1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2010-2011 Alexander V. Chernikov <melifaro@ipfw.ru>
5 * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
6 * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 *
30 * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
31 */
32
33 #include <sys/cdefs.h>
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_route.h"
37 #include <sys/param.h>
38 #include <sys/bitstring.h>
39 #include <sys/systm.h>
40 #include <sys/counter.h>
41 #include <sys/kernel.h>
42 #include <sys/ktr.h>
43 #include <sys/limits.h>
44 #include <sys/mbuf.h>
45 #include <sys/syslog.h>
46 #include <sys/socket.h>
47 #include <vm/uma.h>
48
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/if_var.h>
52 #include <net/if_private.h>
53 #include <net/route.h>
54 #include <net/route/nhop.h>
55 #include <net/route/route_ctl.h>
56 #include <net/ethernet.h>
57 #include <netinet/in.h>
58 #include <netinet/in_fib.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip6.h>
62 #include <netinet/tcp.h>
63 #include <netinet/udp.h>
64
65 #include <netinet6/in6_fib.h>
66
67 #include <netgraph/ng_message.h>
68 #include <netgraph/netgraph.h>
69
70 #include <netgraph/netflow/netflow.h>
71 #include <netgraph/netflow/netflow_v9.h>
72 #include <netgraph/netflow/ng_netflow.h>
73
74 #define NBUCKETS (65536) /* must be power of 2 */
75
76 /* This hash is for TCP or UDP packets. */
77 #define FULL_HASH(addr1, addr2, port1, port2) \
78 (((addr1 ^ (addr1 >> 16) ^ \
79 htons(addr2 ^ (addr2 >> 16))) ^ \
80 port1 ^ htons(port2)) & \
81 (NBUCKETS - 1))
82
83 /* This hash is for all other IP packets. */
84 #define ADDR_HASH(addr1, addr2) \
85 ((addr1 ^ (addr1 >> 16) ^ \
86 htons(addr2 ^ (addr2 >> 16))) & \
87 (NBUCKETS - 1))
88
89 /* Macros to shorten logical constructions */
90 /* XXX: priv must exist in namespace */
91 #define INACTIVE(fle) (time_uptime - fle->f.last > priv->nfinfo_inact_t)
92 #define AGED(fle) (time_uptime - fle->f.first > priv->nfinfo_act_t)
93 #define ISFREE(fle) (fle->f.packets == 0)
94
95 /*
96 * 4 is a magical number: statistically number of 4-packet flows is
97 * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
98 * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
99 * of reachable host and 4-packet otherwise.
100 */
101 #define SMALL(fle) (fle->f.packets <= 4)
102
103 MALLOC_DEFINE(M_NETFLOW_HASH, "netflow_hash", "NetFlow hash");
104
105 static int export_add(item_p, struct flow_entry *);
106 static int export_send(priv_p, fib_export_p, item_p, int);
107
108 #ifdef INET
109 static int hash_insert(priv_p, struct flow_hash_entry *, struct flow_rec *,
110 int, uint8_t, uint8_t);
111 #endif
112 #ifdef INET6
113 static int hash6_insert(priv_p, struct flow_hash_entry *, struct flow6_rec *,
114 int, uint8_t, uint8_t);
115 #endif
116
117 static void expire_flow(priv_p, fib_export_p, struct flow_entry *, int);
118
119 #ifdef INET
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 #endif
144
145 #ifdef INET6
146 /* Generate hash for a given flow6 record. Use lower 4 octets from v6 addresses */
147 static inline uint32_t
ip6_hash(struct flow6_rec * r)148 ip6_hash(struct flow6_rec *r)
149 {
150
151 switch (r->r_ip_p) {
152 case IPPROTO_TCP:
153 case IPPROTO_UDP:
154 return FULL_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
155 r->dst.r_dst6.__u6_addr.__u6_addr32[3], r->r_sport,
156 r->r_dport);
157 default:
158 return ADDR_HASH(r->src.r_src6.__u6_addr.__u6_addr32[3],
159 r->dst.r_dst6.__u6_addr.__u6_addr32[3]);
160 }
161 }
162
163 #endif
164
165 /*
166 * Detach export datagram from priv, if there is any.
167 * If there is no, allocate a new one.
168 */
169 static item_p
get_export_dgram(priv_p priv,fib_export_p fe)170 get_export_dgram(priv_p priv, fib_export_p fe)
171 {
172 item_p item = NULL;
173
174 mtx_lock(&fe->export_mtx);
175 if (fe->exp.item != NULL) {
176 item = fe->exp.item;
177 fe->exp.item = NULL;
178 }
179 mtx_unlock(&fe->export_mtx);
180
181 if (item == NULL) {
182 struct netflow_v5_export_dgram *dgram;
183 struct mbuf *m;
184
185 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
186 if (m == NULL)
187 return (NULL);
188 item = ng_package_data(m, NG_NOFLAGS);
189 if (item == NULL)
190 return (NULL);
191 dgram = mtod(m, struct netflow_v5_export_dgram *);
192 dgram->header.count = 0;
193 dgram->header.version = htons(NETFLOW_V5);
194 dgram->header.pad = 0;
195 }
196
197 return (item);
198 }
199
200 /*
201 * Re-attach incomplete datagram back to priv.
