xref: /dflybsd-src/sys/netgraph7/netflow/netflow.c (revision 805c8e8e4093ceca2e27510ad3a66d4de8060a55)
1b06ebda0SMatthew Dillon /*-
2b06ebda0SMatthew Dillon  * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
3b06ebda0SMatthew Dillon  * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
4b06ebda0SMatthew Dillon  * All rights reserved.
5b06ebda0SMatthew Dillon  *
6b06ebda0SMatthew Dillon  * Redistribution and use in source and binary forms, with or without
7b06ebda0SMatthew Dillon  * modification, are permitted provided that the following conditions
8b06ebda0SMatthew Dillon  * are met:
9b06ebda0SMatthew Dillon  * 1. Redistributions of source code must retain the above copyright
10b06ebda0SMatthew Dillon  *    notice, this list of conditions and the following disclaimer.
11b06ebda0SMatthew Dillon  * 2. Redistributions in binary form must reproduce the above copyright
12b06ebda0SMatthew Dillon  *    notice, this list of conditions and the following disclaimer in the
13b06ebda0SMatthew Dillon  *    documentation and/or other materials provided with the distribution.
14b06ebda0SMatthew Dillon  *
15b06ebda0SMatthew Dillon  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16b06ebda0SMatthew Dillon  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17b06ebda0SMatthew Dillon  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18b06ebda0SMatthew Dillon  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19b06ebda0SMatthew Dillon  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20b06ebda0SMatthew Dillon  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21b06ebda0SMatthew Dillon  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22b06ebda0SMatthew Dillon  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23b06ebda0SMatthew Dillon  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24b06ebda0SMatthew Dillon  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25b06ebda0SMatthew Dillon  * SUCH DAMAGE.
26b06ebda0SMatthew Dillon  *
27b06ebda0SMatthew Dillon  * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
285a975a3dSMatthew Dillon  * $FreeBSD: src/sys/netgraph/netflow/netflow.c,v 1.29 2008/05/09 23:02:57 julian Exp $
29b06ebda0SMatthew Dillon  */
30b06ebda0SMatthew Dillon 
31b06ebda0SMatthew Dillon #include <sys/param.h>
32b06ebda0SMatthew Dillon #include <sys/kernel.h>
33b06ebda0SMatthew Dillon #include <sys/limits.h>
34*805c8e8eSzrj #include <sys/malloc.h>
35b06ebda0SMatthew Dillon #include <sys/mbuf.h>
36b06ebda0SMatthew Dillon #include <sys/syslog.h>
37b06ebda0SMatthew Dillon #include <sys/systm.h>
38b06ebda0SMatthew Dillon #include <sys/socket.h>
39b06ebda0SMatthew Dillon 
40b06ebda0SMatthew Dillon #include <machine/atomic.h>
41b06ebda0SMatthew Dillon 
42b06ebda0SMatthew Dillon #include <net/if.h>
43b06ebda0SMatthew Dillon #include <net/route.h>
44b06ebda0SMatthew Dillon #include <netinet/in.h>
45b06ebda0SMatthew Dillon #include <netinet/in_systm.h>
46b06ebda0SMatthew Dillon #include <netinet/ip.h>
47b06ebda0SMatthew Dillon #include <netinet/tcp.h>
48b06ebda0SMatthew Dillon #include <netinet/udp.h>
49b06ebda0SMatthew Dillon 
505a975a3dSMatthew Dillon #include "ng_message.h"
515a975a3dSMatthew Dillon #include "netgraph.h"
52b06ebda0SMatthew Dillon 
535a975a3dSMatthew Dillon #include "netflow/netflow.h"
545a975a3dSMatthew Dillon #include "netflow/ng_netflow.h"
55b06ebda0SMatthew Dillon 
56b06ebda0SMatthew Dillon #define	NBUCKETS	(65536)		/* must be power of 2 */
57b06ebda0SMatthew Dillon 
58b06ebda0SMatthew Dillon /* This hash is for TCP or UDP packets. */
59b06ebda0SMatthew Dillon #define FULL_HASH(addr1, addr2, port1, port2)	\
60b06ebda0SMatthew Dillon 	(((addr1 ^ (addr1 >> 16) ^ 		\
61b06ebda0SMatthew Dillon 	htons(addr2 ^ (addr2 >> 16))) ^ 	\
62b06ebda0SMatthew Dillon 	port1 ^ htons(port2)) &			\
63b06ebda0SMatthew Dillon 	(NBUCKETS - 1))
64b06ebda0SMatthew Dillon 
65b06ebda0SMatthew Dillon /* This hash is for all other IP packets. */
66b06ebda0SMatthew Dillon #define ADDR_HASH(addr1, addr2)			\
67b06ebda0SMatthew Dillon 	((addr1 ^ (addr1 >> 16) ^ 		\
68b06ebda0SMatthew Dillon 	htons(addr2 ^ (addr2 >> 16))) &		\
69b06ebda0SMatthew Dillon 	(NBUCKETS - 1))
70b06ebda0SMatthew Dillon 
71b06ebda0SMatthew Dillon /* Macros to shorten logical constructions */
72b06ebda0SMatthew Dillon /* XXX: priv must exist in namespace */
73b06ebda0SMatthew Dillon #define	INACTIVE(fle)	(time_uptime - fle->f.last > priv->info.nfinfo_inact_t)
74b06ebda0SMatthew Dillon #define	AGED(fle)	(time_uptime - fle->f.first > priv->info.nfinfo_act_t)
75b06ebda0SMatthew Dillon #define	ISFREE(fle)	(fle->f.packets == 0)
76b06ebda0SMatthew Dillon 
77b06ebda0SMatthew Dillon /*
78b06ebda0SMatthew Dillon  * 4 is a magical number: statistically number of 4-packet flows is
79b06ebda0SMatthew Dillon  * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
80b06ebda0SMatthew Dillon  * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
81b06ebda0SMatthew Dillon  * of reachable host and 4-packet otherwise.
82b06ebda0SMatthew Dillon  */
83b06ebda0SMatthew Dillon #define	SMALL(fle)	(fle->f.packets <= 4)
84b06ebda0SMatthew Dillon 
85b06ebda0SMatthew Dillon /*
86b06ebda0SMatthew Dillon  * Cisco uses milliseconds for uptime. Bad idea, since it overflows
87b06ebda0SMatthew Dillon  * every 48+ days. But we will do same to keep compatibility. This macro
88b06ebda0SMatthew Dillon  * does overflowable multiplication to 1000.
