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_len == AF_INET)
364 fle->f.next_hop = nh->gw4_sa.sin_addr;
365 fle->f.dst_mask = plen;
366 }
367 }
368
369 /* Do route lookup on source address, to fill in src_mask. */
370 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
371 struct rtentry *rt;
372 struct route_nhop_data rnd;
373
374 rt = fib4_lookup_rt(r->fib, fle->f.r.r_src, 0, NHR_NONE, &rnd);
375 if (rt != NULL) {
376 struct in_addr addr;
377 uint32_t scopeid;
378 int plen;
379
380 rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid);
381 fle->f.src_mask = plen;
382 }
383 }
384
385 /* Push new flow at the and of hash. */
386 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
387
388 return (0);
389 }
390 #endif
391
392 #ifdef INET6
393 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)394 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r,
395 int plen, uint8_t flags, uint8_t tcp_flags)
396 {
397 struct flow6_entry *fle6;
398
399 mtx_assert(&hsh6->mtx, MA_OWNED);
400
401 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT);
402 if (fle6 == NULL) {
403 priv->nfinfo_alloc_failed++;
404 return (ENOMEM);
405 }
406
407 /*
408 * Now fle is totally ours. It is detached from all lists,
409 * we can safely edit it.
410 */
411
412 fle6->f.version = IP6VERSION;
413 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec));
414 fle6->f.bytes = plen;
415 fle6->f.packets = 1;
416 fle6->f.tcp_flags = tcp_flags;
417
418 fle6->f.first = fle6->f.last = time_uptime;
419
420 /*
421 * First we do route table lookup on destination address. So we can
422 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
423 */
424 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
425 struct rtentry *rt;
426 struct route_nhop_data rnd;
427
428 rt = fib6_lookup_rt(r->fib, &fle6->f.r.dst.r_dst6, 0, NHR_NONE, &rnd);
429 if (rt != NULL) {
430 struct in6_addr addr;
431 uint32_t scopeid;
432 struct nhop_object *nh = nhop_select_func(rnd.rnd_nhop, 0);
433 int plen;
434
435 rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
436 fle6->f.fle_o_ifx = nh->nh_ifp->if_index;
437 if (nh->gw_sa.sa_len == AF_INET6)
438 fle6->f.n.next_hop6 = nh->gw6_sa.sin6_addr;
439 fle6->f.dst_mask = plen;
440 }
441 }
442
443 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
444 /* Do route lookup on source address, to fill in src_mask. */
445 struct rtentry *rt;
446 struct route_nhop_data rnd;
447
448 rt = fib6_lookup_rt(r->fib, &fle6->f.r.src.r_src6, 0, NHR_NONE, &rnd);
449 if (rt != NULL) {
450 struct in6_addr addr;
451 uint32_t scopeid;
452 int plen;
453
454 rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
455 fle6->f.src_mask = plen;
456 }
457 }
458
459 /* Push new flow at the and of hash. */
460 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash);
461
462 return (0);
463 }
464 #endif
465
466 /*
467 * Non-static functions called from ng_netflow.c
468 */
469
470 /* Allocate memory and set up flow cache */
471 void
ng_netflow_cache_init(priv_p priv)472 ng_netflow_cache_init(priv_p priv)
473 {
474 struct flow_hash_entry *hsh;
475 int i;
476
477 /* Initialize cache UMA zone. */
478 priv->zone = uma_zcreate("NetFlow IPv4 cache",
479 sizeof(struct flow_entry), NULL, NULL, NULL, NULL,
480 UMA_ALIGN_CACHE, 0);
481 uma_zone_set_max(priv->zone, CACHESIZE);
482 #ifdef INET6
483 priv->zone6 = uma_zcreate("NetFlow IPv6 cache",
484 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL,
485 UMA_ALIGN_CACHE, 0);
486 uma_zone_set_max(priv->zone6, CACHESIZE);
487 #endif
488
489 /* Allocate hash. */
490 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
491 M_NETFLOW_HASH, M_WAITOK | M_ZERO);
492
493 /* Initialize hash. */
494 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
495 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
496 TAILQ_INIT(&hsh->head);
497 }
498
499 #ifdef INET6
500 /* Allocate hash. */
501 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
502 M_NETFLOW_HASH, M_WAITOK | M_ZERO);