202 * If there is already another one, then send incomplete. */
203 static void
return_export_dgram(priv_p priv,fib_export_p fe,item_p item,int flags)204 return_export_dgram(priv_p priv, fib_export_p fe, item_p item, int flags)
205 {
206
207 /*
208 * It may happen on SMP, that some thread has already
209 * put its item there, in this case we bail out and
210 * send what we have to collector.
211 */
212 mtx_lock(&fe->export_mtx);
213 if (fe->exp.item == NULL) {
214 fe->exp.item = item;
215 mtx_unlock(&fe->export_mtx);
216 } else {
217 mtx_unlock(&fe->export_mtx);
218 export_send(priv, fe, item, flags);
219 }
220 }
221
222 /*
223 * The flow is over. Call export_add() and free it. If datagram is
224 * full, then call export_send().
225 */
226 static void
expire_flow(priv_p priv,fib_export_p fe,struct flow_entry * fle,int flags)227 expire_flow(priv_p priv, fib_export_p fe, struct flow_entry *fle, int flags)
228 {
229 struct netflow_export_item exp;
230 uint16_t version = fle->f.version;
231
232 if ((priv->export != NULL) && (version == IPVERSION)) {
233 exp.item = get_export_dgram(priv, fe);
234 if (exp.item == NULL) {
235 priv->nfinfo_export_failed++;
236 if (priv->export9 != NULL)
237 priv->nfinfo_export9_failed++;
238 /* fle definitely contains IPv4 flow. */
239 uma_zfree_arg(priv->zone, fle, priv);
240 return;
241 }
242
243 if (export_add(exp.item, fle) > 0)
244 export_send(priv, fe, exp.item, flags);
245 else
246 return_export_dgram(priv, fe, exp.item, NG_QUEUE);
247 }
248
249 if (priv->export9 != NULL) {
250 exp.item9 = get_export9_dgram(priv, fe, &exp.item9_opt);
251 if (exp.item9 == NULL) {
252 priv->nfinfo_export9_failed++;
253 if (version == IPVERSION)
254 uma_zfree_arg(priv->zone, fle, priv);
255 #ifdef INET6
256 else if (version == IP6VERSION)
257 uma_zfree_arg(priv->zone6, fle, priv);
258 #endif
259 else
260 panic("ng_netflow: Unknown IP proto: %d",
261 version);
262 return;
263 }
264
265 if (export9_add(exp.item9, exp.item9_opt, fle) > 0)
266 export9_send(priv, fe, exp.item9, exp.item9_opt, flags);
267 else
268 return_export9_dgram(priv, fe, exp.item9,
269 exp.item9_opt, NG_QUEUE);
270 }
271
272 if (version == IPVERSION)
273 uma_zfree_arg(priv->zone, fle, priv);
274 #ifdef INET6
275 else if (version == IP6VERSION)
276 uma_zfree_arg(priv->zone6, fle, priv);
277 #endif
278 }
279
280 /* Get a snapshot of node statistics */
281 void
ng_netflow_copyinfo(priv_p priv,struct ng_netflow_info * i)282 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
283 {
284
285 i->nfinfo_bytes = counter_u64_fetch(priv->nfinfo_bytes);
286 i->nfinfo_packets = counter_u64_fetch(priv->nfinfo_packets);
287 i->nfinfo_bytes6 = counter_u64_fetch(priv->nfinfo_bytes6);
288 i->nfinfo_packets6 = counter_u64_fetch(priv->nfinfo_packets6);
289 i->nfinfo_sbytes = counter_u64_fetch(priv->nfinfo_sbytes);
290 i->nfinfo_spackets = counter_u64_fetch(priv->nfinfo_spackets);
291 i->nfinfo_sbytes6 = counter_u64_fetch(priv->nfinfo_sbytes6);
292 i->nfinfo_spackets6 = counter_u64_fetch(priv->nfinfo_spackets6);
293 i->nfinfo_act_exp = counter_u64_fetch(priv->nfinfo_act_exp);
294 i->nfinfo_inact_exp = counter_u64_fetch(priv->nfinfo_inact_exp);
295
296 i->nfinfo_used = uma_zone_get_cur(priv->zone);
297 #ifdef INET6
298 i->nfinfo_used6 = uma_zone_get_cur(priv->zone6);
299 #endif
300
301 i->nfinfo_alloc_failed = priv->nfinfo_alloc_failed;
302 i->nfinfo_export_failed = priv->nfinfo_export_failed;
303 i->nfinfo_export9_failed = priv->nfinfo_export9_failed;
304 i->nfinfo_realloc_mbuf = priv->nfinfo_realloc_mbuf;
305 i->nfinfo_alloc_fibs = priv->nfinfo_alloc_fibs;
306 i->nfinfo_inact_t = priv->nfinfo_inact_t;
307 i->nfinfo_act_t = priv->nfinfo_act_t;
308 }
309
310 /*
311 * Insert a record into defined slot.
