A netgraph node implementing Netflow version 5.

Supported by:	Bestcom ISP, Rinet ISP
Approved by:	julian (mentor)

4 files changed, 1670 insertions, 0 deletions

diff --git a/sys/netgraph/netflow/netflow.c b/sys/netgraph/netflow/netflow.c
new file mode 100644
index 0000000..1075838
--- /dev/null
+++ b/sys/netgraph/netflow/netflow.c
@@ -0,0 +1,736 @@
+/*-
+ * Copyright (c) 2004 Gleb Smirnoff <glebius@cell.sick.ru>
+ * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * 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. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ * This product includes software developed by Gleb Smirnoff and
+ * contributors.
+ * 4. Neither the name of the author nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * 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 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.
+ *
+ * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
+ */
+
+static const char rcs_id[] =
+    "@(#) $FreeBSD$";
+
+#include <sys/param.h>
+#include <sys/kernel.h>
+#include <sys/limits.h>
+#include <sys/mbuf.h>
+#include <sys/systm.h>
+#include <sys/socket.h>
+
+#include <net/if.h>
+#include <net/if_var.h>
+#include <net/if_dl.h>
+#include <net/route.h>
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+
+#include <netgraph/ng_message.h>
+#include <netgraph/netgraph.h>
+
+#include <netgraph/netflow/netflow.h>
+#include <netgraph/netflow/ng_netflow.h>
+
+#define	NBUCKETS	(4096)	/* must be power of 2 */
+
+/* This hash is for TCP or UDP packets */
+#define FULL_HASH(addr1,addr2,port1,port2)\
+	(((addr1 >> 16) ^		\
+	  (addr2 & 0x00FF) ^		\
+	  ((port1 ^ port2) << 8) )&	\
+	 (NBUCKETS - 1))
+
+/* This hash for all other IP packets */
+#define ADDR_HASH(addr1,addr2)\
+	(((addr1 >> 16) ^		\
+	  (addr2 & 0x00FF) )&		\
+	 (NBUCKETS - 1))
+
+/* Macros to shorten logical constructions */
+/* XXX: priv must exist in namespace */
+#define	INACTIVE(fle)	(time_uptime - fle->f.last > priv->info.nfinfo_inact_t)
+#define	AGED(fle)	(time_uptime - fle->f.first > priv->info.nfinfo_act_t)
+#define	ISFREE(fle)	(fle->f.packets == 0)
+
+/*
+ * 4 is a magical number: statistically number of 4-packet flows is
+ * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
+ * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
+ * of reachable host and 4-packet otherwise.
+ */
+#define	SMALL(fle)	(fle->f.packets <= 4)
+	
+MALLOC_DECLARE(M_NETFLOW);
+MALLOC_DEFINE(M_NETFLOW, "NetFlow", "flow cache");
+
+static int export_add(priv_p , struct flow_entry *);
+static int export_send(priv_p );
+
+/* Generate hash for a given flow record */
+static __inline uint32_t
+ip_hash(struct flow_rec *r)
+{
+	switch (r->r_ip_p) {
+	case IPPROTO_TCP:
+	case IPPROTO_UDP:
+		return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
+		    r->r_sport, r->r_dport);
+	default:
+		return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
+	}
+}
+
+/* Lookup for record in given slot */
+static __inline struct flow_entry *
+hash_lookup(struct flow_hash_entry *h, int slot, struct flow_rec *r)
+{
+	struct flow_entry *fle;
+
+	LIST_FOREACH(fle, &(h[slot].head), fle_hash)
+		if (bcmp(r, &fle->f.r, sizeof(struct flow_rec)) == 0)
+			return (fle);
+
+	return (NULL);
+}
+
+/* Get a flow entry from free list */
+static __inline struct flow_entry *
+alloc_flow(priv_p priv, int *flows)
+{
+	register struct flow_entry	*fle;
+
+	mtx_lock(&priv->free_mtx);
+
+	if (SLIST_EMPTY(&priv->free_list)) {
+		mtx_unlock(&priv->free_mtx);
+		return(NULL);
+	}
+
+	fle = SLIST_FIRST(&priv->free_list);
+	SLIST_REMOVE_HEAD(&priv->free_list, fle_free);
+
+	priv->info.nfinfo_used++;
+	priv->info.nfinfo_free--;
+
+	if (flows != NULL)
+		*flows = priv->info.nfinfo_used;
+
+	mtx_unlock(&priv->free_mtx);
+
+	return (fle);
+}
+
+/* Insert flow entry into a free list. */
+static __inline int
+free_flow(priv_p priv, struct flow_entry *fle)
+{
+	int flows;
+
+	mtx_lock(&priv->free_mtx);
+	fle->f.packets = 0;
+	SLIST_INSERT_HEAD(&priv->free_list, fle, fle_free);
+	flows = priv->info.nfinfo_used--;
+	priv->info.nfinfo_free++;
+	mtx_unlock(&priv->free_mtx);
+
+	return flows;
+}
+
+#define	NGNF_GETUSED(priv, rval)	do {	\
+	mtx_lock(&priv->free_mtx);		\
+	rval = priv->info.nfinfo_used;		\
+	mtx_unlock(&priv->free_mtx);		\
+	} while (0)
+
+/* Insert flow entry into expire list. */
+/* XXX: Flow must be detached from work queue, but not from cache */
+static __inline void
+expire_flow(priv_p priv, struct flow_entry *fle)
+{
+	mtx_assert(&priv->work_mtx, MA_OWNED);
+	LIST_REMOVE(fle, fle_hash);
+
+	mtx_lock(&priv->expire_mtx);
+	SLIST_INSERT_HEAD(&priv->expire_list, fle, fle_free);
+	mtx_unlock(&priv->expire_mtx);
+}
+
+/* Get a snapshot of node statistics */
+void
+ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
+{
+	mtx_lock(&priv->free_mtx);
+	memcpy((void *)i, (void *)&priv->info, sizeof(priv->info));
+	mtx_unlock(&priv->free_mtx);
+}
+
+/* Calculate number of bits in netmask */
+#define	g21	0x55555555ul	/* = 0101_0101_0101_0101_0101_0101_0101_0101 */
+#define	g22	0x33333333ul	/* = 0011_0011_0011_0011_0011_0011_0011_0011 */
+#define	g23	0x0f0f0f0ful	/* = 0000_1111_0000_1111_0000_1111_0000_1111 */
+static __inline u_char
+bit_count(uint32_t v)
+{
+	v = (v & g21) + ((v >> 1) & g21);
+	v = (v & g22) + ((v >> 2) & g22);
+	v = (v + (v >> 4)) & g23;
+	return (v + (v >> 8) + (v >> 16) + (v >> 24)) & 0x3f;
+}
+
+/*
+ * Insert a record into defined slot.
+ *
+ * First we get for us a free flow entry, then fill in all
+ * possible fields in it. Then obtain lock on flow cache
+ * and insert flow entry.
+ */
+static __inline int
+hash_insert(priv_p priv, int slot, struct flow_rec *r, int plen)
+{
+	struct flow_hash_entry	*h = priv->hash;
+	struct flow_entry	*fle;
+	struct route ro;
+	struct sockaddr_in *sin;
+
+	fle = alloc_flow(priv, NULL);
+	if (fle == NULL)
+		return (ENOMEM);
+
+	/*
+	 * Now fle is totally ours. It is detached from all lists,
+	 * we can safely edit it.