503
504 /* Initialize hash. */
505 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) {
506 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
507 TAILQ_INIT(&hsh->head);
508 }
509 #endif
510
511 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK);
512 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK);
513 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK);
514 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK);
515 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK);
516 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK);
517 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK);
518 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK);
519 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK);
520 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK);
521
522 ng_netflow_v9_cache_init(priv);
523 CTR0(KTR_NET, "ng_netflow startup()");
524 }
525
526 /* Initialize new FIB table for v5 and v9 */
527 int
ng_netflow_fib_init(priv_p priv,int fib)528 ng_netflow_fib_init(priv_p priv, int fib)
529 {
530 fib_export_p fe = priv_to_fib(priv, fib);
531
532 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib);
533
534 if (fe != NULL)
535 return (0);
536
537 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH,
538 M_NOWAIT | M_ZERO)) == NULL)
539 return (ENOMEM);
540
541 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF);
542 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF);
543 fe->fib = fib;
544 fe->domain_id = fib;
545
546 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib],
547 (uintptr_t)NULL, (uintptr_t)fe) == 0) {
548 /* FIB already set up by other ISR */
549 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p",
550 fib, fe, priv_to_fib(priv, fib));
551 mtx_destroy(&fe->export_mtx);
552 mtx_destroy(&fe->export9_mtx);
553 free(fe, M_NETGRAPH);
554 } else {
555 /* Increase counter for statistics */
556 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)",
557 fib, fe, priv_to_fib(priv, fib));
558 priv->nfinfo_alloc_fibs++;
559 }
560
561 return (0);
562 }
563
564 /* Free all flow cache memory. Called from node close method. */
565 void
ng_netflow_cache_flush(priv_p priv)566 ng_netflow_cache_flush(priv_p priv)
567 {
568 struct flow_entry *fle, *fle1;
569 struct flow_hash_entry *hsh;
570 struct netflow_export_item exp;
571 fib_export_p fe;
572 int i;
573
574 bzero(&exp, sizeof(exp));
575
576 /*
577 * We are going to free probably billable data.
578 * Expire everything before freeing it.
579 * No locking is required since callout is already drained.
580 */
581 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
582 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
583 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
584 fe = priv_to_fib(priv, fle->f.r.fib);
585 expire_flow(priv, fe, fle, NG_QUEUE);
586 }
587 #ifdef INET6
588 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++)
589 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
590 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
591 fe = priv_to_fib(priv, fle->f.r.fib);
592 expire_flow(priv, fe, fle, NG_QUEUE);
593 }
594 #endif
595
596 uma_zdestroy(priv->zone);
597 /* Destroy hash mutexes. */
598 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
599 mtx_destroy(&hsh->mtx);
600
601 /* Free hash memory. */
602 if (priv->hash != NULL)
603 free(priv->hash, M_NETFLOW_HASH);
604 #ifdef INET6
605 uma_zdestroy(priv->zone6);
606 /* Destroy hash mutexes. */
607 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++)
608 mtx_destroy(&hsh->mtx);
609
610 /* Free hash memory. */
611 if (priv->hash6 != NULL)
612 free(priv->hash6, M_NETFLOW_HASH);
613 #endif
614
615 for (i = 0; i < priv->maxfibs; i++) {
616 if ((fe = priv_to_fib(priv, i)) == NULL)
617 continue;
618
619 if (fe->exp.item != NULL)
620 export_send(priv, fe, fe->exp.item, NG_QUEUE);
621
622 if (fe->exp.item9 != NULL)
623 export9_send(priv, fe, fe->exp.item9,
624 fe->exp.item9_opt, NG_QUEUE);
625
