uint32_t	caplen;
+	uint32_t	len;
+};
+
+struct nm_stat {	/* same as pcap_stat	*/
+	u_int	ps_recv;
+	u_int	ps_drop;
+	u_int	ps_ifdrop;
+#ifdef WIN32
+	u_int	bs_capt;
+#endif /* WIN32 */
+};
+
+#define NM_ERRBUF_SIZE	512
+
+struct nm_desc {
+	struct nm_desc *self; /* point to self if netmap. */
+	int fd;
+	void *mem;
+	int memsize;
+	int done_mmap;	/* set if mem is the result of mmap */
+	struct netmap_if * const nifp;
+	uint16_t first_tx_ring, last_tx_ring, cur_tx_ring;
+	uint16_t first_rx_ring, last_rx_ring, cur_rx_ring;
+	struct nmreq req;	/* also contains the nr_name = ifname */
+	struct nm_pkthdr hdr;
+
+	/*
+	 * The memory contains netmap_if, rings and then buffers.
+	 * Given a pointer (e.g. to nm_inject) we can compare with
+	 * mem/buf_start/buf_end to tell if it is a buffer or
+	 * some other descriptor in our region.
+	 * We also store a pointer to some ring as it helps in the
+	 * translation from buffer indexes to addresses.
+	 */
+	struct netmap_ring * const some_ring;
+	void * const buf_start;
+	void * const buf_end;
+	/* parameters from pcap_open_live */
+	int snaplen;
+	int promisc;
+	int to_ms;
+	char *errbuf;
+
+	/* save flags so we can restore them on close */
+	uint32_t if_flags;
+        uint32_t if_reqcap;
+        uint32_t if_curcap;
+
+	struct nm_stat st;
+	char msg[NM_ERRBUF_SIZE];
+};
+
+/*
+ * when the descriptor is open correctly, d->self == d
+ * Eventually we should also use some magic number.
+ */
+#define P2NMD(p)		((struct nm_desc *)(p))
+#define IS_NETMAP_DESC(d)	((d) && P2NMD(d)->self == P2NMD(d))
+#define NETMAP_FD(d)		(P2NMD(d)->fd)
+
+
+/*
+ * this is a slightly optimized copy routine which rounds
+ * to multiple of 64 bytes and is often faster than dealing
+ * with other odd sizes. We assume there is enough room
+ * in the source and destination buffers.
+ *
+ * XXX only for multiples of 64 bytes, non overlapped.
+ */
+static inline void
+nm_pkt_copy(const void *_src, void *_dst, int l)
+{
+	const uint64_t *src = (const uint64_t *)_src;
+	uint64_t *dst = (uint64_t *)_dst;
+
+	if (unlikely(l >= 1024)) {
+		memcpy(dst, src, l);
+		return;
+	}
+	for (; likely(l > 0); l-=64) {
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+		*dst++ = *src++;
+	}
+}
+
+
+/*
+ * The callback, invoked on each received packet. Same as libpcap
+ */
+typedef void (*nm_cb_t)(u_char *, const struct nm_pkthdr *, const u_char *d);
+
+/*
+ *--- the pcap-like API ---
+ *
+ * nm_open() opens a file descriptor, binds to a port and maps memory.
+ *
+ * ifname	(netmap:foo or vale:foo) is the port name
+ *		a suffix can indicate the follwing:
+ *		^		bind the host (sw) ring pair
+ *		*		bind host and NIC ring pairs (transparent)
+ *		-NN		bind individual NIC ring pair
+ *		{NN		bind master side of pipe NN
+ *		}NN		bind slave side of pipe NN
+ *
+ * req		provides the initial values of nmreq before parsing ifname.
+ *		Remember that the ifname parsing will override the ring
+ *		number in nm_ringid, and part of nm_flags;
+ * flags	special functions, normally 0
+ *		indicates which fields of *arg are significant
+ * arg		special functions, normally NULL
+ *		if passed a netmap_desc with mem != NULL,
+ *		use that memory instead of mmap.