89b06ebda0SMatthew Dillon  */
90b06ebda0SMatthew Dillon #define	MILLIUPTIME(t)	(((t) << 9) +	/* 512 */	\
91b06ebda0SMatthew Dillon 			 ((t) << 8) +	/* 256 */	\
92b06ebda0SMatthew Dillon 			 ((t) << 7) +	/* 128 */	\
93b06ebda0SMatthew Dillon 			 ((t) << 6) +	/* 64  */	\
94b06ebda0SMatthew Dillon 			 ((t) << 5) +	/* 32  */	\
95b06ebda0SMatthew Dillon 			 ((t) << 3))	/* 8   */
96b06ebda0SMatthew Dillon 
97b06ebda0SMatthew Dillon MALLOC_DECLARE(M_NETFLOW_HASH);
98b06ebda0SMatthew Dillon MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
99b06ebda0SMatthew Dillon 
100b06ebda0SMatthew Dillon static int export_add(item_p, struct flow_entry *);
101b06ebda0SMatthew Dillon static int export_send(priv_p, item_p, int flags);
102b06ebda0SMatthew Dillon 
103b06ebda0SMatthew Dillon /* Generate hash for a given flow record. */
104b06ebda0SMatthew Dillon static __inline uint32_t
ip_hash(struct flow_rec * r)105b06ebda0SMatthew Dillon ip_hash(struct flow_rec *r)
106b06ebda0SMatthew Dillon {
107b06ebda0SMatthew Dillon 	switch (r->r_ip_p) {
108b06ebda0SMatthew Dillon 	case IPPROTO_TCP:
109b06ebda0SMatthew Dillon 	case IPPROTO_UDP:
110b06ebda0SMatthew Dillon 		return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
111b06ebda0SMatthew Dillon 		    r->r_sport, r->r_dport);
112b06ebda0SMatthew Dillon 	default:
113b06ebda0SMatthew Dillon 		return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
114b06ebda0SMatthew Dillon 	}
115b06ebda0SMatthew Dillon }
116b06ebda0SMatthew Dillon 
117b06ebda0SMatthew Dillon /* This is callback from uma(9), called on alloc. */
118b06ebda0SMatthew Dillon static int
uma_ctor_flow(void * mem,int size,void * arg,int how)119b06ebda0SMatthew Dillon uma_ctor_flow(void *mem, int size, void *arg, int how)
120b06ebda0SMatthew Dillon {
121b06ebda0SMatthew Dillon 	priv_p priv = (priv_p )arg;
122b06ebda0SMatthew Dillon 
123b06ebda0SMatthew Dillon 	if (atomic_load_acq_32(&priv->info.nfinfo_used) >= CACHESIZE)
124b06ebda0SMatthew Dillon 		return (ENOMEM);
125b06ebda0SMatthew Dillon 
126b06ebda0SMatthew Dillon 	atomic_add_32(&priv->info.nfinfo_used, 1);
127b06ebda0SMatthew Dillon 
128b06ebda0SMatthew Dillon 	return (0);
129b06ebda0SMatthew Dillon }
130b06ebda0SMatthew Dillon 
131b06ebda0SMatthew Dillon /* This is callback from uma(9), called on free. */
132b06ebda0SMatthew Dillon static void
uma_dtor_flow(void * mem,int size,void * arg)133b06ebda0SMatthew Dillon uma_dtor_flow(void *mem, int size, void *arg)
134b06ebda0SMatthew Dillon {
135b06ebda0SMatthew Dillon 	priv_p priv = (priv_p )arg;
136b06ebda0SMatthew Dillon 
137b06ebda0SMatthew Dillon 	atomic_subtract_32(&priv->info.nfinfo_used, 1);
138b06ebda0SMatthew Dillon }
139b06ebda0SMatthew Dillon 
140b06ebda0SMatthew Dillon /*
141b06ebda0SMatthew Dillon  * Detach export datagram from priv, if there is any.
142b06ebda0SMatthew Dillon  * If there is no, allocate a new one.
143b06ebda0SMatthew Dillon  */
144b06ebda0SMatthew Dillon static item_p
get_export_dgram(priv_p priv)145b06ebda0SMatthew Dillon get_export_dgram(priv_p priv)
146b06ebda0SMatthew Dillon {
147b06ebda0SMatthew Dillon 	item_p	item = NULL;
148b06ebda0SMatthew Dillon 
149b06ebda0SMatthew Dillon 	mtx_lock(&priv->export_mtx);
150b06ebda0SMatthew Dillon 	if (priv->export_item != NULL) {
151b06ebda0SMatthew Dillon 		item = priv->export_item;
152b06ebda0SMatthew Dillon 		priv->export_item = NULL;
153b06ebda0SMatthew Dillon 	}
154b06ebda0SMatthew Dillon 	mtx_unlock(&priv->export_mtx);
155b06ebda0SMatthew Dillon 
156b06ebda0SMatthew Dillon 	if (item == NULL) {
157b06ebda0SMatthew Dillon 		struct netflow_v5_export_dgram *dgram;
158b06ebda0SMatthew Dillon 		struct mbuf *m;
159b06ebda0SMatthew Dillon 
160b5523eacSSascha Wildner 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
161b06ebda0SMatthew Dillon 		if (m == NULL)
162b06ebda0SMatthew Dillon 			return (NULL);
163b06ebda0SMatthew Dillon 		item = ng_package_data(m, NG_NOFLAGS);
164b06ebda0SMatthew Dillon 		if (item == NULL)
165b06ebda0SMatthew Dillon 			return (NULL);
166b06ebda0SMatthew Dillon 		dgram = mtod(m, struct netflow_v5_export_dgram *);
167b06ebda0SMatthew Dillon 		dgram->header.count = 0;
168b06ebda0SMatthew Dillon 		dgram->header.version = htons(NETFLOW_V5);
169b06ebda0SMatthew Dillon 
170b06ebda0SMatthew Dillon 	}
171b06ebda0SMatthew Dillon 
172b06ebda0SMatthew Dillon 	return (item);
173b06ebda0SMatthew Dillon }
174b06ebda0SMatthew Dillon 
175b06ebda0SMatthew Dillon /*
176b06ebda0SMatthew Dillon  * Re-attach incomplete datagram back to priv.
