1 /*-
2 * Copyright (c) 2010-2011 Alexander V. Chernikov <melifaro@ipfw.ru>
3 * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
4 * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 *
28 * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
29 */
30
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33
34 #include "opt_inet6.h"
35 #include "opt_route.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/counter.h>
39 #include <sys/kernel.h>
40 #include <sys/ktr.h>
41 #include <sys/limits.h>
42 #include <sys/mbuf.h>
43 #include <sys/syslog.h>
44 #include <sys/socket.h>
45
46 #include <net/if.h>
47 #include <net/if_var.h>
48 #include <net/route.h>
49 #include <net/ethernet.h>
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/ip6.h>
54 #include <netinet/tcp.h>
55 #include <netinet/udp.h>
56
57 #include <netgraph/ng_message.h>
58 #include <netgraph/netgraph.h>
59
60 #include <netgraph/netflow/netflow.h>
61 #include <netgraph/netflow/netflow_v9.h>
62 #include <netgraph/netflow/ng_netflow.h>
63
64 #define NBUCKETS (65536) /* must be power of 2 */
65
66 /* This hash is for TCP or UDP packets. */
67 #define FULL_HASH(addr1, addr2, port1, port2) \
68 (((addr1 ^ (addr1 >> 16) ^ \
69 htons(addr2 ^ (addr2 >> 16))) ^ \
70 port1 ^ htons(port2)) & \
71 (NBUCKETS - 1))
72
73 /* This hash is for all other IP packets. */
74 #define ADDR_HASH(addr1, addr2) \
75 ((addr1 ^ (addr1 >> 16) ^ \
76 htons(addr2 ^ (addr2 >> 16))) & \
77 (NBUCKETS - 1))
78
79 /* Macros to shorten logical constructions */
80 /* XXX: priv must exist in namespace */
81 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t)
82 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t)
83 #define ISFREE(fle) (fle->f.packets == 0)
84
85 /*
86 * 4 is a magical number: statistically number of 4-packet flows is
87 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
88 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
89 * of reachable host and 4-packet otherwise.
90 */
91 #define SMALL(fle) (fle->f.packets <= 4)
92
93 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
94
95 static int export_add(item_p, struct flow_entry *);
96 static int export_send(priv_p, fib_export_p, item_p, int);
97
98 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *,
99 int, uint8_t, uint8_t);
100 #ifdef INET6
101 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *,
102 int, uint8_t, uint8_t);
103 #endif
104
105 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int);
106
107 /*
108 * Generate hash for a given flow record.
109 *
110 * FIB is not used here, because:
111 * most VRFS will carry public IPv4 addresses which are unique even
112 * without FIB private addresses can overlap, but this is worked out
113 * via flow_rec bcmp() containing fib id. In IPv6 world addresses are
114 * all globally unique (it's not fully true, there is FC00::/7 for example,
115 * but chances of address overlap are MUCH smaller)
116 */
117 static inline uint32_t
ip_hash(struct flow_rec * r)118 ip_hash(struct flow_rec *r)
119 {
120
121 switch (r->r_ip_p) {
122 case IPPROTO_TCP:
123 case IPPROTO_UDP:
124 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
125 r->r_sport, r->r_dport);
126 default:
127 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
128 }
129 }
130
131 #ifdef INET6
132 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */
133 static inline uint32_t
ip6_hash(struct flow6_rec * r)134 ip6_hash(struct flow6_rec *r)
135 {
136
137 switch (r->r_ip_p) {
138 case IPPROTO_TCP:
139 case IPPROTO_UDP:
140 return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
141 r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport,
142 r->r_dport);
143 default:
144 return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
145 r->dst.r_dst6.__u6_addr.__u6_addr32[3]);
146 }
147 }
148 #endif
149
150 /*
151 * Detach export datagram from priv, if there is any.
152 * If there is no, allocate a new one.
153 */
154 static item_p
get_export_dgram(priv_p priv,fib_export_p fe)155 get_export_dgram(priv_p priv, fib_export_p fe)
156 {
157 item_p item = NULL;
158
159 mtx_lock(&fe->export_mtx);
160 if (fe->exp.item != NULL) {
161 item = fe->exp.item;
162 fe->exp.item = NULL;
163 }
164 mtx_unlock(&fe->export_mtx);
165
166 if (item == NULL) {
167 struct netflow_v5_export_dgram *dgram;
168 struct mbuf *m;
169
170 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
171 if (m == NULL)
172 return (NULL);
173 item = ng_package_data(m, NG_NOFLAGS);
174 if (item == NULL)
175 return (NULL);
176 dgram = mtod(m, struct netflow_v5_export_dgram *);
177 dgram->header.count = 0;
178 dgram->header.version = htons(NETFLOW_V5);
179 dgram->header.pad = 0;
180 }
181
182 return (item);
183 }
184
185 /*
186 * Re-attach incomplete datagram back to priv.
