xref: /freebsd/sys/netinet/in_rmx.c (revision 13de33a5dc2304b13d595d75d48c51793958474f)
1 /*-
2  * Copyright 1994, 1995 Massachusetts Institute of Technology
3  *
4  * Permission to use, copy, modify, and distribute this software and
5  * its documentation for any purpose and without fee is hereby
6  * granted, provided that both the above copyright notice and this
7  * permission notice appear in all copies, that both the above
8  * copyright notice and this permission notice appear in all
9  * supporting documentation, and that the name of M.I.T. not be used
10  * in advertising or publicity pertaining to distribution of the
11  * software without specific, written prior permission.  M.I.T. makes
12  * no representations about the suitability of this software for any
13  * purpose.  It is provided "as is" without express or implied
14  * warranty.
15  *
16  * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''.  M.I.T. DISCLAIMS
17  * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE,
18  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
19  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT
20  * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
23  * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
25  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
26  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sysctl.h>
37 #include <sys/socket.h>
38 #include <sys/mbuf.h>
39 #include <sys/syslog.h>
40 #include <sys/callout.h>
41 
42 #include <net/if.h>
43 #include <net/if_var.h>
44 #include <net/route.h>
45 #include <net/vnet.h>
46 
47 #include <netinet/in.h>
48 #include <netinet/in_var.h>
49 #include <netinet/ip.h>
50 #include <netinet/ip_icmp.h>
51 #include <netinet/ip_var.h>
52 
53 extern int	in_inithead(void **head, int off);
54 #ifdef VIMAGE
55 extern int	in_detachhead(void **head, int off);
56 #endif
57 
58 #define RTPRF_OURS		RTF_PROTO3	/* set on routes we manage */
59 
60 /*
61  * Do what we need to do when inserting a route.
62  */
63 static struct radix_node *
64 in_addroute(void *v_arg, void *n_arg, struct radix_node_head *head,
65     struct radix_node *treenodes)
66 {
67 	struct rtentry *rt = (struct rtentry *)treenodes;
68 	struct sockaddr_in *sin = (struct sockaddr_in *)rt_key(rt);
69 
70 	RADIX_NODE_HEAD_WLOCK_ASSERT(head);
71 	/*
72 	 * A little bit of help for both IP output and input:
73 	 *   For host routes, we make sure that RTF_BROADCAST
74 	 *   is set for anything that looks like a broadcast address.
75 	 *   This way, we can avoid an expensive call to in_broadcast()
76 	 *   in ip_output() most of the time (because the route passed
77 	 *   to ip_output() is almost always a host route).
78 	 *
79 	 *   We also do the same for local addresses, with the thought
80 	 *   that this might one day be used to speed up ip_input().
81 	 *
82 	 * We also mark routes to multicast addresses as such, because
83 	 * it's easy to do and might be useful (but this is much more
84 	 * dubious since it's so easy to inspect the address).
85 	 */
86 	if (rt->rt_flags & RTF_HOST) {
87 		if (in_broadcast(sin->sin_addr, rt->rt_ifp)) {
88 			rt->rt_flags |= RTF_BROADCAST;
89 		} else if (satosin(rt->rt_ifa->ifa_addr)->sin_addr.s_addr ==
90 		    sin->sin_addr.s_addr) {
91 			rt->rt_flags |= RTF_LOCAL;
92 		}
93 	}
94 	if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
95 		rt->rt_flags |= RTF_MULTICAST;
96 
97 	if (!rt->rt_rmx.rmx_mtu && rt->rt_ifp)
98 		rt->rt_rmx.rmx_mtu = rt->rt_ifp->if_mtu;
99 
100 	return (rn_addroute(v_arg, n_arg, head, treenodes));
101 }
102 
103 /*
104  * This code is the inverse of in_clsroute: on first reference, if we
105  * were managing the route, stop doing so and set the expiration timer
106  * back off again.