626 mtx_destroy(&fe->export_mtx);
627 mtx_destroy(&fe->export9_mtx);
628 free(fe, M_NETGRAPH);
629 }
630
631 counter_u64_free(priv->nfinfo_bytes);
632 counter_u64_free(priv->nfinfo_packets);
633 counter_u64_free(priv->nfinfo_bytes6);
634 counter_u64_free(priv->nfinfo_packets6);
635 counter_u64_free(priv->nfinfo_sbytes);
636 counter_u64_free(priv->nfinfo_spackets);
637 counter_u64_free(priv->nfinfo_sbytes6);
638 counter_u64_free(priv->nfinfo_spackets6);
639 counter_u64_free(priv->nfinfo_act_exp);
640 counter_u64_free(priv->nfinfo_inact_exp);
641
642 ng_netflow_v9_cache_flush(priv);
643 }
644
645 #ifdef INET
646 /* Insert packet from into flow cache. */
647 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)648 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip,
649 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
650 unsigned int src_if_index)
651 {
652 struct flow_entry *fle, *fle1;
653 struct flow_hash_entry *hsh;
654 struct flow_rec r;
655 int hlen, plen;
656 int error = 0;
657 uint16_t eproto;
658 uint8_t tcp_flags = 0;
659
660 bzero(&r, sizeof(r));
661
662 if (ip->ip_v != IPVERSION)
663 return (EINVAL);
664
665 hlen = ip->ip_hl << 2;
666 if (hlen < sizeof(struct ip))
667 return (EINVAL);
668
669 eproto = ETHERTYPE_IP;
670 /* Assume L4 template by default */
671 r.flow_type = NETFLOW_V9_FLOW_V4_L4;
672
673 r.r_src = ip->ip_src;
674 r.r_dst = ip->ip_dst;
675 r.fib = fe->fib;
676
677 plen = ntohs(ip->ip_len);
678
679 r.r_ip_p = ip->ip_p;
680 r.r_tos = ip->ip_tos;
681
682 r.r_i_ifx = src_if_index;
683
684 /*
685 * XXX NOTE: only first fragment of fragmented TCP, UDP and
686 * ICMP packet will be recorded with proper s_port and d_port.
687 * Following fragments will be recorded simply as IP packet with
688 * ip_proto = ip->ip_p and s_port, d_port set to zero.
689 * I know, it looks like bug. But I don't want to re-implement
690 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
691 * and nobody complains yet :)
692 */
693 if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
694 switch(r.r_ip_p) {
695 case IPPROTO_TCP:
696 {
697 struct tcphdr *tcp;
698
699 tcp = (struct tcphdr *)((caddr_t )ip + hlen);
700 r.r_sport = tcp->th_sport;
701 r.r_dport = tcp->th_dport;
702 tcp_flags = tcp->th_flags;
703 break;
704 }
705 case IPPROTO_UDP:
706 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
707 break;
708 }
709
710 counter_u64_add(priv->nfinfo_packets, 1);
711 counter_u64_add(priv->nfinfo_bytes, plen);
712
713 /* Find hash slot. */
714 hsh = &priv->hash[ip_hash(&r)];
715
716 mtx_lock(&hsh->mtx);
717
718 /*
719 * Go through hash and find our entry. If we encounter an
720 * entry, that should be expired, purge it. We do a reverse
721 * search since most active entries are first, and most
722 * searches are done on most active entries.
723 */
724 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
725 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
726 break;
727 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
728 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
729 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
730 fle, NG_QUEUE);
731 counter_u64_add(priv->nfinfo_act_exp, 1);
732 }
733 }
734
735 if (fle) { /* An existent entry. */
736
737 fle->f.bytes += plen;
738 fle->f.packets ++;
739 fle->f.tcp_flags |= tcp_flags;
740 fle->f.last = time_uptime;
741
742 /*
743 * We have the following reasons to expire flow in active way:
744 * - it hit active timeout
745 * - a TCP connection closed
746 * - it is going to overflow counter
747 */
748 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
749 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
750 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
751 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
752 fle, NG_QUEUE);
753 counter_u64_add(priv->nfinfo_act_exp, 1);
754 } else {
755 /*
756 * It is the newest, move it to the tail,
757 * if it isn't there already. Next search will
758 * locate it quicker.
