1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2010-2011 Alexander V. Chernikov <[email protected]>
5 * Copyright (c) 2004-2005 Gleb Smirnoff <[email protected]>
6 * Copyright (c) 2001-2003 Roman V. Palagin <[email protected]>
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 __FBSDID("$FreeBSD$");
35
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_route.h"
39 #include <sys/param.h>
40 #include <sys/bitstring.h>
41 #include <sys/systm.h>
42 #include <sys/counter.h>
43 #include <sys/kernel.h>
44 #include <sys/ktr.h>
45 #include <sys/limits.h>
46 #include <sys/mbuf.h>
47 #include <sys/syslog.h>
48 #include <sys/socket.h>
49 #include <vm/uma.h>
50
51 #include <net/if.h>
52 #include <net/if_dl.h>
53 #include <net/if_var.h>
54 #include <net/route.h>
55 #include <net/route/nhop.h>
56 #include <net/route/route_ctl.h>
57 #include <net/ethernet.h>
58 #include <netinet/in.h>
59 #include <netinet/in_fib.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/ip6.h>
63 #include <netinet/tcp.h>
64 #include <netinet/udp.h>
65
66 #include <netinet6/in6_fib.h>
67
68 #include <netgraph/ng_message.h>
69 #include <netgraph/netgraph.h>
70
71 #include <netgraph/netflow/netflow.h>
72 #include <netgraph/netflow/netflow_v9.h>
73 #include <netgraph/netflow/ng_netflow.h>
74
75 #define NBUCKETS (65536) /* must be power of 2 */
76
77 /* This hash is for TCP or UDP packets. */
78 #define FULL_HASH(addr1, addr2, port1, port2) \
79 (((addr1 ^ (addr1 >> 16) ^ \
80 htons(addr2 ^ (addr2 >> 16))) ^ \
81 port1 ^ htons(port2)) & \
82 (NBUCKETS - 1))
83
84 /* This hash is for all other IP packets. */
85 #define ADDR_HASH(addr1, addr2) \
86 ((addr1 ^ (addr1 >> 16) ^ \
87 htons(addr2 ^ (addr2 >> 16))) & \
88 (NBUCKETS - 1))
89
90 /* Macros to shorten logical constructions */
91 /* XXX: priv must exist in namespace */
92 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t)
93 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t)
94 #define ISFREE(fle) (fle->f.packets == 0)
95
96 /*
97 * 4 is a magical number: statistically number of 4-packet flows is
98 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
99 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
100 * of reachable host and 4-packet otherwise.
101 */
102 #define SMALL(fle) (fle->f.packets <= 4)
103
104 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
105
106 static int export_add(item_p, struct flow_entry *);
107 static int export_send(priv_p, fib_export_p, item_p, int);
108
109 #ifdef INET
110 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *,
111 int, uint8_t, uint8_t);
112 #endif
113 #ifdef INET6
114 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *,
115 int, uint8_t, uint8_t);
116 #endif
117
118 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int);
119
120 /*
121 * Generate hash for a given flow record.
