xref: /freebsd/sbin/routed/table.c (revision da5432eda807c4b7232d030d5157d5b417ea4f52)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1983, 1988, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include "defs.h"
33 static struct rt_spare *rts_better(struct rt_entry *);
34 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0,0};
35 static void  set_need_flash(void);
36 #ifdef _HAVE_SIN_LEN
37 static void masktrim(struct sockaddr_in *ap);
38 #else
39 static void masktrim(struct sockaddr_in_new *ap);
40 #endif
41 static void rtbad(struct rt_entry *);
42 
43 
44 struct radix_node_head *rhead;		/* root of the radix tree */
45 
46 int	need_flash = 1;			/* flash update needed
47 					 * start =1 to suppress the 1st
48 					 */
49 
50 struct timeval age_timer;		/* next check of old routes */
51 struct timeval need_kern = {		/* need to update kernel table */
52 	EPOCH+MIN_WAITTIME-1, 0
53 };
54 
55 int	stopint;
56 
57 int	total_routes;
58 
59 /* zap any old routes through this gateway */
60 static naddr age_bad_gate;
61 
62 
63 /* It is desirable to "aggregate" routes, to combine differing routes of
64  * the same metric and next hop into a common route with a smaller netmask
65  * or to suppress redundant routes, routes that add no information to
66  * routes with smaller netmasks.
67  *
68  * A route is redundant if and only if any and all routes with smaller
69  * but matching netmasks and nets are the same.  Since routes are
70  * kept sorted in the radix tree, redundant routes always come second.
71  *
72  * There are two kinds of aggregations.  First, two routes of the same bit
73  * mask and differing only in the least significant bit of the network
74  * number can be combined into a single route with a coarser mask.
75  *
76  * Second, a route can be suppressed in favor of another route with a more
77  * coarse mask provided no incompatible routes with intermediate masks
78  * are present.  The second kind of aggregation involves suppressing routes.
79  * A route must not be suppressed if an incompatible route exists with
80  * an intermediate mask, since the suppressed route would be covered
81  * by the intermediate.
82  *
83  * This code relies on the radix tree walk encountering routes
84  * sorted first by address, with the smallest address first.
85  */
86 
87 static struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest;
88 
89 /* #define DEBUG_AG */
90 #ifdef DEBUG_AG
91 #define CHECK_AG() {int acnt = 0; struct ag_info *cag;		\
92 	for (cag = ag_avail; cag != NULL; cag = cag->ag_fine)	\
93 		acnt++;						\
94 	for (cag = ag_corsest; cag != NULL; cag = cag->ag_fine)	\
95 		acnt++;						\
96 	if (acnt != NUM_AG_SLOTS) {				\
97 		(void)fflush(stderr);				\
98 		abort();					\
99 	}							\
100 }
101 #else
102 #define CHECK_AG()
103 #endif
104 
105 
106 /* Output the contents of an aggregation table slot.
107  *	This function must always be immediately followed with the deletion
108  *	of the target slot.
109  */
110 static void
ag_out(struct ag_info * ag,void (* out)(struct ag_info *))111 ag_out(struct ag_info *ag,
112 	 void (*out)(struct ag_info *))
113 {
114 	struct ag_info *ag_cors;
115 	naddr bit;
116 
117 
118 	/* Forget it if this route should not be output for split-horizon. */
119 	if (ag->ag_state & AGS_SPLIT_HZ)
120 		return;
121 
122 	/* If we output both the even and odd twins, then the immediate parent,
123 	 * if it is present, is redundant, unless the parent manages to
124 	 * aggregate into something coarser.
125 	 * On successive calls, this code detects the even and odd twins,
126 	 * and marks the parent.
127 	 *
128 	 * Note that the order in which the radix tree code emits routes
129 	 * ensures that the twins are seen before the parent is emitted.
130 	 */
131 	ag_cors = ag->ag_cors;
132 	if (ag_cors != NULL
133 	    && ag_cors->ag_mask == ag->ag_mask<<1
134 	    && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) {
135 		ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h)
136 				      ? AGS_REDUN0
137 				      : AGS_REDUN1);
138 	}
139 
140 	/* Skip it if this route is itself redundant.
141 	 *
142 	 * It is ok to change the contents of the slot here, since it is
143 	 * always deleted next.
144 	 */
145 	if (ag->ag_state & AGS_REDUN0) {
146 		if (ag->ag_state & AGS_REDUN1)
147 			return;		/* quit if fully redundant */
148 		/* make it finer if it is half-redundant */
149 		bit = (-ag->ag_mask) >> 1;
150 		ag->ag_dst_h |= bit;
151 		ag->ag_mask |= bit;
152 
153 	} else if (ag->ag_state & AGS_REDUN1) {
154 		/* make it finer if it is half-redundant */
155 		bit = (-ag->ag_mask) >> 1;
156 		ag->ag_mask |= bit;
157 	}
158 	out(ag);
159 }
160 
161 
162 static void
ag_del(struct ag_info * ag)163 ag_del(struct ag_info *ag)
164 {
165 	CHECK_AG();
166 
167 	if (ag->ag_cors == NULL)
168 		ag_corsest = ag->ag_fine;
169 	else
170 		ag->ag_cors->ag_fine = ag->ag_fine;
171 
172 	if (ag->ag_fine == NULL)
173 		ag_finest = ag->ag_cors;
174 	else
175 		ag->ag_fine->ag_cors = ag->ag_cors;
176 
177 	ag->ag_fine = ag_avail;
178 	ag_avail = ag;
179 
180 	CHECK_AG();
181 }
182 
183 
184 /* Flush routes waiting for aggregation.
185  *	This must not suppress a route unless it is known that among all
186  *	routes with coarser masks that match it, the one with the longest
187  *	mask is appropriate.  This is ensured by scanning the routes
188  *	in lexical order, and with the most restrictive mask first
189  *	among routes to the same destination.
190  */
191 void
ag_flush(naddr lim_dst_h,naddr lim_mask,void (* out)(struct ag_info *))192 ag_flush(naddr lim_dst_h,		/* flush routes to here */
193 	 naddr lim_mask,		/* matching this mask */
194 	 void (*out)(struct ag_info *))
195 {
196 	struct ag_info *ag, *ag_cors;
197 	naddr dst_h;
198 
199 
200 	for (ag = ag_finest;
201 	     ag != NULL && ag->ag_mask >= lim_mask;
202 	     ag = ag_cors) {
203 		ag_cors = ag->ag_cors;
204 
205 		/* work on only the specified routes */
206 		dst_h = ag->ag_dst_h;
207 		if ((dst_h & lim_mask) != lim_dst_h)
208 			continue;
209 
210 		if (!(ag->ag_state & AGS_SUPPRESS))
211 			ag_out(ag, out);
212 
213 		else for ( ; ; ag_cors = ag_cors->ag_cors) {
214 			/* Look for a route that can suppress the
215 			 * current route */
216 			if (ag_cors == NULL) {
217 				/* failed, so output it and look for
218 				 * another route to work on
219 				 */
220 				ag_out(ag, out);
221 				break;
222 			}
223 
224 			if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) {
225 				/* We found a route with a coarser mask that
226 				 * aggregates the current target.
227 				 *
228 				 * If it has a different next hop, it
229 				 * cannot replace the target, so output
230 				 * the target.
231 				 */
232 				if (ag->ag_gate != ag_cors->ag_gate
233 				    && !(ag->ag_state & AGS_FINE_GATE)
234 				    && !(ag_cors->ag_state & AGS_CORS_GATE)) {
235 					ag_out(ag, out);
236 					break;
237 				}
238 
239 				/* If the coarse route has a good enough
240 				 * metric, it suppresses the target.
241 				 * If the suppressed target was redundant,
242 				 * then mark the suppressor redundant.
243 				 */
244 				if (ag_cors->ag_pref <= ag->ag_pref) {
245 				    if (AG_IS_REDUN(ag->ag_state)
246 					&& ag_cors->ag_mask==ag->ag_mask<<1) {
247 					if (ag_cors->ag_dst_h == dst_h)
248 					    ag_cors->ag_state |= AGS_REDUN0;
249 					else
250 					    ag_cors->ag_state |= AGS_REDUN1;
251 				    }
252 				    if (ag->ag_tag != ag_cors->ag_tag)
253 					    ag_cors->ag_tag = 0;
254 				    if (ag->ag_nhop != ag_cors->ag_nhop)
255 					    ag_cors->ag_nhop = 0;
256 				    break;
257 				}
258 			}
259 		}
260 
261 		/* That route has either been output or suppressed */
262 		ag_cors = ag->ag_cors;
263 		ag_del(ag);
264 	}
265 
266 	CHECK_AG();
267 }
268 
269 
270 /* Try to aggregate a route with previous routes.
271  */
272 void
ag_check(naddr dst,naddr mask,naddr gate,naddr nhop,char metric,char pref,u_int new_seqno,u_short tag,u_short state,void (* out)(struct ag_info *))273 ag_check(naddr	dst,
274 	 naddr	mask,
275 	 naddr	gate,
276 	 naddr	nhop,
277 	 char	metric,
278 	 char	pref,
279 	 u_int	new_seqno,
280 	 u_short tag,
281 	 u_short state,
282 	 void (*out)(struct ag_info *))	/* output using this */
283 {
284 	struct ag_info *ag, *nag, *ag_cors;
285 	naddr xaddr;
286 	int x;
287 
288 	dst = ntohl(dst);
289 
290 	/* Punt non-contiguous subnet masks.
291 	 *
292 	 * (X & -X) contains a single bit if and only if X is a power of 2.
293 	 * (X + (X & -X)) == 0 if and only if X is a power of 2.