+	 */
+
+	bcopy(r, &fle->f.r, sizeof(struct flow_rec));
+	fle->f.bytes = plen;
+	fle->f.packets = 1;
+
+	priv->info.nfinfo_bytes += plen;
+
+	fle->f.first = fle->f.last = time_uptime;
+
+	/*
+	 * First we do route table lookup on destination address. So we can
+	 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
+	 */
+	bzero((caddr_t)&ro, sizeof(ro));
+	sin = (struct sockaddr_in *)&ro.ro_dst;
+	sin->sin_len = sizeof(*sin);
+	sin->sin_family = AF_INET;
+	sin->sin_addr = fle->f.r.r_dst;
+	rtalloc_ign(&ro, RTF_CLONING);
+	if (ro.ro_rt != NULL) {
+		struct rtentry *rt = ro.ro_rt;
+
+		fle->f.fle_o_ifx = rt->rt_ifp->if_index;
+
+		if (rt->rt_flags & RTF_GATEWAY &&
+		    rt->rt_gateway->sa_family == AF_INET)
+			fle->f.next_hop =
+			    ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
+
+		if (rt_mask(rt))
+			fle->f.dst_mask =
+			    bit_count(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
+		else if (rt->rt_flags & RTF_HOST)
+			/* Give up. We can't determine mask :( */
+			fle->f.dst_mask = 32;
+
+		RTFREE(ro.ro_rt);
+	}
+
+	/* Do route lookup on source address, to fill in src_mask. */
+
+	bzero((caddr_t)&ro, sizeof(ro));
+	sin = (struct sockaddr_in *)&ro.ro_dst;
+	sin->sin_len = sizeof(*sin);
+	sin->sin_family = AF_INET;
+	sin->sin_addr = fle->f.r.r_src;
+	rtalloc_ign(&ro, RTF_CLONING);
+	if (ro.ro_rt != NULL) {
+		struct rtentry *rt = ro.ro_rt;
+
+		if (rt_mask(rt))
+			fle->f.src_mask =
+			    bit_count(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
+		else if (rt->rt_flags & RTF_HOST)
+			/* Give up. We can't determine mask :( */
+			fle->f.src_mask = 32;
+
+		RTFREE(ro.ro_rt);
+	}
+
+	/* Push new flow entry into flow cache */
+	mtx_lock(&priv->work_mtx);
+	LIST_INSERT_HEAD(&(h[slot].head), fle, fle_hash);
+	TAILQ_INSERT_TAIL(&priv->work_queue, fle, fle_work);
+	mtx_unlock(&priv->work_mtx);
+
+	return (0);
+}
+
+static __inline int
+make_flow_rec(struct mbuf **m, int *plen, struct flow_rec *r, uint8_t *tcp_flags,
+	u_int16_t i_ifx)
+{
+	register struct ip *ip;
+	int hlen;
+	int error = 0;
+
+	ip = mtod(*m, struct ip*);
+
+	/* check version */
+	if (ip->ip_v != IPVERSION)
+		return (EINVAL);
+
+	/* verify min header length */
+	hlen = ip->ip_hl << 2;
+
+	if (hlen < sizeof(struct ip))
+		return (EINVAL);
+
+	r->r_src = ip->ip_src;
+	r->r_dst = ip->ip_dst;
+
+	/* save packet length */
+	*plen = ntohs(ip->ip_len);
+
+	r->r_ip_p = ip->ip_p;
+	r->r_tos = ip->ip_tos;
+
+	if ((*m)->m_pkthdr.rcvif)
+		r->r_i_ifx = (*m)->m_pkthdr.rcvif->if_index;
+	else
+		r->r_i_ifx = i_ifx;
+
+	/*
+	 * XXX NOTE: only first fragment of fragmented TCP, UDP and
+	 * ICMP packet will be recorded with proper s_port and d_port.
+	 * Following fragments will be recorded simply as IP packet with
+	 * ip_proto = ip->ip_p and s_port, d_port set to zero.
+	 * I know, it looks like bug. But I don't want to re-implement
+	 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
+	 * and nobody complains yet :)
+	 */
+	if(ip->ip_off & htons(IP_OFFMASK))
+		return (0);
+
+	/* skip IP header */
+	m_adj(*m, hlen);
+
+	switch(r->r_ip_p) {
+	case IPPROTO_TCP:
+	{
+		register struct tcphdr *tcp;
+
+		/* verify that packet is not truncated */
+		if (CHECK_MLEN(*m, sizeof(struct tcphdr)))
+			ERROUT(EINVAL);
+
+		if (CHECK_PULLUP(*m, sizeof(struct tcphdr)))
+			ERROUT(ENOBUFS);
+
+		tcp = mtod(*m, struct tcphdr*);
+		r->r_sport = tcp->th_sport;
+		r->r_dport = tcp->th_dport;
+		*tcp_flags = tcp->th_flags;
+		break;
+	}
+	case IPPROTO_UDP:
+		/* verify that packet is not truncated */
+		if (CHECK_MLEN(*m, sizeof(struct udphdr)))
+			ERROUT(EINVAL);
+
+		if (CHECK_PULLUP(*m, sizeof(struct udphdr)))
+			ERROUT(ENOBUFS);
+
+		r->r_ports = *(mtod(*m, uint32_t *));
+		break;
+	}
+
+done:
+	return (error);
+}
+
+/*
+ * Non-static functions called from ng_netflow.c
+ */
+
+/* Allocate memory and set up flow cache */
+int
+ng_netflow_cache_init(priv_p priv)
+{
+	struct flow_entry *fle;
+	int i;
+
+	/* allocate cache */
+	MALLOC(priv->cache, struct flow_entry *,
+	    CACHESIZE * sizeof(struct flow_entry),
+	    M_NETFLOW, M_WAITOK | M_ZERO);
+
+	if (priv->cache == NULL)
+		return (ENOMEM);
+
+	/* allocate hash */
+	MALLOC(priv->hash, struct flow_hash_entry *,
+	    NBUCKETS * sizeof(struct flow_hash_entry),
+	    M_NETFLOW, M_WAITOK | M_ZERO);
+
+	if (priv->hash == NULL)
+		return (ENOMEM);
+
+	TAILQ_INIT(&priv->work_queue);
+	SLIST_INIT(&priv->free_list);
+	SLIST_INIT(&priv->expire_list);
+
+	mtx_init(&priv->work_mtx, "ng_netflow cache mutex", NULL, MTX_DEF);
+	mtx_init(&priv->free_mtx, "ng_netflow free mutex", NULL, MTX_DEF);
+	mtx_init(&priv->expire_mtx, "ng_netflow expire mutex", NULL, MTX_DEF);
+
+	/* build free list */
+	for (i = 0, fle = priv->cache; i < CACHESIZE; i++, fle++)
+		SLIST_INSERT_HEAD(&priv->free_list, fle, fle_free);
+
+	priv->info.nfinfo_free = CACHESIZE;
+
+	return (0);
+}
+
+/* Free all flow cache memory. Called from node close method. */
+void
+ng_netflow_cache_flush(priv_p priv)
+{
+	register struct flow_entry	*fle;
+	int i;
+
+	/*
+	 * We are going to free probably billable data.
