xref: /netbsd-src/sys/netinet6/ip6_flow.c (revision 74809f044aa1c34df701bf79c7bfb9794896fdf8) 
1 /*	$NetBSD: ip6_flow.c,v 1.17 2008/04/28 20:24:10 martin Exp $	*/
2 
3 /*-
4  * Copyright (c) 2007 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 Liam J. Foy
9  * <liamjfoy@netbsd.org> and Matt Thomas <matt@netbsd.org>.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  *
32  * IPv6 version was developed by Liam J. Foy. Original source existed in IPv4
33  * format developed by Matt Thomas. Thanks to Joerg Sonnenberger, Matt
34  * Thomas and Christos Zoulas.
35  *
36  * Thanks to Liverpool John Moores University, especially Dr. David Llewellyn-Jones
37  * for providing resources (to test) and Professor Madjid Merabti.
38  */
39 
40 #include <sys/cdefs.h>
41 __KERNEL_RCSID(0, "$NetBSD: ip6_flow.c,v 1.17 2008/04/28 20:24:10 martin Exp $");
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/domain.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/pool.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/if.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
59 #include <net/pfil.h>
60 
61 #include <netinet/in.h>
62 #include <netinet6/in6_var.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/ip6_private.h>
67 
68 /*
69  * IPv6 Fast Forward caches/hashes flows from one source to destination.
70  *
71  * Upon a successful forward IPv6FF caches and hashes details such as the
72  * route, source and destination. Once another packet is received matching
73  * the source and destination the packet is forwarded straight onto if_output
74  * using the cached details.
75  *
76  * Example:
77  * ether/fddi_input -> ip6flow_fastfoward -> if_output
78  */
79 
80 POOL_INIT(ip6flow_pool, sizeof(struct ip6flow), 0, 0, 0, "ip6flowpl", NULL,
81     IPL_NET);
82 
83 LIST_HEAD(ip6flowhead, ip6flow);
84 
85 /*
86  * We could use IPv4 defines (IPFLOW_HASHBITS) but we'll
87  * use our own (possibly for future expansion).
88  */
89 #define	IP6FLOW_TIMER		(5 * PR_SLOWHZ)
90 #define	IP6FLOW_DEFAULT_HASHSIZE	(1 << IP6FLOW_HASHBITS)
91 
92 static struct ip6flowhead *ip6flowtable = NULL;
93 static struct ip6flowhead ip6flowlist;
94 static int ip6flow_inuse;
95 
96 /*
97  * Insert an ip6flow into the list.
98  */
99 #define	IP6FLOW_INSERT(bucket, ip6f) \
100 do { \
101 	LIST_INSERT_HEAD((bucket), (ip6f), ip6f_hash); \
102 	LIST_INSERT_HEAD(&ip6flowlist, (ip6f), ip6f_list); \
103 } while (/*CONSTCOND*/ 0)
104 
105 /*
106  * Remove an ip6flow from the list.
107  */
108 #define	IP6FLOW_REMOVE(ip6f) \
109 do { \
110 	LIST_REMOVE((ip6f), ip6f_hash); \
111 	LIST_REMOVE((ip6f), ip6f_list); \
112 } while (/*CONSTCOND*/ 0)
113 
114 #ifndef IP6FLOW_DEFAULT
115 #define	IP6FLOW_DEFAULT		256
116 #endif
117 
118 int ip6_maxflows = IP6FLOW_DEFAULT;
119 int ip6_hashsize = IP6FLOW_DEFAULT_HASHSIZE;
120 
121 /*
122  * Calculate hash table position.
123  */
124 static size_t
125 ip6flow_hash(const struct ip6_hdr *ip6)
126 {
127 	size_t hash;
128 	uint32_t dst_sum, src_sum;
129 	size_t idx;
130 
131 	src_sum = ip6->ip6_src.s6_addr32[0] + ip6->ip6_src.s6_addr32[1]
132 	    + ip6->ip6_src.s6_addr32[2] + ip6->ip6_src.s6_addr32[3];
133 	dst_sum = ip6->ip6_dst.s6_addr32[0] + ip6->ip6_dst.s6_addr32[1]
134 	    + ip6->ip6_dst.s6_addr32[2] + ip6->ip6_dst.s6_addr32[3];
135 
136 	hash = ip6->ip6_flow;
137 
138 	for (idx = 0; idx < 32; idx += IP6FLOW_HASHBITS)
139 		hash += (dst_sum >> (32 - idx)) + (src_sum >> idx);
140 
141 	return hash & (ip6_hashsize-1);
142 }
143 
144 /*
145  * Check to see if a flow already exists - if so return it.