294 	 */
295 	if ((mask & -mask) + mask != 0) {
296 		struct ag_info nc_ag;
297 
298 		nc_ag.ag_dst_h = dst;
299 		nc_ag.ag_mask = mask;
300 		nc_ag.ag_gate = gate;
301 		nc_ag.ag_nhop = nhop;
302 		nc_ag.ag_metric = metric;
303 		nc_ag.ag_pref = pref;
304 		nc_ag.ag_tag = tag;
305 		nc_ag.ag_state = state;
306 		nc_ag.ag_seqno = new_seqno;
307 		out(&nc_ag);
308 		return;
309 	}
310 
311 	/* Search for the right slot in the aggregation table.
312 	 */
313 	ag_cors = NULL;
314 	ag = ag_corsest;
315 	while (ag != NULL) {
316 		if (ag->ag_mask >= mask)
317 			break;
318 
319 		/* Suppress old routes (i.e. combine with compatible routes
320 		 * with coarser masks) as we look for the right slot in the
321 		 * aggregation table for the new route.
322 		 * A route to an address less than the current destination
323 		 * will not be affected by the current route or any route
324 		 * seen hereafter.  That means it is safe to suppress it.
325 		 * This check keeps poor routes (e.g. with large hop counts)
326 		 * from preventing suppression of finer routes.
327 		 */
328 		if (ag_cors != NULL
329 		    && ag->ag_dst_h < dst
330 		    && (ag->ag_state & AGS_SUPPRESS)
331 		    && ag_cors->ag_pref <= ag->ag_pref
332 		    && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h
333 		    && (ag_cors->ag_gate == ag->ag_gate
334 			|| (ag->ag_state & AGS_FINE_GATE)
335 			|| (ag_cors->ag_state & AGS_CORS_GATE))) {
336 			/*  If the suppressed target was redundant,
337 			 * then mark the suppressor redundant.
338 			 */
339 			if (AG_IS_REDUN(ag->ag_state)
340 			    && ag_cors->ag_mask == ag->ag_mask<<1) {
341 				if (ag_cors->ag_dst_h == dst)
342 					ag_cors->ag_state |= AGS_REDUN0;
343 				else
344 					ag_cors->ag_state |= AGS_REDUN1;
345 			}
346 			if (ag->ag_tag != ag_cors->ag_tag)
347 				ag_cors->ag_tag = 0;
348 			if (ag->ag_nhop != ag_cors->ag_nhop)
349 				ag_cors->ag_nhop = 0;
350 			ag_del(ag);
351 			CHECK_AG();
352 		} else {
353 			ag_cors = ag;
354 		}
355 		ag = ag_cors->ag_fine;
356 	}
357 
358 	/* If we find the even/odd twin of the new route, and if the
359 	 * masks and so forth are equal, we can aggregate them.
360 	 * We can probably promote one of the pair.
361 	 *
362 	 * Since the routes are encountered in lexical order,
363 	 * the new route must be odd.  However, the second or later
364 	 * times around this loop, it could be the even twin promoted
365 	 * from the even/odd pair of twins of the finer route.
366 	 */
367 	while (ag != NULL
368 	       && ag->ag_mask == mask
369 	       && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) {
370 
371 		/* Here we know the target route and the route in the current
372 		 * slot have the same netmasks and differ by at most the
373 		 * last bit.  They are either for the same destination, or
374 		 * for an even/odd pair of destinations.
375 		 */
376 		if (ag->ag_dst_h == dst) {
377 			/* We have two routes to the same destination.
378 			 * Routes are encountered in lexical order, so a
379 			 * route is never promoted until the parent route is
380 			 * already present.  So we know that the new route is
381 			 * a promoted (or aggregated) pair and the route
382 			 * already in the slot is the explicit route.
383 			 *
384 			 * Prefer the best route if their metrics differ,
385 			 * or the aggregated one if not, following a sort
386 			 * of longest-match rule.
387 			 */
388 			if (pref <= ag->ag_pref) {
389 				ag->ag_gate = gate;
390 				ag->ag_nhop = nhop;
391 				ag->ag_tag = tag;
392 				ag->ag_metric = metric;
393 				ag->ag_pref = pref;
394 				if (ag->ag_seqno < new_seqno)
395 					ag->ag_seqno = new_seqno;
396 				x = ag->ag_state;
397 				ag->ag_state = state;
398 				state = x;
399 			}
400 
401 			/* Some bits are set if they are set on either route,
402 			 * except when the route is for an interface.
403 			 */
404 			if (!(ag->ag_state & AGS_IF))
405 				ag->ag_state |= (state & (AGS_AGGREGATE_EITHER
406 							| AGS_REDUN0
407 							| AGS_REDUN1));
408 			return;
409 		}
410 
411 		/* If one of the routes can be promoted and the other can
412 		 * be suppressed, it may be possible to combine them or
413 		 * worthwhile to promote one.
414 		 *
415 		 * Any route that can be promoted is always
416 		 * marked to be eligible to be suppressed.
417 		 */
418 		if (!((state & AGS_AGGREGATE)
419 		      && (ag->ag_state & AGS_SUPPRESS))
420 		    && !((ag->ag_state & AGS_AGGREGATE)
421 			 && (state & AGS_SUPPRESS)))
422 			break;
423 
424 		/* A pair of even/odd twin routes can be combined
425 		 * if either is redundant, or if they are via the
426 		 * same gateway and have the same metric.
427 		 */
428 		if (AG_IS_REDUN(ag->ag_state)
429 		    || AG_IS_REDUN(state)
430 		    || (ag->ag_gate == gate
431 			&& ag->ag_pref == pref
432 			&& (state & ag->ag_state & AGS_AGGREGATE) != 0)) {
433 
434 			/* We have both the even and odd pairs.
435 			 * Since the routes are encountered in order,
436 			 * the route in the slot must be the even twin.
437 			 *
438 			 * Combine and promote (aggregate) the pair of routes.
439 			 */
440 			if (new_seqno < ag->ag_seqno)
441 				new_seqno = ag->ag_seqno;
442 			if (!AG_IS_REDUN(state))
443 				state &= ~AGS_REDUN1;
444 			if (AG_IS_REDUN(ag->ag_state))
445 				state |= AGS_REDUN0;
446 			else
447 				state &= ~AGS_REDUN0;
448 			state |= (ag->ag_state & AGS_AGGREGATE_EITHER);
449 			if (ag->ag_tag != tag)
450 				tag = 0;
451 			if (ag->ag_nhop != nhop)
452 				nhop = 0;
453 
454 			/* Get rid of the even twin that was already
455 			 * in the slot.
456 			 */
457 			ag_del(ag);
458 
459 		} else if (ag->ag_pref >= pref
460 			   && (ag->ag_state & AGS_AGGREGATE)) {
461 			/* If we cannot combine the pair, maybe the route
462 			 * with the worse metric can be promoted.
463 			 *
464 			 * Promote the old, even twin, by giving its slot
465 			 * in the table to the new, odd twin.
466 			 */
467 			ag->ag_dst_h = dst;
468 
469 			xaddr = ag->ag_gate;
470 			ag->ag_gate = gate;
471 			gate = xaddr;
472 
473 			xaddr = ag->ag_nhop;
474 			ag->ag_nhop = nhop;
475 			nhop = xaddr;
476 
477 			x = ag->ag_tag;
478 			ag->ag_tag = tag;
479 			tag = x;
480 
481 			/* The promoted route is even-redundant only if the
482 			 * even twin was fully redundant.  It is not
483 			 * odd-redundant because the odd-twin will still be
484 			 * in the table.
485 			 */
486 			x = ag->ag_state;
487 			if (!AG_IS_REDUN(x))
488 				x &= ~AGS_REDUN0;
489 			x &= ~AGS_REDUN1;
490 			ag->ag_state = state;
491 			state = x;
492 
493 			x = ag->ag_metric;
494 			ag->ag_metric = metric;
495 			metric = x;
496 
497 			x = ag->ag_pref;
498 			ag->ag_pref = pref;
499 			pref = x;
500 
501 			/* take the newest sequence number */
502 			if (new_seqno <= ag->ag_seqno)
503 				new_seqno = ag->ag_seqno;
504 			else
505 				ag->ag_seqno = new_seqno;
506 
507 		} else {
508 			if (!(state & AGS_AGGREGATE))
509 				break;	/* cannot promote either twin */
510 
511 			/* Promote the new, odd twin by shaving its
512 			 * mask and address.
513 			 * The promoted route is odd-redundant only if the
514 			 * odd twin was fully redundant.  It is not
515 			 * even-redundant because the even twin is still in
516 			 * the table.
517 			 */
518 			if (!AG_IS_REDUN(state))
519 				state &= ~AGS_REDUN1;
520 			state &= ~AGS_REDUN0;
521 			if (new_seqno < ag->ag_seqno)
522 				new_seqno = ag->ag_seqno;
523 			else
524 				ag->ag_seqno = new_seqno;
525 		}
526 
527 		mask <<= 1;
528 		dst &= mask;
529 
530 		if (ag_cors == NULL) {
531 			ag = ag_corsest;
532 			break;
533 		}
534 		ag = ag_cors;
535 		ag_cors = ag->ag_cors;
536 	}
537 
538 	/* When we can no longer promote and combine routes,
539 	 * flush the old route in the target slot.  Also flush
540 	 * any finer routes that we know will never be aggregated by
541 	 * the new route.
542 	 *
543 	 * In case we moved toward coarser masks,
544 	 * get back where we belong
545 	 */
546 	if (ag != NULL
547 	    && ag->ag_mask < mask) {
548 		ag_cors = ag;
549 		ag = ag->ag_fine;
550 	}
551 
552 	/* Empty the target slot
553 	 */
554 	if (ag != NULL && ag->ag_mask == mask) {
555 		ag_flush(ag->ag_dst_h, ag->ag_mask, out);
556 		ag = (ag_cors == NULL) ? ag_corsest : ag_cors->ag_fine;
557 	}
558 
559 #ifdef DEBUG_AG
560 	(void)fflush(stderr);
561 	if (ag == NULL && ag_cors != ag_finest)
562 		abort();
563 	if (ag_cors == NULL && ag != ag_corsest)
564 		abort();
565 	if (ag != NULL && ag->ag_cors != ag_cors)
566 		abort();
567 	if (ag_cors != NULL && ag_cors->ag_fine != ag)
568 		abort();
569 	CHECK_AG();
570 #endif
571 
572 	/* Save the new route on the end of the table.