177b06ebda0SMatthew Dillon  * If there is already another one, then send incomplete. */
178b06ebda0SMatthew Dillon static void
return_export_dgram(priv_p priv,item_p item,int flags)179b06ebda0SMatthew Dillon return_export_dgram(priv_p priv, item_p item, int flags)
180b06ebda0SMatthew Dillon {
181b06ebda0SMatthew Dillon 	/*
182b06ebda0SMatthew Dillon 	 * It may happen on SMP, that some thread has already
183b06ebda0SMatthew Dillon 	 * put its item there, in this case we bail out and
184b06ebda0SMatthew Dillon 	 * send what we have to collector.
185b06ebda0SMatthew Dillon 	 */
186b06ebda0SMatthew Dillon 	mtx_lock(&priv->export_mtx);
187b06ebda0SMatthew Dillon 	if (priv->export_item == NULL) {
188b06ebda0SMatthew Dillon 		priv->export_item = item;
189b06ebda0SMatthew Dillon 		mtx_unlock(&priv->export_mtx);
190b06ebda0SMatthew Dillon 	} else {
191b06ebda0SMatthew Dillon 		mtx_unlock(&priv->export_mtx);
192b06ebda0SMatthew Dillon 		export_send(priv, item, flags);
193b06ebda0SMatthew Dillon 	}
194b06ebda0SMatthew Dillon }
195b06ebda0SMatthew Dillon 
196b06ebda0SMatthew Dillon /*
197b06ebda0SMatthew Dillon  * The flow is over. Call export_add() and free it. If datagram is
198b06ebda0SMatthew Dillon  * full, then call export_send().
199b06ebda0SMatthew Dillon  */
200b06ebda0SMatthew Dillon static __inline void
expire_flow(priv_p priv,item_p * item,struct flow_entry * fle,int flags)201b06ebda0SMatthew Dillon expire_flow(priv_p priv, item_p *item, struct flow_entry *fle, int flags)
202b06ebda0SMatthew Dillon {
203b06ebda0SMatthew Dillon 	if (*item == NULL)
204b06ebda0SMatthew Dillon 		*item = get_export_dgram(priv);
205b06ebda0SMatthew Dillon 	if (*item == NULL) {
206b06ebda0SMatthew Dillon 		atomic_add_32(&priv->info.nfinfo_export_failed, 1);
207b06ebda0SMatthew Dillon 		uma_zfree_arg(priv->zone, fle, priv);
208b06ebda0SMatthew Dillon 		return;
209b06ebda0SMatthew Dillon 	}
210b06ebda0SMatthew Dillon 	if (export_add(*item, fle) > 0) {
211b06ebda0SMatthew Dillon 		export_send(priv, *item, flags);
212b06ebda0SMatthew Dillon 		*item = NULL;
213b06ebda0SMatthew Dillon 	}
214b06ebda0SMatthew Dillon 	uma_zfree_arg(priv->zone, fle, priv);
215b06ebda0SMatthew Dillon }
216b06ebda0SMatthew Dillon 
217b06ebda0SMatthew Dillon /* Get a snapshot of node statistics */
218b06ebda0SMatthew Dillon void
ng_netflow_copyinfo(priv_p priv,struct ng_netflow_info * i)219b06ebda0SMatthew Dillon ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
220b06ebda0SMatthew Dillon {
221b06ebda0SMatthew Dillon 	/* XXX: atomic */
222b06ebda0SMatthew Dillon 	memcpy((void *)i, (void *)&priv->info, sizeof(priv->info));
223b06ebda0SMatthew Dillon }
224b06ebda0SMatthew Dillon 
225b06ebda0SMatthew Dillon /*
226b06ebda0SMatthew Dillon  * Insert a record into defined slot.
227b06ebda0SMatthew Dillon  *
228b06ebda0SMatthew Dillon  * First we get for us a free flow entry, then fill in all
229b06ebda0SMatthew Dillon  * possible fields in it.
230b06ebda0SMatthew Dillon  *
231b06ebda0SMatthew Dillon  * TODO: consider dropping hash mutex while filling in datagram,
232b06ebda0SMatthew Dillon  * as this was done in previous version. Need to test & profile
233b06ebda0SMatthew Dillon  * to be sure.
234b06ebda0SMatthew Dillon  */
235b06ebda0SMatthew Dillon static __inline int
hash_insert(priv_p priv,struct flow_hash_entry * hsh,struct flow_rec * r,int plen,uint8_t tcp_flags)236b06ebda0SMatthew Dillon hash_insert(priv_p priv, struct flow_hash_entry  *hsh, struct flow_rec *r,
237b06ebda0SMatthew Dillon 	int plen, uint8_t tcp_flags)
238b06ebda0SMatthew Dillon {
239b06ebda0SMatthew Dillon 	struct flow_entry *fle;
240b06ebda0SMatthew Dillon 	struct sockaddr_in sin;
241b06ebda0SMatthew Dillon 	struct rtentry *rt;
242b06ebda0SMatthew Dillon 
243a6c72860SNuno Antunes 	KKASSERT(mtx_owned(&hsh->mtx));
244b06ebda0SMatthew Dillon 
2455a975a3dSMatthew Dillon 	fle = uma_zalloc_arg(priv->zone, priv, M_WAITOK | M_NULLOK);
246b06ebda0SMatthew Dillon 	if (fle == NULL) {
247b06ebda0SMatthew Dillon 		atomic_add_32(&priv->info.nfinfo_alloc_failed, 1);
248b06ebda0SMatthew Dillon 		return (ENOMEM);
249b06ebda0SMatthew Dillon 	}
250b06ebda0SMatthew Dillon 
251b06ebda0SMatthew Dillon 	/*
252b06ebda0SMatthew Dillon 	 * Now fle is totally ours. It is detached from all lists,
253b06ebda0SMatthew Dillon 	 * we can safely edit it.