312 *
313 * First we get for us a free flow entry, then fill in all
314 * possible fields in it.
315 *
316 * TODO: consider dropping hash mutex while filling in datagram,
317 * as this was done in previous version. Need to test & profile
318 * to be sure.
319 */
320 #ifdef INET
321 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)322 hash_insert(priv_p priv, struct flow_hash_entry *hsh, struct flow_rec *r,
323 int plen, uint8_t flags, uint8_t tcp_flags)
324 {
325 struct flow_entry *fle;
326
327 mtx_assert(&hsh->mtx, MA_OWNED);
328
329 fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT);
330 if (fle == NULL) {
331 priv->nfinfo_alloc_failed++;
332 return (ENOMEM);
333 }
334
335 /*
336 * Now fle is totally ours. It is detached from all lists,
337 * we can safely edit it.
338 */
339 fle->f.version = IPVERSION;
340 bcopy(r, &fle->f.r, sizeof(struct flow_rec));
341 fle->f.bytes = plen;
342 fle->f.packets = 1;
343 fle->f.tcp_flags = tcp_flags;
344
345 fle->f.first = fle->f.last = time_uptime;
346
347 /*
348 * First we do route table lookup on destination address. So we can
349 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
350 */
351 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
352 struct rtentry *rt;
353 struct route_nhop_data rnd;
354
355 rt = fib4_lookup_rt(r->fib, fle->f.r.r_dst, 0, NHR_NONE, &rnd);
356 if (rt != NULL) {
357 struct in_addr addr;
358 uint32_t scopeid;
359 struct nhop_object *nh = nhop_select_func(rnd.rnd_nhop, 0);
360 int plen;
361
362 rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid);
363 fle->f.fle_o_ifx = nh->nh_ifp->if_index;
364 if (nh->gw_sa.sa_family == AF_INET)
365 fle->f.next_hop = nh->gw4_sa.sin_addr;
366 /*
367 * XXX we're leaving an empty gateway here for
368 * IPv6 nexthops.
369 */
370 fle->f.dst_mask = plen;
371 }
372 }
373
374 /* Do route lookup on source address, to fill in src_mask. */
375 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
376 struct rtentry *rt;
377 struct route_nhop_data rnd;
378
379 rt = fib4_lookup_rt(r->fib, fle->f.r.r_src, 0, NHR_NONE, &rnd);
380 if (rt != NULL) {
381 struct in_addr addr;
382 uint32_t scopeid;
383 int plen;
384
385 rt_get_inet_prefix_plen(rt, &addr, &plen, &scopeid);
386 fle->f.src_mask = plen;
387 }
388 }
389
390 /* Push new flow at the and of hash. */
391 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
392
393 return (0);
394 }
395 #endif
396
397 #ifdef INET6
398 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)399 hash6_insert(priv_p priv, struct flow_hash_entry *hsh6, struct flow6_rec *r,
400 int plen, uint8_t flags, uint8_t tcp_flags)
401 {
402 struct flow6_entry *fle6;
403
404 mtx_assert(&hsh6->mtx, MA_OWNED);
405
406 fle6 = uma_zalloc_arg(priv->zone6, priv, M_NOWAIT);
407 if (fle6 == NULL) {
408 priv->nfinfo_alloc_failed++;
409 return (ENOMEM);
410 }
411
412 /*
413 * Now fle is totally ours. It is detached from all lists,
414 * we can safely edit it.
415 */
416
417 fle6->f.version = IP6VERSION;
418 bcopy(r, &fle6->f.r, sizeof(struct flow6_rec));
419 fle6->f.bytes = plen;
420 fle6->f.packets = 1;
421 fle6->f.tcp_flags = tcp_flags;
422
423 fle6->f.first = fle6->f.last = time_uptime;
424
425 /*
426 * First we do route table lookup on destination address. So we can
427 * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
428 */
429 if ((flags & NG_NETFLOW_CONF_NODSTLOOKUP) == 0) {
430 struct rtentry *rt;
431 struct route_nhop_data rnd;
432
433 rt = fib6_lookup_rt(r->fib, &fle6->f.r.dst.r_dst6, 0, NHR_NONE, &rnd);
434 if (rt != NULL) {
435 struct in6_addr addr;
436 uint32_t scopeid;
437 struct nhop_object *nh = nhop_select_func(rnd.rnd_nhop, 0);
438 int plen;
439
440 rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
441 fle6->f.fle_o_ifx = nh->nh_ifp->if_index;
442 if (nh->gw_sa.sa_family == AF_INET6)
443 fle6->f.n.next_hop6 = nh->gw6_sa.sin6_addr;
444 fle6->f.dst_mask = plen;
445 }
446 }
447
448 if ((flags & NG_NETFLOW_CONF_NOSRCLOOKUP) == 0) {
449 /* Do route lookup on source address, to fill in src_mask. */
450 struct rtentry *rt;