187 * If there is already another one, then send incomplete. */
188 static void
return_export_dgram(priv_p priv,fib_export_p fe,item_p item,int flags)189 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags)
190 {
191
192 /*
193 * It may happen on SMP, that some thread has already
194 * put its item there, in this case we bail out and
195 * send what we have to collector.
196 */
197 mtx_lock(&fe->export_mtx);
198 if (fe->exp.item == NULL) {
199 fe->exp.item = item;
200 mtx_unlock(&fe->export_mtx);
201 } else {
202 mtx_unlock(&fe->export_mtx);
203 export_send(priv, fe, item, flags);
204 }
205 }
206
207 /*
208 * The flow is over. Call export_add() and free it. If datagram is
209 * full, then call export_send().
210 */
211 static void
expire_flow(priv_p priv,fib_export_p fe,struct flow_entry * fle,int flags)212 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags)
213 {
214 struct netflow_export_item exp;
215 uint16_t version = fle->f.version;
216
217 if ((priv->export != NULL) && (version == IPVERSION)) {
218 exp.item = get_export_dgram(priv, fe);
219 if (exp.item == NULL) {
220 priv->nfinfo_export_failed++;
221 if (priv->export9 != NULL)
222 priv->nfinfo_export9_failed++;
223 /* fle definitely contains IPv4 flow. */
224 uma_zfree_arg(priv->zone, fle, priv);
225 return;
226 }
227
228 if (export_add(exp.item, fle) > 0)
229 export_send(priv, fe, exp.item, flags);
230 else
231 return_export_dgram(priv, fe, exp.item, NG_QUEUE);
232 }
233
234 if (priv->export9 != NULL) {
235 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt);
236 if (exp.item9 == NULL) {
237 priv->nfinfo_export9_failed++;
238 if (version == IPVERSION)
239 uma_zfree_arg(priv->zone, fle, priv);
240 #ifdef INET6
241 else if (version == IP6VERSION)
242 uma_zfree_arg(priv->zone6, fle, priv);
243 #endif
244 else
245 panic("ng_netflow: Unknown IP proto: %d",
246 version);
247 return;
248 }
249
250 if (export9_add(exp.item9, exp.item9_opt, fle) > 0)
251 export9_send(priv, fe, exp.item9, exp.item9_opt, flags);
252 else
253 return_export9_dgram(priv, fe, exp.item9,
254 exp.item9_opt, NG_QUEUE);
255 }
256
257 if (version == IPVERSION)
258 uma_zfree_arg(priv->zone, fle, priv);
259 #ifdef INET6
260 else if (version == IP6VERSION)
261 uma_zfree_arg(priv->zone6, fle, priv);
262 #endif
263 }
264
265 /* Get a snapshot of node statistics */
266 void
ng_netflow_copyinfo(priv_p priv,struct ng_netflow_info * i)267 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
268 {
269
270 i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes);
271 i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets);
272 i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6);
273 i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6);
274 i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes);
275 i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets);
276 i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6);
277 i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6);
278 i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp);
279 i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp);
280
281 i->nfinfo_used = uma_zone_get_cur(priv->zone);
282 #ifdef INET6
283 i->nfinfo_used6 = uma_zone_get_cur(priv->zone6);
284 #endif
285
286 i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed;
287 i->nfinfo_export_failed = priv->nfinfo_export_failed;
288 i->nfinfo_export9_failed = priv->nfinfo_export9_failed;
289 i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf;
290 i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs;
291 i->nfinfo_inact_t = priv->nfinfo_inact_t;
292 i->nfinfo_act_t = priv->nfinfo_act_t;
293 }
294
295 /*
296 * Insert a record into defined slot.
297 *
298 * First we get for us a free flow entry, then fill in all
299 * possible fields in it.
300 *
301 * TODO: consider dropping hash mutex while filling in datagram,
302 * as this was done in previous version. Need to test & profile
303 * to be sure.
304 */
305 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)306 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r,
307 int plen, uint8_t flags, uint8_t tcp_flags)
308 {
309 struct flow_entry *fle;
310 struct sockaddr_in sin;
311 struct rtentry *rt;
312
313 mtx_assert(&hsh->mtx, MA_OWNED);
314
315 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT);
316 if (fle == NULL) {
317 priv->nfinfo_alloc_failed++;
318 return (ENOMEM);
319 }
320
321 /*
322 * Now fle is totally ours. It is detached from all lists,
323 * we can safely edit it.
