xref: /freebsd/sys/netgraph/netflow/netflow.c (revision a3cefe7f2b4df0f70ff92d4570ce18e517af43ec)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2010-2011 Alexander V. Chernikov <melifaro@ipfw.ru>
5  * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
6  * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
31  */
32 
33 #include <sys/cdefs.h>
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include <sys/param.h>
37 #include <sys/bitstring.h>
38 #include <sys/systm.h>
39 #include <sys/counter.h>
40 #include <sys/kernel.h>
41 #include <sys/ktr.h>
42 #include <sys/limits.h>
43 #include <sys/mbuf.h>
44 #include <sys/syslog.h>
45 #include <sys/socket.h>
46 #include <vm/uma.h>
47 
48 #include <net/if.h>
49 #include <net/if_dl.h>
50 #include <net/if_var.h>
51 #include <net/if_private.h>
52 #include <net/route.h>
53 #include <net/route/nhop.h>
54 #include <net/route/route_ctl.h>
55 #include <net/ethernet.h>
56 #include <netinet/in.h>
57 #include <netinet/in_fib.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip6.h>
61 #include <netinet/tcp.h>
62 #include <netinet/udp.h>
63 
64 #include <netinet6/in6_fib.h>
65 
66 #include <netgraph/ng_message.h>
67 #include <netgraph/netgraph.h>
68 
69 #include <netgraph/netflow/netflow.h>
70 #include <netgraph/netflow/netflow_v9.h>
71 #include <netgraph/netflow/ng_netflow.h>
72 
73 #define	NBUCKETS	(65536)		/* must be power of 2 */
74 
75 /* This hash is for TCP or UDP packets. */
76 #define FULL_HASH(addr1, addr2, port1, port2)	\
77 	(((addr1 ^ (addr1 >> 16) ^ 		\
78 	htons(addr2 ^ (addr2 >> 16))) ^ 	\
79 	port1 ^ htons(port2)) &			\
80 	(NBUCKETS - 1))
81 
82 /* This hash is for all other IP packets. */
83 #define ADDR_HASH(addr1, addr2)			\
84 	((addr1 ^ (addr1 >> 16) ^ 		\
85 	htons(addr2 ^ (addr2 >> 16))) &		\
86 	(NBUCKETS - 1))
87 
88 /* Macros to shorten logical constructions */
89 /* XXX: priv must exist in namespace */
90 #define	INACTIVE(fle)	(time_uptime - fle->f.last > priv->nfinfo_inact_t)
91 #define	AGED(fle)	(time_uptime - fle->f.first > priv->nfinfo_act_t)
92 #define	ISFREE(fle)	(fle->f.packets == 0)
93 
94 /*
95  * 4 is a magical number: statistically number of 4-packet flows is
96  * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
97  * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
98  * of reachable host and 4-packet otherwise.
99  */
100 #define	SMALL(fle)	(fle->f.packets <= 4)
101 
102 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
103 
104 static int export_add(item_p, struct flow_entry *);
105 static int export_send(priv_p, fib_export_p, item_p, int);
106 
107 #ifdef INET
108 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *,
109     int, uint8_t, uint16_t);
110 #endif
111 #ifdef INET6
112 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *,
113     int, uint8_t, uint16_t);
114 #endif
115 
116 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int);
117 
118 #ifdef INET
119 /*
120  * Generate hash for a given flow record.
121  *
122  * FIB is not used here, because:
123  * most VRFS will carry public IPv4 addresses which are unique even
124  * without FIB private addresses can overlap, but this is worked out
125  * via flow_rec bcmp() containing fib id. In IPv6 world addresses are
126  * all globally unique (it's not fully true, there is FC00::/7 for example,
127  * but chances of address overlap are MUCH smaller)
128  */
129 static inline uint32_t
130 ip_hash(struct flow_rec *r)
131 {
132 
133 	switch (r->r_ip_p) {
134 	case IPPROTO_TCP:
135 	case IPPROTO_UDP:
136 		return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
137 		    r->r_sport, r->r_dport);
138 	default:
139 		return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
140 	}
141 }
142 #endif
143 
144 #ifdef INET6
145 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */
146 static inline uint32_t
147 ip6_hash(struct flow6_rec *r)
148 {
149 
150 	switch (r->r_ip_p) {
151 	case IPPROTO_TCP:
152 	case IPPROTO_UDP:
153 		return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
154 		    r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport,
155 		    r->r_dport);
156 	default:
157 		return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
158 		    r->dst.r_dst6.__u6_addr.__u6_addr32[3]);
159  	}
160 }
161 
162 #endif
163 
164 /*
165  * Detach export datagram from priv, if there is any.
166  * If there is no, allocate a new one.