107  */
108 static struct radix_node *
109 in_matroute(void *v_arg, struct radix_node_head *head)
110 {
111 	struct radix_node *rn = rn_match(v_arg, head);
112 	struct rtentry *rt = (struct rtentry *)rn;
113 
114 	if (rt) {
115 		RT_LOCK(rt);
116 		if (rt->rt_flags & RTPRF_OURS) {
117 			rt->rt_flags &= ~RTPRF_OURS;
118 			rt->rt_rmx.rmx_expire = 0;
119 		}
120 		RT_UNLOCK(rt);
121 	}
122 	return rn;
123 }
124 
125 static VNET_DEFINE(int, rtq_reallyold) = 60*60; /* one hour is "really old" */
126 #define	V_rtq_reallyold		VNET(rtq_reallyold)
127 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_RTEXPIRE, rtexpire, CTLFLAG_RW,
128     &VNET_NAME(rtq_reallyold), 0,
129     "Default expiration time on dynamically learned routes");
130 
131 /* never automatically crank down to less */
132 static VNET_DEFINE(int, rtq_minreallyold) = 10;
133 #define	V_rtq_minreallyold	VNET(rtq_minreallyold)
134 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_RTMINEXPIRE, rtminexpire, CTLFLAG_RW,
135     &VNET_NAME(rtq_minreallyold), 0,
136     "Minimum time to attempt to hold onto dynamically learned routes");
137 
138 /* 128 cached routes is "too many" */
139 static VNET_DEFINE(int, rtq_toomany) = 128;
140 #define	V_rtq_toomany		VNET(rtq_toomany)
141 SYSCTL_VNET_INT(_net_inet_ip, IPCTL_RTMAXCACHE, rtmaxcache, CTLFLAG_RW,
142     &VNET_NAME(rtq_toomany), 0,
143     "Upper limit on dynamically learned routes");
144 
145 /*
146  * On last reference drop, mark the route as belong to us so that it can be
147  * timed out.
148  */
149 static void
150 in_clsroute(struct radix_node *rn, struct radix_node_head *head)
151 {
152 	struct rtentry *rt = (struct rtentry *)rn;
153 
154 	RT_LOCK_ASSERT(rt);
155 
156 	if (!(rt->rt_flags & RTF_UP))
157 		return;			/* prophylactic measures */
158 
159 	if (rt->rt_flags & RTPRF_OURS)
160 		return;
161 
162 	if (!(rt->rt_flags & RTF_DYNAMIC))
163 		return;
164 
165 	/*
166 	 * If rtq_reallyold is 0, just delete the route without
167 	 * waiting for a timeout cycle to kill it.
168 	 */
169 	if (V_rtq_reallyold != 0) {
170 		rt->rt_flags |= RTPRF_OURS;
171 		rt->rt_rmx.rmx_expire = time_uptime + V_rtq_reallyold;
172 	} else {
173 		rtexpunge(rt);
174 	}
175 }
176 
177 struct rtqk_arg {
178 	struct radix_node_head *rnh;
179 	int draining;
180 	int killed;
181 	int found;
182 	int updating;
183 	time_t nextstop;
184 };
185 
186 /*
187  * Get rid of old routes.  When draining, this deletes everything, even when
188  * the timeout is not expired yet.  When updating, this makes sure that
189  * nothing has a timeout longer than the current value of rtq_reallyold.
190  */
191 static int
192 in_rtqkill(struct radix_node *rn, void *rock)
193 {
194 	struct rtqk_arg *ap = rock;
195 	struct rtentry *rt = (struct rtentry *)rn;
196 	int err;
197 
198 	RADIX_NODE_HEAD_WLOCK_ASSERT(ap->rnh);
199 
200 	if (rt->rt_flags & RTPRF_OURS) {
201 		ap->found++;
202 
203 		if (ap->draining || rt->rt_rmx.rmx_expire <= time_uptime) {
204 			if (rt->rt_refcnt > 0)
205 				panic("rtqkill route really not free");
206 
207 			err = in_rtrequest(RTM_DELETE,
208 					(struct sockaddr *)rt_key(rt),
209 					rt->rt_gateway, rt_mask(rt),
210 					rt->rt_flags | RTF_RNH_LOCKED, 0,
211 					rt->rt_fibnum);
212 			if (err) {
213 				log(LOG_WARNING, "in_rtqkill: error %d\n", err);
214 			} else {
215 				ap->killed++;
216 			}
217 		} else {
218 			if (ap->updating &&
219 			    (rt->rt_rmx.rmx_expire - time_uptime >
220 			     V_rtq_reallyold)) {
221 				rt->rt_rmx.rmx_expire =
222 				    time_uptime + V_rtq_reallyold;
223 			}
224 			ap->nextstop = lmin(ap->nextstop,
225 					    rt->rt_rmx.rmx_expire);
226 		}
227 	}
228 
229 	return 0;
230 }
231 
232 #define RTQ_TIMEOUT	60*10	/* run no less than once every ten minutes */
233 static VNET_DEFINE(int, rtq_timeout) = RTQ_TIMEOUT;
234 static VNET_DEFINE(struct callout, rtq_timer);
235 
236 #define	V_rtq_timeout		VNET(rtq_timeout)
237 #define	V_rtq_timer		VNET(rtq_timer)
238 
239 static void in_rtqtimo_one(void *rock);
240 
241 static void
242 in_rtqtimo(void *rock)
243 {
244 	CURVNET_SET((struct vnet *) rock);
245 	int fibnum;
246 	void *newrock;
247 	struct timeval atv;
248 
249 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
250 		newrock = rt_tables_get_rnh(fibnum, AF_INET);
251 		if (newrock != NULL)
252 			in_rtqtimo_one(newrock);
253 	}
254 	atv.tv_usec = 0;
255 	atv.tv_sec = V_rtq_timeout;
256 	callout_reset(&V_rtq_timer, tvtohz(&atv), in_rtqtimo, rock);
257 	CURVNET_RESTORE();
258 }
259 
260 static void
261 in_rtqtimo_one(void *rock)
262 {
263 	struct radix_node_head *rnh = rock;
264 	struct rtqk_arg arg;
265 	static time_t last_adjusted_timeout = 0;
266 
267 	arg.found = arg.killed = 0;
268 	arg.rnh = rnh;
269 	arg.nextstop = time_uptime + V_rtq_timeout;
270 	arg.draining = arg.updating = 0;
271 	RADIX_NODE_HEAD_LOCK(rnh);
272 	rnh->rnh_walktree(rnh, in_rtqkill, &arg);
273 	RADIX_NODE_HEAD_UNLOCK(rnh);
274 
275 	/*
276 	 * Attempt to be somewhat dynamic about this:
277 	 * If there are ``too many'' routes sitting around taking up space,
278 	 * then crank down the timeout, and see if we can't make some more
279 	 * go away.  However, we make sure that we will never adjust more
280 	 * than once in rtq_timeout seconds, to keep from cranking down too
281 	 * hard.