+	 * Expire everything before freeing it.
+	 * No locking is required since callout is already drained.
+	 */
+
+	for (i = 0, fle = priv->cache; i < CACHESIZE; i++, fle++)
+		if (!ISFREE(fle))
+			/* ignore errors now */
+			(void )export_add(priv, fle);
+
+	mtx_destroy(&priv->work_mtx);
+	mtx_destroy(&priv->free_mtx);
+	mtx_destroy(&priv->expire_mtx);
+
+	/* free hash memory */
+	if (priv->hash)
+		FREE(priv->hash, M_NETFLOW);
+
+	/* free flow cache */
+	if (priv->cache)
+		FREE(priv->cache, M_NETFLOW);
+
+}
+
+/* Insert packet from &m into flow cache. */
+int
+ng_netflow_flow_add(priv_p priv, struct mbuf **m, iface_p iface)
+{
+	struct flow_hash_entry		*h = priv->hash;
+	register struct flow_entry	*fle;
+	struct flow_rec		r;
+	int			plen;
+	int			error = 1;
+	uint32_t		slot;
+	uint8_t			tcp_flags = 0;
+
+	priv->info.nfinfo_packets ++;
+
+	/* Try to fill *rec */
+	bzero(&r, sizeof(r));
+	if ((error = make_flow_rec(m, &plen, &r, &tcp_flags, iface->info.ifinfo_index)))
+		return (error);
+
+	slot = ip_hash(&r);
+
+	mtx_lock(&priv->work_mtx);
+	fle = hash_lookup(h, slot, &r); /* New flow entry or existent? */
+
+	if (fle) {	/* an existent entry */
+
+		TAILQ_REMOVE(&priv->work_queue, fle, fle_work);
+
+		fle->f.bytes += plen;
+		fle->f.packets ++;
+		fle->f.tcp_flags |= tcp_flags;
+		fle->f.last = time_uptime;
+
+		/*
+		 * We have the following reasons to expire flow in active way:
+		 * - it hit active timeout
+		 * - a TCP connection closed
+		 * - it is going to overflow counter
+		 */
+		if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
+		    (fle->f.bytes >= (UINT_MAX - IF_MAXMTU)) )
+			expire_flow(priv, fle);
+		else
+			TAILQ_INSERT_TAIL(&priv->work_queue, fle, fle_work);
+
+		mtx_unlock(&priv->work_mtx);
+		priv->info.nfinfo_bytes += plen;
+
+	} else {	/* a new flow entry */
+
+		mtx_unlock(&priv->work_mtx);
+		return hash_insert(priv, slot, &r, plen);
+
+	}
+
+	mtx_assert(&priv->work_mtx, MA_NOTOWNED);
+	mtx_assert(&priv->expire_mtx, MA_NOTOWNED);
+	mtx_assert(&priv->free_mtx, MA_NOTOWNED);
+	return (0);
+}
+
+/*
+ * Return records from cache. netgraph(4) guarantees us that we
+ * are locked against ng_netflow_rcvdata(). However we can
+ * work with ng_netflow_expire() in parrallel. XXX: Is it dangerous?
+ *
+ * TODO: matching particular IP should be done in kernel, here.
+ */
+int
+ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp)
+{
+	struct flow_entry *fle;
+	struct ngnf_flows *data;
+
+	data = (struct ngnf_flows *)resp->data;
+	data->last = 0;
+	data->nentries = 0;
+
+	/* Check if this is a first run */
+	if (last == 0)
+		fle = priv->cache;
+	else {
+		if (last > CACHESIZE-1)
+			return (EINVAL);
+		fle = priv->cache + last;
+	}
+
+	/*
+	 * We will transfer not more than NREC_AT_ONCE. More data
+	 * will come in next message.
+	 * We send current stop point to userland, and userland should return
+	 * it back to us.
+	 */
+	for (; last < CACHESIZE; fle++, last++) {
+		if (ISFREE(fle))
+			continue;
+		bcopy(&fle->f, &(data->entries[data->nentries]), sizeof(fle->f));
+		data->nentries ++;
+		if (data->nentries == NREC_AT_ONCE) {
+			if (++last < CACHESIZE)
+				data->last = (++fle - priv->cache);
+			return (0);
+		}
+     	}
+
+	return (0);
+}
+
+/* We have full datagram in privdata. Send it to export hook. */
+static int
+export_send(priv_p priv)
+{
+	struct netflow_v5_header *header = &priv->dgram.header;
+	struct timespec ts;
+	struct mbuf *m;
+	int error = 0;
+	int mlen;
+
+	header->sys_uptime = htonl(time_uptime);
+
+	getnanotime(&ts);
+	header->unix_secs  = htonl(ts.tv_sec);
+	header->unix_nsecs = htonl(ts.tv_nsec);
+
+	/* Flow sequence contains number of first record */
+	header->flow_seq = htonl(priv->flow_seq - header->count);
+
+	mlen = sizeof(struct netflow_v5_header) +
+	    sizeof(struct netflow_v5_record) * header->count;
+
+	header->count = htons(header->count);
+	if ((m = m_devget((caddr_t)header, mlen, 0, NULL, NULL)) == NULL) {
+		printf("ng_netflow: m_devget() failed, losing export dgram\n");
+		header->count = 0;
+		return(ENOBUFS);
+	}
+
+	header->count = 0;
+
+	/* Giant is required in sosend() at this moment. */
+	NET_LOCK_GIANT();
+	NG_SEND_DATA_ONLY(error, priv->export, m);
+	NET_UNLOCK_GIANT();
+
+	if (error)
+		NG_FREE_M(m);
+
+	return (error);
+}
+
+
+/* Create export datagram. */
+static int
+export_add(priv_p priv, struct flow_entry *fle)
+{
+	struct netflow_v5_header *header = &priv->dgram.header;
+	struct netflow_v5_record *rec;
+
+	if (header->count == 0 ) {	/* first record */
+		rec = &priv->dgram.r[0];
+		header->count = 1;
+	} else {			/* continue filling datagram */
+		rec = &priv->dgram.r[header->count];
+		header->count ++;
+	}
+
+	/* Fill in export record */
+	rec->src_addr = fle->f.r.r_src.s_addr;
+	rec->dst_addr = fle->f.r.r_dst.s_addr;
+	rec->next_hop = fle->f.next_hop.s_addr;
+	rec->i_ifx    = htons(fle->f.fle_i_ifx);
+	rec->o_ifx    = htons(fle->f.fle_o_ifx);
+	rec->packets  = htonl(fle->f.packets);
+	rec->octets   = htonl(fle->f.bytes);
+	rec->first    = htonl(fle->f.first);
+	rec->last     = htonl(fle->f.last);
+	rec->s_port   = fle->f.r.r_sport;
+	rec->d_port   = fle->f.r.r_dport;
+	rec->flags    = fle->f.tcp_flags;
+	rec->prot     = fle->f.r.r_ip_p;
+	rec->tos      = fle->f.r.r_tos;
+	rec->dst_mask = fle->f.dst_mask;
+	rec->src_mask = fle->f.src_mask;
+
+	priv->flow_seq++;
+
+	if (header->count == NETFLOW_V5_MAX_RECORDS) /* end of datagram */
+		return export_send(priv);
+
+	return (0);	
+}
+
+/* Periodic flow expiry run. */
+void
+ng_netflow_expire(void *arg)
+{
+	register struct flow_entry	*fle, *fle1;
+	priv_p priv = (priv_p )arg;
+	uint32_t used;
+	int error = 0;
+
+	/* First pack actively expired entries */
+	mtx_lock(&priv->expire_mtx);
+	while (!SLIST_EMPTY(&(priv->expire_list))) {
+		fle = SLIST_FIRST(&(priv->expire_list));
+		SLIST_REMOVE_HEAD(&(priv->expire_list), fle_free);
+		mtx_unlock(&priv->expire_mtx);
+
+		/*
+		 * While we have dropped the lock, expire_flow() may
+		 * insert another flow into top of the list.