167  */
168 static item_p
169 get_export_dgram(priv_p priv, fib_export_p fe)
170 {
171 	item_p	item = NULL;
172 
173 	mtx_lock(&fe->export_mtx);
174 	if (fe->exp.item != NULL) {
175 		item = fe->exp.item;
176 		fe->exp.item = NULL;
177 	}
178 	mtx_unlock(&fe->export_mtx);
179 
180 	if (item == NULL) {
181 		struct netflow_v5_export_dgram *dgram;
182 		struct mbuf *m;
183 
184 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
185 		if (m == NULL)
186 			return (NULL);
187 		item = ng_package_data(m, NG_NOFLAGS);
188 		if (item == NULL)
189 			return (NULL);
190 		dgram = mtod(m, struct netflow_v5_export_dgram *);
191 		dgram->header.count = 0;
192 		dgram->header.version = htons(NETFLOW_V5);
193 		dgram->header.pad = 0;
194 	}
195 
196 	return (item);
197 }
198 
199 /*
200  * Re-attach incomplete datagram back to priv.
201  * If there is already another one, then send incomplete. */
202 static void
203 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags)
204 {
205 
206 	/*
207 	 * It may happen on SMP, that some thread has already
208 	 * put its item there, in this case we bail out and
209 	 * send what we have to collector.
210 	 */
211 	mtx_lock(&fe->export_mtx);
212 	if (fe->exp.item == NULL) {
213 		fe->exp.item = item;
214 		mtx_unlock(&fe->export_mtx);
215 	} else {
216 		mtx_unlock(&fe->export_mtx);
217 		export_send(priv, fe, item, flags);
218 	}
219 }
220 
221 /*
222  * The flow is over. Call export_add() and free it. If datagram is
223  * full, then call export_send().
224  */
225 static void
226 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags)
227 {
228 	struct netflow_export_item exp;
229 	uint16_t version = fle->f.version;
230 
231 	if ((priv->export != NULL) && (version == IPVERSION)) {
232 		exp.item = get_export_dgram(priv, fe);
233 		if (exp.item == NULL) {
234 			priv->nfinfo_export_failed++;
235 			if (priv->export9 != NULL)
236 				priv->nfinfo_export9_failed++;
237 			/* fle definitely contains IPv4 flow. */
238 			uma_zfree_arg(priv->zone, fle, priv);
239 			return;
240 		}
241 
242 		if (export_add(exp.item, fle) > 0)
243 			export_send(priv, fe, exp.item, flags);
244 		else
245 			return_export_dgram(priv, fe, exp.item, NG_QUEUE);
246 	}
247 
248 	if (priv->export9 != NULL) {
249 		exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt);
250 		if (exp.item9 == NULL) {
251 			priv->nfinfo_export9_failed++;
252 			if (version == IPVERSION)
253 				uma_zfree_arg(priv->zone, fle, priv);
254 #ifdef INET6
255 			else if (version == IP6VERSION)
256 				uma_zfree_arg(priv->zone6, fle, priv);
257 #endif
258 			else
259 				panic("ng_netflow: Unknown IP proto: %d",
260 				    version);
261 			return;
262 		}
263 
264 		if (export9_add(exp.item9, exp.item9_opt, fle) > 0)
265 			export9_send(priv, fe, exp.item9, exp.item9_opt, flags);
266 		else
267 			return_export9_dgram(priv, fe, exp.item9,
268 			    exp.item9_opt, NG_QUEUE);
269 	}
270 
271 	if (version == IPVERSION)
272 		uma_zfree_arg(priv->zone, fle, priv);
273 #ifdef INET6
274 	else if (version == IP6VERSION)
275 		uma_zfree_arg(priv->zone6, fle, priv);
276 #endif
277 }
278 
279 /* Get a snapshot of node statistics */
280 void
281 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
282 {
283 
284 	i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes);
285 	i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets);
286 	i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6);
287 	i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6);
288 	i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes);
289 	i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets);
290 	i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6);
291 	i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6);
292 	i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp);
293 	i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp);
294 
295 	i->nfinfo_used = uma_zone_get_cur(priv->zone);
296 #ifdef INET6
297 	i->nfinfo_used6 = uma_zone_get_cur(priv->zone6);
298 #endif
299 
300 	i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed;
301 	i->nfinfo_export_failed = priv->nfinfo_export_failed;
302 	i->nfinfo_export9_failed = priv->nfinfo_export9_failed;
303 	i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf;
304 	i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs;
305 	i->nfinfo_inact_t = priv->nfinfo_inact_t;
306 	i->nfinfo_act_t = priv->nfinfo_act_t;
307 }
308 
309 /*
310  * Insert a record into defined slot.
311  *
312  * First we get for us a free flow entry, then fill in all
313  * possible fields in it.
314  *
315  * TODO: consider dropping hash mutex while filling in datagram,
316  * as this was done in previous version. Need to test & profile
317  * to be sure.