122 *
123 * FIB is not used here, because:
124 * most VRFS will carry public IPv4 addresses which are unique even
125 * without FIB private addresses can overlap, but this is worked out
126 * via flow_rec bcmp() containing fib id. In IPv6 world addresses are
127 * all globally unique (it's not fully true, there is FC00::/7 for example,
128 * but chances of address overlap are MUCH smaller)
129 */
130 static inline uint32_t
ip_hash(struct flow_rec * r)131 ip_hash(struct flow_rec *r)
132 {
133
134 switch (r->r_ip_p) {
135 case IPPROTO_TCP:
136 case IPPROTO_UDP:
137 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
138 r->r_sport, r->r_dport);
139 default:
140 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
141 }
142 }
143
144 #ifdef INET6
145 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */
146 static inline uint32_t
ip6_hash(struct flow6_rec * r)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
get_export_dgram(priv_p priv,fib_export_p fe)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
return_export_dgram(priv_p priv,fib_export_p fe,item_p item,int flags)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
expire_flow(priv_p priv,fib_export_p fe,struct flow_entry * fle,int flags)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
ng_netflow_copyinfo(priv_p priv,struct ng_netflow_info * i)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
hash_insert(priv_p priv,struct flow_hash_entry * hsh,struct flow_rec * r,int plen,uint8_t flags,uint8_t tcp_flags)321 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r,
322 int plen, uint8_t flags, uint8_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
hash6_insert(priv_p priv,struct flow_hash_entry * hsh6,struct flow6_rec * r,int plen,uint8_t flags,uint8_t tcp_flags)398 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r,
399 int plen, uint8_t flags, uint8_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
ng_netflow_cache_init(priv_p priv)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
ng_netflow_fib_init(priv_p priv,int fib)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
ng_netflow_cache_flush(priv_p priv)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
ng_netflow_flow_add(priv_p priv,fib_export_p fe,struct ip * ip,caddr_t upper_ptr,uint8_t upper_proto,uint8_t flags,unsigned int src_if_index)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 eproto;
662 uint8_t tcp_flags = 0;
663
664 bzero(&r, sizeof(r));
665
666 if (ip->ip_v != IPVERSION)
667 return (EINVAL);
668
669 hlen = ip->ip_hl << 2;
670 if (hlen < sizeof(struct ip))
671 return (EINVAL);
672
673 eproto = ETHERTYPE_IP;
674 /* Assume L4 template by default */
675 r.flow_type = NETFLOW_V9_FLOW_V4_L4;
676
677 r.r_src = ip->ip_src;
678 r.r_dst = ip->ip_dst;
679 r.fib = fe->fib;
680
681 plen = ntohs(ip->ip_len);
682
683 r.r_ip_p = ip->ip_p;
684 r.r_tos = ip->ip_tos;
685
686 r.r_i_ifx = src_if_index;
687
688 /*
689 * XXX NOTE: only first fragment of fragmented TCP, UDP and
690 * ICMP packet will be recorded with proper s_port and d_port.
691 * Following fragments will be recorded simply as IP packet with
692 * ip_proto = ip->ip_p and s_port, d_port set to zero.
693 * I know, it looks like bug. But I don't want to re-implement
694 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
695 * and nobody complains yet :)
696 */
697 if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
698 switch(r.r_ip_p) {
699 case IPPROTO_TCP:
700 {
701 struct tcphdr *tcp;
702
703 tcp = (struct tcphdr *)((caddr_t )ip + hlen);
704 r.r_sport = tcp->th_sport;
705 r.r_dport = tcp->th_dport;
706 tcp_flags = tcp->th_flags;
707 break;
708 }
709 case IPPROTO_UDP:
710 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
711 break;
712 }
713
714 counter_u64_add(priv->nfinfo_packets, 1);
715 counter_u64_add(priv->nfinfo_bytes, plen);
716
717 /* Find hash slot. */
718 hsh = &priv->hash[ip_hash(&r)];
719
720 mtx_lock(&hsh->mtx);
721
722 /*
723 * Go through hash and find our entry. If we encounter an
724 * entry, that should be expired, purge it. We do a reverse
725 * search since most active entries are first, and most
726 * searches are done on most active entries.
727 */
728 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
729 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
730 break;
731 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
732 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
733 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
734 fle, NG_QUEUE);
735 counter_u64_add(priv->nfinfo_act_exp, 1);
736 }
737 }
738
739 if (fle) { /* An existent entry. */
740
741 fle->f.bytes += plen;
742 fle->f.packets ++;
743 fle->f.tcp_flags |= tcp_flags;
744 fle->f.last = time_uptime;
745
746 /*
747 * We have the following reasons to expire flow in active way:
748 * - it hit active timeout
749 * - a TCP connection closed
750 * - it is going to overflow counter
751 */
752 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
753 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
754 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
755 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
756 fle, NG_QUEUE);
757 counter_u64_add(priv->nfinfo_act_exp, 1);
758 } else {
759 /*
760 * It is the newest, move it to the tail,
761 * if it isn't there already. Next search will
762 * locate it quicker.
