xref: /netbsd-src/sys/netinet/ip_flow.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: ip_flow.c,v 1.66 2015/03/23 18:33:17 roy Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by the 3am Software Foundry ("3am").  It was developed by Matt Thomas.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/cdefs.h>
33 __KERNEL_RCSID(0, "$NetBSD: ip_flow.c,v 1.66 2015/03/23 18:33:17 roy Exp $");
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/domain.h>
40 #include <sys/protosw.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/errno.h>
44 #include <sys/time.h>
45 #include <sys/kernel.h>
46 #include <sys/pool.h>
47 #include <sys/sysctl.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/route.h>
52 #include <net/pfil.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/ip.h>
57 #include <netinet/in_pcb.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/ip_private.h>
61 
62 /*
63  * Similar code is very well commented in netinet6/ip6_flow.c
64  */
65 
66 #define	IPFLOW_HASHBITS		6	/* should not be a multiple of 8 */
67 
68 static struct pool ipflow_pool;
69 
70 LIST_HEAD(ipflowhead, ipflow);
71 
72 #define	IPFLOW_TIMER		(5 * PR_SLOWHZ)
73 #define	IPFLOW_DEFAULT_HASHSIZE	(1 << IPFLOW_HASHBITS)
74 
75 static struct ipflowhead *ipflowtable = NULL;
76 static struct ipflowhead ipflowlist;
77 static int ipflow_inuse;
78 
79 #define	IPFLOW_INSERT(bucket, ipf) \
80 do { \
81 	LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \
82 	LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \
83 } while (/*CONSTCOND*/ 0)
84 
85 #define	IPFLOW_REMOVE(ipf) \
86 do { \
87 	LIST_REMOVE((ipf), ipf_hash); \
88 	LIST_REMOVE((ipf), ipf_list); \
89 } while (/*CONSTCOND*/ 0)
90 
91 #ifndef IPFLOW_MAX
92 #define	IPFLOW_MAX		256
93 #endif
94 static int ip_maxflows = IPFLOW_MAX;
95 static int ip_hashsize = IPFLOW_DEFAULT_HASHSIZE;
96 
97 static void ipflow_sysctl_init(struct sysctllog **);
98 
99 static size_t
100 ipflow_hash(const struct ip *ip)
101 {
102 	size_t hash = ip->ip_tos;
103 	size_t idx;
104 
105 	for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) {
106 		hash += (ip->ip_dst.s_addr >> (32 - idx)) +
107 		    (ip->ip_src.s_addr >> idx);
108 	}
109 
110 	return hash & (ip_hashsize-1);
111 }
112 
113 static struct ipflow *
114 ipflow_lookup(const struct ip *ip)
115 {
116 	size_t hash;
117 	struct ipflow *ipf;
118 
119 	hash = ipflow_hash(ip);
120 
121 	LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) {
122 		if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr
123 		    && ip->ip_src.s_addr == ipf->ipf_src.s_addr
124 		    && ip->ip_tos == ipf->ipf_tos)
125 			break;
126 	}
127 	return ipf;
128 }
129 
130 void
131 ipflow_poolinit(void)
132 {
133 
134 	pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl",
135 	    NULL, IPL_NET);
136 }
137 
138 static int
139 ipflow_reinit(int table_size)
140 {
141 	struct ipflowhead *new_table;
142 	size_t i;
143 
144 	new_table = (struct ipflowhead *)malloc(sizeof(struct ipflowhead) *
145 	    table_size, M_RTABLE, M_NOWAIT);
146 
147 	if (new_table == NULL)
148 		return 1;
149 
150 	if (ipflowtable != NULL)
151 		free(ipflowtable, M_RTABLE);
152 
153 	ipflowtable = new_table;
154 	ip_hashsize = table_size;
155 
156 	LIST_INIT(&ipflowlist);
157 	for (i = 0; i < ip_hashsize; i++)
158 		LIST_INIT(&ipflowtable[i]);
159 
160 	return 0;
161 }
162 
163 void
164 ipflow_init(void)
165 {
166 	(void)ipflow_reinit(ip_hashsize);
167 	ipflow_sysctl_init(NULL);
168 }
169 
170 int
171 ipflow_fastforward(struct mbuf *m)
172 {
173 	struct ip *ip;
174 	struct ip ip_store;
175 	struct ipflow *ipf;
176 	struct rtentry *rt;
177 	const struct sockaddr *dst;
178 	int error;
179 	int iplen;
180 
181 	/*
182 	 * Are we forwarding packets?  Big enough for an IP packet?
183 	 */
184 	if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip))
185 		return 0;
186 
187 	/*
188 	 * Was packet received as a link-level multicast or broadcast?
189 	 * If so, don't try to fast forward..
