xref: /freebsd/sys/netinet6/in6_rmx.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
1 /*	$FreeBSD$	*/
2 /*	$KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $	*/
3 
4 /*-
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*-
34  * Copyright 1994, 1995 Massachusetts Institute of Technology
35  *
36  * Permission to use, copy, modify, and distribute this software and
37  * its documentation for any purpose and without fee is hereby
38  * granted, provided that both the above copyright notice and this
39  * permission notice appear in all copies, that both the above
40  * copyright notice and this permission notice appear in all
41  * supporting documentation, and that the name of M.I.T. not be used
42  * in advertising or publicity pertaining to distribution of the
43  * software without specific, written prior permission.  M.I.T. makes
44  * no representations about the suitability of this software for any
45  * purpose.  It is provided "as is" without express or implied
46  * warranty.
47  *
48  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
49  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
50  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
51  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
52  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
54  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
55  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
56  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
57  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
58  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  */
62 
63 /*
64  * This code does two things necessary for the enhanced TCP metrics to
65  * function in a useful manner:
66  *  1) It marks all non-host routes as `cloning', thus ensuring that
67  *     every actual reference to such a route actually gets turned
68  *     into a reference to a host route to the specific destination
69  *     requested.
70  *  2) When such routes lose all their references, it arranges for them
71  *     to be deleted in some random collection of circumstances, so that
72  *     a large quantity of stale routing data is not kept in kernel memory
73  *     indefinitely.  See in6_rtqtimo() below for the exact mechanism.
74  */
75 
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/queue.h>
81 #include <sys/socket.h>
82 #include <sys/socketvar.h>
83 #include <sys/mbuf.h>
84 #include <sys/syslog.h>
85 #include <sys/callout.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 #include <netinet6/nd6.h>
98 
99 #include <netinet/tcp.h>
100 #include <netinet/tcp_seq.h>
101 #include <netinet/tcp_timer.h>
102 #include <netinet/tcp_var.h>
103 
104 extern int	in6_inithead __P((void **head, int off));
105 
106 #define RTPRF_OURS		RTF_PROTO3	/* set on routes we manage */
107 
108 /*
109  * Do what we need to do when inserting a route.
110  */
111 static struct radix_node *
112 in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
113 	    struct radix_node *treenodes)
114 {
115 	struct rtentry *rt = (struct rtentry *)treenodes;
116 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
117 	struct radix_node *ret;
118 
119 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
120 		rt->rt_flags |= RTF_MULTICAST;
121 
122 	/*
123 	 * A little bit of help for both IPv6 output and input:
124 	 *   For local addresses, we make sure that RTF_LOCAL is set,
125 	 *   with the thought that this might one day be used to speed up
126 	 *   ip_input().
127 	 *
128 	 * We also mark routes to multicast addresses as such, because
129 	 * it's easy to do and might be useful (but this is much more
130 	 * dubious since it's so easy to inspect the address).  (This
131 	 * is done above.)
132 	 *
133 	 * XXX
134 	 * should elaborate the code.
135 	 */
136 	if (rt->rt_flags & RTF_HOST) {
137 		if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr)
138 					->sin6_addr,
139 				       &sin6->sin6_addr)) {
140 			rt->rt_flags |= RTF_LOCAL;
141 		}
142 	}
143 
144 	if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
145 		rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp);
146 
147 	ret = rn_addroute(v_arg, n_arg, head, treenodes);
148 	if (ret == NULL && rt->rt_flags & RTF_HOST) {
149 		struct rtentry *rt2;
150 		/*
151 		 * We are trying to add a host route, but can't.
152 		 * Find out if it is because of an
153 		 * ARP entry and delete it if so.
154 		 */
155 		rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_CLONING);
156 		if (rt2) {
157 			if (rt2->rt_flags & RTF_LLINFO &&
158 				rt2->rt_flags & RTF_HOST &&
159 				rt2->rt_gateway &&
160 				rt2->rt_gateway->sa_family == AF_LINK) {
161 				rtexpunge(rt2);
162 				RTFREE_LOCKED(rt2);
163 				ret = rn_addroute(v_arg, n_arg, head,
164 					treenodes);
165 			} else
166 				RTFREE_LOCKED(rt2);
167 		}
168 	} else if (ret == NULL && rt->rt_flags & RTF_CLONING) {
169 		struct rtentry *rt2;
170 		/*
171 		 * We are trying to add a net route, but can't.
172 		 * The following case should be allowed, so we'll make a
173 		 * special check for this:
174 		 *	Two IPv6 addresses with the same prefix is assigned
175 		 *	to a single interrface.
176 		 *	# ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
177 		 *	# ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
178 		 *	In this case, (*1) and (*2) want to add the same
179 		 *	net route entry, 3ffe:0501:: -> if0.
180 		 *	This case should not raise an error.