146  */
147 static struct ip6flow *
148 ip6flow_lookup(const struct ip6_hdr *ip6)
149 {
150 	size_t hash;
151 	struct ip6flow *ip6f;
152 
153 	hash = ip6flow_hash(ip6);
154 
155 	LIST_FOREACH(ip6f, &ip6flowtable[hash], ip6f_hash) {
156 		if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6f->ip6f_dst)
157 		    && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6f->ip6f_src)
158 		    && ip6f->ip6f_flow == ip6->ip6_flow) {
159 		    	/* A cached flow has been found. */
160 			return ip6f;
161 		}
162 	}
163 
164 	return NULL;
165 }
166 
167 /*
168  * Allocate memory and initialise lists. This function is called
169  * from ip6_init and called there after to resize the hash table.
170  * If a newly sized table cannot be malloc'ed we just continue
171  * to use the old one.
172  */
173 int
174 ip6flow_init(int table_size)
175 {
176 	struct ip6flowhead *new_table;
177 	size_t i;
178 
179 	new_table = (struct ip6flowhead *)malloc(sizeof(struct ip6flowhead) *
180 	    table_size, M_RTABLE, M_NOWAIT);
181 
182 	if (new_table == NULL)
183 		return 1;
184 
185 	if (ip6flowtable != NULL)
186 		free(ip6flowtable, M_RTABLE);
187 
188 	ip6flowtable = new_table;
189 	ip6_hashsize = table_size;
190 
191 	LIST_INIT(&ip6flowlist);
192 	for (i = 0; i < ip6_hashsize; i++)
193 		LIST_INIT(&ip6flowtable[i]);
194 
195 	return 0;
196 }
197 
198 /*
199  * IPv6 Fast Forward routine. Attempt to forward the packet -
200  * if any problems are found return to the main IPv6 input
201  * routine to deal with.
202  */
203 int
204 ip6flow_fastforward(struct mbuf *m)
205 {
206 	struct ip6flow *ip6f;
207 	struct ip6_hdr *ip6;
208 	struct rtentry *rt;
209 	const struct sockaddr *dst;
210 	int error;
211 
212 	/*
213 	 * Are we forwarding packets and have flows?
214 	 */
215 	if (!ip6_forwarding || ip6flow_inuse == 0)
216 		return 0;
217 
218 	/*
219 	 * At least size of IPv6 Header?
220 	 */
221 	if (m->m_len < sizeof(struct ip6_hdr))
222 		return 0;
223 	/*
224 	 * Was packet received as a link-level multicast or broadcast?
225 	 * If so, don't try to fast forward.
226 	 */
227 	if ((m->m_flags & (M_BCAST|M_MCAST)) != 0)
228 		return 0;
229 
230 	if (IP6_HDR_ALIGNED_P(mtod(m, const void *)) == 0) {
231 		if ((m = m_copyup(m, sizeof(struct ip6_hdr),
232 				(max_linkhdr + 3) & ~3)) == NULL) {
233 			return 0;
234 		}
235 	} else if (__predict_false(m->m_len < sizeof(struct ip6_hdr))) {
236 		if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
237 			return 0;
238 		}
239 	}
240 
241 	ip6 = mtod(m, struct ip6_hdr *);
242 
243 	if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) {
244 		/* Bad version. */
245 		return 0;
246 	}
247 
248 	/*
249 	 * If we have a hop-by-hop extension we must process it.
250 	 * We just leave this up to ip6_input to deal with.
251 	 */
252 	if (ip6->ip6_nxt == IPPROTO_HOPOPTS)
253 		return 0;
254 
255 	/*
256 	 * Attempt to find a flow.
257 	 */
258 	if ((ip6f = ip6flow_lookup(ip6)) == NULL) {
259 		/* No flow found. */
260 		return 0;
261 	}
262 
263 	/*
264 	 * Route and interface still up?
265 	 */
266 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) == NULL ||
267 	    (rt->rt_ifp->if_flags & IFF_UP) == 0) {
268 	    	/* Route or interface is down */
269 		return 0;
270 	}
271 
272 	/*
273 	 * Packet size greater than MTU?