+		 * This is not harmful for us, since we have already
+		 * detached our own.
+		 */
+
+		if ((error = export_add(priv, fle)) != 0)
+			printf("ng_netflow: export_add() failed: %u\n", error);
+		(void )free_flow(priv, fle);
+
+		mtx_lock(&priv->expire_mtx);
+	}
+	mtx_unlock(&priv->expire_mtx);
+
+	NGNF_GETUSED(priv, used);
+	mtx_lock(&priv->work_mtx);
+	TAILQ_FOREACH_SAFE(fle, &(priv->work_queue), fle_work, fle1) {
+		/*
+		 * When cache size has not reached CACHELOWAT yet, we keep both
+		 * inactive and active flows in cache. Doing this, we reduce number
+		 * of exports, since many inactive flows may wake up and continue
+		 * their life. However, we make an exclusion for scans. It is very
+		 * rare situation that inactive 1-packet flow will wake up.
+		 * When cache has reached CACHELOWAT, we expire all inactive flows,
+		 * until cache gets of sane size.
+		 * 
+		 * When this record's refcount is > 0, we skip it. (XXX)
+		 */
+		if (used <= CACHELOWAT && !INACTIVE(fle))
+			goto finish;
+
+		if (INACTIVE(fle) && (SMALL(fle) || (used > CACHELOWAT))) {
+
+			/* Detach flow entry from cache */
+			LIST_REMOVE(fle, fle_hash);
+			TAILQ_REMOVE(&priv->work_queue, fle, fle_work);
+
+			/*
+			 * While we are sending to collector, unlock cache.
+			 * XXX: it can happen, however with a small probability,
+			 * that item, we are holding now, can be moved to the top
+			 * of flow cache by node thread. In this case our expire
+			 * thread stops checking. Since this is not fatal we will
+			 * just ignore it now.
+			 */
+			mtx_unlock(&priv->work_mtx);
+
+			if ((error = export_add(priv, fle)) != 0)
+				printf("ng_netflow: export_add() failed: %u\n",
+				    error);
+
+			used = free_flow(priv, fle);
+
+			mtx_lock(&priv->work_mtx);
+		}
+     	}
+
+finish:
+	mtx_unlock(&priv->work_mtx);
+
+	mtx_assert(&priv->expire_mtx, MA_NOTOWNED);
+	mtx_assert(&priv->free_mtx, MA_NOTOWNED);
+
+	/* schedule next expire */
+	callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
+	    (void *)priv);
+
+}
diff --git a/sys/netgraph/netflow/netflow.h b/sys/netgraph/netflow/netflow.h
new file mode 100644
index 0000000..a530f92
--- /dev/null
+++ b/sys/netgraph/netflow/netflow.h
@@ -0,0 +1,125 @@
+/*-
+ * Copyright (c) 2004 Gleb Smirnoff <glebius@cell.sick.ru>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * 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. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ * This product includes software developed by Gleb Smirnoff and
+ * contributors.
+ * 4. Neither the name of the author nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * 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 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.
+ *
+ *	 $SourceForge: netflow.h,v 1.8 2004/09/16 17:05:11 glebius Exp $
+ *	 $FreeBSD$
+ */
+
+/* netflow timeouts in seconds */
+
+#define	ACTIVE_TIMEOUT		(30*60)	/* maximum flow lifetime is 30 min */
+#define	INACTIVE_TIMEOUT	15
+
+/* More info can be found in these two Cisco documents:
+ * http://www.cisco.com/warp/public/cc/pd/iosw/ioft/neflct/tech/napps_wp.htm
+ * http://www.cisco.com/en/US/products/sw/netmgtsw/ps1964/prod_installation_guide09186a00800fea56.html#wp1006186
+ * However, they say quite different things.
+ */
+
+#define NETFLOW_V1 1
+#define NETFLOW_V5 5
+
+struct netflow_v1_header
+{
+  uint16_t version;	/* NetFlow version */
+  uint16_t count;	/* Number of records in flow */
+  uint32_t sys_uptime;	/* System uptime */
+  uint32_t unix_secs;	/* Current seconds since 0000 UTC 1970 */
+  uint32_t unix_nsecs;	/* Remaining nanoseconds since 0000 UTC 1970 */
+} __attribute__((__packed__));
+
+struct netflow_v5_header
+{
+  uint16_t version;	/* NetFlow version */
+  uint16_t count;	/* Number of records in flow */
+  uint32_t sys_uptime;	/* System uptime */
+  uint32_t unix_secs;	/* Current seconds since 0000 UTC 1970 */
+  uint32_t unix_nsecs;	/* Remaining nanoseconds since 0000 UTC 1970 */
+  uint32_t flow_seq;	/* Sequence number of the first record */
+  uint8_t engine_type;	/* Type of flow switching engine (RP,VIP,etc.) */
+  uint8_t engine_id;	/* Slot number of the flow switching engine */
+  uint16_t pad;		/* Pad to word boundary */
+} __attribute__((__packed__));
+
+struct netflow_v1_record
+{
+  uint32_t src_addr;	/* Source IP address */
+  uint32_t dst_addr;	/* Destination IP address */
+  uint32_t next_hop;	/* Next hop IP address */
+  uint16_t in_ifx;	/* Source interface index */
+  uint16_t out_ifx;	/* Destination interface index */
+  uint32_t packets;	/* Number of packets in a flow */
+  uint32_t octets;	/* Number of octets in a flow */
+  uint32_t first;	/* System uptime at start of a flow */
+  uint32_t last;	/* System uptime at end of a flow */
+  uint16_t s_port;	/* Source port */
+  uint16_t d_port;	/* Destination port */
+  uint16_t pad1;	/* Pad to word boundary */
+  uint8_t prot;	/* IP protocol */
+  uint8_t tos;		/* IP type of service */
+  uint8_t flags;	/* Cumulative OR of tcp flags */
+  uint8_t pad2;	/* pad to word boundary */
+  uint16_t pad3;	/* Pad to word boundary */
+  uint8_t reserved[5];	/* Reserved for future use */
+} __attribute__((__packed__));
+
+struct netflow_v5_record
+{
+  uint32_t src_addr;	/* Source IP address */
+  uint32_t dst_addr;	/* Destination IP address */
+  uint32_t next_hop;	/* Next hop IP address */
+  uint16_t i_ifx;	/* Source interface index */
+  uint16_t o_ifx;	/* Destination interface index */
+  uint32_t packets;	/* Number of packets in a flow */
+  uint32_t octets;	/* Number of octets in a flow */
+  uint32_t first;	/* System uptime at start of a flow */
+  uint32_t last;	/* System uptime at end of a flow */
+  uint16_t s_port;	/* Source port */
+  uint16_t d_port;	/* Destination port */
+  uint8_t pad1;	/* pad to word boundary */
+  uint8_t flags;	/* Cumulative OR of tcp flags */
+  uint8_t prot;	/* IP protocol */
+  uint8_t tos;		/* IP type of service */
+  uint16_t src_as;	/* Src peer/origin Autonomous System */
+  uint16_t dst_as;	/* Dst peer/origin Autonomous System */
+  uint8_t src_mask;	/* Source route's mask bits */
+  uint8_t dst_mask;	/* Destination route's mask bits */
+  uint16_t pad2;	/* Pad to word boundary */
+} __attribute__((__packed__));
+
+#define NETFLOW_V1_MAX_RECORDS 24
+#define NETFLOW_V5_MAX_RECORDS 30
+
+#define NETFLOW_V1_MAX_SIZE (sizeof(netflow_v1_header)+ \
+			     sizeof(netflow_v1_record)*NETFLOW_V1_MAX_RECORDS)
+#define NETFLOW_V5_MAX_SIZE (sizeof(netflow_v5_header)+ \
+			     sizeof(netflow_v5_record)*NETFLOW_V5_MAX_RECORDS)
diff --git a/sys/netgraph/netflow/ng_netflow.c b/sys/netgraph/netflow/ng_netflow.c
new file mode 100644
index 0000000..4cfbe08
--- /dev/null
+++ b/sys/netgraph/netflow/ng_netflow.c
@@ -0,0 +1,527 @@
+/*-
+ * Copyright (c) 2004 Gleb Smirnoff <glebius@cell.sick.ru>
+ * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * 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. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ * This product includes software developed by Gleb Smirnoff and
+ * contributors.