324 */
325 fle->f.version = IPVERSION;
326 bcopy(r, &fle->f.r, sizeof(struct flow_rec));
327 fle->f.bytes = plen;
328 fle->f.packets = 1;
329 fle->f.tcp_flags = tcp_flags;
330
331 fle->f.first = fle->f.last = time_uptime;
332
333 /*
334 * First we do route table lookup on destination address. So we can
335 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
336 */
337 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
338 bzero(&sin, sizeof(sin));
339 sin.sin_len = sizeof(struct sockaddr_in);
340 sin.sin_family = AF_INET;
341 sin.sin_addr = fle->f.r.r_dst;
342 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib);
343 if (rt != NULL) {
344 fle->f.fle_o_ifx = rt->rt_ifp->if_index;
345
346 if (rt->rt_flags & RTF_GATEWAY &&
347 rt->rt_gateway->sa_family == AF_INET)
348 fle->f.next_hop =
349 ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
350
351 if (rt_mask(rt))
352 fle->f.dst_mask =
353 bitcount32(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
354 else if (rt->rt_flags & RTF_HOST)
355 /* Give up. We can't determine mask :( */
356 fle->f.dst_mask = 32;
357
358 RTFREE_LOCKED(rt);
359 }
360 }
361
362 /* Do route lookup on source address, to fill in src_mask. */
363 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
364 bzero(&sin, sizeof(sin));
365 sin.sin_len = sizeof(struct sockaddr_in);
366 sin.sin_family = AF_INET;
367 sin.sin_addr = fle->f.r.r_src;
368 rt = rtalloc1_fib((struct sockaddr *)&sin, 0, 0, r->fib);
369 if (rt != NULL) {
370 if (rt_mask(rt))
371 fle->f.src_mask =
372 bitcount32(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
373 else if (rt->rt_flags & RTF_HOST)
374 /* Give up. We can't determine mask :( */
375 fle->f.src_mask = 32;
376
377 RTFREE_LOCKED(rt);
378 }
379 }
380
381 /* Push new flow at the and of hash. */
382 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
383
384 return (0);
385 }
386
387 #ifdef INET6
388 /* XXX: make normal function, instead of.. */
389 #define ipv6_masklen(x) bitcount32((x).__u6_addr.__u6_addr32[0]) + \
390 bitcount32((x).__u6_addr.__u6_addr32[1]) + \
391 bitcount32((x).__u6_addr.__u6_addr32[2]) + \
392 bitcount32((x).__u6_addr.__u6_addr32[3])
393 #define RT_MASK6(x) (ipv6_masklen(((struct sockaddr_in6 *)rt_mask(x))->sin6_addr))
394 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)395 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r,
396 int plen, uint8_t flags, uint8_t tcp_flags)
397 {
398 struct flow6_entry *fle6;
399 struct sockaddr_in6 sin6;
400 struct rtentry *rt;
401
402 mtx_assert(&hsh6->mtx, MA_OWNED);
403
404 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT);
405 if (fle6 == NULL) {
406 priv->nfinfo_alloc_failed++;
407 return (ENOMEM);
408 }
409
410 /*
411 * Now fle is totally ours. It is detached from all lists,
412 * we can safely edit it.