451 struct route_nhop_data rnd;
452
453 rt = fib6_lookup_rt(r->fib, &fle6->f.r.src.r_src6, 0, NHR_NONE, &rnd);
454 if (rt != NULL) {
455 struct in6_addr addr;
456 uint32_t scopeid;
457 int plen;
458
459 rt_get_inet6_prefix_plen(rt, &addr, &plen, &scopeid);
460 fle6->f.src_mask = plen;
461 }
462 }
463
464 /* Push new flow at the and of hash. */
465 TAILQ_INSERT_TAIL(&hsh6->head, (struct flow_entry *)fle6, fle_hash);
466
467 return (0);
468 }
469 #endif
470
471 /*
472 * Non-static functions called from ng_netflow.c
473 */
474
475 /* Allocate memory and set up flow cache */
476 void
ng_netflow_cache_init(priv_p priv)477 ng_netflow_cache_init(priv_p priv)
478 {
479 struct flow_hash_entry *hsh;
480 int i;
481
482 /* Initialize cache UMA zone. */
483 priv->zone = uma_zcreate("NetFlow IPv4 cache",
484 sizeof(struct flow_entry), NULL, NULL, NULL, NULL,
485 UMA_ALIGN_CACHE, 0);
486 uma_zone_set_max(priv->zone, CACHESIZE);
487 #ifdef INET6
488 priv->zone6 = uma_zcreate("NetFlow IPv6 cache",
489 sizeof(struct flow6_entry), NULL, NULL, NULL, NULL,
490 UMA_ALIGN_CACHE, 0);
491 uma_zone_set_max(priv->zone6, CACHESIZE);
492 #endif
493
494 /* Allocate hash. */
495 priv->hash = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
496 M_NETFLOW_HASH, M_WAITOK | M_ZERO);
497
498 /* Initialize hash. */
499 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
500 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
501 TAILQ_INIT(&hsh->head);
502 }
503
504 #ifdef INET6
505 /* Allocate hash. */
506 priv->hash6 = malloc(NBUCKETS * sizeof(struct flow_hash_entry),
507 M_NETFLOW_HASH, M_WAITOK | M_ZERO);
508
509 /* Initialize hash. */
510 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++) {
511 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
512 TAILQ_INIT(&hsh->head);
513 }
514 #endif
515
516 priv->nfinfo_bytes = counter_u64_alloc(M_WAITOK);
517 priv->nfinfo_packets = counter_u64_alloc(M_WAITOK);
518 priv->nfinfo_bytes6 = counter_u64_alloc(M_WAITOK);
519 priv->nfinfo_packets6 = counter_u64_alloc(M_WAITOK);
520 priv->nfinfo_sbytes = counter_u64_alloc(M_WAITOK);
521 priv->nfinfo_spackets = counter_u64_alloc(M_WAITOK);
522 priv->nfinfo_sbytes6 = counter_u64_alloc(M_WAITOK);
523 priv->nfinfo_spackets6 = counter_u64_alloc(M_WAITOK);
524 priv->nfinfo_act_exp = counter_u64_alloc(M_WAITOK);
525 priv->nfinfo_inact_exp = counter_u64_alloc(M_WAITOK);
526
527 ng_netflow_v9_cache_init(priv);
528 CTR0(KTR_NET, "ng_netflow startup()");
529 }
530
531 /* Initialize new FIB table for v5 and v9 */
532 int
ng_netflow_fib_init(priv_p priv,int fib)533 ng_netflow_fib_init(priv_p priv, int fib)
534 {
535 fib_export_p fe = priv_to_fib(priv, fib);
536
537 CTR1(KTR_NET, "ng_netflow(): fib init: %d", fib);
538
539 if (fe != NULL)
540 return (0);
541
542 if ((fe = malloc(sizeof(struct fib_export), M_NETGRAPH,
543 M_NOWAIT | M_ZERO)) == NULL)
544 return (ENOMEM);
545
546 mtx_init(&fe->export_mtx, "export dgram lock", NULL, MTX_DEF);
547 mtx_init(&fe->export9_mtx, "export9 dgram lock", NULL, MTX_DEF);
548 fe->fib = fib;
549 fe->domain_id = fib;
550
551 if (atomic_cmpset_ptr((volatile uintptr_t *)&priv->fib_data[fib],
552 (uintptr_t)NULL, (uintptr_t)fe) == 0) {
553 /* FIB already set up by other ISR */
554 CTR3(KTR_NET, "ng_netflow(): fib init: %d setup %p but got %p",
555 fib, fe, priv_to_fib(priv, fib));
556 mtx_destroy(&fe->export_mtx);
557 mtx_destroy(&fe->export9_mtx);
558 free(fe, M_NETGRAPH);
559 } else {
560 /* Increase counter for statistics */
561 CTR3(KTR_NET, "ng_netflow(): fib %d setup to %p (%p)",
562 fib, fe, priv_to_fib(priv, fib));
563 priv->nfinfo_alloc_fibs++;
564 }
565
566 return (0);
567 }
568
569 /* Free all flow cache memory. Called from node close method. */
570 void
ng_netflow_cache_flush(priv_p priv)571 ng_netflow_cache_flush(priv_p priv)
572 {
573 struct flow_entry *fle, *fle1;
574 struct flow_hash_entry *hsh;
575 struct netflow_export_item exp;
576 fib_export_p fe;
577 int i;
578
579 bzero(&exp, sizeof(exp));
580
581 /*
582 * We are going to free probably billable data.