318  */
319 #ifdef INET
320 static int
321 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r,
322 	int plen, uint8_t flags, uint16_t tcp_flags)
323 {
324 	struct flow_entry *fle;
325 
326 	mtx_assert(&hsh->mtx, MA_OWNED);
327 
328 	fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT);
329 	if (fle == NULL) {
330 		priv->nfinfo_alloc_failed++;
331 		return (ENOMEM);
332 	}
333 
334 	/*
335 	 * Now fle is totally ours. It is detached from all lists,
336 	 * we can safely edit it.
337 	 */
338 	fle->f.version = IPVERSION;
339 	bcopy(r, &fle->f.r, sizeof(struct flow_rec));
340 	fle->f.bytes = plen;
341 	fle->f.packets = 1;
342 	fle->f.tcp_flags = tcp_flags;
343 
344 	fle->f.first = fle->f.last = time_uptime;
345 
346 	/*
347 	 * First we do route table lookup on destination address. So we can
348 	 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
349 	 */
350 	if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
351 		struct rtentry *rt;
352 		struct route_nhop_data rnd;
353 
354 		rt = fib4_lookup_rt(r->fib, fle->f.r.r_dst, 0, NHR_NONE, &rnd);
355 		if (rt != NULL) {
356 			struct in_addr addr;
357 			uint32_t scopeid;
358 			struct nhop_object *nh = nhop_select_func(rnd.rnd_nhop, 0);
359 			int plen;
360 
361 			rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid);
362 			fle->f.fle_o_ifx = nh->nh_ifp->if_index;
363 			if (nh->gw_sa.sa_family == AF_INET)
364 				fle->f.next_hop = nh->gw4_sa.sin_addr;
365 			/*
366 			 * XXX we're leaving an empty gateway here for
367 			 * IPv6 nexthops.
368 			 */
369 			fle->f.dst_mask = plen;
370 		}
371 	}
372 
373 	/* Do route lookup on source address, to fill in src_mask. */
374 	if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
375 		struct rtentry *rt;
376 		struct route_nhop_data rnd;
377 
378 		rt = fib4_lookup_rt(r->fib, fle->f.r.r_src, 0, NHR_NONE, &rnd);
379 		if (rt != NULL) {
380 			struct in_addr addr;
381 			uint32_t scopeid;
382 			int plen;
383 
384 			rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid);
385 			fle->f.src_mask = plen;
386 		}
387 	}
388 
389 	/* Push new flow at the and of hash. */
390 	TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
391 
392 	return (0);
393 }
394 #endif
395 
396 #ifdef INET6
397 static int
398 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r,
399 	int plen, uint8_t flags, uint16_t tcp_flags)
400 {
401 	struct flow6_entry *fle6;
402 
403 	mtx_assert(&hsh6->mtx, MA_OWNED);
404 
405 	fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT);
406 	if (fle6 == NULL) {
407 		priv->nfinfo_alloc_failed++;
408 		return (ENOMEM);
409 	}
410 
411 	/*
412 	 * Now fle is totally ours. It is detached from all lists,
413 	 * we can safely edit it.
414 	 */
415 
416 	fle6->f.version = IP6VERSION;
417 	bcopy(r, &fle6->f.r, sizeof(struct flow6_rec));
418 	fle6->f.bytes = plen;
419 	fle6->f.packets = 1;
420 	fle6->f.tcp_flags = tcp_flags;
421 
422 	fle6->f.first = fle6->f.last = time_uptime;
423 
424 	/*
425 	 * First we do route table lookup on destination address. So we can
426 	 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
427 	 */
428 	if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
429 		struct rtentry *rt;
430 		struct route_nhop_data rnd;
431 
432 		rt = fib6_lookup_rt(r->fib, &fle6->f.r.dst.r_dst6, 0, NHR_NONE, &rnd);
433 		if (rt != NULL) {
434 			struct in6_addr addr;
435 			uint32_t scopeid;
436 			struct nhop_object *nh = nhop_select_func(rnd.rnd_nhop, 0);
437 			int plen;
438 
439 			rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
440 			fle6->f.fle_o_ifx = nh->nh_ifp->if_index;
441 			if (nh->gw_sa.sa_family == AF_INET6)
442 				fle6->f.n.next_hop6 = nh->gw6_sa.sin6_addr;
443 			fle6->f.dst_mask = plen;
444 		}
445 	}
446 
447 	if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
448 		/* Do route lookup on source address, to fill in src_mask. */
449 		struct rtentry *rt;
450 		struct route_nhop_data rnd;
451 
452 		rt = fib6_lookup_rt(r->fib, &fle6->f.r.src.r_src6, 0, NHR_NONE, &rnd);
453 		if (rt != NULL) {
454 			struct in6_addr addr;