413 */
414
415 fle6->f.version = IP6VERSION;
416 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec));
417 fle6->f.bytes = plen;
418 fle6->f.packets = 1;
419 fle6->f.tcp_flags = tcp_flags;
420
421 fle6->f.first = fle6->f.last = time_uptime;
422
423 /*
424 * First we do route table lookup on destination address. So we can
425 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
426 */
427 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
428 bzero(&sin6, sizeof(struct sockaddr_in6));
429 sin6.sin6_len = sizeof(struct sockaddr_in6);
430 sin6.sin6_family = AF_INET6;
431 sin6.sin6_addr = r->dst.r_dst6;
432
433 rt = rtalloc1_fib((struct sockaddr *)&sin6, 0, 0, r->fib);
434
435 if (rt != NULL) {
436 fle6->f.fle_o_ifx = rt->rt_ifp->if_index;
437
438 if (rt->rt_flags & RTF_GATEWAY &&
439 rt->rt_gateway->sa_family == AF_INET6)
440 fle6->f.n.next_hop6 =
441 ((struct sockaddr_in6 *)(rt->rt_gateway))->sin6_addr;
442
443 if (rt_mask(rt))
444 fle6->f.dst_mask = RT_MASK6(rt);
445 else
446 fle6->f.dst_mask = 128;
447
448 RTFREE_LOCKED(rt);
449 }
450 }
451
452 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
453 /* Do route lookup on source address, to fill in src_mask. */
454 bzero(&sin6, sizeof(struct sockaddr_in6));
455 sin6.sin6_len = sizeof(struct sockaddr_in6);
456 sin6.sin6_family = AF_INET6;
457 sin6.sin6_addr = r->src.r_src6;
458
459 rt = rtalloc1_fib((struct sockaddr *)&sin6, 0, 0, r->fib);
460
461 if (rt != NULL) {
462 if (rt_mask(rt))
463 fle6->f.src_mask = RT_MASK6(rt);
464 else
465 fle6->f.src_mask = 128;
466
467 RTFREE_LOCKED(rt);
468 }
469 }
470
471 /* Push new flow at the and of hash. */
472 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash);
473
474 return (0);
475 }
476 #undef ipv6_masklen
477 #undef RT_MASK6
478 #endif
479
480
481 /*
482 * Non-static functions called from ng_netflow.c
483 */
484
485 /* Allocate memory and set up flow cache */
486 void
ng_netflow_cache_init(priv_p priv)487 ng_netflow_cache_init(priv_p priv)
488 {
489 struct flow_hash_entry *hsh;
490 int i;
491
492 /* Initialize cache UMA zone. */
493 priv->zone = uma_zcreate("NetFlow IPv4 cache",
494 sizeof(struct flow_entry), NULL, NULL, NULL, NULL,
495 UMA_ALIGN_CACHE, 0);
496 uma_zone_set_max(priv->zone, CACHESIZE);
497 #ifdef INET6
498 priv->zone6 = uma_zcreate("NetFlow IPv6 cache",
499 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL,
500 UMA_ALIGN_CACHE, 0);
501 uma_zone_set_max(priv->zone6, CACHESIZE);
502 #endif
503
504 /* Allocate hash. */
505 priv->hash = 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->hash; i < NBUCKETS; i++, hsh++) {
510 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
511 TAILQ_INIT(&hsh->head);
512 }
513
514 #ifdef INET6
515 /* Allocate hash. */
516 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
517 M_NETFLOW_HASH, M_WAITOK | M_ZERO);
518
519 /* Initialize hash. */
520 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) {
521 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
522 TAILQ_INIT(&hsh->head);
523 }
524 #endif
525
526 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK);
527 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK);
528 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK);
529 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK);
530 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK);
531 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK);
532 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK);
533 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK);
534 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK);
535 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK);
536
537 ng_netflow_v9_cache_init(priv);
538 CTR0(KTR_NET, "ng_netflow startup()");
539 }
540
541 /* Initialize new FIB table for v5 and v9 */
542 int
ng_netflow_fib_init(priv_p priv,int fib)543 ng_netflow_fib_init(priv_p priv, int fib)
544 {
545 fib_export_p fe = priv_to_fib(priv, fib);
546
547 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib);
548
549 if (fe != NULL)
550 return (0);
551
552 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH,
553 M_NOWAIT | M_ZERO)) == NULL)
554 return (ENOMEM);
555
556 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF);
557 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF);
558 fe->fib = fib;
559 fe->domain_id = fib;
560
561 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib],
562 (uintptr_t)NULL, (uintptr_t)fe) == 0) {
563 /* FIB already set up by other ISR */
564 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p",
565 fib, fe, priv_to_fib(priv, fib));
566 mtx_destroy(&fe->export_mtx);
567 mtx_destroy(&fe->export9_mtx);
568 free(fe, M_NETGRAPH);
569 } else {
570 /* Increase counter for statistics */
571 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)",
572 fib, fe, priv_to_fib(priv, fib));
573 priv->nfinfo_alloc_fibs++;
574 }
575
576 return (0);
577 }
578
579 /* Free all flow cache memory. Called from node close method. */
580 void
ng_netflow_cache_flush(priv_p priv)581 ng_netflow_cache_flush(priv_p priv)
582 {
583 struct flow_entry *fle, *fle1;
584 struct flow_hash_entry *hsh;
585 struct netflow_export_item exp;
586 fib_export_p fe;
587 int i;
588
589 bzero(&exp, sizeof(exp));
590
591 /*
592 * We are going to free probably billable data.
593 * Expire everything before freeing it.
594 * No locking is required since callout is already drained.