274 	 */
275 	if (m->m_pkthdr.len > rt->rt_ifp->if_mtu) {
276 		/* Return to main IPv6 input function. */
277 		return 0;
278 	}
279 
280 	if (ip6->ip6_hlim <= IPV6_HLIMDEC)
281 		return 0;
282 
283 	/* Decrement hop limit (same as TTL) */
284 	ip6->ip6_hlim -= IPV6_HLIMDEC;
285 
286 	if (rt->rt_flags & RTF_GATEWAY)
287 		dst = rt->rt_gateway;
288 	else
289 		dst = rtcache_getdst(&ip6f->ip6f_ro);
290 
291 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
292 
293 	ip6f->ip6f_uses++;
294 
295 	/* Send on its way - straight to the interface output routine. */
296 	if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) {
297 		ip6f->ip6f_dropped++;
298 	} else {
299 		ip6f->ip6f_forwarded++;
300 	}
301 
302 	return 1;
303 }
304 
305 /*
306  * Add the IPv6 flow statistics to the main IPv6 statistics.
307  */
308 static void
309 ip6flow_addstats(const struct ip6flow *ip6f)
310 {
311 	struct rtentry *rt;
312 	uint64_t *ip6s;
313 
314 	if ((rt = rtcache_validate(&ip6f->ip6f_ro)) != NULL)
315 		rt->rt_use += ip6f->ip6f_uses;
316 	ip6s = IP6_STAT_GETREF();
317 	ip6s[IP6_STAT_FASTFORWARDFLOWS] = ip6flow_inuse;
318 	ip6s[IP6_STAT_CANTFORWARD] += ip6f->ip6f_dropped;
319 	ip6s[IP6_STAT_ODROPPED] += ip6f->ip6f_dropped;
320 	ip6s[IP6_STAT_TOTAL] += ip6f->ip6f_uses;
321 	ip6s[IP6_STAT_FORWARD] += ip6f->ip6f_forwarded;
322 	ip6s[IP6_STAT_FASTFORWARD] += ip6f->ip6f_forwarded;
323 	IP6_STAT_PUTREF();
324 }
325 
326 /*
327  * Add statistics and free the flow.
328  */
329 static void
330 ip6flow_free(struct ip6flow *ip6f)
331 {
332 	int s;
333 
334 	/*
335 	 * Remove the flow from the hash table (at elevated IPL).
336 	 * Once it's off the list, we can deal with it at normal
337 	 * network IPL.
338 	 */
339 	s = splnet();
340 	IP6FLOW_REMOVE(ip6f);
341 	splx(s);
342 	ip6flow_inuse--;
343 	ip6flow_addstats(ip6f);
344 	rtcache_free(&ip6f->ip6f_ro);
345 	pool_put(&ip6flow_pool, ip6f);
346 }
347 
348 /*
349  * Reap one or more flows - ip6flow_reap may remove
350  * multiple flows if net.inet6.ip6.maxflows is reduced.
351  */
352 struct ip6flow *
353 ip6flow_reap(int just_one)
354 {
355 	while (just_one || ip6flow_inuse > ip6_maxflows) {
356 		struct ip6flow *ip6f, *maybe_ip6f = NULL;
357 		int s;
358 
359 		ip6f = LIST_FIRST(&ip6flowlist);
360 		while (ip6f != NULL) {
361 			/*
362 			 * If this no longer points to a valid route -
363 			 * reclaim it.
364 			 */
365 			if (rtcache_validate(&ip6f->ip6f_ro) == NULL)
366 				goto done;
367 			/*
368 			 * choose the one that's been least recently
369 			 * used or has had the least uses in the
370 			 * last 1.5 intervals.
371 			 */
372 			if (maybe_ip6f == NULL ||
373 			    ip6f->ip6f_timer < maybe_ip6f->ip6f_timer ||
374 			    (ip6f->ip6f_timer == maybe_ip6f->ip6f_timer &&
375 			     ip6f->ip6f_last_uses + ip6f->ip6f_uses <
376 			         maybe_ip6f->ip6f_last_uses +
377 			         maybe_ip6f->ip6f_uses))
378 				maybe_ip6f = ip6f;
379 			ip6f = LIST_NEXT(ip6f, ip6f_list);
380 		}
381 		ip6f = maybe_ip6f;
382 	    done:
383 		/*
384 		 * Remove the entry from the flow table
385 		 */
386 		s = splnet();
387 		IP6FLOW_REMOVE(ip6f);
388 		splx(s);
389 		rtcache_free(&ip6f->ip6f_ro);
390 		if (just_one) {
391 			ip6flow_addstats(ip6f);
392 			return ip6f;
393 		}
394 		ip6flow_inuse--;
395 		ip6flow_addstats(ip6f);
396 		pool_put(&ip6flow_pool, ip6f);
397 	}
398 	return NULL;
399 }
400 
401 void
402 ip6flow_slowtimo(void)
403 {
404 	struct ip6flow *ip6f, *next_ip6f;
405 
406 	mutex_enter(softnet_lock);
407 	KERNEL_LOCK(1, NULL);
408 
409 	for (ip6f = LIST_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
410 		next_ip6f = LIST_NEXT(ip6f, ip6f_list);
411 		if (PRT_SLOW_ISEXPIRED(ip6f->ip6f_timer) ||
412 		    rtcache_validate(&ip6f->ip6f_ro) == NULL) {
413 			ip6flow_free(ip6f);
414 		} else {
415 			ip6f->ip6f_last_uses = ip6f->ip6f_uses;
416 			ip6flow_addstats(ip6f);
417 			ip6f->ip6f_uses = 0;
418 			ip6f->ip6f_dropped = 0;
419 			ip6f->ip6f_forwarded = 0;
420 		}
421 	}
422 
423 	KERNEL_UNLOCK_ONE(NULL);
424 	mutex_exit(softnet_lock);
425 }
426 
427 /*
428  * We have successfully forwarded a packet using the normal
429  * IPv6 stack. Now create/update a flow.