573 	 */
574 	nag = ag_avail;
575 	ag_avail = nag->ag_fine;
576 
577 	nag->ag_dst_h = dst;
578 	nag->ag_mask = mask;
579 	nag->ag_gate = gate;
580 	nag->ag_nhop = nhop;
581 	nag->ag_metric = metric;
582 	nag->ag_pref = pref;
583 	nag->ag_tag = tag;
584 	nag->ag_state = state;
585 	nag->ag_seqno = new_seqno;
586 
587 	nag->ag_fine = ag;
588 	if (ag != NULL)
589 		ag->ag_cors = nag;
590 	else
591 		ag_finest = nag;
592 	nag->ag_cors = ag_cors;
593 	if (ag_cors == NULL)
594 		ag_corsest = nag;
595 	else
596 		ag_cors->ag_fine = nag;
597 	CHECK_AG();
598 }
599 
600 static const char *
rtm_type_name(u_char type)601 rtm_type_name(u_char type)
602 {
603 	static const char * const rtm_types[] = {
604 		"RTM_ADD",
605 		"RTM_DELETE",
606 		"RTM_CHANGE",
607 		"RTM_GET",
608 		"RTM_LOSING",
609 		"RTM_REDIRECT",
610 		"RTM_MISS",
611 		"RTM_LOCK",
612 		"RTM_OLDADD",
613 		"RTM_OLDDEL",
614 		"RTM_RESOLVE",
615 		"RTM_NEWADDR",
616 		"RTM_DELADDR",
617 #ifdef RTM_OIFINFO
618 		"RTM_OIFINFO",
619 #endif
620 		"RTM_IFINFO",
621 		"RTM_NEWMADDR",
622 		"RTM_DELMADDR"
623 	};
624 #define NEW_RTM_PAT "RTM type %#x"
625 	static char name0[sizeof(NEW_RTM_PAT)+2];
626 
627 
628 	if (type > sizeof(rtm_types)/sizeof(rtm_types[0])
629 	    || type == 0) {
630 		snprintf(name0, sizeof(name0), NEW_RTM_PAT, type);
631 		return name0;
632 	} else {
633 		return rtm_types[type-1];
634 	}
635 #undef NEW_RTM_PAT
636 }
637 
638 
639 /* Trim a mask in a sockaddr
640  *	Produce a length of 0 for an address of 0.
641  *	Otherwise produce the index of the first zero byte.
642  */
643 void
644 #ifdef _HAVE_SIN_LEN
masktrim(struct sockaddr_in * ap)645 masktrim(struct sockaddr_in *ap)
646 #else
647 masktrim(struct sockaddr_in_new *ap)
648 #endif
649 {
650 	char *cp;
651 
652 	if (ap->sin_addr.s_addr == 0) {
653 		ap->sin_len = 0;
654 		return;
655 	}
656 	cp = (char *)(&ap->sin_addr.s_addr+1);
657 	while (*--cp == 0)
658 		continue;
659 	ap->sin_len = cp - (char*)ap + 1;
660 }
661 
662 
663 /* Tell the kernel to add, delete or change a route
664  */
665 static void
rtioctl(int action,naddr dst,naddr gate,naddr mask,int metric,int flags)666 rtioctl(int action,			/* RTM_DELETE, etc */
667 	naddr dst,
668 	naddr gate,
669 	naddr mask,
670 	int metric,
671 	int flags)
672 {
673 	struct {
674 		struct rt_msghdr w_rtm;
675 		struct sockaddr_in w_dst;
676 		struct sockaddr_in w_gate;
677 #ifdef _HAVE_SA_LEN
678 		struct sockaddr_in w_mask;
679 #else
680 		struct sockaddr_in_new w_mask;
681 #endif
682 	} w;
683 	long cc;
684 #   define PAT " %-10s %s metric=%d flags=%#x"
685 #   define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags
686 
687 again:
688 	memset(&w, 0, sizeof(w));
689 	w.w_rtm.rtm_msglen = sizeof(w);
690 	w.w_rtm.rtm_version = RTM_VERSION;
691 	w.w_rtm.rtm_type = action;
692 	w.w_rtm.rtm_flags = flags;
693 	w.w_rtm.rtm_seq = ++rt_sock_seqno;
694 	w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY;
695 	if (metric != 0 || action == RTM_CHANGE) {
696 		w.w_rtm.rtm_rmx.rmx_hopcount = metric;
697 		w.w_rtm.rtm_inits |= RTV_HOPCOUNT;
698 	}
699 	w.w_dst.sin_family = AF_INET;
700 	w.w_dst.sin_addr.s_addr = dst;
701 	w.w_gate.sin_family = AF_INET;
702 	w.w_gate.sin_addr.s_addr = gate;
703 #ifdef _HAVE_SA_LEN
704 	w.w_dst.sin_len = sizeof(w.w_dst);
705 	w.w_gate.sin_len = sizeof(w.w_gate);
706 #endif
707 	if (mask == HOST_MASK) {
708 		w.w_rtm.rtm_flags |= RTF_HOST;
709 		w.w_rtm.rtm_msglen -= sizeof(w.w_mask);
710 	} else {
711 		w.w_rtm.rtm_addrs |= RTA_NETMASK;
712 		w.w_mask.sin_addr.s_addr = htonl(mask);
713 #ifdef _HAVE_SA_LEN
714 		masktrim(&w.w_mask);
715 		if (w.w_mask.sin_len == 0)
716 			w.w_mask.sin_len = sizeof(long);
717 		w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len);
718 #endif
719 	}
720 
721 #ifndef NO_INSTALL
722 	cc = write(rt_sock, &w, w.w_rtm.rtm_msglen);
723 	if (cc < 0) {
724 		if (errno == ESRCH
725 		    && (action == RTM_CHANGE || action == RTM_DELETE)) {
726 			trace_act("route disappeared before" PAT, ARGS);
727 			if (action == RTM_CHANGE) {
728 				action = RTM_ADD;
729 				goto again;
730 			}
731 			return;
732 		}
733 		msglog("write(rt_sock)" PAT ": %s", ARGS, strerror(errno));
734 		return;
735 	} else if (cc != w.w_rtm.rtm_msglen) {
736 		msglog("write(rt_sock) wrote %ld instead of %d for" PAT,
737 		       cc, w.w_rtm.rtm_msglen, ARGS);
738 		return;
739 	}
740 #endif
741 	if (TRACEKERNEL)
742 		trace_misc("write kernel" PAT, ARGS);
743 #undef PAT
744 #undef ARGS
745 }
746 
747 
748 #define KHASH_SIZE 71			/* should be prime */
749 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE]
750 static struct khash {
751 	struct khash *k_next;
752 	naddr	k_dst;
753 	naddr	k_mask;
754 	naddr	k_gate;
755 	short	k_metric;
756 	u_short	k_state;
757 #define	    KS_NEW	0x001
758 #define	    KS_DELETE	0x002		/* need to delete the route */
759 #define	    KS_ADD	0x004		/* add to the kernel */
760 #define	    KS_CHANGE	0x008		/* tell kernel to change the route */
761 #define	    KS_DEL_ADD	0x010		/* delete & add to change the kernel */
762 #define	    KS_STATIC	0x020		/* Static flag in kernel */
763 #define	    KS_GATEWAY	0x040		/* G flag in kernel */
764 #define	    KS_DYNAMIC	0x080		/* result of redirect */
765 #define	    KS_DELETED	0x100		/* already deleted from kernel */
766 #define	    KS_CHECK	0x200
767 	time_t	k_keep;
768 #define	    K_KEEP_LIM	30
769 	time_t	k_redirect_time;	/* when redirected route 1st seen */
770 } *khash_bins[KHASH_SIZE];
771 
772 
773 static struct khash*
kern_find(naddr dst,naddr mask,struct khash *** ppk)774 kern_find(naddr dst, naddr mask, struct khash ***ppk)
775 {
776 	struct khash *k, **pk;
777 
778 	for (pk = &KHASH(dst,mask); (k = *pk) != NULL; pk = &k->k_next) {
779 		if (k->k_dst == dst && k->k_mask == mask)
780 			break;
781 	}
782 	if (ppk != NULL)
783 		*ppk = pk;
784 	return k;
785 }
786 
787 
788 static struct khash*
kern_add(naddr dst,naddr mask)789 kern_add(naddr dst, naddr mask)
790 {
791 	struct khash *k, **pk;
792 
793 	k = kern_find(dst, mask, &pk);
794 	if (k != NULL)
795 		return k;
796 
797 	k = (struct khash *)rtmalloc(sizeof(*k), "kern_add");
798 
799 	memset(k, 0, sizeof(*k));
800 	k->k_dst = dst;
801 	k->k_mask = mask;
802 	k->k_state = KS_NEW;
803 	k->k_keep = now.tv_sec;
804 	*pk = k;
805 
806 	return k;
807 }
808 
809 
810 /* If a kernel route has a non-zero metric, check that it is still in the
811  *	daemon table, and not deleted by interfaces coming and going.