595 */
596 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
597 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
598 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
599 fe = priv_to_fib(priv, fle->f.r.fib);
600 expire_flow(priv, fe, fle, NG_QUEUE);
601 }
602 #ifdef INET6
603 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++)
604 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
605 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
606 fe = priv_to_fib(priv, fle->f.r.fib);
607 expire_flow(priv, fe, fle, NG_QUEUE);
608 }
609 #endif
610
611 uma_zdestroy(priv->zone);
612 /* Destroy hash mutexes. */
613 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
614 mtx_destroy(&hsh->mtx);
615
616 /* Free hash memory. */
617 if (priv->hash != NULL)
618 free(priv->hash, M_NETFLOW_HASH);
619 #ifdef INET6
620 uma_zdestroy(priv->zone6);
621 /* Destroy hash mutexes. */
622 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++)
623 mtx_destroy(&hsh->mtx);
624
625 /* Free hash memory. */
626 if (priv->hash6 != NULL)
627 free(priv->hash6, M_NETFLOW_HASH);
628 #endif
629
630 for (i = 0; i < priv->maxfibs; i++) {
631 if ((fe = priv_to_fib(priv, i)) == NULL)
632 continue;
633
634 if (fe->exp.item != NULL)
635 export_send(priv, fe, fe->exp.item, NG_QUEUE);
636
637 if (fe->exp.item9 != NULL)
638 export9_send(priv, fe, fe->exp.item9,
639 fe->exp.item9_opt, NG_QUEUE);
640
641 mtx_destroy(&fe->export_mtx);
642 mtx_destroy(&fe->export9_mtx);
643 free(fe, M_NETGRAPH);
644 }
645
646 counter_u64_free(priv->nfinfo_bytes);
647 counter_u64_free(priv->nfinfo_packets);
648 counter_u64_free(priv->nfinfo_bytes6);
649 counter_u64_free(priv->nfinfo_packets6);
650 counter_u64_free(priv->nfinfo_sbytes);
651 counter_u64_free(priv->nfinfo_spackets);
652 counter_u64_free(priv->nfinfo_sbytes6);
653 counter_u64_free(priv->nfinfo_spackets6);
654 counter_u64_free(priv->nfinfo_act_exp);
655 counter_u64_free(priv->nfinfo_inact_exp);
656
657 ng_netflow_v9_cache_flush(priv);
658 }
659
660 /* Insert packet from into flow cache. */
661 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)662 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip,
663 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
664 unsigned int src_if_index)
665 {
666 struct flow_entry *fle, *fle1;
667 struct flow_hash_entry *hsh;
668 struct flow_rec r;
669 int hlen, plen;
670 int error = 0;
671 uint16_t eproto;
672 uint8_t tcp_flags = 0;
673
674 bzero(&r, sizeof(r));
675
676 if (ip->ip_v != IPVERSION)
677 return (EINVAL);
678
679 hlen = ip->ip_hl << 2;
680 if (hlen < sizeof(struct ip))
681 return (EINVAL);
682
683 eproto = ETHERTYPE_IP;
684 /* Assume L4 template by default */
685 r.flow_type = NETFLOW_V9_FLOW_V4_L4;
686
687 r.r_src = ip->ip_src;
688 r.r_dst = ip->ip_dst;
689 r.fib = fe->fib;
690
691 plen = ntohs(ip->ip_len);
692
693 r.r_ip_p = ip->ip_p;
694 r.r_tos = ip->ip_tos;
695
696 r.r_i_ifx = src_if_index;
697
698 /*
699 * XXX NOTE: only first fragment of fragmented TCP, UDP and
700 * ICMP packet will be recorded with proper s_port and d_port.
701 * Following fragments will be recorded simply as IP packet with
702 * ip_proto = ip->ip_p and s_port, d_port set to zero.
703 * I know, it looks like bug. But I don't want to re-implement
704 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
705 * and nobody complains yet :)
706 */
707 if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
708 switch(r.r_ip_p) {
709 case IPPROTO_TCP:
710 {
711 struct tcphdr *tcp;
712
713 tcp = (struct tcphdr *)((caddr_t )ip + hlen);
714 r.r_sport = tcp->th_sport;
715 r.r_dport = tcp->th_dport;
716 tcp_flags = tcp->th_flags;
717 break;
718 }
719 case IPPROTO_UDP:
720 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
721 break;
722 }
723
724 counter_u64_add(priv->nfinfo_packets, 1);
725 counter_u64_add(priv->nfinfo_bytes, plen);
726
727 /* Find hash slot. */
728 hsh = &priv->hash[ip_hash(&r)];
729
730 mtx_lock(&hsh->mtx);
731
732 /*
733 * Go through hash and find our entry. If we encounter an
734 * entry, that should be expired, purge it. We do a reverse
735 * search since most active entries are first, and most
736 * searches are done on most active entries.
