xref: /dflybsd-src/sys/netinet6/in6_rmx.c (revision 1f8e62c9ab8d2ecdefffbdf2ef5ed3db7376c6ec)
1 /*	$FreeBSD: src/sys/netinet6/in6_rmx.c,v 1.1.2.4 2004/10/06 02:35:17 suz Exp $	*/
2 /*	$DragonFly: src/sys/netinet6/in6_rmx.c,v 1.11 2005/01/06 17:59:32 hsu Exp $	*/
3 /*	$KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $	*/
4 
5 /*
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
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  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright 1994, 1995 Massachusetts Institute of Technology
36  *
37  * Permission to use, copy, modify, and distribute this software and
38  * its documentation for any purpose and without fee is hereby
39  * granted, provided that both the above copyright notice and this
40  * permission notice appear in all copies, that both the above
41  * copyright notice and this permission notice appear in all
42  * supporting documentation, and that the name of M.I.T. not be used
43  * in advertising or publicity pertaining to distribution of the
44  * software without specific, written prior permission.  M.I.T. makes
45  * no representations about the suitability of this software for any
46  * purpose.  It is provided "as is" without express or implied
47  * warranty.
48  *
49  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
50  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
51  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
52  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
53  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
54  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
55  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
56  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
57  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
58  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
59  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  */
63 
64 /*
65  * This code does two things necessary for the enhanced TCP metrics to
66  * function in a useful manner:
67  *  1) It marks all non-host routes as `cloning', thus ensuring that
68  *     every actual reference to such a route actually gets turned
69  *     into a reference to a host route to the specific destination
70  *     requested.
71  *  2) When such routes lose all their references, it arranges for them
72  *     to be deleted in some random collection of circumstances, so that
73  *     a large quantity of stale routing data is not kept in kernel memory
74  *     indefinitely.  See in6_rtqtimo() below for the exact mechanism.
75  */
76 
77 #include <sys/param.h>
78 #include <sys/systm.h>
79 #include <sys/kernel.h>
80 #include <sys/sysctl.h>
81 #include <sys/queue.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/mbuf.h>
85 #include <sys/syslog.h>
86 
87 #include <net/if.h>
88 #include <net/route.h>
89 #include <netinet/in.h>
90 #include <netinet/ip_var.h>
91 #include <netinet/in_var.h>
92 
93 #include <netinet/ip6.h>
94 #include <netinet6/ip6_var.h>
95 
96 #include <netinet/icmp6.h>
97 
98 #include <netinet/tcp.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 
103 static struct callout	in6_rtqtimo_ch;
104 static struct callout	in6_mtutimo_ch;
105 
106 extern int	in6_inithead (void **head, int off);
107 
108 #define RTPRF_OURS		RTF_PROTO3	/* set on routes we manage */
109 
110 /*
111  * Do what we need to do when inserting a route.
112  */
113 static struct radix_node *
114 in6_addroute(char *key, char *mask, struct radix_node_head *head,
115 	     struct radix_node *treenodes)
116 {
117 	struct rtentry *rt = (struct rtentry *)treenodes;
118 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
119 	struct radix_node *ret;
120 
121 	/*
122 	 * For IPv6, all unicast non-host routes are automatically cloning.
123 	 */
124 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
125 		rt->rt_flags |= RTF_MULTICAST;
126 
127 	if (!(rt->rt_flags & (RTF_HOST | RTF_CLONING | RTF_MULTICAST))) {
128 		rt->rt_flags |= RTF_PRCLONING;
129 	}
130 
131 	/*
132 	 * A little bit of help for both IPv6 output and input:
133 	 *   For local addresses, we make sure that RTF_LOCAL is set,
134 	 *   with the thought that this might one day be used to speed up
135 	 *   ip_input().
136 	 *
137 	 * We also mark routes to multicast addresses as such, because
138 	 * it's easy to do and might be useful (but this is much more
139 	 * dubious since it's so easy to inspect the address).  (This
140 	 * is done above.)