+ * 4. Neither the name of the author nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * 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 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.
+ *
+ * $SourceForge: ng_netflow.c,v 1.30 2004/09/05 11:37:43 glebius Exp $
+ */
+
+static const char rcs_id[] =
+    "@(#) $FreeBSD$";
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/kernel.h>
+#include <sys/mbuf.h>
+#include <sys/socket.h>
+#include <sys/ctype.h>
+
+#include <net/if.h>
+#include <net/ethernet.h>
+#include <net/if_arp.h>
+#include <net/if_var.h>
+#include <net/bpf.h>
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/ip.h>
+
+#include <netgraph/ng_message.h>
+#include <netgraph/ng_parse.h>
+#include <netgraph/netgraph.h>
+#include <netgraph/netflow/netflow.h>
+#include <netgraph/netflow/ng_netflow.h>
+
+/* Netgraph methods */
+static ng_constructor_t	ng_netflow_constructor;
+static ng_rcvmsg_t	ng_netflow_rcvmsg;
+static ng_close_t	ng_netflow_close;
+static ng_shutdown_t	ng_netflow_rmnode;
+static ng_newhook_t	ng_netflow_newhook;
+static ng_rcvdata_t	ng_netflow_rcvdata;
+static ng_disconnect_t	ng_netflow_disconnect;
+
+/* Parse type for struct ng_netflow_info */
+static const struct ng_parse_struct_field ng_netflow_info_type_fields[]
+	= NG_NETFLOW_INFO_TYPE;
+static const struct ng_parse_type ng_netflow_info_type = {
+	&ng_parse_struct_type,
+	&ng_netflow_info_type_fields
+};
+
+/*  Parse type for struct ng_netflow_ifinfo */
+static const struct ng_parse_struct_field ng_netflow_ifinfo_type_fields[]
+	= NG_NETFLOW_IFINFO_TYPE;
+static const struct ng_parse_type ng_netflow_ifinfo_type = {
+	&ng_parse_struct_type,
+	&ng_netflow_ifinfo_type_fields
+};
+
+/* Parse type for struct ng_netflow_setdlt */
+static const struct ng_parse_struct_field ng_netflow_setdlt_type_fields[]
+	= NG_NETFLOW_SETDLT_TYPE;
+static const struct ng_parse_type ng_netflow_setdlt_type = {
+	&ng_parse_struct_type,
+	&ng_netflow_setdlt_type_fields
+};
+
+/* Parse type for ng_netflow_setifindex */
+static const struct ng_parse_struct_field ng_netflow_setifindex_type_fields[]
+	= NG_NETFLOW_SETIFINDEX_TYPE;
+static const struct ng_parse_type ng_netflow_setifindex_type = {
+	&ng_parse_struct_type,
+	&ng_netflow_setifindex_type_fields
+};
+
+/* Parse type for ng_netflow_settimeouts */
+static const struct ng_parse_struct_field ng_netflow_settimeouts_type_fields[]
+	= NG_NETFLOW_SETTIMEOUTS_TYPE;
+static const struct ng_parse_type ng_netflow_settimeouts_type = {
+	&ng_parse_struct_type,
+	&ng_netflow_settimeouts_type_fields
+};
+
+/* List of commands and how to convert arguments to/from ASCII */
+static const struct ng_cmdlist ng_netflow_cmds[] = {
+       {
+	 NGM_NETFLOW_COOKIE,
+	 NGM_NETFLOW_INFO,
+	 "info",
+	 NULL,
+	 &ng_netflow_info_type
+       },
+       {
+	NGM_NETFLOW_COOKIE,
+	NGM_NETFLOW_IFINFO,
+	"ifinfo",
+	&ng_parse_uint8_type,
+	&ng_netflow_ifinfo_type
+       },
+       {
+	NGM_NETFLOW_COOKIE,
+	NGM_NETFLOW_SETDLT,
+	"setdlt",
+	&ng_netflow_setdlt_type,
+	NULL
+       },
+       {
+	NGM_NETFLOW_COOKIE,
+	NGM_NETFLOW_SETIFINDEX,
+	"setifindex",
+	&ng_netflow_setifindex_type,
+	NULL
+       },
+       {
+	NGM_NETFLOW_COOKIE,
+	NGM_NETFLOW_SETTIMEOUTS,
+	"settimeouts",
+	&ng_netflow_settimeouts_type,
+	NULL
+       },
+       { 0 }
+};
+
+
+/* Netgraph node type descriptor */
+static struct ng_type ng_netflow_typestruct = {
+	.version =	NG_ABI_VERSION,
+	.name =		NG_NETFLOW_NODE_TYPE,
+	.constructor =	ng_netflow_constructor,
+	.rcvmsg =	ng_netflow_rcvmsg,
+	.close =	ng_netflow_close,
+	.shutdown =	ng_netflow_rmnode,
+	.newhook =	ng_netflow_newhook,
+	.rcvdata =	ng_netflow_rcvdata,
+	.disconnect =	ng_netflow_disconnect,
+	.cmdlist =	ng_netflow_cmds,
+};
+NETGRAPH_INIT(netflow, &ng_netflow_typestruct);
+
+/* Called at node creation */
+static int
+ng_netflow_constructor (node_p node)
+{
+	priv_p priv;
+	int error = 0;
+
+	/* Initialize private data */
+	MALLOC(priv, priv_p, sizeof(*priv), M_NETGRAPH, M_NOWAIT);
+	if (priv == NULL)
+		return (ENOMEM);
+	bzero(priv, sizeof(*priv));
+
+	/* Make node and its data point at each other */
+	NG_NODE_SET_PRIVATE(node, priv);
+	priv->node = node;
+
+	/* Initialize timeouts to default values */
+	priv->info.nfinfo_inact_t = INACTIVE_TIMEOUT;
+	priv->info.nfinfo_act_t = ACTIVE_TIMEOUT;
+
+	/* Initialize callout handle */
+	callout_init(&priv->exp_callout, 1);
+
+	/* Allocate memory and set up flow cache */
+	if ((error = ng_netflow_cache_init(priv)))
+		return (error);
+
+	priv->dgram.header.version = htons(NETFLOW_V5);
+
+	return (0);
+}
+
+/*
+ * ng_netflow supports two hooks: data and export.