737 */
738 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
739 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
740 break;
741 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
742 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
743 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
744 fle, NG_QUEUE);
745 counter_u64_add(priv->nfinfo_act_exp, 1);
746 }
747 }
748
749 if (fle) { /* An existent entry. */
750
751 fle->f.bytes += plen;
752 fle->f.packets ++;
753 fle->f.tcp_flags |= tcp_flags;
754 fle->f.last = time_uptime;
755
756 /*
757 * We have the following reasons to expire flow in active way:
758 * - it hit active timeout
759 * - a TCP connection closed
760 * - it is going to overflow counter
761 */
762 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
763 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
764 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
765 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
766 fle, NG_QUEUE);
767 counter_u64_add(priv->nfinfo_act_exp, 1);
768 } else {
769 /*
770 * It is the newest, move it to the tail,
771 * if it isn't there already. Next search will
772 * locate it quicker.
773 */
774 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
775 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
776 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
777 }
778 }
779 } else /* A new flow entry. */
780 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags);
781
782 mtx_unlock(&hsh->mtx);
783
784 return (error);
785 }
786
787 #ifdef INET6
788 /* Insert IPv6 packet from into flow cache. */
789 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)790 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6,
791 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
792 unsigned int src_if_index)
793 {
794 struct flow_entry *fle = NULL, *fle1;
795 struct flow6_entry *fle6;
796 struct flow_hash_entry *hsh;
797 struct flow6_rec r;
798 int plen;
799 int error = 0;
800 uint8_t tcp_flags = 0;
801
802 /* check version */
803 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
804 return (EINVAL);
805
806 bzero(&r, sizeof(r));
807
808 r.src.r_src6 = ip6->ip6_src;
809 r.dst.r_dst6 = ip6->ip6_dst;
810 r.fib = fe->fib;
811
812 /* Assume L4 template by default */
813 r.flow_type = NETFLOW_V9_FLOW_V6_L4;
814
815 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
816
817 #if 0
818 /* XXX: set DSCP/CoS value */
819 r.r_tos = ip->ip_tos;
820 #endif
821 if ((flags & NG_NETFLOW_IS_FRAG) == 0) {
822 switch(upper_proto) {
823 case IPPROTO_TCP:
824 {
825 struct tcphdr *tcp;
826
827 tcp = (struct tcphdr *)upper_ptr;
828 r.r_ports = *(uint32_t *)upper_ptr;
829 tcp_flags = tcp->th_flags;
830 break;
831 }
832 case IPPROTO_UDP:
833 case IPPROTO_SCTP:
834 r.r_ports = *(uint32_t *)upper_ptr;
835 break;
836 }
837 }
838
839 r.r_ip_p = upper_proto;
840 r.r_i_ifx = src_if_index;
841
842 counter_u64_add(priv->nfinfo_packets6, 1);
843 counter_u64_add(priv->nfinfo_bytes6, plen);
844
845 /* Find hash slot. */
846 hsh = &priv->hash6[ip6_hash(&r)];
847
848 mtx_lock(&hsh->mtx);
849
850 /*
851 * Go through hash and find our entry. If we encounter an
852 * entry, that should be expired, purge it. We do a reverse
853 * search since most active entries are first, and most
854 * searches are done on most active entries.
855 */
856 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
857 if (fle->f.version != IP6VERSION)
858 continue;
859 fle6 = (struct flow6_entry *)fle;
860 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0)
861 break;
862 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) {
863 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
864 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
865 NG_QUEUE);
866 counter_u64_add(priv->nfinfo_act_exp, 1);
867 }
868 }
869
870 if (fle != NULL) { /* An existent entry. */
871 fle6 = (struct flow6_entry *)fle;
872
873 fle6->f.bytes += plen;
874 fle6->f.packets ++;
875 fle6->f.tcp_flags |= tcp_flags;
876 fle6->f.last = time_uptime;
877
878 /*
879 * We have the following reasons to expire flow in active way:
880 * - it hit active timeout
881 * - a TCP connection closed
882 * - it is going to overflow counter
883 */
884 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) ||
885 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
886 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
887 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
888 NG_QUEUE);
889 counter_u64_add(priv->nfinfo_act_exp, 1);
890 } else {
891 /*
892 * It is the newest, move it to the tail,
893 * if it isn't there already. Next search will
894 * locate it quicker.
895 */
896 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
897 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
898 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
899 }
900 }
901 } else /* A new flow entry. */
902 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags);
903
904 mtx_unlock(&hsh->mtx);
905
906 return (error);
907 }
908 #endif
909
910 /*
911 * Return records from cache to userland.