141 	 *
142 	 * XXX
143 	 * should elaborate the code.
144 	 */
145 	if (rt->rt_flags & RTF_HOST) {
146 		if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)
147 					->sin6_addr,
148 				       &sin6->sin6_addr)) {
149 			rt->rt_flags |= RTF_LOCAL;
150 		}
151 	}
152 
153 	if (!rt->rt_rmx.rmx_mtu && !(rt->rt_rmx.rmx_locks & RTV_MTU) &&
154 	    rt->rt_ifp != NULL)
155 		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
156 
157 	ret = rn_addroute(key, mask, head, treenodes);
158 	if (ret == NULL && rt->rt_flags & RTF_HOST) {
159 		struct rtentry *rt2;
160 
161 		/*
162 		 * We are trying to add a host route, but can't.
163 		 * Find out if it is because of an
164 		 * ARP entry and delete it if so.
165 		 */
166 		rt2 = rtpurelookup((struct sockaddr *)sin6);
167 		if (rt2 != NULL) {
168 			--rt2->rt_refcnt;
169 			if (rt2->rt_flags & RTF_LLINFO &&
170 			    rt2->rt_flags & RTF_HOST &&
171 			    rt2->rt_gateway &&
172 			    rt2->rt_gateway->sa_family == AF_LINK) {
173 				rtrequest(RTM_DELETE, rt_key(rt2),
174 					  rt2->rt_gateway, rt_mask(rt2),
175 					  rt2->rt_flags, NULL);
176 				ret = rn_addroute(key, mask, head, treenodes);
177 			}
178 		}
179 	} else if (ret == NULL && rt->rt_flags & RTF_CLONING) {
180 		struct rtentry *rt2;
181 
182 		/*
183 		 * We are trying to add a net route, but can't.
184 		 * The following case should be allowed, so we'll make a
185 		 * special check for this:
186 		 *	Two IPv6 addresses with the same prefix is assigned
187 		 *	to a single interrface.
188 		 *	# ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
189 		 *	# ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
190 		 *	In this case, (*1) and (*2) want to add the same
191 		 *	net route entry, 3ffe:0501:: -> if0.
192 		 *	This case should not raise an error.
193 		 */
194 		rt2 = rtpurelookup((struct sockaddr *)sin6);
195 		if (rt2 != NULL) {
196 			if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY))
197 					== RTF_CLONING &&
198 			    rt2->rt_gateway &&
199 			    rt2->rt_gateway->sa_family == AF_LINK &&
200 			    rt2->rt_ifp == rt->rt_ifp) {
201 				ret = rt2->rt_nodes;
202 			}
203 			--rt2->rt_refcnt;
204 		}
205 	}
206 	return ret;
207 }
208 
209 /*
210  * This code is the inverse of in6_clsroute: on first reference, if we
211  * were managing the route, stop doing so and set the expiration timer
212  * back off again.
213  */
214 static struct radix_node *
215 in6_matchroute(char *key, struct radix_node_head *head)
216 {
217 	struct radix_node *rn = rn_match(key, head);
218 	struct rtentry *rt = (struct rtentry *)rn;
219 
220 	if (rt != NULL && rt->rt_refcnt == 0) { /* this is first reference */
221 		if (rt->rt_flags & RTPRF_OURS) {
222 			rt->rt_flags &= ~RTPRF_OURS;
223 			rt->rt_rmx.rmx_expire = 0;
224 		}
225 	}
226 	return rn;
227 }
228 
229 SYSCTL_DECL(_net_inet6_ip6);
230 
231 static int rtq_reallyold = 60*60;
232 	/* one hour is ``really old'' */
233 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire,
234 	CTLFLAG_RW, &rtq_reallyold , 0, "");
235 
236 static int rtq_minreallyold = 10;
237 	/* never automatically crank down to less */
238 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire,
239 	CTLFLAG_RW, &rtq_minreallyold , 0, "");
240 
241 static int rtq_toomany = 128;
242 	/* 128 cached routes is ``too many'' */
243 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache,
244 	CTLFLAG_RW, &rtq_toomany , 0, "");
245 
246 
247 /*
248  * On last reference drop, mark the route as belong to us so that it can be
249  * timed out.