812  */
813 static void
kern_check_static(struct khash * k,struct interface * ifp)814 kern_check_static(struct khash *k,
815 		  struct interface *ifp)
816 {
817 	struct rt_entry *rt;
818 	struct rt_spare new;
819 
820 	if (k->k_metric == 0)
821 		return;
822 
823 	memset(&new, 0, sizeof(new));
824 	new.rts_ifp = ifp;
825 	new.rts_gate = k->k_gate;
826 	new.rts_router = (ifp != NULL) ? ifp->int_addr : loopaddr;
827 	new.rts_metric = k->k_metric;
828 	new.rts_time = now.tv_sec;
829 
830 	rt = rtget(k->k_dst, k->k_mask);
831 	if (rt != NULL) {
832 		if (!(rt->rt_state & RS_STATIC))
833 			rtchange(rt, rt->rt_state | RS_STATIC, &new, 0);
834 	} else {
835 		rtadd(k->k_dst, k->k_mask, RS_STATIC, &new);
836 	}
837 }
838 
839 
840 /* operate on a kernel entry
841  */
842 static void
kern_ioctl(struct khash * k,int action,int flags)843 kern_ioctl(struct khash *k,
844 	   int action,			/* RTM_DELETE, etc */
845 	   int flags)
846 
847 {
848 	switch (action) {
849 	case RTM_DELETE:
850 		k->k_state &= ~KS_DYNAMIC;
851 		if (k->k_state & KS_DELETED)
852 			return;
853 		k->k_state |= KS_DELETED;
854 		break;
855 	case RTM_ADD:
856 		k->k_state &= ~KS_DELETED;
857 		break;
858 	case RTM_CHANGE:
859 		if (k->k_state & KS_DELETED) {
860 			action = RTM_ADD;
861 			k->k_state &= ~KS_DELETED;
862 		}
863 		break;
864 	}
865 
866 	rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags);
867 }
868 
869 
870 /* add a route the kernel told us
871  */
872 static void
rtm_add(struct rt_msghdr * rtm,struct rt_addrinfo * info,time_t keep)873 rtm_add(struct rt_msghdr *rtm,
874 	struct rt_addrinfo *info,
875 	time_t keep)
876 {
877 	struct khash *k;
878 	struct interface *ifp;
879 	naddr mask;
880 
881 
882 	if (rtm->rtm_flags & RTF_HOST) {
883 		mask = HOST_MASK;
884 	} else if (INFO_MASK(info) != 0) {
885 		mask = ntohl(S_ADDR(INFO_MASK(info)));
886 	} else {
887 		msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type));
888 		return;
889 	}
890 
891 	k = kern_add(S_ADDR(INFO_DST(info)), mask);
892 	if (k->k_state & KS_NEW)
893 		k->k_keep = now.tv_sec+keep;
894 	if (INFO_GATE(info) == 0) {
895 		trace_act("note %s without gateway",
896 			  rtm_type_name(rtm->rtm_type));
897 		k->k_metric = HOPCNT_INFINITY;
898 	} else if (INFO_GATE(info)->sa_family != AF_INET) {
899 		trace_act("note %s with gateway AF=%d",
900 			  rtm_type_name(rtm->rtm_type),
901 			  INFO_GATE(info)->sa_family);
902 		k->k_metric = HOPCNT_INFINITY;
903 	} else {
904 		k->k_gate = S_ADDR(INFO_GATE(info));
905 		k->k_metric = rtm->rtm_rmx.rmx_hopcount;
906 		if (k->k_metric < 0)
907 			k->k_metric = 0;
908 		else if (k->k_metric > HOPCNT_INFINITY-1)
909 			k->k_metric = HOPCNT_INFINITY-1;
910 	}
911 	k->k_state &= ~(KS_DELETE | KS_ADD | KS_CHANGE | KS_DEL_ADD
912 			| KS_DELETED | KS_GATEWAY | KS_STATIC
913 			| KS_NEW | KS_CHECK);
914 	if (rtm->rtm_flags & RTF_GATEWAY)
915 		k->k_state |= KS_GATEWAY;
916 	if (rtm->rtm_flags & RTF_STATIC)
917 		k->k_state |= KS_STATIC;
918 
919 	if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) {
920 		if (INFO_AUTHOR(info) != 0
921 		    && INFO_AUTHOR(info)->sa_family == AF_INET)
922 			ifp = iflookup(S_ADDR(INFO_AUTHOR(info)));
923 		else
924 			ifp = NULL;
925 		if (supplier
926 		    && (ifp == NULL || !(ifp->int_state & IS_REDIRECT_OK))) {
927 			/* Routers are not supposed to listen to redirects,
928 			 * so delete it if it came via an unknown interface
929 			 * or the interface does not have special permission.
930 			 */
931 			k->k_state &= ~KS_DYNAMIC;
932 			k->k_state |= KS_DELETE;
933 			LIM_SEC(need_kern, 0);
934 			trace_act("mark for deletion redirected %s --> %s"
935 				  " via %s",
936 				  addrname(k->k_dst, k->k_mask, 0),
937 				  naddr_ntoa(k->k_gate),
938 				  ifp ? ifp->int_name : "unknown interface");
939 		} else {
940 			k->k_state |= KS_DYNAMIC;
941 			k->k_redirect_time = now.tv_sec;
942 			trace_act("accept redirected %s --> %s via %s",
943 				  addrname(k->k_dst, k->k_mask, 0),
944 				  naddr_ntoa(k->k_gate),
945 				  ifp ? ifp->int_name : "unknown interface");
946 		}
947 		return;
948 	}
949 
950 	/* If it is not a static route, quit until the next comparison
951 	 * between the kernel and daemon tables, when it will be deleted.
952 	 */
953 	if (!(k->k_state & KS_STATIC)) {
954 		k->k_state |= KS_DELETE;
955 		LIM_SEC(need_kern, k->k_keep);
956 		return;
957 	}
958 
959 	/* Put static routes with real metrics into the daemon table so
960 	 * they can be advertised.
961 	 *
962 	 * Find the interface toward the gateway.
963 	 */
964 	ifp = iflookup(k->k_gate);
965 	if (ifp == NULL)
966 		msglog("static route %s --> %s impossibly lacks ifp",
967 		       addrname(S_ADDR(INFO_DST(info)), mask, 0),
968 		       naddr_ntoa(k->k_gate));
969 
970 	kern_check_static(k, ifp);
971 }
972 
973 
974 /* deal with packet loss
975  */
976 static void
rtm_lose(struct rt_msghdr * rtm,struct rt_addrinfo * info)977 rtm_lose(struct rt_msghdr *rtm,
978 	 struct rt_addrinfo *info)
979 {
980 	if (INFO_GATE(info) == 0
981 	    || INFO_GATE(info)->sa_family != AF_INET) {
982 		trace_act("ignore %s without gateway",
983 			  rtm_type_name(rtm->rtm_type));
984 		return;
985 	}
986 
987 	if (rdisc_ok)
988 		rdisc_age(S_ADDR(INFO_GATE(info)));
989 	age(S_ADDR(INFO_GATE(info)));
990 }
991 
992 
993 /* Make the gateway slot of an info structure point to something
994  * useful.  If it is not already useful, but it specifies an interface,
995  * then fill in the sockaddr_in provided and point it there.
996  */
997 static int
get_info_gate(struct sockaddr ** sap,struct sockaddr_in * rsin)998 get_info_gate(struct sockaddr **sap,
999 	      struct sockaddr_in *rsin)
1000 {
1001 	struct sockaddr_dl *sdl = (struct sockaddr_dl *)*sap;
1002 	struct interface *ifp;
1003 
1004 	if (sdl == NULL)
1005 		return 0;
1006 	if ((sdl)->sdl_family == AF_INET)
1007 		return 1;
1008 	if ((sdl)->sdl_family != AF_LINK)
1009 		return 0;
1010 
1011 	ifp = ifwithindex(sdl->sdl_index, 1);
1012 	if (ifp == NULL)
1013 		return 0;
1014 
1015 	rsin->sin_addr.s_addr = ifp->int_addr;
1016 #ifdef _HAVE_SA_LEN
1017 	rsin->sin_len = sizeof(*rsin);
1018 #endif
1019 	rsin->sin_family = AF_INET;
1020 	*sap = (struct sockaddr*)rsin;
1021 
1022 	return 1;
1023 }
1024 
1025 
1026 /* Clean the kernel table by copying it to the daemon image.
1027  * Eventually the daemon will delete any extra routes.
1028  */
1029 void
flush_kern(void)1030 flush_kern(void)
1031 {
1032 	static char *sysctl_buf;
1033 	static size_t sysctl_buf_size = 0;
1034 	size_t needed;
1035 	int mib[6];
1036 	char *next, *lim;
1037 	struct rt_msghdr *rtm;
1038 	struct sockaddr_in gate_sin;
1039 	struct rt_addrinfo info;
1040 	int i;
1041 	struct khash *k;
1042 
1043 
1044 	for (i = 0; i < KHASH_SIZE; i++) {
1045 		for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1046 			k->k_state |= KS_CHECK;
1047 		}
1048 	}
1049 
1050 	mib[0] = CTL_NET;
1051 	mib[1] = PF_ROUTE;
1052 	mib[2] = 0;		/* protocol */
1053 	mib[3] = 0;		/* wildcard address family */
1054 	mib[4] = NET_RT_DUMP;
1055 	mib[5] = 0;		/* no flags */
1056 	for (;;) {
1057 		if ((needed = sysctl_buf_size) != 0) {
1058 			if (sysctl(mib, 6, sysctl_buf,&needed, 0, 0) >= 0)
1059 				break;
1060 			if (errno != ENOMEM && errno != EFAULT)
1061 				BADERR(1,"flush_kern: sysctl(RT_DUMP)");
1062 			free(sysctl_buf);
1063 			needed = 0;
1064 		}
1065 		if (sysctl(mib, 6, 0, &needed, 0, 0) < 0)
1066 			BADERR(1,"flush_kern: sysctl(RT_DUMP) estimate");
1067 		/* Kludge around the habit of some systems, such as
1068 		 * BSD/OS 3.1, to not admit how many routes are in the
1069 		 * kernel, or at least to be quite wrong.
