xref: /freebsd/sys/netinet6/in6_rmx.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $
30  */
31 
32 /*-
33  * Copyright 1994, 1995 Massachusetts Institute of Technology
34  *
35  * Permission to use, copy, modify, and distribute this software and
36  * its documentation for any purpose and without fee is hereby
37  * granted, provided that both the above copyright notice and this
38  * permission notice appear in all copies, that both the above
39  * copyright notice and this permission notice appear in all
40  * supporting documentation, and that the name of M.I.T. not be used
41  * in advertising or publicity pertaining to distribution of the
42  * software without specific, written prior permission.  M.I.T. makes
43  * no representations about the suitability of this software for any
44  * purpose.  It is provided "as is" without express or implied
45  * warranty.
46  *
47  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
48  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
49  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
50  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
51  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
52  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
53  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
54  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
55  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
56  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
57  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58  * SUCH DAMAGE.
59  *
60  */
61 
62 /*
63  * This code does two things necessary for the enhanced TCP metrics to
64  * function in a useful manner:
65  *  1) It marks all non-host routes as `cloning', thus ensuring that
66  *     every actual reference to such a route actually gets turned
67  *     into a reference to a host route to the specific destination
68  *     requested.
69  *  2) When such routes lose all their references, it arranges for them
70  *     to be deleted in some random collection of circumstances, so that
71  *     a large quantity of stale routing data is not kept in kernel memory
72  *     indefinitely.  See in6_rtqtimo() below for the exact mechanism.
73  */
74 
75 #include <sys/cdefs.h>
76 __FBSDID("$FreeBSD$");
77 
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/lock.h>
82 #include <sys/sysctl.h>
83 #include <sys/queue.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/mbuf.h>
87 #include <sys/rwlock.h>
88 #include <sys/syslog.h>
89 #include <sys/callout.h>
90 
91 #include <net/if.h>
92 #include <net/route.h>
93 
94 #include <netinet/in.h>
95 #include <netinet/ip_var.h>
96 #include <netinet/in_var.h>
97 
98 #include <netinet/ip6.h>
99 #include <netinet6/ip6_var.h>
100 
101 #include <netinet/icmp6.h>
102 #include <netinet6/nd6.h>
103 
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 
109 extern int	in6_inithead(void **head, int off);
110 #ifdef VIMAGE
111 extern int	in6_detachhead(void **head, int off);
112 #endif
113 
114 #define RTPRF_OURS		RTF_PROTO3	/* set on routes we manage */
115 
116 /*
117  * Do what we need to do when inserting a route.
118  */
119 static struct radix_node *
120 in6_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
121     struct radix_node *treenodes)
122 {
123 	struct rtentry *rt = (struct rtentry *)treenodes;
124 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt);
125 	struct radix_node *ret;
126 
127 	RADIX_NODE_HEAD_WLOCK_ASSERT(head);
128 	if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
129 		rt->rt_flags |= RTF_MULTICAST;
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_ifp)
154 		rt->rt_rmx.rmx_mtu = IN6_LINKMTU(rt->rt_ifp);
155 
156 	ret = rn_addroute(v_arg, n_arg, head, treenodes);
157 	if (ret == NULL) {
158 		struct rtentry *rt2;
159 		/*
160 		 * We are trying to add a net route, but can't.
161 		 * The following case should be allowed, so we'll make a
162 		 * special check for this:
163 		 *	Two IPv6 addresses with the same prefix is assigned
164 		 *	to a single interrface.
165 		 *	# ifconfig if0 inet6 3ffe:0501::1 prefix 64 alias (*1)
166 		 *	# ifconfig if0 inet6 3ffe:0501::2 prefix 64 alias (*2)
167 		 *	In this case, (*1) and (*2) want to add the same
168 		 *	net route entry, 3ffe:0501:: -> if0.
169 		 *	This case should not raise an error.
170 		 */
171 		rt2 = rtalloc1((struct sockaddr *)sin6, 0, RTF_RNH_LOCKED);
172 		if (rt2) {
173 			if (((rt2->rt_flags & (RTF_HOST|RTF_GATEWAY)) == 0)
174 			 && rt2->rt_gateway
175 			 && rt2->rt_gateway->sa_family == AF_LINK
176 			 && rt2->rt_ifp == rt->rt_ifp) {
177 				ret = rt2->rt_nodes;
178 			}
179 			RTFREE_LOCKED(rt2);
180 		}
181 	}
182 	return (ret);
183 }
184 
185 /*
186  * This code is the inverse of in6_clsroute: on first reference, if we
187  * were managing the route, stop doing so and set the expiration timer
188  * back off again.