250  */
251 static void
252 in6_clsroute(struct radix_node *rn, struct radix_node_head *head)
253 {
254 	struct rtentry *rt = (struct rtentry *)rn;
255 
256 	if (!(rt->rt_flags & RTF_UP))
257 		return;		/* prophylactic measures */
258 
259 	if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
260 		return;
261 
262 	if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED)
263 		return;
264 
265 	/*
266 	 * As requested by David Greenman:
267 	 * If rtq_reallyold is 0, just delete the route without
268 	 * waiting for a timeout cycle to kill it.
269 	 */
270 	if (rtq_reallyold != 0) {
271 		rt->rt_flags |= RTPRF_OURS;
272 		rt->rt_rmx.rmx_expire = time_second + rtq_reallyold;
273 	} else {
274 		/*
275 		 * Remove route from the radix tree, but defer deallocation
276 		 * until we return to rtfree().
277 		 */
278 		rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway, rt_mask(rt),
279 			  rt->rt_flags, &rt);
280 	}
281 }
282 
283 struct rtqk_arg {
284 	struct radix_node_head *rnh;
285 	int mode;
286 	int updating;
287 	int draining;
288 	int killed;
289 	int found;
290 	time_t nextstop;
291 };
292 
293 /*
294  * Get rid of old routes.  When draining, this deletes everything, even when
295  * the timeout is not expired yet.  When updating, this makes sure that
296  * nothing has a timeout longer than the current value of rtq_reallyold.
297  */
298 static int
299 in6_rtqkill(struct radix_node *rn, void *rock)
300 {
301 	struct rtqk_arg *ap = rock;
302 	struct rtentry *rt = (struct rtentry *)rn;
303 	int err;
304 
305 	if (rt->rt_flags & RTPRF_OURS) {
306 		ap->found++;
307 
308 		if (ap->draining || rt->rt_rmx.rmx_expire <= time_second) {
309 			if (rt->rt_refcnt > 0)
310 				panic("rtqkill route really not free");
311 
312 			err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
313 					rt_mask(rt), rt->rt_flags, NULL);
314 			if (err) {
315 				log(LOG_WARNING, "in6_rtqkill: error %d", err);
316 			} else {
317 				ap->killed++;
318 			}
319 		} else {
320 			if (ap->updating
321 			   && (rt->rt_rmx.rmx_expire - time_second
322 			       > rtq_reallyold)) {
323 				rt->rt_rmx.rmx_expire = time_second
324 					+ rtq_reallyold;
325 			}
326 			ap->nextstop = lmin(ap->nextstop,
327 					    rt->rt_rmx.rmx_expire);
328 		}
329 	}
330 
331 	return 0;
332 }
333 
334 #define RTQ_TIMEOUT	60*10	/* run no less than once every ten minutes */
335 static int rtq_timeout = RTQ_TIMEOUT;
336 
337 static void
338 in6_rtqtimo(void *rock)
339 {
340 	struct radix_node_head *rnh = rock;
341 	struct rtqk_arg arg;
342 	struct timeval atv;
343 	static time_t last_adjusted_timeout = 0;
344 	int s;
345 
346 	arg.found = arg.killed = 0;
347 	arg.rnh = rnh;
348 	arg.nextstop = time_second + rtq_timeout;
349 	arg.draining = arg.updating = 0;
350 	s = splnet();
351 	rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
352 	splx(s);
353 
354 	/*
355 	 * Attempt to be somewhat dynamic about this:
356 	 * If there are ``too many'' routes sitting around taking up space,
357 	 * then crank down the timeout, and see if we can't make some more
358 	 * go away.  However, we make sure that we will never adjust more
359 	 * than once in rtq_timeout seconds, to keep from cranking down too
360 	 * hard.