1070 		 */
1071 		needed += 50*(sizeof(*rtm)+5*sizeof(struct sockaddr));
1072 		sysctl_buf = rtmalloc(sysctl_buf_size = needed,
1073 				      "flush_kern sysctl(RT_DUMP)");
1074 	}
1075 
1076 	lim = sysctl_buf + needed;
1077 	for (next = sysctl_buf; next < lim; next += rtm->rtm_msglen) {
1078 		rtm = (struct rt_msghdr *)next;
1079 		if (rtm->rtm_msglen == 0) {
1080 			msglog("zero length kernel route at "
1081 			       " %#lx in buffer %#lx before %#lx",
1082 			       (u_long)rtm, (u_long)sysctl_buf, (u_long)lim);
1083 			break;
1084 		}
1085 
1086 		rt_xaddrs(&info,
1087 			  (struct sockaddr *)(rtm+1),
1088 			  (struct sockaddr *)(next + rtm->rtm_msglen),
1089 			  rtm->rtm_addrs);
1090 
1091 		if (INFO_DST(&info) == 0
1092 		    || INFO_DST(&info)->sa_family != AF_INET)
1093 			continue;
1094 
1095 #if defined (RTF_LLINFO)
1096 		/* ignore ARP table entries on systems with a merged route
1097 		 * and ARP table.
1098 		 */
1099 		if (rtm->rtm_flags & RTF_LLINFO)
1100 			continue;
1101 #endif
1102 #if defined(RTF_WASCLONED) && defined(__FreeBSD__)
1103 		/* ignore cloned routes
1104 		 */
1105 		if (rtm->rtm_flags & RTF_WASCLONED)
1106 			continue;
1107 #endif
1108 
1109 		/* ignore multicast addresses
1110 		 */
1111 		if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info)))))
1112 			continue;
1113 
1114 		if (!get_info_gate(&INFO_GATE(&info), &gate_sin))
1115 			continue;
1116 
1117 		/* Note static routes and interface routes, and also
1118 		 * preload the image of the kernel table so that
1119 		 * we can later clean it, as well as avoid making
1120 		 * unneeded changes.  Keep the old kernel routes for a
1121 		 * few seconds to allow a RIP or router-discovery
1122 		 * response to be heard.
1123 		 */
1124 		rtm_add(rtm,&info,MIN_WAITTIME);
1125 	}
1126 
1127 	for (i = 0; i < KHASH_SIZE; i++) {
1128 		for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1129 			if (k->k_state & KS_CHECK) {
1130 				msglog("%s --> %s disappeared from kernel",
1131 				       addrname(k->k_dst, k->k_mask, 0),
1132 				       naddr_ntoa(k->k_gate));
1133 				del_static(k->k_dst, k->k_mask, k->k_gate, 1);
1134 			}
1135 		}
1136 	}
1137 }
1138 
1139 
1140 /* Listen to announcements from the kernel
1141  */
1142 void
read_rt(void)1143 read_rt(void)
1144 {
1145 	long cc;
1146 	struct interface *ifp;
1147 	struct sockaddr_in gate_sin;
1148 	naddr mask, gate;
1149 	union {
1150 		struct {
1151 			struct rt_msghdr rtm;
1152 			struct sockaddr addrs[RTAX_MAX];
1153 		} r;
1154 		struct if_msghdr ifm;
1155 	} m;
1156 	char str[100], *strp;
1157 	struct rt_addrinfo info;
1158 
1159 
1160 	for (;;) {
1161 		cc = read(rt_sock, &m, sizeof(m));
1162 		if (cc <= 0) {
1163 			if (cc < 0 && errno != EWOULDBLOCK)
1164 				LOGERR("read(rt_sock)");
1165 			return;
1166 		}
1167 
1168 		if (m.r.rtm.rtm_version != RTM_VERSION) {
1169 			msglog("bogus routing message version %d",
1170 			       m.r.rtm.rtm_version);
1171 			continue;
1172 		}
1173 
1174 		/* Ignore our own results.
1175 		 */
1176 		if (m.r.rtm.rtm_type <= RTM_CHANGE
1177 		    && m.r.rtm.rtm_pid == mypid) {
1178 			static int complained = 0;
1179 			if (!complained) {
1180 				msglog("receiving our own change messages");
1181 				complained = 1;
1182 			}
1183 			continue;
1184 		}
1185 
1186 		if (m.r.rtm.rtm_type == RTM_IFINFO
1187 		    || m.r.rtm.rtm_type == RTM_NEWADDR
1188 		    || m.r.rtm.rtm_type == RTM_DELADDR) {
1189 			ifp = ifwithindex(m.ifm.ifm_index,
1190 					  m.r.rtm.rtm_type != RTM_DELADDR);
1191 			if (ifp == NULL)
1192 				trace_act("note %s with flags %#x"
1193 					  " for unknown interface index #%d",
1194 					  rtm_type_name(m.r.rtm.rtm_type),
1195 					  m.ifm.ifm_flags,
1196 					  m.ifm.ifm_index);
1197 			else
1198 				trace_act("note %s with flags %#x for %s",
1199 					  rtm_type_name(m.r.rtm.rtm_type),
1200 					  m.ifm.ifm_flags,
1201 					  ifp->int_name);
1202 
1203 			/* After being informed of a change to an interface,
1204 			 * check them all now if the check would otherwise
1205 			 * be a long time from now, if the interface is
1206 			 * not known, or if the interface has been turned
1207 			 * off or on.
1208 			 */
1209 			if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL
1210 			    || ifp == NULL
1211 			    || ((ifp->int_if_flags ^ m.ifm.ifm_flags)
1212 				& IFF_UP) != 0)
1213 				ifinit_timer.tv_sec = now.tv_sec;
1214 			continue;
1215 		}
1216 #ifdef RTM_OIFINFO
1217 		if (m.r.rtm.rtm_type == RTM_OIFINFO)
1218 			continue;	/* ignore compat message */
1219 #endif
1220 
1221 		strlcpy(str, rtm_type_name(m.r.rtm.rtm_type), sizeof(str));
1222 		strp = &str[strlen(str)];
1223 		if (m.r.rtm.rtm_type <= RTM_CHANGE)
1224 			strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid);
1225 
1226 		/*
1227 		 * Only messages that use the struct rt_msghdr format are
1228 		 * allowed beyond this point.
1229 		 */
1230 		if (m.r.rtm.rtm_type > RTM_RESOLVE) {
1231 			trace_act("ignore %s", str);
1232 			continue;
1233 		}
1234 
1235 		rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX],
1236 			  m.r.rtm.rtm_addrs);
1237 
1238 		if (INFO_DST(&info) == 0) {
1239 			trace_act("ignore %s without dst", str);
1240 			continue;
1241 		}
1242 
1243 		if (INFO_DST(&info)->sa_family != AF_INET) {
1244 			trace_act("ignore %s for AF %d", str,
1245 				  INFO_DST(&info)->sa_family);
1246 			continue;
1247 		}
1248 
1249 		mask = ((INFO_MASK(&info) != 0)
1250 			? ntohl(S_ADDR(INFO_MASK(&info)))
1251 			: (m.r.rtm.rtm_flags & RTF_HOST)
1252 			? HOST_MASK
1253 			: std_mask(S_ADDR(INFO_DST(&info))));
1254 
1255 		strp += sprintf(strp, ": %s",
1256 				addrname(S_ADDR(INFO_DST(&info)), mask, 0));
1257 
1258 		if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) {
1259 			trace_act("ignore multicast %s", str);
1260 			continue;
1261 		}
1262 
1263 #if defined(RTF_LLINFO)
1264 		if (m.r.rtm.rtm_flags & RTF_LLINFO) {
1265 			trace_act("ignore ARP %s", str);
1266 			continue;
1267 		}
1268 #endif
1269 
1270 #if defined(RTF_WASCLONED) && defined(__FreeBSD__)
1271 		if (m.r.rtm.rtm_flags & RTF_WASCLONED) {
1272 			trace_act("ignore cloned %s", str);
1273 			continue;
1274 		}
1275 #endif
1276 
1277 		if (get_info_gate(&INFO_GATE(&info), &gate_sin)) {
1278 			gate = S_ADDR(INFO_GATE(&info));
1279 			strp += sprintf(strp, " --> %s", naddr_ntoa(gate));
1280 		} else {
1281 			gate = 0;
1282 		}
1283 
1284 		if (INFO_AUTHOR(&info) != 0)
1285 			strp += sprintf(strp, " by authority of %s",
1286 					saddr_ntoa(INFO_AUTHOR(&info)));
1287 
1288 		switch (m.r.rtm.rtm_type) {
1289 		case RTM_ADD:
1290 		case RTM_CHANGE:
1291 		case RTM_REDIRECT:
1292 			if (m.r.rtm.rtm_errno != 0) {
1293 				trace_act("ignore %s with \"%s\" error",
1294 					  str, strerror(m.r.rtm.rtm_errno));
1295 			} else {
1296 				trace_act("%s", str);
1297 				rtm_add(&m.r.rtm,&info,0);
1298 			}
1299 			break;
1300 
1301 		case RTM_DELETE:
1302 			if (m.r.rtm.rtm_errno != 0
1303 			    && m.r.rtm.rtm_errno != ESRCH) {
1304 				trace_act("ignore %s with \"%s\" error",
1305 					  str, strerror(m.r.rtm.rtm_errno));
1306 			} else {
1307 				trace_act("%s", str);
1308 				del_static(S_ADDR(INFO_DST(&info)), mask,
1309 					   gate, 1);
1310 			}
1311 			break;
1312 
1313 		case RTM_LOSING:
1314 			trace_act("%s", str);
1315 			rtm_lose(&m.r.rtm,&info);
1316 			break;
1317 
1318 		default:
1319 			trace_act("ignore %s", str);
1320 			break;
1321 		}
1322 	}
1323 }
1324 
1325 
1326 /* after aggregating, note routes that belong in the kernel
1327  */
1328 static void
kern_out(struct ag_info * ag)1329 kern_out(struct ag_info *ag)
1330 {
1331 	struct khash *k;
1332 
1333 
1334 	/* Do not install bad routes if they are not already present.