455 			uint32_t scopeid;
456 			int plen;
457 
458 			rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
459 			fle6->f.src_mask = plen;
460 		}
461 	}
462 
463 	/* Push new flow at the and of hash. */
464 	TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash);
465 
466 	return (0);
467 }
468 #endif
469 
470 /*
471  * Non-static functions called from ng_netflow.c
472  */
473 
474 /* Allocate memory and set up flow cache */
475 void
476 ng_netflow_cache_init(priv_p priv)
477 {
478 	struct flow_hash_entry *hsh;
479 	int i;
480 
481 	/* Initialize cache UMA zone. */
482 	priv->zone = uma_zcreate("NetFlow IPv4 cache",
483 	    sizeof(struct flow_entry), NULL, NULL, NULL, NULL,
484 	    UMA_ALIGN_CACHE, 0);
485 	uma_zone_set_max(priv->zone, CACHESIZE);
486 #ifdef INET6
487 	priv->zone6 = uma_zcreate("NetFlow IPv6 cache",
488 	    sizeof(struct flow6_entry), NULL, NULL, NULL, NULL,
489 	    UMA_ALIGN_CACHE, 0);
490 	uma_zone_set_max(priv->zone6, CACHESIZE);
491 #endif
492 
493 	/* Allocate hash. */
494 	priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
495 	    M_NETFLOW_HASH, M_WAITOK | M_ZERO);
496 
497 	/* Initialize hash. */
498 	for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
499 		mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
500 		TAILQ_INIT(&hsh->head);
501 	}
502 
503 #ifdef INET6
504 	/* Allocate hash. */
505 	priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
506 	    M_NETFLOW_HASH, M_WAITOK | M_ZERO);
507 
508 	/* Initialize hash. */
509 	for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) {
510 		mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
511 		TAILQ_INIT(&hsh->head);
512 	}
513 #endif
514 
515 	priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK);
516 	priv->nfinfo_packets = counter_u64_alloc(M_WAITOK);
517 	priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK);
518 	priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK);
519 	priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK);
520 	priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK);
521 	priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK);
522 	priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK);
523 	priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK);
524 	priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK);
525 
526 	ng_netflow_v9_cache_init(priv);
527 	CTR0(KTR_NET, "ng_netflow startup()");
528 }
529 
530 /* Initialize new FIB table for v5 and v9 */
531 int
532 ng_netflow_fib_init(priv_p priv, int fib)
533 {
534 	fib_export_p	fe = priv_to_fib(priv, fib);
535 
536 	CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib);
537 
538 	if (fe != NULL)
539 		return (0);
540 
541 	if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH,
542 	    M_NOWAIT | M_ZERO)) == NULL)
543 		return (ENOMEM);
544 
545 	mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF);
546 	mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF);
547 	fe->fib = fib;
548 	fe->domain_id = fib;
549 
550 	if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib],
551 	    (uintptr_t)NULL, (uintptr_t)fe) == 0) {
552 		/* FIB already set up by other ISR */
553 		CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p",
554 		    fib, fe, priv_to_fib(priv, fib));
555 		mtx_destroy(&fe->export_mtx);
556 		mtx_destroy(&fe->export9_mtx);
557 		free(fe, M_NETGRAPH);
558 	} else {
559 		/* Increase counter for statistics */
560 		CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)",
561 		    fib, fe, priv_to_fib(priv, fib));
562 		priv->nfinfo_alloc_fibs++;
563 	}
564 
565 	return (0);
566 }
567 
568 /* Free all flow cache memory. Called from node close method. */
569 void
570 ng_netflow_cache_flush(priv_p priv)
571 {
572 	struct flow_entry	*fle, *fle1;
573 	struct flow_hash_entry	*hsh;
574 	struct netflow_export_item exp;
575 	fib_export_p fe;
576 	int i;
577 
578 	bzero(&exp, sizeof(exp));
579 
580 	/*
581 	 * We are going to free probably billable data.
582 	 * Expire everything before freeing it.
583 	 * No locking is required since callout is already drained.