759 */
760 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
761 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
762 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
763 }
764 }
765 } else /* A new flow entry. */
766 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags);
767
768 mtx_unlock(&hsh->mtx);
769
770 return (error);
771 }
772 #endif
773
774 #ifdef INET6
775 /* Insert IPv6 packet from into flow cache. */
776 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)777 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6,
778 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
779 unsigned int src_if_index)
780 {
781 struct flow_entry *fle = NULL, *fle1;
782 struct flow6_entry *fle6;
783 struct flow_hash_entry *hsh;
784 struct flow6_rec r;
785 int plen;
786 int error = 0;
787 uint8_t tcp_flags = 0;
788
789 /* check version */
790 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
791 return (EINVAL);
792
793 bzero(&r, sizeof(r));
794
795 r.src.r_src6 = ip6->ip6_src;
796 r.dst.r_dst6 = ip6->ip6_dst;
797 r.fib = fe->fib;
798
799 /* Assume L4 template by default */
800 r.flow_type = NETFLOW_V9_FLOW_V6_L4;
801
802 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
803
804 #if 0
805 /* XXX: set DSCP/CoS value */
806 r.r_tos = ip->ip_tos;
807 #endif
808 if ((flags & NG_NETFLOW_IS_FRAG) == 0) {
809 switch(upper_proto) {
810 case IPPROTO_TCP:
811 {
812 struct tcphdr *tcp;
813
814 tcp = (struct tcphdr *)upper_ptr;
815 r.r_ports = *(uint32_t *)upper_ptr;
816 tcp_flags = tcp->th_flags;
817 break;
818 }
819 case IPPROTO_UDP:
820 case IPPROTO_SCTP:
821 r.r_ports = *(uint32_t *)upper_ptr;
822 break;
823 }
824 }
825
826 r.r_ip_p = upper_proto;
827 r.r_i_ifx = src_if_index;
828
829 counter_u64_add(priv->nfinfo_packets6, 1);
830 counter_u64_add(priv->nfinfo_bytes6, plen);
831
832 /* Find hash slot. */
833 hsh = &priv->hash6[ip6_hash(&r)];
834
835 mtx_lock(&hsh->mtx);
836
837 /*
838 * Go through hash and find our entry. If we encounter an
839 * entry, that should be expired, purge it. We do a reverse
840 * search since most active entries are first, and most
841 * searches are done on most active entries.
842 */
843 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
844 if (fle->f.version != IP6VERSION)
845 continue;
846 fle6 = (struct flow6_entry *)fle;
847 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0)
848 break;
849 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) {
850 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
851 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
852 NG_QUEUE);
853 counter_u64_add(priv->nfinfo_act_exp, 1);
854 }
855 }
856
857 if (fle != NULL) { /* An existent entry. */
858 fle6 = (struct flow6_entry *)fle;
859
860 fle6->f.bytes += plen;
861 fle6->f.packets ++;
862 fle6->f.tcp_flags |= tcp_flags;
863 fle6->f.last = time_uptime;
864
865 /*
866 * We have the following reasons to expire flow in active way:
867 * - it hit active timeout
868 * - a TCP connection closed
869 * - it is going to overflow counter
870 */
871 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) ||
872 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
873 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
874 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
875 NG_QUEUE);
876 counter_u64_add(priv->nfinfo_act_exp, 1);
877 } else {
878 /*
879 * It is the newest, move it to the tail,
880 * if it isn't there already. Next search will
881 * locate it quicker.
882 */
883 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
884 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
885 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
886 }
887 }
888 } else /* A new flow entry. */
889 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags);
890
891 mtx_unlock(&hsh->mtx);
892
893 return (error);
894 }
895 #endif
896
897 /*
898 * Return records from cache to userland.
899 *
900 * TODO: matching particular IP should be done in kernel, here.
901 */
902 int
ng_netflow_flow_show(priv_p priv,struct ngnf_show_header * req,struct ngnf_show_header * resp)903 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req,
904 struct ngnf_show_header *resp)
905 {
906 struct flow_hash_entry *hsh;
907 struct flow_entry *fle;
908 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1);
909 #ifdef INET6
910 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1);
911 #endif
912 int i, max;
913
914 i = req->hash_id;
915 if (i > NBUCKETS-1)
916 return (EINVAL);
917
918 #ifdef INET6
919 if (req->version == 6) {
920 resp->version = 6;
921 hsh = priv->hash6 + i;
922 max = NREC6_AT_ONCE;
923 } else
924 #endif
925 if (req->version == 4) {
926 resp->version = 4;
927 hsh = priv->hash + i;
928 max = NREC_AT_ONCE;
929 } else
930 return (EINVAL);
931
932 /*
933 * We will transfer not more than NREC_AT_ONCE. More data
934 * will come in next message.