282 	 */
283 	if ((arg.found - arg.killed > V_rtq_toomany) &&
284 	    (time_uptime - last_adjusted_timeout >= V_rtq_timeout) &&
285 	    V_rtq_reallyold > V_rtq_minreallyold) {
286 		V_rtq_reallyold = 2 * V_rtq_reallyold / 3;
287 		if (V_rtq_reallyold < V_rtq_minreallyold) {
288 			V_rtq_reallyold = V_rtq_minreallyold;
289 		}
290 
291 		last_adjusted_timeout = time_uptime;
292 #ifdef DIAGNOSTIC
293 		log(LOG_DEBUG, "in_rtqtimo: adjusted rtq_reallyold to %d\n",
294 		    V_rtq_reallyold);
295 #endif
296 		arg.found = arg.killed = 0;
297 		arg.updating = 1;
298 		RADIX_NODE_HEAD_LOCK(rnh);
299 		rnh->rnh_walktree(rnh, in_rtqkill, &arg);
300 		RADIX_NODE_HEAD_UNLOCK(rnh);
301 	}
302 
303 }
304 
305 void
306 in_rtqdrain(void)
307 {
308 	VNET_ITERATOR_DECL(vnet_iter);
309 	struct radix_node_head *rnh;
310 	struct rtqk_arg arg;
311 	int 	fibnum;
312 
313 	VNET_LIST_RLOCK_NOSLEEP();
314 	VNET_FOREACH(vnet_iter) {
315 		CURVNET_SET(vnet_iter);
316 
317 		for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
318 			rnh = rt_tables_get_rnh(fibnum, AF_INET);
319 			arg.found = arg.killed = 0;
320 			arg.rnh = rnh;
321 			arg.nextstop = 0;
322 			arg.draining = 1;
323 			arg.updating = 0;
324 			RADIX_NODE_HEAD_LOCK(rnh);
325 			rnh->rnh_walktree(rnh, in_rtqkill, &arg);
326 			RADIX_NODE_HEAD_UNLOCK(rnh);
327 		}
328 		CURVNET_RESTORE();
329 	}
330 	VNET_LIST_RUNLOCK_NOSLEEP();
331 }
332 
333 void
334 in_setmatchfunc(struct radix_node_head *rnh, int val)
335 {
336 
337 	rnh->rnh_matchaddr = (val != 0) ? rn_match : in_matroute;
338 }
339 
340 static int _in_rt_was_here;
341 /*
342  * Initialize our routing tree.
343  */
344 int
345 in_inithead(void **head, int off)
346 {
347 	struct radix_node_head *rnh;
348 
349 	/* XXX MRT
350 	 * This can be called from vfs_export.c too in which case 'off'
351 	 * will be 0. We know the correct value so just use that and
352 	 * return directly if it was 0.
353 	 * This is a hack that replaces an even worse hack on a bad hack
354 	 * on a bad design. After RELENG_7 this should be fixed but that
355 	 * will change the ABI, so for now do it this way.