584 	 */
585 	for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
586 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
587 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
588 			fe = priv_to_fib(priv, fle->f.r.fib);
589 			expire_flow(priv, fe, fle, NG_QUEUE);
590 		}
591 #ifdef INET6
592 	for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++)
593 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
594 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
595 			fe = priv_to_fib(priv, fle->f.r.fib);
596 			expire_flow(priv, fe, fle, NG_QUEUE);
597 		}
598 #endif
599 
600 	uma_zdestroy(priv->zone);
601 	/* Destroy hash mutexes. */
602 	for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
603 		mtx_destroy(&hsh->mtx);
604 
605 	/* Free hash memory. */
606 	if (priv->hash != NULL)
607 		free(priv->hash, M_NETFLOW_HASH);
608 #ifdef INET6
609 	uma_zdestroy(priv->zone6);
610 	/* Destroy hash mutexes. */
611 	for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++)
612 		mtx_destroy(&hsh->mtx);
613 
614 	/* Free hash memory. */
615 	if (priv->hash6 != NULL)
616 		free(priv->hash6, M_NETFLOW_HASH);
617 #endif
618 
619 	for (i = 0; i < priv->maxfibs; i++) {
620 		if ((fe = priv_to_fib(priv, i)) == NULL)
621 			continue;
622 
623 		if (fe->exp.item != NULL)
624 			export_send(priv, fe, fe->exp.item, NG_QUEUE);
625 
626 		if (fe->exp.item9 != NULL)
627 			export9_send(priv, fe, fe->exp.item9,
628 			    fe->exp.item9_opt, NG_QUEUE);
629 
630 		mtx_destroy(&fe->export_mtx);
631 		mtx_destroy(&fe->export9_mtx);
632 		free(fe, M_NETGRAPH);
633 	}
634 
635 	counter_u64_free(priv->nfinfo_bytes);
636 	counter_u64_free(priv->nfinfo_packets);
637 	counter_u64_free(priv->nfinfo_bytes6);
638 	counter_u64_free(priv->nfinfo_packets6);
639 	counter_u64_free(priv->nfinfo_sbytes);
640 	counter_u64_free(priv->nfinfo_spackets);
641 	counter_u64_free(priv->nfinfo_sbytes6);
642 	counter_u64_free(priv->nfinfo_spackets6);
643 	counter_u64_free(priv->nfinfo_act_exp);
644 	counter_u64_free(priv->nfinfo_inact_exp);
645 
646 	ng_netflow_v9_cache_flush(priv);
647 }
648 
649 #ifdef INET
650 /* Insert packet from into flow cache. */
651 int
652 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip,
653     caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
654     unsigned int src_if_index)
655 {
656 	struct flow_entry	*fle, *fle1;
657 	struct flow_hash_entry	*hsh;
658 	struct flow_rec		r;
659 	int			hlen, plen;
660 	int			error = 0;
661 	uint16_t		tcp_flags = 0;
662 
663 	bzero(&r, sizeof(r));
664 
665 	if (ip->ip_v != IPVERSION)
666 		return (EINVAL);
667 
668 	hlen = ip->ip_hl << 2;
669 	if (hlen < sizeof(struct ip))
670 		return (EINVAL);
671 
672 	/* Assume L4 template by default */
673 	r.flow_type = NETFLOW_V9_FLOW_V4_L4;
674 
675 	r.r_src = ip->ip_src;
676 	r.r_dst = ip->ip_dst;
677 	r.fib = fe->fib;
678 
679 	plen = ntohs(ip->ip_len);
680 
681 	r.r_ip_p = ip->ip_p;
682 	r.r_tos = ip->ip_tos;
683 
684 	r.r_i_ifx = src_if_index;
685 
686 	/*
687 	 * XXX NOTE: only first fragment of fragmented TCP, UDP and
688 	 * ICMP packet will be recorded with proper s_port and d_port.
689 	 * Following fragments will be recorded simply as IP packet with
690 	 * ip_proto = ip->ip_p and s_port, d_port set to zero.
691 	 * I know, it looks like bug. But I don't want to re-implement
692 	 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
693 	 * and nobody complains yet :)
694 	 */
695 	if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
696 		switch(r.r_ip_p) {
697 		case IPPROTO_TCP:
698 		    {
699 			struct tcphdr *tcp;
700 
701 			tcp = (struct tcphdr *)((caddr_t )ip + hlen);
702 			r.r_sport = tcp->th_sport;
703 			r.r_dport = tcp->th_dport;
704 			tcp_flags = tcp_get_flags(tcp);
705 			break;
706 		    }
707 		case IPPROTO_UDP:
708 			r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
709 			break;
710 		}
711 
712 	counter_u64_add(priv->nfinfo_packets, 1);
713 	counter_u64_add(priv->nfinfo_bytes, plen);
714 
715 	/* Find hash slot. */
716 	hsh = &priv->hash[ip_hash(&r)];
717 
718 	mtx_lock(&hsh->mtx);
719 
720 	/*
721 	 * Go through hash and find our entry. If we encounter an
722 	 * entry, that should be expired, purge it. We do a reverse
723 	 * search since most active entries are first, and most
724 	 * searches are done on most active entries.
725 	 */
726 	TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
727 		if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
728 			break;
729 		if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
730 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
731 			expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
732 			    fle, NG_QUEUE);
733 			counter_u64_add(priv->nfinfo_act_exp, 1);
734 		}
735 	}
736 
737 	if (fle) {			/* An existent entry. */
738 
739 		fle->f.bytes += plen;
740 		fle->f.packets ++;
741 		fle->f.tcp_flags |= tcp_flags;
742 		fle->f.last = time_uptime;
743 
744 		/*
745 		 * We have the following reasons to expire flow in active way:
746 		 * - it hit active timeout
747 		 * - a TCP connection closed
748 		 * - it is going to overflow counter
749 		 */
750 		if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
751 		    (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
752 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
753 			expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
754 			    fle, NG_QUEUE);
755 			counter_u64_add(priv->nfinfo_act_exp, 1);
756 		} else {
757 			/*
758 			 * It is the newest, move it to the tail,
759 			 * if it isn't there already. Next search will
760 			 * locate it quicker.