254b06ebda0SMatthew Dillon 	 */
255b06ebda0SMatthew Dillon 
256b06ebda0SMatthew Dillon 	bcopy(r, &fle->f.r, sizeof(struct flow_rec));
257b06ebda0SMatthew Dillon 	fle->f.bytes = plen;
258b06ebda0SMatthew Dillon 	fle->f.packets = 1;
259b06ebda0SMatthew Dillon 	fle->f.tcp_flags = tcp_flags;
260b06ebda0SMatthew Dillon 
261b06ebda0SMatthew Dillon 	fle->f.first = fle->f.last = time_uptime;
262b06ebda0SMatthew Dillon 
263b06ebda0SMatthew Dillon 	/*
264b06ebda0SMatthew Dillon 	 * First we do route table lookup on destination address. So we can
265b06ebda0SMatthew Dillon 	 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
266b06ebda0SMatthew Dillon 	 */
267b06ebda0SMatthew Dillon 	bzero(&sin, sizeof(sin));
268b06ebda0SMatthew Dillon 	sin.sin_len = sizeof(struct sockaddr_in);
269b06ebda0SMatthew Dillon 	sin.sin_family = AF_INET;
270b06ebda0SMatthew Dillon 	sin.sin_addr = fle->f.r.r_dst;
271b06ebda0SMatthew Dillon 	/* XXX MRT 0 as a default.. need the m here to get fib */
272b06ebda0SMatthew Dillon 	rt = rtalloc1_fib((struct sockaddr *)&sin, 0, RTF_CLONING, 0);
273b06ebda0SMatthew Dillon 	if (rt != NULL) {
274b06ebda0SMatthew Dillon 		fle->f.fle_o_ifx = rt->rt_ifp->if_index;
275b06ebda0SMatthew Dillon 
276b06ebda0SMatthew Dillon 		if (rt->rt_flags & RTF_GATEWAY &&
277b06ebda0SMatthew Dillon 		    rt->rt_gateway->sa_family == AF_INET)
278b06ebda0SMatthew Dillon 			fle->f.next_hop =
279b06ebda0SMatthew Dillon 			    ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
280b06ebda0SMatthew Dillon 
281b06ebda0SMatthew Dillon 		if (rt_mask(rt))
282b06ebda0SMatthew Dillon 			fle->f.dst_mask = bitcount32(((struct sockaddr_in *)
283b06ebda0SMatthew Dillon 			    rt_mask(rt))->sin_addr.s_addr);
284b06ebda0SMatthew Dillon 		else if (rt->rt_flags & RTF_HOST)
285b06ebda0SMatthew Dillon 			/* Give up. We can't determine mask :( */
286b06ebda0SMatthew Dillon 			fle->f.dst_mask = 32;
287b06ebda0SMatthew Dillon 
288b06ebda0SMatthew Dillon 		RTFREE_LOCKED(rt);
289b06ebda0SMatthew Dillon 	}
290b06ebda0SMatthew Dillon 
291b06ebda0SMatthew Dillon 	/* Do route lookup on source address, to fill in src_mask. */
292b06ebda0SMatthew Dillon 	bzero(&sin, sizeof(sin));
293b06ebda0SMatthew Dillon 	sin.sin_len = sizeof(struct sockaddr_in);
294b06ebda0SMatthew Dillon 	sin.sin_family = AF_INET;
295b06ebda0SMatthew Dillon 	sin.sin_addr = fle->f.r.r_src;
296b06ebda0SMatthew Dillon 	/* XXX MRT 0 as a default  revisit.  need the mbuf for fib*/
297b06ebda0SMatthew Dillon 	rt = rtalloc1_fib((struct sockaddr *)&sin, 0, RTF_CLONING, 0);
298b06ebda0SMatthew Dillon 	if (rt != NULL) {
299b06ebda0SMatthew Dillon 		if (rt_mask(rt))
300b06ebda0SMatthew Dillon 			fle->f.src_mask = bitcount32(((struct sockaddr_in *)
301b06ebda0SMatthew Dillon 			    rt_mask(rt))->sin_addr.s_addr);
302b06ebda0SMatthew Dillon 		else if (rt->rt_flags & RTF_HOST)
303b06ebda0SMatthew Dillon 			/* Give up. We can't determine mask :( */
304b06ebda0SMatthew Dillon 			fle->f.src_mask = 32;
305b06ebda0SMatthew Dillon 
306b06ebda0SMatthew Dillon 		RTFREE_LOCKED(rt);
307b06ebda0SMatthew Dillon 	}
308b06ebda0SMatthew Dillon 
309b06ebda0SMatthew Dillon 	/* Push new flow at the and of hash. */
310b06ebda0SMatthew Dillon 	TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
311b06ebda0SMatthew Dillon 
312b06ebda0SMatthew Dillon 	return (0);
313b06ebda0SMatthew Dillon }
314b06ebda0SMatthew Dillon 
315b06ebda0SMatthew Dillon 
316b06ebda0SMatthew Dillon /*
317b06ebda0SMatthew Dillon  * Non-static functions called from ng_netflow.c
318b06ebda0SMatthew Dillon  */
319b06ebda0SMatthew Dillon 
320b06ebda0SMatthew Dillon /* Allocate memory and set up flow cache */
321b06ebda0SMatthew Dillon int
ng_netflow_cache_init(priv_p priv)322b06ebda0SMatthew Dillon ng_netflow_cache_init(priv_p priv)
323b06ebda0SMatthew Dillon {
324b06ebda0SMatthew Dillon 	struct flow_hash_entry	*hsh;
325b06ebda0SMatthew Dillon 	int i;
326b06ebda0SMatthew Dillon 
327b06ebda0SMatthew Dillon 	/* Initialize cache UMA zone. */
328b06ebda0SMatthew Dillon 	priv->zone = uma_zcreate("NetFlow cache", sizeof(struct flow_entry),
329b06ebda0SMatthew Dillon 	    uma_ctor_flow, uma_dtor_flow, NULL, NULL, UMA_ALIGN_CACHE, 0);
330b06ebda0SMatthew Dillon 	uma_zone_set_max(priv->zone, CACHESIZE);
331b06ebda0SMatthew Dillon 
332b06ebda0SMatthew Dillon 	/* Allocate hash. */
333fc025606SSascha Wildner 	priv->hash = kmalloc(NBUCKETS * sizeof(struct flow_hash_entry),
334b06ebda0SMatthew Dillon 			     M_NETFLOW_HASH, M_WAITOK | M_ZERO);
335b06ebda0SMatthew Dillon 
336b06ebda0SMatthew Dillon 	/* Initialize hash. */
337b06ebda0SMatthew Dillon 	for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
338b06ebda0SMatthew Dillon 		mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
339b06ebda0SMatthew Dillon 		TAILQ_INIT(&hsh->head);
340b06ebda0SMatthew Dillon 	}
341b06ebda0SMatthew Dillon 
342b06ebda0SMatthew Dillon 	mtx_init(&priv->export_mtx, "export dgram lock", NULL, MTX_DEF);
343b06ebda0SMatthew Dillon 
344b06ebda0SMatthew Dillon 	return (0);
345b06ebda0SMatthew Dillon }
346b06ebda0SMatthew Dillon 
347b06ebda0SMatthew Dillon /* Free all flow cache memory. Called from node close method. */
348b06ebda0SMatthew Dillon void
ng_netflow_cache_flush(priv_p priv)349b06ebda0SMatthew Dillon ng_netflow_cache_flush(priv_p priv)
350b06ebda0SMatthew Dillon {
351b06ebda0SMatthew Dillon 	struct flow_entry	*fle, *fle1;
352b06ebda0SMatthew Dillon 	struct flow_hash_entry	*hsh;
353b06ebda0SMatthew Dillon 	item_p			item = NULL;
354b06ebda0SMatthew Dillon 	int i;
355b06ebda0SMatthew Dillon 
356b06ebda0SMatthew Dillon 	/*
357b06ebda0SMatthew Dillon 	 * We are going to free probably billable data.