912 *
913 * TODO: matching particular IP should be done in kernel, here.
914 */
915 int
ng_netflow_flow_show(priv_p priv,struct ngnf_show_header * req,struct ngnf_show_header * resp)916 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req,
917 struct ngnf_show_header *resp)
918 {
919 struct flow_hash_entry *hsh;
920 struct flow_entry *fle;
921 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1);
922 #ifdef INET6
923 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1);
924 #endif
925 int i, max;
926
927 i = req->hash_id;
928 if (i > NBUCKETS-1)
929 return (EINVAL);
930
931 #ifdef INET6
932 if (req->version == 6) {
933 resp->version = 6;
934 hsh = priv->hash6 + i;
935 max = NREC6_AT_ONCE;
936 } else
937 #endif
938 if (req->version == 4) {
939 resp->version = 4;
940 hsh = priv->hash + i;
941 max = NREC_AT_ONCE;
942 } else
943 return (EINVAL);
944
945 /*
946 * We will transfer not more than NREC_AT_ONCE. More data
947 * will come in next message.
948 * We send current hash index and current record number in list
949 * to userland, and userland should return it back to us.
950 * Then, we will restart with new entry.
951 *
952 * The resulting cache snapshot can be inaccurate if flow expiration
953 * is taking place on hash item between userland data requests for
954 * this hash item id.
955 */
956 resp->nentries = 0;
957 for (; i < NBUCKETS; hsh++, i++) {
958 int list_id;
959
960 if (mtx_trylock(&hsh->mtx) == 0) {
961 /*
962 * Requested hash index is not available,
963 * relay decision to skip or re-request data
964 * to userland.
965 */
966 resp->hash_id = i;
967 resp->list_id = 0;
968 return (0);
969 }
970
971 list_id = 0;
972 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
973 if (hsh->mtx.mtx_lock & MTX_CONTESTED) {
974 resp->hash_id = i;
975 resp->list_id = list_id;
976 mtx_unlock(&hsh->mtx);
977 return (0);
978 }
979
980 list_id++;
981 /* Search for particular record in list. */
982 if (req->list_id > 0) {
983 if (list_id < req->list_id)
984 continue;
985
986 /* Requested list position found. */
987 req->list_id = 0;
988 }
989 #ifdef INET6
990 if (req->version == 6) {
991 struct flow6_entry *fle6;
992
993 fle6 = (struct flow6_entry *)fle;
994 bcopy(&fle6->f, data6 + resp->nentries,
995 sizeof(fle6->f));
996 } else
997 #endif
998 bcopy(&fle->f, data + resp->nentries,
999 sizeof(fle->f));
1000 resp->nentries++;
1001 if (resp->nentries == max) {
1002 resp->hash_id = i;
1003 /*
1004 * If it was the last item in list
1005 * we simply skip to next hash_id.
1006 */
1007 resp->list_id = list_id + 1;
1008 mtx_unlock(&hsh->mtx);
1009 return (0);
1010 }
1011 }
1012 mtx_unlock(&hsh->mtx);
1013 }
1014
1015 resp->hash_id = resp->list_id = 0;
1016
1017 return (0);
1018 }
1019
1020 /* We have full datagram in privdata. Send it to export hook. */
1021 static int
export_send(priv_p priv,fib_export_p fe,item_p item,int flags)1022 export_send(priv_p priv, fib_export_p fe, item_p item, int flags)
1023 {
1024 struct mbuf *m = NGI_M(item);
1025 struct netflow_v5_export_dgram *dgram = mtod(m,
1026 struct netflow_v5_export_dgram *);
1027 struct netflow_v5_header *header = &dgram->header;
1028 struct timespec ts;
1029 int error = 0;
1030
1031 /* Fill mbuf header. */
1032 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
1033 header->count + sizeof(struct netflow_v5_header);
1034
1035 /* Fill export header. */
1036 header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
1037 getnanotime(&ts);
1038 header->unix_secs = htonl(ts.tv_sec);
1039 header->unix_nsecs = htonl(ts.tv_nsec);
1040 header->engine_type = 0;
1041 header->engine_id = fe->domain_id;
1042 header->pad = 0;
1043 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq,
1044 header->count));
1045 header->count = htons(header->count);
1046
1047 if (priv->export != NULL)
1048 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
1049 else
1050 NG_FREE_ITEM(item);
1051
1052 return (error);
1053 }