761 			 */
762 			if (fle != TAILQ_LAST(&hsh->head, fhead)) {
763 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
764 				TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
765 			}
766 		}
767 	} else				/* A new flow entry. */
768 		error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags);
769 
770 	mtx_unlock(&hsh->mtx);
771 
772 	return (error);
773 }
774 #endif
775 
776 #ifdef INET6
777 /* Insert IPv6 packet from into flow cache. */
778 int
779 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6,
780     caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
781     unsigned int src_if_index)
782 {
783 	struct flow_entry	*fle = NULL, *fle1;
784 	struct flow6_entry	*fle6;
785 	struct flow_hash_entry	*hsh;
786 	struct flow6_rec	r;
787 	int			plen;
788 	int			error = 0;
789 	uint16_t		tcp_flags = 0;
790 
791 	/* check version */
792 	if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
793 		return (EINVAL);
794 
795 	bzero(&r, sizeof(r));
796 
797 	r.src.r_src6 = ip6->ip6_src;
798 	r.dst.r_dst6 = ip6->ip6_dst;
799 	r.fib = fe->fib;
800 
801 	/* Assume L4 template by default */
802 	r.flow_type = NETFLOW_V9_FLOW_V6_L4;
803 
804 	plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
805 
806 #if 0
807 	/* XXX: set DSCP/CoS value */
808 	r.r_tos = ip->ip_tos;
809 #endif
810 	if ((flags & NG_NETFLOW_IS_FRAG) == 0) {
811 		switch(upper_proto) {
812 		case IPPROTO_TCP:
813 		    {
814 			struct tcphdr *tcp;
815 
816 			tcp = (struct tcphdr *)upper_ptr;
817 			r.r_ports = *(uint32_t *)upper_ptr;
818 			tcp_flags = tcp_get_flags(tcp);
819 			break;
820 		    }
821  		case IPPROTO_UDP:
822 		case IPPROTO_SCTP:
823 			r.r_ports = *(uint32_t *)upper_ptr;
824 			break;
825 		}
826 	}
827 
828 	r.r_ip_p = upper_proto;
829 	r.r_i_ifx = src_if_index;
830 
831 	counter_u64_add(priv->nfinfo_packets6, 1);
832 	counter_u64_add(priv->nfinfo_bytes6, plen);
833 
834 	/* Find hash slot. */
835 	hsh = &priv->hash6[ip6_hash(&r)];
836 
837 	mtx_lock(&hsh->mtx);
838 
839 	/*
840 	 * Go through hash and find our entry. If we encounter an
841 	 * entry, that should be expired, purge it. We do a reverse
842 	 * search since most active entries are first, and most
843 	 * searches are done on most active entries.
844 	 */
845 	TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
846 		if (fle->f.version != IP6VERSION)
847 			continue;
848 		fle6 = (struct flow6_entry *)fle;
849 		if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0)
850 			break;
851 		if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) {
852 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
853 			expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
854 			    NG_QUEUE);
855 			counter_u64_add(priv->nfinfo_act_exp, 1);
856 		}
857 	}
858 
859 	if (fle != NULL) {			/* An existent entry. */
860 		fle6 = (struct flow6_entry *)fle;
861 
862 		fle6->f.bytes += plen;
863 		fle6->f.packets ++;
864 		fle6->f.tcp_flags |= tcp_flags;
865 		fle6->f.last = time_uptime;
866 
867 		/*
868 		 * We have the following reasons to expire flow in active way:
869 		 * - it hit active timeout
870 		 * - a TCP connection closed
871 		 * - it is going to overflow counter
872 		 */
873 		if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) ||
874 		    (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
875 			TAILQ_REMOVE(&hsh->head, fle, fle_hash);
876 			expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
877 			    NG_QUEUE);
878 			counter_u64_add(priv->nfinfo_act_exp, 1);
879 		} else {
880 			/*
881 			 * It is the newest, move it to the tail,
882 			 * if it isn't there already. Next search will
883 			 * locate it quicker.
884 			 */
885 			if (fle != TAILQ_LAST(&hsh->head, fhead)) {
886 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
887 				TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
888 			}
889 		}
890 	} else				/* A new flow entry. */
891 		error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags);
892 
893 	mtx_unlock(&hsh->mtx);
894 
895 	return (error);
896 }
897 #endif
898 
899 /*
900  * Return records from cache to userland.
901  *
902  * TODO: matching particular IP should be done in kernel, here.