1335 	 * This includes routes that had RS_NET_SYN for interfaces that
1336 	 * recently died.
1337 	 */
1338 	if (ag->ag_metric == HOPCNT_INFINITY) {
1339 		k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0);
1340 		if (k == NULL)
1341 			return;
1342 	} else {
1343 		k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask);
1344 	}
1345 
1346 	if (k->k_state & KS_NEW) {
1347 		/* will need to add new entry to the kernel table */
1348 		k->k_state = KS_ADD;
1349 		if (ag->ag_state & AGS_GATEWAY)
1350 			k->k_state |= KS_GATEWAY;
1351 		k->k_gate = ag->ag_gate;
1352 		k->k_metric = ag->ag_metric;
1353 		return;
1354 	}
1355 
1356 	if (k->k_state & KS_STATIC)
1357 		return;
1358 
1359 	/* modify existing kernel entry if necessary */
1360 	if (k->k_gate != ag->ag_gate
1361 	    || k->k_metric != ag->ag_metric) {
1362 		/* Must delete bad interface routes etc. to change them. */
1363 		if (k->k_metric == HOPCNT_INFINITY)
1364 			k->k_state |= KS_DEL_ADD;
1365 		k->k_gate = ag->ag_gate;
1366 		k->k_metric = ag->ag_metric;
1367 		k->k_state |= KS_CHANGE;
1368 	}
1369 
1370 	/* If the daemon thinks the route should exist, forget
1371 	 * about any redirections.
1372 	 * If the daemon thinks the route should exist, eventually
1373 	 * override manual intervention by the operator.
1374 	 */
1375 	if ((k->k_state & (KS_DYNAMIC | KS_DELETED)) != 0) {
1376 		k->k_state &= ~KS_DYNAMIC;
1377 		k->k_state |= (KS_ADD | KS_DEL_ADD);
1378 	}
1379 
1380 	if ((k->k_state & KS_GATEWAY)
1381 	    && !(ag->ag_state & AGS_GATEWAY)) {
1382 		k->k_state &= ~KS_GATEWAY;
1383 		k->k_state |= (KS_ADD | KS_DEL_ADD);
1384 	} else if (!(k->k_state & KS_GATEWAY)
1385 		   && (ag->ag_state & AGS_GATEWAY)) {
1386 		k->k_state |= KS_GATEWAY;
1387 		k->k_state |= (KS_ADD | KS_DEL_ADD);
1388 	}
1389 
1390 	/* Deleting-and-adding is necessary to change aspects of a route.
1391 	 * Just delete instead of deleting and then adding a bad route.
1392 	 * Otherwise, we want to keep the route in the kernel.
1393 	 */
1394 	if (k->k_metric == HOPCNT_INFINITY
1395 	    && (k->k_state & KS_DEL_ADD))
1396 		k->k_state |= KS_DELETE;
1397 	else
1398 		k->k_state &= ~KS_DELETE;
1399 #undef RT
1400 }
1401 
1402 
1403 /* ARGSUSED */
1404 static int
walk_kern(struct radix_node * rn,struct walkarg * argp UNUSED)1405 walk_kern(struct radix_node *rn,
1406 	  struct walkarg *argp UNUSED)
1407 {
1408 #define RT ((struct rt_entry *)rn)
1409 	char metric, pref;
1410 	u_int ags = 0;
1411 
1412 
1413 	/* Do not install synthetic routes */
1414 	if (RT->rt_state & RS_NET_SYN)
1415 		return 0;
1416 
1417 	if (!(RT->rt_state & RS_IF)) {
1418 		/* This is an ordinary route, not for an interface.
1419 		 */
1420 
1421 		/* aggregate, ordinary good routes without regard to
1422 		 * their metric
1423 		 */
1424 		pref = 1;
1425 		ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1426 
1427 		/* Do not install host routes directly to hosts, to avoid
1428 		 * interfering with ARP entries in the kernel table.
1429 		 */
1430 		if (RT_ISHOST(RT)
1431 		    && ntohl(RT->rt_dst) == RT->rt_gate)
1432 			return 0;
1433 
1434 	} else {
1435 		/* This is an interface route.
1436 		 * Do not install routes for "external" remote interfaces.
1437 		 */
1438 		if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL))
1439 			return 0;
1440 
1441 		/* Interfaces should override received routes.
1442 		 */
1443 		pref = 0;
1444 		ags |= (AGS_IF | AGS_CORS_GATE);
1445 
1446 		/* If it is not an interface, or an alias for an interface,
1447 		 * it must be a "gateway."
1448 		 *
1449 		 * If it is a "remote" interface, it is also a "gateway" to
1450 		 * the kernel if is not an alias.
1451 		 */
1452 		if (RT->rt_ifp == 0
1453 		    || (RT->rt_ifp->int_state & IS_REMOTE))
1454 			ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1455 	}
1456 
1457 	/* If RIP is off and IRDP is on, let the route to the discovered
1458 	 * route suppress any RIP routes.  Eventually the RIP routes
1459 	 * will time-out and be deleted.  This reaches the steady-state
1460 	 * quicker.
1461 	 */
1462 	if ((RT->rt_state & RS_RDISC) && rip_sock < 0)
1463 		ags |= AGS_CORS_GATE;
1464 
1465 	metric = RT->rt_metric;
1466 	if (metric == HOPCNT_INFINITY) {
1467 		/* if the route is dead, so try hard to aggregate. */
1468 		pref = HOPCNT_INFINITY;
1469 		ags |= (AGS_FINE_GATE | AGS_SUPPRESS);
1470 		ags &= ~(AGS_IF | AGS_CORS_GATE);
1471 	}
1472 
1473 	ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0,
1474 		 metric,pref, 0, 0, ags, kern_out);
1475 	return 0;
1476 #undef RT
1477 }
1478 
1479 
1480 /* Update the kernel table to match the daemon table.
1481  */
1482 static void
fix_kern(void)1483 fix_kern(void)
1484 {
1485 	int i;
1486 	struct khash *k, **pk;
1487 
1488 
1489 	need_kern = age_timer;
1490 
1491 	/* Walk daemon table, updating the copy of the kernel table.
1492 	 */
1493 	(void)rn_walktree(rhead, walk_kern, 0);
1494 	ag_flush(0,0,kern_out);
1495 
1496 	for (i = 0; i < KHASH_SIZE; i++) {
1497 		for (pk = &khash_bins[i]; (k = *pk) != NULL; ) {
1498 			/* Do not touch static routes */
1499 			if (k->k_state & KS_STATIC) {
1500 				kern_check_static(k,0);
1501 				pk = &k->k_next;
1502 				continue;
1503 			}
1504 
1505 			/* check hold on routes deleted by the operator */
1506 			if (k->k_keep > now.tv_sec) {
1507 				/* ensure we check when the hold is over */
1508 				LIM_SEC(need_kern, k->k_keep);
1509 				/* mark for the next cycle */
1510 				k->k_state |= KS_DELETE;
1511 				pk = &k->k_next;
1512 				continue;
1513 			}
1514 
1515 			if ((k->k_state & KS_DELETE)
1516 			    && !(k->k_state & KS_DYNAMIC)) {
1517 				kern_ioctl(k, RTM_DELETE, 0);
1518 				*pk = k->k_next;
1519 				free(k);
1520 				continue;
1521 			}
1522 
1523 			if (k->k_state & KS_DEL_ADD)
1524 				kern_ioctl(k, RTM_DELETE, 0);
1525 
1526 			if (k->k_state & KS_ADD) {
1527 				kern_ioctl(k, RTM_ADD,
1528 					   ((0 != (k->k_state & (KS_GATEWAY
1529 							| KS_DYNAMIC)))
1530 					    ? RTF_GATEWAY : 0));
1531 			} else if (k->k_state & KS_CHANGE) {
1532 				kern_ioctl(k,  RTM_CHANGE,
1533 					   ((0 != (k->k_state & (KS_GATEWAY
1534 							| KS_DYNAMIC)))
1535 					    ? RTF_GATEWAY : 0));
1536 			}
1537 			k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD);
1538 
1539 			/* Mark this route to be deleted in the next cycle.
1540 			 * This deletes routes that disappear from the
1541 			 * daemon table, since the normal aging code
1542 			 * will clear the bit for routes that have not
1543 			 * disappeared from the daemon table.
1544 			 */
1545 			k->k_state |= KS_DELETE;
1546 			pk = &k->k_next;
1547 		}
1548 	}
1549 }
1550 
1551 
1552 /* Delete a static route in the image of the kernel table.
1553  */
1554 void
del_static(naddr dst,naddr mask,naddr gate,int gone)1555 del_static(naddr dst,
1556 	   naddr mask,
1557 	   naddr gate,
1558 	   int gone)
1559 {
1560 	struct khash *k;
1561 	struct rt_entry *rt;
1562 
1563 	/* Just mark it in the table to be deleted next time the kernel
1564 	 * table is updated.
1565 	 * If it has already been deleted, mark it as such, and set its
1566 	 * keep-timer so that it will not be deleted again for a while.
1567 	 * This lets the operator delete a route added by the daemon
1568 	 * and add a replacement.
1569 	 */
1570 	k = kern_find(dst, mask, 0);
1571 	if (k != NULL && (gate == 0 || k->k_gate == gate)) {
1572 		k->k_state &= ~(KS_STATIC | KS_DYNAMIC | KS_CHECK);
1573 		k->k_state |= KS_DELETE;
1574 		if (gone) {
1575 			k->k_state |= KS_DELETED;
1576 			k->k_keep = now.tv_sec + K_KEEP_LIM;
1577 		}
1578 	}
1579 
1580 	rt = rtget(dst, mask);
1581 	if (rt != NULL && (rt->rt_state & RS_STATIC))
1582 		rtbad(rt);
1583 }
1584 
1585 
1586 /* Delete all routes generated from ICMP Redirects that use a given gateway,
1587  * as well as old redirected routes.