+ * Incoming traffic is expected on data, and expired
+ * netflow datagrams are sent to export.
+ */
+static int
+ng_netflow_newhook(node_p node, hook_p hook, const char *name)
+{
+	const priv_p priv = NG_NODE_PRIVATE(node);
+
+	if (strncmp(name, NG_NETFLOW_HOOK_DATA,	/* an iface hook? */
+	    strlen(NG_NETFLOW_HOOK_DATA)) == 0) {
+		iface_p iface;
+		int ifnum = -1;
+		const char *cp;
+		char *eptr;
+
+		cp = name + strlen(NG_NETFLOW_HOOK_DATA);
+		if (!isdigit(*cp) || (cp[0] == '0' && cp[1] != '\0'))
+			return (EINVAL);
+
+		ifnum = (int)strtoul(cp, &eptr, 10);
+		if (*eptr != '\0' || ifnum < 0 || ifnum >= NG_NETFLOW_MAXIFACES)
+			return (EINVAL);
+
+		/* See if hook is already connected */
+		if (priv->ifaces[ifnum].hook != NULL)
+			return (EISCONN);
+
+		iface = &priv->ifaces[ifnum];
+
+		/* Link private info and hook together */
+		NG_HOOK_SET_PRIVATE(hook, iface);
+		iface->hook = hook;
+
+		/*
+		 * In most cases traffic accounting is done on an
+		 * Ethernet interface, so default data link type
+		 * will be DLT_EN10MB.
+		 */
+		iface->info.ifinfo_dlt = DLT_EN10MB;
+
+	} else if (strcmp(name, NG_NETFLOW_HOOK_EXPORT) == 0) {
+
+		if (priv->export != NULL)
+			return (EISCONN);
+
+		priv->export = hook;
+
+		/* Exporter is ready. Let's schedule expiry. */
+		callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
+		    (void *)priv);
+	} else
+		return (EINVAL);
+
+	return (0);
+}
+
+/* Get a netgraph control message. */
+static int
+ng_netflow_rcvmsg (node_p node, item_p item, hook_p lasthook)
+{
+	const priv_p priv = NG_NODE_PRIVATE(node);
+	struct ng_mesg *resp = NULL;
+	int error = 0;
+	struct ng_mesg *msg;
+
+	NGI_GET_MSG(item, msg);
+
+	/* Deal with message according to cookie and command */
+	switch (msg->header.typecookie) {
+	case NGM_NETFLOW_COOKIE:
+		switch (msg->header.cmd) {
+		case NGM_NETFLOW_INFO:
+		{
+			struct ng_netflow_info *i;
+
+			NG_MKRESPONSE(resp, msg, sizeof(struct ng_netflow_info),
+			    M_NOWAIT);
+			i = (struct ng_netflow_info *)resp->data;
+			ng_netflow_copyinfo(priv, i);
+
+			break;
+		}
+		case NGM_NETFLOW_IFINFO:
+		{
+			struct ng_netflow_ifinfo *i;
+			const uint8_t *index;
+
+			if (msg->header.arglen != sizeof(uint8_t))
+				 ERROUT(EINVAL);
+
+			index  = (uint8_t *)msg->data;
+
+			/* connected iface? */
+			if (priv->ifaces[*index].hook == NULL)
+				 ERROUT(EINVAL);
+
+			NG_MKRESPONSE(resp, msg,
+			     sizeof(struct ng_netflow_ifinfo), M_NOWAIT);
+			i = (struct ng_netflow_ifinfo *)resp->data;
+			memcpy((void *)i, (void *)&priv->ifaces[*index].info,
+			    sizeof(priv->ifaces[*index].info));
+
+			break;
+		}
+		case NGM_NETFLOW_SETDLT:
+		{
+			struct ng_netflow_setdlt *set;
+			struct ng_netflow_iface *iface;
+
+			if (msg->header.arglen != sizeof(struct ng_netflow_setdlt))
+				ERROUT(EINVAL);
+
+			set = (struct ng_netflow_setdlt *)msg->data;
+			iface = &priv->ifaces[set->iface];
+
+			/* connected iface? */
+			if (iface->hook == NULL)
+				ERROUT(EINVAL);
+
+			switch (set->dlt) {
+			case	DLT_EN10MB:
+				iface->info.ifinfo_dlt = DLT_EN10MB;
+				break;
+			case	DLT_RAW:
+				iface->info.ifinfo_dlt = DLT_RAW;
+				break;
+			default:
+				ERROUT(EINVAL);
+			}
+			break;
+		}
+		case NGM_NETFLOW_SETIFINDEX:
+		{
+			struct ng_netflow_setifindex *set;
+			struct ng_netflow_iface *iface;
+
+			if (msg->header.arglen != sizeof(struct ng_netflow_setifindex))
+				ERROUT(EINVAL);
+
+			set = (struct ng_netflow_setifindex *)msg->data;
+			iface = &priv->ifaces[set->iface];
+
+			/* connected iface? */
+			if (iface->hook == NULL)
+				ERROUT(EINVAL);
+
+			iface->info.ifinfo_index = set->index;
+
+			break;
+		}
+		case NGM_NETFLOW_SETTIMEOUTS:
+		{
+			struct ng_netflow_settimeouts *set;
+
+			if (msg->header.arglen != sizeof(struct ng_netflow_settimeouts))
+				ERROUT(EINVAL);
+
+			set = (struct ng_netflow_settimeouts *)msg->data;
+
+			priv->info.nfinfo_inact_t = set->inactive_timeout;
+			priv->info.nfinfo_act_t = set->active_timeout;
+
+			break;
+		}
+		case NGM_NETFLOW_SHOW:
+		{
+			uint32_t *last;
+
+			if (msg->header.arglen != sizeof(uint32_t))
+				ERROUT(EINVAL);
+
+			last = (uint32_t *)msg->data;
+
+			NG_MKRESPONSE(resp, msg, NGRESP_SIZE, M_NOWAIT);
+
+			if (!resp)
+				ERROUT(ENOMEM);
+
+			error = ng_netflow_flow_show(priv, *last, resp);
+
+			break;
+		}
+		default:
+			ERROUT(EINVAL);		/* unknown command */
+			break;
+		}
+		break;
+	default:
+		ERROUT(EINVAL);		/* incorrect cookie */
+		break;
+	}
+
+	/*
+	 * Take care of synchronous response, if any.
+	 * Free memory and return.