903  */
904 int
905 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req,
906 struct ngnf_show_header *resp)
907 {
908 	struct flow_hash_entry	*hsh;
909 	struct flow_entry	*fle;
910 	struct flow_entry_data	*data = (struct flow_entry_data *)(resp + 1);
911 #ifdef INET6
912 	struct flow6_entry_data	*data6 = (struct flow6_entry_data *)(resp + 1);
913 #endif
914 	int	i, max;
915 
916 	i = req->hash_id;
917 	if (i > NBUCKETS-1)
918 		return (EINVAL);
919 
920 #ifdef INET6
921 	if (req->version == 6) {
922 		resp->version = 6;
923 		hsh = priv->hash6 + i;
924 		max = NREC6_AT_ONCE;
925 	} else
926 #endif
927 	if (req->version == 4) {
928 		resp->version = 4;
929 		hsh = priv->hash + i;
930 		max = NREC_AT_ONCE;
931 	} else
932 		return (EINVAL);
933 
934 	/*
935 	 * We will transfer not more than NREC_AT_ONCE. More data
936 	 * will come in next message.
937 	 * We send current hash index and current record number in list
938 	 * to userland, and userland should return it back to us.
939 	 * Then, we will restart with new entry.
940 	 *
941 	 * The resulting cache snapshot can be inaccurate if flow expiration
942 	 * is taking place on hash item between userland data requests for
943 	 * this hash item id.
944 	 */
945 	resp->nentries = 0;
946 	for (; i < NBUCKETS; hsh++, i++) {
947 		int list_id;
948 
949 		if (mtx_trylock(&hsh->mtx) == 0) {
950 			/*
951 			 * Requested hash index is not available,
952 			 * relay decision to skip or re-request data
953 			 * to userland.
954 			 */
955 			resp->hash_id = i;
956 			resp->list_id = 0;
957 			return (0);
958 		}
959 
960 		list_id = 0;
961 		TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
962 			if (hsh->mtx.mtx_lock & MTX_WAITERS) {
963 				resp->hash_id = i;
964 				resp->list_id = list_id;
965 				mtx_unlock(&hsh->mtx);
966 				return (0);
967 			}
968 
969 			list_id++;
970 			/* Search for particular record in list. */
971 			if (req->list_id > 0) {
972 				if (list_id < req->list_id)
973 					continue;
974 
975 				/* Requested list position found. */
976 				req->list_id = 0;
977 			}
978 #ifdef INET6
979 			if (req->version == 6) {
980 				struct flow6_entry *fle6;
981 
982 				fle6 = (struct flow6_entry *)fle;
983 				bcopy(&fle6->f, data6 + resp->nentries,
984 				    sizeof(fle6->f));
985 			} else
986 #endif
987 				bcopy(&fle->f, data + resp->nentries,
988 				    sizeof(fle->f));
989 			resp->nentries++;
990 			if (resp->nentries == max) {
991 				resp->hash_id = i;
992 				/*
993 				 * If it was the last item in list
994 				 * we simply skip to next hash_id.
995 				 */
996 				resp->list_id = list_id + 1;
997 				mtx_unlock(&hsh->mtx);
998 				return (0);
999 			}
1000 		}
1001 		mtx_unlock(&hsh->mtx);
1002 	}
1003 
1004 	resp->hash_id = resp->list_id = 0;
1005 
1006 	return (0);
1007 }
1008 
1009 /* We have full datagram in privdata. Send it to export hook. */
1010 static int
1011 export_send(priv_p priv, fib_export_p fe, item_p item, int flags)
1012 {
1013 	struct mbuf *m = NGI_M(item);
1014 	struct netflow_v5_export_dgram *dgram = mtod(m,
1015 					struct netflow_v5_export_dgram *);
1016 	struct netflow_v5_header *header = &dgram->header;
1017 	struct timespec ts;
1018 	int error = 0;
1019 
1020 	/* Fill mbuf header. */
1021 	m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
1022 	   header->count + sizeof(struct netflow_v5_header);
1023 
1024 	/* Fill export header. */
1025 	header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
1026 	getnanotime(&ts);
1027 	header->unix_secs  = htonl(ts.tv_sec);
1028 	header->unix_nsecs = htonl(ts.tv_nsec);
1029 	header->engine_type = 0;
1030 	header->engine_id = fe->domain_id;
1031 	header->pad = 0;
1032 	header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq,
1033 	    header->count));
1034 	header->count = htons(header->count);
1035 
1036 	if (priv->export != NULL)
1037 		NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
1038 	else
1039 		NG_FREE_ITEM(item);
1040 
1041 	return (error);
1042 }
1043 
1044 /* Add export record to dgram. */
1045 static int