1588  */
1589 void
del_redirects(naddr bad_gate,time_t old)1590 del_redirects(naddr bad_gate,
1591 	      time_t old)
1592 {
1593 	int i;
1594 	struct khash *k;
1595 
1596 
1597 	for (i = 0; i < KHASH_SIZE; i++) {
1598 		for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1599 			if (!(k->k_state & KS_DYNAMIC)
1600 			    || (k->k_state & KS_STATIC))
1601 				continue;
1602 
1603 			if (k->k_gate != bad_gate
1604 			    && k->k_redirect_time > old
1605 			    && !supplier)
1606 				continue;
1607 
1608 			k->k_state |= KS_DELETE;
1609 			k->k_state &= ~KS_DYNAMIC;
1610 			need_kern.tv_sec = now.tv_sec;
1611 			trace_act("mark redirected %s --> %s for deletion",
1612 				  addrname(k->k_dst, k->k_mask, 0),
1613 				  naddr_ntoa(k->k_gate));
1614 		}
1615 	}
1616 }
1617 
1618 
1619 /* Start the daemon tables.
1620  */
1621 extern int max_keylen;
1622 
1623 void
rtinit(void)1624 rtinit(void)
1625 {
1626 	int i;
1627 	struct ag_info *ag;
1628 
1629 	/* Initialize the radix trees */
1630 	max_keylen = sizeof(struct sockaddr_in);
1631 	rn_init();
1632 	rn_inithead(&rhead, 32);
1633 
1634 	/* mark all of the slots in the table free */
1635 	ag_avail = ag_slots;
1636 	for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) {
1637 		ag->ag_fine = ag+1;
1638 		ag++;
1639 	}
1640 }
1641 
1642 
1643 #ifdef _HAVE_SIN_LEN
1644 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET, 0, {0}, {0}};
1645 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET, 0, {0}, {0}};
1646 #else
1647 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET};
1648 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET};
1649 #endif
1650 
1651 
1652 static void
set_need_flash(void)1653 set_need_flash(void)
1654 {
1655 	if (!need_flash) {
1656 		need_flash = 1;
1657 		/* Do not send the flash update immediately.  Wait a little
1658 		 * while to hear from other routers.
1659 		 */
1660 		no_flash.tv_sec = now.tv_sec + MIN_WAITTIME;
1661 	}
1662 }
1663 
1664 
1665 /* Get a particular routing table entry
1666  */
1667 struct rt_entry *
rtget(naddr dst,naddr mask)1668 rtget(naddr dst, naddr mask)
1669 {
1670 	struct rt_entry *rt;
1671 
1672 	dst_sock.sin_addr.s_addr = dst;
1673 	mask_sock.sin_addr.s_addr = htonl(mask);
1674 	masktrim(&mask_sock);
1675 	rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead);
1676 	if (!rt
1677 	    || rt->rt_dst != dst
1678 	    || rt->rt_mask != mask)
1679 		return 0;
1680 
1681 	return rt;
1682 }
1683 
1684 
1685 /* Find a route to dst as the kernel would.
1686  */
1687 struct rt_entry *
rtfind(naddr dst)1688 rtfind(naddr dst)
1689 {
1690 	dst_sock.sin_addr.s_addr = dst;
1691 	return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead);
1692 }
1693 
1694 
1695 /* add a route to the table
1696  */
1697 void
rtadd(naddr dst,naddr mask,u_int state,struct rt_spare * new)1698 rtadd(naddr	dst,
1699       naddr	mask,
1700       u_int	state,			/* rt_state for the entry */
1701       struct	rt_spare *new)
1702 {
1703 	struct rt_entry *rt;
1704 	naddr smask;
1705 	int i;
1706 	struct rt_spare *rts;
1707 
1708 	rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd");
1709 	memset(rt, 0, sizeof(*rt));
1710 	for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++)
1711 		rts->rts_metric = HOPCNT_INFINITY;
1712 
1713 	rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock;
1714 	rt->rt_dst = dst;
1715 	rt->rt_dst_sock.sin_family = AF_INET;
1716 #ifdef _HAVE_SIN_LEN
1717 	rt->rt_dst_sock.sin_len = dst_sock.sin_len;
1718 #endif
1719 	if (mask != HOST_MASK) {
1720 		smask = std_mask(dst);
1721 		if ((smask & ~mask) == 0 && mask > smask)
1722 			state |= RS_SUBNET;
1723 	}
1724 	mask_sock.sin_addr.s_addr = htonl(mask);
1725 	masktrim(&mask_sock);
1726 	rt->rt_mask = mask;
1727 	rt->rt_state = state;
1728 	rt->rt_spares[0] = *new;
1729 	rt->rt_time = now.tv_sec;
1730 	rt->rt_poison_metric = HOPCNT_INFINITY;
1731 	rt->rt_seqno = update_seqno;
1732 
1733 	if (++total_routes == MAX_ROUTES)
1734 		msglog("have maximum (%d) routes", total_routes);
1735 	if (TRACEACTIONS)
1736 		trace_add_del("Add", rt);
1737 
1738 	need_kern.tv_sec = now.tv_sec;
1739 	set_need_flash();
1740 
1741 	if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock,
1742 				    rhead, rt->rt_nodes)) {
1743 		msglog("rnh_addaddr() failed for %s mask=%#lx",
1744 		       naddr_ntoa(dst), (u_long)mask);
1745 		free(rt);
1746 	}
1747 }
1748 
1749 
1750 /* notice a changed route
1751  */
1752 void
rtchange(struct rt_entry * rt,u_int state,struct rt_spare * new,char * label)1753 rtchange(struct rt_entry *rt,
1754 	 u_int	state,			/* new state bits */
1755 	 struct rt_spare *new,
1756 	 char	*label)
1757 {
1758 	if (rt->rt_metric != new->rts_metric) {
1759 		/* Fix the kernel immediately if it seems the route
1760 		 * has gone bad, since there may be a working route that
1761 		 * aggregates this route.
1762 		 */
1763 		if (new->rts_metric == HOPCNT_INFINITY) {
1764 			need_kern.tv_sec = now.tv_sec;
1765 			if (new->rts_time >= now.tv_sec - EXPIRE_TIME)
1766 				new->rts_time = now.tv_sec - EXPIRE_TIME;
1767 		}
1768 		rt->rt_seqno = update_seqno;
1769 		set_need_flash();
1770 	}
1771 
1772 	if (rt->rt_gate != new->rts_gate) {
1773 		need_kern.tv_sec = now.tv_sec;
1774 		rt->rt_seqno = update_seqno;
1775 		set_need_flash();
1776 	}
1777 
1778 	state |= (rt->rt_state & RS_SUBNET);
1779 
1780 	/* Keep various things from deciding ageless routes are stale.
1781 	 */
1782 	if (!AGE_RT(state, new->rts_ifp))
1783 		new->rts_time = now.tv_sec;
1784 
1785 	if (TRACEACTIONS)
1786 		trace_change(rt, state, new,
1787 			     label ? label : "Chg   ");
1788 
1789 	rt->rt_state = state;
1790 	rt->rt_spares[0] = *new;
1791 }
1792 
1793 
1794 /* check for a better route among the spares
1795  */
1796 static struct rt_spare *
rts_better(struct rt_entry * rt)1797 rts_better(struct rt_entry *rt)
1798 {
1799 	struct rt_spare *rts, *rts1;
1800 	int i;
1801 
1802 	/* find the best alternative among the spares */
1803 	rts = rt->rt_spares+1;
1804 	for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) {
1805 		if (BETTER_LINK(rt,rts1,rts))
1806 			rts = rts1;
1807 	}
1808 
1809 	return rts;
1810 }
1811 
1812 
1813 /* switch to a backup route
1814  */
1815 void
rtswitch(struct rt_entry * rt,struct rt_spare * rts)1816 rtswitch(struct rt_entry *rt,
1817 	 struct rt_spare *rts)
1818 {
1819 	struct rt_spare swap;
1820 	char label[10];
1821 
1822 
1823 	/* Do not change permanent routes */
1824 	if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC
1825 				  | RS_NET_SYN | RS_IF)))
1826 		return;
1827 
1828 	/* find the best alternative among the spares */
1829 	if (rts == NULL)
1830 		rts = rts_better(rt);
1831 
1832 	/* Do not bother if it is not worthwhile.
1833 	 */
1834 	if (!BETTER_LINK(rt, rts, rt->rt_spares))
1835 		return;
1836 
1837 	swap = rt->rt_spares[0];
1838 	(void)sprintf(label, "Use #%d", (int)(rts - rt->rt_spares));
1839 	rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), rts, label);
1840 	if (swap.rts_metric == HOPCNT_INFINITY) {
1841 		*rts = rts_empty;
1842 	} else {
1843 		*rts = swap;
1844 	}
1845 }
1846 
1847 
1848 void
rtdelete(struct rt_entry * rt)1849 rtdelete(struct rt_entry *rt)
1850 {
1851 	struct khash *k;
1852 
1853 
1854 	if (TRACEACTIONS)
1855 		trace_add_del("Del", rt);
1856 
1857 	k = kern_find(rt->rt_dst, rt->rt_mask, 0);
1858 	if (k != NULL) {
1859 		k->k_state |= KS_DELETE;
1860 		need_kern.tv_sec = now.tv_sec;
1861 	}
1862 
1863 	dst_sock.sin_addr.s_addr = rt->rt_dst;
1864 	mask_sock.sin_addr.s_addr = htonl(rt->rt_mask);
1865 	masktrim(&mask_sock);
1866 	if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock,
1867 							rhead)) {
1868 		msglog("rnh_deladdr() failed");
1869 	} else {
1870 		free(rt);
1871 		total_routes--;
1872 	}
1873 }
1874 
1875 
1876 void
rts_delete(struct rt_entry * rt,struct rt_spare * rts)1877 rts_delete(struct rt_entry *rt,
1878 	   struct rt_spare *rts)
1879 {
1880 	trace_upslot(rt, rts, &rts_empty);
1881 	*rts = rts_empty;
1882 }
1883 
1884 
1885 /* Get rid of a bad route, and try to switch to a replacement.