583 * Expire everything before freeing it.
584 * No locking is required since callout is already drained.
585 */
586 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
587 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
588 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
589 fe = priv_to_fib(priv, fle->f.r.fib);
590 expire_flow(priv, fe, fle, NG_QUEUE);
591 }
592 #ifdef INET6
593 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++)
594 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
595 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
596 fe = priv_to_fib(priv, fle->f.r.fib);
597 expire_flow(priv, fe, fle, NG_QUEUE);
598 }
599 #endif
600
601 uma_zdestroy(priv->zone);
602 /* Destroy hash mutexes. */
603 for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
604 mtx_destroy(&hsh->mtx);
605
606 /* Free hash memory. */
607 if (priv->hash != NULL)
608 free(priv->hash, M_NETFLOW_HASH);
609 #ifdef INET6
610 uma_zdestroy(priv->zone6);
611 /* Destroy hash mutexes. */
612 for (i = 0, hsh = priv->hash6; i < NBUCKETS; i++, hsh++)
613 mtx_destroy(&hsh->mtx);
614
615 /* Free hash memory. */
616 if (priv->hash6 != NULL)
617 free(priv->hash6, M_NETFLOW_HASH);
618 #endif
619
620 for (i = 0; i < priv->maxfibs; i++) {
621 if ((fe = priv_to_fib(priv, i)) == NULL)
622 continue;
623
624 if (fe->exp.item != NULL)
625 export_send(priv, fe, fe->exp.item, NG_QUEUE);
626
627 if (fe->exp.item9 != NULL)
628 export9_send(priv, fe, fe->exp.item9,
629 fe->exp.item9_opt, NG_QUEUE);
630
631 mtx_destroy(&fe->export_mtx);
632 mtx_destroy(&fe->export9_mtx);
633 free(fe, M_NETGRAPH);
634 }
635
636 counter_u64_free(priv->nfinfo_bytes);
637 counter_u64_free(priv->nfinfo_packets);
638 counter_u64_free(priv->nfinfo_bytes6);
639 counter_u64_free(priv->nfinfo_packets6);
640 counter_u64_free(priv->nfinfo_sbytes);
641 counter_u64_free(priv->nfinfo_spackets);
642 counter_u64_free(priv->nfinfo_sbytes6);
643 counter_u64_free(priv->nfinfo_spackets6);
644 counter_u64_free(priv->nfinfo_act_exp);
645 counter_u64_free(priv->nfinfo_inact_exp);
646
647 ng_netflow_v9_cache_flush(priv);
648 }
649
650 #ifdef INET
651 /* Insert packet from into flow cache. */
652 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)653 ng_netflow_flow_add(priv_p priv, fib_export_p fe, struct ip *ip,
654 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
655 unsigned int src_if_index)
656 {
657 struct flow_entry *fle, *fle1;
658 struct flow_hash_entry *hsh;
659 struct flow_rec r;
660 int hlen, plen;
661 int error = 0;
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 /* Assume L4 template by default */
674 r.flow_type = NETFLOW_V9_FLOW_V4_L4;
675
676 r.r_src = ip->ip_src;
677 r.r_dst = ip->ip_dst;
678 r.fib = fe->fib;
679
680 plen = ntohs(ip->ip_len);
681
682 r.r_ip_p = ip->ip_p;
683 r.r_tos = ip->ip_tos;
684
685 r.r_i_ifx = src_if_index;
686
687 /*
688 * XXX NOTE: only first fragment of fragmented TCP, UDP and
689 * ICMP packet will be recorded with proper s_port and d_port.
690 * Following fragments will be recorded simply as IP packet with
691 * ip_proto = ip->ip_p and s_port, d_port set to zero.
692 * I know, it looks like bug. But I don't want to re-implement
693 * ip packet assebmling here. Anyway, (in)famous trafd works this way -
694 * and nobody complains yet :)
695 */
696 if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
697 switch(r.r_ip_p) {
698 case IPPROTO_TCP:
699 {
700 struct tcphdr *tcp;
701
702 tcp = (struct tcphdr *)((caddr_t )ip + hlen);
703 r.r_sport = tcp->th_sport;
704 r.r_dport = tcp->th_dport;
705 tcp_flags = tcp->th_flags;
706 break;
707 }
708 case IPPROTO_UDP:
709 r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
710 break;
711 }
712
713 counter_u64_add(priv->nfinfo_packets, 1);
714 counter_u64_add(priv->nfinfo_bytes, plen);
715
716 /* Find hash slot. */
717 hsh = &priv->hash[ip_hash(&r)];
718
719 mtx_lock(&hsh->mtx);
720
721 /*
722 * Go through hash and find our entry. If we encounter an
723 * entry, that should be expired, purge it. We do a reverse
724 * search since most active entries are first, and most
725 * searches are done on most active entries.