361 	 */
362 	if ((arg.found - arg.killed > rtq_toomany)
363 	   && (time_second - last_adjusted_timeout >= rtq_timeout)
364 	   && rtq_reallyold > rtq_minreallyold) {
365 		rtq_reallyold = 2*rtq_reallyold / 3;
366 		if (rtq_reallyold < rtq_minreallyold) {
367 			rtq_reallyold = rtq_minreallyold;
368 		}
369 
370 		last_adjusted_timeout = time_second;
371 #ifdef DIAGNOSTIC
372 		log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d",
373 		    rtq_reallyold);
374 #endif
375 		arg.found = arg.killed = 0;
376 		arg.updating = 1;
377 		s = splnet();
378 		rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
379 		splx(s);
380 	}
381 
382 	atv.tv_usec = 0;
383 	atv.tv_sec = arg.nextstop;
384 	callout_reset(&in6_rtqtimo_ch, tvtohz_high(&atv), in6_rtqtimo, rock);
385 }
386 
387 /*
388  * Age old PMTUs.
389  */
390 struct mtuex_arg {
391 	struct radix_node_head *rnh;
392 	time_t nextstop;
393 };
394 
395 static int
396 in6_mtuexpire(struct radix_node *rn, void *rock)
397 {
398 	struct rtentry *rt = (struct rtentry *)rn;
399 	struct mtuex_arg *ap = rock;
400 
401 	/* sanity */
402 	if (!rt)
403 		panic("rt == NULL in in6_mtuexpire");
404 
405 	if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
406 		if (rt->rt_rmx.rmx_expire <= time_second) {
407 			rt->rt_flags |= RTF_PROBEMTU;
408 		} else {
409 			ap->nextstop = lmin(ap->nextstop,
410 					rt->rt_rmx.rmx_expire);
411 		}
412 	}
413 
414 	return 0;
415 }
416 
417 #define	MTUTIMO_DEFAULT	(60*1)
418 
419 static void
420 in6_mtutimo(void *rock)
421 {
422 	struct radix_node_head *rnh = rock;
423 	struct mtuex_arg arg;
424 	struct timeval atv;
425 	int s;
426 
427 	arg.rnh = rnh;
428 	arg.nextstop = time_second + MTUTIMO_DEFAULT;
429 	s = splnet();
430 	rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
431 	splx(s);
432 
433 	atv.tv_usec = 0;
434 	atv.tv_sec = arg.nextstop;
435 	if (atv.tv_sec < time_second) {
436 		printf("invalid mtu expiration time on routing table\n");
437 		arg.nextstop = time_second + 30;	/* last resort */
438 	}
439 	callout_reset(&in6_mtutimo_ch, tvtohz_high(&atv), in6_mtutimo, rock);
440 }
441 
442 #if 0
443 void
444 in6_rtqdrain(void)
445 {
446 	struct radix_node_head *rnh = rt_tables[AF_INET6];
447 	struct rtqk_arg arg;
448 	int s;
449 	arg.found = arg.killed = 0;
450 	arg.rnh = rnh;
451 	arg.nextstop = 0;
452 	arg.draining = 1;
453 	arg.updating = 0;
454 	s = splnet();
455 	rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
456 	splx(s);
457 }
458 #endif
459 
460 /*
461  * Initialize our routing tree.
462  */
463 int
464 in6_inithead(void **head, int off)
465 {
466 	struct radix_node_head *rnh;
467 
468 	if (!rn_inithead(head, off))
469 		return 0;
470 
471 	if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */
472 		return 1;	/* only do this for the real routing table */
473 
474 	rnh = *head;
475 	rnh->rnh_addaddr = in6_addroute;
476 	rnh->rnh_matchaddr = in6_matchroute;
477 	rnh->rnh_close = in6_clsroute;
478 	callout_init(&in6_mtutimo_ch);
479 	callout_init(&in6_rtqtimo_ch);
480 	in6_rtqtimo(rnh);	/* kick off timeout first time */
481 	in6_mtutimo(rnh);	/* kick off timeout first time */
482 	return 1;
483 }
484