+ */
+ 
+static struct nm_desc *nm_open(const char *ifname, const struct nmreq *req,
+	uint64_t flags, const struct nm_desc *arg);
+
+/*
+ * nm_open can import some fields from the parent descriptor.
+ * These flags control which ones.
+ * Also in flags you can specify NETMAP_NO_TX_POLL and NETMAP_DO_RX_POLL,
+ * which set the initial value for these flags.
+ * Note that the 16 low bits of the flags are reserved for data
+ * that may go into the nmreq.
+ */
+enum {
+	NM_OPEN_NO_MMAP =	0x040000, /* reuse mmap from parent */
+	NM_OPEN_IFNAME =	0x080000, /* nr_name, nr_ringid, nr_flags */
+	NM_OPEN_ARG1 =		0x100000,
+	NM_OPEN_ARG2 =		0x200000,
+	NM_OPEN_ARG3 =		0x400000,
+	NM_OPEN_RING_CFG =	0x800000, /* tx|rx rings|slots */
+};
+
+
+/*
+ * nm_close()	closes and restores the port to its previous state
+ */
+
+static int nm_close(struct nm_desc *);
+
+/*
+ * nm_inject() is the same as pcap_inject()
+ * nm_dispatch() is the same as pcap_dispatch()
+ * nm_nextpkt() is the same as pcap_next()
+ */
+
+static int nm_inject(struct nm_desc *, const void *, size_t);
+static int nm_dispatch(struct nm_desc *, int, nm_cb_t, u_char *);
+static u_char *nm_nextpkt(struct nm_desc *, struct nm_pkthdr *);
+
+
+/*
+ * Try to open, return descriptor if successful, NULL otherwise.
+ * An invalid netmap name will return errno = 0;
+ * You can pass a pointer to a pre-filled nm_desc to add special
+ * parameters. Flags is used as follows
+ * NM_OPEN_NO_MMAP	use the memory from arg, only
+ *			if the nr_arg2 (memory block) matches.
+ * NM_OPEN_ARG1		use req.nr_arg1 from arg
+ * NM_OPEN_ARG2		use req.nr_arg2 from arg
+ * NM_OPEN_RING_CFG	user ring config from arg
+ */
+static struct nm_desc *
+nm_open(const char *ifname, const struct nmreq *req,
+	uint64_t new_flags, const struct nm_desc *arg)
+{
+	struct nm_desc *d = NULL;
+	const struct nm_desc *parent = arg;
+	u_int namelen;
+	uint32_t nr_ringid = 0, nr_flags;
+	const char *port = NULL;
+	const char *errmsg = NULL;
+
+	if (strncmp(ifname, "netmap:", 7) && strncmp(ifname, "vale", 4)) {
+		errno = 0; /* name not recognised, not an error */
+		return NULL;
+	}
+	if (ifname[0] == 'n')
+		ifname += 7;
+	/* scan for a separator */
+	for (port = ifname; *port && !index("-*^{}", *port); port++)
+		;
+	namelen = port - ifname;
+	if (namelen >= sizeof(d->req.nr_name)) {
+		errmsg = "name too long";
+		goto fail;
+	}
+	switch (*port) {
+	default:  /* '\0', no suffix */
+		nr_flags = NR_REG_ALL_NIC;
+		break;
+	case '-': /* one NIC */
+		nr_flags = NR_REG_ONE_NIC;
+		nr_ringid = atoi(port + 1);
+		break;
+	case '*': /* NIC and SW, ignore port */
+		nr_flags = NR_REG_NIC_SW;
+		if (port[1]) {
+			errmsg = "invalid port for nic+sw";