358b06ebda0SMatthew Dillon 	 * Expire everything before freeing it.
359b06ebda0SMatthew Dillon 	 * No locking is required since callout is already drained.
360b06ebda0SMatthew Dillon 	 */
361b06ebda0SMatthew Dillon 	for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
362b06ebda0SMatthew Dillon 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
363b06ebda0SMatthew Dillon 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
364b06ebda0SMatthew Dillon 			expire_flow(priv, &item, fle, NG_QUEUE);
365b06ebda0SMatthew Dillon 		}
366b06ebda0SMatthew Dillon 
367b06ebda0SMatthew Dillon 	if (item != NULL)
368b06ebda0SMatthew Dillon 		export_send(priv, item, NG_QUEUE);
369b06ebda0SMatthew Dillon 
370b06ebda0SMatthew Dillon 	uma_zdestroy(priv->zone);
371b06ebda0SMatthew Dillon 
372b06ebda0SMatthew Dillon 	/* Destroy hash mutexes. */
373b06ebda0SMatthew Dillon 	for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
374b06ebda0SMatthew Dillon 		mtx_destroy(&hsh->mtx);
375b06ebda0SMatthew Dillon 
376b06ebda0SMatthew Dillon 	/* Free hash memory. */
377b06ebda0SMatthew Dillon 	if (priv->hash)
378fc025606SSascha Wildner 		kfree(priv->hash, M_NETFLOW_HASH);
379b06ebda0SMatthew Dillon 
380b06ebda0SMatthew Dillon 	mtx_destroy(&priv->export_mtx);
381b06ebda0SMatthew Dillon }
382b06ebda0SMatthew Dillon 
383b06ebda0SMatthew Dillon /* Insert packet from into flow cache. */
384b06ebda0SMatthew Dillon int
ng_netflow_flow_add(priv_p priv,struct ip * ip,iface_p iface,struct ifnet * ifp)385b06ebda0SMatthew Dillon ng_netflow_flow_add(priv_p priv, struct ip *ip, iface_p iface,
386b06ebda0SMatthew Dillon 	struct ifnet *ifp)
387b06ebda0SMatthew Dillon {
388b06ebda0SMatthew Dillon 	register struct flow_entry	*fle, *fle1;
389b06ebda0SMatthew Dillon 	struct flow_hash_entry		*hsh;
390b06ebda0SMatthew Dillon 	struct flow_rec		r;
391b06ebda0SMatthew Dillon 	item_p			item = NULL;
392b06ebda0SMatthew Dillon 	int			hlen, plen;
393b06ebda0SMatthew Dillon 	int			error = 0;
394b06ebda0SMatthew Dillon 	uint8_t			tcp_flags = 0;
395b06ebda0SMatthew Dillon 
396b06ebda0SMatthew Dillon 	/* Try to fill flow_rec r */
397b06ebda0SMatthew Dillon 	bzero(&r, sizeof(r));
398b06ebda0SMatthew Dillon 	/* check version */
399b06ebda0SMatthew Dillon 	if (ip->ip_v != IPVERSION)
400b06ebda0SMatthew Dillon 		return (EINVAL);
401b06ebda0SMatthew Dillon 
402b06ebda0SMatthew Dillon 	/* verify min header length */
403b06ebda0SMatthew Dillon 	hlen = ip->ip_hl << 2;
404b06ebda0SMatthew Dillon 
405b06ebda0SMatthew Dillon 	if (hlen < sizeof(struct ip))
406b06ebda0SMatthew Dillon 		return (EINVAL);
407b06ebda0SMatthew Dillon 
408b06ebda0SMatthew Dillon 	r.r_src = ip->ip_src;
409b06ebda0SMatthew Dillon 	r.r_dst = ip->ip_dst;
410b06ebda0SMatthew Dillon 
411b06ebda0SMatthew Dillon 	/* save packet length */
412b06ebda0SMatthew Dillon 	plen = ntohs(ip->ip_len);
413b06ebda0SMatthew Dillon 
414b06ebda0SMatthew Dillon 	r.r_ip_p = ip->ip_p;
415b06ebda0SMatthew Dillon 	r.r_tos = ip->ip_tos;
416b06ebda0SMatthew Dillon 
417b06ebda0SMatthew Dillon 	/* Configured in_ifx overrides mbuf's */
418b06ebda0SMatthew Dillon 	if (iface->info.ifinfo_index == 0) {
419b06ebda0SMatthew Dillon 		if (ifp != NULL)
420b06ebda0SMatthew Dillon 			r.r_i_ifx = ifp->if_index;
421b06ebda0SMatthew Dillon 	} else
422b06ebda0SMatthew Dillon 		r.r_i_ifx = iface->info.ifinfo_index;
423b06ebda0SMatthew Dillon 
424b06ebda0SMatthew Dillon 	/*
425b06ebda0SMatthew Dillon 	 * XXX NOTE: only first fragment of fragmented TCP, UDP and
426b06ebda0SMatthew Dillon 	 * ICMP packet will be recorded with proper s_port and d_port.
427b06ebda0SMatthew Dillon 	 * Following fragments will be recorded simply as IP packet with
428b06ebda0SMatthew Dillon 	 * ip_proto = ip->ip_p and s_port, d_port set to zero.