181 		 */
182 		rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_CLONING);
183 		if (rt2) {
184 			if ((rt2->rt_flags & (RTF_CLONING|RTF_HOST|RTF_GATEWAY))
185 					== RTF_CLONING
186 			 && rt2->rt_gateway
187 			 && rt2->rt_gateway->sa_family == AF_LINK
188 			 && rt2->rt_ifp == rt->rt_ifp) {
189 				ret = rt2->rt_nodes;
190 			}
191 			RTFREE_LOCKED(rt2);
192 		}
193 	}
194 	return ret;
195 }
196 
197 /*
198  * This code is the inverse of in6_clsroute: on first reference, if we
199  * were managing the route, stop doing so and set the expiration timer
200  * back off again.
201  */
202 static struct radix_node *
203 in6_matroute(void *v_arg, struct radix_node_head *head)
204 {
205 	struct radix_node *rn = rn_match(v_arg, head);
206 	struct rtentry *rt = (struct rtentry *)rn;
207 
208 	if (rt && rt->rt_refcnt == 0) { /* this is first reference */
209 		if (rt->rt_flags & RTPRF_OURS) {
210 			rt->rt_flags &= ~RTPRF_OURS;
211 			rt->rt_rmx.rmx_expire = 0;
212 		}
213 	}
214 	return rn;
215 }
216 
217 SYSCTL_DECL(_net_inet6_ip6);
218 
219 static int rtq_reallyold = 60*60;
220 	/* one hour is ``really old'' */
221 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire,
222 	CTLFLAG_RW, &rtq_reallyold , 0, "");
223 
224 static int rtq_minreallyold = 10;
225 	/* never automatically crank down to less */
226 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire,
227 	CTLFLAG_RW, &rtq_minreallyold , 0, "");
228 
229 static int rtq_toomany = 128;
230 	/* 128 cached routes is ``too many'' */
231 SYSCTL_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache,
232 	CTLFLAG_RW, &rtq_toomany , 0, "");
233 
234 
235 /*
236  * On last reference drop, mark the route as belong to us so that it can be
237  * timed out.
238  */
239 static void
240 in6_clsroute(struct radix_node *rn, struct radix_node_head *head)
241 {
242 	struct rtentry *rt = (struct rtentry *)rn;
243 
244 	RT_LOCK_ASSERT(rt);
245 
246 	if (!(rt->rt_flags & RTF_UP))
247 		return;		/* prophylactic measures */
248 
249 	if ((rt->rt_flags & (RTF_LLINFO | RTF_HOST)) != RTF_HOST)
250 		return;
251 
252 	if ((rt->rt_flags & (RTF_WASCLONED | RTPRF_OURS)) != RTF_WASCLONED)
253 		return;
254 
255 	/*
256 	 * As requested by David Greenman:
257 	 * If rtq_reallyold is 0, just delete the route without
258 	 * waiting for a timeout cycle to kill it.
259 	 */
260 	if (rtq_reallyold != 0) {
261 		rt->rt_flags |= RTPRF_OURS;
262 		rt->rt_rmx.rmx_expire = time_uptime + rtq_reallyold;
263 	} else {
264 		rtexpunge(rt);
265 	}
266 }
267 
268 struct rtqk_arg {
269 	struct radix_node_head *rnh;
270 	int mode;
271 	int updating;
272 	int draining;
273 	int killed;
274 	int found;
275 	time_t nextstop;
276 };
277 
278 /*
279  * Get rid of old routes.  When draining, this deletes everything, even when
280  * the timeout is not expired yet.  When updating, this makes sure that
281  * nothing has a timeout longer than the current value of rtq_reallyold.
282  */
283 static int
284 in6_rtqkill(struct radix_node *rn, void *rock)
285 {
286 	struct rtqk_arg *ap = rock;
287 	struct rtentry *rt = (struct rtentry *)rn;
288 	int err;
289 
290 	if (rt->rt_flags & RTPRF_OURS) {
291 		ap->found++;
292 
293 		if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
294 			if (rt->rt_refcnt > 0)
295 				panic("rtqkill route really not free");
296 
297 			err = rtrequest(RTM_DELETE,
298 					(struct sockaddr *)rt_key(rt),
299 					rt->rt_gateway, rt_mask(rt),
300 					rt->rt_flags, 0);
301 			if (err) {
302 				log(LOG_WARNING, "in6_rtqkill: error %d", err);
303 			} else {
304 				ap->killed++;
305 			}
306 		} else {
307 			if (ap->updating
308 			   && (rt->rt_rmx.rmx_expire - time_uptime
309 			       > rtq_reallyold)) {
310 				rt->rt_rmx.rmx_expire = time_uptime
311 					+ rtq_reallyold;
312 			}
313 			ap->nextstop = lmin(ap->nextstop,
314 					    rt->rt_rmx.rmx_expire);
315 		}
316 	}
317 
318 	return 0;
319 }
320 
321 #define RTQ_TIMEOUT	60*10	/* run no less than once every ten minutes */
322 static int rtq_timeout = RTQ_TIMEOUT;
323 static struct callout rtq_timer;
324 
325 static void
326 in6_rtqtimo(void *rock)
327 {
328 	struct radix_node_head *rnh = rock;
329 	struct rtqk_arg arg;
330 	struct timeval atv;
331 	static time_t last_adjusted_timeout = 0;
332 
333 	arg.found = arg.killed = 0;
334 	arg.rnh = rnh;
335 	arg.nextstop = time_uptime + rtq_timeout;
336 	arg.draining = arg.updating = 0;
337 	RADIX_NODE_HEAD_LOCK(rnh);
338 	rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
339 	RADIX_NODE_HEAD_UNLOCK(rnh);
340 
341 	/*
342 	 * Attempt to be somewhat dynamic about this:
343 	 * If there are ``too many'' routes sitting around taking up space,
344 	 * then crank down the timeout, and see if we can't make some more
345 	 * go away.  However, we make sure that we will never adjust more
346 	 * than once in rtq_timeout seconds, to keep from cranking down too
347 	 * hard.