+			goto fail;
+		}
+		break;
+	case '^': /* only sw ring */
+		nr_flags = NR_REG_SW;
+		if (port[1]) {
+			errmsg = "invalid port for sw ring";
+			goto fail;
+		}
+		break;
+	case '{':
+		nr_flags = NR_REG_PIPE_MASTER;
+		nr_ringid = atoi(port + 1);
+		break;
+	case '}':
+		nr_flags = NR_REG_PIPE_SLAVE;
+		nr_ringid = atoi(port + 1);
+		break;
+	}
+
+	if (nr_ringid >= NETMAP_RING_MASK) {
+		errmsg = "invalid ringid";
+		goto fail;
+	}
+	/* add the *XPOLL flags */
+	nr_ringid |= new_flags & (NETMAP_NO_TX_POLL | NETMAP_DO_RX_POLL);
+
+	d = (struct nm_desc *)calloc(1, sizeof(*d));
+	if (d == NULL) {
+		errmsg = "nm_desc alloc failure";
+		errno = ENOMEM;
+		return NULL;
+	}
+	d->self = d;	/* set this early so nm_close() works */
+	d->fd = open("/dev/netmap", O_RDWR);
+	if (d->fd < 0) {
+		errmsg = "cannot open /dev/netmap";
+		goto fail;
+	}
+
+	if (req)
+		d->req = *req;
+	d->req.nr_version = NETMAP_API;
+	d->req.nr_ringid &= ~NETMAP_RING_MASK;
+
+	/* these fields are overridden by ifname and flags processing */
+	d->req.nr_ringid |= nr_ringid;
+	d->req.nr_flags = nr_flags;
+	memcpy(d->req.nr_name, ifname, namelen);
+	d->req.nr_name[namelen] = '\0';
+	/* optionally import info from parent */
+	if (IS_NETMAP_DESC(parent) && new_flags) {
+		if (new_flags & NM_OPEN_ARG1)
+			D("overriding ARG1 %d", parent->req.nr_arg1);
+		d->req.nr_arg1 = new_flags & NM_OPEN_ARG1 ?
+			parent->req.nr_arg1 : 4;
+		if (new_flags & NM_OPEN_ARG2)
+			D("overriding ARG2 %d", parent->req.nr_arg2);
+		d->req.nr_arg2 = new_flags & NM_OPEN_ARG2 ?
+			parent->req.nr_arg2 : 0;
+		if (new_flags & NM_OPEN_ARG3)
+			D("overriding ARG3 %d", parent->req.nr_arg3);
+		d->req.nr_arg3 = new_flags & NM_OPEN_ARG3 ?
+			parent->req.nr_arg3 : 0;
+		if (new_flags & NM_OPEN_RING_CFG) {
+			D("overriding RING_CFG");
+			d->req.nr_tx_slots = parent->req.nr_tx_slots;
+			d->req.nr_rx_slots = parent->req.nr_rx_slots;
+			d->req.nr_tx_rings = parent->req.nr_tx_rings;
+			d->req.nr_rx_rings = parent->req.nr_rx_rings;
+		}
+		if (new_flags & NM_OPEN_IFNAME) {
+			D("overriding ifname %s ringid 0x%x flags 0x%x",
+				parent->req.nr_name, parent->req.nr_ringid,
+				parent->req.nr_flags);
+			memcpy(d->req.nr_name, parent->req.nr_name,
+				sizeof(d->req.nr_name));
+			d->req.nr_ringid = parent->req.nr_ringid;
+			d->req.nr_flags = parent->req.nr_flags;
+		}
+	}
+	if (ioctl(d->fd, NIOCREGIF, &d->req)) {
+		errmsg = "NIOCREGIF failed";
+		goto fail;
+	}
+
+	if (IS_NETMAP_DESC(parent) && parent->mem &&
+	    parent->req.nr_arg2 == d->req.nr_arg2) {
+		/* do not mmap, inherit from parent */
+		d->memsize = parent->memsize;
+		d->mem = parent->mem;
+	} else {
+		d->memsize = d->req.nr_memsize;