429b06ebda0SMatthew Dillon 	 * I know, it looks like bug. But I don't want to re-implement
430b06ebda0SMatthew Dillon 	 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
431b06ebda0SMatthew Dillon 	 * and nobody complains yet :)
432b06ebda0SMatthew Dillon 	 */
433b06ebda0SMatthew Dillon 	if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
434b06ebda0SMatthew Dillon 		switch(r.r_ip_p) {
435b06ebda0SMatthew Dillon 		case IPPROTO_TCP:
436b06ebda0SMatthew Dillon 		{
437b06ebda0SMatthew Dillon 			register struct tcphdr *tcp;
438b06ebda0SMatthew Dillon 
439b06ebda0SMatthew Dillon 			tcp = (struct tcphdr *)((caddr_t )ip + hlen);
440b06ebda0SMatthew Dillon 			r.r_sport = tcp->th_sport;
441b06ebda0SMatthew Dillon 			r.r_dport = tcp->th_dport;
442b06ebda0SMatthew Dillon 			tcp_flags = tcp->th_flags;
443b06ebda0SMatthew Dillon 			break;
444b06ebda0SMatthew Dillon 		}
445b06ebda0SMatthew Dillon 			case IPPROTO_UDP:
446b06ebda0SMatthew Dillon 			r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
447b06ebda0SMatthew Dillon 			break;
448b06ebda0SMatthew Dillon 		}
449b06ebda0SMatthew Dillon 
450b06ebda0SMatthew Dillon 	/* Update node statistics. XXX: race... */
451b06ebda0SMatthew Dillon 	priv->info.nfinfo_packets ++;
452b06ebda0SMatthew Dillon 	priv->info.nfinfo_bytes += plen;
453b06ebda0SMatthew Dillon 
454b06ebda0SMatthew Dillon 	/* Find hash slot. */
455b06ebda0SMatthew Dillon 	hsh = &priv->hash[ip_hash(&r)];
456b06ebda0SMatthew Dillon 
457b06ebda0SMatthew Dillon 	mtx_lock(&hsh->mtx);
458b06ebda0SMatthew Dillon 
459b06ebda0SMatthew Dillon 	/*
460b06ebda0SMatthew Dillon 	 * Go through hash and find our entry. If we encounter an
461b06ebda0SMatthew Dillon 	 * entry, that should be expired, purge it. We do a reverse
462b06ebda0SMatthew Dillon 	 * search since most active entries are first, and most
463b06ebda0SMatthew Dillon 	 * searches are done on most active entries.
464b06ebda0SMatthew Dillon 	 */
465b06ebda0SMatthew Dillon 	TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
466b06ebda0SMatthew Dillon 		if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
467b06ebda0SMatthew Dillon 			break;
468b06ebda0SMatthew Dillon 		if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
469b06ebda0SMatthew Dillon 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
470b06ebda0SMatthew Dillon 			expire_flow(priv, &item, fle, NG_QUEUE);
471b06ebda0SMatthew Dillon 			atomic_add_32(&priv->info.nfinfo_act_exp, 1);
472b06ebda0SMatthew Dillon 		}
473b06ebda0SMatthew Dillon 	}
474b06ebda0SMatthew Dillon 
475b06ebda0SMatthew Dillon 	if (fle) {			/* An existent entry. */
476b06ebda0SMatthew Dillon 
477b06ebda0SMatthew Dillon 		fle->f.bytes += plen;
478b06ebda0SMatthew Dillon 		fle->f.packets ++;
479b06ebda0SMatthew Dillon 		fle->f.tcp_flags |= tcp_flags;
480b06ebda0SMatthew Dillon 		fle->f.last = time_uptime;
481b06ebda0SMatthew Dillon 
482b06ebda0SMatthew Dillon 		/*
483b06ebda0SMatthew Dillon 		 * We have the following reasons to expire flow in active way:
484b06ebda0SMatthew Dillon 		 * - it hit active timeout
485b06ebda0SMatthew Dillon 		 * - a TCP connection closed
486b06ebda0SMatthew Dillon 		 * - it is going to overflow counter
487b06ebda0SMatthew Dillon 		 */
488b06ebda0SMatthew Dillon 		if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
489b06ebda0SMatthew Dillon 		    (fle->f.bytes >= (UINT_MAX - IF_MAXMTU)) ) {
490b06ebda0SMatthew Dillon 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
491b06ebda0SMatthew Dillon 			expire_flow(priv, &item, fle, NG_QUEUE);
492b06ebda0SMatthew Dillon 			atomic_add_32(&priv->info.nfinfo_act_exp, 1);
493b06ebda0SMatthew Dillon 		} else {
494b06ebda0SMatthew Dillon 			/*
495b06ebda0SMatthew Dillon 			 * It is the newest, move it to the tail,
496b06ebda0SMatthew Dillon 			 * if it isn't there already. Next search will
497b06ebda0SMatthew Dillon 			 * locate it quicker.
498b06ebda0SMatthew Dillon 			 */
499b06ebda0SMatthew Dillon 			if (fle != TAILQ_LAST(&hsh->head, fhead)) {
500b06ebda0SMatthew Dillon 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
501b06ebda0SMatthew Dillon 				TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
502b06ebda0SMatthew Dillon 			}
503b06ebda0SMatthew Dillon 		}
504b06ebda0SMatthew Dillon 	} else				/* A new flow entry. */
505b06ebda0SMatthew Dillon 		error = hash_insert(priv, hsh, &r, plen, tcp_flags);
506b06ebda0SMatthew Dillon 
507b06ebda0SMatthew Dillon 	mtx_unlock(&hsh->mtx);
508b06ebda0SMatthew Dillon 
509b06ebda0SMatthew Dillon 	if (item != NULL)
510b06ebda0SMatthew Dillon 		return_export_dgram(priv, item, NG_QUEUE);
511b06ebda0SMatthew Dillon 
512b06ebda0SMatthew Dillon 	return (error);
513b06ebda0SMatthew Dillon }
514b06ebda0SMatthew Dillon 
515b06ebda0SMatthew Dillon /*
516b06ebda0SMatthew Dillon  * Return records from cache to userland.
517b06ebda0SMatthew Dillon  *
518b06ebda0SMatthew Dillon  * TODO: matching particular IP should be done in kernel, here.
519b06ebda0SMatthew Dillon  */
520b06ebda0SMatthew Dillon int
ng_netflow_flow_show(priv_p priv,uint32_t last,struct ng_mesg * resp)521b06ebda0SMatthew Dillon ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp)
522b06ebda0SMatthew Dillon {
523b06ebda0SMatthew Dillon 	struct flow_hash_entry *hsh;
524b06ebda0SMatthew Dillon 	struct flow_entry *fle;
525b06ebda0SMatthew Dillon 	struct ngnf_flows *data;
526b06ebda0SMatthew Dillon 	int i;
527b06ebda0SMatthew Dillon 
528b06ebda0SMatthew Dillon 	data = (struct ngnf_flows *)resp->data;
529b06ebda0SMatthew Dillon 	data->last = 0;
530b06ebda0SMatthew Dillon 	data->nentries = 0;
531b06ebda0SMatthew Dillon 
532b06ebda0SMatthew Dillon 	/* Check if this is a first run */
533b06ebda0SMatthew Dillon 	if (last == 0) {
534b06ebda0SMatthew Dillon 		hsh = priv->hash;
535b06ebda0SMatthew Dillon 		i = 0;
536b06ebda0SMatthew Dillon 	} else {
537b06ebda0SMatthew Dillon 		if (last > NBUCKETS-1)
538b06ebda0SMatthew Dillon 			return (EINVAL);
539b06ebda0SMatthew Dillon 		hsh = priv->hash + last;
540b06ebda0SMatthew Dillon 		i = last;
541b06ebda0SMatthew Dillon 	}
542b06ebda0SMatthew Dillon 
543b06ebda0SMatthew Dillon 	/*
544b06ebda0SMatthew Dillon 	 * We will transfer not more than NREC_AT_ONCE. More data
545b06ebda0SMatthew Dillon 	 * will come in next message.