190 	 */
191 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
192 		return 0;
193 
194 	/*
195 	 * IP header with no option and valid version and length
196 	 */
197 	if (IP_HDR_ALIGNED_P(mtod(m, const void *)))
198 		ip = mtod(m, struct ip *);
199 	else {
200 		memcpy(&ip_store, mtod(m, const void *), sizeof(ip_store));
201 		ip = &ip_store;
202 	}
203 	iplen = ntohs(ip->ip_len);
204 	if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) ||
205 	    iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len)
206 		return 0;
207 	/*
208 	 * Find a flow.
209 	 */
210 	if ((ipf = ipflow_lookup(ip)) == NULL)
211 		return 0;
212 
213 	/*
214 	 * Verify the IP header checksum.
215 	 */
216 	switch (m->m_pkthdr.csum_flags &
217 		((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_IPv4) |
218 		 M_CSUM_IPv4_BAD)) {
219 	case M_CSUM_IPv4|M_CSUM_IPv4_BAD:
220 		return (0);
221 
222 	case M_CSUM_IPv4:
223 		/* Checksum was okay. */
224 		break;
225 
226 	default:
227 		/* Must compute it ourselves. */
228 		if (in_cksum(m, sizeof(struct ip)) != 0)
229 			return (0);
230 		break;
231 	}
232 
233 	/*
234 	 * Route and interface still up?
235 	 */
236 	if ((rt = rtcache_validate(&ipf->ipf_ro)) == NULL ||
237 	    (rt->rt_ifp->if_flags & IFF_UP) == 0 ||
238 	    (rt->rt_flags & (RTF_BLACKHOLE | RTF_BROADCAST)) != 0)
239 		return 0;
240 
241 	/*
242 	 * Packet size OK?  TTL?
243 	 */
244 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC)
245 		return 0;
246 
247 	/*
248 	 * Clear any in-bound checksum flags for this packet.
249 	 */
250 	m->m_pkthdr.csum_flags = 0;
251 
252 	/*
253 	 * Everything checks out and so we can forward this packet.
254 	 * Modify the TTL and incrementally change the checksum.
255 	 *
256 	 * This method of adding the checksum works on either endian CPU.
257 	 * If htons() is inlined, all the arithmetic is folded; otherwise
258 	 * the htons()s are combined by CSE due to the const attribute.
259 	 *
260 	 * Don't bother using HW checksumming here -- the incremental
261 	 * update is pretty fast.
262 	 */
263 	ip->ip_ttl -= IPTTLDEC;
264 	if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8))
265 		ip->ip_sum -= ~htons(IPTTLDEC << 8);
266 	else
267 		ip->ip_sum += htons(IPTTLDEC << 8);
268 
269 	/*
270 	 * Done modifying the header; copy it back, if necessary.
271 	 *
272 	 * XXX Use m_copyback_cow(9) here? --dyoung
273 	 */
274 	if (IP_HDR_ALIGNED_P(mtod(m, void *)) == 0)
275 		memcpy(mtod(m, void *), &ip_store, sizeof(ip_store));
276 
277 	/*
278 	 * Trim the packet in case it's too long..
279 	 */
280 	if (m->m_pkthdr.len > iplen) {
281 		if (m->m_len == m->m_pkthdr.len) {
282 			m->m_len = iplen;
283 			m->m_pkthdr.len = iplen;
284 		} else
285 			m_adj(m, iplen - m->m_pkthdr.len);
286 	}
287 
288 	/*
289 	 * Send the packet on its way.  All we can get back is ENOBUFS
290 	 */
291 	ipf->ipf_uses++;
292 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
293 
294 	if (rt->rt_flags & RTF_GATEWAY)
295 		dst = rt->rt_gateway;
296 	else
297 		dst = rtcache_getdst(&ipf->ipf_ro);
298 
299 	KERNEL_LOCK(1, NULL);
300 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
301 		if (error == ENOBUFS)
302 			ipf->ipf_dropped++;
303 		else
304 			ipf->ipf_errors++;
305 	}
306 	KERNEL_UNLOCK_ONE(NULL);
307 	return 1;
308 }
309 
310 static void
311 ipflow_addstats(struct ipflow *ipf)
312 {
313 	struct rtentry *rt;
314 	uint64_t *ips;
315 
316 	if ((rt = rtcache_validate(&ipf->ipf_ro)) != NULL)
317 		rt->rt_use += ipf->ipf_uses;
318 
319 	ips = IP_STAT_GETREF();
320 	ips[IP_STAT_CANTFORWARD] += ipf->ipf_errors + ipf->ipf_dropped;
321 	ips[IP_STAT_TOTAL] += ipf->ipf_uses;
322 	ips[IP_STAT_FORWARD] += ipf->ipf_uses;
323 	ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
324 	IP_STAT_PUTREF();
325 }
326 
327 static void
328 ipflow_free(struct ipflow *ipf)
329 {
330 	int s;
331 	/*
332 	 * Remove the flow from the hash table (at elevated IPL).