+		d->mem = mmap(0, d->memsize, PROT_WRITE | PROT_READ, MAP_SHARED,
+				d->fd, 0);
+		if (d->mem == NULL) {
+			errmsg = "mmap failed";
+			goto fail;
+		}
+		d->done_mmap = 1;
+	}
+	{
+		struct netmap_if *nifp = NETMAP_IF(d->mem, d->req.nr_offset);
+		struct netmap_ring *r = NETMAP_RXRING(nifp, );
+
+		*(struct netmap_if **)(uintptr_t)&(d->nifp) = nifp;
+		*(struct netmap_ring **)(uintptr_t)&d->some_ring = r;
+		*(void **)(uintptr_t)&d->buf_start = NETMAP_BUF(r, 0);
+		*(void **)(uintptr_t)&d->buf_end =
+			(char *)d->mem + d->memsize;
+	}
+
+	if (nr_flags ==  NR_REG_SW) { /* host stack */
+		d->first_tx_ring = d->last_tx_ring = d->req.nr_tx_rings;
+		d->first_rx_ring = d->last_rx_ring = d->req.nr_rx_rings;
+	} else if (nr_flags ==  NR_REG_ALL_NIC) { /* only nic */
+		d->first_tx_ring = 0;
+		d->first_rx_ring = 0;
+		d->last_tx_ring = d->req.nr_tx_rings - 1;
+		d->last_rx_ring = d->req.nr_rx_rings - 1;
+	} else if (nr_flags ==  NR_REG_NIC_SW) {
+		d->first_tx_ring = 0;
+		d->first_rx_ring = 0;
+		d->last_tx_ring = d->req.nr_tx_rings;
+		d->last_rx_ring = d->req.nr_rx_rings;
+	} else if (nr_flags == NR_REG_ONE_NIC) {
+		/* XXX check validity */
+		d->first_tx_ring = d->last_tx_ring =
+		d->first_rx_ring = d->last_rx_ring = nr_ringid;
+	} else { /* pipes */
+		d->first_tx_ring = d->last_tx_ring = 0;
+		d->first_rx_ring = d->last_rx_ring = 0;
+	}
+
+#ifdef DEBUG_NETMAP_USER
+    { /* debugging code */
+	int i;
+
+	D("%s tx %d .. %d %d rx %d .. %d %d", ifname,
+		d->first_tx_ring, d->last_tx_ring, d->req.nr_tx_rings,
+                d->first_rx_ring, d->last_rx_ring, d->req.nr_rx_rings);
+	for (i = 0; i <= d->req.nr_tx_rings; i++) {
+		struct netmap_ring *r = NETMAP_TXRING(d->nifp, i);
+		D("TX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
+	}
+	for (i = 0; i <= d->req.nr_rx_rings; i++) {
+		struct netmap_ring *r = NETMAP_RXRING(d->nifp, i);
+		D("RX%d %p h %d c %d t %d", i, r, r->head, r->cur, r->tail);
+	}
+    }
+#endif /* debugging */
+		
+	d->cur_tx_ring = d->first_tx_ring;
+	d->cur_rx_ring = d->first_rx_ring;
+	return d;
+
+fail:
+	nm_close(d);
+	if (errmsg)
+		D("%s %s", errmsg, ifname);
+	errno = EINVAL;
+	return NULL;
+}
+
+
+static int
+nm_close(struct nm_desc *d)
+{
+	/*
+	 * ugly trick to avoid unused warnings
+	 */
+	static void *__xxzt[] __attribute__ ((unused))  =
+		{ (void *)nm_open, (void *)nm_inject,
+		  (void *)nm_dispatch, (void *)nm_nextpkt } ;
+
+	if (d == NULL || d->self != d)
+		return EINVAL;
+	if (d->done_mmap && d->mem)
+		munmap(d->mem, d->memsize);
+	if (d->fd != -1)
+		close(d->fd);
+	bzero(d, sizeof(*d));
+	free(d);
+	return 0;
+}
+
+
+/*
+ * Same prototype as pcap_inject(), only need to cast.