935 * We send current hash index and current record number in list
936 * to userland, and userland should return it back to us.
937 * Then, we will restart with new entry.
938 *
939 * The resulting cache snapshot can be inaccurate if flow expiration
940 * is taking place on hash item between userland data requests for
941 * this hash item id.
942 */
943 resp->nentries = 0;
944 for (; i < NBUCKETS; hsh++, i++) {
945 int list_id;
946
947 if (mtx_trylock(&hsh->mtx) == 0) {
948 /*
949 * Requested hash index is not available,
950 * relay decision to skip or re-request data
951 * to userland.
952 */
953 resp->hash_id = i;
954 resp->list_id = 0;
955 return (0);
956 }
957
958 list_id = 0;
959 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
960 if (hsh->mtx.mtx_lock & MTX_CONTESTED) {
961 resp->hash_id = i;
962 resp->list_id = list_id;
963 mtx_unlock(&hsh->mtx);
964 return (0);
965 }
966
967 list_id++;
968 /* Search for particular record in list. */
969 if (req->list_id > 0) {
970 if (list_id < req->list_id)
971 continue;
972
973 /* Requested list position found. */
974 req->list_id = 0;
975 }
976 #ifdef INET6
977 if (req->version == 6) {
978 struct flow6_entry *fle6;
979
980 fle6 = (struct flow6_entry *)fle;
981 bcopy(&fle6->f, data6 + resp->nentries,
982 sizeof(fle6->f));
983 } else
984 #endif
985 bcopy(&fle->f, data + resp->nentries,
986 sizeof(fle->f));
987 resp->nentries++;
988 if (resp->nentries == max) {
989 resp->hash_id = i;
990 /*
991 * If it was the last item in list
992 * we simply skip to next hash_id.
993 */
994 resp->list_id = list_id + 1;
995 mtx_unlock(&hsh->mtx);
996 return (0);
997 }
998 }
999 mtx_unlock(&hsh->mtx);
1000 }
1001
1002 resp->hash_id = resp->list_id = 0;
1003
1004 return (0);
1005 }
1006
1007 /* We have full datagram in privdata. Send it to export hook. */
1008 static int
export_send(priv_p priv,fib_export_p fe,item_p item,int flags)1009 export_send(priv_p priv, fib_export_p fe, item_p item, int flags)
1010 {
1011 struct mbuf *m = NGI_M(item);
1012 struct netflow_v5_export_dgram *dgram = mtod(m,
1013 struct netflow_v5_export_dgram *);
1014 struct netflow_v5_header *header = &dgram->header;
1015 struct timespec ts;
1016 int error = 0;
1017
1018 /* Fill mbuf header. */
1019 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
1020 header->count + sizeof(struct netflow_v5_header);
1021
1022 /* Fill export header. */
1023 header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
1024 getnanotime(&ts);
1025 header->unix_secs = htonl(ts.tv_sec);
1026 header->unix_nsecs = htonl(ts.tv_nsec);
1027 header->engine_type = 0;
1028 header->engine_id = fe->domain_id;
1029 header->pad = 0;
1030 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq,
1031 header->count));
1032 header->count = htons(header->count);
1033
1034 if (priv->export != NULL)
1035 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
1036 else
1037 NG_FREE_ITEM(item);
1038
1039 return (error);
1040 }
1041
1042 /* Add export record to dgram. */
1043 static int
export_add(item_p item,struct flow_entry * fle)1044 export_add(item_p item, struct flow_entry *fle)
1045 {
1046 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
1047 struct netflow_v5_export_dgram *);
1048 struct netflow_v5_header *header = &dgram->header;
1049 struct netflow_v5_record *rec;
1050
1051 rec = &dgram->r[header->count];
1052 header->count ++;
1053
1054 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