+	 */
+done:
+	NG_RESPOND_MSG(error, node, item, resp);
+	NG_FREE_MSG(msg);
+
+	return (error);
+}
+
+/* Receive data on hook. */
+static int
+ng_netflow_rcvdata (hook_p hook, item_p item)
+{
+	const node_p node = NG_HOOK_NODE(hook);
+	const priv_p priv = NG_NODE_PRIVATE(node);
+	const iface_p iface = NG_HOOK_PRIVATE(hook);
+	struct mbuf *m;
+	int error = 0;
+
+	NGI_GET_M(item, m);
+	if (hook == priv->export) {
+		/*
+		 * Data arrived on export hook.
+		 * This must not happen.
+		 */
+		printf("ng_netflow: incoming data on export hook!\n");
+		ERROUT(EINVAL);
+	};
+
+	/* increase counters */
+	iface->info.ifinfo_packets++;
+
+	switch (iface->info.ifinfo_dlt) {
+	case DLT_EN10MB:	/* Ethernet */
+	{
+		struct ether_header *eh;
+		uint16_t etype;
+
+		if (CHECK_MLEN(m, (sizeof(struct ether_header))))
+			ERROUT(EINVAL);
+
+		if (CHECK_PULLUP(m, (sizeof(struct ether_header))))
+			ERROUT(ENOBUFS);
+
+		eh = mtod(m, struct ether_header *);
+
+		/* make sure this is IP frame */
+		etype = ntohs(eh->ether_type);
+		switch (etype) {
+		case ETHERTYPE_IP:
+			m_adj(m, sizeof(struct ether_header));
+			break;
+		default:
+			ERROUT(EINVAL);	/* ignore this frame */
+		}
+
+		break;
+	}
+	case DLT_RAW:
+		break;
+	default:
+		ERROUT(EINVAL);
+		break;
+	}
+
+	if (CHECK_MLEN(m, sizeof(struct ip)))
+		ERROUT(EINVAL);
+
+	if (CHECK_PULLUP(m, sizeof(struct ip)))
+		ERROUT(ENOBUFS);
+
+	error = ng_netflow_flow_add(priv, &m, iface);
+
+done:
+	if (m) {
+		if (item)
+			NG_FREE_ITEM(item);
+		NG_FREE_M(m);
+	}
+
+	return (error);	
+}
+
+/* We will be shut down in a moment */
+static int
+ng_netflow_close(node_p node)
+{
+	const priv_p priv = NG_NODE_PRIVATE(node);
+
+	callout_drain(&priv->exp_callout);
+	ng_netflow_cache_flush(priv);
+
+	return (0);
+}
+
+/* Do local shutdown processing. */
+static int
+ng_netflow_rmnode(node_p node)
+{
+	const priv_p priv = NG_NODE_PRIVATE(node);
+
+	NG_NODE_SET_PRIVATE(node, NULL);
+	NG_NODE_UNREF(priv->node);
+
+	FREE(priv, M_NETGRAPH);
+
+	return (0);
+}
+
+/* Hook disconnection. */
+static int
+ng_netflow_disconnect(hook_p hook)
+{
+	node_p node = NG_HOOK_NODE(hook);
+	priv_p priv = NG_NODE_PRIVATE(node);
+	iface_p iface = NG_HOOK_PRIVATE(hook);
+
+	if (iface != NULL)
+		iface->hook = NULL;
+
+	/* if export hook disconnected stop running expire(). */
+	if (hook == priv->export) {
+		callout_drain(&priv->exp_callout);
+		priv->export = NULL;
+	}
+
+	/* Removal of the last link destroys the node. */
+	if (NG_NODE_NUMHOOKS(node) == 0)
+		ng_rmnode_self(node);
+
+	return (0);
+}
diff --git a/sys/netgraph/netflow/ng_netflow.h b/sys/netgraph/netflow/ng_netflow.h
new file mode 100644
index 0000000..f668b5d
--- /dev/null
+++ b/sys/netgraph/netflow/ng_netflow.h
@@ -0,0 +1,282 @@
+/*-
+ * Copyright (c) 2004 Gleb Smirnoff <glebius@cell.sick.ru>
+ * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * 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. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ * This product includes software developed by Gleb Smirnoff and
+ * contributors.
+ * 4. Neither the name of the author nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * 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 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.
+ *
+ *	 $SourceForge: ng_netflow.h,v 1.26 2004/09/04 15:44:55 glebius Exp $
+ *	 $FreeBSD$
+ */
+
+#ifndef	_NG_NETFLOW_H_
+#define	_NG_NETFLOW_H_
+
+#define NG_NETFLOW_NODE_TYPE	"netflow"
+#define NGM_NETFLOW_COOKIE	1095355665
+
+#define	NG_NETFLOW_MAXIFACES	64
+
+/* Hook names */
+
+#define	NG_NETFLOW_HOOK_DATA	"iface"
+#define NG_NETFLOW_HOOK_EXPORT	"export"
+
+/* Netgraph commands understood by netflow node */
+enum {
+    NGM_NETFLOW_INFO = 1,	/* get node info */
+    NGM_NETFLOW_IFINFO,		/* get iface info */
+    NGM_NETFLOW_SHOW,		/* show ip cache flow */
+    NGM_NETFLOW_SETDLT,		/* set data-link type */	
+    NGM_NETFLOW_SETIFINDEX, 	/* set interface index */
+    NGM_NETFLOW_SETTIMEOUTS, 	/* set active/inactive flow timeouts */
+};
+
+/* This structure is returned by the NGM_NETFLOW_INFO message */
+struct ng_netflow_info {
+	uint64_t	nfinfo_bytes;	/* total number of accounted bytes */
+	uint32_t	nfinfo_packets;	/* total number of accounted packets */
+	uint32_t	nfinfo_used;	/* number of used cache records */
+	uint32_t	nfinfo_free;	/* number of free records */
+	uint32_t	nfinfo_inact_t;	/* flow inactive timeout */
+	uint32_t	nfinfo_act_t;	/* flow active timeout */
+};
+
+/* This structure is returned by the NGM_NETFLOW_IFINFO message */
+struct ng_netflow_ifinfo {
+	uint32_t	ifinfo_packets;	/* number of packets for this iface */
+	uint8_t		ifinfo_dlt;	/* Data Link Type, DLT_XXX */
+#define	MAXDLTNAMELEN	20
+	u_int16_t	ifinfo_index;	/* connected iface index */
+};
+
+
+/* This structure is passed to NGM_NETFLOW_SETDLT message */
+struct ng_netflow_setdlt {
+	uint16_t iface;		/* which iface dlt change */
+	uint8_t  dlt;			/* DLT_XXX from bpf.h */
+};
+
+/* This structure is passed to NGM_NETFLOW_SETIFINDEX */
+struct ng_netflow_setifindex {
+	u_int16_t iface;		/* which iface index change */
+	u_int16_t index;		/* new index */
+};
+
+/* This structure is passed to NGM_NETFLOW_SETTIMEOUTS */
+struct ng_netflow_settimeouts {
+	uint32_t	inactive_timeout;	/* flow inactive timeout */
+	uint32_t	active_timeout;		/* flow active timeout */
+};
+
+/* This is unique data, which identifies flow */