348 	 */
349 	if ((arg.found - arg.killed > rtq_toomany)
350 	   && (time_uptime - last_adjusted_timeout >= rtq_timeout)
351 	   && rtq_reallyold > rtq_minreallyold) {
352 		rtq_reallyold = 2*rtq_reallyold / 3;
353 		if (rtq_reallyold < rtq_minreallyold) {
354 			rtq_reallyold = rtq_minreallyold;
355 		}
356 
357 		last_adjusted_timeout = time_uptime;
358 #ifdef DIAGNOSTIC
359 		log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold to %d",
360 		    rtq_reallyold);
361 #endif
362 		arg.found = arg.killed = 0;
363 		arg.updating = 1;
364 		RADIX_NODE_HEAD_LOCK(rnh);
365 		rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
366 		RADIX_NODE_HEAD_UNLOCK(rnh);
367 	}
368 
369 	atv.tv_usec = 0;
370 	atv.tv_sec = arg.nextstop - time_uptime;
371 	callout_reset(&rtq_timer, tvtohz(&atv), in6_rtqtimo, rock);
372 }
373 
374 /*
375  * Age old PMTUs.
376  */
377 struct mtuex_arg {
378 	struct radix_node_head *rnh;
379 	time_t nextstop;
380 };
381 static struct callout rtq_mtutimer;
382 
383 static int
384 in6_mtuexpire(struct radix_node *rn, void *rock)
385 {
386 	struct rtentry *rt = (struct rtentry *)rn;
387 	struct mtuex_arg *ap = rock;
388 
389 	/* sanity */
390 	if (!rt)
391 		panic("rt == NULL in in6_mtuexpire");
392 
393 	if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
394 		if (rt->rt_rmx.rmx_expire <= time_uptime) {
395 			rt->rt_flags |= RTF_PROBEMTU;
396 		} else {
397 			ap->nextstop = lmin(ap->nextstop,
398 					rt->rt_rmx.rmx_expire);
399 		}
400 	}
401 
402 	return 0;
403 }
404 
405 #define	MTUTIMO_DEFAULT	(60*1)
406 
407 static void
408 in6_mtutimo(void *rock)
409 {
410 	struct radix_node_head *rnh = rock;
411 	struct mtuex_arg arg;
412 	struct timeval atv;
413 
414 	arg.rnh = rnh;
415 	arg.nextstop = time_uptime + MTUTIMO_DEFAULT;
416 	RADIX_NODE_HEAD_LOCK(rnh);
417 	rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
418 	RADIX_NODE_HEAD_UNLOCK(rnh);
419 
420 	atv.tv_usec = 0;
421 	atv.tv_sec = arg.nextstop - time_uptime;
422 	if (atv.tv_sec < 0) {
423 		printf("invalid mtu expiration time on routing table\n");
424 		arg.nextstop = time_uptime + 30;	/* last resort */
425 		atv.tv_sec = 30;
426 	}
427 	callout_reset(&rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock);
428 }
429 
430 #if 0
431 void
432 in6_rtqdrain()
433 {
434 	struct radix_node_head *rnh = rt_tables[AF_INET6];
435 	struct rtqk_arg arg;
436 
437 	arg.found = arg.killed = 0;
438 	arg.rnh = rnh;
439 	arg.nextstop = 0;
440 	arg.draining = 1;
441 	arg.updating = 0;
442 	RADIX_NODE_HEAD_LOCK(rnh);
443 	rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
444 	RADIX_NODE_HEAD_UNLOCK(rnh);
445 }
446 #endif
447 
448 /*
449  * Initialize our routing tree.
450  */
451 int
452 in6_inithead(void **head, int off)
453 {
454 	struct radix_node_head *rnh;
455 
456 	if (!rn_inithead(head, off))
457 		return 0;
458 
459 	if (head != (void **)&rt_tables[AF_INET6]) /* BOGUS! */
460 		return 1;	/* only do this for the real routing table */
461 
462 	rnh = *head;
463 	rnh->rnh_addaddr = in6_addroute;
464 	rnh->rnh_matchaddr = in6_matroute;
465 	rnh->rnh_close = in6_clsroute;
466 	callout_init(&rtq_timer, CALLOUT_MPSAFE);
467 	in6_rtqtimo(rnh);	/* kick off timeout first time */
468 	callout_init(&rtq_mtutimer, CALLOUT_MPSAFE);
469 	in6_mtutimo(rnh);	/* kick off timeout first time */
470 	return 1;
471 }
472