333 	 * Once it's off the list, we can deal with it at normal
334 	 * network IPL.
335 	 */
336 	s = splnet();
337 	IPFLOW_REMOVE(ipf);
338 	splx(s);
339 	ipflow_addstats(ipf);
340 	rtcache_free(&ipf->ipf_ro);
341 	ipflow_inuse--;
342 	s = splnet();
343 	pool_put(&ipflow_pool, ipf);
344 	splx(s);
345 }
346 
347 struct ipflow *
348 ipflow_reap(bool just_one)
349 {
350 	while (just_one || ipflow_inuse > ip_maxflows) {
351 		struct ipflow *ipf, *maybe_ipf = NULL;
352 		int s;
353 
354 		ipf = LIST_FIRST(&ipflowlist);
355 		while (ipf != NULL) {
356 			/*
357 			 * If this no longer points to a valid route
358 			 * reclaim it.
359 			 */
360 			if (rtcache_validate(&ipf->ipf_ro) == NULL)
361 				goto done;
362 			/*
363 			 * choose the one that's been least recently
364 			 * used or has had the least uses in the
365 			 * last 1.5 intervals.
366 			 */
367 			if (maybe_ipf == NULL ||
368 			    ipf->ipf_timer < maybe_ipf->ipf_timer ||
369 			    (ipf->ipf_timer == maybe_ipf->ipf_timer &&
370 			     ipf->ipf_last_uses + ipf->ipf_uses <
371 			         maybe_ipf->ipf_last_uses +
372 			         maybe_ipf->ipf_uses))
373 				maybe_ipf = ipf;
374 			ipf = LIST_NEXT(ipf, ipf_list);
375 		}
376 		ipf = maybe_ipf;
377 	    done:
378 		/*
379 		 * Remove the entry from the flow table.
380 		 */
381 		s = splnet();
382 		IPFLOW_REMOVE(ipf);
383 		splx(s);
384 		ipflow_addstats(ipf);
385 		rtcache_free(&ipf->ipf_ro);
386 		if (just_one)
387 			return ipf;
388 		pool_put(&ipflow_pool, ipf);
389 		ipflow_inuse--;
390 	}
391 	return NULL;
392 }
393 
394 void
395 ipflow_slowtimo(void)
396 {
397 	struct rtentry *rt;
398 	struct ipflow *ipf, *next_ipf;
399 	uint64_t *ips;
400 
401 	mutex_enter(softnet_lock);
402 	KERNEL_LOCK(1, NULL);
403 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
404 		next_ipf = LIST_NEXT(ipf, ipf_list);
405 		if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer) ||
406 		    (rt = rtcache_validate(&ipf->ipf_ro)) == NULL) {
407 			ipflow_free(ipf);
408 		} else {
409 			ipf->ipf_last_uses = ipf->ipf_uses;
410 			rt->rt_use += ipf->ipf_uses;
411 			ips = IP_STAT_GETREF();
412 			ips[IP_STAT_TOTAL] += ipf->ipf_uses;
413 			ips[IP_STAT_FORWARD] += ipf->ipf_uses;
414 			ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses;
415 			IP_STAT_PUTREF();
416 			ipf->ipf_uses = 0;
417 		}
418 	}
419 	KERNEL_UNLOCK_ONE(NULL);
420 	mutex_exit(softnet_lock);
421 }
422 
423 void
424 ipflow_create(const struct route *ro, struct mbuf *m)
425 {
426 	const struct ip *const ip = mtod(m, const struct ip *);
427 	struct ipflow *ipf;
428 	size_t hash;
429 	int s;
430 
431 	/*
432 	 * Don't create cache entries for ICMP messages.
433 	 */
434 	if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP)
435 		return;
436 
437 	KERNEL_LOCK(1, NULL);
438 
439 	/*
440 	 * See if an existing flow struct exists.  If so remove it from its
441 	 * list and free the old route.  If not, try to malloc a new one
442 	 * (if we aren't at our limit).
443 	 */
444 	ipf = ipflow_lookup(ip);
445 	if (ipf == NULL) {
446 		if (ipflow_inuse >= ip_maxflows) {
447 			ipf = ipflow_reap(true);
448 		} else {
449 			s = splnet();
450 			ipf = pool_get(&ipflow_pool, PR_NOWAIT);
451 			splx(s);
452 			if (ipf == NULL)
453 				goto out;
454 			ipflow_inuse++;
455 		}
456 		memset(ipf, 0, sizeof(*ipf));
457 	} else {
458 		s = splnet();
459 		IPFLOW_REMOVE(ipf);
460 		splx(s);
461 		ipflow_addstats(ipf);
462 		rtcache_free(&ipf->ipf_ro);
463 		ipf->ipf_uses = ipf->ipf_last_uses = 0;
464 		ipf->ipf_errors = ipf->ipf_dropped = 0;
465 	}
466 
467 	/*
468 	 * Fill in the updated information.