+ */
+static int
+nm_inject(struct nm_desc *d, const void *buf, size_t size)
+{
+	u_int c, n = d->last_tx_ring - d->first_tx_ring + 1;
+
+	for (c = 0; c < n ; c++) {
+		/* compute current ring to use */
+		struct netmap_ring *ring;
+		uint32_t i, idx;
+		uint32_t ri = d->cur_tx_ring + c;
+
+		if (ri > d->last_tx_ring)
+			ri = d->first_tx_ring;
+		ring = NETMAP_TXRING(d->nifp, ri);
+		if (nm_ring_empty(ring)) {
+			continue;
+		}
+		i = ring->cur;
+		idx = ring->slot[i].buf_idx;
+		ring->slot[i].len = size;
+		nm_pkt_copy(buf, NETMAP_BUF(ring, idx), size);
+		d->cur_tx_ring = ri;
+		ring->head = ring->cur = nm_ring_next(ring, i);
+		return size;
+	}
+	return 0; /* fail */
+}
+
+
+/*
+ * Same prototype as pcap_dispatch(), only need to cast.
+ */
+static int
+nm_dispatch(struct nm_desc *d, int cnt, nm_cb_t cb, u_char *arg)
+{
+	int n = d->last_rx_ring - d->first_rx_ring + 1;
+	int c, got = 0, ri = d->cur_rx_ring;
+
+	if (cnt == 0)
+		cnt = -1;
+	/* cnt == -1 means infinite, but rings have a finite amount
+	 * of buffers and the int is large enough that we never wrap,
+	 * so we can omit checking for -1
+	 */
+	for (c=0; c < n && cnt != got; c++) {
+		/* compute current ring to use */
+		struct netmap_ring *ring;
+
+		ri = d->cur_rx_ring + c;
+		if (ri > d->last_rx_ring)
+			ri = d->first_rx_ring;
+		ring = NETMAP_RXRING(d->nifp, ri);
+		for ( ; !nm_ring_empty(ring) && cnt != got; got++) {
+			u_int i = ring->cur;
+			u_int idx = ring->slot[i].buf_idx;
+			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
+
+			// __builtin_prefetch(buf);
+			d->hdr.len = d->hdr.caplen = ring->slot[i].len;
+			d->hdr.ts = ring->ts;
+			cb(arg, &d->hdr, buf);
+			ring->head = ring->cur = nm_ring_next(ring, i);
+		}
+	}
+	d->cur_rx_ring = ri;
+	return got;
+}
+
+static u_char *
+nm_nextpkt(struct nm_desc *d, struct nm_pkthdr *hdr)
+{
+	int ri = d->cur_rx_ring;
+
+	do {
+		/* compute current ring to use */
+		struct netmap_ring *ring = NETMAP_RXRING(d->nifp, ri);
+		if (!nm_ring_empty(ring)) {
+			u_int i = ring->cur;
+			u_int idx = ring->slot[i].buf_idx;
+			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
+
+			// __builtin_prefetch(buf);
+			hdr->ts = ring->ts;
+			hdr->len = hdr->caplen = ring->slot[i].len;
+			ring->cur = nm_ring_next(ring, i);
+			/* we could postpone advancing head if we want
+			 * to hold the buffer. This can be supported in
+			 * the future.
+			 */
+			ring->head = ring->cur;
+			d->cur_rx_ring = ri;
+			return buf;
+		}
+		ri++;
+		if (ri > d->last_rx_ring)
+			ri = d->first_rx_ring;
+	} while (ri != d->cur_rx_ring);
+	return NULL; /* nothing found */
+}
+
+#endif /* !HAVE_NETMAP_WITH_LIBS */
+
+#endif /* NETMAP_WITH_LIBS */
 
 #endif /* _NET_NETMAP_USER_H_ */