726 */
727 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
728 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
729 break;
730 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
731 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
732 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
733 fle, NG_QUEUE);
734 counter_u64_add(priv->nfinfo_act_exp, 1);
735 }
736 }
737
738 if (fle) { /* An existent entry. */
739
740 fle->f.bytes += plen;
741 fle->f.packets ++;
742 fle->f.tcp_flags |= tcp_flags;
743 fle->f.last = time_uptime;
744
745 /*
746 * We have the following reasons to expire flow in active way:
747 * - it hit active timeout
748 * - a TCP connection closed
749 * - it is going to overflow counter
750 */
751 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
752 (fle->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
753 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
754 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib),
755 fle, NG_QUEUE);
756 counter_u64_add(priv->nfinfo_act_exp, 1);
757 } else {
758 /*
759 * It is the newest, move it to the tail,
760 * if it isn't there already. Next search will
761 * locate it quicker.
762 */
763 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
764 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
765 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
766 }
767 }
768 } else /* A new flow entry. */
769 error = hash_insert(priv, hsh, &r, plen, flags, tcp_flags);
770
771 mtx_unlock(&hsh->mtx);
772
773 return (error);
774 }
775 #endif
776
777 #ifdef INET6
778 /* Insert IPv6 packet from into flow cache. */
779 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)780 ng_netflow_flow6_add(priv_p priv, fib_export_p fe, struct ip6_hdr *ip6,
781 caddr_t upper_ptr, uint8_t upper_proto, uint8_t flags,
782 unsigned int src_if_index)
783 {
784 struct flow_entry *fle = NULL, *fle1;
785 struct flow6_entry *fle6;
786 struct flow_hash_entry *hsh;
787 struct flow6_rec r;
788 int plen;
789 int error = 0;
790 uint8_t tcp_flags = 0;
791
792 /* check version */
793 if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
794 return (EINVAL);
795
796 bzero(&r, sizeof(r));
797
798 r.src.r_src6 = ip6->ip6_src;
799 r.dst.r_dst6 = ip6->ip6_dst;
800 r.fib = fe->fib;
801
802 /* Assume L4 template by default */
803 r.flow_type = NETFLOW_V9_FLOW_V6_L4;
804
805 plen = ntohs(ip6->ip6_plen) + sizeof(struct ip6_hdr);
806
807 #if 0
808 /* XXX: set DSCP/CoS value */
809 r.r_tos = ip->ip_tos;
810 #endif
811 if ((flags & NG_NETFLOW_IS_FRAG) == 0) {
812 switch(upper_proto) {
813 case IPPROTO_TCP:
814 {
815 struct tcphdr *tcp;
816
817 tcp = (struct tcphdr *)upper_ptr;
818 r.r_ports = *(uint32_t *)upper_ptr;
819 tcp_flags = tcp->th_flags;
820 break;
821 }
822 case IPPROTO_UDP:
823 case IPPROTO_SCTP:
824 r.r_ports = *(uint32_t *)upper_ptr;
825 break;
826 }
827 }
828
829 r.r_ip_p = upper_proto;
830 r.r_i_ifx = src_if_index;
831
832 counter_u64_add(priv->nfinfo_packets6, 1);
833 counter_u64_add(priv->nfinfo_bytes6, plen);
834
835 /* Find hash slot. */
836 hsh = &priv->hash6[ip6_hash(&r)];
837
838 mtx_lock(&hsh->mtx);
839
840 /*
841 * Go through hash and find our entry. If we encounter an
842 * entry, that should be expired, purge it. We do a reverse
843 * search since most active entries are first, and most
844 * searches are done on most active entries.
845 */
846 TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
847 if (fle->f.version != IP6VERSION)
848 continue;
849 fle6 = (struct flow6_entry *)fle;
850 if (bcmp(&r, &fle6->f.r, sizeof(struct flow6_rec)) == 0)
851 break;
852 if ((INACTIVE(fle6) && SMALL(fle6)) || AGED(fle6)) {
853 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
854 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
855 NG_QUEUE);
856 counter_u64_add(priv->nfinfo_act_exp, 1);
857 }
858 }
859
860 if (fle != NULL) { /* An existent entry. */
861 fle6 = (struct flow6_entry *)fle;
862
863 fle6->f.bytes += plen;
864 fle6->f.packets ++;
865 fle6->f.tcp_flags |= tcp_flags;
866 fle6->f.last = time_uptime;
867
868 /*
869 * We have the following reasons to expire flow in active way:
870 * - it hit active timeout
871 * - a TCP connection closed
872 * - it is going to overflow counter
873 */
874 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle6) ||
875 (fle6->f.bytes >= (CNTR_MAX - IF_MAXMTU)) ) {
876 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
877 expire_flow(priv, priv_to_fib(priv, fle->f.r.fib), fle,
878 NG_QUEUE);
879 counter_u64_add(priv->nfinfo_act_exp, 1);
880 } else {
881 /*
882 * It is the newest, move it to the tail,
883 * if it isn't there already. Next search will
884 * locate it quicker.