546b06ebda0SMatthew Dillon 	 * We send current hash index to userland, and userland should
547b06ebda0SMatthew Dillon 	 * return it back to us. Then, we will restart with new entry.
548b06ebda0SMatthew Dillon 	 *
549b06ebda0SMatthew Dillon 	 * The resulting cache snapshot is inaccurate for the
550b06ebda0SMatthew Dillon 	 * following reasons:
551b06ebda0SMatthew Dillon 	 *  - we skip locked hash entries
552b06ebda0SMatthew Dillon 	 *  - we bail out, if someone wants our entry
553b06ebda0SMatthew Dillon 	 *  - we skip rest of entry, when hit NREC_AT_ONCE
554b06ebda0SMatthew Dillon 	 */
555b06ebda0SMatthew Dillon 	for (; i < NBUCKETS; hsh++, i++) {
556b06ebda0SMatthew Dillon 		if (mtx_trylock(&hsh->mtx) == 0)
557b06ebda0SMatthew Dillon 			continue;
558b06ebda0SMatthew Dillon 
559b06ebda0SMatthew Dillon 		TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
560023c80aeSSascha Wildner 			if (mtx_contested(&hsh->mtx))
561b06ebda0SMatthew Dillon 				break;
562b06ebda0SMatthew Dillon 
563b06ebda0SMatthew Dillon 			bcopy(&fle->f, &(data->entries[data->nentries]),
564b06ebda0SMatthew Dillon 			    sizeof(fle->f));
565b06ebda0SMatthew Dillon 			data->nentries++;
566b06ebda0SMatthew Dillon 			if (data->nentries == NREC_AT_ONCE) {
567b06ebda0SMatthew Dillon 				mtx_unlock(&hsh->mtx);
568b06ebda0SMatthew Dillon 				if (++i < NBUCKETS)
569b06ebda0SMatthew Dillon 					data->last = i;
570b06ebda0SMatthew Dillon 				return (0);
571b06ebda0SMatthew Dillon 			}
572b06ebda0SMatthew Dillon 		}
573b06ebda0SMatthew Dillon 		mtx_unlock(&hsh->mtx);
574b06ebda0SMatthew Dillon 	}
575b06ebda0SMatthew Dillon 
576b06ebda0SMatthew Dillon 	return (0);
577b06ebda0SMatthew Dillon }
578b06ebda0SMatthew Dillon 
579b06ebda0SMatthew Dillon /* We have full datagram in privdata. Send it to export hook. */
580b06ebda0SMatthew Dillon static int
export_send(priv_p priv,item_p item,int flags)581b06ebda0SMatthew Dillon export_send(priv_p priv, item_p item, int flags)
582b06ebda0SMatthew Dillon {
583b06ebda0SMatthew Dillon 	struct mbuf *m = NGI_M(item);
584b06ebda0SMatthew Dillon 	struct netflow_v5_export_dgram *dgram = mtod(m,
585b06ebda0SMatthew Dillon 					struct netflow_v5_export_dgram *);
586b06ebda0SMatthew Dillon 	struct netflow_v5_header *header = &dgram->header;
587b06ebda0SMatthew Dillon 	struct timespec ts;
588b06ebda0SMatthew Dillon 	int error = 0;
589b06ebda0SMatthew Dillon 
590b06ebda0SMatthew Dillon 	/* Fill mbuf header. */
591b06ebda0SMatthew Dillon 	m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
592b06ebda0SMatthew Dillon 	   header->count + sizeof(struct netflow_v5_header);
593b06ebda0SMatthew Dillon 
594b06ebda0SMatthew Dillon 	/* Fill export header. */
595b06ebda0SMatthew Dillon 	header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
596b06ebda0SMatthew Dillon 	getnanotime(&ts);
597b06ebda0SMatthew Dillon 	header->unix_secs  = htonl(ts.tv_sec);
598b06ebda0SMatthew Dillon 	header->unix_nsecs = htonl(ts.tv_nsec);
599b06ebda0SMatthew Dillon 	header->engine_type = 0;
600b06ebda0SMatthew Dillon 	header->engine_id = 0;
601b06ebda0SMatthew Dillon 	header->pad = 0;
602b06ebda0SMatthew Dillon 	header->flow_seq = htonl(atomic_fetchadd_32(&priv->flow_seq,
603b06ebda0SMatthew Dillon 	    header->count));
604b06ebda0SMatthew Dillon 	header->count = htons(header->count);
605b06ebda0SMatthew Dillon 
606b06ebda0SMatthew Dillon 	if (priv->export != NULL)
607b06ebda0SMatthew Dillon 		NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
608b06ebda0SMatthew Dillon 	else
609b06ebda0SMatthew Dillon 		NG_FREE_ITEM(item);
610b06ebda0SMatthew Dillon 
611b06ebda0SMatthew Dillon 	return (error);
612b06ebda0SMatthew Dillon }
613b06ebda0SMatthew Dillon 
614b06ebda0SMatthew Dillon 
615b06ebda0SMatthew Dillon /* Add export record to dgram. */
616b06ebda0SMatthew Dillon static int
export_add(item_p item,struct flow_entry * fle)617b06ebda0SMatthew Dillon export_add(item_p item, struct flow_entry *fle)
618b06ebda0SMatthew Dillon {
619b06ebda0SMatthew Dillon 	struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
620b06ebda0SMatthew Dillon 					struct netflow_v5_export_dgram *);
621b06ebda0SMatthew Dillon 	struct netflow_v5_header *header = &dgram->header;
622b06ebda0SMatthew Dillon 	struct netflow_v5_record *rec;
623b06ebda0SMatthew Dillon 
624b06ebda0SMatthew Dillon 	rec = &dgram->r[header->count];
625b06ebda0SMatthew Dillon 	header->count ++;
626b06ebda0SMatthew Dillon 
627b06ebda0SMatthew Dillon 	KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