189  */
190 static struct radix_node *
191 in6_matroute(void *v_arg, struct radix_node_head *head)
192 {
193 	struct radix_node *rn = rn_match(v_arg, head);
194 	struct rtentry *rt = (struct rtentry *)rn;
195 
196 	if (rt) {
197 		RT_LOCK(rt);
198 		if (rt->rt_flags & RTPRF_OURS) {
199 			rt->rt_flags &= ~RTPRF_OURS;
200 			rt->rt_rmx.rmx_expire = 0;
201 		}
202 		RT_UNLOCK(rt);
203 	}
204 	return rn;
205 }
206 
207 SYSCTL_DECL(_net_inet6_ip6);
208 
209 static VNET_DEFINE(int, rtq_reallyold6) = 60*60;
210 	/* one hour is ``really old'' */
211 #define	V_rtq_reallyold6		VNET(rtq_reallyold6)
212 SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
213     &VNET_NAME(rtq_reallyold6) , 0, "");
214 
215 static VNET_DEFINE(int, rtq_minreallyold6) = 10;
216 	/* never automatically crank down to less */
217 #define	V_rtq_minreallyold6		VNET(rtq_minreallyold6)
218 SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
219     &VNET_NAME(rtq_minreallyold6) , 0, "");
220 
221 static VNET_DEFINE(int, rtq_toomany6) = 128;
222 	/* 128 cached routes is ``too many'' */
223 #define	V_rtq_toomany6			VNET(rtq_toomany6)
224 SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
225     &VNET_NAME(rtq_toomany6) , 0, "");
226 
227 struct rtqk_arg {
228 	struct radix_node_head *rnh;
229 	int mode;
230 	int updating;
231 	int draining;
232 	int killed;
233 	int found;
234 	time_t nextstop;
235 };
236 
237 /*
238  * Get rid of old routes.  When draining, this deletes everything, even when
239  * the timeout is not expired yet.  When updating, this makes sure that
240  * nothing has a timeout longer than the current value of rtq_reallyold6.
241  */
242 static int
243 in6_rtqkill(struct radix_node *rn, void *rock)
244 {
245 	struct rtqk_arg *ap = rock;
246 	struct rtentry *rt = (struct rtentry *)rn;
247 	int err;
248 
249 	RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh);
250 
251 	if (rt->rt_flags & RTPRF_OURS) {
252 		ap->found++;
253 
254 		if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
255 			if (rt->rt_refcnt > 0)
256 				panic("rtqkill route really not free");
257 
258 			err = rtrequest(RTM_DELETE,
259 					(struct sockaddr *)rt_key(rt),
260 					rt->rt_gateway, rt_mask(rt),
261 					rt->rt_flags|RTF_RNH_LOCKED, 0);
262 			if (err) {
263 				log(LOG_WARNING, "in6_rtqkill: error %d", err);
264 			} else {
265 				ap->killed++;
266 			}
267 		} else {
268 			if (ap->updating
269 			   && (rt->rt_rmx.rmx_expire - time_uptime
270 			       > V_rtq_reallyold6)) {
271 				rt->rt_rmx.rmx_expire = time_uptime
272 					+ V_rtq_reallyold6;
273 			}
274 			ap->nextstop = lmin(ap->nextstop,
275 					    rt->rt_rmx.rmx_expire);
276 		}
277 	}
278 
279 	return 0;
280 }
281 
282 #define RTQ_TIMEOUT	60*10	/* run no less than once every ten minutes */
283 static VNET_DEFINE(int, rtq_timeout6) = RTQ_TIMEOUT;
284 static VNET_DEFINE(struct callout, rtq_timer6);
285 
286 #define	V_rtq_timeout6			VNET(rtq_timeout6)
287 #define	V_rtq_timer6			VNET(rtq_timer6)
288 
289 static void
290 in6_rtqtimo(void *rock)
291 {
292 	CURVNET_SET_QUIET((struct vnet *) rock);
293 	struct radix_node_head *rnh;
294 	struct rtqk_arg arg;
295 	struct timeval atv;
296 	static time_t last_adjusted_timeout = 0;
297 
298 	rnh = rt_tables_get_rnh(0, AF_INET6);
299 	if (rnh == NULL) {
300 		CURVNET_RESTORE();
301 		return;
302 	}
303 	arg.found = arg.killed = 0;
304 	arg.rnh = rnh;
305 	arg.nextstop = time_uptime + V_rtq_timeout6;
306 	arg.draining = arg.updating = 0;
307 	RADIX_NODE_HEAD_LOCK(rnh);
308 	rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
309 	RADIX_NODE_HEAD_UNLOCK(rnh);
310 
311 	/*
312 	 * Attempt to be somewhat dynamic about this:
313 	 * If there are ``too many'' routes sitting around taking up space,
314 	 * then crank down the timeout, and see if we can't make some more
315 	 * go away.  However, we make sure that we will never adjust more
316 	 * than once in rtq_timeout6 seconds, to keep from cranking down too
317 	 * hard.