+struct flow_rec {
+	struct in_addr	r_src;
+	struct in_addr	r_dst;
+	union {
+		struct {
+			uint16_t	s_port;	/* source TCP/UDP port */
+			uint16_t	d_port; /* destination TCP/UDP port */
+		} dir;
+		uint32_t both;
+	} ports;
+	union {
+		struct {
+			u_char		prot;	/* IP protocol */
+			u_char		tos;	/* IP TOS */
+			uint16_t	i_ifx;	/* input interface index */
+		} i;
+		uint32_t all;
+	} misc;
+};
+
+#define	r_ip_p	misc.i.prot
+#define	r_tos	misc.i.tos
+#define	r_i_ifx	misc.i.i_ifx
+#define r_misc	misc.all
+#define r_ports	ports.both
+#define r_sport	ports.dir.s_port
+#define r_dport	ports.dir.d_port
+	
+/* A flow entry which accumulates statistics */
+struct flow_entry_data {
+	struct flow_rec		r;
+	struct in_addr		next_hop;
+	uint16_t		fle_o_ifx;	/* output interface index */
+#define				fle_i_ifx	r.misc.i.i_ifx
+	uint8_t		dst_mask;	/* destination route mask bits */
+	uint8_t		src_mask;	/* source route mask bits */
+	u_long			packets;
+	u_long			bytes;
+	long			first;	/* uptime on first packet */
+	long			last;	/* uptime on last packet */
+	u_char			tcp_flags;	/* cumulative OR */
+};
+
+/*
+ * How many flow records we will transfer at once
+ * without overflowing socket receive buffer
+ */
+#define NREC_AT_ONCE		1000
+#define NGRESP_SIZE		(sizeof(struct ngnf_flows) + (NREC_AT_ONCE * \
+				sizeof(struct flow_entry_data)))
+#define SORCVBUF_SIZE		(NGRESP_SIZE + 2 * sizeof(struct ng_mesg))
+
+/* This struct is returned to userland, when "show cache ip flow" */
+struct ngnf_flows {
+	uint32_t		nentries;
+	uint32_t		last;
+	struct flow_entry_data	entries[0];
+};
+
+/* Everything below is for kernel */
+
+#ifdef _KERNEL
+
+struct flow_entry {
+	struct flow_entry_data	f;
+
+	LIST_ENTRY(flow_entry)	fle_hash;	/* entries in one hash item */
+	TAILQ_ENTRY(flow_entry)	fle_work;	/* entries in work queue*/
+	SLIST_ENTRY(flow_entry)	fle_free;	/* entries in free stack */
+};
+
+/* Parsing declarations */
+
+/* Parse the info structure */
+#define	NG_NETFLOW_INFO_TYPE	{			\
+	{ "Bytes",	&ng_parse_uint64_type },	\
+	{ "Packets",	&ng_parse_uint32_type },	\
+	{ "Records used",	&ng_parse_uint32_type },\
+	{ "Records free",	&ng_parse_uint32_type },\
+	{ "Inactive timeout",	&ng_parse_uint32_type },\
+	{ "Active timeout",	&ng_parse_uint32_type },\
+	{ NULL }					\
+}
+
+/* Parse the ifinfo structure */
+#define NG_NETFLOW_IFINFO_TYPE	{			\
+	{ "packets",	&ng_parse_uint32_type },	\
+	{ "data link type", &ng_parse_uint8_type },	\
+	{ "index", &ng_parse_uint16_type },		\
+	{ NULL }					\
+}
+
+/* Parse the setdlt structure */
+#define	NG_NETFLOW_SETDLT_TYPE {			\
+	{ "iface",	&ng_parse_uint16_type },	\
+	{ "dlt",	&ng_parse_uint8_type  },	\
+	{ NULL }					\
+}
+
+/* Parse the setifindex structure */
+#define	NG_NETFLOW_SETIFINDEX_TYPE {			\
+	{ "iface",	&ng_parse_uint16_type },	\
+	{ "index",	&ng_parse_uint16_type },	\
+	{ NULL }					\
+}
+
+/* Parse the settimeouts structure */
+#define NG_NETFLOW_SETTIMEOUTS_TYPE {			\
+	{ "inactive",	&ng_parse_uint32_type },	\
+	{ "active",	&ng_parse_uint32_type },	\
+	{ NULL }					\
+}
+
+/* Private hook data */
+struct ng_netflow_iface {
+	hook_p		hook;		/* NULL when disconnected */
+	struct ng_netflow_ifinfo	info;
+};
+
+typedef struct ng_netflow_iface *iface_p;
+typedef struct ng_netflow_ifinfo *ifinfo_p;
+
+/* Structure describing our flow engine */
+struct netflow {
+	node_p			node;		/* link to the node itself */
+
+	struct ng_netflow_iface	ifaces[NG_NETFLOW_MAXIFACES];	/* incoming */
+	hook_p			export;		/* export data goes there */
+
+	struct ng_netflow_info	info;
+	uint32_t		flow_seq;	/* current flow sequence */
+
+	struct callout		exp_callout;
+
+	/* Flow cache is a big chunk of memory referenced by 'cache'.
+	 * Accounting engine searches for its record using hashing index
+	 * 'hash'. Expiry engine searches for its record from begining of
+	 * tail queue 'expire_q'. Allocation is performed using last free
+	 * stack held in singly linked list 'free_l' */
+#define	CACHESIZE			65536
+#define	CACHELOWAT			(CACHESIZE * 3/4)
+#define	CACHEHIGHWAT			(CACHESIZE * 9/10)
+	struct flow_entry		*cache;
+	struct flow_hash_entry		*hash;
+	TAILQ_HEAD( , flow_entry)	work_queue;
+	SLIST_HEAD( , flow_entry)	free_list;
+	SLIST_HEAD( , flow_entry)	expire_list;
+
+	/* Mutexes to protect above lists */
+	struct mtx			work_mtx;
+	struct mtx			free_mtx;
+	struct mtx			expire_mtx;
+
+	/* ng_netflow_export_send() forms its datagram here. */
+	struct netflow_export_dgram {
+		struct netflow_v5_header	header;
+		struct netflow_v5_record	r[NETFLOW_V5_MAX_RECORDS];
+	} __attribute__((__packed__)) dgram;
+};
+
+typedef struct netflow *priv_p;
+
+/* Header of a small list in hash cell */
+struct flow_hash_entry {
+	LIST_HEAD( ,flow_entry) head;
+};
+
+/* Make sure packet large enough to contain len bytes */
+#define	CHECK_MLEN(m, length)	((m)->m_pkthdr.len < (length))
+#define CHECK_PULLUP(m, length)	((m)->m_len < (length) && \
+				(((m) = m_pullup((m),(length))) == NULL))
+
+#define	ERROUT(x)	{ error = (x); goto done; }
+
+/* Prototypes for netflow.c */
+int	ng_netflow_cache_init(priv_p);
+void	ng_netflow_cache_flush(priv_p);
+void	ng_netflow_copyinfo(priv_p, struct ng_netflow_info *);
+timeout_t ng_netflow_expire;
+int	ng_netflow_flow_add(priv_p, struct mbuf **, iface_p);
+int	ng_netflow_flow_show(priv_p, uint32_t last, struct ng_mesg *);
+
+#endif	/* _KERNEL */
+#endif	/* _NG_NETFLOW_H_ */