469 	 */
470 	rtcache_copy(&ipf->ipf_ro, ro);
471 	ipf->ipf_dst = ip->ip_dst;
472 	ipf->ipf_src = ip->ip_src;
473 	ipf->ipf_tos = ip->ip_tos;
474 	PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER);
475 
476 	/*
477 	 * Insert into the approriate bucket of the flow table.
478 	 */
479 	hash = ipflow_hash(ip);
480 	s = splnet();
481 	IPFLOW_INSERT(&ipflowtable[hash], ipf);
482 	splx(s);
483 
484  out:
485 	KERNEL_UNLOCK_ONE(NULL);
486 }
487 
488 int
489 ipflow_invalidate_all(int new_size)
490 {
491 	struct ipflow *ipf, *next_ipf;
492 	int s, error;
493 
494 	error = 0;
495 	s = splnet();
496 	for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) {
497 		next_ipf = LIST_NEXT(ipf, ipf_list);
498 		ipflow_free(ipf);
499 	}
500 
501 	if (new_size)
502 		error = ipflow_reinit(new_size);
503 	splx(s);
504 
505 	return error;
506 }
507 
508 #ifdef GATEWAY
509 /*
510  * sysctl helper routine for net.inet.ip.maxflows.
511  */
512 static int
513 sysctl_net_inet_ip_maxflows(SYSCTLFN_ARGS)
514 {
515 	int error;
516 
517 	error = sysctl_lookup(SYSCTLFN_CALL(rnode));
518 	if (error || newp == NULL)
519 		return (error);
520 
521 	mutex_enter(softnet_lock);
522 	KERNEL_LOCK(1, NULL);
523 
524 	ipflow_reap(false);
525 
526 	KERNEL_UNLOCK_ONE(NULL);
527 	mutex_exit(softnet_lock);
528 
529 	return (0);
530 }
531 
532 static int
533 sysctl_net_inet_ip_hashsize(SYSCTLFN_ARGS)
534 {
535 	int error, tmp;
536 	struct sysctlnode node;
537 
538 	node = *rnode;
539 	tmp = ip_hashsize;
540 	node.sysctl_data = &tmp;
541 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
542 	if (error || newp == NULL)
543 		return (error);
544 
545 	if ((tmp & (tmp - 1)) == 0 && tmp != 0) {
546 		/*
547 		 * Can only fail due to malloc()
548 		 */
549 		mutex_enter(softnet_lock);
550 		KERNEL_LOCK(1, NULL);
551 
552 		error = ipflow_invalidate_all(tmp);
553 
554 		KERNEL_UNLOCK_ONE(NULL);
555 		mutex_exit(softnet_lock);
556 
557 	} else {
558 		/*
559 		 * EINVAL if not a power of 2
560 	         */
561 		error = EINVAL;
562 	}
563 
564 	return error;
565 }
566 #endif /* GATEWAY */
567 
568 static void
569 ipflow_sysctl_init(struct sysctllog **clog)
570 {
571 	sysctl_createv(clog, 0, NULL, NULL,
572 		       CTLFLAG_PERMANENT,
573 		       CTLTYPE_NODE, "inet",
574 		       SYSCTL_DESCR("PF_INET related settings"),
575 		       NULL, 0, NULL, 0,
576 		       CTL_NET, PF_INET, CTL_EOL);
577 	sysctl_createv(clog, 0, NULL, NULL,
578 		       CTLFLAG_PERMANENT,
579 		       CTLTYPE_NODE, "ip",
580 		       SYSCTL_DESCR("IPv4 related settings"),
581 		       NULL, 0, NULL, 0,
582 		       CTL_NET, PF_INET, IPPROTO_IP, CTL_EOL);
583 
584 #ifdef GATEWAY
585 	sysctl_createv(clog, 0, NULL, NULL,
586 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
587 		       CTLTYPE_INT, "maxflows",
588 		       SYSCTL_DESCR("Number of flows for fast forwarding"),
589 		       sysctl_net_inet_ip_maxflows, 0, &ip_maxflows, 0,
590 		       CTL_NET, PF_INET, IPPROTO_IP,
591 		       IPCTL_MAXFLOWS, CTL_EOL);
592 	sysctl_createv(clog, 0, NULL, NULL,
593 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
594 			CTLTYPE_INT, "hashsize",
595 			SYSCTL_DESCR("Size of hash table for fast forwarding (IPv4)"),
596 			sysctl_net_inet_ip_hashsize, 0, &ip_hashsize, 0,
597 			CTL_NET, PF_INET, IPPROTO_IP,
598 			CTL_CREATE, CTL_EOL);
599 #endif /* GATEWAY */
600 }
601