628b06ebda0SMatthew Dillon 	    ("ng_netflow: export too big"));
629b06ebda0SMatthew Dillon 
630b06ebda0SMatthew Dillon 	/* Fill in export record. */
631b06ebda0SMatthew Dillon 	rec->src_addr = fle->f.r.r_src.s_addr;
632b06ebda0SMatthew Dillon 	rec->dst_addr = fle->f.r.r_dst.s_addr;
633b06ebda0SMatthew Dillon 	rec->next_hop = fle->f.next_hop.s_addr;
634b06ebda0SMatthew Dillon 	rec->i_ifx    = htons(fle->f.fle_i_ifx);
635b06ebda0SMatthew Dillon 	rec->o_ifx    = htons(fle->f.fle_o_ifx);
636b06ebda0SMatthew Dillon 	rec->packets  = htonl(fle->f.packets);
637b06ebda0SMatthew Dillon 	rec->octets   = htonl(fle->f.bytes);
638b06ebda0SMatthew Dillon 	rec->first    = htonl(MILLIUPTIME(fle->f.first));
639b06ebda0SMatthew Dillon 	rec->last     = htonl(MILLIUPTIME(fle->f.last));
640b06ebda0SMatthew Dillon 	rec->s_port   = fle->f.r.r_sport;
641b06ebda0SMatthew Dillon 	rec->d_port   = fle->f.r.r_dport;
642b06ebda0SMatthew Dillon 	rec->flags    = fle->f.tcp_flags;
643b06ebda0SMatthew Dillon 	rec->prot     = fle->f.r.r_ip_p;
644b06ebda0SMatthew Dillon 	rec->tos      = fle->f.r.r_tos;
645b06ebda0SMatthew Dillon 	rec->dst_mask = fle->f.dst_mask;
646b06ebda0SMatthew Dillon 	rec->src_mask = fle->f.src_mask;
647b06ebda0SMatthew Dillon 
648b06ebda0SMatthew Dillon 	/* Not supported fields. */
649b06ebda0SMatthew Dillon 	rec->src_as = rec->dst_as = 0;
650b06ebda0SMatthew Dillon 
651b06ebda0SMatthew Dillon 	if (header->count == NETFLOW_V5_MAX_RECORDS)
652b06ebda0SMatthew Dillon 		return (1); /* end of datagram */
653b06ebda0SMatthew Dillon 	else
654b06ebda0SMatthew Dillon 		return (0);
655b06ebda0SMatthew Dillon }
656b06ebda0SMatthew Dillon 
657b06ebda0SMatthew Dillon /* Periodic flow expiry run. */
658b06ebda0SMatthew Dillon void
ng_netflow_expire(void * arg)659b06ebda0SMatthew Dillon ng_netflow_expire(void *arg)
660b06ebda0SMatthew Dillon {
661b06ebda0SMatthew Dillon 	struct flow_entry	*fle, *fle1;
662b06ebda0SMatthew Dillon 	struct flow_hash_entry	*hsh;
663b06ebda0SMatthew Dillon 	priv_p			priv = (priv_p )arg;
664b06ebda0SMatthew Dillon 	item_p			item = NULL;
665b06ebda0SMatthew Dillon 	uint32_t		used;
666b06ebda0SMatthew Dillon 	int			i;
667b06ebda0SMatthew Dillon 
668b06ebda0SMatthew Dillon 	/*
669b06ebda0SMatthew Dillon 	 * Going through all the cache.
670b06ebda0SMatthew Dillon 	 */
671b06ebda0SMatthew Dillon 	for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
672b06ebda0SMatthew Dillon 		/*
673b06ebda0SMatthew Dillon 		 * Skip entries, that are already being worked on.
674b06ebda0SMatthew Dillon 		 */
675b06ebda0SMatthew Dillon 		if (mtx_trylock(&hsh->mtx) == 0)
676b06ebda0SMatthew Dillon 			continue;
677b06ebda0SMatthew Dillon 
678b06ebda0SMatthew Dillon 		used = atomic_load_acq_32(&priv->info.nfinfo_used);
679b06ebda0SMatthew Dillon 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
680b06ebda0SMatthew Dillon 			/*
681b06ebda0SMatthew Dillon 			 * Interrupt thread wants this entry!
682b06ebda0SMatthew Dillon 			 * Quick! Quick! Bail out!
683b06ebda0SMatthew Dillon 			 */
6845a975a3dSMatthew Dillon 			if (mtx_contested(&hsh->mtx))
685b06ebda0SMatthew Dillon 				break;
686b06ebda0SMatthew Dillon 
687b06ebda0SMatthew Dillon 			/*
688b06ebda0SMatthew Dillon 			 * Don't expire aggressively while hash collision
689b06ebda0SMatthew Dillon 			 * ratio is predicted small.
690b06ebda0SMatthew Dillon 			 */
691b06ebda0SMatthew Dillon 			if (used <= (NBUCKETS*2) && !INACTIVE(fle))
692b06ebda0SMatthew Dillon 				break;
693b06ebda0SMatthew Dillon 
694b06ebda0SMatthew Dillon 			if ((INACTIVE(fle) && (SMALL(fle) ||
695b06ebda0SMatthew Dillon 			    (used > (NBUCKETS*2)))) || AGED(fle)) {
696b06ebda0SMatthew Dillon 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
697b06ebda0SMatthew Dillon 				expire_flow(priv, &item, fle, NG_NOFLAGS);
698b06ebda0SMatthew Dillon 				used--;
699b06ebda0SMatthew Dillon 				atomic_add_32(&priv->info.nfinfo_inact_exp, 1);
700b06ebda0SMatthew Dillon 			}
701b06ebda0SMatthew Dillon 		}
702b06ebda0SMatthew Dillon 		mtx_unlock(&hsh->mtx);
703b06ebda0SMatthew Dillon 	}
704b06ebda0SMatthew Dillon 
705b06ebda0SMatthew Dillon 	if (item != NULL)
706b06ebda0SMatthew Dillon 		return_export_dgram(priv, item, NG_NOFLAGS);
707b06ebda0SMatthew Dillon 
708b06ebda0SMatthew Dillon 	/* Schedule next expire. */
709b06ebda0SMatthew Dillon 	callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
710b06ebda0SMatthew Dillon 	    (void *)priv);
711b06ebda0SMatthew Dillon }
712