318 	 */
319 	if ((arg.found - arg.killed > V_rtq_toomany6)
320 	   && (time_uptime - last_adjusted_timeout >= V_rtq_timeout6)
321 	   && V_rtq_reallyold6 > V_rtq_minreallyold6) {
322 		V_rtq_reallyold6 = 2*V_rtq_reallyold6 / 3;
323 		if (V_rtq_reallyold6 < V_rtq_minreallyold6) {
324 			V_rtq_reallyold6 = V_rtq_minreallyold6;
325 		}
326 
327 		last_adjusted_timeout = time_uptime;
328 #ifdef DIAGNOSTIC
329 		log(LOG_DEBUG, "in6_rtqtimo: adjusted rtq_reallyold6 to %d",
330 		    V_rtq_reallyold6);
331 #endif
332 		arg.found = arg.killed = 0;
333 		arg.updating = 1;
334 		RADIX_NODE_HEAD_LOCK(rnh);
335 		rnh->rnh_walktree(rnh, in6_rtqkill, &arg);
336 		RADIX_NODE_HEAD_UNLOCK(rnh);
337 	}
338 
339 	atv.tv_usec = 0;
340 	atv.tv_sec = arg.nextstop - time_uptime;
341 	callout_reset(&V_rtq_timer6, tvtohz(&atv), in6_rtqtimo, rock);
342 	CURVNET_RESTORE();
343 }
344 
345 /*
346  * Age old PMTUs.
347  */
348 struct mtuex_arg {
349 	struct radix_node_head *rnh;
350 	time_t nextstop;
351 };
352 static VNET_DEFINE(struct callout, rtq_mtutimer);
353 #define	V_rtq_mtutimer			VNET(rtq_mtutimer)
354 
355 static int
356 in6_mtuexpire(struct radix_node *rn, void *rock)
357 {
358 	struct rtentry *rt = (struct rtentry *)rn;
359 	struct mtuex_arg *ap = rock;
360 
361 	/* sanity */
362 	if (!rt)
363 		panic("rt == NULL in in6_mtuexpire");
364 
365 	if (rt->rt_rmx.rmx_expire && !(rt->rt_flags & RTF_PROBEMTU)) {
366 		if (rt->rt_rmx.rmx_expire <= time_uptime) {
367 			rt->rt_flags |= RTF_PROBEMTU;
368 		} else {
369 			ap->nextstop = lmin(ap->nextstop,
370 					rt->rt_rmx.rmx_expire);
371 		}
372 	}
373 
374 	return 0;
375 }
376 
377 #define	MTUTIMO_DEFAULT	(60*1)
378 
379 static void
380 in6_mtutimo(void *rock)
381 {
382 	CURVNET_SET_QUIET((struct vnet *) rock);
383 	struct radix_node_head *rnh;
384 	struct mtuex_arg arg;
385 	struct timeval atv;
386 
387 	rnh = rt_tables_get_rnh(0, AF_INET6);
388 	if (rnh == NULL) {
389 		CURVNET_RESTORE();
390 		return;
391 	}
392 	arg.rnh = rnh;
393 	arg.nextstop = time_uptime + MTUTIMO_DEFAULT;
394 	RADIX_NODE_HEAD_LOCK(rnh);
395 	rnh->rnh_walktree(rnh, in6_mtuexpire, &arg);
396 	RADIX_NODE_HEAD_UNLOCK(rnh);
397 
398 	atv.tv_usec = 0;
399 	atv.tv_sec = arg.nextstop - time_uptime;
400 	if (atv.tv_sec < 0) {
401 		printf("invalid mtu expiration time on routing table\n");
402 		arg.nextstop = time_uptime + 30;	/* last resort */
403 		atv.tv_sec = 30;
404 	}
405 	callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock);
406 	CURVNET_RESTORE();
407 }
408 
409 /*
410  * Initialize our routing tree.
411  * XXX MRT When off == 0, we are being called from vfs_export.c
412  * so just set up their table and leave. (we know what the correct
413  * value should be so just use that).. FIX AFTER RELENG_7 is MFC'd
414  * see also comments in in_inithead() vfs_export.c and domain.h
415  */
416 int
417 in6_inithead(void **head, int off)
418 {
419 	struct radix_node_head *rnh;
420 
421 	if (!rn_inithead(head, offsetof(struct sockaddr_in6, sin6_addr) << 3))
422 		return 0;		/* See above */
423 
424 	if (off == 0)		/* See above */
425 		return 1;	/* only do the rest for the real thing */
426 
427 	rnh = *head;
428 	KASSERT(rnh == rt_tables_get_rnh(0, AF_INET6), ("rnh?"));
429 	rnh->rnh_addaddr = in6_addroute;
430 	rnh->rnh_matchaddr = in6_matroute;
431 	callout_init(&V_rtq_timer6, CALLOUT_MPSAFE);
432 	callout_init(&V_rtq_mtutimer, CALLOUT_MPSAFE);
433 	in6_rtqtimo(curvnet);	/* kick off timeout first time */
434 	in6_mtutimo(curvnet);	/* kick off timeout first time */
435 	return 1;
436 }
437 
438 #ifdef VIMAGE
439 int
440 in6_detachhead(void **head, int off)
441 {
442 
443 	callout_drain(&V_rtq_timer6);
444 	callout_drain(&V_rtq_mtutimer);
445 	return (1);
446 }
447 #endif
448