885 */
886 if (fle != TAILQ_LAST(&hsh->head, fhead)) {
887 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
888 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
889 }
890 }
891 } else /* A new flow entry. */
892 error = hash6_insert(priv, hsh, &r, plen, flags, tcp_flags);
893
894 mtx_unlock(&hsh->mtx);
895
896 return (error);
897 }
898 #endif
899
900 /*
901 * Return records from cache to userland.
902 *
903 * TODO: matching particular IP should be done in kernel, here.
904 */
905 int
ng_netflow_flow_show(priv_p priv,struct ngnf_show_header * req,struct ngnf_show_header * resp)906 ng_netflow_flow_show(priv_p priv, struct ngnf_show_header *req,
907 struct ngnf_show_header *resp)
908 {
909 struct flow_hash_entry *hsh;
910 struct flow_entry *fle;
911 struct flow_entry_data *data = (struct flow_entry_data *)(resp + 1);
912 #ifdef INET6
913 struct flow6_entry_data *data6 = (struct flow6_entry_data *)(resp + 1);
914 #endif
915 int i, max;
916
917 i = req->hash_id;
918 if (i > NBUCKETS-1)
919 return (EINVAL);
920
921 #ifdef INET6
922 if (req->version == 6) {
923 resp->version = 6;
924 hsh = priv->hash6 + i;
925 max = NREC6_AT_ONCE;
926 } else
927 #endif
928 if (req->version == 4) {
929 resp->version = 4;
930 hsh = priv->hash + i;
931 max = NREC_AT_ONCE;
932 } else
933 return (EINVAL);
934
935 /*
936 * We will transfer not more than NREC_AT_ONCE. More data
937 * will come in next message.
938 * We send current hash index and current record number in list
939 * to userland, and userland should return it back to us.
940 * Then, we will restart with new entry.
941 *
942 * The resulting cache snapshot can be inaccurate if flow expiration
943 * is taking place on hash item between userland data requests for
944 * this hash item id.
945 */
946 resp->nentries = 0;
947 for (; i < NBUCKETS; hsh++, i++) {
948 int list_id;
949
950 if (mtx_trylock(&hsh->mtx) == 0) {
951 /*
952 * Requested hash index is not available,
953 * relay decision to skip or re-request data
954 * to userland.
955 */
956 resp->hash_id = i;
957 resp->list_id = 0;
958 return (0);
959 }
960
961 list_id = 0;
962 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
963 if (hsh->mtx.mtx_lock & MTX_CONTESTED) {
964 resp->hash_id = i;
965 resp->list_id = list_id;
966 mtx_unlock(&hsh->mtx);
967 return (0);
968 }
969
970 list_id++;
971 /* Search for particular record in list. */
972 if (req->list_id > 0) {
973 if (list_id < req->list_id)
974 continue;
975
976 /* Requested list position found. */
977 req->list_id = 0;
978 }
979 #ifdef INET6
980 if (req->version == 6) {
981 struct flow6_entry *fle6;
982
983 fle6 = (struct flow6_entry *)fle;
984 bcopy(&fle6->f, data6 + resp->nentries,
985 sizeof(fle6->f));
986 } else
987 #endif
988 bcopy(&fle->f, data + resp->nentries,
989 sizeof(fle->f));
990 resp->nentries++;
991 if (resp->nentries == max) {
992 resp->hash_id = i;
993 /*
994 * If it was the last item in list
995 * we simply skip to next hash_id.
996 */
997 resp->list_id = list_id + 1;
998 mtx_unlock(&hsh->mtx);
999 return (0);
1000 }
1001 }
1002 mtx_unlock(&hsh->mtx);
1003 }
1004
1005 resp->hash_id = resp->list_id = 0;
1006
1007 return (0);
1008 }
1009
1010 /* We have full datagram in privdata. Send it to export hook. */
1011 static int
export_send(priv_p priv,fib_export_p fe,item_p item,int flags)1012 export_send(priv_p priv, fib_export_p fe, item_p item, int flags)
1013 {
1014 struct mbuf *m = NGI_M(item);
1015 struct netflow_v5_export_dgram *dgram = mtod(m,
1016 struct netflow_v5_export_dgram *);
1017 struct netflow_v5_header *header = &dgram->header;
1018 struct timespec ts;
1019 int error = 0;
1020
1021 /* Fill mbuf header. */
1022 m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
1023 header->count + sizeof(struct netflow_v5_header);
1024
1025 /* Fill export header. */
1026 header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
1027 getnanotime(&ts);
1028 header->unix_secs = htonl(ts.tv_sec);
1029 header->unix_nsecs = htonl(ts.tv_nsec);
1030 header->engine_type = 0;
1031 header->engine_id = fe->domain_id;
1032 header->pad = 0;
1033 header->flow_seq = htonl(atomic_fetchadd_32(&fe->flow_seq,