356 	 */
357 	if (!rn_inithead(head, 32))
358 		return 0;
359 
360 	if (off == 0)		/* XXX MRT  see above */
361 		return 1;	/* only do the rest for a real routing table */
362 
363 	rnh = *head;
364 	rnh->rnh_addaddr = in_addroute;
365 	in_setmatchfunc(rnh, V_drop_redirect);
366 	rnh->rnh_close = in_clsroute;
367 	if (_in_rt_was_here == 0 ) {
368 		callout_init(&V_rtq_timer, CALLOUT_MPSAFE);
369 		callout_reset(&V_rtq_timer, 1, in_rtqtimo, curvnet);
370 		_in_rt_was_here = 1;
371 	}
372 	return 1;
373 }
374 
375 #ifdef VIMAGE
376 int
377 in_detachhead(void **head, int off)
378 {
379 
380 	callout_drain(&V_rtq_timer);
381 	return (1);
382 }
383 #endif
384 
385 /*
386  * This zaps old routes when the interface goes down or interface
387  * address is deleted.  In the latter case, it deletes static routes
388  * that point to this address.  If we don't do this, we may end up
389  * using the old address in the future.  The ones we always want to
390  * get rid of are things like ARP entries, since the user might down
391  * the interface, walk over to a completely different network, and
392  * plug back in.
393  */
394 struct in_ifadown_arg {
395 	struct ifaddr *ifa;
396 	int del;
397 };
398 
399 static int
400 in_ifadownkill(struct radix_node *rn, void *xap)
401 {
402 	struct in_ifadown_arg *ap = xap;
403 	struct rtentry *rt = (struct rtentry *)rn;
404 
405 	RT_LOCK(rt);
406 	if (rt->rt_ifa == ap->ifa &&
407 	    (ap->del || !(rt->rt_flags & RTF_STATIC))) {
408 		/*
409 		 * Aquire a reference so that it can later be freed
410 		 * as the refcount would be 0 here in case of at least
411 		 * ap->del.
412 		 */
413 		RT_ADDREF(rt);
414 		/*
415 		 * Disconnect it from the tree and permit protocols
416 		 * to cleanup.
417 		 */
418 		rtexpunge(rt);
419 		/*
420 		 * At this point it is an rttrash node, and in case
421 		 * the above is the only reference we must free it.
422 		 * If we do not noone will have a pointer and the
423 		 * rtentry will be leaked forever.
424 		 * In case someone else holds a reference, we are
425 		 * fine as we only decrement the refcount. In that
426 		 * case if the other entity calls RT_REMREF, we
427 		 * will still be leaking but at least we tried.
428 		 */
429 		RTFREE_LOCKED(rt);
430 		return (0);
431 	}
432 	RT_UNLOCK(rt);
433 	return 0;
434 }
435 
436 void
437 in_ifadown(struct ifaddr *ifa, int delete)
438 {
439 	struct in_ifadown_arg arg;
440 	struct radix_node_head *rnh;
441 	int	fibnum;
442 
443 	KASSERT(ifa->ifa_addr->sa_family == AF_INET,
444 	    ("%s: wrong family", __func__));
445 
446 	for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) {
447 		rnh = rt_tables_get_rnh(fibnum, AF_INET);
448 		arg.ifa = ifa;
449 		arg.del = delete;
450 		RADIX_NODE_HEAD_LOCK(rnh);
451 		rnh->rnh_walktree(rnh, in_ifadownkill, &arg);
452 		RADIX_NODE_HEAD_UNLOCK(rnh);
453 		ifa->ifa_flags &= ~IFA_ROUTE;		/* XXXlocking? */
454 	}
455 }
456 
457 /*
458  * inet versions of rt functions. These have fib extensions and
459  * for now will just reference the _fib variants.
460  * eventually this order will be reversed,
461  */
462 void
463 in_rtalloc_ign(struct route *ro, u_long ignflags, u_int fibnum)
464 {
465 	rtalloc_ign_fib(ro, ignflags, fibnum);
466 }
467 
468 int
469 in_rtrequest( int req,
470 	struct sockaddr *dst,
471 	struct sockaddr *gateway,
472 	struct sockaddr *netmask,
473 	int flags,
474 	struct rtentry **ret_nrt,
475 	u_int fibnum)
476 {
477 	return (rtrequest_fib(req, dst, gateway, netmask,
478 	    flags, ret_nrt, fibnum));
479 }
480 
481 struct rtentry *
482 in_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum)
483 {
484 	return (rtalloc1_fib(dst, report, ignflags, fibnum));
485 }
486 
487 void
488 in_rtredirect(struct sockaddr *dst,
489 	struct sockaddr *gateway,
490 	struct sockaddr *netmask,
491 	int flags,
492 	struct sockaddr *src,
493 	u_int fibnum)
494 {
495 	rtredirect_fib(dst, gateway, netmask, flags, src, fibnum);
496 }
497 
498 void
499 in_rtalloc(struct route *ro, u_int fibnum)
500 {
501 	rtalloc_ign_fib(ro, 0UL, fibnum);
502 }
503 
504 #if 0
505 int	 in_rt_getifa(struct rt_addrinfo *, u_int fibnum);
506 int	 in_rtioctl(u_long, caddr_t, u_int);
507 int	 in_rtrequest1(int, struct rt_addrinfo *, struct rtentry **, u_int);
508 #endif
509 
510 
511