763 */
764 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
765 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
766 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
767 }
768 }
769 } else /* A new flow entry. */
770 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags);
771
772 mtx_unlock(&hsh->mtx);
773
774 return (error);
775 }
776 #endif
777
778 #ifdef INET6
779 /* Insert IPv6 packet from into flow cache. */
780 int
ng_netflow_flow6_add(priv_p priv,fib_export_p fe,struct ip6_hdr * ip6,caddr_t upper_ptr,uint8_t upper_proto,uint8_t flags,unsigned int src_if_index)781 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6,
782 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
783 unsigned int src_if_index)
784 {
785 struct flow_entry *fle = NULL, *fle1;
786 struct flow6_entry *fle6;
787 struct flow_hash_entry *hsh;
788 struct flow6_rec r;
789 int plen;
790 int error = 0;
791 uint8_t tcp_flags = 0;
792
793 /* check version */
794 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
795 return (EINVAL);
796
797 bzero(&r, sizeof(r));
798
799 r.src.r_src6 = ip6->ip6_src;
800 r.dst.r_dst6 = ip6->ip6_dst;
801 r.fib = fe->fib;
802
803 /* Assume L4 template by default */
804 r.flow_type = NETFLOW_V9_FLOW_V6_L4;
805
806 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
807
808 #if 0
809 /* XXX: set DSCP/CoS value */
810 r.r_tos = ip->ip_tos;
811 #endif
812 if ((flags & NG_NETFLOW_IS_FRAG) == 0) {
813 switch(upper_proto) {
814 case IPPROTO_TCP:
815 {
816 struct tcphdr *tcp;
817
818 tcp = (struct tcphdr *)upper_ptr;
819 r.r_ports = *(uint32_t *)upper_ptr;
820 tcp_flags = tcp->th_flags;
821 break;
822 }
823 case IPPROTO_UDP:
824 case IPPROTO_SCTP:
825 r.r_ports = *(uint32_t *)upper_ptr;
826 break;
827 }
828 }
829
830 r.r_ip_p = upper_proto;
831 r.r_i_ifx = src_if_index;
832
833 counter_u64_add(priv->nfinfo_packets6, 1);
834 counter_u64_add(priv->nfinfo_bytes6, plen);
835
836 /* Find hash slot. */
837 hsh = &priv->hash6[ip6_hash(&r)];
838
839 mtx_lock(&hsh->mtx);
840
841 /*
842 * Go through hash and find our entry. If we encounter an
843 * entry, that should be expired, purge it. We do a reverse
844 * search since most active entries are first, and most
845 * searches are done on most active entries.
846 */
847 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
848 if (fle->f.version != IP6VERSION)
849 continue;
850 fle6 = (struct flow6_entry *)fle;
851 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0)
852 break;
853 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) {
854 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
855 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
856 NG_QUEUE);
857 counter_u64_add(priv->nfinfo_act_exp, 1);
858 }
859 }
860
861 if (fle != NULL) { /* An existent entry. */
862 fle6 = (struct flow6_entry *)fle;
863
864 fle6->f.bytes += plen;
865 fle6->f.packets ++;
866 fle6->f.tcp_flags |= tcp_flags;
867 fle6->f.last = time_uptime;
868
869 /*
870 * We have the following reasons to expire flow in active way:
871 * - it hit active timeout
872 * - a TCP connection closed
873 * - it is going to overflow counter
874 */
875 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) ||
876 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
877 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
878 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
879 NG_QUEUE);
880 counter_u64_add(priv->nfinfo_act_exp, 1);
881 } else {
882 /*
883 * It is the newest, move it to the tail,
884 * if it isn't there already. Next search will
885 * locate it quicker.
886 */
887 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
888 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
889 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
890 }
891 }
892 } else /* A new flow entry. */
893 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags);
894
895 mtx_unlock(&hsh->mtx);
896
897 return (error);
898 }
899 #endif
900
901 /*
902 * Return records from cache to userland.
903 *
904 * TODO: matching particular IP should be done in kernel, here.
905 */
906 int
ng_netflow_flow_show(priv_p priv,struct ngnf_show_header * req,struct ngnf_show_header * resp)907 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req,
908 struct ngnf_show_header *resp)
909 {
910 struct flow_hash_entry *hsh;
911 struct flow_entry *fle;
912 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1);
913 #ifdef INET6
914 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1);
915 #endif
916 int i, max;
917
918 i = req->hash_id;
919 if (i > NBUCKETS-1)
920 return (EINVAL);
921
922 #ifdef INET6
923 if (req->version == 6) {
924 resp->version = 6;
925 hsh = priv->hash6 + i;
926 max = NREC6_AT_ONCE;
927 } else
928 #endif
929 if (req->version == 4) {
930 resp->version = 4;
931 hsh = priv->hash + i;
932 max = NREC_AT_ONCE;
933 } else
934 return (EINVAL);
935
936 /*
937 * We will transfer not more than NREC_AT_ONCE. More data
938 * will come in next message.