1034 header->count));
1035 header->count = htons(header->count);
1036
1037 if (priv->export != NULL)
1038 NG_FWD_ITEM_HOOK_FLAGS(error, item, priv->export, flags);
1039 else
1040 NG_FREE_ITEM(item);
1041
1042 return (error);
1043 }
1044
1045 /* Add export record to dgram. */
1046 static int
export_add(item_p item,struct flow_entry * fle)1047 export_add(item_p item, struct flow_entry *fle)
1048 {
1049 struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
1050 struct netflow_v5_export_dgram *);
1051 struct netflow_v5_header *header = &dgram->header;
1052 struct netflow_v5_record *rec;
1053
1054 rec = &dgram->r[header->count];
1055 header->count ++;
1056
1057 KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
1058 ("ng_netflow: export too big"));
1059
1060 /* Fill in export record. */
1061 rec->src_addr = fle->f.r.r_src.s_addr;
1062 rec->dst_addr = fle->f.r.r_dst.s_addr;
1063 rec->next_hop = fle->f.next_hop.s_addr;
1064 rec->i_ifx = htons(fle->f.fle_i_ifx);
1065 rec->o_ifx = htons(fle->f.fle_o_ifx);
1066 rec->packets = htonl(fle->f.packets);
1067 rec->octets = htonl(fle->f.bytes);
1068 rec->first = htonl(MILLIUPTIME(fle->f.first));
1069 rec->last = htonl(MILLIUPTIME(fle->f.last));
1070 rec->s_port = fle->f.r.r_sport;
1071 rec->d_port = fle->f.r.r_dport;
1072 rec->flags = fle->f.tcp_flags;
1073 rec->prot = fle->f.r.r_ip_p;
1074 rec->tos = fle->f.r.r_tos;
1075 rec->dst_mask = fle->f.dst_mask;
1076 rec->src_mask = fle->f.src_mask;
1077 rec->pad1 = 0;
1078 rec->pad2 = 0;
1079
1080 /* Not supported fields. */
1081 rec->src_as = rec->dst_as = 0;
1082
1083 if (header->count == NETFLOW_V5_MAX_RECORDS)
1084 return (1); /* end of datagram */
1085 else
1086 return (0);
1087 }
1088
1089 /* Periodic flow expiry run. */
1090 void
ng_netflow_expire(void * arg)1091 ng_netflow_expire(void *arg)
1092 {
1093 struct flow_entry *fle, *fle1;
1094 struct flow_hash_entry *hsh;
1095 priv_p priv = (priv_p )arg;
1096 int used, i;
1097
1098 /*
1099 * Going through all the cache.
1100 */
1101 used = uma_zone_get_cur(priv->zone);
1102 for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
1103 /*
1104 * Skip entries, that are already being worked on.
1105 */
1106 if (mtx_trylock(&hsh->mtx) == 0)
1107 continue;
1108
1109 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1110 /*
1111 * Interrupt thread wants this entry!
1112 * Quick! Quick! Bail out!
1113 */
1114 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1115 break;
1116
1117 /*
1118 * Don't expire aggressively while hash collision
1119 * ratio is predicted small.
1120 */
1121 if (used <= (NBUCKETS*2) && !INACTIVE(fle))
1122 break;
1123
1124 if ((INACTIVE(fle) && (SMALL(fle) ||
1125 (used > (NBUCKETS*2)))) || AGED(fle)) {
1126 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1127 expire_flow(priv, priv_to_fib(priv,
1128 fle->f.r.fib), fle, NG_NOFLAGS);
1129 used--;
1130 counter_u64_add(priv->nfinfo_inact_exp, 1);
1131 }
1132 }
1133 mtx_unlock(&hsh->mtx);
1134 }
1135
1136 #ifdef INET6
1137 used = uma_zone_get_cur(priv->zone6);
1138 for (hsh = priv->hash6, i = 0; i < NBUCKETS; hsh++, i++) {
1139 struct flow6_entry *fle6;
1140
1141 /*
1142 * Skip entries, that are already being worked on.
1143 */
1144 if (mtx_trylock(&hsh->mtx) == 0)
1145 continue;
1146
1147 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
1148 fle6 = (struct flow6_entry *)fle;
1149 /*
1150 * Interrupt thread wants this entry!
1151 * Quick! Quick! Bail out!
1152 */
1153 if (hsh->mtx.mtx_lock & MTX_CONTESTED)
1154 break;
1155
1156 /*
1157 * Don't expire aggressively while hash collision
1158 * ratio is predicted small.
1159 */
1160 if (used <= (NBUCKETS*2) && !INACTIVE(fle6))
1161 break;
1162
1163 if ((INACTIVE(fle6) && (SMALL(fle6) ||
1164 (used > (NBUCKETS*2)))) || AGED(fle6)) {
1165 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
1166 expire_flow(priv, priv_to_fib(priv,
1167 fle->f.r.fib), fle, NG_NOFLAGS);
1168 used--;
1169 counter_u64_add(priv->nfinfo_inact_exp, 1);
1170 }
1171 }
1172 mtx_unlock(&hsh->mtx);
1173 }
1174 #endif
1175
1176 /* Schedule next expire. */
1177 callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
1178 (void *)priv);
1179 }
1180