430  */
431 void
432 ip6flow_create(const struct route *ro, struct mbuf *m)
433 {
434 	const struct ip6_hdr *ip6;
435 	struct ip6flow *ip6f;
436 	size_t hash;
437 	int s;
438 
439 	ip6 = mtod(m, const struct ip6_hdr *);
440 
441 	/*
442 	 * If IPv6 Fast Forward is disabled, don't create a flow.
443 	 * It can be disabled by setting net.inet6.ip6.maxflows to 0.
444 	 *
445 	 * Don't create a flow for ICMPv6 messages.
446 	 */
447 	if (ip6_maxflows == 0 || ip6->ip6_nxt == IPPROTO_IPV6_ICMP)
448 		return;
449 
450 	/*
451 	 * See if an existing flow exists.  If so:
452 	 *	- Remove the flow
453 	 *	- Add flow statistics
454 	 *	- Free the route
455 	 *	- Reset statistics
456 	 *
457 	 * If a flow doesn't exist allocate a new one if
458 	 * ip6_maxflows hasn't reached its limit. If it has
459 	 * been reached, reap some flows.
460 	 */
461 	ip6f = ip6flow_lookup(ip6);
462 	if (ip6f == NULL) {
463 		if (ip6flow_inuse >= ip6_maxflows) {
464 			ip6f = ip6flow_reap(1);
465 		} else {
466 			ip6f = pool_get(&ip6flow_pool, PR_NOWAIT);
467 			if (ip6f == NULL)
468 				return;
469 			ip6flow_inuse++;
470 		}
471 		memset(ip6f, 0, sizeof(*ip6f));
472 	} else {
473 		s = splnet();
474 		IP6FLOW_REMOVE(ip6f);
475 		splx(s);
476 		ip6flow_addstats(ip6f);
477 		rtcache_free(&ip6f->ip6f_ro);
478 		ip6f->ip6f_uses = 0;
479 		ip6f->ip6f_last_uses = 0;
480 		ip6f->ip6f_dropped = 0;
481 		ip6f->ip6f_forwarded = 0;
482 	}
483 
484 	/*
485 	 * Fill in the updated/new details.
486 	 */
487 	rtcache_copy(&ip6f->ip6f_ro, ro);
488 	ip6f->ip6f_dst = ip6->ip6_dst;
489 	ip6f->ip6f_src = ip6->ip6_src;
490 	ip6f->ip6f_flow = ip6->ip6_flow;
491 	PRT_SLOW_ARM(ip6f->ip6f_timer, IP6FLOW_TIMER);
492 	ip6f->ip6f_start = time_uptime;
493 
494 	/*
495 	 * Insert into the approriate bucket of the flow table.
496 	 */
497 	hash = ip6flow_hash(ip6);
498 	s = splnet();
499 	IP6FLOW_INSERT(&ip6flowtable[hash], ip6f);
500 	splx(s);
501 }
502 
503 /*
504  * Invalidate/remove all flows - if new_size is positive we
505  * resize the hash table.
506  */
507 int
508 ip6flow_invalidate_all(int new_size)
509 {
510 	struct ip6flow *ip6f, *next_ip6f;
511 	int s, error;
512 
513 	error = 0;
514 	s = splnet();
515 	for (ip6f = LIST_FIRST(&ip6flowlist); ip6f != NULL; ip6f = next_ip6f) {
516 		next_ip6f = LIST_NEXT(ip6f, ip6f_list);
517 		ip6flow_free(ip6f);
518 	}
519 
520 	if (new_size)
521 		error = ip6flow_init(new_size);
522 	splx(s);
523 
524 	return error;
525 }
526