939 * We send current hash index and current record number in list
940 * to userland, and userland should return it back to us.
941 * Then, we will restart with new entry.
942 *
943 * The resulting cache snapshot can be inaccurate if flow expiration
944 * is taking place on hash item between userland data requests for
945 * this hash item id.
946 */
947 resp->nentries = 0;
948 for (; i < NBUCKETS; hsh++, i++) {
949 int list_id;
950
951 if (mtx_trylock(&hsh->mtx) == 0) {
952 /*
953 * Requested hash index is not available,
954 * relay decision to skip or re-request data
955 * to userland.
956 */
957 resp->hash_id = i;
958 resp->list_id = 0;
959 return (0);
960 }
961
962 list_id = 0;
963 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
964 if (hsh->mtx.mtx_lock & MTX_CONTESTED) {
965 resp->hash_id = i;
966 resp->list_id = list_id;
967 mtx_unlock(&hsh->mtx);
968 return (0);
969 }
970
971 list_id++;
972 /* Search for particular record in list. */
973 if (req->list_id > 0) {
974 if (list_id < req->list_id)
975 continue;
976
977 /* Requested list position found. */
978 req->list_id = 0;
979 }
980 #ifdef INET6
981 if (req->version == 6) {
982 struct flow6_entry *fle6;
983
984 fle6 = (struct flow6_entry *)fle;
985 bcopy(&fle6->f, data6 + resp->nentries,
986 sizeof(fle6->f));
987 } else
988 #endif
989 bcopy(&fle->f, data + resp->nentries,
990 sizeof(fle->f));
991 resp->nentries++;
992 if (resp->nentries == max) {
993 resp->hash_id = i;
994 /*
995 * If it was the last item in list
996 * we simply skip to next hash_id.
997 */
998 resp->list_id = list_id + 1;
999 mtx_unlock(&hsh->mtx);
1000 return (0);
1001 }
1002 }
1003 mtx_unlock(&hsh->mtx);
1004 }
1005
1006 resp->hash_id = resp->list_id = 0;
1007
1008 return (0);
1009 }
1010
1011 /* We have full datagram in privdata. Send it to export hook. */
1012 static int
export_send(priv_p priv,fib_export_p fe,item_p item,int flags)1013 export_send(priv_p priv, fib_export_p fe, item_p item, int flags)
1014 {
1015 struct mbuf *m = NGI_M(item);
1016 struct netflow_v5_export_dgram *dgram = mtod(m,
1017 struct netflow_v5_export_dgram *);
1018 struct netflow_v5_header *header = &dgram->header;
1019 struct timespec ts;
1020 int error = 0;
1021
1022 /* Fill mbuf header. */
1023 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
1024 header->count + sizeof(struct netflow_v5_header);
1025
1026 /* Fill export header. */
1027 header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
1028 getnanotime(&ts);
1029 header->unix_secs = htonl(ts.tv_sec);
1030 header->unix_nsecs = htonl(ts.tv_nsec);
1031 header->engine_type = 0;
1032 header->engine_id = fe->domain_id;
1033 header->pad = 0;
1034 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq,
1035 header->count));
1036 header->count = htons(header->count);
1037
1038 if (priv->export != NULL)
1039 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
1040 else
1041 NG_FREE_ITEM(item);
1042
1043 return (error);
1044 }
1045
1046 /* Add export record to dgram. */
1047 static int
export_add(item_p item,struct flow_entry * fle)1048 export_add(item_p item, struct flow_entry *fle)
1049 {
1050 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
1051 struct netflow_v5_export_dgram *);
1052 struct netflow_v5_header *header = &dgram->header;
1053 struct netflow_v5_record *rec;
1054
1055 rec = &dgram->r[header->count];
1056 header->count ++;
1057
1058 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