1886  */
1887 static void
rtbad(struct rt_entry * rt)1888 rtbad(struct rt_entry *rt)
1889 {
1890 	struct rt_spare new;
1891 
1892 	/* Poison the route */
1893 	new = rt->rt_spares[0];
1894 	new.rts_metric = HOPCNT_INFINITY;
1895 	rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), &new, 0);
1896 	rtswitch(rt, 0);
1897 }
1898 
1899 
1900 /* Junk a RS_NET_SYN or RS_LOCAL route,
1901  *	unless it is needed by another interface.
1902  */
1903 void
rtbad_sub(struct rt_entry * rt)1904 rtbad_sub(struct rt_entry *rt)
1905 {
1906 	struct interface *ifp, *ifp1;
1907 	struct intnet *intnetp;
1908 	u_int state;
1909 
1910 
1911 	ifp1 = NULL;
1912 	state = 0;
1913 
1914 	if (rt->rt_state & RS_LOCAL) {
1915 		/* Is this the route through loopback for the interface?
1916 		 * If so, see if it is used by any other interfaces, such
1917 		 * as a point-to-point interface with the same local address.
1918 		 */
1919 		LIST_FOREACH(ifp, &ifnet, int_list) {
1920 			/* Retain it if another interface needs it.
1921 			 */
1922 			if (ifp->int_addr == rt->rt_ifp->int_addr) {
1923 				state |= RS_LOCAL;
1924 				ifp1 = ifp;
1925 				break;
1926 			}
1927 		}
1928 
1929 	}
1930 
1931 	if (!(state & RS_LOCAL)) {
1932 		/* Retain RIPv1 logical network route if there is another
1933 		 * interface that justifies it.
1934 		 */
1935 		if (rt->rt_state & RS_NET_SYN) {
1936 			LIST_FOREACH(ifp, &ifnet, int_list) {
1937 				if ((ifp->int_state & IS_NEED_NET_SYN)
1938 				    && rt->rt_mask == ifp->int_std_mask
1939 				    && rt->rt_dst == ifp->int_std_addr) {
1940 					state |= RS_NET_SYN;
1941 					ifp1 = ifp;
1942 					break;
1943 				}
1944 			}
1945 		}
1946 
1947 		/* or if there is an authority route that needs it. */
1948 		for (intnetp = intnets;
1949 		     intnetp != NULL;
1950 		     intnetp = intnetp->intnet_next) {
1951 			if (intnetp->intnet_addr == rt->rt_dst
1952 			    && intnetp->intnet_mask == rt->rt_mask) {
1953 				state |= (RS_NET_SYN | RS_NET_INT);
1954 				break;
1955 			}
1956 		}
1957 	}
1958 
1959 	if (ifp1 != NULL || (state & RS_NET_SYN)) {
1960 		struct rt_spare new = rt->rt_spares[0];
1961 		new.rts_ifp = ifp1;
1962 		rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN|RS_LOCAL)) | state),
1963 			 &new, 0);
1964 	} else {
1965 		rtbad(rt);
1966 	}
1967 }
1968 
1969 
1970 /* Called while walking the table looking for sick interfaces
1971  * or after a time change.
1972  */
1973 /* ARGSUSED */
1974 int
walk_bad(struct radix_node * rn,struct walkarg * argp UNUSED)1975 walk_bad(struct radix_node *rn,
1976 	 struct walkarg *argp UNUSED)
1977 {
1978 #define RT ((struct rt_entry *)rn)
1979 	struct rt_spare *rts;
1980 	int i;
1981 
1982 
1983 	/* fix any spare routes through the interface
1984 	 */
1985 	rts = RT->rt_spares;
1986 	for (i = NUM_SPARES; i != 1; i--) {
1987 		rts++;
1988 		if (rts->rts_metric < HOPCNT_INFINITY
1989 		    && (rts->rts_ifp == NULL
1990 			|| (rts->rts_ifp->int_state & IS_BROKE)))
1991 			rts_delete(RT, rts);
1992 	}
1993 
1994 	/* Deal with the main route
1995 	 */
1996 	/* finished if it has been handled before or if its interface is ok
1997 	 */
1998 	if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE))
1999 		return 0;
2000 
2001 	/* Bad routes for other than interfaces are easy.
2002 	 */
2003 	if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) {
2004 		rtbad(RT);
2005 		return 0;
2006 	}
2007 
2008 	rtbad_sub(RT);
2009 	return 0;
2010 #undef RT
2011 }
2012 
2013 
2014 /* Check the age of an individual route.
2015  */
2016 /* ARGSUSED */
2017 static int
walk_age(struct radix_node * rn,struct walkarg * argp UNUSED)2018 walk_age(struct radix_node *rn,
2019 	   struct walkarg *argp UNUSED)
2020 {
2021 #define RT ((struct rt_entry *)rn)
2022 	struct interface *ifp;
2023 	struct rt_spare *rts;
2024 	int i;
2025 
2026 
2027 	/* age all of the spare routes, including the primary route
2028 	 * currently in use
2029 	 */
2030 	rts = RT->rt_spares;
2031 	for (i = NUM_SPARES; i != 0; i--, rts++) {
2032 
2033 		ifp = rts->rts_ifp;
2034 		if (i == NUM_SPARES) {
2035 			if (!AGE_RT(RT->rt_state, ifp)) {
2036 				/* Keep various things from deciding ageless
2037 				 * routes are stale
2038 				 */
2039 				rts->rts_time = now.tv_sec;
2040 				continue;
2041 			}
2042 
2043 			/* forget RIP routes after RIP has been turned off.
2044 			 */
2045 			if (rip_sock < 0) {
2046 				rtdelete(RT);
2047 				return 0;
2048 			}
2049 		}
2050 
2051 		/* age failing routes
2052 		 */
2053 		if (age_bad_gate == rts->rts_gate
2054 		    && rts->rts_time >= now_stale) {
2055 			rts->rts_time -= SUPPLY_INTERVAL;
2056 		}
2057 
2058 		/* trash the spare routes when they go bad */
2059 		if (rts->rts_metric < HOPCNT_INFINITY
2060 		    && now_garbage > rts->rts_time
2061 		    && i != NUM_SPARES)
2062 			rts_delete(RT, rts);
2063 	}
2064 
2065 
2066 	/* finished if the active route is still fresh */
2067 	if (now_stale <= RT->rt_time)
2068 		return 0;
2069 
2070 	/* try to switch to an alternative */
2071 	rtswitch(RT, 0);
2072 
2073 	/* Delete a dead route after it has been publicly mourned. */
2074 	if (now_garbage > RT->rt_time) {
2075 		rtdelete(RT);
2076 		return 0;
2077 	}
2078 
2079 	/* Start poisoning a bad route before deleting it. */
2080 	if (now.tv_sec - RT->rt_time > EXPIRE_TIME) {
2081 		struct rt_spare new = RT->rt_spares[0];
2082 		new.rts_metric = HOPCNT_INFINITY;
2083 		rtchange(RT, RT->rt_state, &new, 0);
2084 	}
2085 	return 0;
2086 }
2087 
2088 
2089 /* Watch for dead routes and interfaces.
2090  */
2091 void
age(naddr bad_gate)2092 age(naddr bad_gate)
2093 {
2094 	struct interface *ifp;
2095 	int need_query = 0;
2096 
2097 	/* If not listening to RIP, there is no need to age the routes in
2098 	 * the table.
2099 	 */
2100 	age_timer.tv_sec = (now.tv_sec
2101 			    + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL));
2102 
2103 	/* Check for dead IS_REMOTE interfaces by timing their
2104 	 * transmissions.
2105 	 */
2106 	LIST_FOREACH(ifp, &ifnet, int_list) {
2107 		if (!(ifp->int_state & IS_REMOTE))
2108 			continue;
2109 
2110 		/* ignore unreachable remote interfaces */
2111 		if (!check_remote(ifp))
2112 			continue;
2113 
2114 		/* Restore remote interface that has become reachable
2115 		 */
2116 		if (ifp->int_state & IS_BROKE)
2117 			if_ok(ifp, "remote ");
2118 
2119 		if (ifp->int_act_time != NEVER
2120 		    && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) {
2121 			msglog("remote interface %s to %s timed out after"
2122 			       " %ld:%ld",
2123 			       ifp->int_name,
2124 			       naddr_ntoa(ifp->int_dstaddr),
2125 			       (long)(now.tv_sec - ifp->int_act_time)/60,
2126 			       (long)(now.tv_sec - ifp->int_act_time)%60);
2127 			if_sick(ifp);
2128 		}
2129 
2130 		/* If we have not heard from the other router
2131 		 * recently, ask it.
2132 		 */
2133 		if (now.tv_sec >= ifp->int_query_time) {
2134 			ifp->int_query_time = NEVER;
2135 			need_query = 1;
2136 		}
2137 	}
2138 
2139 	/* Age routes. */
2140 	age_bad_gate = bad_gate;
2141 	(void)rn_walktree(rhead, walk_age, 0);
2142 
2143 	/* delete old redirected routes to keep the kernel table small
2144 	 * and prevent blackholes
2145 	 */
2146 	del_redirects(bad_gate, now.tv_sec-STALE_TIME);
2147 
2148 	/* Update the kernel routing table. */
2149 	fix_kern();
2150 
2151 	/* poke reticent remote gateways */
2152 	if (need_query)
2153 		rip_query();
2154 }
2155