1054
1055
1056 /* Add export record to dgram. */
1057 static int
export_add(item_p item,struct flow_entry * fle)1058 export_add(item_p item, struct flow_entry *fle)
1059 {
1060 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
1061 struct netflow_v5_export_dgram *);
1062 struct netflow_v5_header *header = &dgram->header;
1063 struct netflow_v5_record *rec;
1064
1065 rec = &dgram->r[header->count];
1066 header->count ++;
1067
1068 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
1069 ("ng_netflow: export too big"));
1070
1071 /* Fill in export record. */
1072 rec->src_addr = fle->f.r.r_src.s_addr;
1073 rec->dst_addr = fle->f.r.r_dst.s_addr;
1074 rec->next_hop = fle->f.next_hop.s_addr;
1075 rec->i_ifx = htons(fle->f.fle_i_ifx);
1076 rec->o_ifx = htons(fle->f.fle_o_ifx);
1077 rec->packets = htonl(fle->f.packets);
1078 rec->octets = htonl(fle->f.bytes);
1079 rec->first = htonl(MILLIUPTIME(fle->f.first));
1080 rec->last = htonl(MILLIUPTIME(fle->f.last));
1081 rec->s_port = fle->f.r.r_sport;
1082 rec->d_port = fle->f.r.r_dport;
1083 rec->flags = fle->f.tcp_flags;
1084 rec->prot = fle->f.r.r_ip_p;
1085 rec->tos = fle->f.r.r_tos;
1086 rec->dst_mask = fle->f.dst_mask;
1087 rec->src_mask = fle->f.src_mask;
1088 rec->pad1 = 0;
1089 rec->pad2 = 0;
1090
1091 /* Not supported fields. */
1092 rec->src_as = rec->dst_as = 0;
1093
1094 if (header->count == NETFLOW_V5_MAX_RECORDS)
1095 return (1); /* end of datagram */
1096 else
1097 return (0);
1098 }
1099
1100 /* Periodic flow expiry run. */
1101 void
ng_netflow_expire(void * arg)1102 ng_netflow_expire(void *arg)
1103 {
1104 struct flow_entry *fle, *fle1;
1105 struct flow_hash_entry *hsh;
1106 priv_p priv = (priv_p )arg;
1107 int used, i;
1108
1109 /*
1110 * Going through all the cache.
1111 */
1112 used = uma_zone_get_cur(priv->zone);
1113 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
1114 /*
1115 * Skip entries, that are already being worked on.
1116 */
1117 if (mtx_trylock(&hsh->mtx) == 0)
1118 continue;
1119
1120 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1121 /*
1122 * Interrupt thread wants this entry!
1123 * Quick! Quick! Bail out!
1124 */
1125 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1126 break;
1127
1128 /*
1129 * Don't expire aggressively while hash collision
1130 * ratio is predicted small.
1131 */
1132 if (used <= (NBUCKETS*2) && !INACTIVE(fle))
1133 break;
1134
1135 if ((INACTIVE(fle) && (SMALL(fle) ||
1136 (used > (NBUCKETS*2)))) || AGED(fle)) {
1137 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1138 expire_flow(priv, priv_to_fib(priv,
1139 fle->f.r.fib), fle, NG_NOFLAGS);
1140 used--;
1141 counter_u64_add(priv->nfinfo_inact_exp, 1);
1142 }
1143 }
1144 mtx_unlock(&hsh->mtx);
1145 }
1146
1147 #ifdef INET6
1148 used = uma_zone_get_cur(priv->zone6);
1149 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) {
1150 struct flow6_entry *fle6;
1151
1152 /*
1153 * Skip entries, that are already being worked on.
1154 */
1155 if (mtx_trylock(&hsh->mtx) == 0)
1156 continue;
1157
1158 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1159 fle6 = (struct flow6_entry *)fle;
1160 /*
1161 * Interrupt thread wants this entry!
1162 * Quick! Quick! Bail out!
1163 */
1164 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1165 break;
1166
1167 /*
1168 * Don't expire aggressively while hash collision
1169 * ratio is predicted small.
1170 */
1171 if (used <= (NBUCKETS*2) && !INACTIVE(fle6))
1172 break;
1173
1174 if ((INACTIVE(fle6) && (SMALL(fle6) ||
1175 (used > (NBUCKETS*2)))) || AGED(fle6)) {
1176 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1177 expire_flow(priv, priv_to_fib(priv,
1178 fle->f.r.fib), fle, NG_NOFLAGS);
1179 used--;
1180 counter_u64_add(priv->nfinfo_inact_exp, 1);
1181 }
1182 }
1183 mtx_unlock(&hsh->mtx);
1184 }
1185 #endif
1186
1187 /* Schedule next expire. */
1188 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
1189 (void *)priv);
1190 }
1191