1059 ("ng_netflow: export too big"));
1060
1061 /* Fill in export record. */
1062 rec->src_addr = fle->f.r.r_src.s_addr;
1063 rec->dst_addr = fle->f.r.r_dst.s_addr;
1064 rec->next_hop = fle->f.next_hop.s_addr;
1065 rec->i_ifx = htons(fle->f.fle_i_ifx);
1066 rec->o_ifx = htons(fle->f.fle_o_ifx);
1067 rec->packets = htonl(fle->f.packets);
1068 rec->octets = htonl(fle->f.bytes);
1069 rec->first = htonl(MILLIUPTIME(fle->f.first));
1070 rec->last = htonl(MILLIUPTIME(fle->f.last));
1071 rec->s_port = fle->f.r.r_sport;
1072 rec->d_port = fle->f.r.r_dport;
1073 rec->flags = fle->f.tcp_flags;
1074 rec->prot = fle->f.r.r_ip_p;
1075 rec->tos = fle->f.r.r_tos;
1076 rec->dst_mask = fle->f.dst_mask;
1077 rec->src_mask = fle->f.src_mask;
1078 rec->pad1 = 0;
1079 rec->pad2 = 0;
1080
1081 /* Not supported fields. */
1082 rec->src_as = rec->dst_as = 0;
1083
1084 if (header->count == NETFLOW_V5_MAX_RECORDS)
1085 return (1); /* end of datagram */
1086 else
1087 return (0);
1088 }
1089
1090 /* Periodic flow expiry run. */
1091 void
ng_netflow_expire(void * arg)1092 ng_netflow_expire(void *arg)
1093 {
1094 struct flow_entry *fle, *fle1;
1095 struct flow_hash_entry *hsh;
1096 priv_p priv = (priv_p )arg;
1097 int used, i;
1098
1099 /*
1100 * Going through all the cache.
1101 */
1102 used = uma_zone_get_cur(priv->zone);
1103 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
1104 /*
1105 * Skip entries, that are already being worked on.
1106 */
1107 if (mtx_trylock(&hsh->mtx) == 0)
1108 continue;
1109
1110 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1111 /*
1112 * Interrupt thread wants this entry!
1113 * Quick! Quick! Bail out!
1114 */
1115 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1116 break;
1117
1118 /*
1119 * Don't expire aggressively while hash collision
1120 * ratio is predicted small.
1121 */
1122 if (used <= (NBUCKETS*2) && !INACTIVE(fle))
1123 break;
1124
1125 if ((INACTIVE(fle) && (SMALL(fle) ||
1126 (used > (NBUCKETS*2)))) || AGED(fle)) {
1127 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1128 expire_flow(priv, priv_to_fib(priv,
1129 fle->f.r.fib), fle, NG_NOFLAGS);
1130 used--;
1131 counter_u64_add(priv->nfinfo_inact_exp, 1);
1132 }
1133 }
1134 mtx_unlock(&hsh->mtx);
1135 }
1136
1137 #ifdef INET6
1138 used = uma_zone_get_cur(priv->zone6);
1139 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) {
1140 struct flow6_entry *fle6;
1141
1142 /*
1143 * Skip entries, that are already being worked on.
1144 */
1145 if (mtx_trylock(&hsh->mtx) == 0)
1146 continue;
1147
1148 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1149 fle6 = (struct flow6_entry *)fle;
1150 /*
1151 * Interrupt thread wants this entry!
1152 * Quick! Quick! Bail out!
1153 */
1154 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1155 break;
1156
1157 /*
1158 * Don't expire aggressively while hash collision
1159 * ratio is predicted small.
1160 */
1161 if (used <= (NBUCKETS*2) && !INACTIVE(fle6))
1162 break;
1163
1164 if ((INACTIVE(fle6) && (SMALL(fle6) ||
1165 (used > (NBUCKETS*2)))) || AGED(fle6)) {
1166 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1167 expire_flow(priv, priv_to_fib(priv,
1168 fle->f.r.fib), fle, NG_NOFLAGS);
1169 used--;
1170 counter_u64_add(priv->nfinfo_inact_exp, 1);
1171 }
1172 }
1173 mtx_unlock(&hsh->mtx);
1174 }
1175 #endif
1176
1177 /* Schedule next expire. */
1178 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
1179 (void *)priv);
1180 }
1181