1055 ("ng_netflow: export too big"));
1056
1057 /* Fill in export record. */
1058 rec->src_addr = fle->f.r.r_src.s_addr;
1059 rec->dst_addr = fle->f.r.r_dst.s_addr;
1060 rec->next_hop = fle->f.next_hop.s_addr;
1061 rec->i_ifx = htons(fle->f.fle_i_ifx);
1062 rec->o_ifx = htons(fle->f.fle_o_ifx);
1063 rec->packets = htonl(fle->f.packets);
1064 rec->octets = htonl(fle->f.bytes);
1065 rec->first = htonl(MILLIUPTIME(fle->f.first));
1066 rec->last = htonl(MILLIUPTIME(fle->f.last));
1067 rec->s_port = fle->f.r.r_sport;
1068 rec->d_port = fle->f.r.r_dport;
1069 rec->flags = fle->f.tcp_flags;
1070 rec->prot = fle->f.r.r_ip_p;
1071 rec->tos = fle->f.r.r_tos;
1072 rec->dst_mask = fle->f.dst_mask;
1073 rec->src_mask = fle->f.src_mask;
1074 rec->pad1 = 0;
1075 rec->pad2 = 0;
1076
1077 /* Not supported fields. */
1078 rec->src_as = rec->dst_as = 0;
1079
1080 if (header->count == NETFLOW_V5_MAX_RECORDS)
1081 return (1); /* end of datagram */
1082 else
1083 return (0);
1084 }
1085
1086 /* Periodic flow expiry run. */
1087 void
ng_netflow_expire(void * arg)1088 ng_netflow_expire(void *arg)
1089 {
1090 struct flow_entry *fle, *fle1;
1091 struct flow_hash_entry *hsh;
1092 priv_p priv = (priv_p )arg;
1093 int used, i;
1094
1095 /*
1096 * Going through all the cache.
1097 */
1098 used = uma_zone_get_cur(priv->zone);
1099 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
1100 /*
1101 * Skip entries, that are already being worked on.
1102 */
1103 if (mtx_trylock(&hsh->mtx) == 0)
1104 continue;
1105
1106 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1107 /*
1108 * Interrupt thread wants this entry!
1109 * Quick! Quick! Bail out!
1110 */
1111 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1112 break;
1113
1114 /*
1115 * Don't expire aggressively while hash collision
1116 * ratio is predicted small.
1117 */
1118 if (used <= (NBUCKETS*2) && !INACTIVE(fle))
1119 break;
1120
1121 if ((INACTIVE(fle) && (SMALL(fle) ||
1122 (used > (NBUCKETS*2)))) || AGED(fle)) {
1123 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1124 expire_flow(priv, priv_to_fib(priv,
1125 fle->f.r.fib), fle, NG_NOFLAGS);
1126 used--;
1127 counter_u64_add(priv->nfinfo_inact_exp, 1);
1128 }
1129 }
1130 mtx_unlock(&hsh->mtx);
1131 }
1132
1133 #ifdef INET6
1134 used = uma_zone_get_cur(priv->zone6);
1135 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) {
1136 struct flow6_entry *fle6;
1137
1138 /*
1139 * Skip entries, that are already being worked on.
1140 */
1141 if (mtx_trylock(&hsh->mtx) == 0)
1142 continue;
1143
1144 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1145 fle6 = (struct flow6_entry *)fle;
1146 /*
1147 * Interrupt thread wants this entry!
1148 * Quick! Quick! Bail out!
1149 */
1150 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1151 break;
1152
1153 /*
1154 * Don't expire aggressively while hash collision
1155 * ratio is predicted small.
1156 */
1157 if (used <= (NBUCKETS*2) && !INACTIVE(fle6))
1158 break;
1159
1160 if ((INACTIVE(fle6) && (SMALL(fle6) ||
1161 (used > (NBUCKETS*2)))) || AGED(fle6)) {
1162 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1163 expire_flow(priv, priv_to_fib(priv,
1164 fle->f.r.fib), fle, NG_NOFLAGS);
1165 used--;
1166 counter_u64_add(priv->nfinfo_inact_exp, 1);
1167 }
1168 }
1169 mtx_unlock(&hsh->mtx);
1170 }
1171 #endif
1172
1173 /* Schedule next expire. */
1174 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
1175 (void *)priv);
1176 }
1177