1046 export_add(item_p item, struct flow_entry *fle)
1047 {
1048 	struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
1049 					struct netflow_v5_export_dgram *);
1050 	struct netflow_v5_header *header = &dgram->header;
1051 	struct netflow_v5_record *rec;
1052 
1053 	rec = &dgram->r[header->count];
1054 	header->count ++;
1055 
1056 	KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
1057 	    ("ng_netflow: export too big"));
1058 
1059 	/* Fill in export record. */
1060 	rec->src_addr = fle->f.r.r_src.s_addr;
1061 	rec->dst_addr = fle->f.r.r_dst.s_addr;
1062 	rec->next_hop = fle->f.next_hop.s_addr;
1063 	rec->i_ifx    = htons(fle->f.fle_i_ifx);
1064 	rec->o_ifx    = htons(fle->f.fle_o_ifx);
1065 	rec->packets  = htonl(fle->f.packets);
1066 	rec->octets   = htonl(fle->f.bytes);
1067 	rec->first    = htonl(MILLIUPTIME(fle->f.first));
1068 	rec->last     = htonl(MILLIUPTIME(fle->f.last));
1069 	rec->s_port   = fle->f.r.r_sport;
1070 	rec->d_port   = fle->f.r.r_dport;
1071 	rec->flags    = fle->f.tcp_flags;
1072 	rec->prot     = fle->f.r.r_ip_p;
1073 	rec->tos      = fle->f.r.r_tos;
1074 	rec->dst_mask = fle->f.dst_mask;
1075 	rec->src_mask = fle->f.src_mask;
1076 	rec->pad1     = 0;
1077 	rec->pad2     = 0;
1078 
1079 	/* Not supported fields. */
1080 	rec->src_as = rec->dst_as = 0;
1081 
1082 	if (header->count == NETFLOW_V5_MAX_RECORDS)
1083 		return (1); /* end of datagram */
1084 	else
1085 		return (0);
1086 }
1087 
1088 /* Periodic flow expiry run. */
1089 void
1090 ng_netflow_expire(void *arg)
1091 {
1092 	struct flow_entry	*fle, *fle1;
1093 	struct flow_hash_entry	*hsh;
1094 	priv_p			priv = (priv_p )arg;
1095 	int			used, i;
1096 
1097 	/*
1098 	 * Going through all the cache.
1099 	 */
1100 	used = uma_zone_get_cur(priv->zone);
1101 	for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
1102 		/*
1103 		 * Skip entries, that are already being worked on.
1104 		 */
1105 		if (mtx_trylock(&hsh->mtx) == 0)
1106 			continue;
1107 
1108 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1109 			/*
1110 			 * Interrupt thread wants this entry!
1111 			 * Quick! Quick! Bail out!
1112 			 */
1113 			if (hsh->mtx.mtx_lock & MTX_WAITERS)
1114 				break;
1115 
1116 			/*
1117 			 * Don't expire aggressively while hash collision
1118 			 * ratio is predicted small.
1119 			 */
1120 			if (used <= (NBUCKETS*2) && !INACTIVE(fle))
1121 				break;
1122 
1123 			if ((INACTIVE(fle) && (SMALL(fle) ||
1124 			    (used > (NBUCKETS*2)))) || AGED(fle)) {
1125 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1126 				expire_flow(priv, priv_to_fib(priv,
1127 				    fle->f.r.fib), fle, NG_NOFLAGS);
1128 				used--;
1129 				counter_u64_add(priv->nfinfo_inact_exp, 1);
1130 			}
1131 		}
1132 		mtx_unlock(&hsh->mtx);
1133 	}
1134 
1135 #ifdef INET6
1136 	used = uma_zone_get_cur(priv->zone6);
1137 	for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) {
1138 		struct flow6_entry	*fle6;
1139 
1140 		/*
1141 		 * Skip entries, that are already being worked on.
1142 		 */
1143 		if (mtx_trylock(&hsh->mtx) == 0)
1144 			continue;
1145 
1146 		TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1147 			fle6 = (struct flow6_entry *)fle;
1148 			/*
1149 			 * Interrupt thread wants this entry!
1150 			 * Quick! Quick! Bail out!
1151 			 */
1152 			if (hsh->mtx.mtx_lock & MTX_WAITERS)
1153 				break;
1154 
1155 			/*
1156 			 * Don't expire aggressively while hash collision
1157 			 * ratio is predicted small.
1158 			 */
1159 			if (used <= (NBUCKETS*2) && !INACTIVE(fle6))
1160 				break;
1161 
1162 			if ((INACTIVE(fle6) && (SMALL(fle6) ||
1163 			    (used > (NBUCKETS*2)))) || AGED(fle6)) {
1164 				TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1165 				expire_flow(priv, priv_to_fib(priv,
1166 				    fle->f.r.fib), fle, NG_NOFLAGS);
1167 				used--;
1168 				counter_u64_add(priv->nfinfo_inact_exp, 1);
1169 			}
1170 		}
1171 		mtx_unlock(&hsh->mtx);
1172 	}
1173 #endif
1174 
1175 	/* Schedule next expire. */
1176 	callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
1177 	    (void *)priv);
1178 }
1179